Geometries: A. B. Sossinsky

S T U D E N T M AT H E M AT I C A L L I B R A RY
Volume 64
Geometries
A. B. Sossinsky
Geometries
S T U D E N T M AT H E M AT I C A L L I B R A RY
Volume 64
Geometries
A. B. Sossinsky
American Mathematical Society

Providence, Rhode Island
Editorial Board
Gerald B. Folland Brad G. Osgood (Chair)
Robin Forman John Stillwell
2010 Mathematics Subject Classification. Primary 51-01;

Secondary 51-02, 01A20, 01A55, 18-01.
For additional information and updates on this book, visit

www.ams.org/bookpages/stml-64
Library of Congress Cataloging-in-Publication Data

Sosinskii, A. B. (Aleksei Bronislavovich)
Geometries / A. B. Sossinsky.
p. cm. – (Student mathematical library ; v. 64)
Includes bibliographical references and index.
ISBN 978-0-8218-7571-1 (alk. paper)
1. Geometry–Textbooks. I. Title.
QA445.S593 2012
516–dc23 2012002357
Copying and reprinting. Individual readers of this publication, and nonprofit

libraries acting for them, are permitted to make fair use of the material, such as to
copy a chapter for use in teaching or research. Permission is granted to quote brief
passages from this publication in reviews, provided the customary acknowledgment of
the source is given.
Republication, systematic copying, or multiple reproduction of any material in this
publication is permitted only under license from the American Mathematical Society.
Requests for such permission should be addressed to the Acquisitions Department,
American Mathematical Society, 201 Charles Street, Providence, Rhode Island 02904-
2294 USA. Requests can also be made by e-mail to [email protected].

c 2012 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10 9 8 7 6 5 4 3 2 1 17 16 15 14 13 12
Contents
Preface xiii
Chapter 0. About Euclidean Geometry 1

§0.1. The axioms of Euclidean plane geometry 2
§0.2. Commentary 5
§0.3. Rotations 7
§0.4. Parallel translations and vectors 11
§0.5. Triangles: congruence, properties 13
§0.6. Homothety and similitude 15
§0.7. Angle measure and trigonometry 18
§0.8. Properties of the circle 20
§0.9. Isometries of the plane 24
§0.10. Space geometry 28
Chapter 1. Toy Geometries and Main Definitions 33

§1.1. Isometries of the Euclidean plane and space 33
§1.2. Symmetries of some figures 35
§1.3. Transformation groups 41
§1.4. The category of geometries 46
§1.5. Some philosophical remarks 49
v
vi Contents
§1.6. Problems 50
Chapter 2. Abstract Groups and Group Presentations 53

§2.1. Abstract groups 53
§2.2. Morphisms of Groups 57
§2.3. Subgroups 58
§2.4. The Lagrange theorem 59
§2.5. Quotient groups 60
§2.6. Free groups and permutations 61
§2.7. Group presentations 62
§2.8. Cayley’s theorem 64
§2.9. Problems 65
Chapter 3. Finite Subgroups of SO(3) and the Platonic Bodies 67

§3.1. The Platonic bodies in art, philosophy, and science 68
§3.2. Finite subgroups of SO(3) 70
§3.3. The five regular polyhedra 77
§3.4. The five Kepler cubes 78
§3.5. Regular polyhedra in higher dimensions 79
§3.6. Problems 81
Chapter 4. Discrete Subgroups of the Isometry Group of the

Plane and Tilings 85
§4.1. Tilings in architecture, art, and science 85
§4.2. Tilings and crystallography 87
§4.3. Isometries of the plane 89
§4.4. Discrete groups and discrete geometries 90
§4.5. The seventeen regular tilings 90
§4.6. The 230 crystallographic groups 95
§4.7. Problems 95
Chapter 5. Reflection Groups and Coxeter Geometries 99

§5.1. An example: the kaleidoscope 99
Contents vii
§5.2. Coxeter polygons and polyhedra 100

§5.3. Coxeter geometries on the plane 101
§5.4. Coxeter geometries in Euclidean space R 3
103
§5.5. Coxeter schemes and the classification theorem 105
§5.6. Problems 107
Chapter 6. Spherical Geometry 109

§6.1. A list of classical continuous geometries 109
§6.2. Some basic facts from Euclidean plane geometry 113
§6.3. Lines, distances, angles, polars, and perpendiculars 114
§6.4. Biangles and triangles in S 2
116
§6.5. Other theorems about triangles 120
§6.6. Coxeter triangles on the sphere S 2
121
§6.7. Two-dimensional elliptic geometry 121
§6.8. Problems 123
Chapter 7. The Poincaré Disk Model of Hyperbolic Geometry 125

§7.1. Inversion and orthogonal circles 126
§7.2. Definition of the disk model 131
§7.3. Points and lines in the hyperbolic plane 133
§7.4. Perpendiculars 134
§7.5. Parallels and nonintersecting lines 134
§7.6. Sum of the angles of a triangle 135
§7.7. Rotations and circles in the hyperbolic plane 136
§7.8. Hyperbolic geometry and the physical world 138
§7.9. Problems 139
Chapter 8. The Poincaré Half-Plane Model 143

§8.1. Affine and linear-fractional transformations of C 144
§8.2. The Poincaré half-plane model 147
§8.3. Perpendiculars and parallels 148
§8.4. Isometries w.r.t. Möbius distance 150
§8.5. Problems 151
viii Contents
Chapter 9. The Cayley–Klein Model 153

§9.1. Isometry and the Cayley–Klein model 153
§9.2. Parallels in the Cayley–Klein model 156
§9.3. Perpendiculars in the Cayley–Klein model 158
§9.4. The hyperbolic line and relativity 159
§9.5. Problems 160
Chapter 10. Hyperbolic Trigonometry and Absolute Constants 163

§10.1. Isomorphism between the two disk models 163
§10.2. Isomorphism between the two Poincaré models 168
§10.3. Hyperbolic functions 169
§10.4. Trigonometry on the hyperbolic plane 170
§10.5. Angle of parallelism and Schweikart constant 170
§10.6. Problems 173
Chapter 11. History of Non-Euclidean Geometry 177

§11.1. Euclid’s Fifth Postulate 177
§11.2. Statements equivalent to the Fifth Postulate 178
§11.3. Gauss 179
§11.4. Lobachevsky 180
§11.5. Bolyai 182
§11.6. Beltrami, Helmholtz, Lie, Cayley, Klein, Poincaré 183
§11.7. Hilbert 184
Chapter 12. Projective Geometry 185

§12.1. The projective plane as a geometry 185
§12.2. Homogeneous coordinates 186
§12.3. Projective transformations 188
§12.4. Cross ratio of collinear points 191
§12.5. Projective duality 192
§12.6. Conics in RP 2
194
§12.7. The Desargues, Pappus, and Pascal theorems 194
§12.8. Projective space RP 3 199
Contents ix
Chapter 13. “Projective Geometry Is All Geometry” 203

§13.1. Subgeometries 203
§13.2. The Euclidean plane as a subgeometry of the
projective plane RP 2 204
§13.3. The hyperbolic plane as a subgeometry of the
projective plane RP 2 205
§13.4. The elliptic plane as a subgeometry of RP 2 207
Chapter 14. Finite Geometries 211

§14.1. Small finite geometries 212
§14.2. Finite fields 212
§14.3. Example: the finite affine plane over F(5) 213
§14.4. Example: the finite affine plane over F(2 )2
215
§14.5. Example of a finite projective plane 216
§14.6. Axioms for finite affine planes 217
§14.7. Axioms for finite projective planes 218
§14.8. Constructing projective planes
over finite fields 220
§14.9. The Desargues theorem 221
§14.10. Algebraic structures in finite projective planes 223
§14.11. Open problems and conjectures 226
Chapter 15. The Hierarchy of Geometries 229

§15.1. Dimension one: lines 230
§15.2. Dimension two: planes 232
§15.3. From metric to affine to projective 234
§15.4. Three-dimensional space geometries 235
§15.5. Finite and discrete geometries 236
§15.6. The hierarchy of geometries 236
x Contents
Chapter 16. Morphisms of Geometries 241

