P.S.

Aleksandrov

COMBINATORIAL
TOPOLOGY
Volume 1
COMBINATORIAL TOPOLOGY
VOLUME 1
 OTHER GRAY LOCK PUBLICATIONS

ALEKSANDROV: C om bin atorial Topology

Vol. 1: Introduction. C om plexes.
C o verin g s. D im ension
Vol. 2: The B e tti Groups
Vol. 3: H om ological M an ifolds. The D uality
T h eo rem s. Cohom ology Groups o f C om pacta.
Continuous M appings o f P olyh edra

KHINCHIN: T h ree P e a r ls o f N um ber T heory

M ath em atical Foundations o f Quantum
S ta tis tic s

KOLMOGOROV and FOMIN: E lem en ts o f the T heory o f Functions

and F unctional A n a ly sis
Vol. 1: M e tric and N o rm ed Spaces
Vol. 2: M easu re. The L ebesgu e In tegral.
H ilb ert Space

NOVOZHILOV: Foundations o f the N onlinear T heory o f

E la s tic ity

PETROVSKII: L e c tu re s on the T h eory o f In teg ra l Equations

PONTRYAGIN: Foundations o f C o m bin atorial T opology

TIETZE: F am ous P ro b le m s o f M a th em a tics

COMBINATORIAL TOPOLOGY

VOLUM E 1

BY

P. S. ALEKSANDROV

G R A Y L O C K P R E S S

B A L T IM O R E , M D .

1956
TRANSLATED FROM THE FIRST (1947) RUSSIAN EDITION

BY
HORACE KOMM

Copyright, 1956, by

GRAYLOCK PRESS
428 E. Preston Street
Baltimore, Md. 21202

Third Printing, October 1969

All rights reserved. This book, or

parts thereof, may not be reproduced
in any form, or translated, without
permission in writing from the
publishers.

L ibrary of Congress Catalog Card Number 56-13930

Second Printing January 1965

Manufactured in the United States o f America

 T R A N S L A T O R ’S N O T E
This volume is a translation of the first third of P. S. Aleksandrov’s
Kombinatornaya Topologiya. An appendix on the analytic geometry of
Euclidean n-space is also included. The volume, complete in itself, deals
with certain classical problems such as the Jordan curve theorem and the
classification of closed surfaces without using the formal techniques of
homology theory. The elementary but rigorous treatm ent of these prob­
lems, the introductory chapters on complexes and coverings and their
applications to dimension theory, and the large number of examples and
pictures should provide an excellent intuitive background for further
study in combinatorial topology.
In Chapter I the references have been expanded to include a number of
standard works in English. References to these and to the books and papers
cited in Chapter I of the original are listed at the end of the chapter and
correspond to the numbers enclosed in brackets in the body of the text.
References in the remaining chapters are enclosed in brackets, capital
letters referring to books and lower case letters to papers. These refer to
the bibliography at the end of the book. The bibliography includes all
papers mentioned in the original edition and a few which have been added
by the translator. English translations are cited wherever possible. Cross-
references to items in the text are made by citing chapter and section.
Where the chapter number is omitted, the reference is to a section of the
chapter being read. The system of transliteration used is that of the M athe­
matical Reviews. This may be confusing only in the cases of Aleksandrov
and Tihonov, whose names are usually written in English as Alexandroff
and Ty chon off.

16138
 C O N TEN TS
Preface............................................................................................................. xiii

PART ONE
INTRODUCTION
Chapter
I. S urvey of the E lementary P roperties of T opological
Spaces
1. Notation of Set Theory.............................................................. 2
1.1. Operations on sets................................................................ 2
1.2. Mappings.............................................................................. 3
1.3. Indexed sets; systems of sets; order of a system of sets;
coverings............................................................................... 3
2. Topological Spaces....................................................................... 4
2.1. Definition of topological spaces and basic related no­
tions. . : ................................................................................. 4
2.2. The neighborhood topology.............................................. 6
2.3. Metric and metrizable spaces.................. 7
2.4. Continuous mappings......................................................... 9
2.5. Uniform convergence of mappings.................................... 10
2.G. Topological product of spaces.......................................... 10
3. Connectedness.............................................................................. 13
3.1. Definition and basic theorems........................................... 13
3.2. Components.......................................................................... 15
4. Separation Axioms. Compactness.............................................. 16
4.1. Separation axioms............................................................... 16
4.2. Theorems on continuous functions in normal spaces. . . . 17
4.3. Compactness......................................................................... 17
4.4. Further theorems on bicompacta. Metrization and im­
bedding theorems................................................................. 18
4.5. Coptinuous mappings of bicompacta................................ 18
5. Upper Semi-Continuous Decompositions of Compacta and
Their Relation to Continuous Mappings (Identifications).
Locally Compact Spaces.Topological Manifolds. Examples. . 18
5.1. Upper semi-continuous decompositions. The space of a
given decomposition............................................................ 18
5.2. Examples of upper semi-continuous decompositions and
identifications. The projective n-space..............: ............. 20
5.3. Locally compact spaces. Topological manifolds. Exam­
ples ......................................................................................... 24
vii
vm CONTENTS

Chaplet
6. Partially Ordered Sets and Discrete Spaces............................. 25
6.1. Definitions............................................................................. 25
6.2. Examples of partially orderedsets...................................... 26
6.3. The sets -Ao(p) and Oe(p) .................................................... 27
6.4. Duality of partially ordered sets........................................ 27
6.5. Discrete spaces..................................................................... 28
7. Complete Metric Spaces and Com pacta................................... 29
7.1. Definitions and simplest properties of complete metric
spaces..................................................................................... 29
7.2. c-nets in compacta................................................................ 30
7.3. The space of continuous mappings.............‘ ...................... 31
7.4. Deformations. Homotopy classes of mappings of a com-
pactum X into a compactum Y ......................................... 32
8. Coverings of Normal Spaces and, In Particular, of Compacta 32
8.1. Closed and open coverings of topological spaces. The
method of combinatorial topology.................................... 32
8.2. Similar coverings.................................................................. 33
8.3. e-coverings of compacta. Lebesgue numbers of a cover­
in g .......................................................................................... 35
8.4. Definition of dimension....................................................... 36
II. T he J ordan T heorem
1. Formulation of the Jordan Theorem. Common Boundaries of
Domains........................................................................................ 39
1.1. Formulation of the Jordan theorem.................................. 39
1.2. Domains in R n and their boundaries................................ 40
1.3. Outline of the proof of the Jordan theorem ..................... 44
1.4. Notation. Orientation of simple arcs and simple closed
curves..................................................................................... 45
2. The Angle Function of a Continuous Mapping of a Segment
Into the Plane. The Index of a Point Relative to a Closed
Path in the Plane......................................................................... 46
2.1. The functions Fa(p, C, x) and f(p, C, xi , x2) .................. 46
2.2. The index of a point relative to a closed p a th ................ 47
2.3. The addition form ula.......................................................... 49
2.4. The index of a point relative to a Jordan curve.............. 50
2.5. The index of a point p 6 R 2 relative to a continuous
mapping of a circumference into R 2\ p ; degree of a
continuous mapping of a circumference into a circum­
ference .................................................................................... 52
3. Theorem: A Simple Arc Does Not Separate the Plane.......... 55
4. Proof of the Jordan Theorem..................................................... 60
 CONTENTS ix

Chapter
4.1. Fundamental auxiliary construction................................. 60
4.2. The case a>(s, 4>~*) = 0; the set T = R 2 \ consists of
at least two components..................................................... 61
4.3. Conclusion of the proof of the Jordan theorem .............. 63
III. S urfaces
1. Elementary Curves and 1-complexes......................................... 66
1.1. Elementary curves and their subdivision into arcs........ 66
1.2. The connectivity of a curve (one-dimensional Betti
num ber)................................................................................. 68
2. Surfaces and Their Triangulations............................................ 71
2.1. 2-complexes and polyhedra................................................ 71
2.2. Closed surfaces..................................................................... 73
2.3. Surfaces with boundary...................................................... 76
2.4. Subdivisions of triangulations............................................ 77
2.5. Skeleton complexes.............................................................. 78
3. Cuts and Identifications.............................................................. 79
3.1. Identification of elements in a skeleton complex.............. 79
3.2. Cut lines and semi-stars of the vertices of cut lines........ 82
3.3. The cut operation................................................................ 86
3.4. Reduction of holes............................................................... 90
4. Orientability of Surfaces............................................................. 93
4.1. Definitions............................................................................ 93
5. The Connectivity of a Surface.................................................. 98
5.1................................................................................................... 98
6. Simple Surfaces........................................................................... 100
6.1. Closed c u ts............................................................................ 100
6.2. Definition of simple triangulations. Invariance under
regular subdivisions............................................................. 102
6.3. Elementary lemmas............................................................. 104
6.4. Classification of simple surfaces......................................... 104
7. Classification of Closed Surfaces................................................ 108
7.1. Genus of a surface. Normal surface of a given genus. . . . 108
7.2. The fundamental theorem of the topology of surfaces. . 110

PART TWO
COMPLEXES. COVERINGS. DIM ENSION

IV. C om plexes
Introductory Section: Preliminary Remarks on Simplexes.......... 116
0.1. Simplexes and their skeletons............................................ 116
0.2. Faces...................................................................................... 116
X CONTENTS

Chapter
0.3. Combinatorial sum .............................................................. 117
0. 4. Closure of a simplex......................................................... 118
1. Basic Definitions........................................................................... 118
1.1. Triangulations. Examples: | T n | and T n......................... 118
1.2. Polyhedral complexes.......................................................... 119
1.3. Skeleton complexes.............................................................. 120
1.4. General definition of a simplicial complex........................ 121
1.5. Examples of simplicial complexes...................................... 122
1. G. Simplicial mappings and isomorphisms of skeleton and
simplicial complexes............................................................ 123
1.7. Definition of an abstract complex..................................... 125
1.8. Closed and open subcomplexes of a complex. Combina­
torial closures and stars...................................................... 126
1.9. Theorem on imbedding in R 2n+l........................................ 127
2. Some Notable Skeleton Complexes............................................ 128
2.1. The nerve of a finite system of sets.................................. 128
2.2. Barycentric derivation and barycentric subdivisions. . . 131
2.3. The cone of a complex........................................................ 134
2.4. Prisms over a skeleton complex......................................... 135
2.5. The prism spanned by a skeleton complex and its sim­
plicial image.......................................................................... 135
3. The Body of a Complex. Polyhedra.......................................... 136
3.1. Definitions............................................................................. 136
3.2. Star neighborhoods. Open stars......................................... 137
3.3. Simplicial mappings of triangulated polyhedra................. 138
4. Subdivisions of Polyhedral Complexes......... •.......................... 139
4.1. Definition of a subdivision.................................................. 139
4.2. Successive barycentric subdivisions.................................. 140
4.3. Central and elementary subdivisionsof complexes.......... 140
4.4. Subdivisions of nonclosed subcomplexes of polyhedral
complexes.............................................................................. 143
5. Barycentric S tars.......................................................................... 143
5.1. Barycentric sta rs.................................................................. 143
5.2. The dual complex of a triangulation................................. 144
5.3. Closed barycentric sta rs...................................................... 146
5.4. Subcomplexes of the complex K*-, their bodies and bary­
centric subdivisions.............................................................. 147
6. Topological Complexes and TopologicalPolyhedra................. 149
6.1. Definitions............................................................................. 149
6.2. ?i-manifolds............................................................................ 150
7. Connectedness of Complexes....................................................... 151
 CONTENTS xi

Chapter
7.1. Connected complexes. Components.................................. 152
7.2. The case of unrestricted simplicial complexes................ 153
7.3. The components of || K || and K ...................................... 154
V. Spehneh’s Lemma and Its Corollaries
1. Preliminary Remarks................................................................... 156
1.1. Triangulations and subdivisions of a closed simplex. . . . 156
2. Sperner’s Lemma......................................................................... 160
2.1. Sperner’s lemma.................................................................... 160
2.2. A corollary of Sperner’s lemma. Conclusion of the proof
of the Pflastersatz................................................................ 161
2.3. Invariance of the dimension number of R n...................... 163
3. Invariance of Interior Points...................................................... 164
3.1................................................................................................... 164
3.2. Invariance of interior points of topological manifolds. . . 168
4. The Fixed Point Theorem for an n-cell.................................... 168
VI. I ntroduction to D imension T heory
1. Theorems one-displacements andImbedding in R n................... 170
1.1. Definition of e-mappings and e-displacemcnts. Outline
of the section........................................................................ 170
1.2. Retraction............................................................................. 173
1.3. Barycentric mapping of a space into the nerve of an open
covering................................................................................. 175
1.4. Theorem on e-mappings...................................................... 175
1.5. Barycentric approximations of a continuous mapping of
a compactum into R n. Theorem on e-displacements. .. . 177
1.6. Imbedding theorem for compacta..................................... 179
2. Theorem on Essential Mappings............................................... 180
2.1. Definition. Statement of the theorem............................... 180
2.2. Proof of Theorem 2.1.......................................................... 182
3. The Sum Theorem. Inductive Dimension................................ 184
3.1. Statement and proof ofthe sum theorem......................... 184
3.2. Inductive dimension............................................................ 186
4. Sequences of Subdivisions........................................................... 188
4.1. Irreducible ^coverings........................................................ 188
4.2. Subdivisions.......................................................................... 190
5. Some Applications to Topological Manifolds and Polyhedra 198
5.1. The case of topological manifolds (in particular, R n and
S n)................................................................................................. : 198
.

5.2. Strong connectedness.......................................................... 199

Appendix 1................................................................................................... 202
B i b l i o g r a p h y ............................................................................................... 216
I ndex ............................................................................................................. 218
 PREFACE

This book is an introduction to modern homology theory. It can be

understood by anyone familiar with general set-theoretic and algebraic
concepts and has been designed for the reader striving to acquire a knowl­
edge of topology through a systematic study of its essentials. Hence, in
this book, a reader can become acquainted with the ideas of modern
topology only by a detailed study of the fundamental topological facts. I
have endeavored to present these facts together with the necessary tech­
nical apparatus, often cumbersome and not at all attractive, with all
logical rigor and at the risk of being boring and tiresome at times. The
book can serve as a text for graduate students specializing in topology or
any other branch of mathematics related to topology.
In writing this book I have made extensive use of Topologie I (see Alexan-
droff [A-H]), a joint work of the well known Swiss mathematician H.
Hopf and myself. In particular, Chapter XVII of the present work is a
translation, and Chapter XVI a revision, of Chapters XIV and X II,
respectively, of Topologie I. I deem it especially necessary to emphasize
this debt, since these two chapters of Topologie I were written in their
entirety by H. Hopf. Both appendices of the present book (on Abelian
groups and on the analytic geometry of n-dimensional space) are also in
essence borrowed from the book of Alexandroff-Hopf, but are, in fact, con­
siderably abridged. Besides these fundamental extracts, there are more
superficial ones scattered throughout the text. It is hardly necessary to
enumerate these individually.
Chapters V III and IX are central to the whole book. They deal with
what is called combinatorial homology theory. The necessary technical
apparatus, complexes and chains, is presented in Chapters IV and VII.
The approach in Chapter IV is combinatorial-geometric, while that in
Chapter VII is algebraic. Chapters V III and IX study not only the usual
(“lower” or A-) homology but also the so called “upper” or V-homology
(cohomology). The latter is constantly acquiring more and more signifi­
cance in modern topological research. I have tried to expound the whole
theory as simply and clearly as possible, elucidating it with a great many
elementary examples whose assimilation is important for an understanding
of the essential material.
The topological invariance of homology theory is proved in Chapters
X and X I, which contain several different ways of proving the invariance
of the Betti groups. In this connection I have completely dispensed with
the so called “continuous” (singular) cycles, since it seems to me necessary
to look to the “true” (proper) cycles, rather than to the singular cycles,
xiii
XIV PREFACE

as the basic path of application of homology concepts to arbitrary com-

pacta. True cycles are also convenient in practice in the study of poly-
hedra, in all cases where it is reasonable to use concepts whose very formu­
lation is topologically invariant (e.g., in applications to the calculus of
variations). In the principal questions of the construction of a general
homology theory of compacta (particularly if one has in mind the general­
ization of this theory to the nonmetrizable case, first of all to arbitrary
bicompacta) it is of course preferable to use “spectral theory” (developed
in Chapter XIV), based on the direct investigation of the partially ordered
set of all finite open or closed coverings of the space. For the study of
the homology properties of a concrete individual compactum, however,
true cycles defined in terms of the'm etric of the space are the most ex­
pedient.
A further development of the homology theory of modern topology is its
localization, which reduces (by means of the notions of relative cycle and
relative homology) to the definition of the Betti groups in a point. This
makes it possible to study, e.g., manifolds with boundary, and also to de­
velop the homology dimension theory. Of the latter we give in this book
merely the definition of the homology dimension and a proof of the fact
that, for polyhedra, this dimension is synonymous with the dimension
in the elementary sense. Chapter X II is devoted to relative cycles and their
applications.
At the present time we consider homology theory to be the fundamental
core of topology because an extraordinarily significant number of the new
geometric facts discovered in topology are formulated in terms of homology.
Among the most important of these are the Poinear6 and Alexander-Pon-
tryagin dualities. Chapters X III-X V are essentially devoted to these.
The concept of homological manifold, the elements of the theory of inter­
sections, as well as the theory of linking, also find their natural place here.
The latter theory, in its elementary form also turns out to be a conveni­
ent method for investigating the simplest classical questions of the theory
of continuous mappings. Chapter XIV gives an account of this theory.
Finally, the celebrated Lefsehetz-Hopf formula for the algebraic number
of fixed points of a continuous mapping of a polyhedron into itself is proved
in Chapter XVII (borrowed in its entirety, as has already been said, from
my joint work with Hopf).
The reader may choose from all the wealth of concrete factual material
to which homology theory leads that which he finds interesting or useful.
For example, one can, studying Chapter V III and taking from Chapters
X -X I any one of the proofs of the invariance of the Betti groups, pass at
once to the theory of linking (Chapter XV) and then to the theory of con­
tinuous mappings (Chapters XVI and XVII), limiting oneself, if desirable,
 PKEFACE XV

to the classical topics presented in Chapter XVI. Or, one can, after Chap­
ters VIII, IX , X, go on to Chapters X III and XIV, i.e., basically, to the
duality laws. Finally, the reader who wants to acquire the very elementary
topological notions can, in more or less arbitrary order, read Chapters II,
III, V ; or, if the predominant interest is in questions of set-theoretic charac­
ter, he can read Chapters V and VI, consulting the preceding chapters
(basically Chapter IV) merely as a reference. Introductory remarks at the
beginning of each part and chapter of the book will generally help the
reader to choose a suitable sequence for the study of various topics.

Needless to say, this book does not begin to exhaust even the basic
branches of modern combinatorial topology. Being oriented towards prob­
lems which are quite general and at the same time sufficiently elementary,
this book, as is made clear above, seeks such problems in the domain of
homology theory. The entire immense field of homotopy methods, from the
classical fundamental group of Poincare (see Seifert-Threlfall [S-T, Ch. 7])
to the homotopy groups of Hurewicz, and the recent research in homotopy
problems of H. Hopf, Pontryagin, Eilenberg, and many others, remain
entirely outside the scope of the book. Neither, however, have I touched on
many deep questions of homology theory: the theory of intersections and
products is represented merely by the wholly elementary case p + q =
n, i.e., only when a zero-dimensional image is obtained in the intersection.
It follows as a m atter of course that all the results of the theory of con­
tinuous mappings which depend on the methods of "products” and "inter­
sections” (Product Method of Lefschetz), as well as Hopf’s extension of the
theory, have been omitted. A series of papers which are due to appear in
forthcoming topological issues of Uspehi Matematiceskih Nauk may be
consulted on some of the topics mentioned (see Glezerman [a]). For a fur­
ther development of questions related to the duality laws see Aleksandrov
[f], [I propose to devote a special monograph to questions of duality and
homology dimension theory, in the hope of according these questions a
fuller and more lucid treatm ent than they have received in the literature
(see Aleksandrov [h, i, j]). Unfortunately, the methods developed in this
paper, written after the completion of the book, could not be used here;
otherwise, various improvements could have been made in Chapters X III
and XIV. I therefore recommend this paper especially as supplementary
literature. Hurewicz-Wallman’s Dimension Theory is recommended in the
same vein.
The manuscript of this book was completed in the summer of 1941.
Naturally, the outbreak of events delayed publication for a long time.
After five years, as the book was about to appear, I was strongly tempted
to consider many desirable changes in its text because of the advances in
XVI PREFACE

Soviet and foreign scientific thought. I had to consider, first, the question
of notation and terminology (see Aleksandrov ffj). However, each minor
change in so unwieldy a work would have entailed others and would, in
the end, have caused me to undertake a revision of the whole book, again
delaying publication. I therefore decided to publish this manuscript in
its original form and to postpone all possible improvements (whose neces­
sity is more obvious to me than to anyone else) to a second edition.

In conclusion, I would like to thank all those who helped me in one way
or another in the preparation of the book.
First of all, I would like to express my gratitude to L. S. Pontryagin for
a close reading of the essential portions of the manuscript and for con­
tributing a whole series of valuable remarks and suggestions. He frequently
enabled me to improve the exposition substantially. I have noted in the
text all cases of new and simpler proofs of propositions (as, e.g., in Chapters
IV and X III) communicated to me by L. S. Pontryagin. The continual
friendly communication with L. S. Pontryagin over a long period of time
has helped me a great deal in the preparation of the book and has been of
great benefit to it. M. R. Sura-Bura has carefully read all of the book in
proof, and I am obliged to him for many improvements in the exposition
(especially in Chapter III).
I am obliged to A. N. Kolmogorov, A. S. Parhomenko, and A. M.
Rodnyanskii for a number of valuable remarks. I wish to thank A. N.
Kolmogorov, in addition, for the execution of many of the figures.
My assistant U. Smirnov, a student at the University of Moscow, to
whom I express here my sincere thanks, rendered me great assistance in
the preparation of the final text of this book and a number of the figures.
I am indebted to S. V. Fomin for his help in reading proof and for com­
piling the index.
Finally, I am very grateful to the publishers (OGIZ) in the persons of
its director G. F. Rybkin and the chief editor of mathematical literature
A. I. MarkuseviS for their scrupulous attention to my wishes in connec­
tion with the publication of this book.
P. A l e k s a n d r o v
Bolsevo, Komarovko
June 22, 1946
 Part One
IN T R O D U C T IO N

P art One consists of three chapters. Chapter I is auxiliary in character.

Chapters II and III are independent of each other and are concerned with
the most elementary questions of “geometric” topology: the Jordan theo­
rem and the elementary theory of closed surfaces. Chapters II and III,
together with Chapter V, constitute a central core of interesting and im­
portant topological ideas. Without utilizing the general concepts of com­
binatorial topology, these chapters make the importance of such concepts
quite clear.
 Chapter I
P R O P E R T IE S O F TO PO LO G IC A L SPA C E S

Chapter I is an outline of the elementary theory of topological spaces. It

is not meant for obligatory study. It may be used (after reading §1) merely
as a reference. On the other hand, the student may read the chapter as
part of the book and either furnish independent proofs where these are
inadequate or omitted, or look up the proofs in, e.g., Ilausdorff (see Trans­
lator’s Note and [1]).
The results of Chapter I will not be applied often in the sequel, but we
shall, of course, make systematic use of some of the concepts and proposi­
tions established here. In addition to such elementary ideas as continuous
mapping, closed and open sets, etc., these include the concepts of deforma­
tion, Lebesgue numbers of a covering, partially ordered sets, and the
theorem on the completeness of the metric space of continuous mappings
of one compactum into another. The latter is applied in Chapter VI.
The notion of a bicompactum is used just once in the book, in the defini­
tion of the cohomology groups of bicompacta (see XIV). It would be per­
fectly possible to confine oneself here again to compacta (the reader may
do so). Hence, all the elementary concepts introduced in Chapter I could
have been introduced for metric spaces alone. The student can be guided
by this suggestion as well in using Chapter I.
The exposition in this chapter is detailed in those cases which deal with
concepts essential to the applications in the sequel. The same can be said
of theorems cited here which are not to be found in Ilausdorff (see Transla­
tor’s Note and [1]). In the remaining cases the exposition has been abridged.
Needless to say, all definitions and the statements of all theorems have
been given in full.
§1. Notation of set theory
§1.1. Operations on sets. The union of the sets A, B, C, ■• • , i.e., the
set of all elements contained in at least one of the sets A, B, C, • • • , is de­
noted by A u B u C ■■• . Correspondingly, the union of the sets A a ,
where a is an index (see 1.3) which assumes a finite or infinite number of
values, is denoted by UaAa .
The intersection of the sets A, B, C, • • • , denoted by A n B n C • • • , is
the set of elements contained in all of the sets A, B, C, • • • . Similarly,
Dai4a denotes the intersection of the sets A a .
The difference of two sets A and B, denoted by A \ B, consists of those
elements of the set A which are not elements of the set B. This definition
does not presuppose that I? is a subset of A, so that
A \ B = A \ (A n B).
2
§ 1] NOTATION OF SET THEORY 3

The notation A Cl B, or B 3 A, means that the set A is a subset of

the set B, i.e., that every element of A is an element of B.
A a B, or B Z) A, means that A is a proper subset of B, i.e., that every
element of A is an element of B but that there is at least one element of B
which is not an element of A.
The relation “a is an element of the set A ” is written as a £ A.
The negation of the relations expressed by the symbols <Z, £, and the
like, is denoted by a line through these symbols. For example, a <£ A means
th at a is not an element of A.

§1.2. Mappings. A mapping f of a set X into a set Y is an assignment

to every element x of X of a definite element y = f(x ) of Y. The element
f(x) is called the image of the element x under the mapping /., If A C l ,
f(A ) is called the image of the set A under the m apping/ and denotes the
set of all elements in Y which are images of elements in A. If B C Y, the
set of all x £ X for which f(x) £ B is called the inverse image of the set
B under the mapping / and is denoted b y / -1 (B). If B, in particular, con­
sists of a single element yo £ Y, f~*(B) is denoted by / — 1(?/o) and is called
the inverse image of the element y0 under the m apping/. If the image of
A" under the mapping / is all of Y, f is said to be a mapping onto Y. The
mapping / of the set X onto the set Y is said to be (1-1) if the inverse image
of every element of Y under / consists of a single element of X .
Given mappings f 2 of a set X i into a set X 2 and / 23 of X 2 into a set X 3,
assign to each xi 6 Xi the elem ent/V /V ^i)] € X 3. The result is a mapping
f\ = f\f\
of Xi into X 3 . Accordingly, for every A3 C X 3,

( j \ T l(Az) = ( f \ r 1[ ( f \ r 1(A3)].
R em ark. Instead of f(x), f(A ), f~ l(y), r \ B ) , we will often write fx,
fA , r V , r lB.
§1.3. Indexed sets; systems of sets; order of a system of sets; coverings.
Let / be a mapping of a set A into a set M [f need not be (1-1)] and denote
a pair a £ A, m £ M , w here/(a) = m, by ma . It is sometimes convenient
to call the set of such pairs “elements of M indexed by the set (a) = A ”
or simply to say th at M is indexed by A . The elements of M may, of course,
themselves be sets. In that case, we have a family or system of sets in­
dexed by A , or an indexed system of sets.
For instance, let fix, 6) be a function of two variables, x and 6, defined
for all pairs of values x, 6, 0 < x < 1, 0 < 6 < I. Then for each fixed
value of 6, the function
fe(x) = f(x, 6)
4 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

is a function of one variable x, and the functions fe,(x) and fet (x) may coin­
cide for two distinct values 0i and 02 of 0. The set of functions fe(x) is said
to be indexed by 0, 0 < 0 < 1.
Usually, instead of “a set indexed by A ”, we shall speak of a family
or system of elements (which may themselves be sets) depending on the index
or parameter a; for instance, we say that the family of functions f»(x) de­
pends on the parameter 0.
Another important example is that of a system of subsets of a set R.
Let us assign to each element i of a set of indices a subset A, of the set R,
where it is understood that distinct indices may correspond to identical
subsets A i , i.e., subsets consisting of the same elemepts. The resulting
indexed system of subsets A ,• of the set R is briefly referred to as a system of
subsets of R,
A system of subsets of a set R is called a covering of R if the union of the
sets of this system is all of R.
The systems and, in particular, the coverings considered in this book
will consist almost exclusively of a finite number of sets. Accordingly, a
basic notion will be the order of a covering. The order of a finite system of
sets is the greatest integer n for which the system has n elements with
nonempty intersection.
R e m a r k . A system of sets is said to be simple if every two elements of
the system are distinct, i.e., if to every two distinct indices i and j cor­
respond distinct sets A* and A y. We could, of course, consider only simple
systems, but it is sometimes convenient not to be bound by this restriction.
For examples of this see XIV.
§2. Topological spaces [2]
§2.1. Definition of topological spaces and basic related notions.
D e f in it io n 2.11. A topological space is a s e t R co m p o se d o f e le m e n ts of
a rb itra r y n a tu r e in w h ic h c e rta in s u b s e ts A C R t c a lled closed sets o f th e
to p o lo g ic a l sp a c e R, h a v e b een d efin ed so a s t o s a tis fy th e fo llo w in g c o n d i­
tio n s , ca lled th e axioms of a topological space:
lx . The intersection of any number and the union of any finite number of
closed sets is a closed set.
2x . The whole set R and the empty set are closed sets.
R e m a r k 1. The elements of the set R are called points of the topological
space.
The sets complementary to the closed sets of R, i.e., the sets of the form
R \ A , where A is closed, are called open sets of the topological space R.
They clearly satisfy the following conditions:
1r • The union of any number and the intersection of any finite number of
open sets is an open set.
2r . The whole set R and the empty set are open.
§2] TOPOLOGICAL SPACES 5

R em ark 2. A topological space could be defined as a set R in which

certain subsets, called open sets, have been singled out and which satisfy
li^ 2 r • Then the sets complementary to the open sets satisfy l^-2x
and are said to be closed.
D e f in it io n 2.12. The intersection of all closed sets c o n ta in in g a set M
is called the closure of M in the topological space R and is denoted by M .
The points of M will be referred to as the contact points of M.
2.13. The closure operation, i.e., the passage from a set M to its closure
M , satisfies the following conditions:
1. M u N = M u N .
2. M C M .
3. M = M for arbitrary M .
4. The closure of the empty set is empty.
R e m a r k 3. A topological space could be defined as a set in which every
subset M has been assigned a closure M satisfying 1 — 4. Then closed
sets are defined as sets which coincide with their closures. This leads to
precisely the topological spaces which were defined at the beginning of this
section.
D e f in it io n 2.14. A neighborhood of a point p in a topological space R is
any open set containing p.
A point p £ M is called an interior point of the set M relative to the
topological space R if p has a neighborhood contained in M . The set of all
interior points of a set M , the interior of M , is an open set T C M . The
set M \ r is called the boundary of M in R. A point of M \ r is a bound­
ary point of M . A point of R is said to be isolated in A if the set consisting
of this one point is open.
2.141. p 6 M if, and only if, every neighborhood of the point p contains
at least one point of the set M.
A point p is said to be a point of accumulation or a limit point of a set
M if every neighborhood of p contains an infinite set of points of M .
D e f in it io n 2.15. A set M is said to be dense in R if M = R.
R e m a r k 4. A topological space is a composite of two concepts: the set
of points of the topological space and the topology defined in this set, i.e., a
system of sets exhibited as closed (open) sets of the topological space
[or defined in the space by the closure operation (see Remark 3)].
2.16. Let R be a topological space and let M C R. The topology of R
induces a topology in M (the relative topology) in the following way: the
sets closed in M are, by definition, the intersections with M of the closed sets
of R.
Whenever we refer to an arbitrary subset M of a topological space R
as a topological space, we shall always have the relative topology in mind.
This topology obviously satisfies I a~2a •
It is easy to prove
6 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

2.17. The sets open in M are the intersections with M of the open sets
of R.

§2.2. Neighborhood topology.

Let R be any set and call its elements points. With each
T h e o r e m 2 .2 1 .
point p 6 R associate certain subsets V(p) of R which contain p and are
called neighborhoods of p in the given neighborhood system 35. We shall as­
sume that the given neighborhood system satisfies the following conditions:
l v. The intersection of any two neighborhoods of a point p 6 R contains a
neighborhood of p.
2y. I f q is any point contained in a neighborhood V(pf of p 6 R, there
exists a neighborhood V(q) of q contained in V(p).
Let us call a point p a contact point of a set M if every neighborhood of p
contains at least one point of the set M , and let us define the closure M of M
to be the set of all contact points of M. The closure operation defined in this
way satisfies 1-4 of 2.13 and consequently converts the set R into a topo­
logical space in which all the neighborhoods V(p) are open sets of R (condi­
tion 2y).
The collection of all neighborhoods of points of a topological space R
defined in 2.14, i.e., the collection of all open sets of R, is conveniently
referred to as the absolute system of neighborhoods of R] this system satis­
fies 1 y 2y. The topology of the given topological space can therefore be
defined by applying Theorem 2.21 to the absolute system of neighborhoods
of the space, i.e., the neighborhood topology induced by using the system
of open sets of R as the defining neighborhood system of 2.21 is the same
as the original open set, closed set, or closure topology in R.
2.22. A collection 33 of open sets T of a topological space R is called a
basis for the space R if every open set of R is a union of sets of 33.
It is clear th a t:
2.23. In order that a collection 33 of open sets of a topological space R
be a basis for R, it is necessary and sufficient that for every neighborhood
V of an arbitrary point p (i.e., for every open set containing p) there exist
an element r of 33 such that
p 6 r C V.
It follows from 2.141 that:
2.24. Let 33 be a basis for R. Then p is a contact point of a set M if, and
only if, every neighborhood of p, which is at the same time an element of 33,
contains at least one point of M .
D e f in it io n 2.25. Let R be a topological space. Every system of neigh­
borhoods satisfying the conditions of Theorem 2.21 and inducing, in
accordance with Theorem 2.21, the very same topology which was origi­
nally given in R is called a system of neighborhoods of the topological space R.
§2] TOPOLOGICAL SPACES 7

I t follows from 2.24 that the elements of a system of neighborhoods of R

constitute a basis in R ; conversely, if each element F of a basis 33 is taken
to be a neighborhood of every point x £ r , the basis forms a system of
neighborhoods of the space R.
E x a m p le . The real line with its usual topology (as in analysis) has,
among others, the following neighborhood systems: 1. A neighborhood of a
point p is any open interval containing p. 2. A neighborhood of a point p
is any open interval with rational endpoints containing p. 3. A neighbor­
hood of p is any interval of the form (p — r, p + r), where r is any positive
rational number.
D e f in it io n 2.251. A topological space is said to have &countable basis
if it has a basis consisting of a countable set of elements. Such a space is
also said to be a space of “countable w e i g h t In general, the least cardinal
number r such that a topological space has a basis of power r is called
the weight of the space.
E x a m p l e . The set of all open intervals with rational endpoints is a
countable basis for the real line.
2.26. Let 33 be a basis for the space R and let M <Z R. To obtain a basis
for M , it suffices to take the sets which are the intersections of M with the
elements of 33. Hence the weight of M does not exceed the weight of R.

§2.3. Metric and metrizable spaces.

D e f in it io n 2.3. A set of arbitrary elements (points of the space) in which
every two elements x and y are assigned a nonnegative number p(x, y), the
distance between x and y, is called a metric space if the following conditions
are satisfied-.
1. p{x, y) — 0 if, and only if, x and y are identical (axiom of identity).
2. p(x, y) = p(y, x) (axiom of symmetry).
3. I f x, y, z are any three points of the space,
p(x, y) + p(V, z) > p(x, z)
{triangle axiom).
A nonnegative function p(x, y) of two variable points x, y of the space
R satisfying these conditions is called a metric of the metric space.
R e m a r k . Every subset of a metric space R is itself a metric space with
metric the same as that in R.
A metric of a metric space defines a topology in the space in the follow­
ing way. Let us call the greatest lower bound of the nonnegative numbers
p(p, x), x 6 M , the distance between the point p and the set M.
Define the closure of a set M in the metric space R to be the set of all
points p for which p(p, M ) = 0. Finally, in accord with Remark 3 of 2.1,
call the set M closed if it coincides with its own closure [or, M is closed if it
contains every point p such that p(p, M) = 0]. It is easily verified that
8 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

this topology satisfies the axioms of a topological space. We shall say

that the topology just defined in a metric space is the topology induced by
the metric or the natural topology of the metric space.
2.31. Let 72 be a metric space. The least upper bound d < °o of the
numbers p(x, y), x, y £ R, is called the diameter of R.
2.32. Let e > .0. We shall call the set S(P, e) of all points x £ R such
that p(P, x) < e an e-neighborhood of the set (or point) P of the metric
space R.
R e m a r k . An e-neighborhood is also called a spherical neighborhood (of
radius e). I t is easily seen that spherical neighborhoods are open sets.
2.33. Let M be a subset of a metric space 72; then p if, and only if,
every spherical neighborhood S(p, e) of p contains at least one point of M.
2.331. The spherical neighborhoods of the points of a metric space R
form a neighborhood system in R (in the sense of Def. 2.25).
2.34. A sequence
(2.34) x i , x2, ,xn,
of points of a metric space R converges, by definition, to a point x £ R if
lim„_oo p(x, x n) = 0, i.e., if every (spherical) neighborhood of x contains
all but a finite number of the points of the sequence.
2.35. A point p of a metric space 72 is a contact point of a set M if, and
only if, there is a sequence of points of M which converges to p. A point
p is an accumulation point of M if, and only if, M contains a sequence of
distinct points of M converging to p.
2.36. A metric space 72 has a countable basis if, and only if, it contains a
countable set dense in 72.
Indeed, if U\ , • • • , Un , • • • is a basis for the metric space 72 and x n 6
Un , then the set of all x„ is dense in 72. Conversely, if D = {a:„} is dense
in 72, then the set of all S(xn , r), where x n £ D and r is any positive ra­
tional number, is a basis for 72.
I t follows from 2.36 and 2.26 that:
2.37. If a metric space 72 contains a denumerable set dense in 72 and
A d 72 is infinite, then A also contains a denumerable dense set.
E x a m p le s o f M e t r i c S p a c e s . 1. The Euclidean n-space R n is the space
of all sequences of n real numbers x = (xx, • • • , xr), with the distance
between two points x = {xx, • • • , x n), y = G/i, • • • , y„) given by
pOl y) = [fri — yi)2 + • • • + (xn — yn)2]1
2. Every subset of a Euclidean space is also a metric space. Among
these subsets we note particularly:
the closed solid n-sphere E n with center a and radius p which consists
of all points x £ 72" at a distance <p from the point a;
§2] TOPOLOGICAL SPACES 9

the open solid n-sphere E n with center a and radius p consisting of all
points x £ R n at a distance <p from the point a;
the (n — I)-sphere 5 n_1 = E n \ E n with center a and radius p consisting
of all points x £ R n whose distance from a is equal to p.
3. The Hilbert space R00 has as points the set of all infinite sequences of
real numbers
x = ( x i, •• • , xn , •• •)
such th at %n < 00 ; the distance between two points x — ( x i, • • • ,
x„ , • • •) and y = ( y i, • • • , yn , • • •) is given by the formula

p(z, y) = E n - i (x n - ynf ] \
This is an immediate generalization of the distance formula in Euclidean
space [3].
4. The set of all points x = {xY, • • • , x n , • • ■) of. Hilbert spacesuch
th at | x n | < (§)", n = 1, 2, • • • , is called the Hilbert parallelotope (in analogy
with the ordinary parallelotope whose width is half its length and whose
thickness is half its width).
R e m a r k . (Urysohn’s theorem.) Every metric space which contains a
countable dense subset is homeomorphic (see 2.45) to a subset of the Hilbert
parallelotope [4].

§2.4. A continuous mapping of a topological space X into a topological

space Y may be defined in each of the following equivalent ways:
2.411. The mapping C is continuous if the inverse image CT1 (B ) of every
set B closed in Y is closed in X .
2.412 . The mapping C is continuous if the inverse image of every set
open in Y is open in X .
2.413 . The mapping C is continuous if C(A) C C(A) for every subset
A C X.
2.42. (Cauchy’s definition of continuity.) A mapping C of a topological
space X into a topological space Y is continuous at a point a £ X if for
every neighborhood V(b) of the point b = C(a) 6 Y there exists a neigh­
borhood V{a) of the point a such th at C'[T(a)] C V(b).
2.43. A mapping C is continuous in the sense of Defs. 2.41i-2.413 if, and
only if, it is continuous a t every point of the space X in the sense of Def.
2.42.
Moreover, in the case of metric spaces, we have
2.44. A mapping C of a metric space X into a metric space Y is continu­
ous if, and only if, the convergence of an arbitrary sequence
X i , **• , , • • • in X
10 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

to a point x 6 X always implies the convergence of the sequence

C(x0, C(x2), • • • , C(xn), • • • in F
to the point C(x) £ F.
2.45. Let C be a continuous (1-1) mapping of a topological space X onto
a topological space Y. If the mapping C-1 of F onto X, the inverse of C,
is continuous, the mapping C is said to be bicontinuom or topological;
topological mappings are also known as homeomorphisms. Two topological
spaces are homeomorpkic if one of them can be mapped topologically onto
the other.
A topological space which is homeomorphic to a metric space is called a
metrizable space.

§2.5. Uniform convergence of mappings.

D e f in it io n 2.51. A sequence

Ci , C2 , • • • , C„ ,
of mappings of a set X into a metric space 'Y is uniformly convergent to
the mapping C of X into Y if for every e > 0 there exists a natural number
n(t) such that
P[C(X), Cn(x)] < t
for n > n(t) and all x £ X . The proof of the following theorem is the same
as that given in books on analysis:
2.52. The limit of a uniformly convergent sequence of continuous map­
pings of a topological space X into a metric space F is a continuous map­
ping of X into F.

§2.6. Topological product of spaces. Let us apply 2.2 to define the

topological product of topological spaces.
The product of two sets is, after Cantor, the collection of all pairs (x, y),
x £ X , y £ F. The product of the sets X a of a system X = {A^a} is the
set of all systems of the form

£ = {*<»}
containing a single element x a of each set A”,, . The points of the topo­
logical product of the topological spaces X a are, by definition, the elements
of the product of the sets X a . The points x a £ Ara are referred to as the
“coordinates” of the point £ = {a:a}. We shall define a topology first for
the product of two spaces X and F : a neighborhood of a point £ = (x, y)
is the product of arbitrary neighborhoods of the points x and y in X and
F, respectively. It is not difficult to prove that this topology satisfies con­
ditions 1 y—2y of Theorem 2.21.
§2] TOPOLOGICAL SPACES 11

A generalization of this topology to an arbitrary number of spaces X a

was first found by A. N. Tihonov: to obtain a neighborhood of a point
£ = {x first choose any finite number of coordinates x°a of this point,
say x0ai. (1 < i < s), and then choose neighborhoods T(x°ai) C X a{ of
each of these coordinates. A neighborhood of the point £° consists, by defi­
nition, of all £ = (x„) such that x a{ £ V(xa{) (1 < i < s), with the re­
maining coordinates assuming all possible values. We may again show
th at the conditions of Theorem 2.21 are satisfied.
E x a m p l e s . 1. The plane is the topological product of two lines, the
torus [the surface obtained by rotating a circumference about an axis
lying in the plane of the circumference and not intersecting it (Fig. 1)] is
the topological product of two 1-spheres, three-dimensional space is the
product of three lines. In general, the Euclidean n-space (with the topol­
ogy defined by its usual metric) is the topological product of n lines.
The topological product of n 1-spheres is known as an n-dimensional
torus.

2. Let oxyz be a system of coordinates in three-dimensional space.

In the plane 2 = 0 consider the unit circumference iS1 = x + if = 1 and
a mechanism M i consisting of two rods oa and ac, each of unit length and
fastened at the point a by a spherical hinge in such a way th at the rod
oa can rotate freely in the plane 2 = 0 around the fixed point o, while the
rod ac is free to rotate in the three-dimensional space around the point a.
Let (oa, ac) be a definite position of the system of two rods. As a neigh­
borhood of this position it is natural to take the set of all positions (oa',
a'c') for which the distances p(a, a'), p(c, c') are less than a given e > 0.
This converts the set of all positions of the mechanism into a topological
space P.
We shall prove that the space P is homeomorphic to the topological
product of the 1-sphere S 1 and a 2-sphere S2. Indeed, a definite position
of the system of two rods oa and ac determines the point a of the 1-sphere
S 1 = x + y2 = 1 and the point of the 2-sphere S2 = x + y -f z = 1
in which this sphere intersects the ray emanating from the point o in the
direction of the vector ac. Conversely, every pair of points, p £ S 1, p' 6 S2
determines a position op of the first rod and also a position of the second
12 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

rod, obtained by laying off from the point p a vector equal to the vector
op'. This defines a (1-1) correspondence between the space P and the
topological product of a 1-sphere and a 2-sphere. It is easy to see that this
correspondence is bicontinuous. We can say, briefly, th a t the manifold
(see 5.3) of all positions of the mechanism M i is the topological product of the
1-sphere S 1 and the 2-sphere S2.
3. Let us now consider a mechanism M 2 consisting of a rod oa of unit
length, which can rotate freely in space around the point o (the origin of
coordinates), and a rod ac (also of unit length) fastened to the rod oa at
the point a in such a way that it can rotate freely in a plane passing through
the point a and perpendicular to the line oa, i.e., in the plane tangent to
the sphere S 2 at the point a.
The set of positions of this mechanism may again be converted into a
topological space Q in the following natural way: as a neighborhood of a
position (oa, ac) take all positions (oa', a'c') for which the distances p(a', a)
and p(c', c) are less than a given e > 0. The reasoning which established
the topological equivalence of the space P (Example 2) with the topo­
logical product of a 1-sphere and a 2-sphere is not applicable to this case
(the reader should verify this immediately). The space Q is not homeomor-
phic to the topological product of a 1-sphere and a 2-sphere.
We shall not prove this assertion in full, but will show th at Q, in any
case, cannot be decomposed into the product of a 1-sphere and a 2-sphere
in so natural a way as in the case of P. To each point of the space Q, i.e.,
to each position (oa, ac) of the mechanism, there corresponds a definite
point of the sphere S2. We shall prove that there does not exist a homeo-
morphism between Q and the topological product 5 1 X S2 of S l and S2
satisfying the following condition: to each point q — (oa, ac) of Q there cor­
responds a point (p, a) 6 S l X S2, where p is any point of S l and the
point a is determined by the condition q = (oa, ac).
Let us suppose that such a homeomorphism exists and consider the set
$ of all points (p0, p') £ S 1 X S2, where pQis a fixed point of S 1 and p'
takes all values on S2. Our assumptions imply that a definite vector V(p')
of unit length emanating from the point p' of S2 and lying in the plane
tangent to this sphere at the point p' corresponds to each (po, p') £ $ Q
S 1 X S2. Hence the vector V(p') is a continuous function of p'. We have
therefore obtained on the sphere S2 a continuous field of tangent vectors
which vanish nowhere. We shall prove in Chapter XVI, however, th at
such continuous vector fields do not exist on the 2-sphere (XVI, 5.510).
4. Consider a mechanical system consisting of a point moving on a cir­
cumference with arbitrary speed. Each state of this system is determined
by two data: the position of the point on the circumference and its speed.
Topologizing this collection of states in a natural way (proximity of posi-
§3) CONNECTEDNESS 13

tions and proximity of speeds), we obtain as the manifold of states (“phase

space”) the product of a 1-sphere and a line, i.e., an infinite cylinder.
5. Let us call a space consisting of two isolated points a doublet. The
product of a denumerable number of doublets is homeomorphic to the
Cantor perfect set. If r is any cardinal number, we shall denote by Dr
(dyadic discontinuum) the very remarkable space which is the product of
r doublets.
The product of r spaces each of which is homeomorphic to a closed seg­
ment of the real line is also very important ;*it is denoted by RT. The space
Rt<o, i.e., the topological product of a denumerable number of closed seg­
ments, is homeomorphic to the Hilbert parallelotope (see 2.3, Example 4).
The reader should prove this assertion.
6. Let R be a space consisting of two points a, b with the following
topology: there are exactly three closed sets in R, the empty set, the set
consisting of the point a, and the set consisting of both points a, b. We
shall call this space a connected doublet (see 3.1 for a definition of connected­
ness).
Let r be an arbitrary cardinal number and denote by FT the topological
product of r connected doublets. The properties of the space Fr are in­
teresting even for finite r.

§3. Connectedness
§3.1. Definition and basic theorems.
D e f in it io n 3.1. A topological space is said to be connected if it is not the
union of two disjoint nonempty closed sets.
Since every subset of a topological space is itself a topological space
(see 2.16), this definition of connectedness applies also to subsets of topo­
logical spaces.
E x a m p l e 1. The real line is connected, as are open and closed segments
[5].
We shall prove several theorems on connectedness.
F u n d a m e n t a l L em m a 3.11. I f A and B are two disjoint closed subsets of a
topological space R and a set M C A u B is connected, then either M C A
or M C B.
Indeed, in the contrary case, M = (M n A ) u (M n B), where M n A
and M n B are disjoint nonempty sets closed in M .
3.12. I f every two points of a space R are contained in a connected set
M C R, then R is connected.
In fact, let
R = Ai u A 2,
14 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

where Ai and A 2 are disjoint nonempty closed sets. Let pi € A i , p2 G A 2,

and let M be a connected subset of R containing pi and p2.
By virtue of 3 .1 1 , the set M is contained in one of the sets A i , A 2 .
Then both points pi , p2 are contained in one of the sets A i , A 2, a contra­
diction.
R e m a r k 1. The connectedness of a segment and 3 .1 2 imply that every
convex set is connected. In particular, the space R n is connected.
3 .1 3 . The union of two connected subsets A and B of a topological space R
such that A n B ^ 0 is connected.
For, let A u B = C.
If
C = Ci u C2,
where C i, C2 are disjoint sets closed in C, then, by 3.11, each of the sets
A, B is contained in one of the two sets C i, C2 , say A C C i. Then A n B C
C i, whence B C C i, i.e., C2 is empty, q.e.d.
D e f in it io n 3.14. A finite sequence of sets

(3.14) A i , • • • , A.
is called a chain of sets (more precisely: a chain connecting A i to A ,) if
A j n Aj+i 9^ 0 (1 < j < s — 1).
R e m a r k 2. The chain (3.14) is said to be closed if, in addition, 4 , n i i ^ 0.
A system (of any power) of sets is said to be chained if any two sets of
the system can be connected by a chain made up of elements of this sys­
tem.
It follows from 3.13 that:
3.14 The union of a chain of connected sets of a topological space R is
connected.
Furthermore, 3.14 and 3.12 imply
3.15. Let A a be a chained system of connected sets of a topological space R.
Then the union of the sets A a is connected.
In particular, we have
3.16. The union of any number of connected sets of a topological space R,
each pair of which has a nonempty intersection, is connected.
D e f in it io n 3.17. An open connected set of a topological space R is
called a domain of R.
3.18. Let © be a system of nonempty domains Fa in a topological space
R. The union I1 of the domains r a £ © is connected only i f © is a chained
system.
Proof. Suppose that the system © is not chained. Then there exist two
domains D , F2 £ © which cannot be connected by any chain of elements
of the system ©. Denote by F' the union of all the elements of the system
 §3] CONN ECTEDNESS 15

® which can be connected to Ti by chains of ele­

ments of © and denote by T" the union of all the
remaining elements of the system ©. The sets T'
and T" are disjoint nonempty open sets and their
union is r . I t follows th at T is not connected.
This proves 3.18.
3.19. I f A C R is connected and A C B C A,
then B is also connected.
Proof. Let
B = B\ u B 2,
where Bi and B 2 are disjoint and closed in B. We
shall show th at one of the sets B h B2 is empty.
Since A is connected, A is contained in one of the
sets B i , B 2, say B i . But Bi is closed in B. Hence
the set of all contact points of A contained in B,
i.e., A n B = B, is contained in B i . Therefore, B2
is empty, which was to be proved.
E x a m p l e 2 . Let the set A consist of all the
points of- the curve y = sin 1/x, 0 < x < 1
(Fig. 2). Since this curve is homeomorphic to the
real line, A is connected. If any set of points lying
on the segment —1 < y < 1 of the axis of ordi­
nates (e.g., the two points y = — y = is added
to A , the result is a connected set B.
§3.2. Components. Let p be a point of a topological space R. Since
a set consisting of one point is connected, there is a connected set in R
containing p. The union of all connected sets containing p is connected
(see 3.16) and is the largest connected set in R containing p; it is called the
component of the point p in R. It follows from 3.19, that the component' of
each point p of a topological space R is a closed set. 3.13 implies that the
components of two points p and p' are either identical or disjoint.
Therefore,
3.21. Every topological space can be partitioned into the components of its
points.
E x a m p l e 3. Let P be the Cantor perfect set on the segment 0 < x < 1
of the axis of abscissas, and let Q consist of all points (x , y) of the plane
such th at x £ P, 0 < y < 1. The sets Qx , each of which consists of the
points (x , y) 6 Q with x fixed and 0 < y < 1, are components of the set Q.
The set of these components has the power of the continuum.
E x e r c i s e . Prove that a continuous image of a connected topological
space is a connected topological space.
16 PROPERTIES OP TOPOLOGICAL SPACES [CH. I

§4. Separation axioms. Compactness [6 ]

§4.1. Separation axioms. A set M C R is said to be degenerate if it con­
sists of just one point.
4.1. Every open set containing a set M is called a neighborhood of the
set M .
4.11 A topological space R is said to be a T0-space if every two distinct
degenerate subsets of R have distinct closures in R (see the examples at the
end of this article).
I t is easy to see th at this definition is equivalent to the following:
4.11'. A space R is called a TVspace if, given any two distinct points of
R, at least one of the points has a neighborhood not containing the other
point.
4.12. A topological space is called a Ti-space if all its degenerate sets are
closed.
An equivalent definition is
4.12'. A space R is a TVspace if, given any two distinct points of R,
each of the points has a neighborhood not containing the other point.
4.13. A topological space is called a T 2-space, or a Hausdorff space, if
every two distinct points have disjoint neighborhoods.
T h e o r e m 4.131. Let R be a TV, a 7V, or a TVspace, respectively. Let
21 be a system of neighborhoods of R. Then the neighborhoods mentioned
in 4.11', 4.12', and 4.13, respectively, can be chosen from among the
neighborhoods of the system 21.
4.14. A Hausdorff space is said to be normal if any two of its disjoint
closed sets have disjoint neighborhoods.
This definition may also be given the following form:
4 .1 4 1 . A Hausdorff space R is normal if, for every closed set A C R
and neighborhood V(A) of A, there exists a neighborhood Vi{A) of A
such th at Vi(A) C V(A).
4.15. Every metric space is normal [7].
4.16 (Urysohn). A space with a countable basis is metrizable if, and only
if, it is normal [8 ].
The proof of Theorem 4.16 depends on Theorem 4.21.
E x a m p l e s . 1. The space R consists of two points. The only sets closed
in R are the empty set and R. R is not a TVspace.
2. A connected doublet (2.6, Example 6) is a TVspace, but not a TV
space. The space Fr (2.6, Example 6) is also a TVspace, but not a TVspace,
for every r. The space FT has the following universal property: it contains
the topological image of every T0-space of weight < r (see Def. 2.251 and
[9]). Conversely, every subset of the space FT is a 7Vspace of weight < r.
Finite TVspaces are the most important special cases of so called discrete
§4] SEPARATION AXIOMS. COMPACTNESS 17

spaces (see §6). The only discrete spaces which are interesting are those
which are not 7Vspaces (since all finite 7\-spaces consist of isolated points).
3. Let us adjoin to the segment 0 < x < 1 with its usual topology a new
point £. As a neighborhood of the point £ take any set consisting of the
point £ and all except an arbitrary finite number of points of the segment
0 < x < 1. The resulting topological space is a 7Yspace, but not a 7V
space.
4. The spaces DT and RT , r > , are examples of nonmetric normal
spaces. Every normal space of weight < r is homeomorphic to a subset
of the space R r (Theorem of A. N. Tihonov [10]). All the subspaces of the
space R r are 7Vspaces of weight < r , but not all of them are normal.

§4.2. Theorems on continuous functions in normal spaces [11].

4.21 (Urysohn). In order that a Hausdorff space R be normal, it is neces­
sary and sufficient that, for every two disjoint nonempty closed subsets A 0
and Ai of R, there exist a real f unction f(x) continuous on all of R, such that
0 < f(x) < 1 for all x f R and such that f(x) = 0 for x £ A 0 and f(x) = 1
for x 6 A i [12].
4.22 (Brouwer-Urysohn). Every continuous function fix) defined on a
closed set A of a normal space R can be extended to all of R, i.e., a real func­
tion Fix) can be constructed which is continuous on all of R and coincides
w ithf(x) on A [13].
C o r o l l a r y . Since a continuous mapping of a space R into Euclidean
n-space R n defines a system of n real functions = f f x ) , x d R (where
X\ , • • • , xn are the coordinates in R n), 4.22 implies
4.23. Every continuous mapping / of a closed set A of a normal space R
into R n can be extended to all of R.

§4.3. Compactness [14].

D e f in it io n 4.31. A topological space R is said to be compact if every
open covering of R contains a finite open covering of R.
A compact Hausdorff space is called a bicompactum. A metrizable com­
pact space is called a compactum. A compactum is obviously a special
case of a bicompactum.
4.32. A metrizable space is a compactum if, and only if, every sequence of
points of the space contains a convergent subsequence [15].
4.33. Among the subsets of Euclidean spaces the compacta are charac­
terized as those which are closed and bounded [16].
4.34. Every closed subset of a bicompactum (compactum) is itself a bicom­
pactum (compactum) [17].
4.35. I f a bicompactum <t> is a subset of a Hausdorff space R, then 4> is o
closed subset of R [18].
18 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

§4.4. Further theorems on bicompacta. Metrization and imbedding

theorems.
4.41. Every bicompadum is a normal space [19].
4.42 (Urysohn). In order that a bicompactum be a compacium, i.e., be
metrizable, it is necessary and sufficient that it have a countable basis [20].
4.43 (Tihonov). The topological product of an arbitrary finite or infinite
number of bicompacta is a bicompactum [21].
4.431. The topological product of a countable number of compacta is a
compadum [22].
4.44 (Aleksandrov). Among the Hausdorff spaces the compacta are charac­
terized as those which are continuous images of the Cantor perfect set [23].
4.45 (Aleksandrov). Among the Hausdorff spaces the bicompacta of weight
< r are characterized as those which are continuous images of closed subsets
of the space Dr [24].
R e m a r k . This result cannot be improved: for every noncountable t
there is a bicompactum of weight r which is not a continuous image of the
whole space Dr (an example of such a bicompactum is the set of all or­
dinals <coT, where a neighborhood of a given ordinal is any interval of
ordinals containing it and a>T is the least ordinal of power r).

§4.5. Continuous mappings of bicompacta.

4.51. Every Hausdorff space which is a continuous image of a bicompactum
is a bicompactum [25].
4.511. Every metric space which is a continuous image of a bicompactum
(in particular, of a compacium) is a compadum [26].
R e m a r k . It is somewhat harder to prove that every Hausdorff space
which is a continuous image of a compactum is a compactum [27].
Consequently,
4.52. A (1-1) continuous mapping of a bicompactum X onto a Hausdorff
space Y is a topological mapping.
4.53. Every continuous mapping of a compactum X onto a compactum Y
is uniformly continuous [28], i.e., for every e > 0 there exists a $ > 0 such
that p[C(x), C(x')\ < e for x £ X , x' £ X , and p(x, x') < 5.

§5. Upper semi-continuous decompositions of compacta and their relation

to continuous mappings (identifications). Locally compact spaces.
Topological manifolds. Examples
§5.1. Upper semi-continuous decompositions. The space of a given
decomposition [29], Every continuous mapping C of a compactum X
onto a compactum Y induces a decomposition © of A into mutually dis­
joint closed sets, the inverse images C~1 (y) of the points y £ Y.
This decomposition has the following property:
§5] UPPER SEMI-CONTINUOUS DECOMPOSITIONS OF COMPACTS 19

5.11. I f A q is an element of the decomposition <5 and r is an arbitrary

neighborhood of the set Ao in the space X , there exists a neighborhood Ti C T
of the set Ao in X such that every set A £ S which intersects Ti is contained
in T.
Indeed, let A 0 = C~1(yo) and suppose that condition 5.11 is not satisfied
(we shall regard the compactum X as defined by a metric). Then there is a
sequence of sets
an= cr\Vn) e <s
with the following properties:
1. There is a point x 'n £ A„ whose distance from A 0 < 1/n.
2. There is a point x "n £ A n whose distance from Ao is greater than
some fixed e > 0. Without loss of generality we may suppose that both
sequences {x'n\ and {x"n} converge:

lim x'n = x' £ A 0,

lim x "n = x" £ X \ Ao,
where x" £ X \ A 0 implies that C(x") 9* yo. Since the mapping C is con­
tinuous, yn = C(x'v) converges to C(x') = C(Af) = yo. On the other hand,
since x "n £ A n , C(x"„) = yn and, by the continuity of C,
lim yn = lim C(x"n) = C(x") 9^ y0.
This contradiction proves the assertion.
D e f in it io n 5.12. A decomposition of a compactum X into mutually dis­
joint closed sets A is said to be upper semi-continuous if it has property 5.11.
5.13. Every decomposition S of a compactum X into mutually disjoint
closed sets A induces a topological space F, the decomposition space, in the
following way: the sets A are, by definition, the points of the space F; to
define a neighborhood of a point A 0 £ F, choose any neighborhood r of
the set A 0 in the space X and let the corresponding neighborhood Ur(-Ao)
of the point Ao in the space F consist of all ^4 G F for which A C r in X.
It is easy to verify that the resulting neighborhoods in F satisfy the con­
ditions of Theorem 2.21. Hence F is a topological space and, since X is
normal, F is a Hausdorff space.
Furthermore,
5.14. I f the decomposition S is upper semi-continuous, the assignment to
each point x £ X of the set A £ Y containing x is a continuous mapping C
of X into Y.
In fact, let x G A 0 £ F; then C(x) = A 0 . Let Uv{A0) be any neighbor­
hood of the point A 0 G F. By 5.11, the neighborhood T of the set A0 in
X which generates the neighborhood f/r(A 0) contains a neighborhood
20 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

Ti of A 0 in X with the property that every i f S which intersects Ti

is contained in r . Then Id is a neighborhood of the point i in Z and C
maps Ti into the given neighborhood Uy(A<s) of the point A q = C(x) £ Y.
This proves the continuity of the mapping (Cauchy’s criterion 2.42).
Since a Hausdorff space which is a continuous image of a compactum is
itself a compactum (see 4.5, Remark), it follows that:
5.15. The space induced by an upper semi-continuous decomposition
of a compactum X is a compactum, since it is a continuous image of the
compactum X.
R e m a r k . If a compactum Y is the space of the upper semi-continuous
decomposition 6 of a compactum X or is homeomorphic to this space,
we shall say th at the compactum Y is the result of identification of certain
points of X ; namely, all the points of a set A £ S are identified and so give
rise to a point of the space Y. If every A is a finite set, this identification
may be thought of as effected by “pasting together” the points involved.