§16.1. Examples of geometric covering spaces 242
§16.2. Examples of geometric G-bundles 245
§16.3. Lie groups 247
§16.4. Examples of geometric vector bundles 248
§16.5. Geometric G-bundles 250
§16.6. The Milnor construction 251
Appendix A. Excerpts from Euclid’s “Elements” 255

Postulates of Book I 256
The Common Notions 257
The Definitions of Book I 258
The Propositions of Book I 262
Conclusion 269
Appendix B. Hilbert’s Axioms for Plane Geometry 271

I. Axioms of connection 272
II. Axioms of order 274
III. Axiom of parallels 275
IV. Axioms of congruence 276
V. Axiom of continuity 279
Consistency of Hilbert’s axioms 280
Conclusion 281
Answers & Hints 283

Chapter 1 283
Chapter 2 285
Chapter 3 286
Chapter 4 288
Chapter 5 290
Contents xi
Chapter 6 290
Chapter 7 291
Chapter 8 294
Chapter 9 294
Chapter 10 295
Chapter 12 295
Chapter 13 296
Chapter 14 296
Chapter 15 296
Chapter 16 296
Bibliography 297
Index 299
Preface
This book is dedicated to the proposition that all geometries are cre-
ated equal. This was first pointed out by Felix Klein, who declared
that each individual geometry is a set with a transformation group
acting on it. Here we shall study geometries from this point of view
not only as individual objects, but also in their social life, i.e., in their
relationships (called morphisms or equivariant maps) within their so-
cium: the category of geometries.
Of course, some geometries are more equal than others. Accord-
ingly, we will ignore the most common ones (affine and Euclidean ge-
ometries, vector spaces), assuming that they are known to the reader,
and concentrate on the most distinguished and beautiful ones. (We
assume that our readers are familiar with elementary Euclidean ge-
ometry; those who aren’t may refer to Chapter 0, which is a précis of
the subject, whenever the need arises.)
The reader should not be deceived by the words groups, mor-
phisms, categories into thinking that this is a formal algebraic or
(heaven forbid) an analytic (coordinate) treatment of geometric top-
ics; it suffices to glance at the numerous figures in the book to realize
that we constantly privilege the visual aspect.
Category theory is not used in this book, but we do use some basic
category language, which, as the reader will see, is extremely natural
xiii
xiv Preface
in the geometric context. Thus, Cayley’s famous phrase: “projec-

tive geometry is all geometry” can be given a precise mathematical
meaning by using the term subgeometry (which means “image by an
injective equivariant map”). In the context of this book, it may be
rephrased as follows: “The geometries studied in this book (includ-
ing the three classical ones – hyperbolic, elliptic, and Euclidean) are
(almost all) subgeometries of projective geometry.”
There is very little in the main body of this book about the ax-
iomatic approach to geometry. This is one of the author’s biases: I
believe that the classical axiom systems for, say, Euclidean and hy-
perbolic geometry are hopelessly outdated and no longer belong in
contemporary mathematics. Their place is in the history of mathe-
matics and in the philosophy of science. Accordingly, here they only
appear in one chapter, devoted to the fascinating history of the cre-
ation of non-Euclidean geometry, while a detailed treatment of the
axiom systems of Euclid and Hilbert is relegated to Appendices A
and B.
The use of the plural (Geometries) in the title of the book indi-
cates that, to my mind, there is no such subject as “geometry”, but
there are some concrete mathematical objects called geometries. In
the singular, the word “geometry” should be understood as a way
of thinking about mathematics, in fact the original one: in Ancient
Greece, the word “geometry” was used as a synonym for “mathemat-
ics”. One can and should think geometrically not only when working
with circles and triangles, but also when using commutative diagrams,
morphisms, or groups. The famous phrase written above the entrance
to Plato’s Academy
Let no one enter who is not a geometer
should also be displayed on the gates leading to the world of mathe-

matics.
* * *
I will not give a systematic summary of the contents of this course
in this Preface, referring the reader to the Table of Contents. Look-
ing at it, the well-prepared reader may wonder why some of her/his
Preface xv
favorite geometric topics do not appear in this book among the “dis-
tinguished and beautiful” ones promised above. Let me comment on
some missing topics, explaining why they are not treated here.
First, there is no algebraic geometry in this book. This is because
the author believes that this beautiful field of mathematics belongs to
algebra, not geometry. Indeed, the mathematicians doing algebraic
geometry are typically algebraists, and this is not only true of the
great French school (following Grothendieck and his schemes), but
also of the more classical Russian school.
Neither is there any differential geometry: in its classical low-
dimensional aspect it is usually developed in calculus books (where it
indeed belongs); in its higher-dimensional modern aspect it is a part
of analysis (under the title “Calculus on Manifolds”) and topology
(under the title “Differential Topology”).
Other missing topics include convex geometry (part of analysis
and more specifically optimization theory as “convex analysis”), sym-
plectic geometry (part of classical mechanics and dynamical systems),
contact geometry (part of differential equations), etc.
Of course, contact geometry (say) is formally a geometry in the
sense of Klein. In fact, the ideology of transformation groups comes
from Sophus Lie as much as (if not more than) from Felix Klein, but
the context of Lie’s beautiful contact geometry is definitely differential
equations.
* * *
This book is based on lectures given in the framework of semester

courses taught in Russian at the Independent University of Moscow to
first-year students in 2003 and 2006, for which I prepared handouts,
written in “simple English” and posted on the IUM web site. These
lecture notes were published as a 100-page booklet by the Moscow
Center of Continuous Mathematical Education in 2006 and used in
geometry courses taught to Math in Moscow students of the Indepen-
dent University.
xvi Preface
The brevity of a short semester course (13 lectures) made me

restrict the study of such classical geometries as hyperbolic and pro-
jective to dimension two, regretfully shelving the three-dimensional
case. But then there are many occasions in the course for developing
one’s intuition of space, and indeed the general case is easier to treat
from the linear algebra coordinate point of view than from the rather
visual synthetic approach characterizing this course. For the reader
who wants to go further, I strongly recommend the book by Marcel
Berger [2]. I should add that, although my approach to the subject is
very different from Berger’s, I am heavily indebted to that remarkable
book in several specific parts of the exposition. For those who would
like to learn more about the axiomatic approach to the classical ge-
ometries, there is no better book, to my mind, than N.V. Efimov’s
Higher Geometry [6].
An important, if not the most important, aspect of this book
are the problems, which appear at the end of each chapter. It is by
solving these problems, much more than by learning the theory, that
the reader will become capable of thinking and working geometri-
cally. The sources of the problems are varied. Many were “stolen”
from books written by my friend and favorite co-author Victor Pra-
solov. In many cases, I simply don’t know where they originally
come from. As handouts for the exercise classes, they were grouped
together by Irina Paramonova, who contributed several, as did the
other instructors conducting the exercise classes (Vladimir Ivanov and
Oleg Karpenkov). I am grateful to all of the people mentioned above,
and also to Mikhail Panov, Anton Ponkrashov, and Victor Shuvalov,
who produced the computer versions of most of the illustrations, to
M.I. Bykova, who corrected many errors in the original handouts,
to Victor Prasolov, who found many more in the first draft of this
book, and to the anonymous referees, whose constructive criticism
was very helpful. Finally, I am indebted to Sergei Gelfand, without
whose encouragement this book would never have been written.
Chapter 0
About Euclidean
Geometry
The first chapter of this book is Chapter 1 (see p. 33 below); the

present chapter (numbered 0) contains some standard facts about Eu-
clidean geometry that can be regarded as prerequisites for reading this
book. The author feels that the majority of readers should omit this
chapter, even if they did not have a sound Euclidean geometry course
in high school or college, and go right on to Chapter 1. Some concrete
facts from Euclidean geometry are briefly recalled in the main text
of the book when they are needed for our further exposition, and if
they turn out to be unfamiliar or not too well remembered, then, by
returning to Chapter 0, the reader can brush up on them, replacing
these facts in the context of a systematic exposition of elementary
Euclidean geometry.
Our exposition of Euclidean plane geometry is axiomatic. The
axioms are chosen on the basis of Klein’s approach to geometry, so
that transformations play the key role, with a distance function speci-
fied from the outset, as is done in Kolmogorov’s high school geometry
textbooks. Our approach presupposes the knowledge of the real num-
bers and their main properties, some familiarity with the language of
naive set theory (no serious knowledge of set theory itself is assumed),
1
2 0. About Euclidean Geometry
and some experience in the kind of logical reasoning used in math-

ematics (e.g. arguing by contradiction or by induction, splitting a
proof into various cases, and so on).
0.1. The axioms of Euclidean plane geometry

There are three types of undefined notions in our theory: points,
(straight) lines, and the (Euclidean) plane. Points are denoted by var-
ious capital letters (A, B, C, . . . , P, Q, . . . , sometimes supplied
with subscripts or superscripts), lines are denoted by small italics
(l, m, n, . . . , also possibly endowed with subscripts or superscripts),
and the plane is denoted by E2 . We are also given a distance function
d, which assigns a nonnegative real number to each pair of points,
d : (A, B) → d(AB) ∈ R+ .
In what follows, we will also use the more traditional notation |AB|
for the value of the distance d(A, B) between two points.
These objects satisfy the following eight axioms.
I. The plane E2 is the set of all points.
II. The family L of all lines consists of nonempty subsets of the

plane E2 .
III. For any two distinct points A and B there exists one and
only one line l = AB ∈ L containing these two points.
IV. For any line l ∈ L there exists at least one point P ∈ E2 not
contained in that line.
To state the next axiom, we need a definition. Two lines l1 and l2
are called parallel if they coincide or if they have no common points;
if the lines l1 and l2 are parallel, we write l1 l2 .
V. For any point P ∈ E2 and any line l ∈ L there exists one and
only one line l ∈ L parallel to l and containing P .
VI. The distance function d possesses the following properties:

(i) d(A, B) = 0 if and only if A = B;
(ii) d(A, B) = d(B, A) for all A, B ∈ E2 ;
0.1. The axioms of Euclidean plane geometry 3
(iii) d(A, B) + d(B, C) ≥ d(A, C) for all A, B, C ∈ E2 ; the

points A, B, C lie on the same line if and only if the previous inequality
is in fact an equality;
(iv) for any point P on a line l ∈ L and any positive number ρ
there are exactly two points A and B such that d(A, P ) = d(P, B) = ρ.
A few more definitions are needed to state the last axioms.

The first is that of isometry, which plays the key role in Klein’s
approach to the study of Euclidean geometry and of many other ge-
ometries (see Section 1.4 in Chapter 1). An isometry is a bijection β of
E2 (i.e., a one-to-one transformation of E2 onto itself) that preserves
distances:
d(β(P ), β(Q)) = d(P, Q) for all A, B ∈ E2 .
Following the traditional terminology, we will often call subsets of

the plane E2 (geometric) figures. Two figures F1 , F2 ⊂ E2 are called
congruent if there exists an isometry ϕ of the plane that takes one
onto the other, ϕ(F1 ) = F2 .
The next definition is that of the “between” relation: if for three
distinct points A, B, C, we have d(A, B) + d(B, C) = d(A, C), then
we say that the point B lies between the points A and C on the line
l = AC. (Note that by axiom III we have l = AC = CA = AB =
BA = BC = CB.) The set of all points lying between two distinct
points A and B is denoted by (A, B) and said to be the open interval
with endpoints A and B. If we add the endpoints to the points of an
open interval (A, B), we obtain the closed interval (or segment) with
endpoints A and B, which is denoted by [A, B]. Two distinct points
O and A determine the ray with origin O passing through A, denoted
by [O, A, as the union of the segment [O, A] and all points B such
that A lies between O and B.
Remark. In some expositions of Euclidean geometry (e.g., Hilbert’s,

see Appendix B), the “between” relation is an undefined relation. In
most elementary textbooks on plane geometry, nothing is said about
this relation itself, the exposition leans heavily on the illustrations,
and a point is declared to lie between two other points provided it
appears that way on the corresponding figure.
The union of two rays [O, A and [O, B with common origin O is
called an angle and denoted ∠AOB or ∠BOA; the point O is called
the vertex of the angle. Two intersecting lines AA ∩ BB = O, where
O lies between A and A , and also between B and B , determine four
angles: ∠AOB, ∠A OB , ∠AOB , ∠A OB; the first two, as well as the
last two, are called vertical to each other. Two angles with a common
ray located between their two other rays are called adjacent. If we
remove the common ray of two adjacent angles, we obtain an angle
that we call the (geometric) sum of the two given angles.
Two intersecting lines that form four congruent angles are called
perpendicular, and these four angles are said to be right angles. If
the two rays of a given angle ∠AOB lie on one line and have no
common points other than the vertex, we say that ∠AOB forms two
right angles.
We are now ready to formulate the next to last axiom.
VII. For any line l ∈ L and any point P ∈ E2 , there exists a

unique perpendicular to l containing P .
Figure 0.1. Dropping and raising perpendiculars.
Note that the axiom does not specify whether P lies on l or

not, so that accordingly two different pictures illustrate this axiom
(Figure 0.1). In the traditional teaching of geometry, one often uses
the expression raise the perpendicular to l from P in the second case
and drop the perpendicular from P to l in the first one. In both cases
the intersection point of the perpendicular with l is often called the
foot of the perpendicular.
0.2. Commentary 5
One more definition, a very important one, is needed to formulate

the last axiom. Let l ∈ L be a line; then the reflection in that line is
the assignment Sl : E2 → E2 that leaves each point of l fixed and takes
any point P ∈/ l to the point P := Sl (P ) lying on the perpendicular
drawn from P to l and such that |P H| = |HP |, where H is the foot
of the perpendicular and lies between P and P .
VIII. The reflection in any line is an isometry.
If l is any given line and P is a point not on l, then the half-plane

determined by l and P is the set of all points M of the plane such
that the segment [P, M ] does not intersect l. It follows immediately
from the definition of reflection (and the axioms) that any reflection
in l interchanges the two half-planes determined by l.
0.2. Commentary
0.2.1. The axioms listed above, except for axiom V (the axiom of
parallels) are intuitively obvious. In the standard school model of
elementary school geometry, in which the plane (or rather part of it)
is a piece of paper lying on a flat table, points are marks on that
paper made by a well-sharpened pencil, lines are obtained by sliding
the pencil along the edge of a ruler, and distances between points
are measured in the usual way by means of that ruler, axioms I–IV
and VI–VIII may actually be regarded as experimental facts. This
is not true of the axiom of parallels (lines being infinitely long, we
cannot extend them “to infinity” to verify experimentally how many
parallels there are, if any). As to the reflection axiom (VIII), it can
be modeled by placing a piece of tracing paper with a line drawn on
it on our “paper plane” (on which a line l is also drawn) so that the
two lines coincide, and then flipping the tracing paper around and
placing it back on the paper plane so that the lines coincide as they
did before; if a point P on the paper plane is given and its position on
the tracing paper is denoted by P1 , the position of the point P1 after
the tracing paper is turned over will play the role of the image P of
P under the reflection; the fact that distances are preserved seems
obvious (one cannot stretch or shrink the tracing paper).
0.2.2. Independence. The axioms I–VIII are not independent:

some of their assertions can be derived from the other axioms. More-
over, the undefined notion of “straight line” can actually be rigorously
defined by using axiom VI, the notion of set, and the undefined notion
of “point”. If that is done, axiom III becomes a theorem.
Hilbert constructed a rigorous axiom system for Euclidean geom-
etry (see Appendix B) in which the axioms are independent (in the
sense that there is no axiom that can be logically deduced from all the
others). His approach is conceptually important for the foundations
of mathematics, but is not very satisfactory from the pedagogical
point of view.
0.2.3. Consistency. The axiom system above is consistent (i.e., no

contradiction can be obtained from its axioms by correct logical argu-
ments) provided that the theory of real numbers is consistent. This
can be proved by constructing a model for this axiom system, i.e., sup-
plying the undefined terms with concrete meanings within the theory
of real numbers so that the axioms are theorems in that theory. This is
done as follows: the undefined notion of “plane” is interpreted as R2 ,
the set of ordered pairs of real numbers, “points” are pairs (x, y) ∈ R2 ,
“lines” are sets of pairs satisfying linear equations αx + βy + γ = 0,
and the distance function is defined by the formula