§5.2. Examples of upper semi-continuous decompositions and identi­

fications. The projective n-space.
1. Let X be a 2-sphere with a system of geographic coordinates defined
on it. Let the elements of the upper semi-continuous decomposition be:
a) the pair of poles of the given system of coordinates, and b) the circles
of latitude of this system. The space of this decomposition is clearly homeo­
morphic to a 1-sphere.
2. Let X be a torus and consider the upper semi-continuous decomposi­
tion of the torus into its meridians \p = c (Fig. 3; the meridian = 0 is
drawn as a solid curve). The space of this decomposition is also homeomor­
phic to a 1-sphere.
3. In general, if Z is the topological product of compacta X and Y,
the sets A v consisting of all points (x , y) £ Z, where y £ Y is fixed and x
runs over the whole space X, are the elements of an upper semi-continuous

F ig. 3 Fig. 4
§5] UPPER SEMI-CONTINUOUS DECOMPOSITIONS OK COMPACTA 21

.............................
•
•
4
4
•
•
•
•
•
•
•
•
•
•
•

F ig. 5 F ig. 6

decomposition of the space Z; the space of this decomposition is homeo-

morphic to F.
4. Let us denote by 6 the upper semi-continuous decomposition of a
circle represented in Fig. 4 (the circumference together with both mutually
perpendicular diameters form one element of the decomposition). The
space of this decomposition is homeomorphic to a compactum consisting
of four closed straight line segments with one common endpoint.
5. Let X be the square 0 < x < l , 0 < ? / < l i n the plane (x, y). Con­
sider the decomposition (5 of the square X whose elements are: a) the in­
dividual interior points of the square; b) the pairs of points {(x, 0), (x, 1)},
0 < x < 1; c) the pairs of points {(0, y), (1, y )), 0 < y < 1; d) the quad­
ruple of points {(0, 0), (0, 1), (1, 0), (1, 1)}. The space of this decomposition
is homeomorphic to a torus. In brief, the torus is the result of identifying
corresponding points on opposite sides of the square (Fig. 5).
6. In an analogous sense, identification of corresponding points of each
pair of opposite faces of a solid cube yields the three-dimensional torus
(topological product of three 1-spheres).
7. Let Q? be the closure of the domain of three-dimensional space con­
tained between two concentric spheres S 2 and s2, and identify correspond­
ing points of both spheres (i.e., the points of intersection of a ray emanating
from the center with the two spheres). The result is the topological product
of a 2-sphere and a 1-sphere. [Let the 2-sphere be one of the concentric
spheres. The segment joining a point of this sphere to the corresponding
point of the other sphere becomes a 1-sphere after identification of its
endpoints (Trans.).]
This identification can be represented in still another way (Fig. G).
Let us cut the solid Q3 with an equatorial plane into two congruent solids
Q2u and Q3i . This divides each of the spheres S2 and s2 into two hemi­
spheres: upper U2 and u , and lower L2 and I1. Denote the equatorial cir­
cumferences of the spheres S 2 and s2 by S' and s', respectively.
22 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

The solid Q3U can be thought of as a deformed solid cylinder with bases
U2 and u and lateral surface H2 = I12U, the latter in the form of a plane
ring which is hatched in the sketch.
The solid Q3t is also a cylinder with lateral surface II2 = II2j and bases
L2 and I2.
In each of the solids Q3U and ( fi corresponding points of the upper and
lower bases are to be identified; this identification converts both solids
into two three-dimensional rings (a three-dimensional ring is a bounded
solid whose boundary is a torus). Finally, it remains only to do away with
the equatorial cut, i.c., to identify corresponding points of both toruses
II2„ and n 2; .

Hence the topological product of a 2-sphere and a 1-sphcre is obtained

by “doubling” a three-dimensional ring, i.e., by identifying corresponding
points of the boundaries of two congruent three-dimensional rings.
R e m a r k . It is easily and immediately verified that the space resulting
from the doubling of a three-dimensional ring is the topological product of a
2-sphere and a 1-sphere. [For the 1-sphere take a circle of latitude of one
of the toruses. At each point of this circle the meridian circle passing
through the point is identified with the corresponding meridian of the
other torus, but the interiors of the meridian circles are not identified.
This gives a 2-sphere at each point of the 1-sphcre (Trans.).]
E x e r c i s e . Prove that a three-dimensional torus is obtained from the
domain included between two coaxial toruses (Fig. 7) by identifying cor­
responding points of the two bounding toruses.
8. Projective space. Let x0, ■■• , xn ; yQ, • • • , yn be two sequences of
n + 1 real numbers, each of which contains at least one number different
from zero. The two sequences are said to be proportional or to define the
same ratio if x{yk — XkVi (0 < i, lc < n). It is easy to see that the relation
of proportionality of numerical sequences (consisting of the same number
of elements)'is reflexive, symmetric, and transitive. Therefore, the set of
all sequences xq , • • • , xn (n fixed and at least one of the x,- different from
zero) is partitioned into disjoint classes, each class consisting of propor­
tional sequences. These classes are called ratios. Each ratio is a class of
§5] UPPER SEMI-CONTINUOUS DECOMPOSITIONS OF COMPACTA 23

proportional sequences. The ratio of the sequence x0 , • • • , xn is denoted

by (xo • x\ : ■• • : xn). The sequence whose terms are all zero has no ratio.
Given a ratio (x0 : x\ : • • • : x n), it is always possible to choose a sequence
x0, • • • , xn in the ratio such that x£ + • • • + x n2 = 1.
We may now define the projective n-space as follows: the points of the
projective n-space are the ratios x = (xo : • • • : xn) of all sequences of
n + 1 real numbers .t0 , • • • , xn of which at least one is different from zero.
A neighborhood of a point a = (a0 : ai : an) is defined by choosing
in the ratio (a0 : ai : • • • : a„) a sequence a0, ■■■ , a„ such that
2 i i 2 -j
CIO + • • • + < ! „ = 1

and then taking all sequences x0, • • • , xn for which x02 + • • • + x 2 = 1

and | a,- — x ,-1 < e. The points (ratios) of the projective space correspond­
ing to such sequences form, by definition, a neighborhood S(a, e) of the
point a. This definition permits the construction of several models of the
projective space. Thus, for instance, each point (x0 : xi : ■• ■ : x„) 6 P n
can be made to correspond to the line of Euclidean (n + l)-space 7?n+1
(£o ; • • • ; £„) passing through the origin of coordinates and defined by
the equations
(£o/zo) = (£i/zi) = ••• = (Zn/xn),
where the £,• are running coordinates.
It is left to the reader to prove that if the distance between two straight
lines is defined as the angle <p between them, 0 < <p < 7 t / 2 , then the set
of all these straight lines is converted into a metric space; and that the
correspondence established above between this metric space and the space
P n is topological.
We have thus constructed a geometrical model of the projective space,
in which the points are straight lines passing through the origin. Now let
R n be any n-plane in R n+l which does not contain the origin and, as the
representative of each straight line (passing through the origin), let us
take its point of intersection with R n. This yields the usual interpretation
of the projective space as a Euclidean space closed by elements at infinity.
We obtain another model by taking as the representative of each straight
line its two points of intersection with the unit sphere
£o2 + £i2 + ------b £n2 = 1
in R n+1. This corresponds to the choice in each ratio (x0 : x\ : • • • : x„) of a
sequence x0 , • • • , x n satisfying the condition xq + • • • + x 2 — 1.
Hence it is clear that the n-sphere S*, after identification of its dia­
metrically opposite points, becomes the projective n-space.
Let us divide the sphere £o2 + • • • + £n2 = 1 into two hemispheres > 0
24 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

and < 0 with the equatorial plane £„ = 0, and let us add the equator
£n = o, £02 + • • • + tn = 1 to the upper hemisphere £„ > 0, denoting
the resulting compactum £02 + • • • + £n2 = 1, £« > 0 by Qn.
Each pair of diametrically opposite points of S n not lying on the equa­
tor S ”-1 has its (unique) representative in Q", so that only diametrically
opposite points of the equator need be identified. To obtain projective n-
space it, therefore, suffices to odd to one hemisphere of the n-sphere S n its
equator S n~x and to identify diametrically opposite points of the equator.
Since the compactum Qn is homeomorphic to a closed solid n-sphere, projective
n-space is obtained from a closed solid n-sphere by identifying diametrically
opposite points of its boundary.

§6.3. Locally compact spaces. Topological manifolds. Examples. A

topological space is said to be locally compact if every point of the space
has a neighborhood whose closure is compact.
For instance, Euclidean n-space is locally compact (but, clearly, not
compact).
A connected locally compact space with a countable basis, each point
of which has a neighborhood homeomorphic to Euclidean ?i-space, is
called an n-dimensional topological manifold or, briefly, an n-manifold. A
manifold is said to be closed if it is compact; manifolds which are not
closed are called open.
E xam ples of C l o se d M a n if o l d s .
1. The n-sphere (see 2.3, Example 2).
2. The n-dimensional torus, i.e., the product of n 1-spheres; in general,
the topological product of an arbitrary finite number of closed manifolds
is a closed manifold (in particular, a topological product of n-spheres is a
closed manifold).
3. [30]. Let us consider all sextuplets of real numbers
P l2 > P n i P u , P 34 , P i 2 , P'2Z

whose elements are not all zero and which satisfy

(5-31) PviPu + PnPii + PuPn = 0.
Each such sextuplet determines a line in projective 3-space [if, e.g., pn ^ 0,
the sextuplet determines the line passing through the points (0: pn : pn : pu)
and ( —P12 : 0: p23 •’ P24)] whose Plucker coordinates are the six numbers
Pa ; and two sextuplets determine the same line if, and only if, they are
proportional to each other. [If a line passes through the points (xx : :
x-i : xf) and (y\ : y2 : y2 ■yf), its Thicker coordinates (determined to within
Xi Xj
proportionality) are p,j = ; hence the notation p l7 . Then the deter-
Vi Vi
§6] PARTIALLY ORDERED SETS 25

minant having first and third rows x 4 , x2, x3 , x4

and second and fourth rows iji , y2 , y3 , y4 is clearly
equal to zero; this is also expressed by (5.31).]
Conversely, every line determines a unique class
of proportional sextuplets. Hence the lines of pro­
jective 3-space are in ( 1- 1 ) correspondence with the
ratios of sextuplets of real numbers which satisfy
(5.31).
Neighborhoods in the space of lines are defined
by analogy with the neighborhoods already defined
in the projective space. Let a line be determined by
the ratio (p n : p 43 : p 44 : p 34 : p 42 : p 23) and let
0 0
p 12 P 13 P°i4 , p°34 , p°42, p°23 be a “normal” repre­
sentative of this ratio, i.e., a sextuplet of the ratio
for which (p°12)2 + • • ■ + (p°23? = L Now let p12, pn , p14, p3i, pi2, p23
be all sextuplets such that pn + • • • + P23 = 1 and | p \ 2 — pi2 \ < e,
■■• , | p°23 — P23 | < «• The lines (pl2 : pi3 : • • • : p23) corresponding to
these sextuplets generate an e-neighborhood of the line (pV : • • ■ : P°23).
This topology converts the set of all lines into a closed 4-manifold.
Let us now consider the projective 5-space P 5. The points of this space
are ratios (21 : x2 : x3 : z 4 : x 6 : x6). The hypersurface of second degree
defined in P 5 by the equation
XiX4 + x2xs + x3x6 = 0

is clearly homeomorphic to the 4-manifold of all lines of projective 3-space.

R e m a r k . Every domain (connected open set) of a closed n-manifold is
an example of an open n-manifold. However by no means every open n-
manifold is homeomorphic to a domain of a closed manifold. The open 2 -
manifold pictured in Fig. 8 is not homeomorphic to any subset of any
closed 2 -manifold.
G e n e r a l R e m a r k C o n c e r n in g M a n if o l d s . The following fundamental
theorem will be proved in Chapter V:
A topological space cannot be, for n 5^ m, both an n-manifold and an m-
manifold.
§6. Partially ordered sets and discrete spaces [31]
§6.1. Definitions.
D e f in it io n 6.11. A set 0 of arbitrary elements is said to be partially
ordered if an order is defined in 0 , i.e., if a rule is given by which it is pos­
sible to determine, for any pair of elements, whether one of the pair fol­
lows (or precedes) the other. The order must satisfy the following condi­
tions (order axioms):
2G PROPERTIES OF TOPOLOGICAL SPACES [CH. I

1. If &follows a, then a does not follow b.

2. If b follows a, and c follows b, then c follows a.
D e f in it io n 6.12. A partially ordered set 9 is said to be (simply) ordered
if, for every pair of distinct elements a and b of 9, either a follows b or b
follows a.
R e m a r k 1. If 6 follows a, we also say that a precedes b and write b > a
or a < b.
R e m a r k 2. The first order axiom asserts that the relation a < b excludes
the relation a > b. Hence, in particular, a < a cannot occur, since this
would imply both a < a and a > a.
D e f in it io n 6.13. A (1-1) mapping / of a partially ordered set 9i onto
a partially ordered set 92 is said to be a similarity mapping (or transforma­
tion) if / is order preserving, i.e., if
/(p) > f(p') in ©2
is equivalent to p > p' in 9 i . Two partially ordered sets 9i and 9 2 are
said to be similar if there is a similarity transformation which maps one
onto the other.

§6.2. Examples of partially ordered sets.

1. Any set of simplexes or, in general, convex polyhedral domains (see
Appendix 1, §3) is partially ordered by defining T% > T i , where T i and T2
are any two polyhedral domains, to mean that T i is a proper face of T2 .
This is called the geometric order in the set of convex polyhedral domains.
2. The set consisting of the points, straight lines, and planes of a E u­
clidean space is partially ordered by letting a point (straight line) precede
any straight line (plane) containing it.
3. The set 9 of real functions defined on a point set X is partially ordered
if /i > /2 is taken to mean th at/i(x ) > / 2(x) for all x £ X and /i(x) > / 2(x)
for at least one x £ X.
The following example is well known and fundamental:
4. The partially ordered set of subsets of an arbitrary set M . Let 9 be a set
whose elements are subsets A a of a set M,
9 = | A a), A a C M.
If A a and A$ are any two elements of 9, put A a > Ap if Ap is a proper
subset of A a , i.e., if A a Z) Ap . The resulting order is called the natural
order in the collection of sets 9; it obviously converts 9 into a partially or­
dered set.
5. If M consists of n + 1 elements, the set of all nonempty subsets of
M, with the natural order, is similar to the set of all faces of an n-simplex,
with the geometric order.
§6] PARTIALLY ORDERED SETS 27

§6.3. The sets Aep and Oep.. Let 9 be a partially ordered set and let
P € 9.
Denote by A ep the set of all x 6 9 such that x < p. The set A ep is
called the combinatorial closure of p in 9. Clearly, p £ A ep.
The set of all x G 9 such that x > p will be denoted by Oep. This set is
called the star of p in 9. Obviously, p G Oep.
E x a m p le s . 1. Let 9 be the partially ordered set of all points, straight
lines, and planes of three-dimensional space. If p is a point of 9, A Qp con­
sists of just the point p; if p is a straight line, A ep is the set consisting of
the straight line p and all points lying on this line; if p is a plane, Aep is
the plane p and all lines and points lying on this plane. In an analogous
fashion, Oep consists of the point p and all straight lines and planes passing
through p, if p is a point; of the straight line p and
all planes containing p, if p is a straight line; and of
just the plane p, if p is a plane.
2. Let 9 be the set of all triangles, sides, and ver­
tices sketched in Fig. 9. If p G 0, Oep consists of all
simplexes having p as a face. For example, for the
case shown in Fig. 9:
If p is a triangle, Oep consists of the single ele­ F ig . 9
ment p, and ^4ep is the triangle p, its sides, and
vertices; if p is a side, Oep consists of this side and of the triangles having
p as a side (in the figure there are one, two, or four such triangles, depend­
ing on the side chosen), and Aop consists of the side p and its endpoints; if
p is a vertex, Oep consists of the triangles having p as a vertex and of
the sides having p as an endpoint; while A Qp is just the vertex p.
T h e o r e m 6.3. Every partially ordered set 9 is similar to its set 21 of sub­
sets {Aep\, p G 9, partially ordered by the natural order.
The proof is left to the reader as a simple exercise.

§6.4. Duality of partially ordered sets. Let 9 be a partially ordered

set and let 9' be the partially ordered set with the same elements as those
of 0 but with the converse order relation, i.e., a < b in 9 ' if a > b in 9.
The partially ordered set 9 ' is called the dual of 9. The relation of duality is
clearly symmetric: if 9 ' is dual to 9, then 9 is dual to 9'.
The closures of elements in 9 ' are synonymous with the stars of elements
in 9 and conversely: A&p = Oep, Oe>p = Aep.
We have seen that every partially ordered set is similar to the set of
closures of its elements naturally ordered. Therefore, 9', the dual of 9,
is similar to the set of stars of the elements of 9 with the natural order.
E x a m p l e s . 1. Let 9 consist of the faces, edges, and vertices of a cube
with the geometric partial order. Its dual 0 ' is similar to the set of faces,
 28 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

edges, and vertices of an octahedron also with the geometric partial order.
2. The partially ordered set of all proper faces of an n-siinplex geo­
metrically ordered is similar to its dual. To realize the correspondence, it
suffices to associate with each face of the simplex the face opposite it.
3. Let us partition three-dimensional space into congruent cubes of
side 1 and with vertices having integral coordinates. Let © consist of all
these cubes, their faces, edges, and vertices geometrically ordered. The dual
of 0 is similar to 0. To realize a similarity transformation, it suffices to
displace the whole system of cubes along the vector with components
(h i) and to associate with each element p' of the displaced system the
element p of the old system having a common center with p'.

§6.5. Discrete spaces.

D efin it io n " G.51. A 7T
0-space is called a discrete space if the union of an
arbitrary number of closed sets of the space is closed (or, equivalently, if the
intersection of any number of open sets of the space is open).
T h e o r e m 6.52. We shall say that a subset A of a partially ordered set 0 is
closed i f p 6 A and p' < p imply p' £ A.
This topology converts 0 into a discrete space R = /(0 ). Conversely, every
discrete space R can be turned into a partially ordered set © = <p(R) if, for
any two distinct elements p, p' 6 R,

P' < P
is taken to mean that

p' € p.
It follows that

MR)) = R
and

*>[/(©)] = e.
The proof consists of a routine verification of the axioms of a discrete
space and of a partially ordered set and may be left to the reader.
In accordance with Theorem 6.52, partially ordered sets can be identified
with discrete spaces. Then the combinatorial closure of an element p of 0
is synonymous with the closure in the space/(0) of the set consisting of the
single point p, and the star of p 6 0 is the minimal neighborhood of the
point p in /(©), i.e., the intersection of all open sets of the space/(0) con­
taining p.
The transition from a partially ordered set 0 to its dual 0 / corresponds
§7 ] COMPLETE METRIC SPACES AND COMPACTA 29

to the transition from the discrete space R to its dual space R', whose
closed sets are the open sets of the space R (and conversely).
A similarity mapping of one partially ordered set onto another is, there­
fore, equivalent to a topological mapping of the corresponding discrete
spaces.
The identification of partially ordered sets with discrete spaces hence
enables us to carry over to partially ordered sets the various theorems
proved for topological spaces, e.g., all theorems on connectedness.

§7. Complete metric spaces and compacta [32]

§7.1. Definitions and simplest properties of complete metric spaces.
7.11. A sequence of points

■El j %2 j ' * * > 2Cn j

of a metric space R is called a fundamental (or Cauchy) sequence i f for every

t > 0 there is a natural number n(t) such that p > n(e), q > n(e) imply that
p(xp , x q) < e.
I t is obvious that every convergent sequence is a fundamental sequence.
7.12. A metric space is said to be complete if every one of its fundamental
sequences converges.
Of the properties of complete metric spaces we note the following two:
7.13. In a complete metric space R every decreasing sequence of closed sets

Ai 2 A2 2 ••• 2 An 3 ...

whose diameters approach zero has an intersection consisting of one point.

Indeed, since the diameters of the sets A„ approach zero, D.4n cannot
contain two points. On the other hand, if an £ A n , then {o„} is a funda­
mental sequence whose limit is contained in C\A„ .
7.14. I f Ti , r 2 , • • • is a sequence of open sets of a complete
metric space R and each I \ is dense in R, then E — ClF„ is also dense in R.
The proof of Theorem 7.14 is based on the following
L e m m a . I f an open set T of a metric space R is dense in R, then every
open set r 0 C R contains an open set F' (of arbitrarily small diameter)
whose closure is contained in To n T, i.e., F C To n T.
In fact, since F is dense in R, F0 contains a point p of T; and, since
A = (R \ F0) u (R \ T) = R \ (F n F0) is closed, p(p, A) > 0.
Let e > 0 be such that t < p(p, A), and set F' = S(p, e). Every point
of F' is at a distance < e from p and is, therefore, contained in T n To.
This proves the lemma.
We shall now prove Theorem 7.14. To do this, it suffices to show that
there is a point of E in every open set T d R.
30 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

Let r 'i be an open set of diameter < 1 such th at F'i C h n T, and sup­
pose that the sets
r 'i 3 • • • ^ r ' „ , 5 (r\) < (£)*' (1 < i < n),
satisfying the conditions
r ' l C h . - . f ' . c r„
have already been defined. Then, applying the lemma, construct a set
r ' n+i of diameter < (§ )"+1 such that F'n+i Q r'„ n r„+i .
The intersection flF',, is nonempty (it consists of one point) and is con­
tained in T n E. This proves the theorem.
Well known examples of complete metric spaces are: the real line,
Euclidean n-space, Hilbert space, etc.

§7.2. c-nets in compacta. Let R be a metric space. A finite set N C R

is called an e-net if it has the property th at every point x of R is at a dis­
tance < e from some point of AT
If, for a given e, R has no e-net, there exists in R a sequence of points
whose mutual distance from each other is > e.
Indeed, let ai be a point of R. Then ai is not an e-net, since R has no
e-net. Hence there exists a point a2 whose distance from Oi is > e. The set
consisting of ai , a2 is again not an e-net. Hence there exists a point a3
whose distance from ai and a2 is > e. Continuing in this way, we obtain a
sequence
(7.2 1 ) ci\ , a2, • ■* , on ,
of points of R whose mutual distance from each other.is >e.
R e m a r k 1. Since the sequence (7.21) clearly does not have a limit point,
we have proved the proposition:
7.21. A compdetum hds on e-net for every e > 0.
D e f in it io n 7.22. A metric space is sdid to be totally bounded if it contoins
on e-net for every e > 0 .
R e m a r k 2 . The sequence (7.21) obviously has the property th at no sub­
sequence of (7.21) is a fundamental sequence. Hence, if a metric space R
is not totally bounded, it contains a sequence which has no fundamental
subsequence.
On the other hand, if R is totally bounded and
N(e) = {di , • • • , o3}
is an (e/2)-net of R, then the sets S ( d i , t/2) cover R. Since there are only a
finite number of these sets and their diameters are < e, every infinite se­
quence
(7.22) > *** j &n >
§7 ] COMPLETE METRIC SPACES AND COMPACTA 31

of points of R contains an infinite subsequence whose diameter is less than

an arbitrary preassigned t > 0.
Let e„ —> 0 and let
(7.220 a u > ‘ ‘ ‘ , aik , • • •
be a subsequence of (7.22) of diameter <ci . Now construct a subsequence
(7.222) of (7.220 of diameter < e2, etc. The diagonal sequence of all these
sequences is a fundamental subsequence of (7.22).
Hence,
7.23. A metric space is totally bounded i f, and only if, every sequence of
points in the space contains a fundamental subsequence.
Since every compactum is clearly a complete metric space, 7.21 and 7.23
yield
7.24. In order that a metric space be a compactum, it is necessary and suf­
ficient that it be complete and totally bounded.
7.25. Every totally bounded metric space R contains a countable set dense
in R, i.e., R has a countable basis.
In fact, if Nk is an e*-net and e* —» 0 as k —> °o, then the countable set
N = UN* is dense in R.

§7.3. The space of continuous mappings. Let X be a topological space

and let 7 be a bounded (i.e., of finite diameter) metric space. Denote by
£(X, Y) the metric space defined in the following way. The points of
S(N, Y ) are the continuous mappings of X into Y. The distance between
two points Ci 6 £(A", Y) and C2 6 £(X, Y) is defined as the least upper
bound of pfCi(x), C2(x)], x £ X .
7.31. I f Y is a compactum and X is any topological space, the space
( l ( X. Y) is a complete metric space.
Proof. Let
(7.31) C i, C2 , • • • , C„ , • • •
be a fundamental sequence of points of &(X, Y). This means that for
every e > 0 there is an n(t) such that
(7.32) pfCp(x), Cq(x)} < e
for all x 6 X and arbitrary p > n(e), q > n(e). For every x 6 X the se­
quence of points
(7.33) Ci(x), C2(x), • • • , Cn(x), • • •
is, under these conditions, a fundamental sequence, and hence, in virtue of
the compactness of Y, it is a convergent sequence in Y. It then follows from
(7.32) th at the sequence of mappings {C„} converges uniformly to a con-
32 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

tinuous function C € G(X, T). The sequence (7.31) obviously converges

in the metric space (S(X, Y), to C £ G(X, Y ). This proves the theorem.

§7.4. Deformations. Homotopy classes of mappings of a compactum

X into a compactum Y. Let Co and Ci be two continuous mappings of a
compactum X into a compactum Y. Denote by n the topological product
of the compactum X and the segment 7 = a < f l < b o f the real line.
The points of the space X X I = n are the pairs (x, 0), x £ X, 0 £ / . A
continuous mapping C(x, 0) of II into Y is called a deformation of the map­
ping C0 into the mapping Ci if
C(x, a) = C0(x),' C(x, b) = C\{x)
for all x € X. Two continuous mappings Co and Ci of the compactum X
into the compactum Y are said to be homotopic if there exists a deformation
C(x, 0) of one into the other.
Since the homotopy relation is reflexive, symmetric, and transitive, the
set of all continuous mappings of a compactum X into a compactum Y is
partitioned into classes of homotopic mappings or, briefly, into homotopy
classes.
R e m a r k . It is customary to put a = 0 b = 1, to write Ce(x) instead of
C(x, 0), and to speak of a family of continuous mappings Ce(x) of a com­
pactum X into a compactum Y , depending continuously on the parameter
0, 0 < 0 < 1, or briefly, of a continuous family Ce(x) of continuous map­
pings of X into Y.
§8 . Coverings of normal spaces and, in particular, of compacta
§8.1. Closed and open coverings of topological spaces. The method of
combinatorial topology. We already know that a (finite) system of subsets
of a set M whose union is M is called a (finite) covering of M. In the sequel,
we shall consider coverings of a topological space, namely, open coverings,
consisting of open sets, and closed coverings, consisting of closed sets. In
all cases, unless otherwise noted, we shall consider only finite coverings. The
order of a covering a is the greatest integer n for which there exist n elements
of the covering a having a nonempty intersection.
D e f in it io n 8 .1 1 . A system of sets /9 is called a refinement of a system of sets
a if every element of /3 is contained in at least one element of a.
A covering /3 of a set M is said to follow a covering a of M if is a refine­
ment of a and a is not at the same time a refinement of /3. This relation
will be written as “j3 > a ”. It is easily seen that the relation “ > ” between
the coverings of a set M is a partial order and thus converts the set of cover­
ings of M into a partially ordered set. Whenever we speak of a partially
ordered set of coverings of a set M, we will always have in mind the order
defined above.
§8 ] COVERINGS OF NORMAL SPACES 33

Combinatorial topology, in the broad sense, studies the properties of topolog­

ical spaces by investigating the properties of the partially ordered sets of their
open (or closed) coverings.

§8.2. Similar coverings.

D e f i n i t i o n 8 .2 1 . Two fin ite in d e x e d s y s te m s of s e ts

a = (A i, A2, •; • , As),
fi = [By , Bs , • • • , B 6)
are said to be similar if fl^-iA^. ^ 0 is equivalent to f l ^ - i ^ 0, where
the subscript ij stands for any of the integers 1, • • • , s.
The equivalence existing in general combinatorial topology between the
use of partially ordered sets of open and closed coverings is based on the
following two theorems:
8.22. I f
a = {Ay , • • • , A,}
is a system of closed sets of a normal space R, there exists a system of open sets
w = {Oy, ■■• , 0 6}
such that Ay Q Oi (1 < i < s), and such that a is similar to both w and
CO = {Oy, ■■■ , 0 ,\.
A very simple proof of this theorem is given in 8.3 for the case of a com-
pactum R.
8.23. For every open covering
« = {Oi , • • • , 0 8}
of a normal space R there is a closed covering
a = {A i, • • • , A5)
of the space such that
AiQ O i (1 < i < s).
R e m a r k . A s will be evident from the proof, we may even suppose in
Theorem 8.23 th at the A,- are the closures of open sets forming a covering
of R.
Proof of 8.22. Let A be the union of all possible sets of the form ,
where the sets A;,. (1 < j < k) are such that Ay n A;, n • • • n A it = 0.
Since A is closed and A n Aj = 0, \ A is a neighborhood of A y .
Consequently, by 4 .1 4 1 , there exists an open set Oy such that Ay Cl Oy ,
O yQ R \ A .
The system of sets
a = [Ay, • • • , As}
34 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

is similar to the system

<xi = {Oi, A 2, • • • , A„).
Indeed, if A i n A ix n • ■■ n A ik ^ 0, then, a fortiori, Oi n A,-, n • • • n
A ik ^ 0. Conversely, Ai n A,-, n • • • n A ik = 0 implies that A,-, n • • • n
A ik C A which in turn implies that
Oi n A il n • • • n A ik C Oi n A = 0 .
Let us suppose that the open sets 0, 2 A i (1 < i < r < s) have been
constructed in such a way that a is similar to
OCT {Ol , , Or , A T-|_i , " ) A a}. -
Applying the same reasoning to the system a.r and the set A r+i which we
have just applied to the system a and the set A i , we obtain an open set
Or+i ^ A t+i such that the systems aTand
«r+l = { O i, • • • , Or + 1 , A r + 2 , • • • , A ,},
and, by the same token, the systems a and a r+ i, are similar. Continuing
in this way, we arrive at a system a, = {Oi, • • • , 0») which satisfies the
prescribed requirements.
Proof of 8.23. Corresponding to the system
af = { R \ 0 lt ■■■ , R \ 0 . },
there exists, according to the proof above, a system of open sets
a/ = {O' i , ••• ,0 '.}
such that R \ Oi C O'i and such that the system {O 'i, • ■• , O',} is
similar to the system a . Because of this similarity, it follows, in particu­
lar, that
O'i n • • • n O', = 0.
Putting O"i = R \ O' i , we obtain (since R \ O'i is closed)
0"i = R \ O'i C R \ 0 ' i = R \ O'i.
Since R \ O i C O',-, this implies that
0"i Q R \ ( R \ Oi) = O i .
Finally, since
0"i u • • • u O", = R \ ( 0 \ n • • • n O'.) = R,
the required covering is
® {A i, • • ■ , A,},
where Ai = O" ,•.
§8] COVERINGS OF NORMAL SPACES 35

§8.3. e-coverings of compacta. Lebesgue numbers of a covering.

D e f e n it io n 8.31. A covering of a metric space is called an e-covering i f
the elements of this covering are of diameter < e.
L e b e s g u e ’s L e m m a 8.32. Let

a = { ^ i , • • • , As}
he a closed covering of a compactum <f>. Then there is a positive number 8 with
the following property: if M C <f> has diameter < 8 and
(8 .3 2 1 ) A if G a (1 < j < k)
are such that M n A ij ^ 0 (1 < j < k), then fl*=i A tJ. 0.
Proof. The proof is by contradiction. If there is no 8 with the required
property, then for every natural number n there is a finite sequence of
elements (8 .3 2 1 ) of a and a set M n (Z of diameter < 1 / n such that
M„ n A ^ 7^ 0 (1 < j < k) and D^LiA*,. = 0.
Since the number of all combinations (8.321) of different elements of the
covering a. is finite, there is at least one combination (8.321) which, in the
above sense, corresponds to an infinite number of different n ’s. Hence
there is a sequence of natural numbers {ny} and a sequence of sets
j = 1, 2, • • • , of diameter <1 /n , which intersect all of the sets (8.321)
and such th at fly_i A*,. = 0. Let pnj G M n j . Passing, if necessary, to a
subsequence of {p*,.}, we may suppose that the sequence {pn,-} converges
to a point p G <f*. There are clearly points of each of the sets Aij (1 < j < k)
in every neighborhood of p, and since these sets are closed,
p G A is (1 < j < k).
Therefore, fl^Li At. ^ 0, a contradiction. This proves the lemma.
D e f in it io n 8.33. Every positive number 8 which satisfies the conditions of
Lebesgue’s lemma for a given closed e-covering a of a compactum is called a
Lebesgue number of the covering. Hence any sufficiently small positive num­
ber is a Lebesgue number of a closed covering. We shall, in addition, im­
pose one more condition on the Lebesgue numbers of an e-covering: we shall
require th at they be less than one half the difference between e and the
maximum of the diameters of the sets A,-. This definition of a Lebesgue
number immediately implies
8 .3 3 1 . I f 2 a is a Lebesgue number of the closed e-covering

a = l^ i , ■• • , As)
of the compactum $ and i f Oi = S(A ,■, a) is a a-neighborhood of the set A-,-,
then w = {Oi , • • • , } and d> = {Oi , • • • , Os) are, respectively, open and
closed e-coverings of $ which are both similar to the covering a.
L em m a 8 .3 4 . I f
O) = {Oi, • • • , 0 ,}
36 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

is an open covering of a compactum <f>, there is a positive number 77 = 77(a)),

with the following property: every set M CZ of diameter < 77 is wholly con­
tained in at least one of the sets of the covering a;.
Proof. Let 77 be the greatest lower bound of the diameters of all sets
M d «f>which are not contained in any element of the covering w. We shall
show that 77 is positive and this will prove the lemma. Let
M i, M i,
be any sequence of such sets whose diameters approach 77. Let pn € M n .
Passing, if necessary, to a subsequence of {pn\, we may suppose that
converges to a point p 6 4*. The point p is contained in an element 0 t of
the covering a) and consequently has a positive distance 2 e from the closed
set R \ O i , so that
(8.341) S(p, 2 e) C Oi.
By definition, no M n is contained in 0 ,• and, a fortiori, in S(p, 2 e). Since
pn € S(p, e) for all sufficiently large n, it follows that for all such n the
diameter of M n must necessarily be > e [since, otherwise, M n C S(p, 2e)].
However, since 77 is the greatest lower bound of the diameters of the sets
M n , 77 > e. This proves the lemma.
We shall conclude this preliminary investigation of coverings with the
following very general and quite elementary proposition which will be con­
venient in the sequel.
8.35. Let 6 be a property of closed coverings (for example, the property of
having a given order). In order that there exist, for every open covering co of
a compactum 4>, a closed covering which is a refinement of u and which has
property (§, it is necessary and sufficient that there exist, for every e > 0 , a
closed e-covering of the compactum 4> with property @.
Proof of necessity. For any e > 0, take any open e-covering and then a
closed covering which is a refinement of the open covering, and which has
property <$. This yields a closed e-covering with property (5.
Proof of sufficiency. If, for every e > 0, there exist e-coverings with
property @, to obtain a closed covering with property (5 which refines a
given open covering w, it suffices to determine the number 77 = 77(a)) of
Lemma 8.34 for the covering a) and to take an arbitrary closed 77-covering
a with property 6 . According to the definition of 77, the covering a is a
refinement of a).

§8.4. Definition of dimension.

D e f in it io n 8 .4 1 . The dimension of a bicompactum <f>, denoted by d im <f>, is
the least integer n such that every open covering of <f> has a refinement which
is a closed covering of 4> of order < n -f- 1 .
If there is no least integer n for which this is true, dim <f> = 00 .
§8 ] COVERINGS OF NORMAL SPACES 37

R em ark 1. If d im $ = oo th er e e x is ts fo r e v e r y n an o p en c o v e r in g a?
of $ su c h t h a t every c lo sed c o v e r in g w h ic h refin es a> h a s ord er > n + 1 .
E x e r c i s e . Prove that the definition of dimension may also be formulated
as:
8.41'. The dimension of a bicompactum <f>is the least integer n for which
every open covering of $ has a refinement which is a simple closed covering
of order n + 1 .
R e m a r k 2. The dimension of the empty set shall, by definition, be equal
to —1 .
By 8.35, the definition of dimension for compacta can be phrased as:
8.42. The dimension of a compaclum 4> is the least integer n such that for
every t > 0 there exists a closed t-covering of <f>of order <n -f- 1 .
R e m a r k 3. The intuitive meaning of this definition is very simple. For
n = 2 , it asserts th at every two-dimensional “area” (compactum) can be
“paved” with arbitrarily small “bricks” (closed sets) in such a way that
no point of the area is contained in more than three of these bricks, but
th at if the bricks are sufficiently small, at least three have a point in
common. Similarly a “volume” can be filled in with sufficiently small
bricks so th at no point of the volume is in more than four of the bricks, but
if the bricks are sufficiently small at least four will have a point in common.
Hence the theorem, known as the Pflastersatz (see V, Theorem I' and 2.24),
that, e.g., a square has dimension 2 .
The perception of the fact that the order of a covering contains the con­
cept of dimension was a great achievement of mathematical thought. This
achievement is due to Lebesgue, who was the first to formulate and prove
(not, however, completely without error) the Pflastersatz in 1911. The first
complete proof of this theorem was given by Brouwer in 1913.
If there is no n satisfying 8.42, dim 4> = » , by definition. Then for every
n there is an t > 0 such that every closed t-covering is of order > n + 1-.
Since open sets, closed sets, and the order of a covering are invariant
under a topological mapping of a bicompactum 4> onto a bicompactum 4>',
it follows th at two homeomorphic bicompacta have the same dimension,
i.e., dimension is a topological invariant.
A significant part of Chapter V and all of Chapter VI are devoted to the
dimension theory of compacta. We shall close this chapter by deriving one
property of dimension which is an immediate consequence of the definition;
we shall prove it only for compacta, leaving the proof for bicompacta to
the reader.
8.43. I f a compactum <t>0 is contained in a compactum 4>, dim 4>0 < dim 4>.
Indeed, let e > 0 and let a = {Ai , • • • , A.) be a closed t-covering of
the compactum <f>. Then

ao = {4>0 n i i , • • • , 4>0 n A*}

38 PROPERTIES OF TOPOLOGICAL SPACES [CH. I

is an €-covering of the compactum <t>o whose order, obviously, does not ex­
ceed the order of a. This proves 8.43.
R e m a r k 4. Theorems 8.22 and 8.23 imply
8.44. The dimension of a bicompactum <t> may be defined as the least in ­
teger n satisfying the condition:
Every open covering of the bicompactum <t> can be refined by an open cover­
ing of <t> of order < n + 1 .
In particular, the dimension of a compactum <t> can be defined as the least
integer n such that for every e > 0 there exists an open e-covering of <t> of order
< n + 1.
REFERENCES*
1. General references: [A-H; I, II]; [H; V II-IX ]: [H*; VI, VIII]; [L; I]; [N; II-IV];
[P; it]; [Si]; [W; I-II, VII (§§1-3)]; also the relevant sections of [Wi; I-III].
2. [A-H; I (§§1-3, 7, 8)]; [L; I (§§1, 2, 8)]; [P; II (§§7-10, 12, 15)]; [Si; II, VI].
3. [Si, p. 122],
4. [Si; p. 128, Theorem 74],
5. [N ; p. 71, Theorem 1.1 and Corollary],
6. [A-H; I (§§4-6), II (§1)]; [L; I (§§5-6)].
7. See [Si; p. 105, Theorem 63] for a proof of a stronger theorem.
8. See [4] and [7].
9. Aleksandrov [g; §5, Theorem 2],
10. Aleksandrov [g, §3].
H . A l e k s a n d r o v [g, §3]; [A-H; II (§§2, 3)]; [L; I (§§5, 6, 8)]; [Si; V].
12. For necessity see [Si; p. 95, Urysohn’s Lemma]; for sufficiency see [L; p. 27, 34.1],
13. [L; p. 28, 34.2]; [A-H; p. 73, Satz VIII; proof on p. 75]. For the special case of a
metric space see [A-H; pp. 76-78].
14. Bicompactness in the original. The term “compact” has been substituted for
“bicompact” since the book deals mostly with metric spaces and the two no­
tions are equivalent in such spaces (for a proof see [N ; p. 47, Theorem 15.3,
and p. 204, Note 10]; also [L; I, (§5, 21)]).
15. [Si; p. 117, Theorem 70],
16. [N; p. 42, Theorem 14.3].
17. [Wi; p. 35, Lemma 12.10],
18. [Wi; p. 69, Lemma 1.1].
19. [Wi; p. 74, Theorem 1.27],
20. [L; p. 39, Theorem 46.4],
21. [L; p. 19, Theorem 24.1, and p. 24, Theorem 30.2],
22. See [21] and [L; p. 35, 43.8-43.10],
23. [Wi; p. 74, Theorem 2.1],
24. Aleksandrov [g, §5],
25. [Wi; p. 69, Lemma 1.2],
26. Apply 4.51.
27. [Wi; p. 74, Theorem 1.28].
28. [Si; p. 153, Theorem 80].
29. Aleksandrov [a]; [A-H; I (§5), II (§2)]; [W, pp. 122-127],
30. [B; VII (§35)].
31. A l e k s a n d r o v [c ; g, §6].
32. [A-H; II (§4)]; [H*, §26]; [L; pp. 37-38, 45]; [N; pp. 17, 18, 51]; [Si, VII],

* For abbreviations see the Bibliography.

 Chapter I I
TH E JO R D A N TH EO R EM

The proof of the Jordan theorem given in this chapter is due to E.

Schmidt (see Schmidt [a]). This proof is quite elementary and requires
no preliminary knowledge of topology except for the simplest facts about
connectedness (I, 3). These are needed to understand the theorem itself.
A second merit of Schmidt’s proof is that the auxiliary constructions on
which the proof rests (the angle function, the index of a point with respect
to a continuous mapping, etc.) have a wide range of application far beyond
the bounds of the Jordan theorem itself. They are to be numbered among
those elementary concepts with which every student of, e.g., the theory
of differential equations or the theory of functions of a complex variable
needs to be acquainted.
§1. Formulation of the Jordan theorem. Common boundaries of domains
§1.1. Formulation of the Jordan theorem. A compactum <f>homeomorphic
to a circumference is called a simple closed curve or (closed) Jordan curve.
A compactum homeomorphic to a straight line segment is called a simple
or Jordan arc. This chapter is devoted to a proof of the following impor­
tant and celebrated theorem:
T he J ordan T heorem . Let $ be a Jordan curve in the plane R2. The
complementary open set R2 \ $ consists of two disjoint domains (connected
open sets) T0 and Ti whose common boundary is <t>.
In other words:
1. r 2\ $ = r 0u r x, r 0n i \ = o ;
2 . r 0 and lb are domains;
3. <f> = r 0 \ r 0 = i \ \ i \ .
R e m a r k 1. A closed set A of a connected topological space R is said
to separate (or disconnect) R if the open set R \ A is not connected; but
if R \ A is connected, then A, by definition, does not separate R.
R e m a r k 10 . A space is separated by the empty set if, and only if, it is
not connected.
It is customary to state the Jordan theorem in the following more com­
pact way:
Every plane Jordan curve <f> separates the plane R into two domains whose
common boundary is <t).
The Jordan theorem contains three assertions:
1. The open set R 2 \<f> consists of two components, T0 and I\ .
2. Each of these components is an open set.
3 . $ is the common boundary of the domains T0 and .
39
40 THE JORDAN THEOREM [CH. II

R em ark 2. With regard to assertion 2, it should be noted th at the

components of an open subset of a space R need not be open sets in R]
for instance, if R is the Cantor perfect set (with its usual topology), each
point x £ R is its own component in R, i.e., the components of R are
degenerate sets and none of them are open in R. This remark indicates the
significance of Theorems 1.25 and 1.27.
We shall see in the sequel that the following theorem, also very im­
portant, is closely related to the Jordan theorem:
A plane Jordan arc 4> does not disconnect the plane R .
In other words:
I f 4> is a Jordan arc in R2, the open set R2\$ > is connected.
These two theorems, both proved in this chapter, are special cases of a
theorem which will be proved in Chapter XIV and again in Chapter XV.

§1.2. Domains in R n and their boundaries.

Let us note first that:
1.21. Every convex open set in R r‘, e.g., an open solid n-sphere, is con­
nected by virtue of I, 3.12. Hence every set homeomorphic to a convex
domain of R n is also connected.
1.22. An open set T C R n is connected, if, and only if, every pair of
points of T can be joined by a simple broken line (i.e., a broken line with­
out multiple points) contained in I\
Proof. If every pair of points of T can be joined by a broken line (even
one containing multiple points) in T, then T is connected by I, 3.12.
To prove the necessity of the condition, we shall show that if there are
two points, a and b, in T which cannot be joined by a simple broken line
in T, then T is not connected.
Let us denote by r a the set of all points of T which can be joined to a
by broken lines in T, and by A the set of all remaining points of I\
If p is any point of T, all points of R n whose distance from p is less than
e = p(p, R n \ T) are contained in T and can therefore be joined to p by
a segment contained in T. It follows readily that r a and Tb are open in
T. Since r o n Tfc = 0 and a £ r a , b 6 r& , T is not connected. This com­
pletes the proof of Theorem 1.22.
1.23. Let T be a domain of R n, n > 2, and let E C T be a finite set.
Then the set is connected, i.e., is also a domain.
Indeed, let a, b £ T \ E be joined by a broken line in T. An arbitrarily
small displacement of the vertices of this broken line will bring it into
general position (see Appendix 1) with respect to the finite set E. It will
then contain no points of E. Hence the displaced broken line will join a
and b in T \ E.
1.24. Let E n be a closed solid n-sphere in R n. The open set R n \ E n is
connected.
§1] FORMULATION OF THE JORDAN THEOREM 41

In fact, an inversion with respect to the bounding sphere S n~l of E n

maps R n \ E n into E n\ o , where E n is the corresponding open solid
n-sphere and o its center. Since it has been shown that E n\ o is con­
nected, R n \ E n is also connected.
1.25. The components of an open subset of R n are open, i.e., are do­
mains.
Before proving Theorem 1.25, let us make the following definition:
D e f in it io n 1.26. A metric space R is said to be locally connected at a
point p if p has connected neighborhoods of arbitrarily small diameter;
R is locally connected if it is locally connected at all its points.
Since spherical neighborhoods in R n are connected, it follows that R n
and all open sets in R n are locally connected (Def. 1.26 implies that every
open set of a locally connected space is also locally connected). The curve
consisting of the segment —1 < y < 1 of the y-axis and of the graph of
y = sin \ /x, 0 < x < 1/ t (Fig. 2, p. 15) is an example of a compactum
which is not locally connected. This curve is not locally connected at any
of its points on the axis of ordinates.
These remarks make it clear that 1.25 is a special case of the following
im portant theorem:
1.27. The components of a locally connected space are open sets (hence
domains) of the space.
Proof of Theorem 1.27. Let R be a locally connected space, Q a com­
ponent of R, and p a point of Q. Denote by U any connected neighborhood
of p. By I, 3.13, the set Q u U is connected. Since Q is a component, Q =
Q u U. Hence p is an interior point of Q. This proves the theorem.
1.28 If <f> is a compactum in R n, R n \ 4 > has exactly one unbounded
component.
Let E n be a solid sphere containing $ in its interior (such a sphere exists,
since 4> is a bounded set). The set R n \ E His connected. Hence it is con­
tained in one component of R n which is, therefore, unbounded. All
the remaining components of 7 ? " \ $ are contained in E n and, conse­
quently, are bounded.
D e f in it io n 1.29. A compactum 4> is said to be an absolute boundary in
R n if 4> separates R n and is the common boundary of all the components
of the open set R n \4 > .
D e f in it io n 1.291. An absolute boundary in R n is called a regular
boundary if it separates R n into precisely two domains.
Such curves as a circumference, an ellipse, etc., are regular boundaries
of domains in R 2; the sphere and toms are examples of regular boundaries
of domains in R 3. The Jordan theorem asserts that every plane closed Jordan
curve is a regular boundary in R~.
At first glance it would seem that every absolute boundary is regular:
it is difficult to imagine, for instance, a plane closed set which separates
42 THE JORDAN THEOREM [CH. II

the plane into more than two domains of which it is the common bound­
ary. Nevertheless, such closed sets exist; they were first constructed by
Brouwer in 1909. The following example (first published by the Japanese
mathematician Yoneyama) is a slight modification of the original ex­
ample of Brouwer.
Let us imagine an island 4> surrounded by a sea T. We shall represent
the island by a circle (together with its circumference), so th at the sea
T = R 2\ <f> is an open set. Let the island 4> have two circular lakes, one
containing cold water and the other warm (each of the lakes is in the
interior of the circle).
Let us imagine the following program. In the first hour, three dead-end
canals are dug in the interior of the island, one leading out of the sea and
the other two leading out of each lake. Each of these canals is made to
wind in such a way that the distance of every point of dry land of the
island is less than 1 mile both from the sea water and from the water of
each of the lakes (Fig. 10).
In the next half hour, each one of the canals is extended in such a way
that, at the end of this half hour, the distance of every point of dry land
(remaining on the island) from the sea water and the water of both lakes
is less than \ mile. In the following | hour the canals are continued so
§1] FORMULATION OF THE JORDAN THEOREM 43

th at the distance of every point of dry land from the sea, warm fresh
water, and cold fresh water is made less than \ mile, etc. After 1 + \ +
j + • • • + (^)" hours the work on all three canals is continued in a man­
ner to make the distance of each point of dry land from the sea and each
lake less than (^)"+1 miles (this will be accomplished in the next (§)n+1
hours). Throughout, the canals never touch each other or cross them­
selves, i.e., they remain dead-end canals. At the end of two hours, there
remains of the island only a closed set 4>0 which is nowhere dense in the
plane. An arbitrary neighborhood of a point of 4>0 contains points of the
sea as well as points of the two lakes. Hence the closed set 4>0 is the com­
mon boundary of three plane domains, the sea and both lakes.
An absolute boundary in the plane R2 which separates R 2 into an arbi­
trary finite, or even denumerable, number of domains can be constructed
by the same method.
R e m a r k . A compactum 4> C R2 which is the common boundary of two
plane domains (two components of the open set may not be an
absolute boundary: the set <f> (Fig. 1 1 ) consisting of a circumference and
a spiral curve winding inside it separates the plane into three domains,
but is the boundary of only two of these domains (in Fig. 11 one of these
domains has been left blank, the other is stippled; the third is the domain
exterior to the circle and is cross-hatched).
T h e o r e m 1.2. In order that a compactum <f>-fee an absolute boundary in

F i q . 11
44 THE JORDAN THEOREM [CH. II

R", it is necessary and sufficient that it separate the space R n into at least
two domains and that no proper closed subset 4>o of <t> have this property, i.e.,
ij4>0 C $ is closed, then R n \ 4 0 is connected.
Proof. Let <t> separate R n into a countable number of domains (in a
space with a countable basis, every collection of mutually disjoint open
sets is countable at most) T i, r 2, •• • , , ••• and assume th at <t> is
the common boundary of all these domains.
Let <t>0 <Z <f>. Since <t> = f , - \ I \ , it follows from I, 3.19 that each of
the sets
r \ u (4>\4>0)
t
is connected. Hence, by I, 3.16 the set
U jr* u (4>\4>o)] = R n \<*>o
is also connected. This proves the necessity of the condition.
Now let 4> separate R n into domains Tj , r 2, • • • , I \ . , • • • and suppose
that no closed set 4>o (Z 4> separates R n. We shall show that then
4> = f * \ r*
for every k. I t is clear, to begin with, that no point p 6 Ft \ can be
contained in any . Hence p £ 4>, so that
= r* \ r* C <t>.
Since I \ cz R n \ , i k, it follows that R n \ f* is nonempty and
that, therefore,
R n \ * k = Tk u (R n \ T k)
is the union of two non vacuous disjoint open sets I\- and R n \ f k . Hence
Rn is not connected, i.e., separates R n. Since Q 4>, our hypoth­
esis implies th at = 4>for all k. This completes the proof of the theorem.
§1.3. Outline of the proof of the Jordan theorem. Let us first make the
following simple remark:
A plane Jordan arc or Jordan curve 4> is nowhere dense in the plane.
Indeed, every connected subset of 4> is separated by some pair of points
of this subset; the interior of a circle is a connected set which is not sepa­
rated by any of its finite subsets. Therefore, 4> does not contain the in­
terior of any circle. Hence, since is closed, it is nowhere dense in the
plane.
In §2 we introduce a concept basic to the whole proof, the index of a
point relative to a closed curve. In §3 we prove th at a simple arc does
not disconnect the plane, and §4 contains a proof of the fact that a closed
Jordan curve separates the plane into two domains.
§1] FORMULATION OF THE JORDAN THEOREM 45

These two propositions imply the last assertion of the Jordan theorem:
a Jordan curve <t> is the common boundary of the two components of the
open set R 2 \4 > . To prove this it suffices, by 1.2, to show that no closed
set 4>o C 4> separates R 2. Let 4>o C and let £ be any point on the curve
<t> not contained in 4>o. Since 4>n is closed, there exists an open arc (ab) of
<f> containing the point £ and free of points of 4>0. The set 4>i = 4>\ (ab)
is a simple arc lying on 4> and containing 4>0 .
I t is required to prove that R 2\$>o is connected. However, since every
point of 4>i \ 4 >0 is a limit point of R2 \<f>i , the connectedness of R2 \<&o
follows from that of (I, 3.19), i.e.,

R2 \ 4>i Q R2 \ 4>o = /?2 \ f > i U $ i \ $ o ,

which in turn is contained in the closure of R2 i.
Hence to complete the proof of the Jordan theorem, it is enough to
prove that a Jordan curve separates the plane into two domains and that
a simple arc does not disconnect the plane.

§1.4. Notation. Orientation of simple arcs and simple closed curves.

We shall denote a simple arc with endpoints a and b by [ab] and the corre­
sponding open arc, i.e., the set [ab] \ a u b, by (ab). If A denotes a simple
arc, then A will stand for the corresponding open arc.
Since a simple arc is a topological image of a segment, it is possible to
speak of an oriented simple arc, i.e., an arc directed from a to b or from
b to a. An oriented simple arc will be denoted by [abp or [ba]"", respec­
tively; an oriented open arc by (ab)-’ or (ba)-’, respectively.
The direction (or sense) from a to b on a simple arc [ab] may be de­
fined as follows. Let C be any topological mapping of the segment 0 <
x < 1 onto the simple arc [ab] such that (7(0) = a and (7(1) = b. The
mapping C orders the points of the arc [ab]: if y = C(x), y' = C(x') are
two points of [ab], put y < y' if x < x' on [0 1 ].
I t remains to be shown that this order is independent of the choice of
the mapping C.
Let Ci and C2 be topological mappings of the segment [0 1 ] onto the arc
[ab] such th at Ci(0) = C2(0) = a, Ci(l) = C2(l) = b. If Ci and C2 were
to define different orders on [ab], then, for some two points

y = Ci(xi) = Ci(x2) £ [ab]

y ' = Ci(x'i) = C2(x'2) G [ab]

it would follow that, e.g.,

(1.41) Zi < x \ , x2 > x'2 .
46 THE JORDAN THEOREM [CH. II

Then C = C2 lCi is a topological mapping of the segment [01] onto

itself and
(1.42) C(0) = 0, C( 1) = 1,
C(xi) = x2, C(x'i) = x'2 .
However, C, as a topological mapping of [01] onto itself which leaves
the endpoints of the segment fixed, is a strictly monotonic function. Hence
(1.42) is incompatible with (1.41). This proves the assertion.
In the same way, a topological mapping of a circumference onto a
Jordan curve induces a direction of circuit or sense on the latter by means
of a given sense on the former.
More precisely, an oriented Jordan curve is a simple closed curve 4>
on which every simple arc <f>0 a <f> has been assigned a definite sense in
such a way th at if $o C $, $i C $ are any two arcs on 4? and <f>o ^ 4>i,
then the sense assigned to <i>0 coincides with th at assigned to 4^ .
An oriented Jordan curve will be called simply an oriented curve and
will be denoted by d*”*if the curve itself is denoted by 4>. If a Jordan curve
is given by several letters, it will be written as e.g., (abed). An oriented
curve will then be symbolized by (abcd)~\

§2. The angle function of a continuous mapping of a segment into the

plane. The index of a point relative to a closed path in the plane
§2.1. The functions Fa(p, C, x) and f(p, C, x i , x 2). Let C be a continuous
mapping of the segment [eoei] = e0 < x < e\ of the real line into the plane
R 2. Let p 6 R2 be a point of the complement of <f> = C([e0ei]).
Finally, let a be a point of the segment [eoei]. Let us suppose th at in
the plane R2 a definite convention for measuring an angle in the positive
sense (e.g., counterclockwise) has been chosen.
The angle C(a)pC(x) is defined only to within an integral multiple of
2w. However, it can be defined uniquely for all x, e0 < x < et , by choos­
ing a definite value of this angle for a fixed x = x0 and requiring th at it
be a continuous function of x. Indeed, let the points
— Co, fli, • • • , as = e\
divide [eoei] into segments [a0ai], [a1a2], • • • , [a,_ia,] so small that C maps
each of them into the interior of some acute angle with vertex p. If the
value of the angle C(a)pC(x) is defined for a fixed x0 , a t- < x0 < a{+1 , it
is then uniquely defined by continuity for all x, of < x < a i+i ; in par­
ticular, therefore, for x = and x = ai+1 . But the value of the angle
C(a)pC(x) for x = ci;+i uniquely determines the values of this angle for
all x £ [a1+iai+2], and then for all x 6 [a;+2R;+3], etc. In exactly the same
§2] ANGLE FUNCTION OF A MAPPING 47

way, if the angle C(a)pC(x) is defined for x = a,-, its value is uniquely
determined by continuity on the segment [a,-at-_i], etc. Hence the angle
C(a)pC(x) is defined for all x £ [e0Ci].
D e f in it io n 2 .1 1 . The value of the angle C(a)pC(x), uniquely defined
for all x, eo < x < .e \, by the condition that C(a)pC(a) = 0, is denoted
by Fa(p, C, x).
R e m a r k 1. The function Fa(p, C, x) is clearly continuous in each of
x, a, and p. Furthermore, it is easy to see that it is a continuous function
of all its arguments p, C, x, a simultaneously, i.e., for fixed x, a, p, C and
for every e > 0, there exists a b > 0 such that Fa>(p', C , x') is defined
and
Fa(p, C, x) - Fa,(p', C , x')\ < e
for
\x ' ~ x \ < 8, | a' - a | < b, pip1, p) < b, P(C', C) < b.
Here, in accordance with I, 7.3, p(C", C) < 5 means that p[C"(£), C(£)] < b
for all £, eo < £ < e\ .
R e m a r k 2. If a, a' are any two points of the segment [6o€i],

(2.11) Fa’{p, C, x) = FAV , C, a) + Fa(p, C, x).

Hence Faip, C, x2) — Faip, C, xi) is independent of a.
D e f in it io n 2 . 1 2 . Let

(2 .1 2 ) fip , C, Xl , X 2) = Faip, C, Xi) — Faip, C, Xi )

for any a £ [coej; fip , C, X i , x2) is also a continuous function of its argu­
ments.
§2.2. The index of a point relative to a closed path. Let us now assume
th at the mapping C satisfies the condition
Cie 0) = C(d).
The mapping C is then said to define a “closed path” in the plane. The
intuitive geometric meaning of this term is clear. A rigorous definition of
a closed path would have to state the conditions under which two con­
tinuous mappings determine the same closed path. However, we shall
not use this notion and will leave it imprecise.
If C defines a closed path, the function fip , C) = fip, C, e0, ef) is equal
to an integral multiple of 2 t. Hence an integer a>(C, p) = ( l / 2 -7r)/(p, C),
which depends only on C and p, is defined for every point p in the com­
plement of Ci[eoCi]). The number w(C, p) is called the index of the point
p relative to the mapping C (or the index of p relative to the closed path
defined by C).
48 THE JORDAN THEOREM [CH. II

It is clear th at if u(C, p) is defined for a point p, it is also defined for

all points sufficiently near p. Furthermore, since u(C, p) is continuous in
p and assumes only integral values, it follows that
u(C, p') = <*(C, p)
for all p' sufficiently near p. Hence
2.21. The function u(C, p) assumes the same value for all points p con­
tained in one component of the open set R2\ C([eoei]).
Indeed, if pi and p 2 are in the same component T of R2 \ C([e0ei]), they
can be joined by a broken line in F. The function u(C, p) is defined for
all points of this broken line and, because it is continuous and integer­
valued, is constant there. Therefore,
u(C, pi) = «(C, p2).
This proves the assertion.
It is obvious that if the set C([eoei]) is contained in the interior of an
angle a, 0 < a < 2ir, with vertex p, then \f(p, C, Xi , x2)| < a for all
X i, x2 € [eoei].
Hence
2.22. I f C is a continuous mapping of the segment [e<A] into the plane
R2 such that C(e0) = C(ei) and C{[eoe-i\) is contained in one of the two half-
planes into which R2 is separated by some straight line, then u>{C, p) = 0
for any p contained in the other half-plane.
Let us make two important remarks.
2.23. Let C be a continuous mapping of the segment [eo^i] onto a cir­
cumference K, (1-1) on the open interval (eoet) and such that C(e0) =
Ciei). Let p be the center of K. Then u(C, p) = + 1 or —1 depending on
whether, as x varies from e0 to e\ on the segment [e0ei], the point C(x) describes
K in the positive or negative sense.
By virtue of 2 .2 1 , the same assertion is also valid for an arbitrary point
in the interior of K.
The circumference K can obviously be replaced in 2.23 by, e.g., a tri­
angular contour.
2.231. We shall call a deformation Ce , 0 < 6 < 1 , of the mapping
C = Co admissible if the fixed point p never meets the set C«([eo€i]) and
Ce(eo) = Ce(ei)
for all 6, 0 < 8 < 1 . Then u(Ce , p) is defined for 0 < 6 < 1 and, since
Fa(p, c, x) is continuous in C and u(Ce , p) is integer-valued, cc(Ce , p) is
constant, 0 < 6 < 1 . Hence
2.24. The index a(C, p) is invariant under admissible deformations of
the mapping C.
§ 2] ANGLE FUNCTION OF A MAPPING 49

§2.3. The addition formula. Let [coei], [eic2], c0 < ei < e2, be two closed
intervals of the real line and let C i, C2 be continuous mappings of [e0ei]>
[cie2], respectively, into R2 such that
(2.31) C,(c) = C2(ei).
Let C denote the mapping of the segment [eoe2] coinciding with C\ on
[eoei] and with C2 on [cic2]. Finally, let p be a point of R 2 in the comple­
ment of the set
Ci(\eoei\) u C2([eie2]) = CQeo^]).
Then (2.1, Remark 2 )
(2.32) Ff0(p, C, c2) = F„0(p, C, «i) + Fe,(p, C, e2)
Now let
(2.33) C(co) = C(ei) = C (e2) ,
i.e.,
Ci(eo) = Ci(ci) = C2(ei) = C,2(e2).
Then w(Ci , p), u(C 2 , p), w(C, p) are defined and it follows easily from
(2.32) and the definition of these numbers that
(2.34) w(C, p) = « (C i, p) + w(C2, p).
We shall need one application of (2.34) in the sequel.
a = Ci(0) = Ci(2) = Ci(4),
6 = Ci(l) = Ci(3),
[ac6 ]“ = Ci([01]),
[bgar = C,([12]),
[agbr = C,([23]),
[bdar = Ci([34]).
Let Ci be a continuous mapping of the .segment [04] of the real line into
R2 (Fig. 12) such that
Ci(4) = Ci(2) = Ci(0),
(2.35)
Ci(2 + x) = Ci(2 - x), 0 < re < 1.
Hence
Ct(3) = Cj(l).
50 THE JORDAN THEOREM [CH. II

Let p e fl2 be a point in the complement of Cx([04]). Denote by C02

and C24, respectively, the mappings of the segments [0 2 ] and [24] defined
by the mapping Cx . Finally, let Co be the mapping defined by
C0(x) = Ci(x), 0 < x < 1 and 3 < x < 4,
C„(x) = Ci(1) = Ci(3), 1 < x < 3.
Then w(Ci , p), ^(C02, p), a)(C24, p), o>(C0 , p) are defined and
(2.36) w(C0, p) = w(C°2, p) + o>(C24, p).
In fact, (2.34) implies that
u(Ci , p) = w(C°2,- p) + w(C2\ p).
It remains to be shown that u(Co, p) == w(Ci , p).
However, the mapping Co is obtained from Ci by an admissible de­
formation
Cfl(x) = Ci(x),
C.(l + x) = Ci(l + fte), 0 < x < 1,
Ce(2 + x) = Ci[l + 0(1 - x)], 0 < d < 1.