d (x1 , y1 ), (x2 , y2 ) = (x1 − x2 )2 + (y1 − y2 )2 .
The fact that these interpretations of the undefined notions satisfy
axioms I–VIII can be verified in a straightforward way.
0.2.4. Categoricity. The axiom system above is categorical, i.e.,

any two of its models are isomorphic in a certain natural sense that we
do not specify here. Nor do we comment on the fairly straightforward
proof of this fact.
0.2.5. Why the axiomatic approach? The reader may wonder

why Euclidean plane geometry was described here by means of ax-
ioms, when there is a much simpler way of introducing it, namely by
defining E2 := R2 as explained in Subsection 0.2.3. There are two rea-
sons for this. The first is to give a tribute to the traditional teaching
0.3. Rotations 7
of geometry (to some version of which the reader was originally sub-
jected); the second is that the coordinate approach to geometry, to
the author’s mind, is an ugly caricature of what Euclidean geometry
really is.
We must also warn the reader that our exposition is probably
more rigorous and formalized than traditional ones in high school
geometry textbooks, but also less detailed. In particular, here the
proofs are either only sketched or omitted altogether (but the theo-
rems appear in a logical order, so that each easily follows from the
previous theorems and the axioms). Also there are no exercises in
this chapter.
0.3. Rotations
0.3.1. Properties of isometries. Any isometry:
(i) takes lines to lines;
(ii) takes open intervals, segments, rays to open intervals,
segments, rays, respectively;
(iii) takes parallel lines to parallel lines;
(iv) takes perpendiculars to perpendiculars;
(v) takes angles to angles;
(vi) the composition of two isometries is an isometry.
All these properties immediately follow from the corresponding
definitions (and the axioms).
0.3.2. Definition of rotations. An oriented angle is an ordered

pair of rays with common origin O (notation ∠([O, A, [O, B), the
point O is the vertex of the oriented angle). Let α := ∠([O, A, [O, B)
be an oriented angle, let l and m be the lines OA and OB, respec-
tively; assume that l
= m; then the composition of reflections in the
lines l and m, Sl and Sm , performed in that order, will be called the
rotation with center O by the angle 2α and denoted by R2α . (The in-
teger 2 in the last expression is not a misprint, a glance at Figure 0.2
shows where it comes from.)
Figure 0.2. Rotation by 2α.
In the particular case in which the rotation is defined by an ori-

ented angle α = ∠([O, A, [O, B) such that the lines OA and OB
are perpendicular, the rotation with center O by the angle 2α will be
called the (central ) symmetry with center O and denoted by SO .
A circle of center O and radius r > 0 is the set of all points M
such that |OM | = r. If we are given a rotation by a certain angle
α = ∠([O, A, [O, B) and some point P
= O, its image P under
the rotation R2α obviously lies on the circle of center O and radius
|OP |, which justifies the term “rotation” (look at Figure 0.2 again).
Similarly, and just as obviously, the image of some point Q
= O under
the symmetry SO with center O can be constructed by choosing the
point Q on the line OQ so that |QO| = |OQ | and O lies between Q
and Q , which again justifies the terminology.
Proposition 0.3.3. Rotations and central symmetries, being compo-

sitions of isometries, are isometries, and therefore possess properties
(i)–(v).
Theorem 0.3.4. Vertical angles are congruent.

0.3. Rotations 9
This immediately follows by performing a symmetry centered at

the vertex of the angle.
If we are given two parallel but not coinciding lines and a line
cutting them both, then these three lines determine eight angles (see
Figure 0.3) that can be split into two quadruples in each of which all
four angles are congruent.
Figure 0.3. Congruent angles formed by a line cutting two parallels.
Let us formally introduce the notation appearing in Figure 0.3.

Let l = AB be a line, P a point not belonging to l, and l the parallel
to l passing through P ; let m denote the line AP (which cuts the
parallel lines l and l at the points A and P ) and let C be a point on
l in the same half-plane w.r.t. m as B; finally, let D be a point on l
such that P is between C and D, let E be a point on l such that A
is between E and B, let F be a point on m such that A is between F
and P , and let G be a point on m such that P is between G and A.
Theorem 0.3.5. In the situation described above, the following an-
gles are congruent:
∠DP A ∠P AB ∠EAF ∠GP C,
∠DP G ∠F AB ∠EAP ∠AP C.
The congruence ∠DP A ∠P AB follows by performing a sym-

metry with center at the midpoint O of [A, P ]. The congruence
∠DP A ∠GP C follows from Theorem 0.3.4 because these angles

are vertical. The other congruences are proved by the same argu-
ments.
The next assertion (about the sum of angles of a triangle) is one
of the most famous in Euclidean geometry. It is equivalent to Euclid’s
“Fifth Postulate” or to our Axiom V. A triangle ABC, where the
three points A, B, C (called vertices) do not lie on the same line,
is defined as the union of the three segments [A, B], [B, C], [C, A]
(called its sides); the angles ∠ABC, ∠BCA, ∠CAB are its (interior)
angles.
Corollary 0.3.6. The sum of the angles of a triangle is two right
angles.
The proof is immediate: if we draw a parallel to BC through A

and apply Theorem 0.3.4, we see (look at Figure 0.4) that the three
angles fit together neatly at the vertex A, forming two right angles.
Figure 0.4. Sum of the angles of a triangle.
Note that the sum here is understood as the geometric sum; later,
when the measure of angles will be introduced, we will be able to say
in this situation that the sum of measures is equal to 180 degrees or
π radians, depending on the choice of the unit of measure.
Corollary 0.3.7. If two distinct lines are both perpendicular to the
same line, then they are parallel.
This follows from the previous corollary by arguing by contradic-

tion.
0.4. Parallel translations and vectors 11
0.3.8. Parallelograms. Given four points A, B, C, D, no three of

which lie on the same line and such that any two of the segments
[A, B], [B, C], [C, D], [D, A] do not intersect except at common end-
points, the union of the four segments is called a parallelogram pro-
vided that AB CD and BC DA. The points A, B, C, D are said
to be the vertices of the parallelogram, the four segments listed above
are its sides, and the segments [A, C] and [B, D] are its diagonals.
It follows from Theorem 0.3.4 that the two pairs of opposite angles
of a parallelogram ABCD, namely ∠BAD and ∠BCD, ∠ABC and
∠ADC are congruent. It is also true that pairs of opposite sides are
also congruent, but it is more convenient to prove this later.
0.4. Parallel translations and vectors

0.4.1. Definition of parallel translations. The composition of
the reflections Sl and Sm , where (l, m) is an ordered pair of paral-
lel lines, determines an isometry called the parallel translation cor-
responding to the pair (l, m). In the case l = m, the corresponding
parallel translation is the identity. This is not the standard way of in-
troducing the notion of parallel translation, which as the reader surely
knows, is based on the notion of vector. But we have not introduced
the notion of vector yet, we do this in the next subsection.
0.4.2. Vectors. An ordered pair of points {O, A} is said to be a