Ce(3 + x) = Ci(3 + x);

Hence co(Co, p) = co(Ci , p), since aj(C, p) is invariant under admissible
deformations of C, and this at once implies (2.36).
§2.4. The index of a point relative to a Jordan curve. For the remainder
of this section 4> will denote a plane Jordan curve. We shall call a con­
tinuous mapping C of the segment [01] of the real line onto a Jordan curve
4> which maps the open interval (0 1 ) topologically and which satisfies
the condition
C(0) = C(l) = e' e 4>
a simple mapping.
A simple mapping of [01] defines an orientation o f 4 > : i f 0 < a < 6 < l ,
the orientation induced on 4> by C shall be ( e'C(a)C(b)e,)~'.
2.41. If p G 7?2 \4 > and C, C' are two simple mappings of [0 1 ] onto a
Jordan curve 4> which induce the same orientation on 4>, then
o>(C, p) = a>(C', p).
To prove this, it suffices, by 2.24, to show that C and C' can be deformed
into each other by a deformation Ce , 0 < 6 < 1 , such that
(2.41) Ce([01]) = 4>, C.(0 ) = C.( 1 ) = c', 0 < 6 < 1.
§2] ANGLE FUNCTION OF A MAPPING 51

I t is enough to prove the last assertion for a circumference of length

1. In th at case, however, it follows from
2.42. Let Co he any simple mapping of [0 1 ] onto a circumference <t> of
length 1 . Let Ci be the mapping such that
Ci(0) = C0(0 ) = e',
and such th at Ci(x), 0 < x < 1 , is the endpoint of the arc of length x
laid off on the circumference from the point e' in the direction induced
by the mapping Co. Then C0 can be deformed into Ci by a deformation
satisfying (2.41).
Indeed, for each 0 , 0 < 0 < 1, let
Ce(0) = C 9( l ) = Co(0).

Consider the arc L x = Co(x)Ci(x) of 4>, 0 < x < 1 , which has Co(x) and
C\(x) as its endpoints and which does not contain e'; define Co(x) as the
point dividing the arc L x in the ratio 0 : (1 — 0), i.c., such that
[C0(x)C,(x)]:[Ce(x)Ci(x)] = 0:(1 - 0).
Then Ce is the required continuous deformation.
E x e r c i s e . P r o v e t h a t tw o sim p le m a p p in g s Co and Ci w h ich in d u c e
o p p o s ite o r ie n ta tio n s on 4> c a n n o t b e d efo rm ed in to each oth er.
D e f in it io n 2.4. Let4>-’ be an oriented Jordan curve and let p £
R 2\$> ;
the number w(C, p), where C is any simple mapping of [01] onto 4> induc­
ing the given orientation on is called the index of the point p relative
to the oriented Jordan curve <t>~> and is denoted by co(p,
We shall require, finally, the following proposition:
2.43. Consider in R 2 three simple arcs with common endpoints a, b, no
two of which have any points in common except for their endpoints. De­
note these arcs by [ab]i , [ab]2, [ab]3, respectively. From these we form
three oriented Jordan curves
[af>]W u [M 2 ", [a&]i~ u [M3^, M>]-T u [5a]3“>.
Then for any point p not on any of the three arcs,
« (p , [ab]r u [M2"*) + « (P , [M 2 -* u [ba]^)
^ ‘ ’ = w(p, [ab]i~ u [ M O -
This formula follows from (2.3G) and the following almost obvious
proposition:
2.431. Let Co be a continuous mapping of [01] onto a Jordan curve $
such th at Co maps [O5] topologically onto some are Xi = [e'e'i] C 4>, the
segment [3 §] onto e \ , and finally the segment [§L] topologically onto the
arc X2 = [e'ie'} d different from Xi . Then the mapping Co can be dc-
52 THE JORDAN THEOREM [CH. II

formed into a simple mapping by means of a deformation Ce satisfying

(2-41).
It suffices to prove this proposition on the assumption that $ is a cir­
cumference of length 1 ; then G\ and Ce are defined exactly as in the proof
of 2.42.
§2.6. (This article will not be used in the proof of the Jordan theorem.)
The index of a point p 6 R 2 relative to a continuous mapping of a cir­
cumference into R2\ p ; degree of a continuous mapping of a circumference
into a circumference. Let Sg be an oriented circumference of length 1 .
Choose an arbitrary point a,g on Sg . We shall say that Sg is described in
the positive direction if this corresponds to the given orientation of Sg .
Let C^ be a continuous mapping of Sg into R 2 and let p £ R2 \ C^(Sg).
Let Cgl be the continuous mapping of the segment 0 < x < 1 onto the
circumference Sg which assigns to each point x, 0 < x < 1 , the endpoint
of the arc of length x laid off on Sg in the positive sense from the point ag .
Then C = C*Cgl is a mapping of the segment [01] into R2\ p and
C(°) = C(l).
It is easy to see that the index of a point relative to the mapping C is
independent of the choice of the intermediate mapping Cg 1 (i.e., of the
choice of the point ag which completely determines Cg01). Hence the index
of the point p relative to the mapping C = C^Cg1 depends only on the
mapping (P (but also on the chosen orientation of Sg and the sense in
which positive angles are measured in R2). It is therefore natural to call
this index the index of a point relative to the mapping Cff of the circum­
ference Sg into R2\ p. Hence, by definition, let us put
co(C*, p) = <o(C, p). *■
Now let S a be a circumference in R2 with center p, oriented in the same
sense in which positive angles are measured in R2. If C j is a continuous
mapping of Sg into S a , the index of the center p of S a (just defined) rel­
ative to Ca is called the degree of the mapping C j.
In this article we shall show that the degree of C j completely charac­
terizes the homotopy class of C j (see I, 7.4). We shall begin by denning
a normal mapping C j of degree where w is an arbitrary integer.
We shall assume for simplicity that both circumferences S a and Sg are
of length 1 .
Let us choose, once and for all, definite points ag £ Sg and a„ ( £ a .
A mapping v j is called a normal mapping of degree w of Sg into S a if it
assigns to each point £ € Sg the point vj(£) £ S a determined as follows.
Take the arc [a ^ P on Sg in the positive sense. Lay off on S a from the
point aa , in the positive sense if co > 0 and in the negative sense if co < 0 ,
an arc [aa77] ^ whose length is equal to the product of | w | and the length
§2] ANGLE FUNCTION OF A MAPPING 53

of the arc [aa$] The endpoint 77 of the arc [aar]] C S a is, by definition,
the point v/(£). The degree of the mapping vj3 is obviously to; a normal
mapping of degree 0 maps Sp onto the point aa .
Our purpose will be achieved if we prove the following two propositions:
2.51. Every continuous mapping C j of Sp into S a is homotopic to a nor­
mal mapping whose degree is the same as that of C j.
-2.52. Two homotopic continuous mappings of an oriented circumference
Sp into an oriented circumference S a have the same degree.
Proof of 2.51. Let us first note that every continuous mapping of Sp
into S a is homotopic to a mapping C j such that Cj(dp) = aa . (The ho-
motopy is realized by a simple rotation through the appropriate angle.)
Therefore, it suffices to prove Theorem 2.51 for a mapping C j such that
Ca{o,p) = aa • We shall now proceed with the proof.
Let w be the degree of the mapping C j and consider the continuous
function F(Cap, x) defined on the whole real line in the following way.
Let us set
F (C j, x) = Fq(P, C , x), 0 <x < 1,

where C = C jC p 1 and p is the center of S a (the mapping Cpl is defined

as above; the point ap £ Sp is at all times fixed). For all the remaining
values of x let us define F (C j, x) by means of the equation
F (C j, x + 1) = F (C j, x) + 2 ™,

remembering that w(C, p) = (l/'2ir)Fo(p, C, 1).

The function F (C j, x) (for fixed ap) is uniquely defined by the mapping
Ca and is called the angle function (turning function) corresponding to
the mapping C J*.
In general, we shall call any real continuous function F(x) defined on
the whole real line and satisfying the conditions
F(0) = 0,
F(x + 1) = F(x) + 2iry,
where 7 is a fixed integer, an angle function.
2.511. Every angle function F(x) is the angle function corresponding to
some continuous mapping C j of the circumference Sp into S a such that
C a (.dp) da .
Indeed, let us assign to each point f = Cp x of Sp the point 77 of S a
whose polar angle is F(x) (measured from the point aa). Denote the point
77 by Cj(£)- It follows easily from the definition of an angle function that
C j3is a single-valued continuous mapping of Sp into S a .
F(x) is obviously the angle function corresponding to C j . Since, on
54 THE JORDAN THEOREM [CH. II

the other hand, the image of the point £ = Cplx £ Sp under the mapping
C j was uniquely defined by its polar angle F (C j, x), there exists just one
continuous mapping C j [satisfying the condition Ca{af) = aa), to which
the given angle function
F(x) = F (C j, x)
corresponds.
We have thus established a (1-1) correspondence between the con­
tinuous functions C j of S ,3 into S a , for which C j(a f) = aa , and the
angle functions.
Let us note, finally, the following proposition, whose proof may be left
to the reader:
2.512. If Fe(x), 0 < 0 < 1, is a family of angle functions, continuous in
the parameter 0, then the corresponding mappings C j are also continuous
in 0.
The proof of Theorem 2.51 can now be concluded without any difficulty.
Let us set
F$(x) = (1 — 6)F(x) -(- 2irdwx, 0 < 0 < 1,
where F(x) = F0(x) is the angle function corresponding to the mapping
C j of degree u. We shall prove that F$(x) is an angle function. Since
Ft(0) = (1 - d)F(0) = 0,
it suffices to prove that
Fe(x + 1) — F$(x) = 2ttco.
But
Fe(x + 1) - Fe(x) = (1 - 6)[F(x + 1) - F(x)] + 2ir0u.
Since F(x) is the angle function corresponding to the mapping C j of degree
u,
F(x + I) - F(x) = 2wu,
so that
F$(x + 1) — F»(x) = (1 — d)2ivw + 2w0w = 2ttco.
Therefore, all the F0(x) are angle functions, continuous in 0 and de­
forming F0(x) into the angle function
Fi(x) = 2-n-wx,
which obviously corresponds to a normal mapping of degree to. This, by
virtue of 2.512, completes the proof of Theorem 2.51.
§3] theorem : a s im p l e arc does not d is c o n n e c t the plane 55

Proof of 2.52. Let oC / and iC j be two homotopic mappings. It is required

to prove th a t they have the same degree.
Let the deformation eC j , 0 < 6 < 1, carry 0C j into iC j-
Let Cpl be the mapping of the segment [01] onto Sp defined at the
beginning of this article (see p. 52). Then eCjCp01 is an admissible de­
formation of oCjCp01 into \CaCpl, so that, by 2.24, the index of the center
p of S a is the same with respect to both mappings oCjCp01 and iC /C p 1.
But this also means that the mappings 0C j and iC j (of Sp into S a) have
the same degree. This proves Theorem 2.52.
Theorems 2.51, 2.52, and the existence of a normal mapping of degree
co for any integer co may be combined in one proposition.
2.5. There exist continuous mappings of arbitrary degree co of an oriented
circumference Sp into an oriented circumference S a ; two continuous mappings
of Sp into S a are homotopic if, and only if, they have the same degree.
R e m a r k 1. Let C be any continuous mapping of a circumference Sp into
R 2 \ p, S a Cl R2 a circumference with center p, and Ca the central pro­
jection (with center p) of the set R2\ p into S a ■
The index of the point p relative to the mapping C is equal to the degree
of the mapping CaC of Sp into S a ■ (This assertion follows from 2.24.)
R e m a r k 2. The methods of this section admit of very diverse applica­
tions—to analysis, to the theory of functions of a complex variable, etc.
Thus, e.g., the fundamental theorem of algebra can be proved, literally in
tvTo words, by means of the notion of the index of a point relative to a
continuous mapping.
§3. Theorem: A simple arc does not disconnect the plane
D e f in it io n 3.1. Let be a closed subset of the plane 7i2; we shall say
th at 4> does not separate two points p and q of the open set R2 \ «f>, if p
and q are in one component of R2 \ 4>, i.e., if they can be joined by a
simple broken line in R2 \ 4>.
To prove the theorem stated at the head of this section we note first:
3.2. Let 4> = [ab] be a simple arc in R 2 and c ^ 6 a point of 4>; further,
let p, q be points of T = R 2 \ 4>. If the arc [ac] <Z 4> does not separate p
and q (if c = o, the arc [ac] becomes the point a), there exists a point c!
on the arc icb) CZ 4> such th at the arc [ac'] docs not separate p and q.
Indeed, let [pq] be a broken line joining p and q in the complement of
[ac]. Then
p([ac], [pq]) = € > 0.
Let us take a point c', c' ^ c, on the arc [c5] so near to c that

5([cc']) < f.
56 THE JORDAN THEOREM [CH. II

Clearly, [cd], and therefore [ac'] also, has no points in common with [pq].
This proves the assertion.
3.3. I f p and q are not separated by any subarc [ae\ of the simple arc
[ab], c b, then the arc [ab] does not separate p and q.
Before proving this proposition, we shall deduce the theorem stated at
the head of this section from 3.2 and 3.3.
Let us suppose that the simple arc <f> = [ab] disconnects the plane. Then
there are two points p and q which are separated by [ab].
The set consisting of the single point a obviously does not separate
p and q. Therefore, by 3.2, there is a subarc [aa'] of [ab] which does not
separate p and q. Let c be the least upper bound (on the oriented arc
[abP) of the set of those points a' £ [ab], for which the arc [aa'\ C [ab]
does not separate p and q.

By 3.3, the arc [ac] does not separate p and q. In virtue of 3.2, c is identi­
cal with b, i.e., [ab] does not separate p and q. Hence it remains to prove
3.3.
Let K ' be a circle with center b such that a, p, q are in the exterior of
K'. Let d be the last point of [ab] n 7v'0 (on [abP), where K '0 is the cir­
cumference of the circle K '. The arc [ac'] d [ab] has a positive distance
from b. Let K be a circle with center b and radius less than p(b, [ac']);
denote the circumference of this circle by K 0 . Hence the first point (on
[ab] ) of K 0 n [ab] comes, on [abP, after the last point of 7\'0 n [ab], so that
every point which follows a point of K 0 n [ab] on [ab] is in the interior of K '.
We shall now prove 3.3 first for the special case that [ab] n 7\0 consists
of a single point c.
§3 ] th e o k e m : a sim ple akc d oes n o t d isc o n n e c t t h e p la n e 57

Since [ac], by hypothesis, does not separate p and g, there exists a broken
line [pq] not meeting [ac] (Fig. 13). Let px be the first, and qx the last,
point of [pg] n Ko on [pg]~*. Both px and qx are different from c and can
therefore be joined by an arc [pigi] of the circumference K 0 not containing
the point c. The arc [ppx] u [pxqx] u [gig] (where [ppi] C= [pq], [p^] CZ K 0 ,
[gig] Cl [pg]) joins p to g and has no points in common with [afe].
Let us now pass to the case in which [ab] n Ko consists of more than one
point (Fig. 14).

Let us denote by X" = (ddV any open arc of K 0 in the complement

of [a6] n K 0 but with its endpoints in [ab] n Ko and oriented in the positive
sense (i.e., in the sense corresponding to a counterclockwise circuit of K 0).
Let X' = (d'd) ~ be the oriented arc of Ko complementary to the arc X
(hence the orientations of X and X' correspond to a counterclockwise circuit
of Ko).
Denote by C\01 a topological mapping of the segment [01] of the real
line onto the simple arc X, by C,vd~ a topological mapping of [12] onto
the oriented arc [d'd]~* C [ab], and by Cdd<23 the topological mapping of
[23] onto the oriented arc [dd']~' Cl [ofe] defined by the equation
CM' \ 2 + x) = <7^/(2 - x ) , 0 < x < 1.
Finally, let C\-u be a topological mapping of [34] onto X'.
The mappings C'x01, Cd' / 1, Cdd™, C \* \ combined yield a mapping C
of [04] satisfying (2.35); the corresponding mapping C0 is obviously a
mapping of [04] ~ onto Ko oriented counterclockwise. The mapping ( f 2 is a
58 THE JORDAN THEOREM [CH. II

mapping of [02] onto the "closed path” consisting of the arc X-* and the
arc [d 'd p (Z [ab], and C24 is a mapping of [24] onto the closed path con­
sisting of the arcs [dd']"” (Z [ab] and X' . By (2.36),
(3.31) w(C0, x) = ^(C02, x ) + w(C24, x),
i.e.,
(3.31') ^((T02, x) = o)(C0 , x) - a>(C24, x),
for any point x in the complement of Ko u [d'd] (where [d'd] Cl [ab]).
Every point of the arc [d'd] (Z [ab] follows on [ a b ] o n e of the points
d', d of this arc, i.e., it follows some point of the set Ko n [ab]. Hence
[d'd] (Z [ab] is in the interior of the circle K ' , so that CO2([02]) = Xu [d'd]
is also contained in the interior of K '. On the other hand, p and q are in
the exterior of K'. Hence, by 2.22,
(3.32) o;(C°2, p) = 0, «(C” , q) = 0.
Let c be the last point of [ab] n Ko on \ab]
Let us say that a broken line [pq] joining p and q is admissible if it satisfies
the following conditions:
1. [pgj n [ac] = 0.
2. Ko does not contain any vertex of [pq\.
3. No link of the broken line [pq] is tangent to K 0.
An admissible broken line [pq] exists. Indeed, by hypothesis, p and q
can be joined by a broken line having no point in common with the arc
[ac\. Conditions 2 and 3 can be achieved by an arbitrarily small modifica­
tion of this broken line (Fig. 15).
If the admissible broken line [pq] has points in common with [ab], all
these points lie on the arc [cb]~* (Z [ab], i.e., in the interior of the circle K.
Hence, if an admissible broken line [pq] has points in common with [ab], it
necessarily intersects the circumference Ko and, moreover, none of the points
of intersection of [pq] with IC0 are on [ab].

Fia. 15
§3] th e o r e m : a sim ple a r c d oes n o t d isc o n n e c t t h e p la n e 59

3.31. I f th e a d m i s s i b l e b r o k e n l i n e [pq] i n t e r s e c t s Kq , i t c a n be r e p l a c e d
b y a n a d m i s s i b l e b r o k e n l i n e [p q ]', th e n u m b e r o f w h o s e p o i n t s o f i n t e r s e c t i o n
w ith K q i s le s s t h a n t h a t o f [pq] w i t h K 0 .
Proof of 3.31. Among the points of intersection of the oriented broken
line [pqP with K 0 let us distinguish between “entrance points” (points
at which [pq] enters the interior of the circle K) and “exit points” (points
at which [pq]'* leaves the circle K ).
Let pi be the first entrance point on [pq]". It lies on some interval X
contiguous to K q n [ab] (Fig. 14). Since the endpoints d and d' of X are
points of intersection of K q with [ab] and c is the last point of K q n [ab]
on [ab]", it follows that every point of (dd1) " C [ab] precedes c on [ab]".
Hence the broken line [pq] has no p o in ts i n com m on w ith [dd1] C [ob]. Let
x be a variable point on [pq]". Since pi (f C24([24]) = [dd'] u X', u(C24, x)
does not change as x passes through pL. The function w(C0, x ), on the
other hand, obviously jumps by + 1 as x passes through pi or, in general,
through any entrance point situated on X. In consequence of (3.31'),
o)(C°2, x) also jumps by + 1 as a; passes through an entrance point situated
on X.
Since co(C°2, q ) = 0, a point moving on [ p q ] " from p i to q must necessarily
meet points of intersection of [ p q ] with C°2([02]) = [d'd] u X, i.e., points of
intersection of [p q ] with X; among these, in virtue of what has just been
proved, there must be exit points. Let qi be the last exit point on [ p q ] "
contained in X. Let us take two points p \ and q \ on [p#]”’ very near to
Pi and q i , respectively, and such that p \ precedes pi on [ p q ] " and q'\
follows qi on [ p q ] ". If p \ and q \ are taken sufficiently close to pL and ,
they will both obviously be in the exterior of K. Since [pi<7i] C X has a
positive distance from [ab], then (still on the assumption that p \ and q \
have been taken sufficiently near pL and <p) there exists a broken line
[ p \ q ' \ \ contained entirely in the exterior of K, passing very near to [pi<7i],
and having no points in common with [ab].
Let us now take the broken line
[PP'i] u [ p 'iq 'i] u [q \q ],

consisting of the piece [pp'i] of [pq], of the broken line [ p \ q ' i ] just con­
structed, and of the piece [<7'i<?] of [pq].
The resulting broken line [pp'i] u [p'l^'i] u [q\q] is obviously admissible;
the number of its points of intersection with K q is at least two less than
the number of points of intersection of the initial broken line [pq] with K q .
This proves 3.31 and also all of Theorem 3.3, since application of 3.31 a
finite number of times to any admissible broken line [pq] yields, in the
end, an admissible broken line joining p and q and having no point of
intersection with Kq . Such a broken line joining p and q does not have,
as vTe know, points in common with [ab], which was to be proved.
GO THE JORDAN THEOREM [CH. II

§4. Proof of the Jordan theorem

§4.1. Fundamental auxiliary construction. Let s be an arbitrary point
of the plane in the complement of the curve <f>. The point s is to be regarded
as fixed during the entire course of this article.
From the point s let us draw two distinct rays sa' and sb', where a1, b'
are points of the curve. (Rays will be designated by two letters not enclosed
in brackets. The first letter denotes the initial point of the ray, the second
another point of the ray.) Define the sense on each of these rays from s to
a' and from s to b', respectively. Then let a be the first point of intersec­
tion of sa' with <f> and let b be the first point of intersection of sb' with 4>.
Let us take a definite orientation on 4>; we shall agree that this is the
positive sense on 4>. Of the two arcs into which is separated by the points
a and b, denote the one on which the sense from a to b is positive by Ai ,
and the other by A2 .
Now let«i>i^ be the oriented Jordan curve consisting of the arc Ai~ =
(ab) and of the broken line [bsa] and let <f>2^ be the oriented Jordan
curve consisting of the arc A2-* = (ba)~ and of the broken line [asb]N
The conditions of 2.43 are satisfied, so that
(4.11) «(p,4>r) -f w(p, 4 0 = 0,(p,«O
for any point p in the complement of u [asb].
The function u(x, $ 0 is defined and constant for j; f A2. -Denote this
constant value by on . In exactly the same way, u(x, 4>2~) is constant for
x £ Ai ; denote its value for x £ Ai by 0 2 . We shall prove the following
fundamental lemma:
4.11. I f at least one of the inequalities
(4.12) o(p, 4>i ") wi , w(p, <t>2~) ^ a>2,
holds for a point p £ R2 \ 4>, then p and s can be joined by a broken line in
R2 \ < i>.
Proof. Suppose, e.g., that

(4.121) co(p, 4>D 7* wi .

In consequence of the fundamental theorem of §3, there exists a broken
line [ps] in R2 \ Ai . Let pi be the first point on [ps] ”* contained in [asb].
Then there is no point of 4>i on the piece (ppO - of [ps], so that u(x, <t>i~)
is defined and constant for all x £ (ppi). Since, by (4.121), the value of
u(x, 4>i""*) for all x £ [ppj] is different from an , i.e., from the value of the
same function on A2 , it follows that no point of [ppi] is a point of A2 . In
other words, the broken line [ppi], which does not intersect the arc Ai ,
does not intersect A2 either. This means that it does not intersect 4>. Since
P i £ [asb] and is different from the points a, b, it can be connected with s
§4] PROOF OF THE JORDAN THEOREM 61

by the segment [pis] contained in [asb]. The broken line [ppi] u [p^s] joins
p and s in R~ \ 4>. This proves the lemma.
§4.2. The case w(s, 4>") = 0; the set r = R2\ $ consists of at least two
components. Let us draw a straight line D which does not intersect 4>,
so th at <f>is contained in one of the two half-planes into which R 2 is separated
by Z); let s be an arbitrary point of the half-plane which contains no points
of 4>. The point s will remain fixed for the remainder of this article (Fig. 16).
By 2.22,
(4.21) «(«,$O=0.
Under the conditions assumed for s and using the notation of the preceding
article, we shall prove
4.21. One of the numbers wj , a>2 is equal to zero, and the other is ± 1 .
Proof. Let o0 and bo be the points of intersection of the straight line I)
with the rays sa and sb, respectively. Let us draw from s a ray sq0 inter­
secting D in some point qo between o0 and b0. Let q be a point on sqo
between s and g0 • Denote by $ 3”* the oriented Jordan curve consisting of
the arc A U = (ab) C 4> and the broken line [bb0aod\ T Finally, let us
denote by A-* the triangular contour (a0l>os) oriented in the sense (a0b0s)T
Then, by Theorem 2.43,
a>(q, 4>i~) = w(g, 4 0 + w(?, A^).
62 THE JORDAN THEOREM [CH. II

Since
co(g, A”*) = ± 1 , co(g, <?>3 ) = 0,
it follows that
(4.211) w(g, 4>i_>) = ± 1 .
Similarly,
(4.212) co(g, 4 0 = ±1.
As x goes to infinity along the ray sq, co(x, ^ i ”*) and co(x, 4 * 0 will
eventually become, and ever after remain, 0:
co(x, 4 * 0 = 0, co(x, 4>2“*) = 0.
Therefore the ray sq intersects both arcs Ai and A2 .
Let Ai be the first of the two arcs A'i and A2 which is met by sq as x goes
to infinity on sq. We shall show that then
(4.213) co2 = ± 1 , co! = 0.
(This will also prove 4.21. If A2 had been the first arc met by sq, we would
have had co2 = 0, coi = ± 1.)
Let q1 be the first point of the ray sq on the arc Ax ; the segment [ggi]
has no points in common with 4>2, so that co(x, 4>2~>) is constant for points
of this segment. Since co(x, 4>2~’) = ± 1 , co(gi, 4>2~) = ± 1 . Since gi 6 Ai ,
(4.214) co2 = ± 1 .
In particular, if q \ is the last point of sq on Ai , then
(4.215) co(g;i , 4 0 = ±1.
Because of this and the fact that co(x, 4>2~>) = 0 for all points of sq
sufficiently far from the curve 4>, it follows that a point moving along sq
from q \ to infinity will certainly meet the curve 4>2, and consequently
the arc A2. Let g2 be the first point of sq after q \ on A2 . Since g \ is the
last point of Ax on sq, co(x, 4>i”*) is constant on the entire extent of sq from
g'i to infinity, and this constant value of co(x, 4>i~) is zero. Hence

co(g2,4 * 0 = 0.
Since g2 6 A2 , the last equation implies that:
(4.216) co! = 0.
This proves (4.213) and thus 4.21.
Rem ark. Since co(x, 4>i"‘) = 0 for all points x of sq after q \ , it follows,
§4] PROOF OF THE JORDAN THEOREM 63

in particular, that

(4.217) «(512>$ r ) = 0
for any point qn of sq between q\ and g2 •
We shall now show that the curve <t> separates the plane into at least
two components. Since oi(p, = 0 for all p £ R2 \ sufficiently far
from 4>, it suffices, by Theorem 2.21, to exhibit a point p £ R2 \ 4> for
which oo(p, 4>~) 5^ 0. We shall show that, e.g., qn is such a point. Indeed,
since the segment [g'igi2] has no points in common with A2, and therefore
with 4>2 ,
w(?i2, 4>2 ) = w(gh , 4>2 ).
Consequently, by (4.215),
(4.218) w(gi2, 4>2 ) = ± 1 .
(4.218) , (4.217), and (4.11) imply
(4.219) w(gi2, 4 0 = ± 1 , q.e.d.
We shall deduce the following proposition from 4.11 and 4.21:
4.22. Every p £ R2 \ such that
(4.220) u(p,O = 0
can be joined to s by a broken line in R2 \ <f>.
Proof. (4.11) and (4.220) imply that
w(p, S O + u(p, S O = 0.
This and 4.21 imply that at least one of the inequalities
0)(p, 4>i^) Oil, Oi{p,$-T) 9^ 012
must hold. Therefore 4.22 follows from 4.11.
Since oi(x, <!>”') is constant for all x £ [sp] (Z R2 \ the set of all points
satisfying (4.220) is connected; it is, moreover, obviously open and hence
is a domain Ttf C R2 \ 4>. It follows from 2.21, furthermore, that r 0 is a
component of R \ 4>.
§4.3. Conclusion of the proof of the Jordan theorem. We have there­
fore proved:
4.31. The set R2 \ 4 > has at least two components and, moreover, the set of
all p £ R2 \ such that
(4.31) oi{p, O = 0
is a component of R 2 \ <h.
64 THE JORDAN THEOREM [CH. II

I t remains to be proved that:

4.32. The set of all p £ R2 \ 4>for which
(4.32) a>(p, 4 0 5^ 0
is the second component of R2 \ 4>.
To prove this, it suffices to prove in turn that:
4.320. I f s, p £ R2 \ 4> satisfy the conditions
(4.320) oj(s, 4>~) ^ 0, oj(p, 4>~) ^ 0,
there exists a broken line [sp] C R2 \ 4> joining p and s.
Proof of 4.320. We shall regard the point s as fixed for the rest of this
article, and make the same construction as in 4.1. Let us now show that,
under the present conditions and with the notation of 4.1,
(4.321) 0>! = w2 = 0.
Let Q be any circle containing the curves 4>, 4>i, 4>2 in its interior and let
so be any point in the exterior of Q. Then
(4.322) u(s 0 , O * = 0 ,
(4.323) oj(su ,4 * 0 = 0,
(4.324) co(s0 ,4>2"*) = 0.
Let [sos] be a broken line joining so and s in R2 \ Ai (such a broken line
exists according to §3). Since u(x, 4>~*) assumes different values for x = so
and x = s, [sos] intersects 4> and consequently (since it does not intersect
Aj) has points in common with A2. Let c2 be the first point of A2 on [sos]
Then oj(x , 4>~) is constant on (s<jc2) C [sos] and is equal to a>(so, 4>~*). On
the other hand, o>(x, 4>^) is also constant on (sa)~‘ and (sb) and takes
the value w(s, 4>"<) ^ 0. It follows that [soC2] has no points in common
with [asi>] and, therefore, none with the curve 4>i. Therefore, u(x, 4>]' 1) is
constant on (s0c2) and, by (4.323), is equal to zero. Since c2 £ A2, wi = 0.
It can be shown in exactly the same way that w2 = 0.
4.320 follows from (4.321) without any difficulty. Indeed, in consequence
of (4.32), (4.321), and (4.11), at least one of the inequalities
oj(p, 4>1~*) 5^ oji , Q>(p, 4>”*) ^ w2
must hold. Then 4.320 follows from the fundamental lemma 4.11. This
completes the proof of the Jordan theorem.
4.31, 4.32, and (4.218) imply
4.3. I f 4> is a Jordan curve and To, T are the two domains into which it
separates the plane, then a>(p, 4>~*) assumes the same value for every point p
of each of these domains. This value is zero for the exterior (i.e., unbounded)
domain and ± 1 for the interior (i.e., bounded) domain.
 Chapter I I I
SU R F A C E S

In this chapter we shall classify topologically the simplest two-dimen­

sional figures, the closed surfaces.
The exposition of the fundamental properties of these surfaces is pre­
ceded by a brief survey of the topology o’f elementary curves (§1).
Closed surfaces, as well as polyhedra and two-dimensional complexes—
triangulations and their subcomplexes—are defined in §2. §2 also treats
the combinatorial (Euler) characteristic of triangulations of closed surfaces
and introduces the notion of a two-dimensional skeleton complex, by means
of which the identification of the elements of a complex is rigorously defined.
§3 is devoted to a rigorous presentation of all definitions having to do
with cuts and identifications, and to a proof of the relevant elementary
theorems on which the further development (§§5-7) rests to a considerable
extent.
§4 and §5 treat the orientability and connectivity of surfaces. The normal
forms of simple (schlichtartige) surfaces and of closed surfaces are derived
in §6 and §7, respectively. The method used here to derive the normal
forms of closed surfaces is due basically to Alexander (see Alexander [a]).
The need of acquainting the reader with the elementary and geometrically
intuitive material has led to some looseness in the exposition of this chapter.
In contrast with the rest of the book, a great deal of the material in Chapter
III depends on certain invariance theorems, which are not proved until
Chapters V and X and are merely stated in the present chapter.
This procedure seems to me legitimate, since the proofs of these invari­
ance theorems given subsequently are indepenent of the results of Chapter
III. Otherwise, the exposition of the topology of surfaces would have to be
postponed until much later, or burdened with a proof of invariance for
the special case of two dimensions.
Still greater liberty is taken in the proof of the fundamental theorem
of the theory of surfaces itself and in the constructions immediately pre­
ceding it. For the rest, since all the fundamental concepts pertaining to
cuts and identifications have, I hope, been presented with impeccable
rigor, I thought it important to emphasize the applications of these ideas
and to leave to the reader the details of the proofs as exercises. These are
routine (like calculations in classical analysis) because of the completeness
of the definitions; to have included them in the book would only have
served to encumber the exposition and diminish its clarity.
65
6G SURFACES [CH. Ill

§1. Elementary curves and 1-complexes

§1.1. Elementary curves and their subdivision into arcs.
D e f in it io n 1.11. A compactum which is the union of a finite number
of simple arcs having by pairs at most a finite number of points in common
is called an elementary curve. A point p of an elementary curve is said to
be regular if it has a neighborhood relative to <t> which is an open arc, i.e.,
a homeomorph of the straight line. (An open arc is a simple arc without its
two endpoints; an open arc is homeomorphic to the real line, but not every
set homeomorphic to the line is an open arc: for example, the graph of the
function y = sin \/x , 0 < x < l/V, is homeomorphic to the line, but is
not an open arc. However, every subset of an elementary curve homeo­
morphic to the line is either an open arc or a simple closed curve with
one point deleted.) In the contrary case, the point p is said to be singular.
Singular points are, in turn, divided into branch points and endpoints. A
branch point is the common endpoint of three or more simple arcs of <£
which have no points in common except for p. A point p 6 $ is called an
endpoint if it has a neighborhood relative to whose closure is a simple
arc with one endpoint at p. The set of singular points of an elementary
curve is finite.
Let the elementary curve d> be the union of the simple arcs L \ , • • • , Ls .
Subdividing these simple arcs, if necessary, into a finite number of smaller
simple arcs, we may assume, without loss of generality, that each point
contained in more than one of the arcs L i , • •. • , L s is a common endpoint
of all the arcs which contain it and that every two arcs have no more than
one point in common. Let ai , • • • , an be all the points which are endpoints
of at least one of the arcs L \ , • • • , L „. The sequence a i , • • • , an contains
all the singular points of <f>, but not all the points ai , • • • , an are necessar­
ily singular. Let 6i , • • ■, bn be n points in
the three-dimensional space, no four of which
lie in a single plane; then, in particular, no
bz bi
three will lie on one line. If a,- and ay are the
endpoints of an arc Lk , let u§ join 6, and 6,
by a segment L 'h . Two such segments L \
and L'k either have no points in common or
have a common endpoint corresponding to
the common endpoint of the arcs L h and
Lk .
I t follows th at the curve 4> is homeomor­
phic to a polygonal line, i.e., to a line which
is the union of a finite number of segments
having by pairs no common points except,
perhaps, for common endpoints (Fig. 17).
§1] ELEMENTARY CURVES AND 1-COMPLEXES 67

Hence
1.12. Every elementary curve is homcomorphic to a polygonal line in R?.
Therefore, the study of elementary curves is reduced to that of polygonal
lines in R 3.
Let p be any regular point of a polygonal line <£. It is either an interior
point of exactly one segment or the common endpoint of exactly two seg­
ments, i.e., it is, in any case, an interior point of some simple arc of $ pon-
sisting entirely of regular points. This simple arc can be extended in both
directions until a singular point is reached or until a simple closed polygon
is described. I t follows easily from this that:
1.13. Every regular point p of an elementary curve <f>is an interior point
of a uniquely determined curve <i>p Cl $ which is:
either a simple closed curve, the component (I, 3.2) of p in <t>;
or a simple closed curve, containing exactly one singular point of <£;
or a simple arc, both of whose endpoints are singular points of <t>.
The curve 4>p is called the regular component of p.
A regular component of $ contains no singular points if, and only if, it
is both a simple closed curve and a component of $. Every other regular
component contains one singular point if it is closed, and two if it is a
simple arc.
D e f in it io n 1.14. The difference between the number of singular points
of an elementary curve and the number of those of its regular components
which contain singular points is called the {Euler) characteristic of the curve.
Since a topological mapping of an elementary curve preserves singular
points and regular components, two homeomorphic curves have the same
characteristic. It is also clear that the characteristic of a simple arc is 1,
and th at of a simple closed curve is 0. The characteristic of a lemniscate
is —1.
1.15. Let an elementary curve be decomposed into mutually non-
intersecting open arcs and their endpoints; we shall call the resulting open
arcs and their endpoints the one-dimensional and zero-dimensional ele­
ments (1-elements and 0-elements), respectively, of the decomposition of
the curve <t>. This decomposition is referred to as a one-dimensional complex
(1-complex) if no two open arcs have two endpoints in common.
Let us denote the number of 0-elements of the complex K by po and the
number of 1-elements by pi . Then the number po — pi (the Euler char­
acteristic of the complex K ) is equal to the Euler characteristic of the curve <h.
Indeed, replacing two arcs {ah) and (fee), whose common vertex fe is a
regular point of the curve $, by the single arc (oc), decreases p0 and pi by 1,
so th at po — pi is unchanged. Repeating this process of “obliterating” the
0-elements of the complex a finite number of times, we finally arrive at
the regular components, whence 1.15 follows without difficulty.
68 SURF.A CES fCH. I l l

Obviously:
1.16. The Euler characteristic of a curve <i> is equal to the sum of the Euler
characteristics of its components.
§1.2. The connectivity of a curve (the one-dimensional Betti
number). Let us first make the following preliminary remark. Let
M i , • • • , M , be a finite number of sets composed of perfectly arbitrary
elements. The set M = M i + • • • + M„ consisting of those, and only
those, elements each of which is contained in an odd number of the sets
M i , • • • , M s is called the sum (mod 2) of these sets. If there are no such
elements, the sum (mod 2) of the sets M i , • ■• , M s is zero (i.e., the empty
set):
Mi + • • • + M s = 0.
Now let be an elementary curve. Let us consider the following sets
whose elements are regular components of <t>:
1. Sets consisting of a single element, viz., of a regular component which
is a simple closed curve.
2. Sets of regular components whose union is a simple closed curve.
3. The empty set.
These sets of regular components are called simple cycles.
The sum (mod 2) of a number of simple cycles is called a cycle.
The set of all cycles of a curve <t> forms a group if the group operation
is taken to be addition (mod 2). (In this connection, -—z = z for any cycle z,
since z + z = 0.)
The cycles
(1-211) Zi, • • • , 2„

are linearly dependent if a subsequence of (1.211), say Ziy , • • • , z in (all

the ik distinct), has a sum (mod 2) equal to zero. [Instead of this we could
say: if a linear combination of the cycles of (1.211), not all of whose co­
efficients = 0 (mod 2), is equal to zero.]
In the contrary case, the cycles (1.211) are said to be linearly independent.
D e f i n i t i o n 1.21. The connectivity or one-dimensional Betti number it1(3?)
of an elementary curve 3> is the maximum number n such that $ has a
system of n linearly independent cycles.
The number -jt1('f>) could also, obviously, be defined as the greatest number
of linearly independent simple cycles in T.
R emark . Let K be any decomposition of the elementary curve $ into
open arcs and their endpoints (“vertices”).
Let us call every set of 1-elements of K which together with their vertices
form a simple closed curve in $ a simple cycle of the decomposition K; a
§1] ELEMENTARY CURVES AND 1-COMPLEXES 69

sum. (mod 2) of a number of simple cycles we shall call, as before, a cycle.

It is then clear that the cycles of K are in (1—1) correspondence with the
cycles of the curve 4>, so that their respective groups are isomorphic; con­
sequently, the connectivity of <f> is identical with the connectivity of the
complex K , i.e., with the maximum number of linearly independent cycles
of K.
Obviously:
1.22. The connectivity of a curve $ is equal to the sum of the connectivi­
ties of the components of <t>.
1.23. If an elementary c u rv e d is a subset of an elementary curved, then

x'(<f>o) < X*(<!>).

We shall prove the following fundamental proposition, known as the
Euler theorem (for curves).
1.24. The Euler characteristic of an elementary curve is equal to the difference
between the number of its components and its connectivity.
Or, denoting by p0 and pi the number of 0- and 1-elements of any de­
composition K of the curve 4>, by 7r° the number of components of 4>, and
by x1 its one-dimensional Betti number, we have
(1.24) po — pi = 7r° — x1.

For a connected curve, (1.24) becomes

(1.241) Po — pi = 1 — x1.
It suffices to prove (1.241). Then (1.24) follows for an arbitrary curve by
1.16 and 1.22.
Therefore let $ be a connected curve. Without loss of generality, wc
may assume that 4> is a polygonal curve with the given decomposition K
into straight line segments.
We shall reconstruct the given curve 4> in the following way: we shall
begin the construction with an arbitrary segment and then add new seg­
ments one after another, so that at any time the curve already constructed
is connected. This is clearly always possible. With each segment we shall
also add those vertices of this segment which have not yet been added
previously, if such vertices exist.
The addition of new segments may be of two kinds:
1. An addition of the first kind take place when, up to the addition, the
figure already constructed contains only one of the two vertices of the
segment to be added.
2. The addition is of the second kind if both of the vertices of the seg­
ment to be added have been added previously (Fig. 18).
70 SURFACES [CH. I l l

It is easily seen that po — pi and 7r1 are invariant under an addition of

the first kind.
An addition of the second kind obviously diminishes p0 — pi by 1. We
will show that an addition of this kind augments t by 1, so that 1 — t
decreases by 1; this will also prove (1.241) completely: at the beginning of
the construction, i.e., for a curve consisting of a single closed segment, both
sides of (1.241) were equal to 1, so that the formula was valid then; addition
of a new segment did not disturb either side of (1.241), so that (1.241)
remained true. Thus it remains to be shown that an addition of the second
kind augments ir1 by 1.

Let the given addition consist in adding to the connected curve 4>o the
open segment (ab), whose vertices a and b already belong to 4>0 . Since <t>0
is connected, a and b can be joined by a broken line in 4>0. This line, to­
gether with (ab), forms a cycle 20 in <t>o • Now let the connectivity of 4>0 be
7^0 = r, and let

Z\ , • , zT

be a system of r linearly independent cycles of 4>0. Since Zo contains the

segment (ab), which is not in any of the cycles z\ , • • • , zT , the cycles
,
z o , Zy , • • • z T are also linearly independent. It remains to be proved that
the cycles
z, Z0 , Zi , ••• , Zr ,
where z is an arbitrary cycle of the decomposition of 4>o u (ab), are linearly
dependent. This is true by hypothesis, if the cycle 2 is in 4>0. If 2 is not in
4>o, it contains the segment (ab). Then the cycle 2 z0 is in <h0 and conse­
quently is a sum of certain of the cycles z\ , • • • , z T , e.g., 2 + z 0 =
Z\ + • • • + zk . Then 2 + 20 dr 21 -T ■• • + zk = 0, which was to be
proved.
R e m a r k . The number of additions of the second kind is independent of
the choice of the decomposition K of the connected elemental’}'' curve
and of the order in which the segments of K are put together in the con-
§2] SURFACES AND THEIR TRIANGULATIONS 71

struction of $, as long as the curve is connected at every stage of the con­

struction. The number of such additions is equal to the connectivity of
the curve.
Indeed, the assertion is obviously true for a curve consisting of one
segment. Since the number t 1 increases by 1 for every addition of the
second kind and does not change for an addition of the first kind, the
assertion is true in general.

§2. Surfaces and their triangulations

§2.1. 2-complexes and polyhedra. By a triangle we shall mean the in­
terior of a triangle; by a segment or edge, an open segment (a segment with­
out its endpoints).
D e f i n i t i o n 2.11. A finite set K whose elements are triangles, segments,
and individual points of R n is called a triangulation if the following condi­
tions are satisfied:
1. No two elements of K have points in common.
2. All the sides and vertices of any triangle of K and both vertices of
every segment of K are elements of K.
In this chapter, “complex” will refer either to a triangulation or to any
subset of a triangulation. The maximum dimension of the elements of a
complex is called the dimension of the complex (in our case 0, 1, or 2).
The point set union in the given R n of the points of all the elements of a
complex K is called the body of the complex K and is denoted by || K ||. The
body of a triangulation is called a polyhedron. If the polyhedron is the body
of a triangulation K , 3> = || K ||, we shall say that K is a triangulation
of the polyhedron <f>.
It is easily seen that polyhedra are bounded closed subsets of a given
R n, th at is, polyhedra are compacta.
The elements of a complex K may be partially ordered as follows. Let
T £ K . The vertices and sides of T, if T is a triangle, and the vertices of T,
if T is a segment, shall precede T. If T' precedes T, we shall say that T
follows T '.
A subcomplex Ko of a complex K is said to be closed {open) if every ele­
ment of K which precedes (follows) an element of Ko is itself an element
of K o .
Obviously:
2.11. A closed subcomplex of a triangulation is a triangulalion.
D e f i n i t i o n 2.12. A complex is said to be connected if it cannot be ex­
pressed as the union of two nonempty disjoint closed'subcomplexes.
' T h e o r e m 2.13. In order that a triangulation be connected, it is necessary
and sufficient that it be possible to join any two of its vertices by a broken line
consisting of elements of the triangulation.
Proof of necessity. Let a and b be two vertices of the triangulation K
72 SURFACES [CH. I l l

which cannot be joined by a broken line in K . Let K a be the set of all those
elements of K of which at least one (and consequently every) vertex can be
joined by a broken line to a; denote by Kb the remaining elements of K.
It is easily seen that K a and Kb are closed subcomplexes of K ; they are non-
vacuous, since a £ K a , b £ Kb . This contradicts the hypothesis that K is
connected.
Proof of sufficiency. If K is not connected and
K = K in K 2, K i n K 2 = 0,

where K i and K 2 are closed subcomplexes of K , no vertex of K i can be

joined to any vertex of K 2 by a broken line, which was to be proved.
If a polyhedron •is connected, all its 'triangulations are connected (a
partition of a triangulation into two nonempty disjoint closed subcomplexes
corresponds to a partition of its body into two
nonempty disjoint closed sets). Conversely, if a tri­
angulation is connected, its body is also connected: it
follows easily from 2.13 that every two points of
|| K || can be joined by a broken line in || K ||
(Fig. 19). Thus:
2.14. I f at least one triangulation of a polyhedron
is connected, the polyhedron is connected; i f a, poly­
hedron is connected, all its triangulations are con­
nected.
A two-dimensional complex (2-complex) is said to
be pure (rein) if every 1 - and 0 -element of the com-
pIQ> 19 plex precedes some 2-element. A pure 2-complex K is
said to be strongly connected if every two triangles T i
and T, of K can be connected by a chain of triangles

T i , •• • , T.

such that Ti and T l+i, i = 1, • • • , s — 1, have a common side in K .

Finally, let us introduce the following definition:
Let T £ K . The subcomplex consisting of T and of all elements of the
complex K which follow T is called the star OkT of T in K.
Thus, if T is a triangle, OkT = T; if T is a 1-element, OkT consists of
T and of all the triangles of K which have T as a side; if T is a vertex,
Ok T consists of T and of all 1- and 2-elements of K having T as a vertex.
Examples of all these cases are given in Fig. 20.
Let us consider the special case that T is a vertex e of a triangulation K.
Then the star 0 Ke consists of the triangles of the form (eaia2), of the seg­
ments of the form (ea), and of the vertex e. The sides ( a ^ ) of these tri-
§ 2] SURFACES AND THEIR TRIANGULATIONS 73

angles (opposite the vertex e) and the vertices a of the segments (ea) make
up a complex B Ke called the outer boundary of the star 0 Ke (Fig. 20).
A star 0 Ke is said to be cyclic if its outer boundary is a simple closed
broken line (i.e., if its elements are disposed in cyclic order, like the ele­
ments of a circle split up into sectors). If the outer boundary of a star 0 Ke
is a simple broken line which is not closed, the elements of the star have a
natural linear order and the star is said to be semi-cyclic.
For an example of a semi-cyclic star see Fig. 21.
§2.2. Closed surfaces. The following theorem will neither be proved nor
used in this book: Every closed (topological) 2-manifold (I, 5.3) is homeo-
morphic to some polyhedron (see Gawehn [a]). In order not to depend on
this theorem, we shall at the very beginning restrict ourselves to those
closed 2-manifolds which are homeomorphic to polyhedra and simply call
them closed surfaces. The fact of the matter is that the class of closed sur­
faces defined in this way is synonymous with the class of all closed
2-manifolds.
If a triangulation K of a polyhedron^ homeomorphic to a given closed
surface <f>, and a definite topological mapping C of the polyhedron <f>0 onto
the surface are chosen, the triangulation K will be mapped onto a “curved
triangulation” of the surface <f>, consisting of curvilinear triangles, their
sides, and vertices. For example, if an octahedron or an icosahedron in­
scribed in a sphere is centrally projected onto the sphere, the sphere is
decomposed, in the first case into 8, in the second case into 20, spherical
triangles; these triangles, their sides, and vertices are the elements of the
corresponding curved triangulation of the sphere. In exactly the same way,
the triangulation of the polyhedron (Fig. 22) homeomorphic to a torus
goes over into a curved triangulation of the torus under a topological
mapping.
R e m a r k . Since every closed surface is homeomorphic to a polyhedron,
the topological study of surfaces can be confined to those surfaces which
are themselves polyhedra. We shall, therefore, without further reservations,
assume in the sequel that all surfaces considered are polyhedra. This will,
in particular, enable us to avoid completely the use of curved triangulations.
We shall accept without proof the following theorem, which will be
proved in the sequel (V, 3.140).
2.20. Let A and B be two subsets of a topological space R, each of which is
homeomorphic to the plane. Then if A Cl B, A is open in B.
We can now prove the following fundamental theorem:
2.21. In order that a triangulation K be a triangulation of a closed surface,
it is necessary and sufficient that K be connected and that the star of every
vertex of K be cyclic.
Proof of necessity. Let K be a triangulation of the closed surface 4> = || K ||.
By 2.14, the connectedness of K follows from that of 4>. Before proving that
every star 0 Ke, where e i§ a vertex of K , is cyclic, we shall prove that every
segment T 1of K is a side of more than one triangle. This assertion obviously
follows from:
2.210. Let T be a plane triangle, let T be the closure of T in R2, and let
p 6 T \ T . Then there is no set T homeomorphic to the plane such that

p e r C T.

Indeed, suppose that such a set T exists. Then T and Rf satisfy the con­
ditions on the sets A and B, respectively, of Theorem 2.20. Hence T is
§2] SURFACES AND THEIR TRIANGULATIONS 75

open in Rf and p is an interior point of r . Thus, a fortiori, p is an interior

point of T relative to R 2, which is obviously not so. This proves 2.210.
The fact th at every 1-element of K is a side of at least two triangles im­
mediately implies th at there is a cyclicly ordered sequence of triangles of
0 Ke such th at each pair of neighboring elements of the sequence has a
common side of the form (eef). These triangles, their common sides (eet),
and e make up a subcomplex O' of 0 Kc and it suffices to prove that O' = 0*e.
To this end, let r be a neighborhood of e relative to $ homeomorphic to R2.
Let us put
e = p(e, <t>\ T) > 0
and denote by d the maximum of the diameters of the elements of the
complex O'. Let 0 < p < t/d. A similitude with center e and coefficient p
takes the complex O' into a complex 0" such that e 6 || 0" || C r (see

Fig. 23 which shows the complex O', the intersection of its elements with
T, and the complex 0"). Both |] O' || and |] 0" ||, as is easily seen, are homeo­
morphic to the interior of a circle and, hence, to the plane. Therefore, the
sets || 0" || and T satisfy the conditions on the sets A and B of Theorem 2.20.
|| 0" || must be open in T. But if there were an element T of 0 Ke not in
O', this element would not intersect || O' || and, all the more, || 0" ||; on the
other hand, e is a limit point of T and hence cannot be an'interior point of
|| 0" || relative to T. This contradiction proves that O' = 0 Ke.
Proof of sufficiency. Since K is connected, || K || is also connected; we
shall prove th at the fact that every star 0 Ke is cyclic implies th at every
point p £ <f> is contained in an open subset of the polyhedron <t> homeo­
morphic to the plane or, what comes to the same, to the interior of a circle.
First, 0 Ke is, by definition, an open subcomplex of K and the comple-
76 SURFACES [CH. I ll

mentary subcomplex K \ 0 xe is a closed subcomplex of K. Hence

|| K \ 0 Ke ||, being a polyhedron, is closed in <f> = || K ||, while || 0 Ke || is
open in $. The set || 0 Ke || is obviously homeomorphic to the interior of a
circle. Now let p "be any point of 4> and T the unique element of K which
contains p. Let e be any vertex of T. Then p £ || 0 Ke ||. This completes
the proof.
2.22. Every l-element T l of a triangulation K of a closed surface is a side
of exactly two triangles of K.
Indeed, if e is a vertex of the segment T l, all the triangles of K adjoining
T' are elements of the cyclic star 0 xe; since 0 Ke contains precisely two
triangles with side T 1, 2.22 is proved.
§2.3. Surfaces with boundary. In this chapter, we shall understand
by a surface a compactum $ homeomorphic to a polyhedron and with the
following property. The points of 3> can be split into two classes: interior
points, having neighborhoods homeomorphic to the plane, and boundary
points-, in this connection, a point p £ <!> is called a boundary point if it has
a neighborhood which can be mapped topologically onto the union of a
triangle T2 and one of its sides T 1 in such a way that the mapping takes p
into a point of the open segment T l.
The set of all boundary points of a surface is called the boundary of the
surface. If it is nonempty, the surface is called a surface with boundary,
in the contrary case, the surface is closed in the sense of the preceding
article. In agreement until the notation of the preceding article, we will
always assume in the sequel that the surfaces considered are themselves
polyhedra.
2.210 implies
2.31. The division of the points of a surface into interior and boundary
points is, in fact, a decomposition into two disjoint classes: the boundary
points of a surface are characterized as those which do not have neighborhoods
homeomorphic to the plane; a topological mapping of one surface onto another
preserves interior and boundary points.
I t follows from the very definition of boundary point that every boundary
point of a surface is an interior point, relative to the boundary of the sur­
face, of some simple arc consisting entirely of boundary points. On the
other hand, for an arbitrary choice of a triangulation K of the surface 4> all
the boundary points of $ lie either on the 1- or on the 0-elements of K; it
follows from these two observations th at the boundary of a surface is an
elementary curve without singular points and, therefore, is the union of
simple closed curves, no two of which have points in common. These simple
closed curves, which are the components of the boundary of the surface,
are called the contours of the surface.
§2] SURFACES AND THEIR TRIANGULATIONS 77

Let p be a boundary point of the surface and K an arbitrary triangula­

tion of 4>. The point p is either contained in a segment T 1 of K, or is-a vertex
of K. In the first case, by 2.210, T 1 is a side of exactly one triangle of K and
all the points of T l arc boundary points. In the second case p, as a vertex
of K, is the common endpoint of exactly two segments (pei) and (pes) of
K on the boundary of <t>, each of which is a side of one triangle of K. Since
all the remaining elements of the star 0 Kp, except for the point p and the
segments (pei) and (pca), consist of interior points of the surface, every
segment of the form (pe,), 1 ^ i ^ «, is a side of two triangles of K. It
follows easily that every star OkP is semi-cyclic. Conversely, if the star of a
vertex e of K is semi-cyclic, e is a boundary point. Therefore,
2.32. Let K be any triangulation of a surface <f>. In order that a point p be
a boundary point it is necessary and sufficient that it satisfy one of the following
two conditions:
a) p is contained in a segment of K which is a side of only one triangle of K ;
b) p is a vertex of a semi-cyclic star of K.
Those elements (edges and vertices) of K which are on the boundary of
$ are called boundary elements of K. The remaining elements are called
interior elements. The boundary of <t>is the union of the boundary elements
of any triangulation of $>.
2.33. A triangulation of a surface is a strongly connected complex.
Indeed, let T \ and T2, be two triangles of K and suppose it impossible
to connect them by a chain of triangles
(2.33) T \ , • • • , T2S
such that T 2i and T2i+i have a common side. Let Ki be the subcomplex of
K consisting of the triangles which can be connected to T2\ by such chains
and of the sides and vertices of these triangles; let Kz be the subcomplex of
K consisting of all the remaining triangles, edges, and vertices of K. K\
and K 2 are disjoint, by definition. On the other hand, since cyclic and
semi-cyclic stars are strongly connected, the star of every vertex of Ki
(relative to K) is wholly contained in K x and the star of every vertex of Kz
(relative to K) is wholly contained in K 2. Therefore, Ki and K 2 are com­
plementary open, and hence closed, nonempty subcomplexes of K. This
contradicts the fact that K is connected.
§2.4. Subdivisions of triangulations. It is sometimes necessary to pass
from a given triangulation of a surface to a finer triangulation by sub­
dividing the original one. We shall consider only the following types of
subdivisions:
1. A n elementary subdivision of the triangulation of a surface relative to
a given 1-element T 1 of the triangulation consists in dividing T l into two
78 SURFACES [CH. I ll

F i g . 25

segments and then dividing each of the triangles adjoining T l into two
triangles, as in Fig. 24; all the remaining elements of the triangulation re­
main unchanged.
2. The result of performing an elementary subdivision relative to all the
1 -elements of a triangulation is a barycentric subdivision of the triangulation,
whose structure is clear from Fig. 25 (a barycentric subdivision of a tri­
angle consists of the six triangles into which the given triangle is divided
by its three medians).
2.41. A subdivision of a given triangulation, obtained as the result of a
finite number of elementary subdivisions, is said to be a regular subdivision.
In this book it will be necessary to consider only regular subdivisions.
It is easy to show that a barycentric subdivision is regular.
§2.6. Skeleton complexes. The set consisting of the three vertices of a
triangle is called the skeleton [in German: Gerust; sometimes called an
abstract simplex (see IV, Def. 1.41)] of the triangle; the set consisting of
the two vertices of a segment is called the skeleton of the segment; a vertex
is its own skeleton. The elements of a triangulation K correspond (1-1)
to their skeletons and one element of K precedes another if, and only if,
the skeleton of the first element is a proper subset of the skeleton of the
second element.
Hence the study of a triangulation can be successfully replaced by the
study of the set of all skeletons of the triangulation, that is, the skeleton
complex of the triangulation.
In this connection, we shall make the following definitions.
D e f in it io n 2.51. Let E be any set. The elements of E will be called
vertices. Let us assume that certain subsets of E, called skeletons, consisting
of one, two, or three vertices, have been singled out. The number of vertices
in a skeleton less 1 is said to be its dimension. In this chapter a set of skele­
tons will be called a skeleton complex if it satisfies the following conditions:
a) Every nonempty subset of a skeleton is a skeleton.
b) Every set consisting of one vertex is a skeleton.
§3 ] CUTS AND IDENTIFICATIONS 79

The maximum dimension (0, 1, or 2) of the elements of a skeleton com­

plex is called the dimension of the complex.
R e m a r k . Higher dimensional skeleton complexes will be introduced in
Chapter IV.
D e f in it io n 2.52. Two skeleton complexes K and K ' are said to be 'iso­
morphic if there is a (1- 1 ) mapping of the set of vertices of one complex
onto the set of vertices of the other complex which preserves skeletons.
A skeleton complex K and a triangulation K ' are said to be isomorphic
if K and the skeleton complex of K ' are isomorphic. Two triangulations
are isomorphic if their skeleton complexes are isomorphic.
The following important theorem is a special case of a theorem proved
in IV, 1.9:
2.53. Every two-dimensional skeleton complex is isomorphic to some tri­
angulation in R 5.

§3. Cuts and identifications

§3.1. Identification of elements in skeleton complexes. Let K be
a skeleton complex. Let us divide the set E of all vertices of K into mutually
disjoint subsets or classes E \ , • • • , E 's ; and let us assign a new “vertex”
e'i to each class E ' {, with the agreement that a pair or triple of vertices
e‘i shall form a skeleton if, and only if, the corresponding classes E 'i contain
vertices e,- forming a skeleton of K.
The skeletons defined in this way form a skeleton complex K '; the com­
plex K ' is said to he obtained from the complex K by identification of certain
vertices of K ; each vertex e \ of K ' is obtained by identifying all the vertices
of K contained in the class E 'i .
Now let K be any triangulation and let Ki be the skeleton complex ob­
tained by identifying certain vertices of the skeleton complex of the tri­
angulation K. Let K ' be any triangulation isomorphic to Ki (such triangula­
tions exist by virtue of Theorem 2.53).
The triangulation K ' is said to arise from the triangulation K through identi­
fication of certain elements of K.
An identification of the elements of a skeleton complex (or triangulation)
K obviously induces a mapping of the set of all vertices of the complex K
onto the set of all vertices of the complex K ': if e„ 6 E 'i , the vertex e'i is
assigned to the vertex en . This vertex mapping maps every skeleton of the
complex K onto a skeleton of the complex K ' and maps all of K onto K ' . A
mapping of this kind is called a simplicial mapping (for details see IV, 1 .6 )
of the triangulation K onto the triangulation K '. In this chapter we shall
consider only identifications which satisfy the following conditions'.
1. No two vertices which form a skeleton of K are mapped into a single
vertex of K '.
80 SURFACES [CH. Ill

2. Every triangle of K ' is the image of exactly one triangle of K.

3. Every segment of K ' is the image of at most two segments of K.
E x a m p l e s . 1. Consider the triangulation of the rectangle shown in Fig.
26. Let us divide the set of all eight vertices of this triangulation into the
following classes: each of the classes E 't-, i = 1, 2, 3, 4, consists of the one
vertex e» ; E \ consists of the vertices and e7, and £ " 6 of e6 and es . This
identification converts the triangulation K of the rectangle into a triangula­
tion K ' of the lateral surface of a triangular prism (Fig. 27). Both segments
and (e7e8) of K correspond to the same segment (e'ne'e) of K'.
Since the lateral surface of the prism is homeomorphic to the lateral
surface of a cylinder and also to a plane circular ring, we may say that the
identification just described converts the triangulation K of the rectangle
into a triangulation K ' of a cylinder or a plane ring.
2 . The following identification differs from the preceding one only in that
now the class E \ consists of the vertices e5 and e8 , and the class E \ of e$
and e7 . This identification converts the triangulation K of the rectangle
into a triangulation K ' of a surface known as a Mobius band (Fig. 28).
R e m a r k . We shall briefly refer to case 1 as an identification of the di­
rected sides (eB^) ”*and (e7e8) of the rectangle and to case 2 as an identifica­
tion of the directed sides { e ^ ) ^ and (e8e7)~\ We shall use analogous
terminology in other similar cases. A directed segment will be denoted by
parentheses followed by an arrow.
§3] CUTS AND IDENTIFICATIONS 81

3. Let us consider the triangulation of a rectangle shown in Fig. 29. Let

us divide its set of 19 vertices into classes E'i as follows: set E 'i = e i,
i = 1, 2, 3, 4, 17, 18, 19, and let
E't = {e~0 , eg,, eu , eu}, E'i — {ei ,en}, E'i — {ei, 612},

E's = {eg , eje}, E \ = [eio, eu}.

This identification of the directed side (eseu)” vnth the directed side (eaen)-*
and of (e5e8) ”* with (ei4en) ~* converts the original triangulation of the rec­
tangle into a triangulation of the torus (Fig. 30a).
4. The following identification differs from the preceding only in that now
E'i = {ei , 612), E '7 = {ei ,e\i],
i.e., (eseu) as before, is identified with (e8Cu) but (e5es) is now identified
with (cueu)"*. This identification yields a triangulation of a surface known
as a Klein bottle (Fig. 30b), which cannot be imbedded in three-dimensional
space.
5. Let us now set (Fig. 29)
E 'i = {e,j, i = 1, 2, 3, 4, 17, 18, 19; E'i = {e$, eu};
E'i = (e6, e^}; E '7 = {ei, en}; E ’%= {es, eu};
E'i = (e9 , ei5}; E \ 0 = {eio, eu}
[so th at the side (es^)-* is identified with the side (enCi2) ( e 6e7)~ with
(enen)^, • • • , (emeu)”* with (ei6e5)"’]- This identification converts the given
triangulation of the rectangle into a triangulation of the projective plane.
To show this, let us map the rectangle topologically onto the upper hemi­
sphere of the sphere x + i f + z = 1 in such a way that the images of the
82 SURFACES [CH. Ill

points e5 , e6 , • • • , eie divide the equator into 12 equal arcs. Then the
identification can be interpreted as the identification of each pair of dia­
metrically opposite points of the equator. But this (I, 5.2, 8 ) converts the
hemisphere into a topological space homeomorphic to the projective plane.
R emark . The stars of all the vertices of the above triangulation of the
projective plane are obviously cyclic stars. It follows th at the projective
plane is a closed surface.
§3.2. Cut lines and semi-stars of the vertices of cut lines. (Only
cut lines are discussed in this article; the cut operation is not introduced
until 3.3. Therefore, the doubling of the segments in the figures of 3.2
should be ignored; this doubling will have no significance until 3.3.) Let
<£ = || K || be a surface with a definite triangulation K. In this section, a
cut line will mean a simple broken line A consisting of elements of the
triangulation K and satisfying one of the fol­
lowing conditions:
1. A is a closed broken line, all of whose el­
ements are interior elements of the triangula­
tion K (i.e., elements not on the boundaiy of
the surface <f>). This is the case of interior
closed cuts (Fig. 31).
2. A is a simple nonclosed broken line, all of
whose elements are interior elements of K ex­
cept both endpoints of the broken line, which
are on the boundary of the surface (cross cut,
Fig. 32).
3. A is a simple broken line which is not closed and contains at least two
links. One of the endpoints of A may be on the boundary of the surface;
but none of the other elements of A are on the boundary (open cut, Fig. 33).
Let us consider any star 0 Ke whose center e is one of the vertices of a cut
§3] CUTS AND IDENTIFICATIONS 83

F ig. 34

line A; if A is an open cut line, let us assume in addition that the vertex e
is different from the interior endpoint (or endpoints) of the open cut (Fig.
34). Under these assumptions, the set of all elements of the star 0 Ke not
contained in A is the union of two disjoint connected complexes, the two
semi-stars of e; if one of these semi-stars is denoted by Oe+, the other will
be denoted by 0e~ (Fig. 34).
It is easily seen that:
3.21. If e and e'. are two consecutive vertices of A, each of the two semi­
stars of the vertex e meets precisely one of the semi-stars of the vertex e'
(the intersection in each case consists of a triangle having (ee') as a side).
Let us note further that:
If A is an open cut line, e an interior endpoint of A, and e' the vertex of
A adjacent to e, then 0 Ke intersects each of the semi-stars of the vertex
e' and each intersection is a triangle having (ee') as a side (Fig. 33).
I t follows that:
3.220. If [ei ■• • es\ is an open cut line, one of whose endpoints, say e\ ,
is on the boundary of the surface, the semi-stars of the vertices ei , • • • , es-i
and the star of the vertex es form a chain
Oe+ 1 , • • • , Oe+„_i , Ox.es , Oe s_i , • • • , Oe i .
[A chain (or closed chain) of subcomplexes of a given complex, here and
throughout this chapter, is to be taken in the sense of I, Def. 3.14.]
If both endpoints e\ and es of the open cut line are interior points, the
stars 0 Kei and 0 Kes together with the semi-stars of the remaining vertices
form one closed chain
Onei , Oe 2 > ■■■ >Oe «_i, Ok&s >Oe «_i , • ■■ , Oe 2 >O^e 1 .

3.220 in turn implies

3.22. The complement in a triangulation K of an open cut line is a strongly
connected open subcomplex of K.
84 SURFACES [CH. I ll

In fact, every chain of triangles connecting two triangles T i and T v of

K can always be replaced by a chain of triangles connecting T 1 and 7 „
along the circuit of the open cut line (Fig. 35).
R emark . A chain of triangles of a complex K connecting two triangles
7’i , T, 6 K will always mean a sequence of triangles 1 \ , • • • , T v of K
such that Ti and T i+i (1 < i < v - 1) have a common side in K.
Now let A = [ei • ■■e„] be a cross cut, and let 0e+1 be one of the two semi­
stars of the vertex ex . It intersects a completely determined semi-star of

Fig . 35

the vertex e2, which we shall denote by 0e+2 ; 0e+2 in turn intersects a defi­
nite semi-star of e3, 0e+3, etc. We obtain, in this way, a chain of semi-stars
(3.23+) 0e+1 , • • • , 0e+s .
Similarly, starting with the semi-star 0e~i , we obtain a chain of semi-stars
(3.23_) 0e~i , ■• • , 0e~s .
Hence
3.230. The semi-stars of the vertices of a cross cut form two chains, (3.23+.)
and (3.23_).
From 3.230 we shall derive
3.23. The complement in a triangidation K of a cross cut is an open sub­
complex of K which is either itself strongly connected, or is the union of two
disjoint strongly connected complexes. I f the endpoints of the cross cut A lie
on two different contours, K \ A is always a strongly connected complex.
To prove the first assertion, it suffices to show that an arbitrary triangle
Ti £ K can be connected with a triangle of one of the semi-stars of (3.23+)
or (3.23_) by a chain of triangles, consecutive members of which have a
common side not contained in A. To this end let T h be a triangle of, say,
§3] CUTS AND IDENTIFICATIONS 85

one of the semi-stars of (3.23+). Let us connect T\ and 7 \ by a chain of

triangles of K . If every pair of consecutive triangles of this chain has a
common side not contained in A, the proof is complete; in the contrary
case, let T i , 7\+i be the first pair of triangles of the chain with a common
side contained in A. Then Ti is contained in one of the semi-stars of (3.23+)
or (3.23_) and 1\ , ■■• , Ti is the desired chain.
Let us now pass to the second assertion of Theorem 3.23, i.e., if all the
elements of the cross cut A, whose endpoints belong to different contours
7n and 7T2 of the surface || K I], are deleted frorii K, the remainder is a
strongly connected complex.
This assertion is readily proved by the methods used above; it is merely
necessary to make the following very obvious remark. Under the given
conditions, all the semi-stars of the vertices of the cross cut A = [e3 ■• • e„]
together with the stars 0 Ke of all the vertices, different from e3 and es , of
the contours 7ri and 7r2 form a single closed chain (Fig. 40).
Finally, let
A = (ei • • • esei)
be an interior closed cut line. Let Oe+ 1 denote one of the two semi-stars of
e i , Oe+2 the unique semi-star of e2 which intersects Oe+ 1 , Oe+3 the unique
semi-star of e3 which meets Oe+2, etc., until es is reached. This yields a
chain of semi-stars
(3.241) Oe+1 , • • • , Oe+, .
Since es and ei are adjoining vertices, Oe+s meets a definite unique semi­
star of e\ . Then two cases are possible:
1 . Oe+, meets Oe+1 .
2 . Oe~, meets Oe~1 , where Oe~1 is different from Oe+1 .
In case 1 the chain (3.241) is closed; in this case, beginning with the semi­
star Oe~1 and reasoning as before, we obtain a second closed chain of semi­
stars
Oe~1 , • • • , Oe~a
and call the broken line A a closed two-sided ad line (Fig. 31a or Fig. 41).
In case 2 we obtain a closed chain
Oe+ 1 , • • • , Oe+„ , Oe~i ;
Oe~1 meets one definite semi-star of e2 , which differs from Oe+2 (since Oe+2
meets Oe+1), and hence is Oe~2 ; in exactly the same way, Oe~2 meets Oe~3 ,
but not Oe+3 . Continuing in this way, the previous chain of semi-stars is
extended to the chain
Oe+1 , Oe+2, • • • , Oe+„ , Oe~1 , • • • , Oe~, .
86 SURFACES [CH. Ill

The semi-star Oe~s intersects one of the two semi-stars of ey and since it
does not intersect 0e~i (Oe~y meets 0e+s), it intersects 0e + 1 . Hence, in
case 2 , every semi-star occurs in the closed chain
(3.242) Oe+i , • • • , 0e+s , 0 e ~i , • • • , 0e~e , 0e + 1 .
In case 2, the broken line is called a closed one-sided cut line (see Figs.
31b and 42a; the reference is to the “center line” of the Mobius.band which
can be seen in Fig. 31b; this is the line cccccc, drawn as a double line). A
slight change in the proof of 3.22 and 3.23 yields
3.24. The complement of a closed one-sided cut line in a triangulation K
is a strongly connected open subcomplex of K.
3.25. The complement in a triangulation K
of a closed two-sided cut line is either a
strongly connected open subcomplex of K
or the union of two disjoint strongly con­
nected open subcomplexes of K.
In the triangulation of the projective
plane (segments and vertices denoted by
the same symbols are to be identified) shown
in Fig. 36 the closed broken line (pyctfzeifiifitfy)
is a one-sided cut line. The two semi-stars of
ey are distinguished by the signs + and —,
respectively.
§3.3. The cut operation. Having disposed of the preliminary definitions,
propositions, and examples, we may now go on from cut lines to the cut
operation itself. This operation consists of a passage from a given triangula­
tion K to a new triangulation K ', which depends on the cut line A and is
defined only up to an isomorphism. It obviously suffices to define the
skeleton complex of the triangulation K'.
We shall first define the vertices of K'. If A is an open cut, the vertices of
K ' are: 1) all the vertices of K not contained in A and the interior endpoint
(or endpoints) of A; and 2) “new vertices”, c\-, corresponding (1-1) to all
the semi-stars of the vertices of A. If A is a cross cut or a closed interior
cu.t, the vertices of K ' are: 1) all the vertices of K not in A; and 2) new
vertices e'i , corresponding (1-1) to the semi-stars of the vertices of A.
This definition of the vertices of K ' implies that every new vertex of
K ' corresponds to a unique vertex Ci of A, i.e., to th at vertex a , one of
whose semi-stars corresponds to e'i ; it is clear, on the other hand, that eaeh
vertex e; of the broken line (with the exception of one or both endpoints,
if A is an open cut) is placed in correspondence with precisely two new
vertices e't- (the number of semi-stars of ct-). These two new vertices e\- which
§3] CUTS AND IDENTIFICATIONS 87

correspond to the same vertex e, of A will, as the need arises, be denoted

by (e'i)+ and (e'i)_ , or simply by e'i+ and e',-_ .
The elements of the skeleton complex K ' are defined as follows: 1 ) every
skeleton of K, all of whose vertices are in K ', is a skeleton of K'; 2) skeletons
of the form {e\-, e/} and {e\-, e;-, Ck) are in K ' if, and only if, the semi-
star corresponding to the vertex e\- contains {et-, ey} or [ a , e3- , ek), re­
spectively; 3) if A is an open cut, skeletons of the form {e'k , e*} are in K'
where e* is an interior endpoint of the open cut line and e* is the vertex of
A adjacent to it; 4) skeletons of the form [eri , e',}, {e\-, e 'j , e*}, and
{e'i, e'y, e'k] are in K ' if, and only if, the semi-stars corresponding to the
vertices e\-, c'y, and (in the last case) e'k intersect and (in the second case)
this intersection contains {e,-, e j , e*}.

It is immediately verified that the stars of all the vertices of K ' are cyclic
or semi-cyclic. Hence K ' is a triangulation of a surface or of several (in
virtue of 3.22-3.25, no more than two) disjoint surfaces.
It follows readily from the definition of the skeletons of the triangulation
K 1 and from the propositions proved in this article that:
3.31. An open cut one of whose endpoints ei lies on the boundary of the
surface transforms the broken line A = [ei ■■■es+i] into a simple lion-
closed broken line
[e'i+e'2+ ■■■ e's+es+ie's_e'(s_i)_ • • •
which is part of one of the contours of the surface K ' (Fig. 37).
K and K ' have the same number of contours.
An open cut A = [ei ■■■e„+i] both of whose endpoints are interior points
transforms the nonclosed broken line A into a closed broken line
(C iC 2 + C lC S— * * * 6 2—C l )

which appears as a new contour of K'. The number of contours of K ' is

one more than that of K (Fig. 38).
88 SURFACES [CH. Ill

3.32. A cross cut A = [e\ • • • es] replaces the nonclosed broken line A by
two nonclosed broken lines
[e'i+ • • • e',+] and [e\_ • • • e'._].
If both endpoints of the broken line are on the same contour of the surface
K, K ' may have one more contour than K (Fig. 39). Here the contour
(eidi • • • aM
e„6 i • • • byei) is replaced by the two contours
(e'l-di • • • a„e'»-e' ( « _ i ) _ • • • e^-e'i—)
and
(e'i+e'z+ ’ *• e>u-i) +e't+bi • • • bye\+).

F ig . 40
§3] CUTS AND IDENTIFICATIONS 89

B ut if one endpoint of the cross cut A lies on one contour and the other
endpoint on another contour, K ' has one less contour than K , so that the
two contours
(eidi • • • o^ei) and (eabi • ■- byea)
are replaced by the single contour (Fig. 40)

( c • • ■ df^c i— * *c s— b \ * ■ • byC * q i + ) .

3.33. A closed interior two-sided cut replaces the closed broken line by
two closed broken lines (Fig. 41)

(e'i+ • • • e' a+e' i+) and (e \- • • • e'a-e \-),

which appear as two new contours; hence the number of contours increases
by two. For a one-sided cut the closed broken line A = {ei • • • eaei) is re­
placed by the closed broken line
(e\+ • • • e \+ e \- • • • e'a-e \+ )

which appears as a new contour of the surface K'-, hence, K ' has one more
contour than K (Fig. 42).
In each of the complexes K and K ' shown in Fig. 42 identically designated
elements are to be identified. After the identification, the complex A is a
triangulation of a Mobius band on which (e^^ei) is a closed one-sided cut.
This cut converts K into a complex K ' isomorphic to a triangulation of a
plane ring. The complex K has one contour (apip2&<M2a); the complex K '
has two contours:
(fl'pYpibqiqza) and (e/i+e/2+e,3+e/i-e/2-e,3-e/i+).
90 SURFACES [CH. I l l

b 9*
F i g . 42a F i g . 42b

The reader is advised to construct a model of the complex K of the Mobius

band out of paper and to cut it along the line (eie2e3ei) with a pair of scissors.
R emark . Both the open subcomplex K \ A of K and the triangulation K',
which results from cutting K, consist, as is easily seen, of the same number
of mutually disjoint strongly connected complexes; if this number is equal
to 1, we shall say that the cut does not separate the surface; if it is greater
than 1, we shall say that the cut separates the surface and moreover,
because of 3.22-3.25, into two components.
Hence
3.34. Open exits, cross cuts with endpoints on two different contours, and
one-sided closed cuts do not separate a surface; closed two-sided cuts and cross
cxds with endpoints in the same contour either do not separate a surface or
separate it into two components.
§3.4. Reduction of holes. Several different operations which decrease
the number of contours of the surface by 1 or 2 are included under this
heading. We shall discuss two types of reduction of a single hole and of a
pair of holes.
1. R eduction of the F irst K ind of O ne H ole . Consider a contour
formed by a closed broken line with an even number of sides whose con­
secutive vertices are numbered as (Fig. 43)

Ri j Ci , Co , ■ , , a i ,cs, ■ , c 2 , c i , G-i .
The vertices c, and c \ are identified for arbitrary i. As a result of this
identification the closed broken line goes over into a simple broken line
with endpoints ai and a \ . An identification of this type represents an
operation inverse to that of an open cut.
2. R eduction of the S econd K ind of O ne H ole . Let

(zi< t2^ 3 ■ ■ ■ o,sa'i< F ia' 3 • • • a',

§3] CUTS AND IDENTIFICATIONS 91

a*
C,

F ig . 43

be a contour (cf. Fig. 42b, where ni = e!i+ , 02 = er2+ , 0.3 = e73+ , a \ = e'i_ ,
a' 2 = e’2- , a'3 = e'3_ , or Fig. 44). The vertices a{ and a'i are again identi­
fied. An identification of this type annihilates a one-sided cut on the surface.
Application of a reduction of the second kind to one of the two contours
of a plane circular ring, or (what comes to the same) to one of the two bases
of the lateral surface of a cylinder, yields a Mobius band (Fig. 45).
The proof is essentially the reverse of that given in connection with Fig.
42. For greater clarity let us carry out the above identification for another
triangulation of a plane ring. (The identification itself is to be understood
here and everywhere else in this chapter, as the combinatorial operation
on a given, but arbitrarily chosen, triangulation defined in 3.1.)
The identification to be performed is indicated in Fig. 46a (elements de-

F iq . 45
92 SURFACES [CH. I ll

P -----1?

a b c a'
/ /
a. d' c 6' a.

n / r.
b
F i g . 46

noted by identical letters and having identical subscripts, even though

provided with other marks which do not coincide, such as primes, asterisks,
etc., are to be identified). Let us first cut the figure along the lines pa and
a'q (Fig. 46b), and then identify all elements denoted by the same letters
and having the same subscripts. A preliminary rotation through 180°
of the rectangle a*a'*<7*p* about the straight line c'f has no effect on the
result of this identification. [The rotation has to do only with the intuitive
geometric interpretation of the given operation of identification and is
completely unrelated to its combinatorial content, which is determined
solely bjr the identification scheme. This remark also applies to analogous
cases in the sequel (7.1, 7.2).]
The rotation transforms Fig. 46b into Fig. 47a, which in turn, after the
indicated identifications, becomes Fig. 47b.
The one remaining identification of paq* with p*a*q obviously converts
the rectangle into a Mobius band. This proves the assertion.

a b
F i g . 47
§ 4] ORIENTABILITY OF SURFACES 93

This result clarifies the following proposition:

3.41. A reduction of the second kind of a hole with boundary r cut from
a spherical surface (Fig. 48) gives the same result as closing up a somewhat
larger hole with a Mobius band: the hatched ring is converted by an identifi­
cation of the second kind of its inner boundary T into a Mobius band with
boundary T'. (For convenience, a hemisphere with its equatorial plane,
instead of a spherical surface, is shown in the figures. These are obviously
equivalent topologically to a spherical surface; the letters in Fig. 48 refer
to the innermost of the three circumferences.) We will, therefore, refer to
a reduction of the second kind of one hole as a closing up of a hole with a
Mobius band.
3 . R e d u c t io n of T wo H oles by M This consists
e a n s of a H andle.
in pasting together the boundaries of the two holes. Let us consider this
operation in detail in the case of two circular holes cut out of a sphere. Let
the boundaries of these two holes be the closed broken lines
(ai • • • a,ax), (a'i • • • a\a 'i).
The identification may be effected in two ways.
The first case: handle of the first kind. Let a point a describe the contour
r in any one direction, e.g., counterclockwise as seen from outside the
sphere. Let the point a' of the contour r \ which is to be identified with
the point a, describe T' in the opposite sense (i.e., clockwise, as seen from
the exterior of the sphere). In this case we say that the identification of
T with r is an identification of the first kind or that T and T' have been
fitted vnth a handle of the first kind (Fig. 49). Now, as a describes T in a
given sense (counterclockwise as seen from the outside of the sphere) let
the point a' corresponding to a describe T' in the same sense (i.e., also
counterclockwise). This is called an identification of the second kind or we
say th at T and T' have been fitted with a handle of the second kind (Fig. 50).

§4. Orientability of surfaces

§4.1. Definitions.
D e f in it io n 4 .1 0 . An oriented triangle is a triangle together with a pre­
scribed sense of describing its boundary. Each of the two possible senses in

F io. 48 F io. 49 F i g . 50
which the boundary of the triangle may be described is called an orienta­
tion of the triangle. Hence a triangle ABC has two orientations, ABC and
BAC. If one of the orientations of a triangle T 2 is denoted by t2, the other
will be denoted by —t2. Each orientation of a triangle generates an orienta­
tion of each of its sides, the induced orientation of the sides: the orientation
ABC induces on the sides AB, BC, AC the orientations AB, BC, CA
(Fig. 51).
Consider two adjacent triangles (triangles with a common side) of a
triangulation K of a surface and choose a definite orientation of these
triangles. We shall say that the triangles are coherently oriented (non-
coherently oriented) if they induce opposite (identical) orientations on their
common side.
Thus, e.g., in Fig. 52 ABC and ABD are noncoherently oriented (since
the orientations induced on AB by both triangles are the same, namely
AB)-, but ABC and BAD are coherently oriented (since the orientations
induced on AB are opposite).
R e m a r k 1. In this definition it is essential to assume that the triangles
are elements of the same triangulation: in the case shown in Fig. 53 (the
figure is in the plane) the definition of coherent orientation of two triangles
does not apply.
D e f i n i t i o n 4.11. A triangulation A of a surface <t>is said to be orientable
if it is possible to orient all the triangles of K in such a manner that every
two adjacent.triangles are coherently oriented. In the contrary case the
triangulation K is said to be nonorientable.
D e f i n i t i o n 4.111. Let K be an orientable triangulation. Then it is pos­
sible to choose orientations , • • • , t2p for all the triangles T21 , • • • T2P of
K in such a way that every pair of adjacent triangles is coherently oriented.
A set of orientations / \ , • • • , t2p of all the triangles T2,- £ K satisfying this
condition is called an orientation of the triangulation K.
If an orientation / \ of a triangle T 2{ of K is given, the requirement that
every pair of, adjacent triangles be coherently oriented defines uniquely
§4] OKIENTA B1LITY OF SURFACES 95

first the orientations of all the triangles adjacent to T2,- and then step by
step, b y virtue of the strong connectedness of the triangulation K, the
orientations of all the remaining triangles of K. Hence, for any triangle
T i of an orientable triangulation K and for any orientation of T2, , there
is a unique orientation of all the triangles of K which contains the given
orientation of T2,-.
Therefore,
4.112. Every orientable triangulation has 'precisely two orientations.
R e m a r k 2. Defs. 4.1 1 and 4.111, as well as Theorem 4.112, maybe applied
not only to triangulations of surfaces but also to all open strongly connected
subcomplexes of these triangulations, e.g., to cyclic and semi-cyclic stars.
Let Q be a cyclic or semi-cyclic star. The complex Q is obviously iso­
morphic to a plane complex Q'. An orientation of every triangle of Q',
e.g., counterclockwise, yields an orientation
of Q', and hence an orientation of Q.
Hence
4.12. Cyclic and semi-cyclic stars are orientable.
In this connection we have the following
proposition (Fig. 54):
4.121. Let Q be the semi-cyclic star of a ver­
tex e; let (eie) and (ees) be bounding segments
of this star [i.e., segments each of which bor­
ders on a single triangle (eiee2) and (es_iees), respectively, of the star Q\.
Then each orientation of Q induces orientations on the segments (eLe) and
(ees) which are extensions of each other [i.e., the orientations (eie) and
(ees) or the orientations (eei) and (ese)].
It is not difficult to prove this proposition by, e.g., complete induction
on the number of triangles in Q.
In this chapter we shall use without proof the following proposition
which will be proved in Chapter X (naturally, independently of the results
of the present chapter).
4.13. Let K and K ' be arbitrary triangidations of the same surface or of
two homeomorphic surfaces; if one of these triangidations is orientable, the
other is also orientable.
This theorem is called the theorem on the invariance of orientability; in
the light of 4.13, it is natural to make the following definition:
4.14. A surface <t> is said to be orientable if any (arbitrary) triangulation
of 4 > is orientable. In the contrary case, the surface is said to be non-
orientable.
4.13 implies
4.15. I f a surface is orientable (nonorientable), then every surface homeo­
morphic to the given surface is orientable (nonorientable).
96 SURFACES [CH. I l l

D e f i n i t i o n 4.16. Let K be a triangulation of a surface 3>. A sequence

of oriented triangles T \ , T \ , ■■■ , T 2a of this triangulation with orienta­
tions t2! , • • • , t2, is called a disorienting sequence if the following conditions
are satisfied:
For i = 1, ••• , s — 1, the triangles T 2{ and T 2i+1 adjoin each other and
are coherently oriented; the triangles T 2a and T 21 also adjoin each other but
are noncoherently oriented.
If K is an orientable triangulation, it is impossible to construct a dis­
orienting sequence composed of oriented triangles of K. Indeed, if z is that
orientation of K which contains t \ , it is easy to see that the orientations
t22 , • • • , / \ are also contained in z; but then the orientation z would con­
tain two noncoherently oriented adjoining triangles t21 and t2„ . Conversely,
if a triangulation K is nonorientable, it contains a disorienting sequence. In
fact, choose any triangle T 21 of K and give it any orientation t21 . Orient
every triangle adjoining T21 coherently with respect to t21 . In general,
orient all the remaining triangles one by one in such a way that a triangle
adjoining an already oriented triangle receives an orientation coherent

with that of its already oriented neighbor. Since K is, by assumption, non­
orientable, a continuation of this process will eventually yield a triangle
T 2s which receives two opposite orientations depending on which of the
already oriented neighbors of the triangle T 2S is used for the definition of
the orientation of T2S itself. In other words, it is possible to go from T 21
to T2s , passing through consecutive triangles, by two paths, which lead to
opposite orientations of T 2S .
Going from T 21 to T 2S along the first path and then from T2S to T 2\ by
the second, we obtain a disorienting sequence of triangles.
Hence, we have proved
4.17. In order that a triangulation be nonorientable, it is necessary and
sufficient that it contain at least one disorienting sequence of triangles.
E x a m p l e . Let us consider the triangulation of the Mobius band shown
in Fig. 55. Giving the triangles of this triangulation the orientations desig­
nated by arrows in Fig. 55, we obtain a disorienting sequence. Hence, the
Mobius band is a nonorientable surface.
Theorem 4.17 implies that if a subcomplex K 0 of a triangulation K is
§4] ORIENTABILITY OP SURFACES 97

nonorientable, then K is nonorientable. Hence, it follows in turn that if a

triangulation K contains as a subcomplex any triangulation of the Mobius
band, then Tv is a nonorientable triangulation. It is easy to see that the
triangulations of the Klein bottle and the projective plane given above
contain triangulations of Mobius bands. Consequently, both the Klein
bottle and the projective plane are nonorientable surfaces. It is easy to
convince oneself of this immediately.
Let K be a triangulation of a nonorientable surface.
T hat part of || K || which is covered by the triangles of a disorienting
sequence always contains a Mobius band subdivided into triangles of some
regular subdivision (it suffices to take a second order barycentric subdi­
vision) of K. A disorienting sequence is shown in Fig. 56; Fig. 57 shows a
regular subdivision of the triangles of this sequence; the hatched triangles
form a Mobius band.
Hence:
4.18. In order that a surface 4> he nonorientable, it is necessary and sufficient
that some triangulation K\ of the surface 4> contain a subcomplex which is a
triangulation of a Mobius band. A regular subdivision of an arbitrary tri­
angulation K of the surface can be taken as the triangulation K\ .
R e m a r k . It would be possible to prove that every surface containing a
subset homeomorphic to a Mobius band is nonorientable. The converse of
this assertion is contained in what has just been proved.
We shall prove the following fundamental proposition:
4.19. In order that a surface 4> be nonorientable, it is necessary and sufficient
that some triangulation K of contain a one-sided cut line.
Proof. Suppose that K contains a one-sided cut line
A = (eie2 • • • e,ei).
 SURFACES [CH. I ll
98

Let us consider the closed chain of semi-stars

Oe+i , • • • , 0e+s , 0 e ~i , ■■■ , Oe s , Oe^i
[see (3.242)]. Let us form the closed chain (Fig. 08)

Ti — (cjiCid), • • • , Tk = (ces_ies), , Tr (deKe\)

consisting of the triangles of the semi-stars 0e + 1 , • • • , 0 e ¥s , Oe 1 . Let
us orient the semi-star 0e + 1 in any way and let the induced orientation on
(eie2) be, e.g., (eie2) *• The orientation of the semi-star 0e + 1 defines, in par­
ticular, the orientation of the triangle contained in both semi-stars Oe 1 and
0e+2, and consequently of the whole semi-star 0e + 2 . Hence the orienta­
tions of all the semi-stars 0e + 1 , • • • , 0e+s , Oe 1 are defined step by step.

Moreover, all these orientations induce on the broken line A the same
direction which, for the chosen orientation of 0e + 1 , is the sense
(eie2 • • • ese-i) .
Every pair of triangles Ti and T i+1 , i = 1, 2, • • • , s — 1, have been oriented
coherently. However, since = (eseia) and 7'r = (desei) have been oriented
so that the orientations induced on their common side by both triangles are
the same, i.e., (esei)~*, the triangles 7\ and T r are noncoherently oriented.
The chain of triangles
Ti , •• • , T r
with the above orientations forms a disorienting sequence, so that K is a
nonorientable triangulation.
Conversely, if K is a nonorientable triangulation, some regular subdi­
vision of K contains a triangulation of the Mobius band which in turn
contains a one-sided cut. This proves 4.19.
§5. The connectivity of a surface. Euler’s theorem
§6.1. Throughout this section K will denote a triangulation of a sur­
face <f>.
§5] CONNECTIVITY OF A SURFACE. EULER’S THEOREM 99

5.11. Let us denote by po, pi , p2 the number of vertices, edges, and

triangles, respectively, of a triangulation K of a surface 4>; the number
x(K ) — po — pi + pi
is called the Euler characteristic of the triangulation.
In this chapter we shall accept without proof the following theorem
(proved in Chapter X).
T h e o r e m 5.12. ( i n v a r i a n c e o f t h e e u l e r c h a r a c t e r is t ic .) I f K and
K ' are triangulations of the same surface or of two homeomorphic surfaces,
the Eider characteristics of K and K ' are equal.
In virtue of Theorem 5.12 it is natural to call the Euler characteristic of
any triangulation of a surface the Euler characteristic of the surface.
E x e r c i s e . Show (by considering any triangulation of the corresponding
surface) that the Euler characteristic of the sphere and the projective plane
is equal to 2 and 1, respectively, and that the Euler characteristic of a
plane ring (cylinder), Mobius band, torus, and Klein bottle is equal to
zero.
R e m a r k 1. In this chapter we shall say that a one-dimensional closed
subcomplex L of a triangulation K does not separate K, if the open sub­
complex K \ L Cl K is a strongly connected complex.
D e f i n i t i o n 5.13. The connectivity of a triangulation K is the maximum
integer k for which there exists a closed one-dimensional subcomplex of
connectivity k which does not separate K. The connectivity of a triangula­
tion K will be denoted by q{K).
R e m a r k 2. It follows easily from the Jordan theorem that every elemen­
tary curve of positive connectivity separates the sphere.
This implies, without the use of any invariance theorem, that the con­
nectivity of an arbitrary curved triangulation of a sphere, and conse­
quently of every surface homeomorphic to the sphere, is equal to zero. The
reader is advised to prove this assertion.
E u l e r ’s T h e o r e m 5.14.

(5.14) x (/0 = 2 — q(K).

Proof. Let L be a one-dimensional nonseparating closed subcomplex of K
of connectivity q = q(K). Let us enumerate the triangle,s of K in an order
T \ , • • > , T 2r , where r = P2 , such that the triangle T 2i+i , i = 1 , 2 , • • • ,
p2 _ 1 ( will have a side T \ not contained in L in common with one of the
triangles T 21 , • • • , T2,- (this can be done, since L, by hypothesis, docs not
separate K ).
Let us delete the triangle T 21 from K = K 0 and denote the remaining sub­
complex of the triangulation K by K x ; next delete both the triangle T \
and the segment T \ and denote the remaining complex by K 2, etc., until
100 SURFACES [CH. I l l

all P2 triangles are exhausted. We obtain complexes K i , K 2 , ■• • , K r- 1 ,

where T 2t is the only 2 -element of the subcomplex K r-i ; finally,
K t — K r- 1 \ T 2t is a 1 -complex.
We shall prove that all the complexes K { are connected. For Ko = K
this is true by hypothesis. Assuming that is connected, we shall show
that the complex K i+1 = K i \ T 2i+i \ T li is also connected. It suffices to
prove that every pair of vertices a, b of K {+1 can be connected by a broken
line in K{+1 . Let [a5] be a broken line connecting the vertices a and b in
K i . If this broken line is not contained in K i+1 , one of its links is the side
T li of the triangle T2i+1 . If we prove that the other two sides TS and T l2 of
of the triangle T \ +i are contained in K i+ i, then the link T \ of [ab] can be
replaced by a broken line of two links consisting of T1! and T l2 and the
assertion will be proved. But the side T1! (or T x2) does not appear in A 1+]
only if it is a side of some triangle T2h , h < i + 1 . This, however,
cannot be, since, if the segment T li , which is a side of !T\-+ 1 , is also a side
!TY_i, h < i + 1 , of the triangle T \ , then it would have to be a side of
yet a third triangle T2k , k < h , which is impossible. This proves that
K i +1 is connected. The complex K r does not separate K since all the triangles
of K were enumerated in the form of a chain not intersecting K T ; since
K r 3 L, Theorem 1.23 implies that tt1(Kt) > ir(L) = q, and consequently,
by the definition of L and q, 7T1(KT) = 7T1(L) = q.
The 1 -complex K r has p0 vertices and pi — (p2 — 1) edges (since p2 — 1
edges have been deleted from K). Hence, by (1.24),

P0 - fpi - (p2 - 1)] = A K r ) - A K r ) = 1 - q,

or po — pi + p2 = 2 — q, q.e.d.
5.12 and 5.14 imply
5.15. T h e o r e m o n t h e I n v a r ia n c e of C o n n e c t iv it y . T w o a r b itr a r y
t r i a n g u l a t i o n s o f tw o h o m e o m o r p h ic s u r f a c e s h a v e th e s a m e c o n n e c t iv i ty .
It is therefore natural to call the connectivity of any triangulation of a
given surface the c o n n e c t iv i ty o f th e s u r f a c e .
§6. Simple surfaces
§6.1. Closed cuts. In this and in the following sections a triangulation
will mean a triangulation of some surface, and a c lo s e d c u t lin e of a tri­
angulation K will mean any simple closed broken line A consisting of ele­
ments of the triangulation K and such that at least one of its segments is
an interior element of the triangulation K (i.e., one which is not situated
on its boundary).
If the closed cut line A contains more than one vertex belonging to the
boundary of K, then A, as is easily seen, consists of one or more cross cuts
§ 6] SIMPLE SURFACES 101

and of vertices, and perhaps arcs, which lie on the boundary of K (A may
or may not contain such arcs; see Figs. 59-60).
In this case the cut operation itself is realized by the cross cuts contained
in A.
But if A has a single element, a vertex e0 (Fig. 61), on the boundary,
then the cut operation along the line A is defined as follows: take any
second vertex e of A and perform an open cut along one of the two arcs
[ege] CZ A, say along the arc [eofiie], and then a cross cut along the second
arc [ege] contained in A.
R e m a r k . In cases where there can be no misunderstanding, instead of
“cut line” we shall simply say “cut” .
The following proposition -will be required in the sequel:
6.11. If a triangulation K contains a nonseparating closed cut A some of
whose elements lie on the boundary, there is a regular subdivision K ' of
K which contains a closed cut A' which does not separate K ', and such that
all its elements are interior elements.
Passing to the proof of 6.11, we note first that:
In order that a closed cut A not separate the triangulation K, it is suffi­
cient that two triangles Ti and T2 having a common side contained in A
can be joined by a chain of triangles in which every two consecutive tri­
angles have a common side not contained in A. If this is the case, we shall
say that 7\ and T2 are joined by a chain of triangles which avoids the
cut A.
We shall show that, passing, if necessary, to a barycentric subdivision,
it can always be assumed that the “stars” of the vertices e,- and e , , belong­
ing to a contour T, can intersect only if the edge (e,e;-) is in T. Indeed,
let If be a triangulation and K ' its barycentric subdivision. 'It is easily
seen that there are in K ' no interior edges both of whose endpoints lie on the
boundary of the surface. It follows that the triangulation K ' has the fol­
lowing property: the “stars” lying on the boundary of the surface with
102 surfac es [CH. I l l

vertices e' and e" intersect only if both vertices e' and t" belong to a single
contour F and are neighboring vertices on this contour. We shall assume
that the triangulation K has this property to begin with.
To obtain the line A' it is necessary to remove from the boundary every
piece A0 of the cut A which lies on the boundary of the surface (Fig. 62).
This is achieved as follows: take a barycentric subdivision K ' of the triangu­
lation K ; next, consider the part [eTei ■■■es_ie's_i] of the cut, consisting of
the piece A0 = [ei • • es_i] lying on the boundary of the surface, supple­
mented by the two interior edges (e \e i) and ; let us replace this
part of the cut by the broken line Ai = [e\ • • • e's_i] whose vertices are
interior vertices of the triangulation K', lying on the boundary of the
“stars” of the vertices of the piece A0 (in the triangulation K').
Let us prove that this replacement yields a simple closed broken line
A' which does not separate K ' . Indeed, the broken line A i, as is easily seen,
is a simple nonclosed broken line lying in the interior of the triangulation
iv'; in addition, all the vertices lying on Ai , with the exception of e\ and
e 's- i , do not lie on A. Therefore, the broken line A' is the union of two simple
nonclosed broken lines having only their endpoints in common. Hence A'
is a simple closed broken line.
It remains to be shown that the broken line A' does not separate the
triangulation K '. This assertion in turn is an easy consequence of the fol­
lowing. There are two triangles in K ' with a common side in Ai / which
can be joined by a chain of Triangles avoiding A'.
To prove this last assertion, let us consider a chain of triangles
T \ , • • • , T 's of K ' connecting the triangles T \ and T 's with common side
(eVi) and avoiding A. The chain will contain at least one triangle with a
side in A i, and vertices on Ao ; indeed, one (and only one) of the triangles
T \ , T ' , , say T \ , is a triangle of this sort. Let T \ be the last triangle of
the chain with the indicated property (obviously k ^ s). It is then easy to
see that T 'k and TT+i have a common edge contained in Ai and can be
connected by a chain of triangles avoiding A'.
§6.2. Definition of simple triangulations. Invariance under regular
subdivisions.
D e f i n i t i o n 6.21. A triangulation of a surface 4> is said to be sijnple if it
§6] SIMPLE SURFACES 103

is separated by an arbitrary closed cut. An example of a nonsimple triangu­

lation is a triangulation of a Mobius band or of a torus. It can be shown
that: I f a triangulation of a surface $ is simple, then every triangulation of
every surface homeomorphic to $> is also simple.
However, we shall need merely:
6 .2 2 . A ny regular subdivision of a simple triangulation is simple.
Proof. It suffices to deduce from the existence of a nonseparating cut
of a subdivision K ' of the triangulation K the existence of a nonseparating
cut in K. Let K ' be an elementary subdivision with respect to an interior
side (in the case of a side lying on the boundary, the reasoning which
follows below is even simpler) (a c ) which borders on the triangles (a b c ) and
( a d c ) and denote the new vertex on (ac) by e. If A is a nonseparating cut
in K ', which does not appear as such in K, A contains at least one of the
segments {b e ), ( e d ) . If A contains only one of these segments and goes
through, e.g., [bee], then replacing the broken
line [bee] in A by the segment {.be) yields a
nonseparating cut in K.
Let A contain both (be) and (ed). We shall
consider two cases:
1°. Let one of the points a, c, say c, be
contained in either A or a contour T of the
surface containing at least one vertex of
the cut A. The triangle (ode) can be con­
nected, avoiding the cut A, with one of the
triangles (bee), (dec) by a chain which does
not contain the other triangle. If the triangles
joined are (ade) and (bee), the segment (be), the part of T up till the first
vertex of A, and the part of A not containing [deb], form a nonseparating
cut [our assumptions imply that the side (be) is necessarily an interior
side; otherwise [bd] would separate (aed) from (bee)] of the original triangu­
lation.
2°. Suppose that neither a nor c is contained in A. Either, one of the points
a, c, say c, is an interior point, or the contour T which contains this point
has no point of A. Then, replacing the segment [bed] by the broken line
[bed] in A, again yields a nonseparating cut in K.
6.23. Every triangulation K 0 which is a subcomplex of a simple triangula­
tion is simple.
Indeed, every closed cut A in K 0 is a closed cut in K. Let T i and T 2 be
two triangles of K 0 with a common side contained in A. Since it is impos­
sible to join them in A by a chain of triangles avoiding A, it is all the more
impossible to join them by such a chain in K o , which was to be proved.
104 SURFACES [CH. Ill

§6.3. Elementary lemmas. We shall require the following propositions

in the sequel. Their proof is left to the reader.
L emma 6.31. A topological mapping of the circumference S of a circle
Q onto the circumference S' of a circle Q' can be extended to the whole
circle (i.e., there exists a topological mapping of the circle Q onto the circle
Q1 which maps corresponding points of the circumferences into each
other).
Hint. Use polar coordinates.
L emma 6.32. A topological m apping of an arc [afe] of the circumference
S of a circle Q onto the diameter [cd] of a semi-circle Qi can be extended to
a topological mapping of th e whole circle Q onto all of th e sem i-circle Qi
in such a w ay th at the circumference of the circle is mapped onto the
boundary of the semi-circle.
R emark . Let us call every compactum Q which is the topological
image of an ordinary circle Q0 a topological circle. It will be shown in Chapter
V that every topological mapping of Q0 onto Q maps the circumference of
Qo onto the same simple closed curve S a Q, called the topological circum­
ference of Q. If this assertion is accepted, Lemmas 6.31 and 6.32 obviously
remain true if Q and S are taken to mean a topological circle and its topo­
logical circumference.

§6.4. Classification of simple surfaces. A surface is said to be simple if

it has at least one simple triangulation.
T heorem 6.41. Every simple surface whose boundary consists of a single
contour is homeomorphic to a circle.
Proof. As usual, let us denote by p2 the number of triangles of a (simple)
triangulation K of the surfaced. The theorem is obviously true for p2 = 1.
Assuming the theorem true for p2 < n, we shall prove it for p2 = n + 1.
The triangulation K contains a triangle which adjoins the boundary of
the surface 4>. Let T 2 = (eoCie2) be such a triangle.
Three cases may arise:
a) All three vertices, but only two of the sides, (eo^) and (e2ei), of the
triangle T 2 are situated on the boundary of the surface <t>.
b) All three vertices, but only one side, (eo«i), of the triangle T 2 are on
the boundary of the surface <t>.
c) Two vertices, eo and ex , and one side (eaef) of the triangle T 2 are on
the boundary of the surfaced.
In cases a) and c) the closed cut (eoCie2), i.e., in case a) the cross cut
A = [eoei], and in case c) the cross cut A = [e0e2ei] (Figs. 63-65) separates
K into two disjoint triangulations. One of them is the triangle T 2 with its
sides and vertices and the other is some triangulation K\ . The boundary
of Ki is a simple closed broken line consisting of A and the piece [cocei] of
 F ig . 63 (Case a) F ig . 64 (Case b) F ig . 65 (Case c)

the boundary of K. The triangulation K i is simple by virtue of Theorem

6.23 and contains n triangles. Consequently, the surface || K \ || can be
mapped topologically onto a semi-circle Qi of a circle Q in such a way that
A maps onto the diameter of the semi-circle. Mapping T 2 onto the other
half circle Q2 of the circle Q in such a way that this mapping coincides
with the preceding one on A yields a mapping of all of the surface $ onto
the circle Q.
In case b) the points e0 , e i , and e2 separate the boundary of the surface
<f> into three pieces Ax = [e0ce2], A2 = [e-de1], and A3 = [eie0]- The closed cut
along the curve consisting of Ai and the side (e2eo) of the triangle T 2, i.e.,
the cross cut along (eoe2), separates K into two disjoint simple triangula­
tions K \ and K 2, each of which contains no more than n triangles. There­
fore, the surfaces || K i || and || K 2 || are homeomorphic to a circle. Mapping
each of them onto the two half circles of the same circle in such a way that
both mappings coincide on (eoC2), yields a mapping of the surface || K ||
onto a circle, which was to be proved.
T h e o r e m 6.42. A simple closed surface is homeomorphic to the sphere.
Proof. Deleting one triangle T 2 from a simple triangulation K of the
closed surface <f>yields a simple triangulation K \ with one contour. Mapping
l| Ki |] and || T 2 || topologically onto two hemispheres in such a way that
both mappings coincide on the boundary of the triangle T 2 and take this
boundary onto the equator of the sphere, we obtain the required mapping
of the whole surface || K || onto the sphere.
E x e r c i s e . Using reasoning analogous to that in the proof of Theorem
6.41, prove without the invariance theorem, that the Euler characteristic
of an arbitrary triangulation of a surface homeomorphic to the circle is
equal to 1. From this derive (also without using the invariance theorem)
the classical theorem of Euler:
The Euler characteristic of an arbitrary triangulation of a surface
homeomorphic to the sphere is equal to 2.
R e m a r k 1. The Jordan theorem implies that every triangulation of the
sphere is simple. Hence:
106 SURFACES [CH. I l l

I f one triangulation of a closed surface $ is simple, <f> is homeomorphic to

the sphere, and then every triangulation of $ is simple.
D e f i n i t i o n 6.43. A sphere with r circular holes whose boundaries are
disjoint by pairs is called a normal simple surface Qr with r contours. The
boundaries of the holes are part of the surface.
R e m a r k 2. Since a topological mapping of one surface <f> onto another
4>' maps the boundaries of the surfaces <t> and 4>' onto each other (this
assertion will be proved, independently of the results of this chapter, in
Chapter V), and homeomorphic sets (in particular, elementary curves) have
the same number of components, homeomorphic surfaces have the same
number of contours..
R e m a r k 3. To cut a circular hole out of the spherfe is equivalent to
cutting a spherical sector out of it.
R e m a r k 4. The surfaces QT defined in 6.43 are obviously homeomorphic
to the surfaces obtained by removing from a circle Q the interior of r — 1
mutually disjoint circles contained in the interior of Q. The surfaces Qr will
be shown in just this way in the figures.
T h e o r e m 6.44. Every simple surface <f> with r contours is homeomorphic
to a normal simple surface Qr .
This theorem is obviously included in
6.45. Let 4> be a surface with r contours which has a simple triangulation
K. Let C be a topological mapping of one of the contours of the surface <f> onto
one of the contours of a normal surface Qr . The mapping C can be extended
to a topological mapping of the whole surface <f> onto the surface Qr (taking,
in virtue of the invariance of the boundary, the boundary of 4> into the
boundary of Qr).
Proof of Theorem 6.45. Theorem 6.41 and Lemma 6.31 imply that
Theorem 6.45 is true for r = 1. We shall assume Theorem 6.45 true for
r < n and prove it for r = n + 1.
Thus, let K be a simple triangulation of a surface <f>with n + 1 contours,
among which we mark any two contours, Ti and r 2 . Let the surface Qn+i be
represented by a definite triangulation (in order to make it possible to
speak of cuts on the surface). Let us single out two arbitrary contours
r 'i , r ' 2 from among the contours of Qn+1 . Let A be a simple broken line
in K joining Ti and r 2 and let A' be a broken line op Qn+i joining r \ and
r ' 2 . Cuts (Fig. 66) along the broken lines A and A' transform the surfaces
<f> and Qn+i into 4>i = || Ki || and Q„+i,i, respectively, where each pair of
contours: Ti , r 2 and r ' i , r ' 2 , is replaced by the single contour
T = Li u A+ u L2 u A",
and
r ' = V i u A/+ u L 'a u A'“,
§6] SIMPLE SURFACES 107

respectively. Here, Li , L2, etc., are simple arcs arising from Ti and r 2 ,
etc., after the cut (for example, in Fig. 67 Li is the simple arc [ei_Ciei+]).
The surfaces $i and Q*+i,i have only n contours. The surface Qn+i,i can
obviously be identified with the surface Qn .
We shall show that $1 is a simple surface. Indeed, if the triangulation
K 1 , into which the triangulation K is transformed after the cut, were to
contain a nonseparating closed cut, some regular subdivision K \ of K\
would contain a nonseparating closed cut D \ without boundary elements.
Under the identification which returns Ki into K, the subdivision K \ of
K 1 goes over into some regular subdivision K ' of K and the nonseparating
cut D' 1 of K \ into a nonseparating cut of K ' . But this is impossible, since
K ', being a regular subdivision of the simple triangulation K, is simple, by
Theorem 6.22.
Xow let C(Ti , r , ) be a topological mapping of the curve I \ onto r 'i .
It defines a topological mapping C(LX, L \) of the simple arc Li onto the
simple arc L \ , which maps both endpoints ei+ , ei_ of the arc Li onto the
endpoints of the simple arc L \ . The mapping C{L\ , L \) can be extended
to a mapping C(T, T') of the closed curve T onto T', which maps A+, Li , A~,
L2 onto A/+, L \ , A'~, Z/2 , respectively, and which maps two points of the
arcs A+ and A- lying opposite each other into two points of A/+ and A'~
lying opposite each other.
By the inductive hypothesis, the topological mapping C(T, U) can be
extended to a topological mapping C(4>i , Qn) of the surface $1 onto Q„+i,i ,
i.e., onto Qn . If now an identification which annuls the cuts just made
is performed, the homeomorphism C(4>i , Qn) between $1 and Q„+i,i goes
over into a homeomorphism C(d>, Qn+i) between $ and Q*+i which coincides
on lb with the originally given homeomorphism (7(ri , T'i). This proves
Theorem 6.45 and hence 6.44.
108 SURFACES [CH. I l l

6.44 and Remark 2 imply:

6.440. Two simple surfaces are homeomorphic if, and only if, they have
the same number of contours.
Let us now prove that the Euler characteristic of a simple surface with r
contours is 2 — r. Because of the invariance of the Euler characteristic, it
suffices to prove that
(6.46) x(Qr) = 2 - r.
Since the Euler characteristic of the sphere is 2 and because of the
invariance of the Euler characteristic, the equality x = x(Qr) = 2 — r
follows from the following obvious proposition:
Deleting from a triangulation K an arbitrary triangle and retaining its
vertices and sides decreases the Euler characteristic of the triangulation
by one.
E x e r c i s e . Prove by means of the Jordan theorem that every triangu­
lation of a simple surface is simple. For the proof it suffices to consider
arbitrary curvilinear triangulations of simple normal surfaces.

§7. Classification of closed surfaces

§7.1. Genus of a surface. Normal surfaces of a given genus.
D e f i n i t i o n 7.11. Let $ be a closed surface. The genus of the surface <f>
is, by definition, one half its connectivity if is orientable, and its con­
nectivity decreased by 1 if is nonorientable. The genus of a closed surface
is denoted by p(4>)-
Let us clarify the geometric meaning of the term genus.
We note first that fitting a pair of holes of a simple normal surface Qr
with a handle does not change the Euler characteristic of'the surface. Indeed,
an identification of two contours (divided into the same number of arcs)
does not change the number of triangles, while the number of segments
and vertices is decreased by the same number (if the boundary of each
hole had k sides and k vertices, the total number of 1- and the total number
of 0-elements of the triangulation after the matching is decreased by k).
In the same way, closing up one of the holes of the surface Qr with a
Mobius band does not change the Euler characteristic of the surface.
Let p be an arbitrary nonnegative integer and consider the normal
simple surface Q2p , i.e., the sphere with p pairs of holes. Let us denote
by 4>p the closed and, as is easily seen, orientable surface obtained by
fitting the p pairs of holes with handles of the first kind. By the above
remarks, the Euler characteristic of the surface 4>p is equal to the Euler
characteristic of the surface Q2p , i.e., 2 — 2p. Hence
X ( F P) = 2 - 2 p.
§7] CLASSIFICATION OF CLOSED SURFACES 109

Since, on the other hand (see 5.14),

x(4>P) = 2 - q,

it follows th at q = 2p, i.e., p = q/2 is the genus of the surface 4>p .

Hence
7.12. Let p be an arbitrary nonnegative integer; fitting each of the p
pairs of holes of the normal simple surface Q2p with a handle of the first kind
yields a closed orientable surface of genus p.
The surface is called the normal closed orientable surface of genus p
or simply “the sphere with p handles”.
Let us now close up all p + 1 holes of the surface Qp + 1 with Mobius
bands. The result is a closed nonorientable surface 4'p whose Euler char­
acteristic is equal to the Euler characteristic of the surface Qp+j , i.e.,
1 - p.
Euler’s formula implies that p = q — 1, i.e., that p is the genus of the
nonorientable surface 4>p .
Hence
7.13. Closing up all p + 1 holes of the surface QP+i with Mobius bands
yields a nonorientable closed surface of genus p.
I t is called the normal closed nonorientable surface of genus p (or the
sphere with p + 1 cross-caps).
In view of the above construction it is natural to ask what happens if a
pair of holes of the surface Qr is fitted with a handle of the second kind. The
answer to this question is given by the following proposition, which we
shall need.
7.14. Fitting a pair of holes of the surface Qr with a handle of the second
kind is equivalent to fitting each of these holes with a Mobius band (and there­
fore leads to a nonorientable surface).
Proof. Fig. 68 shows a pair of holes fitted with a handle of the second
kind; the points 1, 2, 3, 4, 5, 6, 7, 8 are identified with the points V , 2',
3', 4', 5', 6', 7', 8', respectively. Let us per­
form cross cuts along the lines 33' and 77'
(Fig. 69) and turn the “quadrilateral”
(Figs. 69, 70) 42'8'6 through 180° about
the axis A A ' (Fig. 71).
Matching the elements to be identified
(idenoted by identical numbers), we get Fig.
72.
It now remains merely to close up each
of the two holes 3ab3'a'b' and 7d'c!7'dc as
indicated (i.e., to identify 3 with 3', a with
 F i g . 69 F i g . 70 F i g . 71

a', b with b', etc.). These are reductions of

the second kind, i.e., each of the holes is
closed with a Alobius band.
§7.2. The fundamental theorem of the top­
ology of surfaces. Our purpose will be achiev­
ed by proving the following two propositions,
which together make up the so called funda­
mental theorem of the topology of surfaces.
7.2. Every closed s
phic to a normal closed orient-able surface bp
of some (integral) genus p (if b is orientable)
or to a normal closed nonorientable surface b p of genus p (if $ is non-
orientable).
It follows immediately that the genus p(b) of every closed orientable
surface $ is an integer. Therefore the connectivity is an even number.
Theorem 7.2 and the invariance of the Euler characteristic and orien-
tability of the surface imply
7.21. Two closed surfaces $ and b' are homeomorphic if, and only if, they
are both orientable or both nonorientable and if, in addition, they have either
the same Euler characteristic, or the same connectivity, or the same genus.
Passing to the proof of 7.2, we note first that:
7.211. Let be a closed surface with q = Every subcomplex of an
arbitrary triangulation K of the surface consisting of q + 1 mutually dis­
joint closed curves
E\ , • • • , L q+1,
separates the triangulation.
Indeed, Li u • ■• u L q+1 is a 1-complex of connectivity q + 1, whence
7.211 follows.
§7] CLASSIFICATION OF CLOSED SURFACES 111

Let q' = g'(4>) be the maximum number such that there exist q' mutually
disjoint closed curves L\ , • • • , L q>on some triangulation K of the surface
which form a nonseparating subcomplex L = Li u • • • u L q>of K. It
follows from 7.211 that
q'(<P) < q.
Let us perform a cut along all the curves Li , • • • , L q- . As a result of
this cut, the triangulation K = Ko is converted into a triangulation K\ of
some surface || K\ || = <t>i .
Let us show that K i is a simple triangulation. In fact, if there were a
nonseparating cut on the triangulation Ki , there would exist a nonseparat­
ing cut on some subdivision K \ of K i , consisting entirely of interior elements.
Consequently, q' + 1 mutually disjoint nonseparating cuts could be per­
formed on the corresponding triangulation K ' of the surface 4>. This con­
tradicts the definition of q' .
Hence $1 is a simple surface homeomorphic, by 6.44, to some normal
simple surface Q, .
In order to return to4> from <t>i it is necessary to perform an identification
suppressing the cuts made on the surface <t> along the lines Li , • • • , L q>.
At the same time we will perform the corresponding identification on Qr .
As a result, 4>i becomes <t>and Qr is turned into some closed surface S homeo­
morphic to <t>.
W hat will S represent? To answer this question, let us consider in detail
a cut of <f> along a fixed curve L = Li and the inverse operation, an identi­
fication. The following cases are possible:
1. The cut L is one-sided.
Its suppression is a reduction of the second kind to which, on Qr , corre­
sponds closing up a hole with a Mobius band. If <f>is an orientable surface,
this case cannot arise.
2. The cut L is two-sided; it generates two contours L +, L~, to which
correspond two holes L \ , L \ of the surface Qr . Therefore, if every cut L
is two-sided, r is an even number 2p.
Suppression of the cut L of the surface <J>corresponds on Qr to the reduc­
tion of the pair of holes L \ , IJ 2 , which is realized either by a handle of the
first kind or a handle of the second kind. In the former case we call the
cut L a two-sided cut of the first kind and in the latter case a two-sided
cut of the second kind.
By 7.14, it is impossible to have a two-sided cut of the second kind on an
orientable surface.
Hence, if 4> is an orientable closed surface, r is an even number 2p, and
the surface S, the homeomorph of <t>, is obtained from Qr by fitting each pair
of holes with a handle of the first kind.
112 SURFACES [CH. I ll

In other words:
An integer p > 0 is defined for every orientable closed surface <£ such that
is homeomorphic to a normal orientable surface of genus p (“sphere with p
handles”) and is consequently itself a surface of genus p: two orientable closed
surfaces are homeomorphic if, and only if, they have the same genus or, what
comes to the same, the same Euler characteristic.
Let us now pass to the case of a closed nonorientable surface. Cuts of all
three types may occur when a closed nonorientable surface $ is trans­
formed into a simple surface QT . Moreover, either at least one one-sided
cut, or at least one two-sided cut of the second kind (or both) must occur.
Hence, every closed nonorientable surface is homeomorphic to a closed
surface S resulting from some simple surface QT by means of the following
operations :
a) closing up some of the holes of the surface Qr with Mobius bands;
b) fitting some pairs of holes of the surface Qr with handles of the second
kind;
c) fitting some pairs of holes of the surface Qr with handles of the first
kind.
We shall show that it is always possible to restrict oneself to operations
of type a).
First, by virtue of 7.14, every operation of type b) can be replaced by
two operations of type a).
We shall now prove
7.22. The set of two operations:
1) closing up a hole until a Mobius band,
2) fitting a pair of holes with a handle of the first kind,
is equivalent to the set of two operations:
1) closing up a hole with a Mobius band,
2) fitting a pair of holes with a handle of the second kind.