−→
vector attached to O; we denote it by OA. Using the diagonal of a
parallelogram, it is easy to define the sum of two vectors attached
to the same point (obtaining a vector attached to the same point);
using the axiom of distance, it is easy to define the multiplication
of a vector by a scalar (i.e., by a real number), again obtaining a
vector attached to the same point. It is easy to prove that the set
of all vectors attached to a fixed point in the Euclidean plane is a
two-dimensional vector space. The details are left to the reader.
−→ −−→
Two vectors OA and O A are called equal or equivalent if OAO A
is a parallelogram or if one of the rays [O, A, [O , A contains the
other and |OA| = |O A |. Equality of vectors in an equivalence re-
lation (i.e., it is symmetric, reflexive and transitive) and therefore
the set of all (attached) vectors splits into equivalence classes; each
equivalence class
−→ −→ −−−→
v := OA | OA = O0 A0 ,
where O0 and A0 are given points, is called a free vector. It is easy

to define the sum of two free vectors and the multiplication of a free
vector by a scalar, and then to prove that the set of all free vectors
in the Euclidean plane is a two-dimensional vector space. Here again
the details are left to the reader.
0.4.3. Constructing parallel translations. Let T : E2 → E2 be

the parallel translation corresponding to the ordered pair of parallel
lines (l, l ). Let m be any perpendicular to l and let A and A be
the intersection points of m with l and l , respectively. We will call
the number |AA | the distance between the parallels l and l . Let us
prove that this number is well defined, i.e., it does not depend on the
choice of m. Let m be another perpendicular to l, intersecting l and
and l at B and B , respectively. Then it follows from Theorem 0.3.4
and Corollary 0.3.6 that m and m are both perpendicular to l and
parallel to each other. Now the symmetry with center O, where O is
the midpoint of the diagonal AB , takes [A, A ] to [B , B] (as can be
easily proved), and therefore |AA | = |B B|, as claimed. Now if we
take an arbitrary point P and denote by P = T (P ) its image under
the parallel translation corresponding to the ordered pair of parallel
lines (l, l ), it is easy to see that |P P | = 2|AA |.
We have just shown that any (free) vector v, in particular, the
−−→
free vector determined by the attached vector AA , defines the par-
allel translation Tv : E2 → E2 , the image Q of a point P being the
extremity of the given vector attached to P , i.e., Tv (P ) := Q, where
−−→
P Q ∈ v. This is the construction of parallel translations which is no
doubt familiar to the reader. From this construction it immediately
follows that the composition of two translations by the vectors v1 and
v2 is the translation by the vector v1 + v2 .
Theorem 0.4.4. Any parallel translation is an isometry. The com-

position of two translations is a translation, namely the translation
by the sum of the two vectors defining the given translations.
0.5. Triangles: congruence, properties 13
The first statement of the theorem follows from 0.3.1(vi), the

second one was just established above.
0.5. Triangles: congruence, properties

0.5.1. Congruence tests for triangles. In this subsection, using
the properties of parallel translations, rotations and reflections, we
obtain the three classical tests of the congruence of triangles. These
tests have different names: in continental Europe, they are simply
numbered (“first test”, “second test”, etc.), in the US the more infor-
mative notation SAS, ASA, AAA is used (for example, SAS stands
for side-angle-side). In all three cases, we consider two given trian-
gles: ABC and A B C , with the following notation for the angles
α = ∠BAC, β = ∠ABC, γ = ∠ACB of triangle ABC, and similar
notation α , β , γ for those of A B C .
Theorem 0.5.2 (SAS). If a given triangle has one angle and the two
sides limiting it congruent to an angle and to the sides limiting that
angle in a second triangle, then the two triangles are congruent.
In our notation, the assumptions of the theorem may be written

as
α α , [A, B] [A , B ], [A, C] [A , C ].
−−→
The parallel translation by the vector AA followed by an appropriate
rotation will send [A, B] onto [A , B ]; let C denote the image of C
under the composition of the above translation and rotation. Then
two cases can arise, depending on which half-plane with respect to
A B contains C . If C is in the same half-plane as C , then these two
points coincide and we are done. If not, we first perform a reflection in
the line A B , after which the image of C will coincide with C . Note
that here we are consistently using the fact that the transformations
considered are isometries.
Theorem 0.5.3 (ASA). If a given triangle has one side and the two
angles with vertices at its endpoints congruent to a side and to the
angles with vertices at the endpoints of that side in a second triangle,
then the two triangles are congruent.
The proof of this theorem, as well as that of the next one, is

similar to the proof of Theorem 0.5.2.
Theorem 0.5.4 (SSS). If a given triangle has three sides respectively
congruent to three sides of a second triangle, then the two triangles
are congruent.
Remark 0.5.5. There is no “fourth congruence test” (ASS): it is not
always true that if two triangles have one congruent angle and two
congruent sides, then they are congruent, as Figure 0.5 shows.
Figure 0.5. Noncongruent triangles.
Remark 0.5.6. There is another version of the congruence tests:

instead of assuming the congruence of sides and/or angles, one as-
sumes the equality of the lengths of the sides and the equality of the
measures of the angles. We cannot do this at this stage, because the
measure of angles has not been defined yet.
0.5.7. Special lines in triangles and special triangles. Given

an arbitrary triangle ABC, by drawing the perpendiculars from
the vertices to the opposite sides, we obtain three segments [A, H],
[B, I], [C, J], which are called the altitudes of the triangle. The lines
AR, BS, CT that divide the angles of ABC into pairs of congruent
angles are called the bisectors of ABC. The lines AM , BN , and
CP joining the vertices to the midpoints M , N , and P of the opposite
sides are its medians (see Figure 0.6).
In any triangle, the three altitudes intersect in one point, and so
do the medians and the bisectors; this will be proved later.
A triangle is called isosceles if two of its sides are congruent (these
two sides are called lateral, the third one is the base), and equilateral
if all three sides are congruent.
0.6. Homothety and similitude 15
Figure 0.6. Altitudes, bisectors, and medians of a triangle.
Theorem 0.5.8. In an isosceles triangle, the angles at the base are

congruent, and the median to the base coincides with the correspond-
ing altitude and bisector.
To prove this, one constructs the median [A, M ] to the base and
applies the third congruence test to ABM and ACM .
Theorem 0.5.9. In an equilateral triangle all three angles are con-
gruent, and all the medians coincide with the corresponding altitudes
and bisectors.
This follows from the previous theorem.
0.6. Homothety and similitude

0.6.1. Homothety. A homothety with center O and ratio ρ > 0 is
the transformation that assigns to each point P the extremity P of
the vector given by
−−→ −−→
OP := ρ · OP .
Figures obtained from each other by homothety are called homothetic.
A homothety of ratio ρ = 1 is the identity transformation; if ρ > 1,
then the size of the image of a figure increases but its shape remains
the same, and if ρ < 1, then the shape still does not change, but
the size decreases. The definition of homothety (and the axioms)
immediately imply the following properties of homothety.
Proposition 0.6.2. Any homothety
(i) takes lines to lines, segments to segments, rays to rays;
(ii) takes parallel lines to parallel lines;
(iii) takes perpendiculars to perpendiculars, angles to congruent

angles;
(iv) takes triangles to triangles;
(v) takes circles to circles.
0.6.3. Similitude. Any composition of isometries and homotheties

is called a similitude. Two figures obtained from each other by a
similitude are called similar. Similarity of figures corresponds to the
intuitive notion of having the same shape. The above definition im-
plies that any similitude possesses properties (i)–(v) listed in Sub-
section 0.3.1, that the composition of similitudes is a similitude, and
that any similitude possesses a ratio, i.e., a coefficient ρ > 0 such that
|P Q | = ρ · |P Q|, where P, Q are arbitrary points and P Q are their
images under the given similitude.
The following theorems can be called similarity tests for triangles.
Theorem 0.6.4. Two triangles ABC and A B C such that
|AB| |BC| |CA|
= = = const =: ρ
|A B | |B C | |C A |
are similar.
To prove this, one subjects triangle A B C to a homothety of

ratio ρ and arbitrary center, and then applies the third congruence
test for triangles (Theorem 0.5.4, SSS).
Theorem 0.6.5. If two triangles have all three angles congruent,
then they are similar.
Actually, it suffices to require that two angles be congruent, be-

cause then the third angles are congruent automatically (by Corollary
0.3.6). If the angles at the vertices A, B, C of ABC are, respectively,
congruent to the angles at the vertices A , B , C of A B C , then a
homothety with the ratio ρ := |AB|/|A B | followed by an application
of the second congruence test proves the theorem.
0.6.6. Right triangles. A triangle is said to be a right triangle if

one of its angles is a right angle; the side opposite to the right angle
is known as the hypotenuse, the other two sides are the legs. Since
0.6. Homothety and similitude 17
the sum of angles of a triangle is two right angles, it follows that

the other two angles are acute, i.e., their measure (see 0.7.2) is less
than π/2. The following statement is an immediate consequence of
Theorem 0.6.5.
Corollary 0.6.7. If a right triangle has an acute angle congruent to
an acute angle of another right triangle, then the two triangles are
similar.
We conclude this subsection with one of the most famous theo-