F i g . 73
§7] CLASSIFICATION OF CLOSED SURFACES 113

F i g . 76 F i g . 77

Indeed, let holes I and I I (Fig. 73) be fitted with a handle of the first
kind, and hole I I I be closed with a Mobius band, i.e., its diametrically
opposite points identified (a with a', b with b', etc.).
Let us perform a cut along the curve aABCDEFa' (Fig. 73), and turn
the figure A'B'C'D'E'F'dcb (Fig. 74) through 180° about the horizontal
axis cf (Fig. 75; in Figs. 74 and 75 the axis cf is dotted).
Let us close up the hole I I I , using the indicated identification (Fig. 7G).
Let us add the contour ABCD EFA'B'C'D'E'F'A in the form of a cir­
cumference (Fig. 77).
The resulting circular hole is obviously closed up with a Mobius band,
and the holes I and I I with a handle of the second kind, which was to be
proved.
Suppose that a nonorientable surface <f> is obtained from the surface
Qr by a number of operations of types a), b), and c). Since every operation
of type b) can be replaced by operations of type a), we may suppose that
operations of type a) are present. But in that case every operation of type
114 SURFACES [CH. I l l

c) can be replaced by an operation of type b), so that all are reduced to

operations a) and b).
Replacing all operations of type b) with operations of type a), the entire
construction of a nonoricntable closed surface from a simple surface
Qr is reduced to operations of one type a) alone.
Hence
Every closed nonorientable surface is homeomorphic to a surface obtained
from a normal simple surface Qr (sphere with r holes) by closing up the holes
of the surface Qr with Mobius bands.
This proves Theorem 7.2 in its entirety.
E xam ples.

1. For the projective plane we have

x ($ ) = 1, ?($) = 1, = 0.
The projective plane is obtained by closing up a sphere with one hole
with a Mobius band or (what comes to the same) by pasting together a
Mobius band and a circle at their boundaries.
2. For a Klein bottle we have
X@0 = 0 , ?(*) = 2 , p ($ ) = 1.
The Klein bottle is obtained if two holes cut in the sphere are closed up
with Mobius bands or (what is clearly the same) if two Mobius bands are
pasted together along their boundaries.
R e m a r k 1. From all the above it follows that:
The connectivity of a closed orientable surface <t> is an even number
equal to the number of holes in the sphere it is necessary to fit (by pairs)
with handles of the first kind in order to obtain a surface homeomorphic to
the given surface <t>.
The connnectivity of a closed nonorientable surface <f> is equal to the
number of holes in the sphere which it is necessary to close up with Mobius
bands in order to obtain a surface homeomorphic to the given surface.
R e m a r k 2 . If t h e c o n n e c t i v i t y of a g i v e n c lo s e d n o n o r i e n t a b l e s u r f a c e
is an pven number q = 2p, a surface homeomorphic to <f> can be obtained
by taking the simple surface Q2p (the sphere with 2p holes) and fitting the
p pairs of holes of this surface with handles of the second kind (if p = 1,
the surface is the Klein bottle).
 Part Two
C O M PLEX ES. C O V E R IN G S. D IM E N S IO N

This part, like the preceding one, consists of three chapters. Chapter IV
is a detailed study of the notion of a complex; hence this chapter (and also
Chapter VII) contains a development of the combinatorial (and algebraic)
apparatus of topology; both chapters are essentially auxiliary in character.
Chapter VII can be read immediately after Chapter IV.
Several important topological facts—the invariance of the dimension
number of polyhedra, the invariance of interior points, and the fixed point
theorem for continuous mappings of a simplex—are proved in Chapter V
by elementary means (with the aid of Sperner’s lemma). Chapter V may
be read after Chapter IV.
Chapter VI is devoted to an introduction to dimension theory and is
based on IV; 1, 2.1-2.2, 3, 4.1-4.3, and 5.1-5.3 (and on the portions of
Chapter I indicated in the references). Chapter VI should be read after
Chapter V.
 Chapter IV
CO M PLEXES

Introductory section: preliminary remarks on simplexes

In this section we shall recall the definition of an n-simplex (see Ap­
pendix 1) and introduce the basic notions associated with this definition.
The reader is advised to omit on a first reading the parts of this section
marked with an asterisk. They are not needed until Chapter XIV.
§0.1. Simplexes and their skeletons. Let e0 , ex , • • • . , en , 0 < n < m,
be n + 1 linearly independent points in the Euclidean ?n-space R m\ in
virtue of their linear independence, these points determine an n-plane R n
(eo, e i, • • • , en) C R m and in that plane a system of barycentric coordi­
nates (see Appendix 1, 1.2). Those points of the plane R n (eo, ex , • • • , en)
all of whose barycentric coordinates po , Pi , • • • , are positive with re­
spect to the system (eo, ex , ■■■ , en) form, by definition, an n-simplex
T = (eo ■• • en) with vertices e0, • • • , en .
The number of vertices of a simplex diminished by 1 is called the di­
mension number or dimension of the simplex. The dimension of a simplex
is usually denoted by a superscript. Thus, T n denotes an ?i-simplex.
A 0-simplex is a point. A 1-simplex with vertices e0 and ex is an open
segment (eoex) (a segment without its endpoints). A 2-simplex is a triangle
and a 3-simplex, a tetrahedron (also open, i.e., without their boundary
points).
D e f i n i t i o n 0.11. The set of all n + 1 vertices of an ?i-simplex is called
the skeleton of the simplex.
Simplexes in R m correspond (1-1) to their skeletons.
* R e m a r k . It is sometimes expedient to consider s3rstems of ?i + 1 dis­
tinct, but linearly dependent, points in R m as the skeletons of “degenerate”
?i-simplexes. In that case it is convenient to identify a degenerate n-simplex
with its skeleton.
Hence we arrive at the definition:
Every set of n + 1 points of R'n lying in a plane R k C R m, k < n, is called
a degenerate n-simplex of R m. By the skeleton of a degenerate simplex we
shall understand the degenerate simplex itself. (Wherever degenerate sim-
plexes are used, this will be explicit^ mentioned. Otherwise, simplex will
mean nondegenerate simplex.)*
§0.2. Faces. If the skeleton of a simplex T x is a subset of the skeleton of
a simplex T2 , T x is said to be a face of T2 . H T X = (e0 • • • e,) is a face of
T2 = (e0 ■• • eTer+i ■■• en), T i consists of all the points of the space R n
116
§0] PRELIMINARY REMARKS ON SIMPLEXES 117

Oo, • • • ) £n) whose barycentric coordinates m with respect to e0, ■• • , en

are positive for i < r and equal to zero for i > r.
The number of r-dimensional faces (r-faces) of an n-simplcx is equal to
the number of combinations C(r + 1, n + 1) of n + 1 things taken r + 1
at a time. The 0-faces of a simplex are its vertices. A simplex is its own
face; the remaining faces are said to be proper faces of the simplex.
R e m a r k 1. I t is important to observe th at in this book a face (unless
otherwise noted) of a simplex will always mean either a proper face of the
simplex or the simplex itself.
If a simplex Ti is a proper face of a simplex T2, we shall write Ti < T2
or To > Ti . If is a face of T2, we shall write 7\ < T 2 or T2 > Ti .
The simplexes T\ and T 2 are said to be incident if either T\ < T2 or
T 2 < Ti .

F i g . 78

The faces 7,r1 and T \ of a simplex T n are said to be opposite faces of T n

if every vertex of T n is a vertex of one of the faces T \ or T \ and if these
faces have no vertices in common (Fig. 78). In that case
r + s = n — 1.
I t is sometimes convenient to denote the face T T of a simplex (eo • • • en)
opposite the face T ' of this simplex by placing the symbol “ a ” over the
vertices of the face T s; e.g., (eo • • • &• ■ ■en) denotes the face
(C o * * * C j —. l C i - f - i * c n ) .

If T n = (eo • • • en) is a degenerate simplex (or, what is the

^ R e m a r k 2.
same in our terminology, a degenerate skeleton), every simplex (de­
generate or not) whose skeleton is a proper subset of the skeleton (eo • • • e„)
is called a proper face of T n. This definition implies that a degenerate
simplex may have nondegenerate faces, e.g., three points A, B, C lying
on a straight line form a degenerate triangle which, however, has three
nondegenerate sides AB, BC, C A *
§0.3. Combinatorial sum. Let Ti , T 2, • • • , Tk be any faces of a simplex
T. The union of the skeletons of the simplexes , T2, • • • T k is a subset
of the skeleton of the simplex T and is therefore the skeleton of some face
118 COMPLEXES [CH. IV

To < T. The simplex T0 is culled the combinatorial sum of the simplexes

T\ , 7a , • • • , 1 \ and is denoted by (7'iT2 • • • T k).
The combinatorial sum is therefore defined for simplexes which are
faces of the same simplex.
Obviously:
1. Every simplex is the combinatorial sum of any two opposite faces
of the simplex.
2. Every simplex is the combinatorial sum of all its faces of a given
dimension (in particular, of all its vertices).
§0.4. The closure of a simplex T n = (c0Ci • • • e„) in R m is denoted by
T n = [e0 ■■• en] and is referred to as a closed simplex. A closed simplex
T n = [egei ■■■en\ consists of all those points of the space It71(eo, ei, ■■■, e„)
whose barycentric coordinates with respect to eo, e i, • • • , en are non­
negative.
The point set boundary or frontier T n \ T n of a simplex 7'" is the (point
set) union of all the proper faces of the simplex 71" and is denoted by
|| T n || (for reasons which will be made clear in the sequel).

§1. Basic definitions

§1.1. Triangulatioris. Examples: | T n | and f n.
D e f i n i t i o n 1.11. A finite set of mutually disjoint simplexes situated in
some R n is called a triangulation if every face of every simplex of K is
also an element of K. The set K may be partially ordered as follows: a
simplex T\ 6 K precedes a simplex 712 € K (7\ < Tf) if 7\ is a proper face
of 7V
The maximum dimension of the simplexes of K is .called the dimension
number or the dimension of K .
R e m a r k . The naturalness of this definition follows from, e.g., the reason­
ing and results of Chapter III: a considerable part of topology consists of
the study of the so called polyhedra, i.e., the sets CZ R n which can be
“triangulated” (split up into simplexes which form a triangulation); in
other words, sets which are the (point set) union of the simplexes of some
triangulation. At the same time we shall, of course, study curved (topo­
logical) polyhedra: topological images of triangulable sets. Topological
polyhedra are a far-reaching multi-dimensional generalization of the
closed surfaces of Chapter III. The investigation of the topological proper­
ties of curved polyhedra exhausts the study of polyhedral topology, since
curved polyhedra are homeomorphic to triangulations, and hence both
have the same topological properties. Triangulation is the basic auxiliary
technique of polyhedral topology; not only is it one of the most im portant
parts of topology, but it also suggests techniques for studying more general
5U BASIC DEFINITIONS 119

F i g . 79

topological spaces, to begin with, compacta. Hence, the enormous sig­

nificance of triangulations in modern topology is understandable. Tri-
angulations are not the ultimate objects of study in topology and occupy
a subordinate position in this respect, but they and their immediate
generalizations, as a technique, are basic to modern topological research.
This peculiar position of triangulations, at once subordinate and funda­
mental, should be clear from Chapter III (and Chapter V).
Examples of triangulations are easily constructed. Several examples
of one-, two-, and three-dimensional triangulations are shown in Fig. 79
and also in the figures of Chapter III. The set of all proper faces of a
tetrahedron, an octahedron, or an icosahedron is also an example of a
two-dimensional triangulation.
We shall give special attention to two elementary examples constantly
required in the sequel.
1. The triangulation consisting of an n-simplex T n and all its faces
is denoted by | T n | and called the combinatorial closure of the simplex T n.
2. The (n — l)-dimensional triangulation consisting of all the proper
faces of an n-simplex T n is denoted by T n and is called the (combinatorial)
boundary of the simplex T n (as distinguished from the frontier |] f n || of
T n, which is the point set union of all the proper faces of T n).
§1.2. Polyhedral complexes. The definition of a polyhedral complex
is obtained by replacing the word “simplex” in the definition of a triangula­
tion by “convex polyhedral domain” (see Appendix 1), leaving everything
120 COMPLEXES [CH. IV

else unchanged. In this connection, convex polyhedral domains (as well as

simplexes) are regarded as open, i.e., without frontier points: a square
means the interior of a square, a cube, the interior of a cube, etc. This
concept, at least in significance, is less important than the notion of a
triangulation, but it is so natural a generalization of the latter, th at it is
impossible to give it up. Besides, sometimes replacing a triangulation of a
given polyhedron by a more general polyhedral complex may simplify
some construction. Examples of polyhedral complexes are: the complex
consisting of all the faces of a cube, or a dodecahedron; the decomposition
into quadrilaterals, their sides and vertices, of a surface homeomorphic to
a torus shown in Fig. 80, and many others.
If T n is a convex polyhedral domain, its combinatorial closure | T n |
and its boundary Tn are defined exactly as in the case of a simplex (see
the preceding article).

§1.3. Skeleton complexes. This notion is important in the highest

degree, especially for a geometric treatm ent of the questions of set-
theoretic topology; but, in this book, the applications of'skeleton com­
plexes to the problems of polyhedral topology are essential. We shall
preface the definition of skeleton complexes with the,following remarks.
Simplexes, in particular simplexes of a given triangulation K, corre­
spond (1-1) to their skeletons, where a simplex T x £ K, precedes a
simplex T 2 (i.e., is a proper face of T 2) if, and only if, the skeleton of T\
is a proper subset of the skeleton of T 2 .
In many questions relating to triangulations, the point set character of
simplexes is unessential. What is significant is the order prevailing in the
triangulation as a partially ordered set, and the fact that each simplex is
assigned a dimension. In all purely combinatorial questions it is natural
and convenient to replace the given triangulation by the set of skeletons
corresponding to it, i.e., the set of skeletons of the simplexes of the given
triangulation, where the dimension assigned to each skeleton is the number
of vertices making up the skeleton less 1. Hence we arrive at the so called
skeleton complexes whose definition will now be given in a very general
form.
D e f i n i t i o n 1.31. Let M be a set of elements, called vertices. Lei certain
§1] BASIC DEFINITIONS 121

nonempty finite subsets of M be singled out and called skeletons. The number
of vertices in the skeleton less 1 is called the dimension of the skeleton. The
given set of skeletons is partially ordered by means of the natural order (I,
6.2, Example 4; i.e., T\ precedes T2 i f Ti C 712); it is called a skeleton com­
plex.
Hence a skeleton complex is an arbitrary collection of certain finite subsets
of a set M .
A skeleton complex is said to be unrestricted if every nonempty subset of
a skeleton of i f is also a skeleton (i e., an element of if).
R e m a r k . The skeleton complex of a given triangulation (i.e., the skeleton
complex of all the elements of this triangulation) is a finite unrestricted
skeleton complex. Finite, and especially unrestricted, complexes are the
most im portant of the skeleton complexes. However, restricted complexes
are also often used.
D e f i n i t i o n 1.32. Let i f be a skeleton complex, and let | i f | be the
skeleton complex such that every nonempty subset of every skeleton of if'
is a skeleton of | if |. The skeleton complex | i f | is unrestricted. We shall
call it the combinatorial closure of the complex if.
Obviously, i f is a subcomplex of | i f |. Hence
1.33. Every skeleton complex is a subcomplex of an unrestricted skeleton
complex.

§1.4. General definition of a simplicial complex. We shall begin with

the following example. Let us consider a tetrahedron T inscribed in a
solid sphere. Its vertices lie on the surface of the sphere and determine,
on the one hand, the faces and edges of the tetrahedron and, on the other
hand, the spherical triangles and their sides obtained by a central projection
of these faces and edges. Hence the same skeleton complex corresponds, on
the one hand, to an ordinary triangulation T and, on the other hand, to a
“curvilinear triangulation of the sphere”, whose elements have the same
skeletons as those of the triangulation T. Both kinds of elements are
“simplexes” , the former the usual, the latter the spherical, defined by the
given skeletons; it is natural to treat both sets of “simplexes” as “simplicial
complexes” . Thus we arrive at the following very general and final defi­
nition.
D e f i n i t i o n 1.41. Let i f be an unrestricted skeleton complex; let each
skeleton {e0 , • • • , er) of if be made to correspond to a unique object
(e0 • • ■eT) called an abstract simplex with vertices eo • • • , eT; in this con­
nection it is required th at the correspondence be (1-1) (distinct skeletons
correspond to distinct simplexes) and that it assign to each O-dimensional
skeleton, i.e., to each vertex e, as its simplex the vertex e itself. The skeleton
{eo, • • • , eT} is referred to as the skeleton of the simplex (e0 • • • ef), and
122 COMPLEXES [CH. IV

the dimension r of the skeleton {eo, • • • , er) is also the dimension of the
simplex (co • • • er).
A simplex (e'0 • • • e'T) precedes a simplex (eo • • • cr) or is a proper face
of the simplex (eo • • • cr) if the skeleton (e'o, • • • , c'r) is a proper subset
of the skeleton {eo, • • • , er).
The set of all abstract simplexes constructed in this way is called an
unrestricted simplicial complex and the original skeleton complex K is
called the skeleton complex of the simplicial complex.
In accordance with the above, every unrestricted simplicial complex is
a partially ordered set.
This definition implies that all unrestricted skeleton complexes, as well
as all triangulations, are special cases of unrestricted simplicial complexes.
[In the first case an abstract simplex (e0 • • • er) with skeleton \e0, • • • , er\
is itself th at skeleton, in the second case it is the usual simplex with vertices
e0 , ■• • , er .] A curved triangulation is also an unrestricted simplicial
complex (see §G), whose two-dimensional case was discussed in Chap­
ter III.
D e f i n i t i o n 1.42. Every subset (“subcomplex”) K 0 of an unrestricted
simplicial complex K is called simply a simplicial complex. The elements
of the complex Ko have the same skeletons, the same dimensions and the
same order as in K. The combinatorial closure | Ko | of a complex Ko (in
K ) is, by definition, the unrestricted subcomplex of K, whose skeleton
complex is the combinatorial closure of the skeleton complex K 0 .
D e f i n i t i o n 1.43. If the elements of a simplicial complex K have a
maximum dimension, this maximum is called the dimension (number) of the
simplicial complex A"; if a maximum does not exist, the complex K is said
to be infinite dimensional.
§1.5. Examples of simplicial complexes. 1. Let A be a triangulation,
and Ko a subcomplex of K; Ko is a simplicial complex.
2. Let T be an open subset of R n or the n-sphere S n. A finite set
{Co , ‘ , cr |
contained in F is called a skeleton if its closed convex hull is in T (if V C
S n, it is required, in addition, that the set {eo, • • • , er) be contained in
some hemisphere of S 71 and that closed convex hull be understood in the
sense of the spherical metric in S n). The resulting unrestricted skeleton
complex is denoted by K(F). The subcomplex of A(T) consisting of all
skeletons whose diameter is less than some definite positive number e,
is an unrestricted skeleton complex denoted by A(T, e).
Let us consider the simplicial complex having A(r), K(F, e), respec­
tively, as its skeleton complex, with its simplexes defined as follows: if the
points c0 , • • • , eT of a skeleton are linearly independent in R n (in S n), the
§1] BASIC DEFINITIONS 123

ordinary (or spherical) simplex with vertices e0 , • ■■ , er is called a simplex

(co • • ■er).
But if the points e0 , ■■■ , er are linearly dependent (which, in par­
ticular, will be true if r > ?i), then set
(fo ' ‘ ' Cr) {Po j ' ' ' > >

i.e., identify the simplex (e0 • • ■eT) with its skeleton. The resulting sim-
plicial complex is at times also denoted by K ( T) or /C(r, e), as the case
may be.
3. Let R be a metric space. Denote by K ( R ) the unrestricted skeleton
complex obtained by letting every finite subset of R be a skeleton; K{R)
contains a subcomplex K(R, t) consisting of all the skeletons of the com­
plex K(R) whose diameter is < t.
R e m a r k . The case of R a compactum is especially important.
If R is a subset of R n, the complexes K(R) and K(R, e) can be replaced
by simplicial complexes with the same skeletons but with simplexes de­
fined as at the end of Example 2.
4. Let T be an open subset of a compactum $>. Let us call every finite
subset of <t> which has a nonempty intersection with Y a skeleton. The
resulting skeleton complex is denoted by K($, T); if it were required
th at every skeleton have a diameter <e, the result would be a skeleton
complex K (4>, T, «) C K($, T).
Other important examples of skeleton complexes are discussed in §2.
§1.6. Simplicial mappings and isomorphisms of skeleton and simplicial
complexes.
D e f i n i t i o n 1.61. Let us assign to every vertex ep of an unrestricted
skeleton complex Kp a vertex ea = S j e a of a skeleton complex K a in
such a way that the image of every skeleton \eg0 , • • • , epr\ of Kg is a
skeleton of the complex K a (distinct vertices of Kp may correspond to a
single vertex of K u). In virtue of the condition imposed, the mapping
S j of the set of vertices of Kp into the set of vertices of K a induces a
mapping of the complex Kp into the complex K a , referred to as a simplicial
mapping of the unrestricted skeleton complex Kp into the skeleton complex
K a , also denoted by S j -
If Kp and K a are the unrestricted skeleton complexes of the unrestricted
simplicial complexes K'p and K 'a , respectively, then an assignment to
each simplex (ep0 ■■■epT) € K'p of the simplex of the complex K ' a whose
vertices are £„%„ , • • • Sj e p r yields, by definition, a simplicial mapping
S j of the simplicial complex K'p into the simplicial complex K ' a (induced
by the identically designated simplicial mapping of the skeleton complex
K'p into the skeleton complex K ' a)•
D e f i n i t i o n 1.62. Let Kp be a restricted simplicial complex. A simplicial
124 COMPLEXES [CH. IV

mapping of Kg is any simplicial mapping of the combinatorial closure

| Kg | of Kg restricted to Kg .
D e f in it io n 1.G3. A (1-1) simplicial mapping of an unrestricted
simplicial complex Kg onto (i.e., every skeleton of K a is the image of at
least one skeleton of Kg) an unrestricted simplicial complex K„ is called an
isomorphic mapping or an isomorphism. Two unrestricted simplicial com­
plexes are said to be isomorphic or to have the same combinatorial type if
there is an isomorphic mapping of one onto the other.
Two unrestricted simplicial complexes are isomorphic if, and only if,
their skeleton complexes are isomorphic (in particular, every unrestricted
simplicial complex is isomorphic to its skeleton complex). Hence the
combinatorial type of an unrestricted simplicial complex is uniquely
determined by its skeleton complex.
R e m a r k o n I s o m o r p h i s m . A simplicial mapping S j of an unrestricted
simplicial complex Kg onto an unrestricted simplicial complex K a is an
isomorphism if, and only if, every vertex of K a is the image of exactly one
vertex of Kg .
E xam ples of I s o m o r p h ic S im p l ic ia l C o m p l e x e s .

1. The triangulation consisting of all the proper faces of a convex

polyhedral domain with triangular faces (e.g., a tetrahedron, an octa­
hedron, an icosahedron, etc.) is isomorphic to the “curved triangulation”
of the sphere obtained by projecting the polyhedral domain onto a sphere
in its interior.
2. Let AT be a triangulation. Let us construct the simplicial complex
K ' with the same skeleton complex as that of K but with simplexes which
are defined as the closures of the simplexes of K. : -
The complex K ' is isomorphic to the complex K.
R e m a r k o n I d e n t i f i c a t i o n s . Let Kg and K a be two simplicial complexes
and S j a simplicial mapping of Kg onto K a:
Sa Kg = K a .
To every vertex eai 6 K a corresponds a class e, of vertices e^yof K qsuch that
$ oe @8 j ‘ &a i •

Hence the simplicial mapping S j induces a decomposition of the set

of all vertices egj of the complex Kg into classes. We note that the vertices
fao, • • • , Car 6 K a form a skeleton in K a if, and only if, the corresponding
classes c0 , • • • , er contain vertices e0i 6 Cy, j = 0, 1, • • • , r, which form a
skeleton in Kg .
Hence the simplicial mapping S j 3 is equivalent to an identification
(see I, 5.1 and III, 3.1) of all the vertices of Kg which are in the same
class tj .
§ 1] RASIC DEFINITIONS 125

Conversely, let K be a finite unrestricted simplicial complex and suppose

th at its vertices are divided into classes c,-. A new simplicial complex
K a may be defined as follows: K a is a skeleton complex whose vertices
are the classes c,- of vertices of Kp . The classes to tr form a skeleton
if it is possible to choose vertices ep, ^ = 0, 1, , r, which form a
skeleton in Kp . It is obvious that the result is an unrestricted skeleton
complex.
The simplicial mapping S j 3 of Kp onto K„ is defined in the natural
way, i.e., each vertex ep{ £ /vs is mapped into the class e,- containing it.
Let us suppose that a simplicial mapping of Kp onto K a satisfies the
following condition: S j maps no pair of vertices contained in the same
skeleton of the complex Kp into the same vertex of the complex K a .
Such mappings are called identifications; an identification maps every
element of the complex Kp onto an element of the same dimension of K a
and hence two elements of different dimensions of Kp cannot be mapped
on the same element of K a ; but two elements of Kp having the same
dimension can be identified, i.e., they can be mapped onto the same ele­
ment of K* .
The identifications considered in Chapter III are a special case of this
general concept.
§1.7. Definition of an abstract complex. Considering the various defi­
nitions of complexes given in this section, it may be noted that they are
all special cases of a partially ordered set 0 to each of whose elements 6 is
assigned a nonnegative integer dd, the dimension of the element 8, such that
8i < d2 implies that d8i < dd-2 .
We shall take this, for the time being, as the definition of an abstract com­
plex. The abstract complexes considered in Chapter VII will be of a more
restricted nature.
A similarity mapping of one complex onto another which preserves
dimension is called an isomorphic mapping of the two complexes.
This definition of isomorphism in the case of unrestricted simplicial com­
plexes coincides with that which was given in l.G.
The notions of abstract complex and isomorphism introduced above
enable us to give the definition of an unrestricted simplicial complex the
following irreproachably simple and transparent form:
A n unrestricted simplicial complex is an abstract complex isomorphic to
some unrestricted skeleton complex. Simplicial complexes (not necessarily
unrestricted) are arbitrary subcomplexes of unrestricted simplicial complexes.
(Every subset K 0 C K , whose elements have the same dimensions and tin;
same order as in K, is a subcomplex of K.)
R e m a r k . The notion of isomorphism as applied to restricted simplicial
complexes will not interest 11s: in our exposition every restricted simplicial
126 COMPLEXES [CH. IV

complex will always appear as a subcomplex of some perfectly definite

unrestricted complex K and we shall consider only those isomorphic map­
pings of the complex K 0 which are induced by isomorphisms of the unre­
stricted complex K.
§1.8. Closed and open subcomplexes of a complex K. Combinatorial
closures and stars. (In this article complexes are to be taken in the ab­
stract sense as defined in the preceding article.)
A subcomplex K 0 of a complex K is said to be a closed (open) subcomplex
of K if every element of K which precedes (i.e., is less than) some element
of Ko [which'follows (i.e., is greater than) some element of A 0] is itself an
element of Ko .
The proof of the following propositions may be left to the reader.
1.81. If Ko is closed (open) in K, K \ K 0 is open (closed) in K.
1.82. If Ko is an arbitrary, and Ki a closed (open), subcomplex of a
complex K, K 0 n K\ is a closed (open) subcomplex of the complex Ko ■
1.83. Every closed subcomplex of an unrestricted simplicial complex is
an unrestricted simplicial complex.
D e f i n i t i o n 1.84. If K 0 is any subcomplex of a complex K, the complex
consisting of all the elements of K 0 and of all the elements of K less than
at least one element of A'o (greater than at least one element of Ko) is
called the combinatorial closure | Ko | (star OrKq) of the subcomplex K 0
in the complex K.
1.85. For arbitrary Ko Q K, the complex | K 0 | is closed, and the complex
OkK ois open, in K.
R e m a r k . In the sequel we shall consider only stars O r T of individual
elements of a complex K, i.e., complexes OkK 0 in the case that Ko consists
of a single element T £ K. In this connection, the element T is called the
center of the star O r T . Some examples of stars were given in III, Figs.
20-25.
D e f i n i t i o n 1.86. The complex

(1.86) B rT = jOkT | \ OrT

of all the elements not contained in the star O r T and less than at least one
element of O k T is called the o u t e r b o u n d a r y of the star O r T .
Since O r T is open in K, and hence in | O r T |, B kT, by (1.86) and 1.81,
is closed in | O r T | and consequently in K. Thus
1.87. B r T is a closed subcomplex of K\ if K is an unrestricted simplicial
complex, B r T is an unrestricted simplicial complex.
D e f i n i t i o n 1.88. (This definition is not needed until Chapter X III.)
Let K be a finite simplicial complex and let T v 6 K. Consider the complex
F rT v C I\ consisting of all the simplexes T- (j K satisfying the condition:
§1] BASIC DEFINITIONS 127

there exists a simplex j 1P+T+1 £ OkT v of which T / is the face opposite the
face T p < T p+T+1. The complex FKT P is called the zone of the star OkT p
(in K).
S t a r s in S i m p l i c i a l C o m p l e x e s . Let Ok Ti , 0 KT2, • • • , 0 kT t be stars
of a simplicial complex K. In accordance with the definition of a star, the
intersection Ok Ti n • • • n 0 kT t consists of all the simplexes T £ K which
satisfy all the conditions
1\ < T, T, < T, ■• • , T r < T.
Exam ple. Let the complex K consist of the ten triangles and of the two
vertices p\ and p2 shown in Fig. 81. The intersection of the stars 0 Kp\ and
0 Kp2 consists of the two hatched triangles.
T h e o r e m 1.S9. The intersection of stars Ok Ti , OkT, , • • • , OkTt of
an unrestricted simplicial complex K is nonempty if,and only if, there is a sim­
plex in K having all the simplexes T\ , T2 , • • • ,
T t as faces. I f this is the case, denoting by To the
combinatorial sum of the simplexes i \ , ■• • , T r ,
we have
0 kT ! n 0 K1\ n • • • n 0 KT r = OKT0 6------ .8
F ig . 81
Proof. Since K is unrestricted, it follows that if
the simplexes 1\ , • ■■ , T r of K are faces of the same simplex T of K, then
the combinatorial sum
To = ( T , T , - - - Tt)
of all these simplexes is itself a simplex of K. Obviously, every simplex of
K having I \ , ■• • , T r as faces also has To as a face.
Suppose that a simplex T is in the intersection of the stars OkT x , • • • ,
0 kT t of K.
Then the simplexes 1\ , • • • , T r are faces of T, which consequently also
has To = ( T i T 2- ■-Tr) as a face, and therefore is an element of the star
OkT q .
Conversely, every simplex T 6 OKT0 has T 0 as a face and consequently
has T] , • • • , T r as faces, i.e., is in the intersection of all the stars O k , ■ • • ,
0 kTt . This proves the theorem.
C o r o l l a r y . In particular, the stars OKe0 , 0 Kc\ , • • • , 0 Ker of vertices
e0 , • • • , eT of a complex K have a nonempty intersection if, and only if,
K contains a simplex T 0 = (e0- ■-eT).
§1.9. Theorem on imbedding in R2n+l.
T heorem 1.9. Every finite unrestricted n-dimensional skeleton complex K
(hence also every finite unrestricted simplicial n-complex) is isomorphic to
128 COMPLEXES [CH. IV

some triangulation K ' situated in Euclidean (2n + 1)-space. (The extra­

ordinarily simple proof of Theorem 1.9 given below was communicated to
me by L. S. Pontryagin.)
C o n s t r u c t io n o f t h e T r ia n g u l a t io n K '. Let e\ , • • • , e, be the
vertices of the complex K . Take points e'\ , • ■• , e'a in general position in
R 2n+1 (i.e., such that for k < 2n + 2 every k of the points e \ , • ■■ , e',
are linearly independent; see Appendix 1). Put each skeleton T =
{eiQ , • ■■ , e;r} £ K in correspondence with the simplex T' = (e';0 •' ■• e'iT) d
R 2n+1; this simplex exists since, by "virtue of the general position of the
points e'\ , ■• • , e'„in R 2n+1 and the inequahty r < n, the points e'io , • • • ,
e ir are linearly independent. The simplexes T'i obviously form a complex
K ' isomorphic to K. We shall prove th at K ' is a triangUlation. To do this
it suffices to prove th at every two simplexes T'i € K ', T 'j € K ' are disjoint.
Let the vertices of the simplex P,- be e'i0 , • • • , e'ip and let the vertices of
the simplex T 'j be e'j0 , • ■■ , e'jq , where some of the vertices may be com­
mon to T'i and T ' j . Let e'*0 , • ■• , e'kr be all the points which are vertices
of at least one of the simplexes T ' i , T'j . The number r + 1 of these points
satisfies the inequahty
r + 1 < (p + 1) + (q + 1) < (n + 1) + (n + 1) = 2n + 2,
since the dimensions of the simplexes T'i and T'j do not exceed n. In vir­
tue of the general position of the points e \ , ■• • , e's in R 2n+l, the points
e'io , • • • , e'kT are the vertices of some nondegenerate simplex T 0 of dimen­
sion^ 2n + 1. The simplexes T'i and T 'j are faces of the simplex T 0 and
hence are disjoint if they are distinct.
1.9 implies
1.91. Every n-dimensional finite skeleton complex K .is isomorphic to a sub­
complex of some n-dimensional triangulation whose simplexes are situated in
2n+l

Indeed, the combinatorial closure | K \ is an unrestricted n-dimensional

skeleton complex isomorphic, by 1.9, to a triangulation Q whose simplexes
are in R 2n+l. The isomorphism between | K \ and Q maps the complex
K d | K | onto some subcomplex of Q. This proves 1.91.
§2. Some notable skeleton complexes
§2.1. The nerve of a finite system of sets. This notion will have nu­
merous and very essential applications in this book.
Let
(2.1) « = {Ai, , A.}
be a finite system of sets. To each of the sets A i assign a vertex a,- . These
vertices (elements of a perfectly arbitrary set; see Def. 1.31)
^1 ) *** > 0'S
§ 2] SOME NOTABLE SKELETON COMPLEXES 129

will also be the vertices of a skeleton complex K a , which we shall now de­
fine and call the nerve of the system a. We shall call a given collection of
vertices

a skeleton of the complex K a if, and only if, the sets

•^■*0 > ' l A ir
have a nonempty intersection.
R e m a r k 1. This definition does not exclude the case that several different
elements of the system a may coincide as sets [i.e., differ only in their
indices in (2.1); see I, 1.3],
R e m a r k 2. The definition of nerve immediately implies th at every non­
empty subset of any skeleton of a nerve is also a skeleton of the nerve.
Thus:
2.11. The nerve of every finite system of sets is an unrestricted simplicial
complex.
It follows immediately from the definition of the order of a system of sets
(I, 1.3) and the definition of a nerve that:
2.12. The dimension of the nerve of a system of sets is one less than the order
of the system.
R e m a r k 3. The nature of the vertices a,- is completely immaterial: any
elements which can be put in (1-1) correspondence with the elements of
the system a can be taken as the vertices a , ; from the logical point of view
it is easiest of all to take as the vertices a; the elements themselves.
Then the nerve of the given system of sets would be uniquely defined and
independent of the choice of the vertices.
However, we shall see in the sequel that arbitrariness in the choice of the
vertices and a certain indefiniteness arising from it, however unessential,
is convenient in practice.
We shall therefore regard the nerve of a system of sets as defined only up
to an isomorphism, i.e., we shall call every complex isomorphic to the nerve
of a given system of sets also the nerve of this system of sets.
Passing to examples of nerves, we note first that the Corollary to Theorem
1.89 can be formulated as follows:
2.13. The stars of the vertices of a finite unrestricted simplicial complex K
form a system of subcomplexes of K having the complex K as its nerve.
I t follows from 2.11 and 2.13 that every finite unrestricted simplicial com­
plex is the nerve of a finite system of sets and conversely. Therefore finite
unrestricted simplicial complexes are sometimes referred to as simple nerves.
E x a m p l e s o f N e r v e s . 1. Let us consider-the system of nine closed
squares shown in Fig. 82. The nerve of this system is a 3-complex shown
by dotted lines in the same figure.
130 COMPLEXES [CH. IV

--------- \
\ - / /
\
/

//
*
\
s
\
1
*
.i/
\ V /
\ t ✓
v
' t

t »
. t \
/ \\ i i
\
\ "
\ ,
; x \

V
F ig . 82

2. A second example is obtained by taking as the system a the twenty

closed squares shown in Fig. 83. The nerve of this system is also shown by
dotted lines in the figure.
3. The system a consists of the six closed faces of a cube; its nerve is the
complex consisting of all the 2-faces, edges, and vertices of an octahedron.
4. The system a consists of seven closed sets: a closed triangle, its closed
sides, and its vertices. The nerve of this system is the 3-complex shown in
Fig. 84.
R e m a r k 4. The nerve constructed near a given system of point sets. Let the
elements Ai , • • • , A s of a system of sets
a = {Ai , • • • , A s\
be subsets of a given R n; let e > 0. Let us take as the vertices ai , • • • , a>
of the nerve K a of a any points ai , • • • , as of R n which are distinct and
satisfy the condition
p(fli, A t) c, i 1, 2, • • • , s.
A nerve K a of the system a with vertices ai , • • • , as satisfying these
conditions is called a nerve contained in an t-neighborhood of a.
If n is not less than the dimension m of the nerve K a , let us choose the
points ai , • •• , as in general positon. Then all the skeletons of the nerve
are sets of linearly independent points. Hence simplexes of R n, defined by
the corresponding skeletons of the nerve, can be taken as the simplexes of
the nerve K a . If, in addition, n > 2m -+- 1, the resulting nerve K a will not
only be contained in an t-neighborhood of a but will be a triangulation as well.
Hence
2.14. Let a be a finite system of sets contained in R n, and let t > 0. It is
possible to construct, in every space R m 3 R n of sufficiently high dimension,
a triangulation K a which is a nerve of a contained in an t-neighborhood of a.
The nerve of the system of closed squares shown in Fig. 83 is a nerve
contained in an t-neighborhood of this system for arbitrary t.
§2] SOME NOTABLE SKELETON COMPLEXES 131

§2.2. Barycentric derivation and barycentric subdivisions. Let 0 be a

finite partially ordered set; we shall call its elements p vertices. Those sub­
sets of 0 which arc simply ordered (with respect to the order obtaining in
0), we shall call skeletons.
The resulting (obviously unrestricted) skeleton complex is denoted by
B(0) and is called the barycentric derived of the partially ordered se.t 0.
If 0, in particular, is a polyhedral complex K or a subcomplex if of a
polyhedral complex (see 1.2; the reader may restrict himself to the case of
K a triangulation or a subcomplex of a triangulation), it is possible to con­
struct a triangulation K x isomorphic to the complex B(K) in a particularly
intuitive way. This is done as follows.
Let
t i , •• • , r s

be all the polyhedral domains of K. All these polyhedral domains are con­
tained in some R n. Let us choose in each T *a point e< called the center of
the polyhedral domain; it is customary to choose the centroid of T { as the
point e; . The vertices of the complex K x are the points et- . A number of
vertices e; form a skeleton if the set of polyhedral domains 7\- correspond­
ing to them is simply ordered (with respect to the geometric order, i.e.,
Tj < Ti if Tj is a proper face of Ti). This means that the given vertices
can be written in a sequence
^ ! -0 ) ® t 'l ) ' j ® i'r

such that
(2.20) T i0 > T h > • • • > T ir
(i.e., each of the polyhedral domains T,-0 , • • • , T ir , except T t0 , is a proper
face of the polyhedron preceding it).
From (2.20) and the fact that et- £ T : it follows that the points e;0 ■,
e,-r are linearly independent in R ’\ so that they form a simplex (e,0 • • • eir) (Z
R n.
Hence all the skeletons {e,0 , ••• , elr) are the skeletons of simplexes
(elV • -eir) C R n. These simplexes (e,0---e,r) are also, by definition, the
elements of the complex K x .
The complex K x is isomorphic to the barycentric derived of the arbitrary
complex K , by construction: the isomorphism is realized by assigning the
simplex (e,0- • -eir) of Ki to the skeleton {Tio > • • • > 7\ r ) of the complex
B(K). In consequence, the complex K x is cafied the geometric realization
of the barycentric derived B(K).
R e m a r k 1. If (e,0- • -e,-r) is a simplex of K x and

T !o > ■ > T ir ,
132 COMPLEXES [CH. IV

the vertex eio is called the first, and the vertex e,r the last, vertex of the
simplex (eio- • -eir).
T h e o r e m 2.21. The complex i f i is a triangulation.
Proof. K i is a finite complex, by definition. Since ifi and B (K ) are iso­
morphic and B(K) is an unrestricted complex, ifi is also an unrestricted
complex.
I t remains to be proved that every two distinct simplexes of ifi are dis­
joint.
The assertion is obvious for the O-simplexes of ifi . Before going on to
the general case, let us note the following:
Since two distinct elements T t and Tj of i f are disjoint, two simplexes
(eio- • •«;„), (eJo- • -ejj € ifi are known to be disjoint if their first vertices
eio and ej0 are distinct.
Let us now assume that two simplexes of ifi are disjoint if their dimen­
sions are less than or equal to r. This is true for r = 0. Let us consider
two simplexes (e»0- ‘ ' ev), (eh ' ' ' ei,) € ifi whose dimensions do not ex­
ceed r + 1. If they are not disjoint, the first vertex of both is eio = e,-0 .
But then the simplexes (eioefl- • -el(1) and (ej0eJl ■• -ejf) are the projections
from the point e,-0 = eJ0 of the simplexes (e;i- • -e,f) and (ejl ■■-e7J and
consequently intersect only if (e,-, ■• and (e^- • -ej,) intersect. B ut the
dimensions of ( e ^ - ’ -e^) and do not exceed r. Therefore, if
these simplexes intersect,
M = V i e «'l = ‘ ' J ~ r •

But then (ei0• • -e,^) and (e;o- • -ey„) coincide, which was to be proved.
2.210. Every simplex T Ti of K i is wholly contained in some element T of K
(i.e., is a subset of the polyhedral domain T).
Indeed, of the vertices eio , • • • , eir of the simplex T \ , the first vertex
eio is contained in the element T io £ K and all the rest in its faces. There­
fore the whole simplex Tr1 = (eio • • -eir) is contained in Tio .
Now let i f be a polyhedral complex (until now K could be an arbitrary
subcomplex of a polyhedral complex). Then the triangulation K\ is called
the barycentric subdivision of K and Theorem 2.210 is completed in an
essential way by the proposition:
2.22. Every element T k of the complex K is the union of the simplexes of the
barycentric subdivision K i contained in Tk. (Without great damage the
reader may assume that i f is a triangulation and, accordingly, consider
the polyhedral domains of i f to be simplexes.)
Proof. The theorem is obvious if k = 0, i.e., if Tk is a vertex of if, since
every vertex of i f is at the same time a vertex of i f i .
Let us suppose that Theorem 2.22 has been proved for k < r and prove
it for k = r + 1. Let us consider all the simplexes T \ h of 7fi whose first
vertex is the centroid of one of the proper faces of a given (r + l)-element
§2 ] SOME NOTABLE SKELETON COMPLEXES 133

T T+1 £ K . The collection of these simplexes T'ih constitutes a barycentric

subdivision of the complex T T+1 = \ T r+1 | \ T T+1, where, by assumption,
every point of the boundary | T r+l | \ T r+1 of the polyhedral domain
Tr+1 is contained in some simplex T \ h .
Now let p £ Tr+1; if the point p is the centroid of T r+1, it is a vertex
of K i . Suppose th at p is not the centroid of T T+1; denote by p' the pro­
jection of the point p on T r+1 \ T r+1 from the centroid o of T T+1 and let
T\h = (ex - • -e.) be the simplex of K i containing p'. Then p is a point of
the simplex (oer • -e.) of K i contained in T r+1, which was to be proved.
Let us combine propositions 2.21, 2.210, and 2.22 into one theorem:
2.2. The bary centric subdivision of a 'polyhedral complex K is a triangula­
tion K i such that every simplex of K \ is contained in some element of K and
every point of any element of K is contained in some simplex of K\ .
E x a m p l e s . 1. Fig. 85 shows a) a barycentric subdivision of a two-
dimensional triangulation consisting of two triangles A BC and BCD (to-

B B
134 COMPLEXES [CH. IV

gether with their sides and vertices); b) the realization of the barycentric
derived of the subcomplex (nonclosed) of this triangulation consisting of
the triangles ABC and BCD, the sides AB, AC, and BD and the vertices
A and D.
2. Fig. 8 G shows the barycentric subdivision of a polyhedral complex
consisting of all the faces, edges, and vertices of a cube (the only elements
shown are those turned to the observer).
3. Fig. 87 shows one of the 24 3-simplexes of the barycentric subdivision
of the complex | T 3 \.
§2.3. The cone of a complex. Let K be a simplicial complex. In the se­
quel, we shall identify K with its skeleton complex.
Let o be a vertex not in K.
Let us construct the skeleton complex < o K > , which, by definition,
consists of the following skeletons: the vertex o, all the skeletons of K,
and all sets of the form {o, ei , • • • , e,}, where [ei , • • • , eT) is any skeleton
of K. The complex <oK> , as well as every complex obtained from < o K >
by an isomorphic mapping which leaves invariant all the elements of K
(regarded as a subcomplex of <oK>) , is called a cone-complex (or simply
cone) with vertex o and base K.
The cone-complex of an unrestricted simplicial complex is an unrestricted
simplicial complex.
The complex

oK = <oK> \ K,

i.e., the star of the vertex o in <oK>, is called an open cone with vertex o
and base K.
The following remark, due to L. S. Pontrvagin, will be essential in
Chapter X III.
2.31. Let K be an unrestricted simplicial complex, e a vertex of K, B the
outer boundary of the star 0 Ke, and T an arbitrary element of the complex B.
Then the star of eT in K is the open cone with
vertex e and base 0 B T:
(2.31) 0 KeT = eOBT.
The proof consists in the immediate verifi­
cation of the fact that both sides of (2.31)
consist of the same simplexes. This may be left
to the reader; in Fig. 8 8 the complex consists
of the tetrahedra (ee3e2ei), (ee3c2c4) and all their
faces; T = (e3e2); the complex B consists of
the triangles (e3e2Ci) and (e3e2e4) and their
F ig . 88 faces, 0 bT consists of the segment (e3e2) and
§2 ] SOME NOTABLE SKELETON COMPLEXES 135

the triangles (^362^1) and (636264); both sides of

(2.31) consist of the triangle (66362), and of the two
tetrahedra (ee362ei) and (6636264).

§2.4. Prisms over a skeleton complex. Every

subdivision of a parallelogram by a diagonal into
two triangles, as well as the subdivision of a
trihedral prism into tetrahedra (Fig. 89), known
from textbooks of elementary geometry, are
special cases of the following general construction.
Let K be a nonempty finite unrestricted skele­
ton complex. Let us enumerate in a definite order
all the vertices of K : a i , • • • , as . Let us consider
new vertices bi , • • • , b, which correspond (1—1 ) to the vertices a i , • • • ,
ag . The vertices a,- and bi are, by definition, the vertices of a skeleton com­
plex K m which is defined as follows. The skeleton complex /C[on consists,
by definition, of all nonempty subsets of sets of the form
(2-4) a,0 , ■• ■ , flit , b{k , ■• • , bir ,
where
io < • ■• < ik < • • ■ < iT
and
(o,0 ■• ■Ouj." • -a,r) K
[in particular, of all the skeletons (a,0- • -alr) of the complex K and of all
the skeletons (bio- • •b,T) corresponding to them].
The complex is called the prism over the skeleton complex K. If K
is an r-complex, ifjoi] is an (r -f- l)-dimensional unrestricted skeleton com­
plex.
The complex K is referred to as the lower base of the prism K\oi] (and is
sometimes denoted by K 0). The upper base of the prism /C[0i] is the complex
Ki Cl iC[o]i isomorphic to the complex K and consisting of all the skeletons
(6 ,0- • •b;T) corresponding to the skeletons (a,0- • -a,r) of K .
§2.5. The prism spanned by a skeleton complex and its simplicial
image. Let K 0 be a closed subcomplex of a finite unrestricted skeleton
complex K and let *S° be a simplicial mapping of K 0 into K which satisfies
the condition: for every T 0 6 Ko there is a T £ K such that To and ^ T 0
are faces of the simplex T. Enumerating all the vertices of Ko in a definite
order eu , ■• ■ , a , , let us construct the prism / f [0i] over K 0 ; let the corre­
sponding vertices of the upper base K\ of the prism be b\ , • • ■ , bs .
Let us bow set
= a,-, S01bi = S°ot i = 1 , • • • , s.
136 COMPLEXES [CH. TV

Then S01 maps each skeleton

T = (o,#- • •aikbik • • •bir) 6 if[oi] , i0 < • ■• < iT,
into the skeleton
t f 'T = (ai0--- ah^ a ik- - - ^ a ir) € K,
so th at the mapping 501 of the prism into K is simplicial; its image
t f ' K m is called the prism spanned by K 0 and {fiKo in K. (Since a,-0 , • ■■ ,
a%k , • • • , a-iTform a skeleton in K o , it follows by assumption th at the ver­
tices aio, • ■■ , aik, • • • , a,-r , iS°a<0 , • • • , S°aik, • • • , S°air and, a fortiori,
a,0 , • • • , aik, tfiaik, • • • , S°a,-r form a skeleton in K .)
§3. The body of a complex. Polyhedra
§3.1. Definitions.
D e f in it io n 3 .1 1 . Let K be a complex which is either a triangulation or
a subcomplex of a triangulation. The union of the elements of K (regarded
as point sets of a given R n) is called the body of the complex K and is de­
noted by || K ||.
R e m a r k 1 . The same definition is also applicable to a wider class of com­
plexes, namely, to all complexes whose elements are simplexes or, in general,
polyhedral domains of a given R n.
D e f in it io n 3 .1 2 . A set which is the body of some triangulation is called a
polyhedron.
R e m a r k 2. I t follows from 2 .2 that the body of every polyhedral com­
plex is also the body of some triangulation (for example, the body of the
barycentric subdivision of a complex K ) and, consequently, is a polyhedron.
Hence, polyhedra can be defined as the bodies of polyhedral complexes.
We have the following theorem:
3 .1 3 . Let K be a finite set of convex polyhedral domains of a given R n bound
with only one restriction: every face of an arbitrary element of K is itself an
element of K . The union || K || of all the elements of K {regarded as point
sets of the given R n) is a polyhedron.
The proof of this theorem is left to the reader: let us note a method of
proof. Let us suppose that the theorem has been proved for the case that
all the elements of K are contained in the union of a finite number of
m-planes (for m = 0 the proposition is obvious) and prove it if all the ele­
ments of K are contained in the union of a finite number of {m + 1)-
planes Rim+1, • • • , R ,m+1 of R n. Let R i , • • • , R u be all the planes defined
as the intersection of any two of the planes R;m+1 and R jm+1 and also the
planes carrying any m-element of K. If any of these planes, say R i , has
dimension m, let us write R m,• instead of R { . B ut if the dimension of Ri is
less than m , denote by R m; any m-plane passing through R i . Hence we
[§3 THE BODY OF A COMPLEX. POLYHEDRA 137

obtain m-planes R mi , • • • , R mu . These m-planes divide H = R im+1 u • • • u

R ,m+1 into a finite number of convex sets open in H ; those of the domains
which contain points of the set || K || are convex polyhedral domains in
|| K ||. Let us denote these convex polyhedral domains by Tj , • • • , r r .
By the inductive hypothesis, the set || K || n (Rmi u ; • • u R mu) is a poly­
hedron; a triangulation of this polyhedron, induces a triangulation of the
boundary of each of the convex polyhedral domains I \ . The projection of
each of these triangulations from any interior point of the corresponding
leads to a triangulation of the entire set (Ti u • u Tr) u (|| K || n
{Rm\ u • • • u R mu)) = || K ||, which is therefore also a polyhedron. (For a
detailed proof see Alexandroff [A-H, pp. 141-143].)
From Theorem 3.13 it follows at once that
3.14. The union of a finite number of polyhedra contained in a given R n
is a polyhedron.
E x e r c i s e . Prove the following theorem:
3.15. The intersection of two polyhedra is a polyhedron; the closure of the
difference of two polyhedra is a polyhedron.
Every triangulation whose body is a given polyhedron is called a tri­
angulation of the given polyhedron.
E x a m p l e . Let T n be a simplex. The body of the complex | T n | is ob­
viously the set || | T" | || = T n.
If A is a triangulation and T £ K , then | T | C K , and therefore T C
|| K ||. Hence every polyhedron is the union of the closures of a finite num­
ber of simplexes, i.e., a closed bounded set of Euclidean space. Consequently,
every polyhedron is a compactum.
T h e o r e m 3.1. I f K ' is a subcomplex of a polyhedral complex K , || K ' ||
is closed {open) in || K || if, and only if, K ' is a closed (open) subcomplex of K .
Proof. Since open sets in || K || are the complements of closed sets in
|| K || and open subcomplexes of K are the complements of closed sub­
complexes of K , it suffices to prove the assertion for closed sets and sub­
complexes.
Let K ' be a closed subcomplex of K . Then K ' is a polyhedral complex,
so th at || K ' || is a polyhedron. Hence || K ' || is a compactum, and is there­
fore closed in every set containing it, in particular, in || K ||.
Let K ' be a nonclosed subcomplex of K and let T £ K be a face of any
element of K ' not in K '; all the points of || T || C || K || are limit points
of || K ' || which do not belong to || K ' ||. Hence || K ' || is not closed.
§3.2. Star neighborhoods. Open stars.
D e f i n i t i o n 3.2. Let K be a triangulation. Let p 6 || K ||. The unique
element T{p) of the complex K which contains the point p .is called the
carrier of p in K. The set || 0 Kp || = || 0 KT{p) ||, which is open in || K ||,
is called the star neighborhood of p in K .
138 COMPLEXES [CH. IV

D e f in it io n 3 .2 0 . The star neighborhoods of the vertices of a triangula­

tion K (with respect to this triangulation) are called the open stars of the
triangulation K.
Hence the open stars of a triangulation are open subsets of its body.
T h e o r e m 3 .2 1 . The open stars || 0*e || of a triangulation K cover the poly­
hedron || K ||.
Indeed, let p £ || K ||, let T be the carrier of p, and let e be a vertex of T .
Obviously,
p £ T £ OrS,
p e T Q I! 0 Ke ||.

This proves the theorem.

T h e o r e m 3 .2 2 . The intersection of the bodies of a finite number of sub­
complexes of a polyhedral complex K is the body of the intersection of these
subcomplexes.
It suffices to prove this theorem for two subcomplexes K x C K and
K -2 C K, i.e., to prove that
| | t f i n t f t || = l l ^ l l n \ \Kt \\.

If p £ T C K xn K 2, p £ || tfi || n || K* ||, i.e.-, || K x n K 2 || C || K x || n || K 2 ||.

Conversely, let p £ || K x || n || K> ||. Since the unique element of K con­
taining p is the carrier T{p) of p, T(p) £ K\ , T(p) £ K 2, i.e., T{p) £ K x n
Kz , so th at p £ || T(p) l| Cl |] K x n K t ||. Hence || K x || n || K-t || Cl || K\ n
^ II-
3.23. The open stars || OKe0 ||, • • • , || 0 Ker || of a triangulation K inter­
sect if, and only if, K contains the simplex (eQ■■■er).
This assertion follows from 3.22 and from the Corollary to Theorem
1.89.
In other words:
3.24. Every triangulation is the nerve of its system of open stars.
§3.3. Simplicial mappings of triangulated polyhedra. A simplicial
mapping S j of a triangulation Kb into a triangulation K a induces a con­
tinuous mapping S j of the polyhedron || K$ || into the polyhedron || K a ||
in the following way. Let Tp = (e<>- • ■eT) be any simplex of the complex
K,s . The mapping S j = S j is defined on the vertices of the simplex Tb
and this yields an affine mapping S j of the simplex Tb onto the simplex
SJTb £ K a with vertices S j c 0, S j e x , • • • , Sj e, . The required mapping
S j of the polyhedron || Kb || into || K a || is therefore defined in every sim­
plex Tb £ Kb . The continuity of the mapping S J is easily proved: let
p £ || Kb II, let Tb be the carrier of the point p, and let p = lim pn . W ith­
out loss of generality, we may suppose th at all the pn are contained in the
§4] SUBDIVISIO NS OF POLYHEDRAL COMPLEXES 139

same T'g > Tg . The barycentric coordinates of pn with respect to the skele­
ton of the simplex T'g approach the barycentric coordinates of p. There­
fore the weights assigned to the images of the vertices of T'g to obtain the
point S j p n approach the weights defining the point S j p , whence lim S aBpn
= S j p . The mapping S-J is called a simplicial mapping of the polyhedron
|| Kg || into the polyhedron || K a |) induced by the simplicial mapping S a8 of
the complex Kg into K a .
If a simplicial mapping S a8 of a complex Kg onto a complex K a is (1-1),
the mapping of the polyhedron || Kg || onto the polyhedron || K a || is also
(1 —1) and therefore topological. Whence it follows that:
T h e o r e m 3 .3 1 . If Kg and K a are isomorphic triangulations, the poly-
hedra || Kg || and \\Ka || are homeomorphic.
Since every n-dimensional triangulation is isomorphic to a triangulation
in R 'n+1, we have the following proposition:
T h e o r e m 3 .3 2 . Every n-dimensional polyhedron is homeomorphic to a
polyhedron in Ii2n+1.

§4. Subdivisions of polyhedral complexes

§4.1. Definition of a subdivision.
D e f i n i t i o n 4,11. Let K be an arbitrary polyhedral complex. (The

reader may assume that all the polyhedral complexes mentioned in this
section are triangulations.) A subdivision of the complex K is any poly­
hedral complex K a satisfying the following conditions:
1. The body of the complex K a coincides with the body of the complex K.
2. Every element of the complex K a , considered as a point set, is con­
tained in some element of the complex K.
The elements of K are mutually disjoint. Therefore, if K a is a subdivision
of K , every element of K a is contained in just one element of K . The unique
element T o f K containing a given element T a, of K a is called the carrier of
T ai in K.
If T ai 6 K a has as its carrier T , £ K and Th is an element of K distinct
from T j , then T ai n Th = 0. Indeed, if T ai and Th had a common point,
then Tj and T h would have a common point. This is impossible, since Tj
and T h are distinct elements of K.
Hence
T h e o r e m 4 .1 2 . I f K a is a subdivision of a complex K , every element T a,
of the complex K a is contained in exactly one element of the complex K —-in the
carrier of the element T al— and does not intersect any other element of K.
4.11 and 2.2 imply that the barycentric subdivision of a polyhedral
complex K defined in 2.2 is, in fact, a subdivision in the sense just intro­
duced.
Since a barycentric subdivision of an arbitrary polyhedral complex con­
sists of simplexes, it follows that:
140 COMPLEXES [CH. IV

4.13. Every 'polyhedral complex numbers triangulations among its■sub­

divisions.
D e f i n i t i o n 4.14. A subdivision of a simplex T n is the complex consist­
ing of all the elements of a subdivision of the complex | T n | which are
contained in T n.
§4.2. Successive barycentric subdivisions. Let K x be the barycentric
subdivision of a complex K, the barycentric subdivision of the complex
K \ ; in general, let K , be the barycentric subdivision of the complex 7C-i .
The complex K v is called the barycentric subdivision of order v of K.
If the diameter of an n-simplex T n is d, the simplexes of the barycentric
subdivision of the complex | T n | have diameter <( n f n + 1) d (see Ap­
pendix 1, Theorem 4.2).
Hence it follows that if all the simplexes of a triangulation K are of
diameter <d, then all the simplexes of the complex K v have diameter
< (n /n + 1)” d.
Since lim „-,«(n/n + 1 / = 0, we have the following result:
T h e o r e m 4.21. Every polyhedral complex has subdivisions of arbitrarily
small mesh, i.e., subdivisions consisting of simplexes whose diameters are less
than an arbitrary preassigned positive number.
C o r o l l a r y . Every polyhedron has triangulations of arbitrarily small
mesh.
From 4.21 it follows that
4.22. I f a polyhedron is the body of an n-dimensional triangulation K ,
dim.4> < n.
R e m a r k . See I, 8.4. In the following section (Theorem 5.34) Theorem
4.22 will be proved again; the second proof is simpler because it is based
directly on the definition of dimension (I, 8.42), and not on the compara­
tively complicated Theorem 8.44 of the same chapter.
Proof of Theorem 4.22. Let Ki be a subdivision of the triangulation K,
all of whose simplexes are of diameter <t. Then the open stars of the tri-
angulation K i form an open 2«-covering of the polyhedron <f>, whose nerve
is K i . Theorem 4.22 now follows from I, 8.44.
§4.3. Central and elementary subdivisions of complexes. We have
considered in detail the barycentric subdivision of complexes, since this
is the most important type of subdivision. However, barycentric sub­
divisions are not the very simplest.
a) Let K n be a polyhedral complex. Denote by the complex com­
posed of all the elements of K n of dimension < n — 1, and assume as given
a subdivision K an~l of K n~ \ The subdivision A / -1 of K n~l induces a sub­
division K na of K n called the central subdivision of K n relative to the given
subdivision K an~l of K n~l : let o,- be the centroid of 7in; 6 K n and consider
§4] SUBDIVISIONS OF POLYHEDRAL COMPLEXES 141

the polyhedral domains which are the projections from o, of the elements
of the complex K an~l contained in T", \ T " ,. These polyhedral domains,
the points o,-, and the elements of the complex K an~x are, by definition,
the elements of the complex K nsa .
The proof of the fact th at K n„ is, in fact, a subdivision of K n offers no
difficulty.
The barycentric subdivision of an arbitrary polyhedral complex K n is
reduced to a series of central subdivisions in the following way. The bary­
centric subdivision K °i of the 0-complex K°, consisting of the vertices of
K n, is the complex K° itself. Let us suppose that the barycentric subdi­
vision K ri of the complex K T, consisting of all the simplexes of K n of dimen­
sion < r, has already been constructed. Then we obtain the barycentric
subdivision K { +1 of the complex K T+l as the central subdivision of K T+1
with respect to the barycentric subdivision K T\ of K T.
b) Elementary subdivision. The central subdivision of the complex | T n \
relative to | T n | \ T n is called simply the central subdivision of | T n |;
the simplexes of this subdivision contained in T n form a central subdi­
vision of the simplex T n. The central subdivision of T n or | T n |, respectively,
is also called the elementary subdivision relative to T n. The elementary
subdivision of T n relative to T v < T n consists, by definition, of all the
simplexes (T"’iT’7'“:p~1), where T Ti is any element of the central subdivision
of T v and T n~p~1 is the face of the simplex T n opposite the simplex T v\
here, (Tr,T n~p~1), as always, denotes the simplex whose skeleton is the
union of the skeletons of the simplexes T 7\ and T n p i.e., the combinatorial
sum of T \ and T n~ ^ \

Fig. 91
142 COMPLEXES [CH. IV

The elementary subdivision of a triangulation K relative to a simplex

T p £ K is, by definition, the subdivision obtained by replacing each sim­
plex T 6 0 kT v by its elementary subdivision relative to T v and leaving all
the remaining simplexes (i.e., all the simplexes of the complex K \ OkT p)
unchanged.
Figs. 90 and 91 show various cases of the elementary subdivisions of
2- and 3-simplexes.
The definition of an elementary subdivision immediately implies the
following remarks which are required in Chapters VII and X.
R e m a r k 1. Let = (ep+i- ■•en) be the face of the simplex

7(Co*’ *(Zp@p-j-i ***&n)

opposite the face T p = (e0- • -ep). Let e be the center of the simplex T p.
Then the simplexes of the elementary subdivision of T n relative to T p are
all the simplexes of the form
(eeio • *' 6iT@p-yl' ' *£«), ?o < * ^ L Pi t —P
and only these.
R em a r k 2. Let V n be the elementary subdivision of a simplex T n relative
to one of its faces T p. In order that a subset E of the set of all vertices
of the complex | V n | be a skeleton of | V n | it is necessary and sufficient
that E not contain the skeleton of the simplex Tp; among the skeletons of
| V n | the skeletons of V n are characterized by the fact that they contain
the vertex e and all the vertices of the face T n~p~l < T n opposite the
face T p.
R e m a r k 3. Let us preserve the notation of the preceding remark. If
r < n — 1, an r-simplex T r of the complex V" cannot have among its
vertices more than p — 1 vertices of the simplex T p.
Indeed, in the contrary case, the simplex T r, having among its vertices
the n — p vertices ep+i , • ■• , en of the simplex and also the vertex
e, would have in all at least p + n — P + 1 = n + l vertices; this is im­
possible, since r < n — 1.
Let us show that the barycentric subdivision of a triangulation K is
reducible to a series of elementary subdivisions. To this end, let us first
perform an elementary subdivision on all the original simplexes of the
complex K relative to themselves. Then simplexes of the form T n =
(eo ■• • en) give rise to simplexes (for the notation see the end of Remark
1 of 0.2) T nii = (oeo • • ■e.j ■■• en). Next, let us perform an elementary
subdivision on each of these simplexes relative to the faces (e0 • • • • ■■en).
The resulting simplexes have the form
T (^^i+O ' ' ' &ii ' ' ’ C|2 ' ** Cn).
§4] BARYCENTRIC STARS 143

After the elementary subdivision of these simplexes relative to the faces

(eo • • • e,-, • • • e,2 • • • en), the simplexes T niliti3 are obtained, whose first
three vertices are o, ot, , o,,,, , and whose remaining vertices are vertices of
T". [If Tr is a face of the simplex T n = (e0 • • • en) and etl , • • • , eik are the
vertices of T n which are not the vertices of Tr, then o , i s the centroid of
the simplex T r.]
Continuing the indicated process [the simplex T niv ..ik is defined as
T niv --ik = (ooil • • • oil...ikeik+l • • • ein), we finally obtain simplexes of the
form (oo,-, • • • o , which, together with their faces, form a barycentric
subdivision of the simplex T n.
§4.4. Subdivisions of nonclosed subcomplexes of polyhedral com­
plexes. Let Ar be a subcomplex of a polyhedral complex K ' . Then | K \
is a polyhedral complex. Every complex consisting of all those simplexes
of any subdivision K a of | K | which are contained in elements of K (i.e.,
have these elements as their carriers) is called a subdivision of K.
In particular, if K„ is the barycentric (elementary) subdivision of | K |,
then the subcomplex of K a consisting of all the simplexes of K a having
simplexes of AC as their carriers is called the barycentric (elementary) sub­
division of K.
R e m a r k 1. It follows from this definition that one can speak of the
elementary subdivision of a complex K relative to a simplex not belonging
to this complex (i.e., relative to a simplex T 6 | K | \ K).
R e m a r k 2. If if is a nonclosed subcomplex of a simplicial complex, the
barycentric subdivision of K will not be an unrestricted complex and
therefore cannot be isomorphic to the barycentric derived of K.
§5. Barycentric stars
§6.1. Barycentric stars. Let K be an n-dimensional triangulation, and
let Ki be its barycentric subdivision. Let
T i , T z , ■• • , T.

be all the simplexes of K, and let

T i = Ci,*** , Cu , u po,
be the O-simplexes of K , i.e., the vertices of K . The vertices of K x are the
centers
en , ei2 > • • • , 01*
of the simplexes
T i , To, ■• ■ , T, (en = e{ for i < po).
144 COMPLEXES [CH. IV

We shall write'eu > eu if T x > T jy so that the simplexes of 7£i have the
form
(® lt'o ^ ®1 «‘i ^ ^ ^ l * 'r ) -

D 5.1. Let 71,- G I£. Let us call the subcomplex of Ki con-

e fin it io n

sting of all the simplexes of K\ whose last vertex is the center eu of 71,-
the barycentric star dual to T; (or simply
the dual of 7\). We shall denote the star
dual to T x by T *{, and will refer to the
simplex T t itself as the dual of (the
barycentric star) 7'*,-. The outer bound­
ary of the barycentric star T*x is, by defi­
nition, the complex 71*,- = | T*x | \ T*x ,
i.e., the set of all faces of the simplexes
of T*i not in T*x . The barycentric stars
dual to the vertices ex of K are known as
the major stars. These are the barycentric
Fig. 92
stars T*i , • • • , T *r , r = p0 (Fig. 92).
5.11. If Ti < T j , the dimension of the complex 71*,- is greater than the
dimension of the complex T*}-.
In fact, let the dimension of T*j be r, and let
(ciy, > e,y2 > • > eiir > eu)
be an r-simplex of T*, . Then the simplex
(ciji > cij2 > • • • > e\jr > e{j > eu)
is an (r + l)-simplex of T*{ (Fig. 92).
§5.2. The dual complex of a triangulation.
D e f i n i t i o n 5.2. The set K* whose elements are the complexes T *, is

called the dual (or barycentric) complex of the triangulation K ; the set K*
is in this connection partially ordered by putting
T *. K T *. if T . > T .

This order and the dimensions of the complexes T*t (as subcomplexes,
of K) turn K* into an abstract complex (see 1.7). Hence the term “dual
complex” is fully justified and we may refer to open and closed subcom­
plexes of K*.
We shall derive a series of simple but important properties of K*.
5.21. f i U ' - u r , = Ki .
Indeed, if Tlh G Ki and eu is the last vertex of T Vt, then T xh 6 T*x .
5.21 implies
5.210. || T*, || u • • • u | | T*. || = || /vi || = || K ||.
5.22. If i J, then T*, n T*, = 0.
§ 5] BARYOKNTRIC STARS 145

Indeed, if l \ h £ T*, n , the simplex T ih £ Ki would have two last

vertices eu and eu , which is impossible.
5.220. If i ^ j, then || T*t || n || T*} || = 0.
This follows from 5.22 and Theorem 3.22.
5.23. Every non-major barycentric star is less than at least one major
star.
Indeed, if e* < eu , then T*k > T *c.
5.24. The outer boundary of a barycentric star T*{ is the union of the
barycentric stars less than 71*, .
Proof. It follows from the definition of the complex 71*,- CZ Ki that:
5.240. The complex 71*,- consists of all the simplexes of the form 7\h =
Oiy0 > ei;i > • eljr), where eljr > eu .
Therefore, if
Tih = (ei, 0 > • ■• > C\ jT) £ T *i ,
then eijT> eu and T lh £ T*jr < T*i .
Conversely, if Tih £ T*;- < T * i, then
= (ciio > • • • e,\ jT > Ci;), eu > eu ,
and 7\h £ f *,-.
5.25. The combinatorial closure | T*\ | of the barycentric star T*t in
the complex K i , i.e., the subcomplex of Ki consisting of all the faces of
the simplexes of 7’*, , can be written in the form
(5.25) | T*t \ = T*i u T * i.
5.240 implies
5.250. The complex | T*i \ consists of all the simplexes T lh £ Ki of
the form Tu = (ci;o > eu1 > ■■■ > ei/r), where eur > eu .
Whence it follows that:
5.26. If T*i < T * i, then | T*t \ C | T*j |.
From 5.21, 5.23, 5.24, and 5.25 we infer
5.27. The union of the combinatorial closures of the major barycentric
stars of the complex K is the complex K\ .
T h e o r e m 5.28. A necessary and sufficient condition that

| T*io | n • • • n | T*ir \ ^ 0
is that the simplexes 7 \0 , • • • , T ir be faces of a simplex of K ; in that case
| T \ | n ■• • n | T \ \ = | T \ |,
where 7\ is the combinatorial sum of the simplexes 7\0 , • • • , 7 \r .
Proof, a) Let
(« „■ . > • ■ • > e u „ ) e | r * „ | n . . . n | r * , , | ;
146 COMPLEXES [CH. IV

then, by 5.250
Ci Jp €uk (0 < <T-),
k

i.e., all the simplexes 7\0 , • • • , 7\r are faces of the simplex T ,-p ; hence the
combinatorial sum 7\ of the simplexes T io , ■■■ , 7 \r exists and is a face
of the simplex Tjp:
T ip > T i .
Therefore,
T*h < T * i, (eli0 > > ci,,) 6 T*jp C | 71*,
Hence
71*,,, | n • • • n | 71*,-, | C | 71*,- |.
b) To prove the converse inclusion we note that, according to the defi­
nition of combinatorial sum,
T i > T ik > k = 0, 1, r.
so that
< T*ik , | 7,*i | C | T*ik |,
| T*t | C | T*i0 | n • • • n | T \ |.
This proves 5.28.
In particular, the combinatorial closures of the major stars
T*i o , • • • , T\ , i o, ■■■ , i r < po,

intersect if, and only if, the simplex

(Ci'o ' " " C»r)
is in K.
In other words:
T h e o r e m 5.29. The nerve of the system of combinatorial closures of the
major barycentric stars of a complex K is the complex K itself (Fig. 92).
§6.3. Closed barycentric stars.
D e f i n i t i o n 5.31. The body of the combinatorial closure | T *,• |, i < p0 ,

of a major barycentric star T*i dual to a vertex e,- of a complex K is called

the closed barycentric star dual to the vertex e,-; the vertex e,- is called the
center of the closed barycentric star.
Hence, closed barycentric stars are polyhedra.
5.27 implies that:
§5] BARYCENTRIC STARS 147

I t follows from 5.29 and 3.21 that:

5.33. The closed barycentric stars of a complex K form a closed covering
of the polyhedron ]| K ||, called the barycentric covering of |] K |[ dual to its
triangulation K . The nerve of this covering is the complex K.
Since the simplexes and hence the closed barycentric stars of a complex
K can be assumed to be of arbitrarily small diameter, it follows from
5.33 that:
T h e o r e m 5.34. The body of every n-dimensional triangidation has a
closed (-covering of order n + 1 for arbitrary e.
§6.4. The subcomplexes of the complex K*; their bodies and bary­
centric subdivisions. Let K be any triangulation, K\ its barycentric
subdivision, and K* its dual complex. Let K 0 be any subcomplex of K.
Denote by K *0 the subcomplex of K* consisting of all the barycentric
stars of K dual to the elements of K 0 .
Conversely, to every subcomplex K*0 of K* there corresponds the sub-
complex K 0 Q K consisting of all the elements of K dual to the elements
of K*o . The complexes K 0 and K*0 are said to be dual subcomplexes of K
and K*, respectively.
It" follows th at Ko and K *o are also dual to each other as partially ordered
sets (see I, 6.4). Hence if K 0 is a closed (open) subcomplex of K , K*0 is
an open (closed) subcomplex of K*. The elements of K*0 are barycentric
stars, i.e., certain subcomplexes of Ki ; the union of all the barycentric
stars which appear as elements of K *0 is a subcomplex K*0i of Ki , called
the barycentric subdivision of K *0 .
The body of K*m is, by definition, the body || K \ || of K \ (obviously,
|| K *o || is the union of the bodies of the barycentric stars which are the
elements of K \ ) .
5.41. If K *o is a closed subcomplex of K*, K *oi is a closed subcomplex
of Ki .
Indeed, if T* G K *0, then, since K*0 is closed in K*, the combinatorial
closure of T* £ K \ in K* is contained in K*0 and consequently, by 5.25
and 5.24,
| T* | C K *oi .
This proves 5.41.
C o r o l l a r y . The body of every closed subcomplex K*0 of K* is a poly­
hedron.
R e m a r k . If K 0 is a closed subcomplex of a complex K and K \ is the
subcomplex of K* dual to K 0 , then K* \ K*0 is a closed subcomplex of
K* and hence || K* || \ || K \ || is a polyhedron.
In conclusion, we shall prove the following proposition which is re­
quired in Chapter X:
148 COMPLEXES [CH. IV

5.42. The closure of a simplex T of a triangulation K is contained in the

union of the closed harycentric stars dual to the vertices of T and does not
intersect any closed harycentric star whose center is not a vertex of T.
Proof. In the course of this proof p will denote a point of the closed
simplex T and
Tu = (ei ,-0 > • • • > euT)
the carrier of the point p in the complex K\ . It follows th a t-T ,0 is the
carrier of p in K ; therefore 7\-0 is a face of T. A fortiori,
T ir < T,
so that all the vertices of Tir are also vertices of T.
Let eh = eih be any vertex of T ir; the simplex
T (Cii'o ^ ^ @lir)
is a face of the simplex
(cli0 ^ ^ e\ir ^ Clh)
which is in the harycentric star dual to eh and p is contained in the closed
harycentric star with center eh . But eh , as a vertex of T,-r , is a vertex of T.
This proves the first half of Theorem 5.42.
To prove the second half of the theorem we note th at if p is contained
in the closed harycentric star dual to a vertex ey £ K , then T u is a face of
some simplex of K\ having ey as a vertex; since

Tu — (c«o > • • • > £iir)

and ey is a vertex of K, ey is a vertex of T ir . Hence ey'is a vertex of T, q.e.d.
R e m a r k . Since closed simplexes and closed harycentric stars are com-
pacta, a closed simplex T has a nonempty intersection only with the closed
harycentric stars dual to its vertices and has a positive distance from all
other closed harycentric stars. Since, on the other hand, the closed bary-
centric stars cover the polyhedron || K jj, some neighborhood of the closed
simplex T is contained in the union of the closed harycentric stars dual
to the vertices of T and does not intersect any of the other closed bary-
centric stars. Since K is a finite complex, we have
5.43. If K is a triangulation of a polyhedron || K || there is an
e = e(K) > 0
such th at an e-neighborhood of an arbitrary closed simplex T of K is
contained in the union of the closed harycentric stars dual to the vertices
of T and does not intersect any of the other closed harycentric stars.
§ 6] TOPOLOGICAL COMPLEXES AND TOPOLOGICAL POLYIIEDRA 149

§6. Topological complexes and topological polyhedra

§6.1. Definitions. The definition of a polyhedron (Def. 3.12) is not
topologically invariant; the topological image of a polyhedron is in general
not a polyhedron.
I t is natural to consider the topological images of complexes and poly­
hedra. We introduce the following definitions:
D e f i n i t i o n 6.11. A compactum homeomorphic to a polyhedron is called
a topological polyhedron.
A topological mapping of a polyhedron P onto a topological polyhedron
(P maps the simplexes T, of a given triangulation K of P onto subsets of
(P. The collection of such subsets forms a so called topological complex.
Thus we arrive at the following definition:
6.12. A finite system 3C of subsets 3i , i = 1, 2, • • • , s, of a topological
space R is called a topological complex if the union

|| X || = 3X u • • • u 3 s

can be mapped topologically onto a polyhedron P = \\ K || in such a way

th at under this mapping the sets 3, correspond (1-1) to the simplexes
of some triangulation K of P (i.e., the image of each of the sets 3t- is a
simplex and, conversely, the inverse image of each of the simplexes
Ti is a set 3,).
The union of the elements (“topological simplexes”) 3; of a topological
complex 3C is a topological polyhedron || X ||, the body of 3C; 3C is called a
topological triangulation of || 3C ||.
Examples of topological complexes are shown in Figs. 93 and 94.
R e m a r k 1. Various, but necessarily isomorphic, triangulations K cor­
respond to a topological complex 3C. The dimension number of JC is the
dimension number of a triangulation K isomorphic to 3C. We shall prove
in Chapter V that the dimension (I, 8.4) of a polyhedron is equal to the

F i g . 93 F ig. 94
150 COMPLEXES [CH. IV

dimension number of any of its triangulations. Hence the dimension num­

ber of K (equal to the dimension number of ,TC) is the same as the dimen­
sion of the two homeomorphic eompacta || K || and || X ||. Consequently,
assuming the fundamental theorem of Chapter V, we may state the fol­
lowing proposition:
T h e o r e m G .l. All the topological triangulations of a given topological
polyhedron have the same dimension number, equal to the dimension of the
polyhedron.
R em a r k 2. Every topological triangulation of a topological polyhedron
(P is isomorphic to some (in general, topological) triangulation of a given
fixed polyhedron P homeomorphic to (P. But, on the other hand, every
topological triangulation of (P is isomorphic to a triangulation of some
(arbitrarily chosen) polyhedron P homeomorphic to (?. Hence the set of
all combinatorial types of topological triangulations of a topological
polyhedron may be defined in two ways:
a) as the collection of topological triangulations of a single arbitrarily
chosen polyhedron P homeomorphic to (P;
b) as the collection of triangulations of different polyhedra homeomorphic
to (P.
R em a r k 3. Purely combinatorial conditions (i.e.; conditions independent
of the notion of continuity) which two triangulations must satisfy in order
to have homeomorphic bodies are not known as yet. In particular, the so
called fundamental hypothesis (“Hauptverm utung”) of combinatorial
topology has not yet been proved (see Moise [a], whose papers appeared
several years after this book (Trans.)). This hypothesis asserts th at any
two triangulations of homeomorphic polyhedra have isomorphic subdivisions.
§6.2. n-manifolds. A closed polyhedral manifold is- a polyhedron which
is at the same time a topological n-manifold (I, 5.3); topological polyhedra
which are topological manifolds are usually called simply closed manifolds.
It is not yet known whether the class of closed manifolds defined in this
way coincides with the class of all topological closed manifolds: it is not
known whether there exists a closed topological manifold which is not a
topological polyhedron.
A closed 1-manifold is homeomorphic to a circumference. The closed
2-manifolds are none other than the closed surfaces discussed in Chap­
ter III.
The simplest example of a closed n-manifold is the n-sphere S n. It can
be triangulated by taking an (n + l)-simplex inside the sphere S n C E n+1
and projecting its boundary onto the sphere from some point in its interior.
The three-dimensional torus (i.e., the topological product of three
circumferences, I, 2.G) is also a manifold: to triangulate the three-dimen­
sional torus it suffices to take a second order barycentric subdivision of a
§7] CONNECTEDNESS OF COMPLEXES 151

three-dimensional cube and to identify opposite faces of the cube (I, 5.2,
Example 6).
In the same way it is easy to construct a triangulation of the topological
product of a sphere and a circumference: it is merely necessary to use the
model of this product given in I, 5.2, Example 7.
To construct a triangulation of the projective n-space let us first define
a regular (n + l)-dimensional octahedron in as a convex polyhedral
domain with 2(n + 1) vertices ek , c'k , k = 1, 2, • • • , n + 1, where ck ,
e \ , respectively, have all coordinates equal to zero except the kth which
is equal to 1, —1, respectively. "The boundary of an (n + 1)-dimensional
octahedron, as the boundary of an (n + 1)-dimensional convex polyhedral
domain, is homeomorphic to the n-sphere S n [in the given special case
this is easy to prove directly, e.g., by complete induction on the number
of dimensions: a regular (n + 1)-dimensional octahedron is the union of
two pyramids with vertices (0, 0, • • • ,1 ) and (0, 0, • • • , —1) constructed
on an n-dimensional regular octahedron] and is given in a triangulation
K n possessing the property of central symmetry relative to the origin of
coordinates. This means that every element of K n is mapped by the
symmetric transformation

x \ = —xk , k = 1, 2, • • • , n + 1,

of R n+1 into some other element of K n.

A second order baryccntric subdivision [one could get along more
simply, i.e., with a first order barycentrie subdivision; a second order
subdivision is convenient for some special purposes (see the triangulation
K ni mentioned below)] K n(2) of K n obviously also possesses the property
of central symmetry. Identifying symmetric elements of K n^_, , we obtain
the required triangulation K n0 of the projective space P n.
In addition to K n0 we shall need (in VIII, 4.4) still another triangulation
of the projective space P n, which will be denoted by K nt . It is obtained
by taking a second order barycentrie subdivision of an u-dimcnsional
octahedron and its boundary and identifying elements of the triangulation
of the boundary (i.e., of K '1^1(2>) symmetric relative to the origin of co­
ordinates.
The simplest triangulation of the projective plane P1 is obtained by
identifying symmetric elements (faces, edges, and vertices) of a regular
icosahedron (a polyhedral domain with twenty faces).
§7. Connectedness of complexes
[All the results of this section (except those of 7.3) are true for abstract
complexes, even for all partially ordered sets (discrete spaces), and are
special cases of the corresponding theorems for 7Vspaccs (I, 3).]
152 COMPLEXES [CH. IV

§7.1. Connected complexes. Components. A complex K is said to be

connected if it is not the union of two nonempty disjoint closed subcom­
plexes.
=
R e m a r k . If K K ' u K", K ' n K" —
0 and K ', K " are closed, then
K ' = K \ K " , K" = K \ K ', and consequently K ' and K " are open.
Therefore, the definition of connectedness can also be stated in either
of the following ways:
K is a connected complex if it is not the union of two nonempty dis­
joint open subcomplexes.
K is a connected complex if no proper subcomplex of K is both closed
and open in K.
7.11. If a connected subcomplex 7v0 of a complex K is contained in the
union of two closed (open) disjoint subcomplexes K and K 2 of K , it is
contained either in Ki or in K 2 ■
Inded, if
K o n K r ^ O * K 0n K 2,
then K 0 n K\ and K 0 n K 2 are nonempty closed (open) subcomplexes of
Ko and, since
7 l0 = (A"o n u (K0 n K 2),
K 0 is not connected.
7.12. If Ko is a connected subcomplex of a complex K and T 6 K is
either a face of To or has T0 as a face, wrhere T0 6 Ko , then Ko u T is a
connected complex.
In fact, if
Ko u T = K x u K 2,
w-here K i and K 2 are closed (open) in /v0 u T, then, by 7.11, we may as­
sume that Ko a Ki . But then, since Ki is closed (open) in Ko u T, we
have T £ K^ so that K 2 is empty.
7.13. The union Q of an arbitrary number of connected subcomple.xes
K a of a complex K containing a given fixed element T £ K is connected.
Indeed, let
Q = Qi u Q2
be a decomposition of Q into twro subcomplexes closed in Q; let T € Qi ;
then, by 7.11, every K a is contained in Qt , i.e., Q = Qi and Q2 = 0.
7.13 implies
7.14. The union of all the connected subcomplexes of a complex K
containing a given T 6 K is a connected subcomplex Qk(T), the component
of T in K.
§7] CONNECTEDNESS OF COMPLEXES 153

A component Qk(T ) obviously has the following maximal property:

there does not exist any connected subcomplex Q' c K different from
Qk(T) and containing QK(T).
7.13 further implies th at there cannot exist two nonidentical disjoint
connected subcomplexes Qi and Q2 of K containing T and satisfying the
condition of m axim ally (since Qi u Q2 would be connected and Qi u Q2
=3 Q i, Qi u Q2 => Q.j).
Therefore,
7.15. The component Qk{T) may be defined as the unique maximal
connected subcomplex of K containing T.
Furthermore:
7.16. Two components Q k ( T ) and Q k ( T ' ) having a nonempty inter­
section are identical.
Finally, 7.12 implies
7.17. Every component Qk(T) is both closed and open in K.
Hence
7.1. Every complex is uniquely partitioned into disjoint maximal con­
nected subcomplexes, which are both open and closed. These subcomplexes
are the components of K . A connected complex consists of one component.
We note in conclusion th at 7.12 implies
7.18. If T < r , then QK(T ) = Qk(T').
§7.2. The case of unrestricted simplicial complexes. Let K be an un­
restricted simplicial complex. 7.18 implies th at for an arbitrary T £ K
and arbitrary vertex e < T the components QK{T) and QK(e) are identical.
Hence we need consider only the components of the vertices of K. Further­
more, the definition of connectedness in'the present case may be phrased as:
7.20. An unrestricted simplicial complex K is connected if every two
nonempty closed subcomplexes K i and K 2 whose union is K have at least
one common vertex.
Finally, let us call every finite sequence of edges (1-simplexes) of a
complex K of the form
(e\e2), (62C3), ■■• 1 (es-ic4)
a broken line joining the vertices e-i and es in the complex K.
7.21. The component of a vertex e of an unrestricted simplicial complex K
consists of all the simplexes of K whose vertices can be joined by a broken
line to the vertex e.
7.21 follows easily from:
7.22. An unrestricted simplicial complex K is connected if, and only if,
any two of its vertices can be connected by a broken line in K.
Proof. If K can be represented in the form of a union of two closed
subcomplexes K ' and K " having no common vertices, then no vertex of
154 COMPLEXES [CH. IV

K' can bo joined by a broken line to any vertex of K" [since the broken
line would necessarily have a link (e,e,+i)> where e,- € A ', C/+i € A'", and
consequently this link itself could not belong either to K ' or K"\.
This proves the first part of Theorem 7.22.
To prove the second part of the theorem, let us assume that e' and e"
are two vertices of K which cannot be joined by a broken line; let us
consider the complex K t>consisting of all the elements of the complex K
whose vertices can be joined to d by a broken line. The complement of
K c>, the subcomplex K" = K \ K c' , is nonempty, since it contains the
vertex e " . Both complexes K e>and K" are closed and have no common
vertices. At the same time K = K e>u K " . This proves the theorem.
§7.3. The components of || K || and K. Let K be a triangulation of a
poljdiedron || K ||.
7.31. If K is a connected complex, || K || is a connected polyhedron.
Indeed, let p and p' be two points of || K ||, T and T' their carriers in
K, e and e' any vertices of T and T \ respectively. Let
(eex), (ee2), • • • , (e»_ie,), (e,ef)
be the links of a broken line joining e and e' in K ; and let [pe] and [e'p'}
be straight line segments in T and 7”, respectively.
Then
[peeie2 • • • es_iese'p'}
is a broken line (in the elementary geometric sense) joining p and p' in
I! K ||. This proves that || K || is connected.
Again, it follows from 7.1. 3.1, and 3.22 that:
If Qi > • • • , Qs are components of a complex K, then the sets || Q, || are
mutually disjoint polyhedra which are both closed and open in || K ||.
It follows immediately that every component of a polyhedron || K || is
contained in some polyhedron || Q{ || and, since || Q, || is connected, is
identical with this || Q, |j.
Hence
7.3. The components of a polyhedron || K || coincide with the bodies of
the components of any triangulation of || K ||.
C o r o l l a r y 7.30. All the triangulations of a polyhedron consist of the
same number of components, equal to the number of components of the
polyhedron.
 Chapter V
S P E R N E R ’S L E M M A A N D IT S C O R O L L A R IE S

All the results of this chapter are proved anew in Chapters X and XIV
and again in Chapters X and XV, but in a less elementary fashion than
here. This chapter can be read without reading the preceding chapters.
I t is merely necessary to consult the indicated references to Chapter I and
to read the preliminary introductory section of Chapter IV on simplexes.
In what follows, to avoid ambiguity in the statement of theorems, we
shall use the term “dimension number” to designate the combinatorial
dimension of a simplex or complex, that is, the number of vertices of a
simplex less 1 in the former case and the maximum of the dimension
numbers of the simplexes of a complex in the latter case. In the same way,
a polyhedron will be said to have dimension number n if it has a triangula­
tion whose dimension number is n. The term “dimension” will be re­
served for the topological concept defined in I, 8.42.
Several fundamental topological theorems, first established by Brouwer,
are proved in this chapter. First among these is:
T h e o r e m I. The dimension (I, Def. 8.42) of a closed simplex is equal to
its dimension number (i.e., to the number of its vertices less 1).
In other words (I, 8):
T h e o r e m I' ( t h e P f l a s t e r s a t z f o r a S i m p l e x ) . A closed n-simplex
has closed e-coverings of order n + 1 for every e > 0; if e is sufficiently small,
every closed e-covering of a closed n-simplex is of order > n + 1.
The first assertion of this theorem is contained in IV, 5.34 and 4.22.
Essentially, this assertion is perfectly elementary and the reader may,
without turning to Chapter IV, prove it as an exercise (in this connection
see the hints given in §1).
The second assertion of Theorem V expresses a deep geometric fact; it
also makes up the basic content of this chapter.
An immediate corollary of the second assertion of Theorem I' is (the
reader omitting Chapter IV should at once go on to Theorem III):
T h e o r e m IF. Let be a polyhedron of dimension number n. For every
sufficiently small e > 0 every closed e-covering of the polyhedron 4> has order
> n + 1.
This and IV, 5.34 imply
T h e o r e m II. Every polyhedron <t>of dimension number n has dimension n.
Furthermore, from Theorem I of this chapter we deduce
T h e o r e m III ( i n v a r i a n c e o f t h e d i m e n s i o n n u m b e r o f R n). Two E u­
clidean spaces R n and R m of different dimension numbers n and m are not
homeomorphic.
155
156 SPE R N E R ’S LEMMA AND ITS COROLLARIES [CH. V

The following proposition is an important and stronger form of Theorem

III. It is proved in §3.
T h eo r em I V (In v a r ia n c e of I n t e r io r P o in t s of S u b s e t s of R n
u n d e r T o p o l o g i c a l M a p p i n g s i n t o R n). A topological mapping C of

a set A G R n onto a set B C R n maps every interior point of A (relative to

R n) into an interior point (relative to R n) of B (and consequently maps a
noninterior point of A into a noninterior point of B).
R e m a r k . Theorem IV can be proved not only for the space R n, but also
for all topological manifolds il/n (see I, 5.3).
The proofs of the above theorems given in this chapter differ from the
original proofs of Brouwer; they are based, essentially, on a single ele­
mentary proposition of combinatorial character known as Sperner’s
lemma (formulated and proved in §2). Sperner’s lemma not only leads to
surprisingly simple proofs of Theorems I-IV , but is in itself a remarkable
geometric fact.
Sperner’s lemma also enables us to give a very simple proof of yet
another classical theorem of Brouwer, namely, the fixed point theorem for
an arbitrary continuous mapping of a closed simplex into itself (or, of course,
of an arbitrary compactum homeomorphic to a closed simplex). This proof
of Brouwer’s theorem is due to Knaster, Ivuratowski, and Mazurkiewicz
and has become classical. It is given in §4.
§1. Preliminary remarks
This section is intended only for the reader omitting Chapter IV.

§1.1. Triangulations and barycentric subdivisions of a closed simplex.

Let us recall the propositions of Chapter IV required-in this chapter.
a) We require only triangulations of a closed simplex T n. This term
(triangulation of a closed simplex) denotes a finite set K n of mutually
disjoint simplexes (of various dimension numbers) satisfying the following
conditions:
1. Every face of a simplex which is an element of K n is itself an element
of the set K n.
2. Every element T of I T is contained (as a point set) in one, and
obviously in only one, of the faces (proper or not) of the simplex T n,
called the carrier of T r.
3. The point set union of all the elements of K n is the closed simplex T n.
This definition implies that:
Every face T T of the simplex T n is the union of the simplexes of the
subdivision K n of which T T is the carrier.
Indeed, if p £ T Tand Thi (- K 71contains py then the carrier of the simplex
T i is necessarily T T, since in the contrary case two faces of the simplex
§1] PRELIMINARY REMARKS 157

T n (namely, T T and the carrier of the simplex 7*,-) would have a common
point p. Hence every point p (E T r is contained in a simplex of K n carried
by T T, which proves the assertion.
Furthermore,
If T j < Ti £ K n and T T is the carrier of T j , then the carrier of T3- is a
face of T T.
Indeed, T Tis a closed set which is the union of all the faces of T r. There­
fore every contact point of T r is contained in some face of T T. Since Tj
consists of contact points of Tj , and therefore also of TT, the carrier of
Tj is a face of T r, which was to be proved.
b) Among all the triangulations of a closed simplex T n the most important
is its barycentric subdivision-, the elements of the barycentric subdivision
of a closed simplex T n are all the faces of the (n + 1)! simplexes of the
form
(pOilOjlj2 • • • Ojlt-2...*•„),
where o is the centroid of the simplex T n, otl is the centroid of any of its
(n — l)-faces , o,li2 the centroid of any (n — 2)-face 7,n-\ lt-2 of
T n~l,, , etc., up to the vertex o of T n.
The barycentric subdivision of a closed segment is obtained by dividing
it in two. The barycentric subdivision of a triangle is shown in Fig. 95.
One of the simplexes of the barycentric subdivision of a tetrahedron is
shown in IV, Fig. 87.
I t follows from 4.2 of Appendix 1 that:
If the diameter of a closed simplex T n is equal to d, then the diameter of
an arbitrary simplex of the barycentric subdivision of T n does not exceed
[n/(n -f 1)] d.
Hence it follows that for sufficiently large h the barycentric subdivision
of order h of T n will yield a triangulation of this simplex of arbitrarily
small mesh, i.e., a triangulation all of whose simplexes have diameters
less than an arbitrarily given e > 0. (The mesh of a triangulation is the
maximum of the diameters of its simplexes.)
c) We shall require the following fact: A closed simplex T n can be repre­
sented as the union of n + 1 closed sets A Q, A i , ••• , A„ , corresponding
(1-1) to the vertices eo, e\ , ■• • , en of the simplex T n in such a way that
A , contains the vertex eT and has no point in common with the closed face of
T n opposite the vertex eT.
This fact is implicit in the results of Chapter IV: it suffices to define
A r as the closed barycentric star with center eT, i.e., as the union of all
the simplexes of the barycentric subdivision of T n which have eT as a ,
vertex.
The required sets A r can be constructed somewhat differently: let
158 SPERNICH’S LEMMA AND ITS COROLLARIES [CH. V

F i g . 96s

R rn~l be the plane through the centroid of T n parallel -to the plane of the
face (eo • • • er-ier+i • • • en). This plane cuts off from T n a closed simplex
containing the vertex er which we also denote by A r (Fig. 96).
d) For arbitrary e > 0 the whole space R n can be covered by a countable
number of closed sets A r of diameter <e in such a way that no point of the
space is contained in more than n + 1 of these sets and such that any sphere
xi
2 + i x-i 2 +, ■ ■ ■ +, X n 2 <- 2a

intersects only a finite number of the sets A r .

To prove this, let us first construct a simplicial e-decomposition (“in­
finite triangulation”) K of the space R n, i.e., a set K of mutually disjoint
simplexes (of various dimensions) possessing the following properties:
1. Every face of an arbitrary simplex of the set K is itself an element
of the set K.
2. The union of all the simplexes of K is R n.
3. Every solid sphere of the space R n intersects at inost a finite number
of simplexes of the set K.
4. The mesh of K is < e.
Such a set of simplexes (infinite complex) K can be constructed in the
following way: let us divide R n into the cubes
nii e < Xi < (mi + 1) e' (i = 1, • • • , n),
t' = e/2n% m .j = 0, ± 1 , ± 2 , • • • ,
and their faces. Let us divide each of these cubes barycentrically. This
means: divide all their edges in half, then divide each square into eight
equal triangles, projecting from the center of a square its boundary which
has already been subdivided. Similarly, subdivide all the three-dimensional
cubes, projecting from the center of each cube its already subdivided
boundary, etc.
The resulting simplexes of various dimensions make up the required
complex K.
§1] PRELIMINARY REMARKS 159

Every point p G R n is contained in one and only one -simplex of K, the

carrier of p.
Having constructed a simplicial e-decomposition K of R nr let us repre­
sent each of the closed n-simplexes of this decomposition in the form of a
union of n + 1 closed sets A; satisfying the conditions of c). If em is any
vertex of K , denote by <f>m the union of all the sets A , just constiucted
which contain en . Since the simplexes of the decomposition have diameters
<e, the se ts$ mhave diameters < 2 t. It is clear that an arbitrary bounded
domain of R n intersects only a Unite number of the sets <f>m . We shall
prove, finally, that no point p £ l i n is contained in more than n + 1
sets<hm . Indeed, from the construction of the sets<hm it follows that:
If the carrier of a point p G R n is the simplex T r = (efco ■• • ehr) G K,
then p can be contained only in the sets

It follows from what has just been proved (independently of the results
of Chapter IV) that:
Every compaction in Rn (in particular, every closed simplex of R n) has a
closed t-covering of order < 7 1 + 1 for every e > 0.
This proves the first assertion of Theorem I'.
e) In §3 we shall require the theorem asserting that the boundary of an
n -simplex T n has closed t-coverings of order < n for every e > 0. This
proposition is also a special case of I, 5.34. However, it is also easy to
prove directly.
To this end, it suffices to represent every closed (n — l)-face of the
simplex T n as the union of sets A , satisfying the conditions of c) and to
define <hm for every vertex em of T n as the union of the sets A , containing
the vertex em .
f) Let us note finally that the barycentric subdivisions of all the closed
(n — l)-faces of a given ?i-simplex T n yield, by definition, the barycentric
subdivision of the boundary T H\ T n of T n. Fig. 97 shows the case n = 3
(only the faces turned to the observer are shown).
160 s p e r n e r ’s lem m a and it s c o r o l l a r ie s [CH. V

§2. Spemer’s lemma

§2.1. Spemer’s lemma.
2.1. Let T n = (e0 • • • en) be an n-simplex and let K n be a triangulation
of its closure T n. Let each vertex e'k 6 K n be made to correspond to a vertex
Se'k = e,k of the simplex T n in such a way that the following condition is
satisfied:
2.10. Se'k is a vertex of the carrier of e'k {i.e., a vertex of that face of T n
which contains e'k).
Then there exists an n-simplex
T \ = {e'H ■• • e'*J
of the triangulation K n such that all the vertices Se'k0 , ■• ■ , Se'kn are distinct.
Proof. Let T \ , • • • , T n, be all the n-simplexes of the triangulation
K n. Let us call a simplex 7”\ normal, if all its vertices have been made
to correspond to distinct vertices of T n. We shall prove the following
proposition which is stronger than Sperner’s lemma:
2.11. The number of normal simplexes is odd.
Theorem 2.11 is obvious for n = 0. Let us suppose th at it has been
proved for all (n — l)-simplexes and prove it for an n-simplex T n.
Let us say th at an (n — l)-face of a simplex 71",- is “marked” if its
vertices have been made to correspond to the vertices ex , e2, • • • , en of T n.
We note first that:
2.12. The number of marked faces of a simplex T n{ is either 1, or 2, or 0,
while the number of marked faces of T n{ is equal to 1 if, and only if, T n,
is normal.
Indeed, if a simplex T",- = (e',0 • • • e'if) is normal and
Se ,*q Co, Se ex , • , Se ,n en ,
then (e',! • • • e'if) is obviously the unique marked face of T n,•.
Suppose the simplex is not normal but has, nevertheless, a marked
face (e'i, • • • e',n), so that, for instance, Se'iv = ex , Se'i2 = e2, • • • , S e \n =
en . Then Se',0 is one of the vertices ex , ■■■, en , say Se'io = ex , and the
simplex 7”\- has exactly two marked faces, namely,

(e'u • • • e 'if and (e\-0e'fl • • • e'{f .

This proves 2.12.
Let us denote by the number of marked faces of a simplex T”1,-,
and let us set a = ^»= i G»'•
I t follows from what has just been proved that the number of normal
simplexes has the same parity as the number a. Therefore, it suffices to
prove th at the number a is odd.
§2] s p e r n e r ’s lem m a 161

Let us consider any (n — l)-simplex

T T 1 € K n.
Three cases are possible.
1. T jn 1 is contained in the interior of the simplex T n. Then either T jn~l
is not a marked face of any simplex T n; or the simplex T y"-1 is a marked
face of exactly two simplexes 71”,- and T '\ , which have T,"-1 as their com­
mon face. In the latter case T jn~l will be counted twice in the calculation
of the sum a = a{ .
2. The simplex T jn~l is contained in a face of T n distinct from the face
T - 1 = ( * • ■ ■ en).
In this case, by virtue of 2.10, the simplex T jn~l cannot be a marked
face of any simplex 7771,-.
3. The simplex T jn~l is contained in the face 77"-1 = (ei • • • e„).
This case we divide anew into two cases:
3a. At least two vertices of T jn~l have been mapped into the same
vertex of T n~x = (ei • ■■en). Then T ;n-1 cannot be a marked face of any
simplex T”\- .
3b. All the vertices of T,"-1 have been mapped into distinct vertices of
7”1-1 = (ei • • • en). Hence T ,n”' is a marked face of a unique simplex T ni
From this classification of the different cases it follows that the parity
of the number a is the same as the parity of the number of all (n — 1)-
simplexes 7V -1 which satisfy 3b.
Let us consider case 3b more closely. The simplexes of K n contained in
T n_1 and its faces form a triangulation K n~l of the closed simplex T n~l,
where every vertex of A "-1 has been mapped by S into one of the vertices
of T n_1 so as to satisfy the conditions of Theorem 2.1 for the dimension
(n — 1). Consequently, by the inductive hypothesis, the number of those
(n — l)-simplexes contained in 71"-1, all of whose vertices have been
mapped into distinct vertices of T n~l, is odd. In other words, the number
of simplexes TV-1 which satisfy 3b is odd, and this means that a is also
odd. This completes the proof of 2.1.
§2.2. Corollary of Spemer’s lemma. Conclusion of the proof of the
Pflastersatz. From Sperner’s lemma we deduce the corollary:
2.21. Let T n be an n-simplex with vertices eo, e i , • • • , e„ . Let
a = {A 0 , • • • , A n]

be a covering of the closed simplex T n, consisting o /n + 1 closed sets A o, ■■■ ,

A n satisfying the condition: every closed face [eT-0 • • • et-J of T n is contained in
A io u • ■• u A ir (in particular ei £ A ifo r i = 0, 1, • • • , n).
Then the intersection of all the sets A{ is nonempty.
162 SPEItNEIl’S LEMMA AND ITS COROLLA KIES [CH. V

Proof. Obviously (on the basis of Lcbesgue’s lemma, I, 8.32) 2.21 follows
from: _
2.22. If K n is a triangulation of the closed simplex T n of arbitrarily
small mesh, there is at least one closed simplex T ni of K n which intersects
all of the sets A 0, • • • , A„ .
To prove Theorem 2.22, let us assign to each vertex e'j £ K n a vertex
Se' j = ei of the carrier of e', such that
e'j 6 A i .
Obviously, the mapping S satisfies condition 2.10 of Sperner’s lemma.
Applying Sperner’s lemma to the mapping S, we find a simplex rF ni 6 K n
all of whose vertices have been mapped into distinct vertices of T n. But
this means th at the skeleton of T ni intersects all of the sets A 0, • • • , A n ,
whence 2.22 follows.
From Theorem 2.21 we easily deduce:
2.23. Let
a = [A0 , • • • , A n\
be a closed covering of a closed simplex T n = [e0 ■■■en] such that
1. e i 6 A i , _
2. A i has no points in common with the closed face 7V -1 opposite the
vertex e,-.
Then
A 0 n • • • n A n ^ 0.
To prove 2.23, it sufficesto show that the hypothesis of Theorem 2.23
implies the hypothesis of Theorem 2.21. Let [e,0 • ';,e ,0] be a closed face
of T n and let i be different from i0 , • • • , i r . Then [e!0 • • • e,J — T f ~ l ,
so that A i docs not meet [e,0 ••• eir]. Since this is valid for arbitrary
x , *** , Xr , [c,‘o etj\ A iq U * LI A (r .
Let us now pass to the proof of the second assertion of the Pflastersatz :
2.24. For sufficiently small t > 0 every closed t-covering of a closed n-
simplex T n has order > n + 1.
Proof. Let e be so small that no set of diameter less than e meets all
the closed (n — l)-faees of T n [since the intersection of all the closed
(n — l)-faces of T n is empty, such an e can be found by Lcbesgue’s lemma,
I, 8.32],
With e as above, let
a = [Ao , • • • , A„}
be any e-covering of T n. In virtue of the choice of e, no A , containing a
given vertex of T n intersects the closed face 71,”-1 opposite the vertex
§2 ] SPERNER S LEMMA 1G3

Ci . Hence, in particular, it follows that no A i contains more than one

vertex of T n.
Let us now number the elements of the covering so that
Co (z A o , • • • , en 6 A „ .
If s > n, let A j be any set with index j > n, and let TV'-1 be the closed
face which does not meet A j . Let us delete the element A j from the
covering a and replace Ai by /I, u A j , denoting this last set again by A i .
After this replacement the covering a is converted into a covering a(n,
which has one element less than a. The order of a (1), as easily seen, is in
any case no greater than the order of a. It is also easily shown that no
element of the covering a (n containing a vertex of T n intersects the closed
face opposite this vertex.
Repeating this operation as many times as necessary, we finally obtain
a covering
a (p) = {Ao, ••• , A n}
satisfying all the conditions of Theorem 2.23. Consequently, there exists a
point contained in all the sets A 0, • • • , A n , i.e., the order of the covering
a p) is not less than n + 1. But since the above replacement of one cover­
ing by another did not increase the order of the covering, the original
covering a had order > n + 1. This proves the theorem.
§2.3. The invariance of the dimension number of R n.
2.31. Every compaction <f> <Z R n has dimension <n.
Indeed, every compactum <f> <Z R n, being a bounded set, is contained
in some closed n-simplex T n\ since T n has closed e-coverings of order
n + 1 for every e > 0, <f> also has closed e-coverings of order n + 1 for
every e.
Hence:
2.32. ATo set .4 C /^" containing an interior point (relative to R n) can be
mapped topologically into R m, where m < n.
Indeed, such a mapping would take an n-simplex contained, by assump­
tion, in A, into some compactum of dimension < n. This is impossible,
since the dimension of a compactum is invariant under topological map-
pings.
In particular, R n itself cannot be mapped topologically into R m for
m < n. Hence it follows that:
2.33. I n v a r i a n c e o f t h e D i m e n s i o n N u m b e r o f R n. For n ^ m, the
spaces R n and R m are not homeomorphic.
From 2.33 and the definition of a topological manifold (I, 5.3) it follows
th a t:
2.34. I f m 9^ n, an n-manifold cannot be homeomorphic to an m-manifold.
164 s p e r n e r ’s lem ma and its c o r o l la r ies [CH. V

§3. Invariance of interior points

§3.1. A topological mapping of a set A C R n onto a set B C R n maps
every interior point of A into an interior point of B, and every noninterior
point of A into a noninterior point of B.
Here “interior point” (“noninterior point”) means “interior point”
(“noninterior point”) relative to R n.
The proof is based on the following lemma:
L emma 3.11. Let
« = {^-i j •' ' 1
be a closed covering of order n + 1 of a closed bounded set 4 C R n with the
property that there is exactly one point p € 4* contained in n + 1 elements
of a. Then i f p is not an interior point of 4, the covering a can be transformed
into a covering a' of order < n by a modification of a in an arbitrarily small
neighborhood of p.
[A modification of a covering a in a neighborhood Op of p is a transition
from a = {A\ , ■■■ , A,} to a closed covering a = {A'i , ■■■ , A',) such
that A i \ O p = A 'i \ O p , i = 1, • • • , s.]
Proof of the lemma. Let T n be an arbitrarily small n-simplex containing
p in its interior. Every point of the boundary || T n || of T n is contained in
no more than n elements of a. Therefore, there exists an e > 0 such that
the e-neighborhood of every point of || T n || intersects no more than n
elements of a.
Let us now take any closed e-covering
/3 = [B\ , • • • , B m)
of order n of the set || T n ||.
Let us assign to every set B j £ |9 a set £ a in the following way:
1. If Bj does not intersect <h, let A i(j) be any set A{ which meets T n.
2. If B j meets 4>, then, because of the choice of e, it meets no more than
n elements of the covering a ; let us denote any one of the elements of a
which meets B j by A^j) . In both cases the same set 4,- may, of course,
turn out to be the set At(j) for different B j’s.
Let us now construct a system a" of closed sets A " as follows:
Every A { 6 a which is not a set A ^ for any B j is, by definition, an
element A " of the system a but if A; is a set for one or several
B j’s, then define A " as the union of all such B f s and the set 4,-. Ob­
viously
u || 71" || c U A f , A " G a".
The systems a" and a are in (1-1) correspondence, where, for each i,
either A " = A i or A " is the union of the set A,- and certain subsets of
§3] INVARIANCE OF INTERIOR POINTS 165

the set || T n |j. Therefore, if a point q is contained in the sets A "t l , • • • ,

A " iT , q is contained in either the sets A,-, , • • • , A ir orin || T n ||(these
cases, of course, are not mutually exclusive). Therefore to show that p
is the unique point contained in n + 1 elements of a", it suffices to prove
th at no point of || T n || is contained in more than n elements of a".
Let us prove th at no point of || T n || is contained in more than n ele­
ments of the system a" .
Let q 6 || T n ||. If q $ <f>, then q is contained in no more than n sets
B j , say B h , • • • , B jm; the number of corresponding At-0-)’s, say /I,-,, • • • ,
.1 ,-ni , will not exceed n; the point q is contained only in the elements A",-, ,
• • • , A "im of a", i.e., in no more than n elements.
Xow let q (E || t n || n <f> and let A i , • • • , A h , B i , • • •, B k be all the
elements of the coverings a and containing q. All the sets B\ , • • • , B k
are contained in an t-neighborhood 0(q, e) of q, and certainly among all
the sets Ax, • • • , A* there will be no more than n elements of the cover­
ing a meeting 0(q, t). Thus let
Ai , • • • , A h , • • • , A g , g < n,
be all the elements of a meeting 0(q, e). All the A ia) , ' j = 1, 2, • • • , k,
are among these sets. Therefore Ai" , • • • , A Q" are the only elements of
a" which can contain q. Since g < n, the assertion is proved.
Let us now take in the interior of the simplex T n any point o not con­
tained in <t> (since p is not an interior point of <f>, the required point o can
always be found).
Let us consider those elements of the system a" which meet || T n ||;
let these be
A " )
Ail ... A ".
t Aiv
Denoting by [oA/'] the union of all closed segments of the form [oq],
where q 6 || T n || n A " , set
A ’" - [A " \ (A " n T n)\ u [oA/'}, i = 1, • ■• , v,
and
A l" = A " \ T n, i> v .
The set A / " can differ from A I' only in points of T n. Therefore a point
not in T n cannot be contained in more than n sets A /" . As for the points
of T n, only the point o can be contained i n n -f 1 sets A / " (since, in the
contrary case, if o' were such a point, the endpoint q of [oo'q] on || T ||
would belong to n + 1 sets A " , in contradiction to what was proved
above). Since o (£ <f>, setting
A \ = A / " n <f> for all i,
166 s p e r n e r ’s lem ma and it s c o r o l la r ies [CH. V

we obtain a covering
a ' = M \ , ••• ,A ',}
of the set of order <n.
Obviously, every A 'i can differ from the corresponding A i only in points
belonging to T n, but since the simplex T n can be taken arbitrarily small,
we are justified in saying that the covering a was obtained from the
covering a by a modification of the latter in an arbitrarily small neighbor­
hood of p. Hence the lemma is proved.
From the lemma we deduce
3.12. A topological mapping C of a closed n-simplex, T n into the Euclidean
n-space R n maps every interior point p of T n into an interior point of the
set S> = C (T n).
Proof. Let T n = (e0 • • ■en). Take the barycentric subdivision of the
boundary of T n and project it from the point p. The resulting triangulation
K of the closed simplex T n differs from a barycentric subdivision only in
that the “centroid” of the simplex T n is the point p. Let us denote by A
the union of the closures of all the simplexes of K having as a vertex.
The point p is the unique common point of the sets A 0 , • • • , A „ ; here
both the covering
a = {Ao, • • • , A n\
of the closed simplex T n and every covering which is obtained from a
by a modification of the sets A , in a sufficiently small neighborhood of p
satisfies 2.23.
Hence it follows that:
For a sufficiently small e > 0 every covering a' obtained from a by a
modification of the sets A; in an €-neighborhood of p is a covering of
order > n + 1.
Now let q = C(p) be a boundary point of the set = <7(7'"). Take
5 > 0 so small that the image of a 5-neighborhood of q (relative to <f>) under
the mapping (T 1 is contained in an e-neighborhood of p. Applying Lemma
3.11, modify the sets C (A{) in the 5-neighborhood of q in such a way that
the result is a covering /?' of order <n of the set <t>. C~l maps the covering
,3' into a covering a' of the same order <n, which is a modification of <y
in an e-neighborhood of p. This is impossible, and 3.12 follows.
We can now prove Theorem 3.1 in a few words. Let p be an interior
point of the set A C I f 1-, p is an interior point of some closed n-simplex
T n C A and, therefore, the mapping C of A C R n takes p into an interior
point q of the set C(T") and hence into an interior point of the set (7(A).
C maps every noninterior point p of A into a noninterior point C(p) of
§3] INVARIANCE OF INTERIOR POINTS 167

C(A), since, in the contrary, case C~l would map an interior point C(p) of
C (A ) into a noninterior point p of A .
This completes the proof of Theorem 3.1.
Theorem 3.1 immediately implies the following proposition, which is a
generalization in a merely formal sense, but which is required in the
sequel.
3.13. Let Un and T" be topological spaces homcomorphic to R "; let C be a
topological mapping of a set .1 C ( onto a set B C V n\ then C maps every
interior point of .4 relative to C n into an interior point of B relative to V'\
The following proposition is implicit in 3.*13:
3.14. Let Un and 1’" be homcomorphic to R n; let C be a topological map­
ping of a set .4 open in U" into the space V n. Then (7(A) is open in V n.
In particular, every topological image of the whole space Un in V " is an
open set in T".
Since R m is nowhere dense in R n for m < n, a topological mapping of
R n onto R m is impossible. We have therefore obtained another proof of
Theorem 2.33.
R e m a r k . Wc note separately a special case of Theorem 3.14, which
(for n = 2) was used in Chapter 111. .
3.140. Let l ' n and C" be two sets (contained in some topological space)
homcomorphic to R n. I f Un C T 71, then Un is open in V n.
D e f i n i t i o n 3.15. A compaction homcomorphic to a closed ?i-simplcx
is called a closed n-cell.
Since dimension is topologically invariant and the dimension of a simplex
is equal to its dimension number, the number n in Def. 3.15 is really the
dimension of an n-cell and is therefore unique (i.e., no topological space
can be both an ?i-cell and an m-cell for n ^ ?n).
Wc shall prove the following important proposition:
3.16. Every topological mapping of a closed n-simplex T n onto a closed
n-cell E n takes the boundary T n \ T r of T n onto a unique, set S n~ \ called
the boundary of the closed n-cell E n; the open set E n \ S n~l is dense in E n
and is denoted by E n\ it is called the interior of E n, or an open n-cell.
Proof. Let C\ and C2 be two topological mappings of T n onto E n. If,
e.g., it were true for a point i ( T n \ T n that
C2(x) C Ci(T" \ T n),
then it would be true that C~\ C2(x) (E T n \ Tn; however, C~\ C2 is_a topo­
logical mapping of T 71 onto itself, so that, by 3.1, C i C2 maps T \ T
onto itself. The resulting contradiction proves that
C2( f “ \ T n) C C,{Tn \ T n).
168 s p e r n e r ’s lem ma and its c o r o l l a r ie s [CH. V

In the same way,

Cx(Tn \ T n) C C2(T n \ T n),
i.e.,
Ci(Tn \ T n) = C2(T n \ T n).
Since T n is open and dense in T n, the set E n, being the image of T n
under a topological mapping of T n onto E n, is open and dense' in E n.
This proves Theorem 3.16.
§3.2. Invariance of interior points of topological manifolds. The follow­
ing very important theorem, a generalization of Theorem 3.1, may be
easily derived from 3.1:
3.2. A topological mapping of a subset A of a topological n-manifold
M "] onto a subset B of a topological n-manifold M n2 takes every interior point
of A (relative to M n\) into an interior point of B (;relative to A In2), every
noninterior point of A (relative to M nf) into a noninterior point of B (rela­
tive to M n2).
Proof. We may restrict ourselves to the proof of the assertion for interior
points. After that, the proof for noninterior points is a repetition, word
for word, of the last lines of the proof of Theorem 3.1.
Hence, let C be a topological mapping of A C M ni onto B C M n2 .
Let p be an interior point of A relative to M ni . Let V n be a neighbor­
hood of C(p) in M n2 homeomorphic to R n.
Let Un be a neighborhood of p in A homeomorphic to R n, such th at
C(JJn) C F".
Since Un and V n are homeomorphic to R n, C(Un) is open in F"; but F"
is open in M n2, so that C(Un) is open in M n2 . Since p £ Un Q A , C(p) €
C{Un) C B; this completes the proof.

§4. Fixed point theorem for an n-cell

We shall prove the following theorem:
F i x e d P o i n t T h e o r e m f o r a n w - C e l l . I f C is a continuous mapping
of a closed n-cell E n into itself, C has at least one fixed point [i.e., there is a
point p £ E n such that C(p) = p).
We may obviously suppose, without loss of generality, that E n is a
closed n-simplex
T n = (e0 - • • e„).
L etC be a continuous mapping of T n into itself and let C map the point
p € T n with barycentric coordinates m0 , • • • , mn into the point p' =
C(p) with barycentric coordinates m '0 , • • • , m 'n (the barycentric co­
ordinates are with respect to the system {e0, • • • , en)).
§4] FIXED POINT THEOREM FOR AN 71-CELL 169

Denote by A i the set of all those points p E T n for which m'i < m l . It
is clear th at every set A i is closed. We shall prove that the sets A { form a
covering of T n satisfying the hypothesis of Theorem 2.21, i.e., that an
arbitrary closed face [e,0 • • • elr] of T n is covered by the sets A io , ■• • ,
A ir . Let p be an arbitrary point of the closed face [e,0 ••• eir]. Then
mio + • • • + m ir = 1 > m 'i0 + • • • + m 'ir , whence it follows that
m'ik < m ik for at least one ik . Therefore p G A ik.
Hence Theorem 2.21 can be applied to the sets A i to yield a point p
contained in every set H,-.
For this point p
m'o < m0 , • • • , m 'n < mn .
Hence, in virtue of the conditions
m'o + • • • + m 'n = m0 + • • • + 77in = 1, 0 < m 'i ,
it follows th at
m'o = m0, • ■■ , m 'n = m„ .
In other words, C(p) = p, q.e.d.
 Chapter V I
IN T R O D U C T IO N TO D IM E N S IO N T H E O R Y

All the concepts introduced in I, 8 are presupposed in this chapter.

The definition of dimension was given in I, 8.4. The basic theorem with­
out which the very concept of dimension would have no right to existence,
namely Brouwer’s theorem that the dimension of a closed simplex is equal
to its dimension number, was proved in Chapter V. It is easy to deduce
from this theorem (V, Preliminary Remarks) that the dimension of a poly­
hedron is equal to the dimension number of an arbitrary triangulation of
the polyhedron. This to some extent justifies the general concept of dimen­
sion. However, only to some extent; the introduction of a new fundamental,
and not merely serviceable concept, is justified only if it substantially
clarifies some sufficiently important branch of mathematics, and does not
serve merely to prove one or two theorems. The concept of dimension
fully justifies itself in this respect: the dimension theory developed in the
last quarter of a century represents an integral part, geometric in content,
of general topology. It has provided a deeper insight into various geo­
metric properties, not only of eompacta, but to some extent of their arbi­
trary subsets as well.
The full development of dimension theory is achieved only in the so
called homology dimension theory, which is not presented in this book:
it is based on certain theorems of the combinatorial topology of polyhedra
which are not adduced here. We shall therefore be compelled to restrict
ourselves to the first fundamental theorems of dimension theory, i.e., to
give an introduction to this theory and refer the reader to deeper studies
of the modern theory in the journal literature. (In addition to the classical
papers of Brouwer [a] and Urysohn [a], and Monger's book [M], see Aleksan­
drov fb, d]. A systematic account of dimension theory, with an introduction
to the geometric theory, is to be found in Ilurewiez-Wallman [II-W].)

§1. Theorems on e-displacements and imbedding in R n

§1.1. Definition of e-mappings and e-displacements. Outline of the
section.
D e f in it io n 1.11. A c o n tin u o u s m a p p in g C of a c o m p a c tu m Ar in to a
e o m p a c tu m Y is ealled an t-mapping if th e in v e r se im a g e C~l(y) C A" of
e v e r y p o in t y of Y h a s d ia m e te r < e .
D e f in it io n 1.12. Let the compactum Ar be a subset of a metric space
R. A continuous mapping C of Ar into R is called an e-displacement of the
170
§ 1] THEOREMS OX ^DISPLACEMENTS AND IMBEDDING IN It 171

compactum X in the space It if p(x, Cx) < e for every x £ X . If an e-dis­

placement C maps two points x and x' of X onto the same point y, then
p(x, x') < p(x, y) + p(y, x') < e -f e = 2e.