rems in geometry, attributed to Pythagoras, although it was known
to the Egyptians and Babylonians long before Pythagoras’ time.
Theorem 0.6.8 (Pythagoras, 6th century B.C.). In a right triangle,
the square of (length of ) the hypotenuse is equal to the sum of the
squares of (the lengths of ) the legs.
Figure 0.7. Pythagoras’ pants.
A popular proof of this theorem is shown in Figure 0.7. But this

proof (known to Russian students as “Pythagoras’ pants”) is based
on the notion of area, which we have not introduced yet. There are
many other proofs of this theorem; the reader will find 93 (!) proofs
in a web site easily accessed by googling.
Here we describe a simple proof based on similitude. Let ABC
be our triangle with right angle at C, angles α at A and β at B,
hypotenuse c and legs a and b. Drop the perpendicular CH from C
to AB, where H ∈ [A, B]; denote h := |CH|, denote c1 := |AH| and
c2 := |HB|. Then we obtain two right triangles similar to ABC, so
that we have b/c1 = c/b and a/c2 = c/a, whence a2 = cc2 and b2 =
cc1 . Adding these two equalities and using the fact that c = c1 + c2 ,
we obtain
a2 + b2 = c2 .
Figure 0.8. Proof of the Pythagoras theorem.
0.7. Angle measure and trigonometry

0.7.1. Measuring angles in degrees. The traditional angle mea-
sure comes from navigation, the unit of measure being the degree.
If we divide a circle into 360 congruent arcs and consider the angle
determined by two rays issuing from the center O of the circle and
passing through the endpoints of one of these arcs, we obtain a one
degree (1◦ ) angle; if we take, say, the endpoints of seventeen such suc-
cessive arcs A1 , A2 , . . . , A18 , then the measure of the angle ∠A1 OA18
will be 17◦ , and so on. Right angles measure 90◦ , so that 180◦ an-
gles, for which the rays lie on one line, are those that in the times
of Euclid were referred to as forming “two right angles”. Degrees are
0.7. Angle measure and trigonometry 19
subdivided into “minutes” and minutes into “seconds”, but we will

not need these finer measures.
0.7.2. Measuring angles in radians. A more modern and more

convenient unit of measure from the mathematical point of view is the
radian. Usually, it is defined in calculus courses, but here we define
the radian as the unit of angle measure proportional to the degree and
such that a 180◦ angle is the same as an angle of π radians. Mathe-
maticians prefer radians to degrees and usually say, for example, “this
angle equals π” rather than “this angle measures 180◦ ”.
0.7.3. Trigonometry of the triangle. In elementary geometry

(and in our exposition here) only angles of nonnegative measure less
than or equal to 180◦ are considered, and accordingly the trigono-
metric functions will be defined only for angles in triangles (the more
general case of trigonometric functions for arbitrary values of the ar-
gument is usually studied in calculus courses). Given a right triangle
HAB with hypotenuse [A, B] of length h and legs [H, A] and [H, B]
of length a and b, we define the trigonometric functions sine, cosine,
and tangent of the angle α = ∠BAH as the following ratios:
b a b
sin α := , cos α := , tan α := .
h h a
These definitions immediately imply that
sin α π
tan α = , cos − α = sin α, sin2 α + cos2 α = 1.
cos α 2
In an arbitrary triangle ABC (with angles α, β, γ and opposite
sides of lengths a, b, c), there are classical relationships between the
lengths of the sides and the sines of the angles (or their cosines).
Theorem 0.7.4 (Sine Theorem).

sin α sin β sin γ
= = .
a b c
To prove this, draw the perpendicular CH to AB, where H is
on the line AB; let h := |CH|; we obtain two right triangles which
(by the definition of sine) give sin α = h/b and sin β = h/a (see
Figure 0.9(a), which shows the case in which H ∈ [A, B]; in the
case shown in (b), the same equalities are obtained in a slightly dif-
ferent way). Simple manipulations with these two equalities yield
(sin α)/a = (sin β)/b. The second equality is proved similarly.
(a) (b)
Figure 0.9. Proving the sine and cosine theorems.
Theorem 0.7.5 (Cosine Theorem).

a2 = b2 + c2 − 2 b c cos α.
Using the same construction as in the previous proof, denote c1 :=

[A, H] and c2 := [H, B] (so that c = c1 +c2 ), and apply the Pythagoras
theorem to the two right triangles (Figure 0.9(a)), obtaining a2 =
c22 + h2 and b2 = c21 + h2 ; the definition of the cosine yields cos α =
c1 /b. Now simple manipulations with these three equalities give the
required result. In the case shown in Figure 0.9(b), the result is
obtained in a slightly different way.
0.8. Properties of the circle

Recall that the circle of center O and radius r > 0 is the set of all
points M such that |OM | = r. If A and B are points on the circle,
then the segment [A, B] is said to be a chord. If [A, B] contains the
center O, then O is obviously the midpoint of [A, B] and [A, B] is
called a diameter of the circle. If [A, B] is a chord and C is a point
on the circle, we say that the angle ACB subtends the chord [A, B].
0.8. Properties of the circle 21
Theorem 0.8.1. The line joining the midpoint of a chord to the

center of the circle is perpendicular to the chord.
Let M be the midpoint of the chord [A, B] (Figure 0.10(a)); then

the two triangles OAM and OBM are congruent by the third con-
gruence test SSS (Theorem 0.5.4), hence the two angles at M are
congruent and therefore both must be right.
Figure 0.10. Right angles in the circle.
Theorem 0.8.2. Any angle subtending a diameter of a circle is a

right angle.
Let [A, B] be a diameter and C a point on the circle (Figure

0.10(b)). Join C and O and denote by α and β the two angles formed
at O. Then α + β = π. Let γ be the measure of the two congruent
angles of the isosceles triangle OAC, and δ the same measure in tri-
angle OBC. Then α + 2γ = π and β + 2δ = π (by Corollary 0.3.6).
Extracting α and β from these two equalities and substituting into
α + β = π, we obtain γ + δ = π/2, as required.
A line is called tangent to a circle if it has only one common point

with the circle.
Theorem 0.8.3. Any line passing through a point on a circle and

perpendicular to the radius passing through that point is tangent to
the circle.
Suppose it isn’t, and let B be the intersection point of the line

with the circle other than A, the extremity of the given radius
(Figure 0.10(c)). Then triangle OAB is isosceles and so has right
angles both at A and at B, which means that its angle sum is greater
than two right angles, in contradiction with Corollary 0.3.6 (see Fig-
ure 0.12(b)).
Theorem 0.8.4. Suppose two lines pass through a point T outside

a circle centered at O and are tangent to it at the points A and B.
Then the segments [T, A] and [T B] are congruent, and the ray [T, O]
is the bisector of angle ∠AT B, i.e., the angles ∠AT O and ∠BT O are
congruent.
This readily follows from the congruence of triangles T AO and

T BO (see Figure 0.11).
Figure 0.11. Tangents to a circle.
A circle is said to be inscribed in a triangle if it is tangent to

all three of its sides, and circumscribed to it if it passes through the
triangle’s vertices.
Theorem 0.8.5. All three bisectors of any triangle intersect in one

common point, and this point is the center of the circle inscribed in
the triangle.
This immediately follows from Theorem 0.8.4 (see Figure 0.12(a)).

0.8. Properties of the circle 23
Theorem 0.8.6. All three perpendiculars to the sides of a triangle

that pass through the midpoints of the sides intersect in one common
point, and this point is the center of the circle circumscribed to the
triangle.
This immediately follows from Theorem 0.8.1.
(a) (b)
Figure 0.12. Inscribed and circumscribed circles.
Theorem 0.8.7. Congruent chords in a circle are subtended by con-

gruent angles.
Figure 0.13. Congruent angles in the circle.