Hence the diameter of C~l{y), equal (in viewr of the compactness of X )

to the maximum of p (x,x')io rx 6 C~\y), x' £ C~\y), is less than 2c.Hence
1.13. A n e-displacement C of a compactum X is a2e-mapping of X onto
C(X).
E x a m p l e . Let A" be the Hilbert parallelotope (I, 2.3). Let us assign to
every point

j *** j %n j ' *' ) £ X

the point

C(x) = ( x i, • • • , x n , 0, 0, 0, • • •) € X.
The resulting mapping C of the Hilbert parallelotope onto an n-dimen-
sional rectangular parallelotope is a ( l/2 n-1)-displacement. Therefore:
1.140. For every e > 0 there is an e-displacement of the Hilbert
parallelotope onto a polyhedron, namely a finite-dimensional parallelotope.
Hence:
1.141. Every closed subset of the HilberL parallelotope can be mapped
into a polyhedron by means of an e-displacement for every e > 0.
Since all compacta are homeomorphic to subsets of the Hilbert parallelo­
tope, it follows easily from 1.141 that:
1.142. A compactum can be e-mapped into a polyhedron for every e > 0.
E x e r c i s e . Prove that a closed subset of Hilbert space can be mapped
by an e-displacement for every e > 0 into a polyhedron (and in general
into a closed bounded set of a finite-dimensional Euclidean subspace of
Hilbert space) only if it is a compactum. Prove the analogous proposition
for e-mappings (instead of e-displacements).
O u t l i n e o f t h e F i r s t S e c t i o n . The first fundamental purpose of this
section is to prove the following proposition:
T h e o r e m 1.4 (Theorem on e-mappings). Let$> be a compactum. For every
e > 0, can be e-mapped onto a polyhedron-, in this connection, if dim 4> = n,
then <t>can be e-mapped for every e onto an n-dimensional polyhedron and, for
sufficiently small e > 0, cannot be e-mapped into any polyhedron of dimension
< n. I f the compactum 4> is a subset of Hilbert space or of a Euclidean space,
these e-mappings can be realized by e-displacements for every e.
The proof will proceed in the following order:
First we shall prove that for every n-dimensional compactum 4> there
172 INTRODUCTION TO DIMENSION THEORY [CH. YI

exists an e = e(4>) such that every compactum which is the image of

under any 6-mapping has dimension > n (Theorem 1.16). The proof of
this theorem is based on Lemma 1.15.
Next, we shall prove that the existence of an 6-mapping of the com­
pactum $ into a polyhedron II implies the existence of ail e-mapping of
4> onto a polyhedron n0 C II.
This transition from mappings into a polydedron to mappings onto a
polyhedron is made very simply by means of a retraction (1.2). Along the
way we shall prove the important Theorem 1.25 characterizing n-dimen -
sional compact subsets of R n.
The central part of the proof consists in constructing for every e > 0
an e-mapping of the given n-dimensional compactum into an n-dimensional
polyhedron. This is done in 1.3 and 1.4 by means of the so called barycentric
mappings.
Finally, it remains to be proved that if 4> is a subset of Hilbert space or
of Euclidean space R m, the 6-mapping is realized by an 6-displacement.
This is easily achieved in 1.5 (since the proof is based on IV, 3.1, Remark
2, and consequently makes use of properties of convex polyhedral domains
not proved in this book, we accordingly give several weakened formula­
tions independent of propositions not proved here). The method of proof
consists in constructing for every continuous mapping C of the compactum
4> into the given space R m a “barycentric mapping Ca” (see above) into a
polyhedron of the same dimension as 4>, differing from C by as little as de­
sired.
If C is the identity mapping, Ca is an 6-displacement.
These barycentric approximations to continuous mappings also lead
to the solution of the second basic problem of this section, namely to the
proof of Theorem 1.6, which asserts that every n-dimensional compactum
is homeomorphic to a (closed and bounded) subset of R 2n+l: it turns out
that the barycentric mappings approximating a given continuous mapping
of an n-dimensional compactum <I>into R m, where m > 2n + 1, sufficiently
well, are 6-mappings. Hence for every e the set of all 6-mappings of an n-
dimensional compactum 4> into an m-dimensional closed solid sphere
E m, m > 2n + 1, is dense in the space £.(<!>, E m) of all continuous mappings
of 4> into E m.
Since &(<!>, E m) is a complete metric space and the 6-mappings form an
open set in £(4>, E m), it follows easily th at topological mappings of <t> into
E m not only exist but even form a set dense in £(4>, E m). This will prove
Theorem 1.6 and the stronger Theorem 1.63.
L e m m a 1.15. Let C be an t-mapping of a compactum X into a compactum
Y. There exists a positive number y such that every set B C Y of diameter
<tj has an inverse image C ~\B) of diameter <e,
§1] THEOREMS ON €-DISPLACEMENTS AND IMBEDDING IN Hn 173

Proof by contradiction. Suppose such an 77 does not exist. Then for every
natural number n there is a set B n in Y of diameter < 1 /n whose inverse
image A„ = CT1(Bn) has diameter >e.
Let xn and x 'n be points ol A n for which
p(xn , x 'n) = S(An) > e.
Passing, if necessary, to subsequences, we may suppose that the sequences
{xn} and {x 'n} converge to points x 6 X and s ' 6 X , respectively. Since
p(Cxn , Cx'n) < 1/n, Cx = Cx' = y\ on the other hand, p(x, x') > e, which
contradicts the fact that C is an e-mapping.
Let us apply Lemma 1.15 to the proof of the following im­
portant proposition:
1.16. Let the dimension of a compactum <h be equal to n. There exists an
e = e(<h) > 0 such that every compactum <£' which is the image of <f> under
an e-mapping has dimension >n.
Proof. Choose an e such that the compactum <J>has no closed e-covering
of order < n. Let
= C(*),
where C is an e-mapping. Let us define a number rj for this e-mapping
as in 1.15., and take any closed 77-covering
a' = {A'i, ••• , A' . }
of the compactum Then the sets A t = C~l(A'i) form an e-covering a
of the compactum $ whose order is equal to the order of the covering a'.
By th e d e fin itio n o f e, th e order of th e c o v e r in g a is in a n y ca se > n + 1;
th e sa m e is tr u e fo r th e ord er of th e c o v e r in g a . H e n c e e v e r y 7 7 -c o v erin g

o f th e c o m p a c tu m <£' h a s ord er >n + 1 , w h en ce

dim <J>' > n.

§1.2. Retraction.
1.21. R e t r a c t i o n T h e o r e m . Let $ be a closed subset of a polyhedron
II and let K be a triangulation of IT. There exists a subcomplex K ' of K and
a continuous mapping C of the compactum onto the polyhedron || K ' || with
the following property: if x £ < t>, there is a simplex T £ K whose closure con­
tains both points x and C(x).
Proof. Let 7\ , • • • , T, be all the simplexes of K containing points of $
enumerated in an order such that the dimension of T i+i does not exceed
the dimension of 7\-.
Setting let us denote by Co the identity mapping of $0 and sup­
pose th at the set and the mapping C» have already been defined. If
T i+1 C , set $ i+i = and denote by Ci+1 the identity mapping of .
174 INTRODUCTION TO DIMENSION THEORY [CH. VI

If T i+1 , take a point oi+l 6 T<+i \ 4>i and denote by C<+i the retrac­
tion of 4>t onto $ f \ T;+i , defined on \ T;+i as the identity and on 4>,- n
T {+ 1 as the central projection of 4>,- n T t+1 from the point ot-+1 into that
part of the boundary r i+i \ T {+1 of T i+1 which is contained in $,•. Let
4>i+i = Ci+i($i). Proceeding in this way step by step, we finally arrive at
a compactum 4>s , which is the body of some subcomplex K ' of K , and a
mapping
C = CaCs-i ■■•Co,
which is easily seen to be the desired mapping of 4> onto the polyhedron
II K ' || = * ..
1.22. Let Q be a closed subset of a polyhedron II; for every t > 0 there exists
an e-displacement C of the compactum into the polyhedron II which maps
4> onto a polyhedron II', the body of some subcomplex K ' of some triangulation
K of 4>.
To prove this, it suffices to take a triangulation K of the polyhedron II
of mesh < t, and to'apply Theorem 1.21.
From 1.141 and 1.22 it follows that:
1.23. Every compact subset of the Hilbert parallelotope can be mapped
onto a polyhedron by means of an e-displacement for every e > 0.
From 1.142 and 1.23 it easily follows that:
1.24. Every compactum can be e-mapped onto a polyhedron for every
e > 0.
1.25. In order that a compactum 4> in R n contain interior points (relative
to R n) it is necessary and sufficient that
dim <t> = n.
Indeed, since Cl R n,
dim 4> < n ;
if <f> contains interior points relative to R n and consequently some n-di-
mensional simplex T", then dim <f> > dim T n = n so that dim 4> = n.
Now suppose that the compactum 4> (Z R n does not contain any interior
points.
Since 4> is a closed bounded set in R n, there exists a simplex T n 3 4>.
Take a triangulation K of the closed simplex T" of mesh less than a
given «. Since 4> does not contain any n-simplex, in particular, any ?i-simplex
of the triangulation K, each of the ?i-simplexes Tnt- 6 K contains a point
Oi not belonging to<f>. Let us now apply the retraction of 1.21, i.e., let us
map the set 4> n T ni into T nt \ T n by means of a central projection from
the point ot- . Since this leaves all the points of 4> contained in T",- \ T n
fixed, the composite of the retractions defined in the various ?i-simplexes of
§1] THEOREMS ON t-DISPLACEM ENTS AND IMBEDDING IN R n 175

K is a continuous mapping of $ into an (n — 1)-dimensional polyhedron

which is the union of the boundaries 7in; \ 7int-of all the n-simplexes of K.
Hence it follows that for every t > 0 there exists an t-mapping of the
compactum $ into some (n — l)-dimensional polyhedron. Hence, by virtue
of Theorem 1.16,
dim < ?i — I.
This completes the proof of 1.25.
§1.3. The barycentric mapping of a space into the nerve of an open
covering. Let
w = {Oi, ••• , 0 S}
be an open covering of a metric space X and let K a be a simplicial complex
in some R n isomorphic to the nerve of the covering w. The elements of K u
are simplexes (perhaps degenerate) of R n.
The union ]| K u || of all the nondegenerate simplexes and the closed
convex hulls of the degenerate simplexes of K u will be called the body of
the complex K u . Since K u is an unrestricted complex, its body, in the
sense just explained, can also be defined as the union of the closed convex
hulls of all the degenerate and nondegenerate simplexes of K a .
Let us define in X the continuous functions , i = 1, 2, • • • , s, by
setting
m(x) = p(x, X \ 0 {), x 6X ,
and let us construct a continuous mapping Cu of X into R n as follows: let
x be an arbitrary point of X ; the vertices ax , • ■• , as of K u corresponding
to the elements Oi , • • • , 0, of the covering u are assigned the weights
Pi(x), • • • , Pe(x), respectively; the centroid of these masses is then, by
definition, the point Cu(x). Since the functions m(x) are continuous, Cu
is a continuous mapping of X into R n. Let us prove that Cu is a mapping of
X into the body of K a . To this end, we note that the function n fx ) does
not vanish only at points of the set O i. Therefore, if £ E I and x is con­
tained in 0 {j (0 < j < r), and only in these elements of w, then m ^x) ^
0 (0 < j < r) and all the remaining functions m vanish at x.
Since x £ fl)=o 0,,. , K u contains a simplex with vertices a,-; (0 < j < r)
and Cu(x) is obviously contained in the closed convex hull of the skeleton
(a,-.(0 < j < r)}. This proves the assertion.
D e f i n i t i o n 1.31. The continuous mapping Cu of X into || K u || just
constructed is called the barycentric mapping of corresponding to the
covering u and the given choice of the vertices of the nerve K u of w.
We shall now give several applications of barycentric mappings.
§1.4. Theorem on c-mappings. In the notation of the preceding article,
176 INTRODUCTION TO DIMENSION THEORY [CH. VI

let the order of the covering w be equal to r + 1. Then if n is sufficiently

large, namely if n > 2r + 1, choosing in R n vertices cq , • • ■ , a, in general
position yields a triangulation K a isomorphic to the nerve of the covering
w. Under these conditions, Theorems 1.41 and 1.42 hold:
1.41. The inverse image of the simplex T T = (a,0 ■• • a,-r) £ K w under the
barycentric mapping Cu of X into || K u || is
C - 1^ • • • air) = 0 {j \ U '0 f ,
where U '0t- is the union of all 0 t- for which i ^ io, • ■• , iT•
Indeed, since K u is a triangulation, the various simplexes 7\- £ K a are
mutually disjoint; therefore, if a point v = Cu(x) is contained in a simplex
(a,0 • • • a{r), it is not contained in any other simplex of K u ; in other
words, Ait,(z) > 0 (0 < j < r) and m(x) = 0 (i 9^ io , • • • , iT)- But this
means that
x € aw it-.\ u'o,..
1.42. If o} is an e-covering, e > 0, then the barycentric mapping Cu is
an e-mapping of the space X into the polyhedron || K a || , whose dimension
is one less than the order of the covering w.
Hence, by Theorem 1.16, it follows that:
1.43. Every r-dimensional compactum can be e-mapped into some r-di-
mensional polyhedron for every e > 0 and cannot be e-mapped into any
polyhedron of smaller dimension for a sufficiently small e.
This result may be strengthened as follows:
Let co — {Oi, • ■■ , 0,} be an open e-covcring of order r + 1 of a com­
pactum X of dimension r. Let the triangulation K u situated in some R n
be isomorphic to the nerve of the covering co. Let.us consider the bary­
centric mapping Ca of the compactum X into || K a || ; by Theorem 1.42,
Ca is an e-mapping into || K a || . Let us now subject the set (7U( X ) C
|| K a || to a retraction C (1.21). The result is a closed subcomplex K 'a
of K a and a continuous mapping C’CU of X onto || K 'u || . Recalling the
definition of C', it is easy to see that C'Ca can map onto any point y of a
simplex T' = (aio ■• • air) of K 'a only a point x £ X which is mapped
by Ca into some simplex T having T' as a face and consequently having
the points at-0 , • • • , aif among its vertices. Hence the inverse image of the
point y £ T' under the mapping C'CUis a subset of , so that (C'Ca)~ly
has diameter <€.
Hence
Every r-dimensional compactum can be e-mapped onto some r-dimensional
polyhedron for every e > 0 and cannot be e-mapped into any polyhedron of
dimension <r for sufficiently small <.■
In other words;
§ 1] THEOREMS ON t-DISPLACEM ENTS AND IMBEDDING IN R n 177

1.44. The dimension of a compactum 4> may be defined as the least r with
the property that for every e > 0 there is an e-mapping of 4> onto a polyhedron
of dimension number r. I f there is no number r with this property, the dimen­
sion of 4> is infinite.
§1.5. Barycentric approximations of a given continuous mapping of a
compactum 4> into R n. Theorem on e-displacements. Let C be a continuous
mapping of a compactum 4> into R n and let e > 0. Let a > 0 be so small
th at p(x, x ') < a (in 4>) implies that p(Cx, Cx') < e/4 (in R n)\ for the rest,
ct is perfectly arbitrary. Let
w = {Oi, • • • , Oa)
be an open cr-covering of 4>. In each set 0; choose a point o,- ; choose mu­
tually distinct points a,- £ R n to satisfy the condition
(1.50) p(Co{ , a t) < e/4
and make them the vertices of the nerve K u of the covering «. By the
simplexes of the complex K u we shall again understand the simplexes
(degenerate or not) defined by the corresponding skeletons of K u ■ The
body || K u || of the complex K u and the barycentric mapping Cu of 4> into
|| K a || is defined as in 1.3. Let us call this mapping the barycentric e-ap­
proximation of the mapping C (corresponding to the covering w and the
choice of the vertices a,-). This name is justified by the fact that for every
x 6 4>
(1.51) p (Cx , Cux) < e.
Indeed, let x be an arbitrary point of 4>. Let O,0 , • • • , 0 ir be all the
elements of u containing x. Then the distance of the point x from each of
the points oio , ■• • , 0 {r is less than <r, so that the distance of the point Cx
from each of the points Cot0 , • • • , Coir is less than e/4. In virtue of (1.50),
p(Cx, af) < e/2, i = i0, • • • , iT. In other words, all the points aio , • • •,, air
are contained in the interior of a sphere of radius e/2 with center Cx. The
interior of this sphere contains the closed convex hull of {fl»0 , • • • , u,r) •
Hence the open sphere also contains Cux. This proves (1.51).
Let dim 4> = r. Then the covering w can be taken of order r + 1 so that
K u is an r-complex and || K a || is the union of a finite number of closed
simplexes of dimension < r (in virtue of IV, 3.1, Remark 2, || K u || is an
r-dimensional polyhedron, but we do not wish to use this fact here).
Hence
1.51. For every continuous mapping C of an r-dimensional compactum
into R n there exists a continuous mapping Ca of 4> into a subset || K u ||
of the given R n, differing from C by as little as desired. || K u || is the union
of a finite number of closed simplexes of dimension <r. The mapping Cu
178 INTRODUCTION TO DIMENSION THEORY [CH. VI

can be taken as the barycentric approximation CL of the mapping C con­

structed by means of an arbitrary sufficiently fine open covering « of <t> having
order r + 1.
R em ark. If C maps 4> into the sphere E n, we may also assume that
|| K u || Q E n. Indeed, if p(C, CL) is less than some given e, K u is contained
in a sphere E n0 of radius p + e concentric with E n, where p is the radius
of E n. Let C' be the similitude which takes E n0 onto E n; C' maps each of
the simplexes of K u into a simplex contained in E n and the union of these
simplexes is || K 'u || = C'(|| K u ||) C E n. The mapping C'CU differs from
C by less than e + e = 2e and transforms 4> into || K 'u || CZ E n. We note
in this connection that if Cu is a a-mapping (for some a > 0 ), then, since
C' is (1-1), C'CU is also a a-mapping.
As before, let' dim = r and, in addition, let n > 2r + 1. The positive
number e is arbitrary, a sufficiently small: namely, p(x, x') < cr in ^ im­
plies p{Cx, Cx') < c/4 in R n. Let a a-covering w of $ have order r + 1 .
Choosing the vertices a\ , • • • , a* in R n in general position, we obtain an
r-dimensional triangulation K u . The mapping Cu satisfies both (1.51) and
the hypothesis of Theorem 1.41 (stated at the beginning of 1.4) and is
therefore a o--mapping.
Hence, applying the remark just made, we have
1.52. Let C be an arbitrary continuous mapping of an r-dimensional
compactum 4> into the sphere E n, where n > 2r + 1. For every positive e and
a, there exists a a-mapping CL of <f> into some r-dimensional polyhedron
|| K || C E n such that p(Cx, Cux) < efor all x £ 4>.
The following proposition is essential for the sequel. It follows at once
from 1.52.
1.520. Let n > 2r + 1. For every a > 0 the set of all a-mappings of an
r-dimensional compactum into the closed solid n-sphere E n is dense in the
space G(4>, E n) of all continuous mappings of 4> into E n.
Let us note yet another corollary of Theorem 1.52. Let C be the identitj’
mapping; then the mapping Cu is an t-displacement and we obtain the
following result:
1.521. For every r-dimensional compactum 4> C R n and for every c > 0
there is an e-displacement of 4> onto some r-dimensional polyhedron || K u [|
situated in the space R 2n+1 containing the given space R n.
R emark 1 . Let R m be an arbitrary Euclidean space containing the
r-dimensional compactum 4>. If m > 2r + 1 , set n = m; if m < 2r + 1 ,
set n = 2 r + 1 . Let us choose the vertices a,- C R n of the complex K u in
an arbitrarily given neighborhood of R m C R n.
Let us apply Theorem 1.521 to produce an e-displacement of the com­
pactum 4> into || K u || . Returning the vertices of K u to the space R m by
a small displacement Cx, we obtain a mapping C\ of the polyhedron
§1] THEOREMS ON €-DISPLACEMENTS AND IM BEDDING IN R" 179

|| K u || onto the set (7,(11 K„ || ) which is the union of a finite number of

closed simplexes [of the form C ^T ), where T £ /vu].
Since these simplexes have dimension < r and at least one is an r-sim-
plex, Ci( || K u || ) is an r-dimensional polyhedron (IV, 3.1, Remark 2).
Hence
1.522. For every r-dimensional compactum. $ a R n and for every t > 0
there exists an (.-displacement of <J> onto an r-dimensional polyhedron (Z Rn.
Finally, let be an r-dimensional compactum in the Hilbert parallelo-
tope. Let € > 0 and let
u — [ 0 ,, • • • , 0,}
be an open e-covering of order r + 1 of 4>.
Let C be an e-displacement of $ into some R n, n > 2r + 1 . A barycentric
approximation C„ of C can be chosen so that it will be a mapping of <f>
into a polyhedron || K u || C R n whose triangulation K u is isomorphic to
the nerve of the covering w and has mesh < 2 e.
The mapping Cu differs from the e-displacement C by less than e and is
therefore a 2 e-displacement.
Finally, subjecting Cu($>) C || K a || to the retraction C', which under
our conditions is a 2e-displacement, we obtain a 4e-displacement C'CU of
4> onto a polyhedron || K 'u || , K 'u Q K u , whose dimension is <r. For a
sufficiently small e the dimension of the polyhedron || K 'u || , by 1.16, can­
not be less than r and hence is equal to r.
Therefore
1.530. For every r-dimensional compactum 4> contained in the Hilbert
parallelotope and for every e > 0 there is an t-displacement of 4> onto some
r-dimensional polyhedron.
From what has been proved it follows that:
1.53. The dimension of a compactum <t> contained in the Hilbert parallelo­
tope can be defined as the least number r satisfying the following condition:
for every e > 0 there exists an (-displacement of the compactum 4> into the
Hilbert parallelotope mapping 4> onto some r-dimensional polyhedron.
R emark 2. From the proofs of the theorems of this article it follows
th at it is always possible to take for the polyhedron onto which 4> is mapped
by means of an t-displacement the body of some subcomplex of the nerve
of every sufficiently fine open (or, by I, 8.331, closed) covering of 3>; the
nerve itself must be constructed in a sufficiently small neighborhood of
the given covering w (IV, 2.1, Remark 4).
§1.6. Theorem on imbedding r-dimensional compacta in R 2t+1.
We shall prove the following fundamental theorem:
T heorem 1.6. Every r-dimensional compactum is homeomorphic to a
subset of R 2r+1.
180 INTRODUCTION TO DIMENSION THEORY [CH. VI

The proof is based on Theorem 1.520 and on the following lemma:

L emma 1.61. Let X and Y be arbitrary compacta and let e > 0; let £(X, Y)
be the space of continuous mappings of X into Y. The set of all t-mappings of
X into Y is open in £(X, Y).
It suffices to prove that every mapping C' £ &(X, Y) which is sufficiently
near [in terms of the metric in £(X, F)] an €-mapping C, is itself an e-map-
ping. Let us define for the number e and the given e-mapping C a number
y in accordance with Lemma 1.15 and assume that C' satisfies the in­
equality p(C, C') < y/2. Now let x x and .r2 be two points of X , mapped by
C' into the same point y £ F. Then
P(CXl, Cx2) < P(Cxx, C'xi) + p(C 'xi, C'xf) + p{C’X i, Cx2)
< ^/2 + 0 -f- y / 2 = y,
whence, by the definition of y, p{x1 , xf) < e, i.e., C’ is an e-mapping. This
proves Lemma 1.G1.
Let us now pass to the proof of Theorem 1 .6 .
For every natural number m denote by Tmthe set of all (l/m)-mappings
of an r-dimensional compactum <t> into the (2 r + l)-dimensional closed
solid sphere E n. By Lemma 1.61, Tm is an open subset of £(<!>, E n)\ by
Theorem 1.520, Tm is dense in £(<!>, E n). By I., 7.31, £(<£, E n) is a complete
metric space. Consequently in virtue of I, 7.14, the intersection of all the
r mis a nonvacuous set S [this set is even dense in £(<£, E n)\. The elements of
the set @are continuous mappings of 4> into E n, each of which is an e-map­
ping for arbitrary e > 0. Hence it follows that all the elements of the set
@ are ( 1- 1 ) mappings of 4> into E n, and since 4> and E n are compacta,
all these mappings are topological, ;, q.e.d.
R emark 1 . The proof of the following theorem is implicit in the above
arguments:
T heorem 1.62. For every continuous mapping C of a compactum 4> of
dimension r into the Euclidean space R n, where n > 2r + 1, and for every
e > 0 there exists a topological mapping Co such that
p(Cx, Cox) < t
for all 1
R emark 2. From Theorem 1 .6 .
it follows th at among the topological
spaces the finite dimensional compacta are characterized as those which
are homeomorphic to closed bounded subsets of Euclidean spaces.

§2. Theorem on essential mappings

§2.1. Definition and statement of the theorem. Let C be a continuous
mapping of a compactum $ onto a closed n-cell E n with boundary £ n-1 =
§2] THEOREM ON ESSENTIAL MAPPINGS 181

E \E [V, Theorem 3.16 (in the sequel E n will be either a closed n-sim-
plex or a solid n-sphere)]. Let $o denote the inverse image of S n~l under
the mapping C, i.e. the set of all points of $ mapped by C into <Sn-1. The
mapping C is said to be essential if every continuous mapping Ci of <t> into
E n coinciding with C on $ 0 is a mapping onto E n.
This definition implies that if C is not an essential mapping, there exists
a continuous mapping Ci of $ into E n coinciding with C on <h0 such that
some interior point o of E n is not contained in Ci(<t>). [Indeed, if the entire
interior E n of E n vrere contained in Ci($), then, since Ci($) is closed, we
would have Ci($) = E n, i.e., Ci would be mapped onto E n '\
This definition of essential mappings is very simple but not very intui­
tive: the intuitive meaning of essential mappings is clarified in the very
best wray by the following remark:
2 . 1 1 . Let us call a deformation Ce of a mapping C = Co, 0 < 0 < 1 , for
which all the mappings Ce coincide with the mapping C on <h0, an admissible
deformation. Hie mapping Co is inessential if, and only if, there is an ad­
missible deformation of Co into a mapping Ci such that
C & ) c S n~\
Proof of 2 .1 1 . W ithout affecting the generality of the proof, we may as­
sume th at E n is a solid sphere bounded by *S”-1.
If an admissible deformation Cj exists, the mapping Co is obviously
inessential in the sense of the definition given at the beginning of this
article.
Now let Co be inessential in the first sense. Then there exists a mapping,
which we shall denote by C i, coinciding with C0 on <50 and such that some
point o 6 E n is not contained in the set Ci($). Let us note first that there
is an admissible deformation of Co into C\ ; it suffices, for arbitrary x £ <t>
and arbitrary 0 , 0 < 6 < %, to denote by Cex the point which divides the
directed segment (Cox, Cjx) in the ratio 6: (§ — 0 ).
Let us nowr denote by Ce , h < 0 < 1, the deformation of the mapping
Ci defined by a central projection of the set Ci(4>) from the point o into
S n_1: for each x £ $ and 0, £ < 0 < 1, construct the ray oC{(a:), denote by
Ci(x) the point of intersection of this ray with the sphere <Sn_1, and by
Ce(x) the point which divides the segment C\(x)Ci(x) in the ratio (0 — ^):
(1 — 0 ). The resulting deformation Ce, 0 < 0 < 1, is the desired admissible
deformation.
We shall prove the following proposition in this section. This proposi­
tion, as also the theorem on e-mappings (on e-displacements) proved in
the preceding section, is one of the fundamental theorems of dimension
theory.
T h e o r e m 2.1, A n r-dimensional compactum can be mapped essentially
182 INTRODUCTION TO DIMENSION THEORY [CH. VI

onto a closed r-simplex; every continuous mapping of an r-dimensional com-

padum onto a closed n-cell is inessential for n > r.
R e m a r k 1. In particular, every continuous mapping of a segment onto
a square (Peano curve) is inessential.
R e m a r k 2. Theorem 2.1 implies that the dimension of a compactum $
may be defined as the maximum r for which it is possible to map $ essen­
tially onto a closed r-simplex or, equivalently, onto a solid r-sphere.
§2.2. Proof of Theorem 2.1. We shall prove the second assertion of
Theorem 2.1 first: for n > r every continuous mapping of an r-dimensional
compactum <t>onto an n-cell E n is inessential. W ithout affecting the general­
ity of the result it may obviously be assumed that E n is a closed solid
n-sphere.
Let C be a mapping of $ onto E n\ by Theorem 1.51 (and the remark to
this theorem) there is a mapping Cu , which differs from C by as little as
desired, of $ onto a set || K u || C E ” which is the union of a finite number
of closed simplexes of dimension < r; since r < n, the set || K a || is no­
where dense in E n. Hence, if we prove the following proposition, it will
follow th at C is inessential:
2 .2 1 . Let C be an essential mapping of a compactum <I>into a solid n-sphere
E n\ there exists an t > 0 such that for every continuous mapping C\ of $
into E n differing from C by less than e [i.e., such th at p(Cx, C\x) < e for all
x 6 $] there is a solid n-sphere L ni concentric with E n contained in Ci (<!>).
We shall prove the stronger proposition:
2.22. Let C be an essential mapping of a compactum $ onto a solid n-sphere
E n of radius 1 with center o. Let E n\ be the solid sphere of radius £ concentric
with the sphere E n.
Let S n~l be the bounding sphere of E n and $ 0 = CP1(S n~1).
If e < then for every continuous mapping Ci of into E n satisfying the
condition
p(Cx, Cix) < ( for all x 6 $0 ,
we have E n\ C Ci ("!>).
Proof of 2 .2 2 . Let e < 5 ; let E nr be the solid sphere of radius r with
center 0 . The mapping Ci by hypothesis differs from C = Co by less than
e at all x £ $ 0 . Let x be an arbitrary point of 4> and 0 < 6 < 1 ; denote by
Cex the point which divides the segment Co(x)Ci(x) in the ratio 0:(1 — 6).
Let us now construct the solid sphere E nr of radius r < 1 differing from 1
by so little th at for every x 6 $ for which
p(o, Cx) > r
the segment C(a;)Ci(a;) is in the exterior of the sphere E n\ . Denote the
boundary of the sphere E nr by 5 rn_1. Let us define mappings C'e , 0 < 6 < 1 ,
as follows:
§2] THEOREM ON ESSENTIAL MAPPINGS 183

1. If p(o, Cx) > r, extend the ray oC(x) and consider the segment A
of this ray included between S "-1 and S rn_1 directed to the center o of -the
spheres. Then C'ex is the point of the segment C(x)Ce(x) dividing this
segment in the same ratio in which the point Cx divides the directed seg­
ment A. In particular, if Cx £ S n~\ then C'ex = Cx for all 8.
On the other hand, if Cx 6 5 rn-1 , C'ex = Cex.
2. If Cx is inside the solid sphere E nr , then C'gx = C$x. Since C'o ob­
viously coincides with Co = C, the family of mappings C'e is a deformation
of the mapping C such that C'ex = Cx for all x 6 So and for arbitrary 8.
In other vmrds, C'e is an admissible deformation of the essential mapping
C. Therefore all of the mappings C'e are mappings of $ onto E n. A fortiori,
the entire sphere E n\ is contained in C'«(S) for arbitrary 0 . The mappings
C'e can take only those points of S into E n\ which C maps into E nT ; but
the mappings Ce and C'e coincide in these points. Hence C«(S) ^ E n» .
In particular, Ci(S) 3 JEn\. This proves proposition 2.22 and consequently
the second assertion of Theorem 2.1.
Let us now go on to the first assertion of Theorem 2.1.
Every n-dimensional compaction $ can be mapped essentially onto a closed
n-simplex.
Proof. Let us take c > 0 so small that S cannot be 2e-mapped into any
polyhedron of dimension <n.
Let Co be an c-mapping of S onto an w-dimensional polyhedron II. Let
us define the number 77 in accordance with Lemma 1.15 and consider a
triangulation K n of the polyhedron II of mesh < 7?/2 . Let T n1 , • ■• , T n,
be all the n-simplexes of the triangulation. Set
*,■ = So,• = c - \ ( T \ \ t %)
and define the mapping C0; of S* onto T n{ as the mapping defined by
Co on S,-.
We shall prove that the mapping C0i is essential for at least one i. Sup­
pose the contrary; then for every i there exists an admissible deformation
Cei, 0 < 8 < 1 , taking the mapping C0l- into a mapping Cu of Si into
T ni \ T ni . Cei = Co on all the setsSo,- and hence at all the points contained
in more than one S,-, for arbitrary 8 and for arbitrary i; hence setting
Ce(x) = C0(x), if a; € U.Soi
and
Ce(x) = C$i(x), if x £ Si
for arbitrary 8, 0 < 8 < 1 , we obtain a deformation Ce of the mapping
C = Co of S, where Ci maps S into the (n - 1)-dimensional polyhedron
|| || = || K n || \ (T ni u • • • u T \) . The deformation Ce has the follow'-
ing property: if T 6 K n is the carrier of the point C0(x), then for arbitrary
184 INTRODUCTION TO DIMENSION THEORY [CH. VI

0 and in particular for 0 = 1 , the point Ceix) is contained in T. Hence it

follows th at Ci can map into a given point y £ T £ K n~x only those points
of 4> which are mapped by Co into points of simplexes having the simplex
T as a face; in other words, denoting by r the body of the star O ^ T ,
we have
c~\{x) = c r\{Y ).
Since the diameter of T is less than 77, the set C ^V r), and therefore
C T ^ r), has diameter <e, i.e., Ci is an e-mapping of <I> into the (n — 1 )-
dimensional polyhedron || K n~l || . This contradicts the choice of the num­
ber e.
Hence Coi = C0 is essential on at least one of the sets , say on $1 . Now
assign to each vertex of the simplex T ni this vertex itself; to each vertex of
the complex K n not belonging to the simplex T n\ assign a vertex of the
simplex T n\ . This yields a simplicial mapping C' of the polyhedron II onto
the closed simplex T n1 , such th at C' is the identity mapping on T ni CI n.
Hence it follows easily that the mapping C'C is an essential mapping of
4> onto the closed n-simplex T ni , q.e.d.
§3. The sum theorem. Inductive dimension
§3.1. Statement and proof of the sum theorem.
T h e o r e m 3.1. Let there he given in a compadum 4> a finite number of
closed sets; i f the dimension of each of these sets does not exceed n, then the
dimension of their union does not exceed n.
I t suffices to prove Theorem 3.1 for the case of two sets. Hence let Ai
and A 2 be two closed subsets of 4> of dimension <n. We shall prove that
A = A i u A 2 has dimension <n. .,
Hence it is necessary to prove that for arbitrary e > 0 there exists a
closed e-covering of A of order <n + 1.
Let us take an e-covering
ao = M °i , *• • , -A°S(0)}
of order <n + 1 of the set A 0 = Ai n A 2 and denote by 2a a Lebesgue
number of the covering a0 . Let
= {-4\ , • • • ,
be a closed cr-covering of order < n + -lo f4 x ,X = 1,2. Let
. A/|x1 ) . . . , A4 X„(X)
be all the elements of the covering ax which do not meet the set A 0 . Further
denote by A 1,1 , • • • , AXi,m(x,i) all the elements of ax which intersect
A°i and by A ,,i , • • • , ^4Xi,m(x,i) all the elements of ax which intersect
§3] THE SUM THEOREM. INDUCTIVE DIMENSION 185

A°i (i > 2) but do not intersect a single one of the sets A a\ , • • • , ^40<_i
Let us set
B \ = A Xi, i = 1, 2, ••• ,u(X),
and
B°i = u • • • u A \ ,m ( i , 0 u A 2i'i u • • • u A 2iitn i2 (0 ,

i = 1,2, • • • , s(0) ;
some of the B °t may turn out to be empty.
The sets B °,, B \ , B2t- obviously form a closed 6-covering (1 of the set A.
It remains to be proved th at the order of the covering (3 does not exceed
n + 1 . To this end we note first th at no B1,- intersects any B2k since a point
contained in B1,- and B \ would be contained in Aq . This is impossible since
neither B1,- nor B 2k intersects A 0 by definition.
Further, ao, ai , a2 are s}rstems of sets each of which has order < n + 1;
besides 2a is a Lebesgue number of the system a0 and B°t- Cl B(^40,-, a).
Hence it follows that there cannot exist more than n + 1 sets of each of
the types B°t , B1, , B2, with nonempty intersection (see I, 8.331). There­
fore, if the order of the system /3 exceeds n + 1, there exist n + 2 sets in
/3 of the form
( 3 . 1 1 ) B°ik, B li, , k = 1 , • • • , /x; I = /x + 1 , • • • , n + 2 ,

or of the form
(3.12) B°ik , B 2it , k = 1, • • • , /x ; I = n + 1, • • • , n + 2,
with nonempty intersection. We shall prove that this is impossible. It suf­
fices to consider one of the two cases, say (3.11). Hence suppose there
exists a point p contained in all the sets (3.11). Since p £ B°ik , p is con­
tained in some set A xik,v intersecting A°ik ; in this connection, if X = 2,
p 6 A 2ik,v so that p 6 A 2 and p 6 B \ ii+l , i.e., B1l(i+1 n A 2 ^ 0. Hence
B \ m+1 n Aq 7 ^ 0, which is impossible. Hence
p g n u A \ k,,k n n r i + i B \ , ,
i.e.,
p 6 riL iA 1* n rfi^+iyl .

Here all the sets A \ k,n are distinct, since the set A 1ik,Vk meets A aik and
does not meet any of the sets A ^ j, j < i k ■ Since all the A 1,-(i+r are also
distinct and different from all A 1iktVk , the point p is contained in n + 2
elements of the covering a i , which is impossible,
This proves Theorem 3,1,
186 INTRODUCTION TO DIMENSION THEORY [CH. VI

§3.2. Inductive dimension.

T heorem 3.21. I f every point of a compactum <f> has a neighborhood, of
arbitrarily small diameter, whose boundary (the boundary of an open set T
in a topological space R is the closed set F \ r ) has dimension < n — 1,
then dim <t> < n.
Proof. Given that each point p € $ has for a given arbitrarily small
e > 0 a neighborhood 0 (p ) satisfying the conditions
1. 8[0(p)J < e,
2. dim [0(p) \ 0(p)\ < 7 i — l,
it is required to construct a closed e-covering of <f>of order < n + 1.
Let us construct for every p 6 a neighborhood 0 (p ) satisfying the
conditions 1 and 2 and choose from this collection a finite number of neigh­
borhoods
Oi = 0(pi), • •• , 0 , = 0(p t )
covering the whole space <t>.
By Theorem 3.1 the set
A ' = (Oi \ Oi) u • • • u (0. \ 0 .)
has dimension < n — 1, so that there exists a closed e-covering

a' = { A \ , ••• , A ' U]

of order < n of the set A'.
Let 2<r be a Lebesgue number of the system a' (in <h). Denote by 0 \ a
cr-neighborhood of the set A ' i n <h.
The union of all the O', is a neighborhood O' of the set A '. Let us now
put
A 1 = <5i \ O',

A i = Oi \ (Oi u • • • u 0 { - 1 u O'), i = 1,2, ; s,

and consider the system of sets

« = M i , ••• , A., 0 \ , ••• , O'u).
Since
AiU ■• ■u A ,u 0 ' = <f>,
a is a (closed) covering of the space <J>, and this covering is obviously an
e-covering. Furthermore, since
§3] THE SUM' THEOREM. INDUCTIVE DIMENSION 187

it follows th at for i < k

A i n A k Q Oi n [0* \ (0, u 0')] c 0 , n (0* \ Of) = 0.
Therefore the order of the covering a is at most 1 greater than the order
of the system 0 \ , ■■■ , 0 ' u which in turn does not exceed n. Hence the
order of the covering a is less than or equal to n -f 1, q.e.d.
Theorem 3.21 gives a reason for defining for a topological space R a
certain new topological invariant called the inductive dimension and de­
noted by ind R.
By definition, the only topological space whose inductive dimension is
equal to —1 is the space which does not contain a single point (the empty
set).
Let us suppose that we have already defined every space whose inductive
dimension < n — 1, where n is a nonnegative integer. Then we say that
a topological space R has inductive dimension < n if for every neighborhood
Ox of an arbitrary point x £ R there is a neighborhood 0\X C Ox of x whose
boundary has inductive dimension < n — 1.
If a space R of inductive dimension < n is not at the same time a space
of inductive dimension < n — 1, we say that the inductive dimension of
R is equal to n.
R e m a r k 1. For a metric space we can obviously insert the following
change in this definition:
Assuming that the spaces having inductive dimension < n — 1 have al­
ready been defined, we say that a metric space R has inductive dimension
< n if for every point x 6 R and arbitrary t > 0 there is a neighborhood Ox
of diameter <e whose boundary has inductive dimension < n — 1.
R e m a r k 2. We shall say that a space has inductive dimension < n at a
given point x
ind* R < n

if every neighborhood Ox of x contains a neighborhood 0\X of x whose

boundary has inductive dimension < n — 1.

a )©©O

F ig . 98
188 INTRODUCTION TO DIMENSION THEORY [CH. V I

If a space R has inductive dimension < n at a point x, but not < n — 1,

we shall say that ind* R = n.
E x a m p l e 1. If R is the Cantor perfect set or the set of all irrational num­
bers, then ind* R = 0 at each point x of R.
2. A compactum consisting of a point o and mutually disjoint circles
converging to o (Fig. 98a) has inductive dimension 0 at the point o.
But if each circle is tangent externally to its successor (Fig. 98b) and R
again consists of all these circles and of the point o to which they converge,
then ind„ R = 1.
One of the basic propositions of dimension theory is the following:
T h e o r e m 3.22 (Urysohn). I f $ is a compactum,

(3.22) dim $ = ind <3>.

In this section we shall prove only the inequality
(3.221) dim <3> < ind <3>.

This inequality is easily extended to the case of <3> an arbitrary bicom-

pactum (see Aleksandrov [e]); it is not yet knowm whether (3.22) holds
for arbitrary bicompacta.
The converse inequality dim <3? > ind $ will be proved in the following
section.
The inequality (3.221) easily follows from 3.21.
Indeed, for the empty set (and only for the empty set),

ind 0 = dim 0 = —1.

Hence ind <3? < —1 implies dim <3> < —1.
Let us assume that ind $ < n — 1 implies dim $ < n — 1 and prove
that ind <3? < n implies that dim $ < n. This will also prove inequality
(3.221).
But if ind <3? < n, then every x £ <i>has an arbitrarily small neighborhood
whose boundary has inductive dimension < n — 1. Hence the dimension
of the boundary < n — 1. Hence, by Theorem 3.21, dim<3> < n, q.e.d.
C o r o l l a r y to T h e o r e m 3.22. Every compactum of dimension n contains
a compactum of dimension r for every r < n.
R e m a r k . It is not yet known whether every compactum of infinite di­
mension contains a compactum of arbitrarily preassigned finite dimension
(problem of L. A. Tumarkin).

§4. Sequences of subdivisions

§4.1. Irreducible €-coverings. In this article we shall understand by
covering a covering of a given compactum <3?.
§4] SEQUENCES OF SUBDIVISIONS 189

D e f in it io n 4.11. A closed e-covering

(4.11) a = {A1( , A.}
is said to be irreducible if there does not exist any closed e-covering with a
nerve isomorphic to a proper subcomplex of the nerve K a of the covering a.
Irreducible e-coverings possess certain remarkable properties which we
shall now derive.
D e f i n i t i o n 4.12. A closed e-covering a is said to be a special covering
if its elements are the closures of mutually disjoint open sets of the com-
pactum 4>.
4.13. Every irreducible e-covering (4.11) of an n-dimensional compactum
4> has order < n + 1.
Proof. Denote by 2u a Lebesgue number of the irreducible covering
(4.11) , and let 0 t be a cr-neighborhood of the set A,- in 4>. The covering
(4.131) co = {Oi, ••• , 0 S}
is an open e-covering of $ similar to the covering (4.11) (see I, 8.21, 8.331)-
Now denote by
a' = M 'i, ••• , AV}
any closed <r-covering of 4> of order n + 1. Next denote by A " x the union
of all the A \ which meet A i ; by A "i the union of all the A \ which meet
A i but do not meet any of the sets A x , • • • , A t_i . Then A"; C O i. There­
fore the covering a." consisting of all the nonempty sets A " ; is an e-covering
whose nerve is a subcomplex of the nerve of the covering u or, what comes
to the same, of the nerve of the covering a (because the coverings a and co
are similar). Since the e-covering a is irreducible, the nerve of the covering
a" is identical with the nerve of the covering a. Therefore Theorem 4.13
wall follow if we prove that the order of the covering a" does not exceed
n + 1.
Let
X e n } - i A * if ; ii < i 2 < • • • < u .

Since no summand A \ appears as a summand of two different sums A",-,

there exist mutually distinct elements
A'*, C A "i. (1 < 3 < v)
of the covering a such that
x £ .
Hence it follows th at v does not exceed the order of the covering a', i.e.,
does not exceed n + 1, q.e.d.
190 INTRODUCTION TO DIMENSION THEORY [CH. VI

Since the nerve of every closed e-covering obviously contains a sub­

complex which is the nerve of some irreducible e-covering, it follows from
4.13 that:
4.131. The nerve of an arbitrary e-covering of an n-dimensional com-
pactum contains a subcomplex which is the nerve of some e-covering of
the compactum 4> of order < n + 1.
4.14. Every irreducible e-covering a of a compactum 4> is similar to a special
(-covering a .
Proof. Let
(4.1) a = \Ai , • • • , A,}
be an irreducible e-covering of <t>. Let 2a be a Lebesgue number of the
covering a; denote by 0, a cr-neighborhood of the set A t-. Then
d) = {Oj , • • • , Os}
is a closed e-covering similar to the covering a. Hence for an arbitrary
e, 4> has an irreducible e-covering whose elements are the closures of open
sets.
Let us suppose that the initial covering a already has this property, i.e.,
that A, = f i , i = 1, 2, • • • , s, where I \ is open. Let us introduce the fol­
lowing notation:
A \ = A i , A \ = [At \ .4',]*, • • • , A 'k = [Ak \ (A \ u • • • u A 'k- i)]*,
where [A2 \ A\\*, etc., denotes the closure of the set included in the
brackets. Obviously,
1 . A' k C Ak ,
2. A' k =3 A k \ (Ai u ■• • u Ak-i),
3. A ' k = [r* \ (A'i u • • • u A ' ^ ) ] * .
Therefore the sets A'k form a closed t-covering a whose nerve is a sub­
complex of the nerve K a of the covering a ; the nerve of a ', in view of the
irreducibility of a, is identical with K a . Further, denoting by r\- the set
of all interior points of the set .4', , we have, on the basis of property 3,
T'k = r* \ (A\ u • • • u A'k~\), A'i = f \ ,
and for i < k
T\ C 4> \ A 'i = 4> \ Ft C $ \ T'i ,
which proves Theorem 4.14.
§4.2. Subdivisions. Let us consider anew the coverings of a compac­
tum 4>.
4.21. A closed covering /3 is called a subdivision of the closed covering a
§4] SEQUENCES OF SUBDIVISIONS 191

if every element of a is the union of elements of /3 and every element of /3

is contained in exactly one element of the covering a.
4.22. A sequence of closed coverings ai , a2, ■• ■ , am , ■■■ is called a
chain of subdivisions of the compactum 4> if am is an tm-covering of 4>, lim
t m = 0, and for arbitrary m the covering am+i is a subdivision of the cov­
ering am .
4.23. Two sequences of coverings of a compactum 4>
<*i , • • • , a m , • • •
and
y **' > j
are said to be similar if the coverings am and /3„, are similar.
The following theorem is at the basis of all the results of this section:
4.24. Every sequence
(4.24) , • • • , am , ■■• ,
where am is an irreducible tm-covering of a compactum and lim tm = 0,
contains a sequence similar to a chain of subdimsions of <f>.
Proof. Set mi = 1 and assume that mk has already been defined. Let us
denote by mt+1 the first natural number such that 2tmk+l is a Lebesgue
number of the covering amk . Denote the resulting subsequence of the se­
quence (4.24) again by
(4.24) , • • • , otm , • ■•
and the corresponding t by em .
Hence (4.24) is a sequence of irreducible em-coverings of 4> w'here 2em+i
is a Lebesgue number of <*„, . Hence, in particular, 2tm+i < tm . ^Moreover,
denoting by 8m the maximum of the diameters of the elements of the cover­
ing a m , we have (I, Def. 8.33)
(4.241) ( ^ ”=m+i ot) + 8m < 2tm+i + 8m < .
Let
<*i = {A i, • • • , did) » • • • , -4«}-
Let us assume that all the elements of the covering am are indexed in
the form
Am , • • • , A [,(„,)]
where [1], • • • , [s(m)] denote the various combinations of m indices
i(l), ’ • • , i(m).
Let
A tm , • • • , A [i].,([ii)
192 INTRODUCTION TO DIMENSION THEORY [CH. VI

be all the elements of the covering am+\ which meet A m . In general let
A[k}\ , • • • , A [*],([*])
be all the elements of the covering am+i which meet A w , but do not meet
any of the sets A uj , • • • , A[k-\) ; denote by A [k] the union of the elements
of a m + 1 just selected. We shall prove th at A [k] 9^ 0, i.e., th at there are al­
ways elements of the covering am+i which meet A w but do not meet a
single A[h] , h < k. Indeed, the sets A [k] are contained in an em+i-ncigh-
borhood of the corresponding set A w and consequently form an em-cov-
ering of whose nerve is a subcomplex of the nerve K am of the covering
a m , coinciding, in view of the irreducibility of am , with the whole com­
plex K am ; whence it also follows that none of the sets can be empty.
Let us now consider a sequence of sets of the form
(4.242) A {(i) , A l(i),’(2) , ^4i(i)»(2)i(3) , • ■■ , -4»u)i(2) ••• n™) , • ■• •
Since 5(yli(i)...,-(m)) < ei/2 and ^4i(i)...t(»o n ^ the se"
quence of sets (4.242) converges to a single point .T,-(m(2)--•-!(">)••• .
The set of all points , where the indices
i(l), • • • , i(m) are fixed and all the remaining indices h(m + 1),
h(m + 2), ■• • are free to take on all the values accessible to them, is denoted
by *
We shall prove several properties of the sets .
1. I t follows from the definition of these sets that
B nl)-.-i(m) .
2. The sets B iW.. are closed. Indeed, let

(4.243) Xi(l) *. • »(m)A(l , m + l ) A ( l , m + 2 ) • •• j -^{(1) ••• i(m)h(2 «• • , * ’ * ,

•Ei(l) **•i(m)h(ktm-\-l)h(k,m-\-2) ■■* 3 * * *

be a sequence of points of the set .

Since each of the indices h(k, m-f-1), h(k, m-\-2), • • • can assume only a
finite number of values, there exists a subsequence (4.2431) of (4.243) in
which h(k, m + 1) takes on the same value h{m -f- 1), there exists a subse­
quence (4.2432) of (4.2431) in which h(k, m + 2) takes on a constant value
h{m + 2), etc.
The diagonal subsequence of the sequences (4.2431), (4.2432), etc., con­
verges to the point

£ » ( I ) - ■■ i ( m ) / i ( m + l ) / i ( ? ) i + 2 ) • • • € ,

which proves th at B iw ...{(m) is a compactum and is consequently a closed

subset of 4>.
§4] SEQUENCES OF SUBDIVISIONS 193

3. For every given fixed m the sets B ia)...i(m) form a closed covering /3m
of 4>.
Indeed, let x be an arbitrary point of 4>.
Choose sets
(4.244) A j ( l , l ) , -4»(1,2)»(2,2) , • ' ‘ , ‘ '

to satisfy the condition that the point .r be contained in each of these sets
and argue as in the proof of 2. We obtain a subsequence (4.2441) of the
sequence (4.244) in which the first index i{ 1, m) assumes a constant value
t‘(l), next a subsequence (4.2442) of the sequence (4.2441) in which the
second index assumes a constant value 2), etc. As a result we obtain a
sequence of natural numbers
*‘(1), *'(2), • • • , i(m), • • •
and the sequences (4.244), (4,2441), (4.2442), • • • , (4.244?n), • • • of which
each (except the first) is a subsequence of the preceding, and which are
such th at all the. elements in (4.244m) have as their first indices
f(l)i’(2) • • • i{m). The diagonal subsequence, as is easily seen, converges
to the point £ , - ( i ) i ( 2 ) - - - » ( m ) - . . , and since all the elements of this diagonal sub­
sequence [as elements of the sequence (4.244)] contain the point x, it fol­
lows th at x = xt-(i)»(2)...»cm)--- , so that

4. Since is contained in a (^"=m+i€A:)-, i.e., in a 2€m+i-neigh-

borhood of the set £ am , where 2€m+i is a Lebesgue number of
the covering am , /3mis an em-covering of 4>, having (as a consequence of the
irreducibility of am) the same nerve Kpm = K„m as the covering am ■Hence
The covering (3m is an irreducible em-covering similar to the covering am .
Let us denote by IV)...^*,) the set of those points of B im ...^ m) which
are not contained in any element of the covering /3mdifferent from .
The irreducibility of the covering j3m implies that the set r i(i)...i(m) is non-
vacuous (since if r,a )...;(m) were empty, deleting the element B in)...i{m)
from /3mwould yield an €m-covering /3'mwhose nerve is a proper subcomplex
of the complex 7fyJ. Since is the difference between <f> and the
union of all the elements of the covering /3m different from B im ...^m) ,
r m)... ,-(m) is an open set consisting of all the points i f f 1with the following
property: for every point x represented in the form

& • • •/l(7 7 l)/l(7 7 l4 -l) • • • J

h( l ) = i ( l ) } *• • , h(m) = i(?n). Hence it follows at once that:

^4.24o) f\‘(l) ••• *•.h(m+n) f\*(l) ■*•i(m) •

191 INTRODUCTION TO DIMENSION THEORY [CH. VI

Now let x be an arbitrary point of £ l(i ) . . . , ( m ) and let t be an arbitrary

positive number. Let us take n so large that «m+n < f. The point x is con­
tained in some B ,(1)...;(mWm+i)...Mm+n) and consequently is at a distance
< «m+n < f from r,-(i) •.•i,(mWm+i) ••-/i(m+n) • By (4.245) it follows that
p(x, r,•(!)...t(m)) < t. Since e is arbitrary, x € r t-(1)...t-<m) . Hence
B ^i)...nm) = r»(D■••i(m) • Since two different sets r,(i)...,-(m) (for the same m)
are disjoint by definition, it follows that
5. All the coverings (3m are special coverings.
Property 5 implies that no element of the covering /3m+i can be contained
in more than one element of the covering /9m . Since, on the other hand,
B,\i)■■■i(m)i(ni+i) — Bid) ••■’(*») and every Bid).-.i(m) is the union of the
B;d)•••/(m)i(m+i) contained in it, it follows that
G. The covering /3m+i is a subdivision of the covering /3m .
The verification of all these properties of /3m proves Theorem 4.24.
Let us now assume that dim 4> = r; then all the coverings /3mhave order
<r + 1. Assuming that «i is sufficiently small, we can regard the order of
all the pm as equal to r + 1. The elements of the covering (3m will now be
denoted by B m,•.
7. Each of the coverings (Jm has the following property: the intersection
of any p + 1 elements B m0, • • • , B mp of the covering /3m , 0 < p < r, has
dimension <r — p.
Proof. Given t > 0, it is required to find a closed e-covering of order
<r — p + 1 of the compactum B = B m0 n • • • n B m/>. T^et us take the
,
natural number n so large that e„ < e. Denote by B nik k = 1, • • • , s,-,
all the elements of the covering /3„ contained in a gir'en B mi € /3m • Let
us set
E k — B ^ r\ B i n •n B p ' k — 1, *• • , so ,
and prove that the sets E k form the desired e-covering of the set B. The
sets E k are closed and their diameters are <e by definition. We shall prove
that every point x 6 B is contained in at least one E k . Since x £ B m0 , x
is contained in at least one B ’‘0k ■Moreover, x is contained in all of the sets
B mi , i = ] , ■• • , p. Hence x is contained in some E k . Therefore the sets
Ek form a closed e-covering of B. It remains to be proved that this covering
has order < r — p + 1. Let the point x 6 B be contained in E y , • • • , E„ .
It is necessary to prove that v < r — p + 1. Since x £ B m{ for
all i = 1, • • • , p, there is a set B"iti containing the point x for each
f(l < i < p)- Moreover, x € B n0i n • • • n B n$v . Since all the sets B nik{ ,
i = 1, • • • , p, and B n0k , k = 1, • • • , v, are mutually distinct elements of
the covering having order r + l , p + r < r + l , i.e., v < r — p + 1,
q.e.d.
Let us sum up the propositions proved above.
§4 ] SEQUENCES OF SUBDIVISIONS 195

T h e o r e m 4.25. Let 4> be an r-dimcnsional compactum. Then it is possible

to construct a chain of subdivisions

(4-25) /?1, dl , • • • , dm ,

consisting of coverings dm of <f>with the following properties'.

a) Every dm is a special irreducible tm-covering of order r + 1.
b) The intersection of any p + l , 0 < p < r , elements of the covering dm
has dimension < r — p.
We shall deduce from this the following important proposition.
4.2G. Every point of an r-dimensional compactum $ has an arbitrarily
small neighborhood whose boundary has dimension <r — 1.
Proof. Let € > 0 be given. Take an m such that 2cm < (. For a given
point a; f $ define the neighborhood Omx as the difference between 4> and
the union of all the elements of the covering dm which do not contain the
point x:
0 mx = 4> \ UB '\ , x £ B ”\ .

The closure of the neighborhood 0„,x is contained in the union of all the
B mi E dm which contain x. Hence its diameter is < 2 e m < e. The boundary
of the neighborhood Omx is, on the one hand, contained in the union of all
the B mi E dm containing x and, on the other hand, in the union of all
B mh E dm which do not contain x. In other words, the boundary of 0 mx is
contained in the union of all sets of the form B n\- n B mh , where B mi E dm
contains x, but B m>LE dm does not contain x. Since each of the sets Z?"\ n B m;,
has, by virtue of 4.25, dimension <n — 1, it follows from Theorem 3.1
th at the dimension of their union, and hence also the dimension of the
boundary of the neighborhood 0„,x, is < n — 1, q.e.d.
It is easy to infer from Theorem 4.26 that:
4.260. Let A be a closed set of an n-dimcnsional compactum <f>. For every
neighborhood OA of the set A C 4>, there exists a neighborhood OiA of A con­
tained in OA whose boundary has dimension < n — 1 (and is also contained
in OA).
Proof. For every x E A let Ox be a neighborhood of x satisfying the con­
ditions
' Ox Q OA, dim (Ox \ Ox) < n — 1.
Since A is a compactum, we can choose a finite number of these neigh­
borhoods Oi = Oxi , • • • , Os = 0 x s , covering the whole set A. The union
of these neighborhoods is a neighborhood OxA of A, where the boundary
of OiA is contained, as is easily seen, in the union of the boundaries of the
neighborhoods Oi , • • • 0, . Hence the boundary of OxA has dimension
< n — l. Since OiA C OA, the proof is complete.
196 INTRODUCTION TO DIMENSION THEORY [CH. VI

Let us derive the inequality

(4.261) ind <f> < dim <i>
from Theorem 4.26. This will conclude the proof of Theorem 3.22.
If dim 4> = —1 (i.e., dim < —1), = 0, so that ind 4> < —1. Let
us assume as proved that dim <i> < n — 1 implies ind 4> < n — 1; we
shall prove that dim < n implies ind <t> < n; this will also prove (4.261).
If dim <E> < n, then, by Theorem 4.26, for each x (E there exists an
arbitrarily small neighborhood whose boundary has dimension < n — 1.
Consequently, by hypothesis, the inductive dimension is also < n — 1.
But then, by the definition of inductive dimension, ind < n, q.e.d.
Let us derive from Theorem 4.25 yet another important proposition.
Let us say that a mapping C of a set X onto a set Y has order n if the
inverse image C~1(y) consists of no more than n elements of X for every
y £ Y and if C~1(yo) contains exactly n elements of X for at least one
2/0 e Y.
Theorem 4.27. Every compaclum 4> of dimension n without isolated points
can be represented as the image of the Cantor perfect set under a continuous
mapping of order n + 1; conversely, every compaclum which is the image of
the Cantor perfect set under a continuous mapping of order n + 1 has dimen­
sion < n and does not contain isolated points.
We shall prove the second part of the theorem first. Let the compactum
<t> be the image of the Cantor perfect set P under a continuous mapping C
of order n + 1. Let us divide P into two pieces (by a piece of the Cantor
perfect set we mean its intersection with an arbitrary segment of the real
line whose endpoints do not belong to P ; we shall consider only non-empty
pieces) Pi and P 2 of diameter then each of these into two pieces P n
and P 12 of diameter etc.
The images P,(i>...<(m) = CP,m ...ittm) of the pieces P,-(i)•••»(«) (for a
given m) form an «m-covering /3m of 4>, where lim em = 0. If, for a given m,
a point x £ 4> is contained in v distinct elements £,(i)...,(m) of , then v
distinct pi'eces P,(i)...t(m) contain a point of the inverse image C~1(x). Since
C is a mapping of order n + 1, v < n + 1, i.e., the order of each of the
coverings j8m does not exceed n + 1. Consequently
dim <i> < n.
If were to contain an isolated point x, then CT'x would be an open set in
P and since there are no finite (even countable) open sets in P, C could
not be of finite order.
This proves the second half of Theorem 4.27.
Let us prove the first half of Theorem 4.27. Let 4> be an n-dimensional
compactum without isolated points.
§4] SEQUENCES OF SUBDIVISIONS 197

Let us construct a chain of subdivisions

(4-27) Pi , p2, • • • , Pm ,
for $ satisfying all the conditions of Theorem 4.25 and, moreover, the fol­
lowing additional condition: the covering Pi contains at least two elements
and for arbitrary m the diameter of each element of the covering pm+i is
less than one half the minimum of the diameters of the elements of the
covering pm . Since all the pm are special coverings and 4> has no isolated
points, pm has no degenerate elements, so that the diameters of all the
B mni). ..a,,,) € Pm are different from zero and the supplementary condition
can be fulfilled.
This condition implies that each P'",-ci ) . . . , - ( m) £ Pm contains at least two
Sets B i ( l ) • • • i'(m) i ( m + l ) G Pm-\-1 •

Let us now divide the Cantor perfect set P into the same number of
m u tu al^ disjoint pieces P ,(i) of diameter as the number of ele­
ments B ni) in the covering Pi . Since the diameter of P is equal to 1 and
the number of necessary pieces is >2, the requirement th at <
^ is fulfilled.
Suppose that for a given m the Cantor set P has been divided into
mutually disjoint pieces P,(i)...,(m) of diameter < l / 2 m corresponding
(1 —1) to the elements of the covering pm . Let us divide each piece
Pl ( into the same number of pieces of diameter < l / 2 m+1 as the
number of elements in the covering pm+i contained in P l(i)...£(m). Hence
the pieces P,-(i)...,-(m) are constructed for any m. Assign to each point
£ = P i(i) n P i(i),7 2 ) n • • • n P , ( i ) i ( 2) - - - 2 (m) n • • •

of the set P the point

x = C(£) = B ,-(i) n Bi (1)2(2) n • • • n Pf(i)i(2)---t(m) n • • •

of the compactum <t>. It is easy to see th at this assignment is a continuous

mapping C of P onto <t>. If C maps v distinct points & , • • • , £ „ of P onto
the same point of <t>, let m be such th at no two of the points £i , • • • , £„ are
contained in the same set P,-(i)...£(m). Then the point x = C(£i) = • • • =
C(£„) is contained in v elements of the covering pm . Since all the coverings
pm have order < n + \ , v < n + \ , i.e., the order of the mapping C does
not exceed n + 1, q.e.d.
E x e r c i s e . Prove the following theorem:
Every compactum 4> of dimension < n is the image of a closed subset of the
Cantor perfect set under a continuous mapping of order <n + 1; conversely,
all the compacta which are images of closed subsets of the Cantor perfect set
under continuous mappings of order <n + 1, have dimension <n.
In other words, the dimension of a compactum <t> can be defined as the
198 IN T R O D U C T IO N TO D IM E N S IO N THEORY [C H . V I

least number n satisfying the following condition: there exists a continuous

mapping of order n + \ of a closed subset of the 'Cantor perfect set onto the
compactum <J>.
§6. Some applications to topological manifolds and polyhedra
§6.1. The case of topological manifolds (in particular R n and S"). Let
M n be a closed topological n-manifold. We know that dim M n = n, so
that ind M n = n. Therefore, there exists a point x £ M n for which
imU / " = n. But since all the points x 6 M n have homeoniorphic neigh­
borhoods, it follows that for an arbitrary x 6 fl/n, indx.1/n = n. In partic­
ular,
indxS" = n
and
indx.fi!71 = n
at an arbitrary point x 6 <S", x G fi*", respectively. Hence
5.11. Every topological n-manifold has inductive dimension n at each
of its points.
This implies the following theorem:
5.120. No compact um of dimension <n — 2 separates the space R n.
Proof. Let us assume that a compaction <i> of dimension < n — 2 sepa­
rates R n.
Let T be a bounded component of the open set R n \ <i> and p any point
of the domain r . Let E n be a solid sphere with center p and of sufficiently
large radius to contain <1>in its interior. Let en be a solid sphere of radius e,
e > 0 arbitrary, with center at the origin o of R n.
Let us map the sphere E n onto the sphere en by .means of a similitude.
Then the domain T is mapped into a domain y containing the point o and
the compactum <i> is mapped into a compactum <p of dimension <n — 2.
Here y and <pare contained in the interior of the sphere en and the boundary
of the domain y is contained in y>. Consequently the boundary of y has
dimension < n — 2. Since 7 is a neighborhood of the point 0 having diam­
eter <2e, and e is arbitrary, ind„fi!n < n — 1 , which contradicts Theorem
5.11. This contradiction proves Theorem 5.120.
Theorem 5.120 in turn implies
5.12. A closed solid 71-sphere FA is not separated by any compactum of
dimension <n — 2.
Indeed, suppose that the closed solid n-sphere E n is separated by a
compactum of dimension <n — 2. With no restriction on generality,
we may assume that d> does not contain the center 0 of the sphere.
Our assumptions imply that
En = A u u Dt
§5] A P P L IC A T IO N S TO IV tA N IF O L D S AND POLYHEDRA 199

where A, 4>, D are mutually disjoint and A and D are nonempty open
subsets of E n. For definiteness let o £ A. An inversion maps the sphere
S n = E n \ E n onto itself, the open sets A \ o and D into Ai and A ,
respectively, and the closed set <t> into a closed set <t>i ; here
R n — (A u Ai) u (<£ u <hi) u (Z) u A ).
I t is easy to see that
A u Aj , $ uh, D u Di
are mutually disjoint and that A u A i , D u A are open and nonempty
so th at the compactum u 4>i separates the space R n. This is impossible
since dim 4>! = dim <t> < n — 2 and, by Theorem 3A, dim (4> u <t>i) is
also < n — 2. This proves Theorem 5A2.
5A3. No n-diinensional topological manifold M n is separated by a com­
pactum of dimension < n — 2.
Proof. For every point p £ M n let U(p) be a neighborhood of p homeo-
morphic to R n. Let .C be a homeomorphism of R n onto U(p) and let C
map a closed solid sphere E n C R n onto V(p) C U(p), where V(p) = C(En).
Hence we obtain for each point p £ M n a definite neighborhood V(p)
homeomorphic to R n, where V(p) is homeomorphic to a closed solid sphere.
We can choose a countable, and in the case of a closed manifold M n even
a finite, set of these neighborhoods V(p) whose union is M n.
Let these neighborhoods V(p) be
(5.13) A , 7 S, • • • , A , • • • .
Since M n is connected, the system of sets (5.13) is chained by I, Theorem
3.18.
Xow let $ C M n be a compactum of dimension < n — 2. Each of the
sets <f> n Vi is nowhere dense in V { by Theorem 1.25. Consequently, the
system of sets
A \ < I > , V2\ V , ••• , Vs \ v , •••

is also chained. Each of the sets V { \ V is connected by 5.12 so that by I,

Theorem 3.15, the union of the sets A \ < I >, be., the set M n\ V , is connected.
In particular,
5.131. No domain of the space R n is separated by a compactum of dimen­
sion < n — 2.
§5.2. Strong connectedness.
D e f i n i t i o n 5 . 2 1 . A compactum 4> of dimension n is said to be strongly
connected if no closed set of dimension <n — 2 separates 4> (strongly con­
nected n-dimensional compacta were introduced by Urysohn under the
name of Cantorian manifolds).
200 INTRODUCTION TO DIMENSION THEORY [CH. VI

From this definition and Theorem 3.22 it follows immediately that:

5.22. I f 4> is a strongly connected n-dimensional compactum, then
ind*# = n for every point x 6
In view of the results of the preceding article, all closed topological
manifolds such as the closed solid n-sphere and the compacta homeo-
morphic to it are examples of strongly connected compacta. It is left to
the reader to prove that the closure of every domain of a given closed topo­
logical manifold is a strongly connected compactum.
R e m a r k . The following theorem whose proof will not be given in this
book is valid:
T h e o r e m of Hurewicz-Tumarkin. Every n-dimensional compactum con­
tains a strongly connected n-dimensional compactum.
Let us now go on to consider strongly connected polyhedra.
D e f i n i t i o n 5.23. A finite sequence of n-simplexes

T\ , ••• , T \
of a given simplicial complex K is called a chain of simplexes of K (more
precisely: a chain connecting the simplexes T ni and T n, in K), if the
simplexes T,7lt- and T ni+i , i = 1, 2, • • • , s — 1, have a common (n — 1)-
face in K.
D e f i n i t i o n 5.24. A simplicial n-complex K is said to be pure if each of
its simplexes is a face of an ?i-simplex of K . A pure n-complex K is said
to be strongly connected if every two n-simplexes of K can be connected
by a chain of n-simplexes in K.
5.251. I f a triangulation K is a strongly connected n-complex, || K || is a
strongly connected n-dimensional polyhedron.
Proof. Let T ni , • • • , T ns be all the n-simplexes of the complex K. Then
since K is pure,
(5.251) || If || = u • • • u T n, .
Let«£ C || K || be an arbitrary closed set of dimension n — 2. If T n~1 is an
arbitrary (n — l)-simplex of K,
T n_1\ 4 > ^ 0;
it easily follows from this and the strong connectedness of K that the
system of sets
t w *, • • • , r \ \ 4 >
is chained. Then (by I, 3.15)
|| If || \4 > = ( T ni \4 > ) u • • • u ( T \ \4 > )
is connected. This proves the assertion.
§5] APPLICATIONS TO MANIFOLDS AND POLYHEDKA 201

5.252. Every triangulation K of a strongly connected n-dimensional poly­

hedron 4> is a strongly connected complex.
Indeed, from 5.22 it follows first that K is a pure complex. Suppose that
K is not strongly connected. Then there exist two n-simplexes T ’\ £ K,
T n2 £ K which cannot be connected by any chain in K. Denote by Qi the
set of all ?i.-simplexes of K which can be connected by chains with T ’\ ;
let Qz be the set of all the remaining n-simplexes of K \ both sets Qi and
Q2 are nonempty (since T ’\ 6 Q i, T n» £ Qz). The combinatorial closures
Ki = | Qi | and K 2 = \ Q21
are closed subcomplexes of the triangulation K whose intersection Ko has
dimension < n — 2. Since K = Ki u K 2,

II A 11 = II J C . II u || J f , |l
and
II K I! \ || Ko || = ( || K x || \ || Ko || ) u ( || Ko || \ || Ko || ) .

Since || K\ || \ || K 0 1| and || K 2 j| \ || K 0 1| are disjoint nonempty open

subsets of the polyhedron || K || , || K || is separated by the polyhedron
|| Ko || of dimension <n — 2, which contradicts the strong connectedness
of || K || .
Hence
5.25. Strongly connected polyhedra can be defined as the bodies of strongly
connected triangulations; if one triangulation of a polyhedron is strongly
connected, then every triangulation of every polyhedron homeomorphic to $
has the same property.
 Appendix 1
TV-DIMENSIONAL A N A L Y T I C G E O M E T R Y
Preliminary remarks. Certain definitions and theorems of n-dimensional
analytic geometry applied in the hook are collected in this appendix. It is
assumed th at the reader’s knowledge of this subject approximates the con­
tents of the book of Schreier and Sperner [S-S]. Many of the proofs are
accordingly left to the reader.
We shall use vector notation for calculations involving points of n-space.
If a and b are points of the Euclidean n-space R n with coordinates c*i,
a2, ■■• , an and (3i , @2 , ■■• , (3n , respectively, then a + b is the point
with coordinates
+ /3i , «2 + P2 , • • • , a„ + /3n .
Furthermore, Xa is the point with coordinates
Xai , Xa2 , • • • , Xa„ .
We shall denote the number by | a \ . Then the distance be­
tween the points a and b is
p(o, b) = | a - b | .
The triangle axiom may now be written as
\c — a \ < | c — 6| + | 6 — a | ,
or, replacing c — b by a and b — a by b,
[a + 6 | < | a | + | 6| .
Moreover,
| \a | = | X | • | a | .
This relation combined with the triangle inequality yields
I 2Z X,a; | < H | \ i | • | a,-1 .
§1. The space R n and planes in R n
§1.1. In this book the Euclidean ?t-space is always denoted by R n. An
r-dimensional subspace, 0 < r < n, of R n is denoted by R T or sometimes
by X T or Y T and is called an r-dimensional (hyper)plane of R n or simply
an r-plane. R n is itself the unique h-plane of R n.
Every (n — l)-plane R n~l of R n divides R n into two open half-spaces
H n1 and H n2: if the plane R n 1 is defined by the equation
+ • • • + o,nxn = 0,
202
§1] THE SPACE R n AND PLANES IN R 203

the open half-spaces IV \ and H nn are defined by the inequalities

ai*i + • • • + anx n > 0 and aiX\ + • • • + anx n < 0. Two points of R n are
said to lie on one side of the plane R n 1 if they are in the same half-space
determined by R "-1 and on different sides of R ',-1 if one point is in one half­
space and the other in the second half-space.
The closure of an open half-space //", , i = 1, 2, i.e.,
R \ : = H niU R n~\ i = 1,2
is called a closed half-space.
§1.2. Linear independence of points. Barycentric coordinates. The points
a0 , fli , • • • , ar of R n are said to be linearly independent if they are not
contained in any plane of dimension < r.
Accordingly, every n + 2 or more points of R n are linearly dependent.
The points a,-, i = 0, 1, • • • , r, with coordinates .r(1\ , .t(2\ , • • • , x (n)
are linearly independent if, and only if, the matrix

/ xr (1)o • • • x,
-r(n)o r1 \\

.0 )
X(n)
has rank r. This may be reformulated as follows: if
XoOo -T ■• • + Xrar — 0
and
Xo T Xi -T • • • T Xr = 0
then
Xo — X i — • • • — Xr — 0.

If ao , ai , • • • , ar are r + 1 linearly independent points of R n, there is

exactly one r-plane, the plane R(ao, • • • , ar) spanned b.y the points
a0 , • • • , ar , passing through these points. It is contained in every plane
R s containing the points a0, ai , • ■• , ar .
Barycentric coordinates. The plane R(a0, • • • , ar) consists of those
points of R n which can be represented in the form
(1.2) a = yoQo T Ml°l T • • - T MrOr ,
with the supplementary condition that
(1.21) mo T Mi T • • • + Mr — 1;
on the other hand, the numbers mo , • • • , Mr are uniquely determined by
the point a and the relations (1.2), (1.21); they are called the barycentric
204 JV-DIM ENS10NAL ANALYTIC GEOMETKY [a p p . 1

coordinates of the point a in the coordinate system a0, • • • , aT. This

terminology, introduced by F. A. Mobius, has the following basis: let us
call a point of 72" to which a real number «r, the mass or weight of the
point, is assigned a mass point..If mass points a,- with weights a are given
and m = o-i/(<T0 + • • • + o>), then the point (1.2) is, by definition, the
center of gravity or centroid of this mass distribution.
The numbers crt- may or may not be positive. They are arbitrary real
numbers such that

(To + • • • + 0> ^ 0.

§1.3. Theorems on intersections and linear closures. By the intersec­

tion of two planes R r and R s of R n we understand, as usual, the set of
points common to both planes; the plane spanned by 72r and R *, or the
linear closure of the two planes, is the plane of least dimension which
contains both R r and 72*; it is obviously unique.
The following theorems are easily proved:
1.31. The intersection of two planes R T and R s is either empty or is a plane
Rd, d > r + s - n.
1.32. The dimension h of the linear closure of two planes R T and R s satis­
fies the inequality h < r + s + 1.
1.33. I f the intersection of two planes R r and 72s is nonempty,
then d + h = r -f- s.
Since h < n, 1.31 is contained in 1.33.
§1.4. General position.
D e f i n i t i o n . A set of points of 72" is said to be in general position if every
r points, r < n + 1, of the set are linearly independent.
The theorems of 1.3 imply
1.41. L et the points ao, • • • , ar , b0, • • • , bs be in general position and
let r < n, s < n. Then the following assertions hold for the intersection and
linear closure of the planes R T = R(a0, • • • , ar) and 72s = R(b0, • • • , bs):
I f r + s < n, the linear closure has dimension r + s + 1 and the inter­
section is empty.
I f r -\- s > n, the linear closure has dimension n (hence it coincides with
the space 72") ami the intersection is either empty or has dimension r + s — n.
Every finite set of points can be brought into general position by means
of an arbitrarily small displacement, i.e.,
1 .4 2 . Let { a 0 , • • • , a,} be an arbitrary finite set of points in 72" and e a
positive number. Then there exists a set of points {a\ , • • • , a 's} in general
position such thati

p(a i , af) c i = 0, 1, • • • , s.
§1] THE SPACE R n AND PLANES IN R n 205

Proof. Set a'0 = a0and suppose that points a\- £ £(a,-, e), i < m < s — 1,
have already been found to satisfy the condition that a'o, • • • , a! m are in
general position. Then choose a point a'm+i £ S(am+i , e) not contained in
any of the planes R'{a! ^ , • ■• , a 'lr), where io , • • • , iT do not exceed rn
and r < n. If m = s — 1 , the theorem is proved.
If the points ao, • • ■ , ar are linearly independent, i.e., are not contained
in any plane of dimension <r, the same is true for any points a'0, • • • , a!T
which are at a sufficiently small distance from a0, ■• • , aT, respectively.
From this simple remark it follows that:
1.43. I f the 'points do, • ■• , a* are in general position, there exists a S > 0
such that any points a'0 , ■■• , a!k for which p(ai, a'I) < 5, i = 0 , 1 , • • • , k,
are also in general position.

§1.5. Affine mappings. Concerning affine mappings of R n into itself see

e.g., Schreier and Sperner [S-S, §13]. Of the properties of affine mappings
we recall first the invariance of the centroid (easily established by simple
calculations): if C is an affine mapping of R n into itself, and if a is the
centroid of points ak , k = 0 , 1 , • ■• , r, with weights mk , then C(a) is the
centroid of the mass points C(ak) with the same weights mk .
R e m a r k . I t is assumed here that if C(aA) = C ( ak) = b, then the weight
mk + ink is located at the point b.
The invariance of the centroid implies that:
1.51. I f cio, • • • , an is a set of n + 1 linearly independent points of R n
and bo, • • • , bn is any set of n + 1 points in R n, there exists a unique affine
mapping C of R n onto a linear subspace of R n which takes the points a0, ■• • ,
an , into b0 , • • • , bn , respectively. The mapping C is obtained by assigning
to an arbitrary point a with barycentric coordinates p0, ■■• , (with
respect to ao, • ■• , an) the centroid of the weights po, • ■■ , yn located at
the points b0, ■• • , bn .
An affine mapping is said to be non-singular if it is (1 —1). Otherwise
it is singular. An affine mapping is onto (R n) if, and only if, it is non-singu­
lar. Non-singular mappings may be characterized by the fact that they
map every linearly independent system of n + 1 points of R n onto a
linearly independent system, also consisting of n + 1 points.
This condition is expressed analytically as:
An affine mapping C defined by the relations

(1.51) x'i = uWiX i + • • • + u ^ iX n + U i, i = 1, • • • , n,

(where x { and x '{ are the coordinates of the points of the inverse images
and images, respectively) is 11011-singular if, and only if, the determinant
of the mapping is different from zero:
20G N-DIM ENSIONAL ANALYTIC GEOMETRY [a p p . 1

« (W1 u i

det C = 5* 0 .

U
0) u (n)
An affine mapping is said to positive if its determinant is positive, negative
if its determinant is negative. This definition is legitimate, since the sign
of the determinant is independent of the choice of the coordinate system.
Let Ce , 0 < 8 < 1, be a family of non-singular affine mappings of R n
onto itself depending on the parameter 8 (I, 7.4). All the mappings C» , in
particular Co and C\ , have the same sign since the determinant of C« is
continuous in 8 and hence, never assuming the value zero, cannot change
in sign.
The identity mapping Co of R n onto itself obviously has determinant
+ 1 in every system of coordinates; hence every affine mapping which
can be transformed into the identity mapping by means of a continuous
deformation C# (where all the C« are non-singular affine mappings of R "
onto itself) is positive. As an example of a negative affine mapping of R n
onto itself we shall consider a symmetry relative to an (n — 1 )- plane
R n~' c R \
If the coordinate system is chosen so that the plane R n~l is the co­
ordinate plane x„ = 0 , the mapping can be defined by the equations
x'i = X i, i = 1, • • • , (n — 1 ),
n 2 -n j

whence it is at once clear that the determinant of the mapping in the

chosen coordinate system is —1 . Hence every affine mapping which can
be continuously deformed into a symmetry relative to an (n — l)-plane
by a set of non-singular affine mappings of R n onto itself is negative. We
shall make use of this example in the proof of the following theorem which
is required in Chapter VII:
1.52. Let Ci , • • • , en be a system of n linearly independent points of an
(n — 1Yplane R n 1 C R n, and let e'o and e"o be two points in the comple­
ment of R n~l. Then an affine mapping Co which takes the linearly independent
points e'o, e-i , ■■• , en into e"0, ci , • • • , en , respectively, is positive if e'0
and e"o lie on one side of R n X and negative in the contrary case.
Theorem 1.52 is a special case of the following proposition (which is
utilized in Chapter X):
1.521. Let C be an affine mapping of R n onto itself which maps some (n — 1 )-
plane R n~l onto itself. The affine mapping of R n~l defined by C is denoted
by C>.
§1] THE SPACE R n AND PLANES IN R 207

Let e be a number equal to + 1 if C maps each of the two half-spaces into

which Rn~l divides Rn onto itself and equal to —1 if C maps each of these
half-spaces onto the other. Then the signs of C and C' satisfy the relation
sgn C sg n C' = e
(in particular, if sgn Cr = + 1 , sgn C = t, whence 1.52 also follows).
Proof of 1.521. The theorem is obvious for n = 1 ; let n > 2.
Let .Ti, • • • , xn be a coordinate system in Rn such that R "-1 is the co­
ordinate plane Xi = 0. In this coordinate system we may write C in the form
xi ) i aikx k , x = 1, 2 , • • • , n.
Since R "-1 is transformed into itself, x \ = 0 for xv = 0 and arbitrary
x 2, • • • , x„ , i.e., + • • • + ou xn = 0 for arbitrary x 2, • • • , x„ so
that
a 12 = • • • = O i„ = 0.

In other words, the mapping C has the form

x \ = aux i ,
x'2 = o2i2 i + • • ■ + a2„x„ ,

xn a„iXi ' ** d- onnxn ,

whence it follows that
det C = an det C'.
B ut On has the same sign as e so that sgn C = t sgn C’, q.e.d.
An affine mapping which transposes any two of n + 1 linearly inde­
pendent points eo, • • • , en , i.e., which takes the points Co, • • • , , ■• • ,
ek , • • • , en into e0 , • • • , ek , • • • , et- , • • • , en , respectively, is a negative
mapping. To show this it suffices to take an affine system of coordinates
in Rn whose unit vectors are eo^i , • • • , eoe„ with origin eo : the determinant
of the mapping in this coordinate system is obviously —1 .
I t follows from this remark that:
1.53. If the points eo, • • • , e„ are linearly independent in Rn, the sign of
the affine mapping realizing a given permutation

(:° ::: )
of these points is the same as the sign of the permutation.
§1.6. Definition of an affine mapping of Rn by affine mappings of two
planes X p and Y9, p + q = n.
 208 JV-DIMENSIONAL ANALYTIC GEOMETRY [a p p . 1

Let p + q = n, and let X p, Y 9 be two planes in R n intersecting in a

single point o and such that the linear closure of the planes coincides with
all of R n.
Let Ci and C2 be affine mappings of X p and Y 9, respectively, onto them­
selves such that Ci{o) = C2(o) = o. In these conditions, there exists a
unique affine mapping C of R n onto itself coinciding on X p and Y 9 with
Ci and C2, respectively. The mapping C is called the induced mapping of
Ci and C2 or simply the extension of these mappings over R n.
Proof. Choose a system of coordinates with origin 0 in R n such that the
first p unit vectors of the coordinate system are in X p and the rest in Y 9.
If Ci , C2, in the coordinates Xi , • • • , xp ; xp+i , • • • , xn , are given by the
equations

x 1 anXi -b *■* ~b aipXp x p+i Gp+i,p+i*rp-t-i —

b *** ~b ap+itnXn

xp apixi “b *** “b xn tP+iXp+i ~b *■* “b o.nnxn

respectively, then the desired mapping C in the coordinates *i , • • • ,
xp , ■• • , xn is written as
x 1 • ■auXi ~b *** I aipXp

xp apix 1 1 *** I appxp

x 'p+1 = O p + l.p + l^ p + l + • • • + U p + l , nX n

X n ^ n .p + l^ p + l ~b ’ " • ~b f l n A •

C orollary.

(1.61) det C = det Ci-det C2 .

§2. Convex sets
§2.1. Definition of convex sets. The straight line defined by two points
a, b is the set of all points of the form Xa -f yb, where X + u = 1 (see 1.2).
The subset of this straight line defined by the conditions X > 0, 11 > 0;
X > 0, ju > 0, respectively, is called the closed segment [ah] or the open
segment (ab), respectively.
D e f i n i t i o n 2 . 1 1 . A set M of points of the space is said to be convex if
it contains with every two of its points a and b the w'hole segment [0 6 ].
The simplest examples of convex sets are: the whole space R n and its
planes (of arbitrary dimension), half-spaces, segments, sets consisting of
one point, the empty set.
D e f i n i t i o n 2 . 1 2 . A closed bounded convex subset of R n containing an in­
terior point (relative to R n) is called an n-dimensional convex body.
§2) CONVEX SETS 209

§2.2. Simplest properties of convex sets.

2.21. Every convex set is connected.
In fact, two arbitrary points p and q of a convex set M are contained
in a connected snbset of M (the segment [pg]), i.c., M is connected. It fol­
lows immediately from Dcf. 2.11 that:
2.22. The intersection of an arbitrary {finite or infinite) set of convex sets
is convex.
Furthermore,
2.23. I f M is convex and 8 > 0, then S(M , 8) is convex.
Proof. Let bi , b» be points of the neighborhood S(M , 8) and let b be a
point of the segment [6ifr>]; it is required to show that b is contained in
S(M , 8).
There exist points ai , a2 in M such that | bi — eq | < 8, | b2 — a2 | < 6 .
Then since b is a point of the segment [&i&2],
b = A&i T- ju&2 , A > 0, A "f ju = 1.
If the point Acq + na2 of the segment [aia2] is denoted by a, then
b — a = A(bi — ai) + ju(b2 — a2),
whence
| b — a | < A | bi — fli | -f- fx | b2 — n2 | < (A -f- n)8 — 8
Since a £ M (in virtue of the convexity of M ), the last inequality implies
th at b £ S(M, 5), q.e.d.
C o r o l l a r y . An open solid sphere S(p, 8) is a convex set.
The intersection of all the sets S(M, 8), 8 > 0 arbitrary, is the closure
of M. Therefore, 2.22 and 2.23 imply
2.24. The closure of a convex set is convex.
C o r o l l a r y . A c l o s e d s o li d s p h e r e is a c o n v e x b o d y .

§2.3. Interior and boundary points of a convex set.

2.31. Let M be a convex set, a an arbitrary point of M, b an interior point
of M (relative to tin 3 M ), c a point of the segment [ab] different from a.
Then c is an interior point of M .
Proof. There exists a 5 > 0 such that S(b, 8) C M- furthermore,
c = Aa + y.b, where A + n = 1, A > 0, and, since c 9^ a, n > 0.
We shall prove that S(c, nS) C M.
If d £ S(c, n8), i.e., if | d — c \ = \ d — \a — pb \ < n8, then
| (1/p) d - (A/p)a - 6 | < 5 ;
consequently the point
d' = (1/p) d - (A/p)a
210 JV-DIMENSION'AL ANALYTIC GEOMETRY [a p p . 1

is contained in S(b, 8) C M. Since

d = \a fid',
d is on the segment [ad'} C M , q.e.d*
C o r o l l a r y . The set of interior points of a convex set is convex.
If a is a boundary point and b an interior point of M , then, in particular,
by 2.31, [af>] cannot contain any boundary point different from a. Hence
2.32. Every half-line issuing from an interior point of a bounded convex
set M contains one and only one boundary point of M.
2.32 in turn implies
2.320. I f a convex set M a R n has an interior point,.the set of all interior
points of M is dense in M (and open in R n).
§2.4. The dimension number of a convex set Q is, by definition, the
maximum number r such that Q contains r + 1 linearly independent
points a0 , • • • , ar . If a0, • • • , ar are linearly independent in Q, Q is also
contained in the plane R T(a0, • • • , ar) (since if Q were to contain a point a
not on this plane, Q would have r + 2 linearly independent points a,
a0, • • • , ar). The set of all points x 6 R r(a0, • • • , aT) whose barycentric
coordinates with respect to a0, • • • , aT are positive is an open subset of
R r(a0 , • • • , aT) contained in Q. Hence
2.41. Every convex set of dimension number r is contained in a uniquely
defined r-plane, the carrying plane of Q, and contains interior points with
respect to this plane.
Unless otherwise specified, interior points of a convex set Q will always
refer to interior points of Q relative to its carrying plane. It follows from
2.31 that the interior points of a convex set Q form anr-dimensional convex
set dense in Q.
2.42. All n-di?nensional convex bodies are homeomorphic.
Proof. We shall prove that an n-dimensional convex body Qn <Z I?" is
homeomorphic to a closed solid n-sphere E n. Let us take as the center of
E n any interior point o of the convex set Qn. Let x be an arbitrary point
of Qn different from o. Let us denote by p(x) the boundary point of Q„
lying on the ray ox, by q(x) the point in which the ray ox intersects the
boundary of E n, and by C(x) the point which divides the segment [oj(x)]
in the same ratio as the point x divides the segment [op(x)]. The result is
a mapping C of the convex body Qn onto E n; it is left to the reader to
prove that this mapping is (1 —1) and bicontinuous (see, e.g., Alexandroff-
Hopf [A-H; 601-602]).

§3. Closed convex hull. Simplexes. Convex polyhedral domains

§3.1. The closed convex hull of a set M C R n is defined as the intersec­
tion of all the convex sets containing M . In consequence of 2.22, it is a
§3] CLOSED CONVEX HULL. SIMPLEXES. CONVEX POLYHEDRAL DOMAINS 211

convex set. I t is obvious th at M coincides with its closed convex hull if,
and only if, M is convex.
3.11. T h e d i a m e t e r (the least upper bound of the distances between two
arbitrary points of a set) o f t h e c l o s e d c o n v e x h i d l M * o f a se t M i s e q u a l
to the .d ia m eter o f M .
Proof. It suffices to show that if 8 is such that | x — y \ < 8 for any pair
of points x , y of M , then | a — b | < 8 for any pair of points a , b of M * .
Let a, b be any two fixed points of M * and let x £ M . Since | x — y \ < 8
for all y £ M , M C S ( x , 8). In consequence of the convexity of S ( x , 8)
and the definition of closed convex hull, M * C S ( x , 8) . Hence, in particu­
lar, a £ S ( x , 8). Therefore x £ S ( a , 8) . This is true for any x £ M , so that
M C S ( a , 8) . Hence M * C S ( a , 8) and, in particular, b £ S ( a , 8). There­
fore, | a - b | < 8 , q.e.d.
§3.2. Closed convex hulls of finite sets. Definition of a simplex and of a
closed simplex.
3.21. The closed convex hull of a finite set of points a0, • • • , a s of R n
consists of all the points o £ R n of the form

(3.21) a = "T • ■■ -f- ,

where y Q , ■ • • , y , are arbitrary nonnegative real numbers whose sum is 1.

In other words, the closed convex hull of a finite set a0, • • • , a, consists
of the centroids of all possible nonnegative weights located at the points
a0, • • • , a , respectively. In particular, if the points o0, • • • , o5 are linearly
independent (which implies that s < n ), the closed convex hull of the set
a0 , • • • , a s consists of all the points of the plane R s( a 0, • • • , a«) whose
barycentric coordinates with respect to do, • • • , as are nonnegative.
Proof o f 3.21. A detailed proof can be found in Alexandroff-IIopf [A-H ;602-
604]. We shall give a sketch of this proof here. Denote the set of all points
(3.21) by (d0 , • • • , d„) and the closed convex hull of the set [a0, • •• , a,}
by (do, • • • , as}*.
It is required to prove that ]a0, • • • , as}* = (a0 , • • • , a , ) . To this end,
we state the following three lemmas:
3.211. The set (n0, • • • , a.) is convex.
3.212. If 0 < r < s, and a is an arbitrary point of (a0, • • • , n,), there
exist points a ' £ (a0, • • • , af) and a" £ (ar+1 , • • • , as) such that a is on
the segment [a'n"].
3.213. If the points n0, ■• • , a, are contained in a convex set Q, then
( d o , ■ ■■ , o s) C Q.
Lemma 3.211 can be verified by a routine calculation.
Lemma 3.212 is proved as follows. If a = yo^o + ■• • + >and if

Xr = S o y< ^ 0, — S r + l m« ^
212 JV -D IM E N S IO N A L A N A L Y T IC GEOM ETRY [a p p . 1

put
a' = X o G g A 'K

a" = Z r+ i (Mi/X'Oa,-.

Then a = X'a' + X"a". But if, e.g., X' = 0, then a £ (or+1 , • • • , as) and
the lemma is trivial. Lemma 3.213 is proved by induction on the number s.
If a £ (ao, • • • , a,) and a0 6 Q, • • • , as £ Q, by Lemma 3.212 there is a
point a" £ ( a i, • • • , as) such that a £ [a0a"]. By the inductive hypothesis
(«i , • ■■, a,) C Q; since a0 6 Q and Q is convex, a £ [a0a"] C Q, which
was to be proved. Lemmas 3.211 and 3.213 immediately imply 3.21.
3.22. Let o0, • • • , ar be linearly independent points of R n (so that
r < n). The set of points of the plane R T(a0, • • • , ar) whose barycentric
coordinates with respect to ao, • • • , ar are positive is a convex open set
in R T(a0, • • • , ar) called an r-dimensional simplex or simply an r-simplex
with vertices a0, • • • , or and denoted by (a0 ■■• aT). The closure of the
simplex (ao • • • ar), which is obviously synonymous with the closed convex
hull of the set of points ao, • • • , ar , is called the closed simplex with ver­
tices ao, • • • , ar and is denoted by [a0 • • • ar].
If a l0 , • ■• , aip , 0 < p < r, are vertices of a simplex (a0 • • • ar), the
p-simplex (aio • • • aip) is called a p-face of (a0 • • • ar). In particular, the
r-simplex (ao ■■■ar) is its own unique r-face. The remaining faces will be
called proper faces of the simplex.
Two faces (a10 • ■• a;p) and (iaja • • • a/s) of a simplex (a0 • • • ar) are said
to be opposite faces if every vertex of (a0 • • • ar) is a vertex of exactly one
of the two faces. Obviously, if a p-simplex and a g-simplex are opposite
faces of an r-simplex, then p + q = r — 1. ^-
It is easily proved that
3.23. If T v = (aio • • • a lp) and T 9 = (a/0 • ■■ajQ) are opposite faces of a
simplex T T = (a0 • • • aT), then every point of T r lies on exactly one straight
line segment joining some point of T p with some point of T9.
Further, 3.11 implies
3.24. The diameter of (ao • • • ar) is equal to the maximum of the numbers
p{a,i, af).

§3.3. Convex polyhedral domains. Since the half-spaces of a given R n

are convex sets, the intersection of any number of half-spaces (open or
closed) is convex.
3.31. A bounded nonempty subset of R n which is the intersection of a
finite number of open (closed) half-spaces of R n is called a convex poly­
hedral domain (closed polyhedral domain).
The closure of a convex polyhedral domain is a closed polyhedral do-
§4] C E N T R O ID OF A S IM P L E X 213

main of the same dimension number; conversely, the set of interior points
of a closed convex polyhedral domain is a convex polyhedral domain.
Let Qn be an n-dimensional convex polyhedral domain in l i n. The inter­
section of every (n - l)-plane R n~l C R n with Qn and Qn is convex. A
plane R n 1 is called a plane of support of the polyhedral domain Qn if
Qn n R n_1 ^ 0 and R n~l n Qn = 0.
The intersection of every supporting plane R n~l with the boundary
Qn \ Qn of the polyhedral domain Qn coincides with the set R n~l n Qn
and is therefore a closed convex polyhedral domain QT, r < n — 1; if
r = n — 1, Qr is called an (n — 1)-face of the polyhedral domain Qn.
Hence every (n — 1) -dimensional polyhedral domain which is the interior of
some closed polyhedral domain obtained as the intersection of Qn with a sup­
porting plane of Qn is called an (n — I)-face of the convex polyhedral do­
main Qn.
Furthermore, the (n — 2)-faces of the (n — l)-faces of Qn are called
the (n — 2)-faces of Qn, etc. The 0-faces of Qn are points and are called
vertices of Qn.
3.32. Every two faces of a convex polyhedral domain are disjoint; the union
of all the r-faces of Qn, 0 < r < n — 1, is the boundary Qn \ Qn of Qn.
R e m a r k . A polyhedral domain Qn is its own unique n-face.
The proof of Theorem 3.32, which offers no serious difficulties but is
nevertheless quite long and tiresome (see Alexandroff-Hopf [A-H; 609-614]),
is left to the reader. We note finally that the least possible number of verti­
ces of an n-dimensional convex polyhedral domain is n + 1. Convex poly­
hedral domains which have exactly n + 1 vertices are n-simplexes.
§4. Centroid of a simplex
D e f in it io n 4.1. The centroid of a finite set of points cq , • • • , a* of Rn
is the point
b = (l//c)(ai + • • • + a*),
i.e., the centroid of the system of equal weights located at these points.
The centroid of a simplex (and in general of a convex polyhedral domain)
is, by definition, the centroid of its set of vertices in the above sense.
§4.2. Let M C R n be a finite set of diameter d consisting of k points,
let M i be a nonempty subset of M , and denote by b and bi the centroids
of M and M i , respectively. Then
(4.20) p(b, bi) < [(fc - 1)/k] d.
Indeed, if
M = {ai , • • • , ak), M i = {ai , • • • , a/ei), (1 < ki < k ),
214 ^-D IM EN SIO N AL ANALYTIC GEOMETRY [a p p . 1

then, by the definition of centroid,

b = (l//c) 23<~i a,-, bi = (l/k i) 23yii ai >
so that

h - b = (l/fc i)(E ‘iia y - fci6) = (1/fci) (ay - 6),

a,j — b = (1 /k)(ka j — 23<=ia») = (1/fc) 23<=i (ay — a,),

and
(4 .2 1 ) bi — b = (1/kki) 23* -1 I 3 y = i (ay - a ,).

The fefci summands a,- — a,- contain ki summands such that j = i < A:i
and these summands are equal to zero. This means that the number of
summands different from zero in the sum Z L i E ? i i (a, — a,) is equal
to kki — ki = ki(k — 1). Therefore, (4 .2 1 ) and the assumption
that | ay — a,- | < d imply that
I bi — b | < (l/kk\) k\(k — 1) d = [(& — 1)/^] d, q.e.d.
Let T n be an n-simplex of diameter <d, let T' be a face
C o r o l l a r y 4 .2 .
of arbitrary dimension of T n, and let T" be a face of T ' . Denoting the
centroids of T ', T" by b', b", respectively, we have
P(b', b") < [n/(n + 1)] d.

Indeed, if the dimension of T' is r, in consequence of the above, p(b', b") <
[r/(r + 1)] d. But r/(r + 1) < n /(n + 1), whence the assertion also
follows.
§6. Central projection
Let o be a point of R n, called a center of 'projection. Let M C R n \ o.
We shall call the point set oM([oM]) of R n which is the union of all open
segments (ox) (closed segments [ox]), x £ M, the open (closed) projection
of the set M , or cone with base M and vertex o.
Let N C R n \ o be any set intersecting every half-line joining o with
any point x £ M in a single unique point tt( x ) . The assignment to each
x 6 M of the point tt( x ) £ N yields a mapping r of the set M into the
set N.
The mapping ir, as well as the image tt(M) of M under this mapping,
is called the projection of M into N from o. When there can be no mis­
understanding, we shall use the term projection to denote a projection
as a mapping.
 CENTRAL PROJECTION 215

The proof of the following property of projections is left to the reader:

5.1. Suppose that the intersection of a set M with every half-line issuing
from o consists of a single point and let M i , il/2 be two subsets of M . Then
the projection of the intersection of M x and M i coincides with the intersection
of their projections. B y projection is meant either open or closed projection.
Kazan, December 10, 1941.
 BIBLIOGRAPHY
BOOKS
A l e x a n d r o f f , P. S.
and I I o p f , II.
[A-H] Topologie, I, Berlin, Springer, 1935 (Die Grundlehren der mathematischen
Wissenschaftcn, Bd. 45); in reprint, Ann Arbor, Edwards, 1945.
B 6 c h e r , M.
[B] Introduction to Higher Algebra, New York, Macmillan, 1933.
H a u s d o r f f , F.
[H] Grundzuge der Mengenlehre, Leipzig, von Weit, 1914; in reprint, New York,
Chelsea, 1949.
[H*] Mengenlehre, Berlin, de Gruyter, 1935; in reprint, New York, Dover, 1944.
H u r e w i c z , W. and W a l l m a n , II.
[H-W] Dimension Theory, Princeton University Press, 1941 (Princeton M athe­
matical Series, no. 4).
L e f s c h e t z , S.
[L] Algebraic Topology, New York, 1942 (American Mathematical Society Collo­
quium Publications, vol. 27).
M e n g e r , K.
[M] Dimensionstheorie, Berlin, Teubner, 1928.
N e w m a n , M. H. A.
[N] Elements of the Topology of Plane Sets of Points, Cambridge University Press,
1951.
P o n t r j a g i n , L.
[P] Topological Groups, Princeton University Press, 1939 (Princeton Mathematical
Series, no. 2).
S c H R E I E R , 0 . AND S P E R N E R , E.
[S-S] Introduction to Modern Algebra and Matrix Theory, New York, Chelsea, 1955.
S e i f e r t , H. and T h r e l f a l l , W.
[S-T] Lehrbuch der Topologie, Leipzig, Teubner, 1934; in reprint, New York, Chel­
sea, 1947.
S i e r p i n s k i , W.
[Si] General Topology, University of Toronto Press, 1952.
W h y b u r n , G. T.
[W] Analytic Topology, New York, 1942 (American Mathematical Society Col­
loquium Publications, vol. 28).
W i l d e r , R. L.
[Wi] Topology of Manifolds, New York, 1949 (American Mathematical Society Col­
loquium Publications, vol. 32).

PAPERS
A l e x a n d e r , J. W.
[a] Normal forms for one- and two-sided surfaces, Annals of Mathematics (2), vol.
16 (1915), 158-161.
A l e x a n d r o f f ( or A l e k s a n d r o v ), P. S.
[a] Stetige Abbildungen kompakter Raume, Mathematische Annalen, vol. 96 (1920),
555-573.
[b] Dimensionstheorie, Mathematische Annalen, vol. 106 (1932), 161-238.
[c] Diskrete Raume, Recucil Math(5matique Moscou (MatematiSeskil Sbornik)
(N. S.) 2, (1937), 501-518.
216
 BIBLIOGRAPHY 217

[d] (with Hopf, II. and Pontryagin, L.) tfber den Brouwerschcn Dimensionsbegriff,
Compositio M athemalica, vol. 4 (1937), 239-255.
[c] The sum theorem in the dimension theory of bicompact spaces (Russian), Soob-
sceniya Akademii Nauk Gruzinskol SSR, vol. 2(1941), 1-5.
[f] On homological situation properties of complexes and closed sets, Transactions of
the American Mathematical Society, vol. 54 (1943), 286-339.
[g] On the concept of space in topology (Russian), Uspehi Matematiceskih Nauk
(N. S.) 2, no. 1 (17) (1947), 5-57.
[h] On the dimension of closed sets (Russian), Uspehi Matematiceskih Nauk (N. S.)
4, no. 6 (34) (1949), 17-88.
[i] The present status of the theory of dimension, Russian Translation Project,
American Mathematical Society, vol. 1, 1955.
[j] Duality laws and dimension (Hungarian), Comptes Rendus du Premier Congres
des MathCmaticiens Hongrois, AkadCmiai Kiado, Budapest, 1952, 329-357.
B r o u w e r , L. E. J.
[a] Uber den naturlichen Dimensionsbegriff, Journ. f. Math. (Crelle), vol. 142 (1913).
G a w e h n , I.
[a] Uber unberandete zweidimensionale Mannigfaltigkeiten, Mathematische Annalen,
vol. 98 (1927), 321-354.
G l e z e r m a n , M.
[a] (with Pontryagin, L.) Intersections in manifolds, American Mathematical So­
ciety Translation no. 50.
M o i s e , E. E.
[a] Affine structures in 3-manifolds. V. The triangulation theorem and Hauplver-
mulung, Annals of Mathematics (2), vol. 56 (1952), 96-114.
S c h m i d t , E.
[a] Ueber den Jordanschen Kurvensatz, Sitzungsberichte d. preussischen Akademie
d. Wissenschaften, vol. 28 (1923), 318-329.
U r y s o h n , P.
[a] Memoire sur les m ultiplicity Cantoriennes, Fundamenta Mathematicae, vols.
7-8 (1925-1926).
 IN D E X
Absolute boundary, 41 — for a subspace, 7
------ ■, characterization of, 43 Betti groups, xiii
— system of neighborhoods, 6 in a point, xiv
abstract complex, 125 —■—, invariance of, xiii
— simplex, 121 Betti number, one-dimensional, 68
accumulation point, 5, 8 bicompacta of weight < r, characteri­
addition formula, 49 zation of, 18
admissible broken line, 58 —, theorems on, 17, 18
— deformation, 48, 181 bicompactum, 17
affine mapping, 205 —■, characterization of, 18
Aleksandrov, 1\, xv, xvi, 18, 137, 170, 188 bicontinuous mapping, 10
Alexander, xiv, 65 body, convex, 208
---- Pontryagin duality, xiv, xv — of a complex, 71, 136
Alexandroff (see Aleksandrov) — of a triangulation, 71
angle function, 53 boundary, absolute, 41
arc, Jordan, 39 — elements of a triangulation of a sur­
—, open, 66 face, 77
—, simple, 39 — of a closed n-cell, 167
axioms, closed set, 4 — of a convex polyhedral domain, 120
—, closure, 5 — of a set, 5
— for a metric space, 7 — of a simplex, 119
— for a neighborhood topology, 6 — of a surface, 76
— for a topological space, 4 —, outer, of a star, 73, 126
—, open set, 4 — point of a set, 5
—, partial order, 26 — point of a surface, 76
—, separation, 16 —, regular, 41
—, simple order, 26 branch point of an elementary curve, 66
Brouwer, 17, 37, 42, 155, 156, 170
Baire’s theorem (Theorem 7.14), 29 —'s example, 42 ff.
barycentric complex of a triangulation, ---- Urysohn theorem, 17
144
— coordinates, 203 Cantor, 10
— covering dual to a triangulation, 147 — perfect set, 13, 15, IS, 40, 196
— derived, 131 Cantorian manifold, 199
— (-approximation, 177 carrier, 156
— mapping, 172, 175 — of an element of a subdivided polyhe­
barycentric star, closed, 146
dral complex, 139
------ dual to a simplex, 144
— of a point, 137
barycentric subdivision of a closed sim ­
plex, 157 —■of a simplex, 156
------ of the dual of a complex, 147 carrying plane of a convex set, 210
------ of order v, 140 Cauchy definition of continuity, 9
------ of a polyhedral complex, 132 — sequence, 29
------ of a triangulation, 78 center of a closed barycentric star, 146
basis, countable, 7 — of projection, 214
— for a space, 6 — of a star, 126
218
 IN D EX 219
central subdivision of a polyhedral com­ — topology, fundamental hypothesis of,
plex, 141 150
centroid of a simplex, 213 — topology, method of, 33
— of a system of mass points, 204 — type of simplicial complexes, 124
chain of complexes, 83 compact HausdorIT space, 17
— of semi-stars, S4 — mctrizablc space, 17
— of sets, 14 — space, 17
— of simplexes, 200 compactness, 17
— of subdivisions, 191 —, local, 24
— of triangles, 84 compacts, theorems on, 17
chained system of sets, 14 compaetum, 17
closed barycentric star, 140 — characterization of, 17, IS
— chain of sets, 14 complete metric space, 29
— convex hull, 210 complex, 71
— convex polyhedral domain, 212 —, abstract, 125
— covering, 32 fT. —, connected, 71, 152
— curve, 39 — in a triangulation, 71
— cut line, 100 —, infinite, 158
— 6-covering, 35 —, polyhedral, 119
— manifold, 24, 150 —, pure, 72, 200
— manifolds, examples of, 24 IT. —, simplicial, 122, 125
— a-cell, 167 —, skeleton, 78, 121
— one-sided cut line, S6 —, strongly connected, 72, 200
— path, 47 —, topological, 149
— polyhedral domain, 212 —, unrestricted, 121, 122, 125
— polyhedral manifold, 150 component of a complex, 152 ff.
— projection, 214 — of a space, 15
— set, 4, 7 —, regular, of an elementary curve, 67
— simplex, 212 cone, 214
— solid n-sphere, 8 conc-complex, 134
— subcomplex, 71, 126 connected complex, 71, 152
— surface, 73, 76 — doublet, 13, 16
— topological 7i-manifold, 24 — space, 13
— two-sided cut line, 85 connectedness, 13
closing a hole with a Mobius band, 92 —, local, 41
closure axioms, 5 — of a convex set, 14
— of a set, 5, 6, 7 — of Rn, 14
— of a simplex, 11S — of the real line, 13
— operation, 5 —, strong, 72, 199, 200
coherent orientation, 94 connectivity, invariance of, 100
cohomology, xiii — of a complex, 69
combinatorial boundary of a simplex, 119 — of an elementary curve, 68
combinatorial closure in a complex, 121 — of a surface, 100
------ in a partially ordered set, 27 — of a triangulation, 99
------ of a barycentric star, 145 contact point, 5, 6, 8
------ of a convex polyhedral domain, 120 continuous family of mappings, 32
------ of a simplex, 119 — functions in normal spaces, 17
------ of a simplicial complex, 122 — image of the Cantor perfect set, 18,196
— — of a subcomplex, 126 — image of a connected space, 15
— sum of simplexes, 118 continuous mapping, 9
220 IN D E X

------ , decomposition of a compactum — set, 16

induced by, 18 degree of a mapping, 52 ff.
------ , extension of, 17 dense, 5
------ s of bicompacta, 18 diameter of a metric space, 8
------ s, theory of, xiv difference of sets, 2
contour of a surface, 76 dimension, inductive, 187
convergence, sequential, S —, intuitive meaning of, 37
convex body, 208 — of an abstract simplex, 122
— polyhedral domain, 212 — of a bicompactum, 36, 37, 38
— set, 208 ff. — of a compactum, 37, 38, 177, 179, 197
coordinates, baryccntric, 203 — of a complex, 71
— in a topological product, 10 — of a nerve, 129
countable basis, 7 — of a simplex, 116
— weight, 7 — of a simplicial complex, 122
covering, 4 — of a skeleton complex, 79
—, closed, 32 ff. — of a triangulation, 118
, 35 — theory, 170
—, irreducible 189 —, topological invariance of, 37
—, Lebesgue numbers of a closed, 35 dimension number of a complex, 71, 155
—, open, 32 ff. ------ of a convex set, 210
—, order of, 32 ------ of a polyhedron, 155
—, simple, 37 ------ of a simplex, 116, 155
—s, similar, 33 ------ of a simplicial complex, 122
cross cut, 82 ------ of a topological complex, 149
curve, elementary, 66 ------ of a triangulation, 118
—, Jordan,39 discrete spaces, 16, 28
—, simple closed, 39 ------ and partially ordered sets, 28
curved triangulation, 73 disconnect, 39
cut, 101 disorienting sequence. 96
cut line, 82 displacement, e-, 171
------ , closed, 100 distance, 7, 9
------ , interior closed,'82 — between appoint and a set, 7
------ , nonseparating, 90 — in Euclidean space, S
------ , one-sided closed, 86 — in Hilbert space, 9
------ , open, 82 — in space of continuous mappings, 31
—.—, separating, 90 domain, 14
------ , two-sided closed, 85 doublet, 13
cut operation, 86 ff. —, connected, 13
cycle of an elementary curve, 6S doubling of a three-dimensional ring, 22
— of a 1-complex, 69 dual of a barycentric star, 144
—, relative, xiv — of a complex, 147
—, simple, 68 — of a partially ordered set, 27 ff.
—s, singular, xiii, xiv — of a simplex, 144
—s, true, xiii, xiv — of a triangulation, 144
cyclic star, 73 — subcomplexes, 147
duality laws, xv
Decomposition of an elementary curve, ------ and homology dimension, xv
67 dyadic discontinuum, 13
— space, 19
deformation, 32 Edge, 71
—, admissible, 48, 1S1 Eilenberg, xv
degenerate ra-simplex, 116 elementary curve, 66
 IN D E X 221
elementary subdivision of a complex, 67 — space, 9, 171
------ of a polyhedral complex, 141 homcomorphic spaces, 10
------ of a triangulation, 77 homeomorphism, 10
endpoint of an elementary curve, 66 homological manifold, xiv
e-approximation, barycentric, 177 homology dimension, xiv, 170
e-covering, 35 —, relative, xiv
e-displacement, 171 homology theory, xiii
e-mapping, 170 ------ of bicompacta, xiv
e-neighborhood, S ------ of compacta, xiv
e-net, 30 ------ , topological invariance of, xiii
equivalence of open and closed coverings, homotopic mappings, 32, 53
33 homotopy, xv
essential mapping, 181 — classes of mappings, 32
Euclidean n-space, S — groups, xv
Euler characteristic of an elementary Hopf, H., xiii, xiv, xv
curve, 67 Hurewicz, xv, 200
------ of a 1-complex, 67
------ of a triangulation, 99 Identification, 20, 124
Euler theorem, 69, 99, 105 — of elements of a triangulation, 79
extension of a continuous mapping, 17 — of the first kind, 93
— of an affine mapping, 208 — of partially ordered sets and discrete
spaces, 29
Face of a convex polyhedral domain, 213 — of points of a space, 20
— of a simplex, 116 — of the second kind, 93
—s, opposite, 117, 212 — of vertices of a simplicial complex, 125
family depending on a parameter, 4 — of vertices of a skeleton complex, 79
fixed point theorem, 156, 168 —s, examples of, 20 ff., 80 ff.
frontier of a simplex, 118 image, 3
fundamental group, xv imbedding theorems, 18
— hypothesis of combinatorial topol­ incident simplexes, 117
ogy, 150 index of a point relative to a closed path,
— sequence, 29 47
— theorem of surface topology, 110 ---------- relative to a continuous map­
ping, 47
Gawehn, 73 ---------- relative to a mapping of a cir­
general position, 204 cumference, 52
genus of a nonorientable surface, 108 ---------- relative to an oriented Jordan
— of an orientable surface, 108 curve, 51
geometric partial order, 26 indexed set, 3
geometric realization of a nerve, 131 — system of sets, 3
Glezerman, xv induced orientation, 94
inductive dimension, 187
Half-space, closed, 203 ------ at a point, 187
—, open, 202 infinite dimension, 37
handle of the first kind, 93 — triangulation, 158
---------- second kind, 93 interior closed cut, 82
Hauptvermutung, 150 — element of a surface, 77
Hausdorff, 2 — of a set, 5
— space, 16 interior point, 5
------ , compact, 17 ------ of a convex set, 210
------ , normal, 16 ------ of a set, 5
Hilbert parallelotope, 9, 13, 171, 174, 179 ------ of a surface, 76
222 IN D E X

intersection of sets, 2 —, simple, 50

—s, theory of, xiv, xv —, simplicial, 79, 123
invariance of connectivity, 100 —s, composition of, 3
— of dimension number of Rn, 155, 163 —s, uniform convergence of, 10
— of the Euler characteristic, 99 Mazurkiewicz, 156
— of interior points, 74, 156, 164 ff. mesh, 157
— of orientability, 95 metric, 7
inverse image, 3 metric space, 7
irreducible <-covering, 189 ------ , complete, 29
isolated point, 5 ------ , locally connected, 41
isomorphism of abstract complexes, 125 ------ , totally bounded, 30
— of simplicial complexes, 124 metrizable space, 10
— of skeleton complexes, 79 ------ , compact, 17
metrization theorems, 18
Jordan arc, 39 Mobius, 204
— curve, 39 — band, 80, 91, 96
— theorem, 39 IT. modification of a closed covering, 164 fT.
Moise, 150
Klein bottle, 81, 97, 114
Knaster, 156 Natural order, 26
Kuratowski, 156 — topology of a metric space, 8
n-cell, 167
Lebesgue, 37 negative affine mapping, 206
— lemma, 35, 162 neighborhood, 5
— numbers of a closed covering, 35 — in a decomposition space, 19
Lefschetz, xiv, xv — in a topological product, 11
---- Hopf formula, xiv —, spherical, 8
limit point, 5, 8 — system, 6
linear closure of planes, 204 — topology, 6
linear independence of cycles, 68 —s, absolute system of, 6
------ of points, 203 nerve of a covering, 129 fT.
linking theory, xiv —s, examples-of, 129 fT.
local compactness, 24 n-manifold, 24
— connectedness, 41 —s, examples of, 24 fT.
(n-l)-sphere, 9
Major star, 144 noncoherent orientation, 94
manifold, homological, xiv nonorientable surface, 95
—, topological n-, 24 — triangulation, 94
—s with boundary, xiv nonseparating cut, 90
mapping, 3 — subcomplex, 99
—, affine, 205 non-singular affine mapping, 205
—, bicontinuous, 10
normal closed nonorientable surface of
—, continuous, 9
genus p, 109
170
—, essential, 181 ------ orientable surface of genus p, 109
— into, 3 normal llausdorfT space, 16
—, normal, of degree to, 52 normal mapping of degree a>, 52
— of order n, 196 normal simple surface with r contours,
(1-D, 3 106'
— o n to ,3 n-simplex, 116
—, similarity, 26 n-sphere, 9, 24
 IN D E X 223

1-complex, 67 products, theory of, xv

one-dimensional Betti number of an ele­ projection, 214
mentary curve, 6S projective 5-space, 25
1-element, 67 — n-space, 22 IT., 151
one-sided closed cut line, S6 — plane, 81, 97, 114, 151
1-sphere, 11 — three-space, 24 ff.
(1-1) mapping, 3 proper face, 117, 212
open cone, 134 ------ of a degenerate simplex, 117
— covering, 32 ff. — subset, 3
— cut, S2 proportional sequences, 22
— manifold, 24, 25 pure complex, 72, 200
— n-cell, 167
— projection, 214 Ratio, 22
— set, 4 real line, 7, 13
— simplex, 212 reduction of holes, 90 ff.
— solid rc-sphere, 9 — by means of a handle, 93
— stars of a triangulation, 13S refinement of a system of sets, 32
— subcomplex, 71, 126 regular boundary, 41
opposite faces of a simplex, 212 — component of an elementary curve, 67
order axioms, 26 — point of an elementary curve, 66
------ , partial, 26 — subdivision of a triangulation, 78
------ , simple, 26 relative cycles, xiv
order of a covering, 32 — homology, xiv
— of a finite system of sets, 4 — topology, 5
— of a mapping, 196 retraction, 172, 174
orientable surface, 95 r-face', 117
— triangulation, 94 r-plane, 202
orientation of a Jordan arc, 45 r-simplex, 212
— of a Jordan curve, 46
— of a triangle, 94 Schmidt, E., 39
— of a triangulation, 94 semi-cyclic star, 73
outer boundary of a barvcentric star, 144 semi-star, 83
------ of a star, 73, 126 separate, 39
separation axioms, 16
Partial order in a complex, 71, 118 — of two points, 55
partially ordered set, 25 ff. sequential convergence, 8
p-face, 212 similar coverings, 33
PHastersatz, 37, 155, 162 — partially ordered sets, 26
phase space, 13 similarity mapping, 26
plane, carrying, 210 simple arc, 39
— closed curve, 39
— of support, 213
— covering, 37
Plucker coordinates, 24
— cycle of an elementary curve, 68
Poincar6, xiv, xv — cycle of a 1-complex, 68
point set boundary of a simplex, 118 — mapping, 50
polyhedral complex, 119 — nerve, 129
polyhedron, 71, 118, 136 — surface, 104
—, triangulation of, 137 — system of sets, 4
Pontryagin, xiv, xv, 128, 134 — triangulation of a surface, 102
positive affine mapping, 206 simplex, 116, 212
prism, 135 ff. —, abstract, 121
224 IN D E X

—, boundary of, 119 — ftNo, 13

—, carrier of, 156 — Rr, 13, 17
—, centroid of, 213 —, To -, T\ -, Ti -, 16
—, closed, 118 —, topological, 4
—, degenerate, 116 —, totally bounded metric, 30
—, dimension of, 116 — with countable basis, 7
—, face of, 116 special covering, 189
—, n-, 116 spectral theory, xiv
—, opposite faces of, 117 Sperner’s lemma, 156, 160 ff.
—, skeleton of, 116 sphere, closed solid n-, 8
—, topological, 149 — (« -l)- ,9
—, triangulation of a closed, 156 —, open solid n-, 9
—, vertices of, 116 — with p handles, 109
—es, chain of, 200 — with p + 1 cross-caps, 109
—es, incident, 117 spherical neighborhood, 8
simplicial complex, 122, 125 star, 27, 72
------ es, examples of, 122 ff. —, barycentric, 144
simplicial mapping of a polyhedron, 139 —, center of, 126
------ of a simplicial complex, 123 —, cyclic, 73
------ of a skeleton complex, 123 — in a partially ordered set, 27
singular affine mapping, 205 —, major, 144
— point of an elementary curve, 66 — neighborhood, 137
skeleton, 78, 121, 129 — of a subcomplex, 72, 126
— complex, 78, 120 —, open, 138
------ of a simplicial complex, 122 —, semi-, 83
------ of a triangulation, 78 —, semi-cyclic, 73
-------, unrestricted, 121 —, zone of, 127
------ es, isomorphism of, 79 strongly connected compactum, 199
— of a degenerate simplex, 116 ------ pure complex, 72, 200
skeleton of a simplex, 121 ------ polyhedron, 201
—s, sum (mod 2) of, 68 subcomplex, 71
space, compact Hausdorff, 17 —, closed, 71, 126
—, compact metrizable, 17 —, open, 71, 126
—, compact topological, 17 subdivision, barycentric, 78
—, complete metric, 29 —, central, 140
—, connected, 13 —, elementary, 141
— DT, 13, 17, 18 — of a closed covering, 190
—, decomposition, 19 — of a closed simplex, 140
—, discrete, 16, 28 — of a polyhedral complex, 139
—, Euclidean, 8 — of triangulations, 77 ff.
- , Fr, 13, 16 —, regular, 7S
—, Hausdorff, 16 —s, chain of, 191
—, Hilbert, 9, 171 sum, combinatorial, 118
—, locally compact, 24 — of sets (mod 2), 68
—, locally commected, 41 — theorem, 184 ff.
—, metric, 7 surface, 76
—, metrizable, 10 —, boundary of, 76
—, normal, 16 —, closed, 73, 76
— of continuous mappings, 31 —, contour of, 76
— of countable weight, 7 —, nonorientable, 95
—, projective, 22 fF. —, normal, 109
 IN D E X 225

— of genus p, 108 —, curved, 73

—, orientable, 95 —, dimension of, 118
—, simple, 104 —, infinite, 158
— with boundary, 76 — of a closed simplex, 156
system of subsets, 4 — of a closed surface, 74
, partially ordered, 26 — of a Klein bottle, 81
— of a Mobius band, 80
Three-dimensional ring, 22 — of a polyhedron, 71, 137
Tihonov, 11, 17, IS — of the projective n-space, 151
topological circle, 104 — of the projective plane, 81, 151
— circumference, 104 — of a torus, 81
— complex, 149 —, orientability of, 94
— mapping, 10 —, subdivision of, 77 ff.
— /i-manifold, 24, 19S —s, examples of, 119
— polyhedron, 1IS, 149 true cycles, xiii, xiv
— simplex, 149 Tumarkin, 188, 200
— space, 4 (T. turning function, 53
— triangulation, 149 2-complex, 71
topological product, 10 IT. two-sided cut of the first kind, 111
------ of a denumerable number of dou­ ------ of the second kind, 111
blets, 13 2-SDhere, 10, 105
------ of a finite number of closed mani­
folds, 24 Uniform continuity of a mapping, 18
------ of a 1-sphere and a line, 13 — convergence of mappings, 10
-------of a 1-spherie and a 2-sphere, 11 ff., union of sets, 2
21 ff., 151 unrestricted simplicial complex, 122, 125
------ of closed segments, 13 — skeleton complex, 121
------ of bicompacta, 18
UDDer semi-continuous decomposition of
------ of compacta, 18
a oompactum, 19
------ of n lines, 11
------ of n 1-spheres, 11 ---------- s, examples of, 20 ff.
------ of n-spheres, 24 Urysohn, 9, 16, 17, 18, 170, 188. 189
------ of sets, 10 —’s theorem, 9
------ of spaces, 10 ff.
------ of r connected doublets, 13 Vertex, 68, 128
------ of t doublets, 13 —, first, 132
------ of t closed segments, 13 —, last, 132
------ of two lines, 11 — mapping, 79
------ of two 1-spheres, 11 — of a cone, 214
topology defined in a set, 5 — of a finite system of sets, 128
—, neighborhood, 6 — of a nerve, 128
—, relative, 5 — of a simplex, 212
torus, 11, 21, 22, 81 — of a skeleton complex, 78
—, n-dimensional, 11, 24
—, three-dimensional, 21, 150 Weight of a space, 7
totally bounded metric space, 30
triangle, 71 Yoneyama, 42
— axiom, 7
—, oriented, 93 0-element, 67
triangulation, 71, 118 Zone of a star, 127