Suppose D and C are points of the circle, AB is its chord, and

assume that D and C lie on the same side of line AB (this last assump-
tion is implicitly included in the statement of the theorem). Join the
four points A, B, C, D to the center O of the circle, thereby obtaining
five isosceles triangles, and denote their equal angles by α, β, γ, δ, ε,
as shown in Figure 0.13(a). Since the angle sum of triangles ABC
and ABD is π, we have
2α + 2β + 2ε = π, 2α + 2γ + 2δ = π.
Subtracting the second equality from the first and dividing by 2, we
obtain β + ε = γ + δ, as required.
Theorem 0.8.8. Let the angle ∠ACB subtend the chord [A, B] of
a circle, let BT be the tangent to the circle at B, and assume that T
is on the same side of the line AB as C (see Figure 0.13(b)). Then
the angles ∠ACB and ∠CBT are congruent.
By the previous theorem, we can assume, without loss of generality,

that AC is parallel to BT . Then the required congruence follows from
one of the properties of secants of two parallel lines (Theorem 0.3.4).
0.9. Isometries of the plane

Recall that an isometry of the plane is a transformation of the plane
that preserves distances. Any isometry is a bijection of the plane onto
itself. Examples of isometries that we have encountered are parallel
translations, rotations, and reflections in a line. Another type of
isometry is described in the next subsection.
0.9.1. Glide symmetries. By definition, a glide symmetry GF (l, → −

v)
is the composition of a reflection in the line l and the parallel trans-
lation by the vector −→
v , where −→
v is parallel to l. As can be seen
from the figure, glide symmetries reverse orientation. If −
→v = 0, then
→
−
of course the glide symmetry GF ( v ) is simply the reflection in the
line l.
The next theorem, the main theorem on the structure of isome-
tries of the plane, says that there are no isometries other than the
three mentioned above.
0.9. Isometries of the plane 25
Figure 0.14. Glide symmetry.
Theorem 0.9.2. Any isometry of the plane is a parallel translation,

or a rotation, or a glide symmetry.
Let OAB be an orthonormal frame (this means that OA is per-

pendicular to OB and |OA| = |OB| = 1); then the position of any
point M in the plane is determined by its Cartesian coordinates (x, y)
w.r.t. this frame. Denote by O A B the image of OAB and by M
the image of M under the given isometry. Then the point M has
the coordinates (x, y) w.r.t. O A B . This means that any isometry
is entirely determined by an orthonormal frame and its image.
So let O A B be the image of the orthonormal frame OAB. We
will consider several cases.
First we consider the case in which the lines OA and O A are
not parallel. Then they intersect at a point P . Construct the circle
circumscribed to OO P and denote by Q the intersection point of
this circle with the perpendicular to OO from its midpoint, where Q
lies on the same side of OO as P . Then the rotation by the angle
([QO, [QO ) takes O to O and A to A (because angles ∠QOP
and ∠QO P are congruent as subtending the same chord [Q, P ], see
Theorem 0.8.7).
−→ − −− →
Figure 0.15. OA ∦ O A : rotation and glide symmetry.
Now there are two subcases possible (see Figure 0.15(a) and (b)):
either this rotation takes B to B (in this case the theorem is proved,
because the rotation takes the given frame to its image by the given
isometry, and so that isometry is a rotation, namely the one just
constructed) or it takes B to B̂, where B̂ is the point symmetric to
B w.r.t. the line O A .
In the second subcase we perform a different construction. We
draw the bisector of ∠OP O through the point P and construct the
parallel l to this bisector through the point O . Let Ō be the foot
of the perpendicular drawn from O to l , let M be the midpoint of
[O, Ō]; denote by l the parallel to l through M . Then it is easy to
→ −−→
see that the glide symmetry l, − v , where − →
v = ŌO , takes the frame
OAB to O A B , which proves the theorem in this subcase as well.
Now we pass to the case in which the lines OA and O A are
parallel.
Denote
−→ −−→ −−→ −−−→
e1 := OA, e2 := OB, e1 := O A , e2 := O B .
Let us consider the following four subcases:
(1) the vectors of both pairs (e1 , e1 ) and (e2 , e2 ) point in opposite
directions;
−→ −
−−→
Figure 0.16. OA O A : rotation, glide symmetries or par-
allel translation.
(2) the vectors of the first pair point in opposite directions, and
those of the second pair, in the same direction;
(3) the vectors of the first pair point in the same direction, and
those of the second pair, in opposite ones;
(4) the vectors of both pairs point in the same directions.
In subcase (1), the rotation by π about the midpoint of [O, O ]

will take the frame OAB to O A B (Figure 0.16(a)), in subcases (2)
and (3) this is achieved by glide symmetries (Figure 0.16(b), (c)), and
−−→
finally in subcase (4), by the parallel translation by the vector OO
(Figure 0.16(d)).
0.10. Space geometry

In this subsection, we do not intend to develop Euclidean space geom-
etry (or even to summarize its main constructions and facts). We only
indicate a (very natural, but superfluous) system of axioms for Eu-
clidean space geometry and mention some facts related to isometries
in 3-space, omitting all proofs and not going into detail.
0.10.1. Axioms for Euclidean space geometry. There are four

types of undefined notions in our theory: points, (straight) lines,
planes, and (Euclidean) space. Points are denoted by various capi-
tal letters (A, B, C, . . . , P, Q, . . . , sometimes supplied with subscripts
or superscripts), lines are denoted by small italics (l, m, n, . . . , also
possibly endowed with subscripts or superscripts), planes, by capi-
tal script letters (P, Q, R, . . . ), while space is denoted by E3 . These
objects satisfy the following ten axioms.
IS . The space E3 is the set of all points.
IIS . All lines and all planes are nonempty subsets of the space E3 .
IIIS . The distance d(A, B) between any two points of space is

defined, and it satisfies the same four conditions (i)–(iv) as in plane
geometry.
IVS . Each plane is a Euclidean plane, i.e, points and lines in

it satisfy all the axioms of Euclidean plane geometry; the distance
function in all the planes is the same as the one in axiom IIIS above.
Note that the definition of parallel line used in axiom V of plane
geometry must be modified: two lines are called parallel in space if
they coincide or lie in the same plane and have no common points.
Two planes in space are called parallel if they coincide or have no
common points.
VS . Given a plane P and a point A, there exists a unique plane

containing A and parallel to P.
In a sense this axiom says that space is of dimension no greater
than 3; the next one says its dimension is at least 3.
0.10. Space geometry 29
VIS . For any plane there exists at least one point P ∈ E3 not
contained in that plane.
VIIS . There exists a plane passing through any three points, and
if these points are not contained in a line, this plane is unique.
VIIIS . If two distinct points lie in a plane, then the line passing
through them is contained in the plane.
Using these axioms, it is easy to prove that any two intersecting

lines determine a unique plane containing them. A line l is called
perpendicular to a plane P if it has a common point H with P and is
perpendicular to any line passing through H and contained in P.
IXS . For any plane P and any point A ∈ E3 , there exists a

unique perpendicular to l containing A.
Two points A and B are said to lie on the same side of a given
plane P if the segment [A, B] does not intersect P, and lie in opposite
sides if it does. Thus any plane P determines two open half-spaces,
each consisting of points lying on the same side of the plane, while any
two points from different half-spaces lie on opposite sides of P. (A
closed half-space is defined as the union of an open half-space with the
plane that bounds it.) A reflection in a plane P is the transformation
that takes any point A not in P to the point A that lies on the line
AH perpendicular to P, with H ∈ P, so that H is the midpoint of
AA , and takes any point P ∈ P to itself.
XS . The reflection in any plane is an isometry.
It follows immediately from the definition of reflection (and the

axioms) that any reflection in P interchanges the two half-planes de-
termined by P.
Remark. Actually, axioms IXS and XS can be derived from the

other axioms, in particular, from the very strong axiom IVS . We
have included them for the sake of brevity and in order to stress the
analogy between two-dimensional and three-dimensional Euclidean
geometry.
0.10.2. Convex polyhedra. A subset C of E3 (or of the plane E2 )

is called convex if A, B ∈ C implies that [A, B] ⊂ C. It immediately
follows from definitions that any open half-space is convex, and so is
any closed half-space (the same is true for half-planes).
Proposition 0.10.3. The intersection of any two convex sets is con-

vex.
The intersection of a finite number of closed half-spaces is said to

be a convex polyhedron provided that it is a bounded set and contains
interior points (i.e., points that are the centers of balls entirely lying
in it). The reader is undoubtedly familiar with such convex polyhedra
as the cube, the parallelepiped, various tetrahedra, certain prisms, and
has possibly heard of regular polyhedra, e.g., of the octahedron or the
dodecahedron.
0.10.4. Isometries in 3-space. Recall that an isometry of E3 is a

transformation of E3 that preserves distances. Any isometry is a bi-
jection of E3 . It can be proved that any isometry is the composition of
reflections. (In this subsection by reflection we always mean reflection
in a plane.) Let us list some important examples of isometries.
(i) Parallel translations. The composition τ of two reflections in
parallel planes P Q is said to be a parallel translation. Just as in the
planar case, each parallel translation is determined by a free vector,
−−→
namely the vector 2HK, where the line HK is perpendicular to the
reflection planes and H ∈ P, K ∈ Q.
(ii) Rotations about an axis. The composition ρ of two reflections
in intersecting planes P ∦ Q is said to be a rotation about an axis,
namely about the line l := P∩Q. It is easy to show that the restriction
of ρ to any plane R perpendicular to l is a rotation of that plane about
the point O := R ∩ l by an angle equal to twice the angle between
P and Q. (The definition of the (dihedral) angle between two planes
is left to the reader, as well as the determination of the direction of
rotation.)
A particular case of rotation about an axis is axial symmetry,
which is the composition of two reflections in perpendicular planes,
and may be described as follows: the image A of any point A is
0.10. Space geometry 31
obtained by constructing the perpendicular AH from A to the axis

of rotation l, l H, and extending [A, H] to [A, A ], where H is the
midpoint of [A, A ]. It can also be described as a rotation about l by
180◦ .
(iii) Central symmetries. A central symmetry σ with center C is
the transformation that takes any point A to the point A symmet-
ric to A with respect to C, i.e., to the point A such that C is the
midpoint of [A, A ]. Another way of looking at σ is to define it as
the composition of three reflections in three pairwise perpendicular
planes with common point O.
(iv) Helicoidal motions. A helicoidal motion χ is the composition
of a rotation about an axis and a parallel translation in the direction
of the axis. Thus χ is the composition of four reflections. A helicoidal
motion models the motion of a screw being screwed into a piece of
wood, and therefore is often encountered in mechanics.
Theorem 0.10.5. Any isometry of E3 is a composition of reflections.
The proof is not very difficult and is similar to that of Theo-

rem 0.9.2, i.e., it consists of a case by case study of the image of an
orthonormal frame under the given isometry.
−→ −−→ −−→
0.10.6. Orientation. Let (e1 , e2 , e3 ) = OA, OB, OC be an or-
thonormal frame, i.e., an ordered triple of pairwise perpendicular unit
vectors with common origin O. Another such frame (e1 , e2 , e3 ) has
the same orientation as (e1 , e2 , e3 ) if it can be taken to the other by
the composition of an even number of reflections, and the opposite
orientation if it can be taken to the other by an odd number of reflec-
tions. Thus (e1 , e2 , e3 ) and (e1 , e2 , −e3 ) have opposite orientations.
It is easy to see that the set of all orthonormal frames splits into
two equivalence classes with respect to the relation “to have the same
orientation”. The choice of one of these classes is called a choice of
orientation.
In physics courses (and sometimes in mathematics courses), the
notion of “positively oriented” frame is introduced, along with such
expressions as the “right-hand rule” (e.g. in electrodynamics). Actu-
ally, there is no mathematical way to choose canonically a “preferred”
orientation among the two existing ones, so that the expression “pos-
itively oriented frame” is mathematically meaningless (just as the ex-
pressions “rotation in the positive direction” or “clockwise rotation”
in plane geometry.)
Isometries of E3 that preserve orientation are called motions. Any
motion is the composition of an even number of reflections. Isometries
that do not preserve orientation are said to be orientation-reversing.
0.10.7. Compositions of isometries. The following properties of

the composition of specific types of isometries are not hard to prove:
(i) the composition of two parallel translations is a parallel trans-
lation;
(ii) the composition of two rotations about intersecting axes is a
rotation about an axis passing through the common point of the two
given axes;
(iii) the composition of two reflections is a parallel translation or
a rotation depending on whether the reflection planes are parallel or
not;
(iv) the composition of two central symmetries is a parallel trans-
lation;
(v) the composition of two motions is a motion;
(vi) the composition of two orientation-reversing isometries is a
motion.
In Chapters 1 and 3 of this book, we will be interested in isome-
tries related to various specific two- and three-dimensional objects
(such as the square, the cube, the regular tetrahedron), and we will
need to be able to find, very concretely, what the composition of
two given specific isometries actually is (for instance, find an effective
construction of the axis of rotation in the case of item (ii) above).
Chapter 1
Toy Geometries and

Main Definitions
In this chapter, we study five toy examples of geometries (symme-

tries of the equilateral triangle, the square, the cube, the circle and
the sphere) and a model of the geometry of the so-called elliptic plane.
These examples prepare us for the main definition (given in Sec. 1.4)
of this course: a geometry in the sense of Klein is a set with a transfor-
mation group acting on it. Before that, we present some useful gen-
eral notions related to transformation groups. Further, we study the
relationships (called morphisms or equivariant maps) between differ-
ent geometries, thus introducing the category of all geometries. The
notions introduced in this chapter are illustrated by some problems
(dealing with toy models of geometries) collected at the end of the
chapter.
But before we begin with these topics, we briefly recall some
terminology from elementary Euclidean geometry.
1.1. Isometries of the Euclidean plane and space

We assume that the reader is familiar with the basic notions and
facts of Euclidean geometry in the plane and in space (these notions
and facts are summarized in Chapter 0). One can think of Euclidean
geometry as an axiomatic theory (not too rigorously taught in high
33
34 1. Toy Geometries and Main Definitions
school) or as a small chapter of linear algebra (the plane R2 and the

space R3 supplied with the standard metric). It is irrelevant to us
which of these two points of view is adopted by the reader, and the
aim of this subsection is merely to fix some terminology.
An isometry of the Euclidean plane R2 (or space R3 ) is a map
f : R2 → R2 (respectively, f : R3 → R3 ) which preserves the distance
d between points, i.e.,
d(f (P ), f (Q)) = d(P, Q)
for any pair of points P, Q of the plane (respectively, of space). There

are two types of isometries: those which preserve orientation (they are
called motions, or sometimes rigid motions) and those that reverse
orientation (orientation-reversing isometries).
In the plane, examples of motions are parallel translations (deter-
mined by a fixed translation vector) and rotations (determined by a
pair (C, α), where C is the center of rotation and α is the oriented an-
gle of rotation. In space, examples of motions are parallel translations
and rotations (about an axis). Rotations in space are determined by
pairs (l, α), where l is the axis of rotation, i.e., a straight line with
a specified direction on it, and α is the angle of rotation. The rota-
tion (l, α) maps any point M in space to the point M obtained by
rotating M in the plane Π perpendicular to l and passing through
M by the angle α in a chosen direction. The direction of rotation
is of course the same in each plane parallel to Π; it must be either
clockwise or counterclockwise if one looks at the plane from “above”,
i.e., from some point of l obtained from the point l ∩ Π by moving in
the direction of the axis.
Examples of orientation-reversing isometries in the plane are re-
flections (i.e., symmetries with respect to a line). In space, examples
of orientation-reversing isometries are given by mirror symmetries
(i.e., reflections with respect to planes) and central symmetries (i.e.,
reflections with respect to a point).
All other isometries of the Euclidean plane and space are compo-
sitions of those listed above.
The reader who is uncomfortable with the notions appearing in
this subsection is invited to look at the relevant parts of Chapter 0.
1.2. Symmetries of some figures 35
1.2. Symmetries of some figures

1.2.1. Symmetries of the equilateral triangle. Consider all the
isometries of the equilateral triangle = ABC, i.e., all the distance-
preserving mappings of this triangle onto itself. (To be definite, we
assume that the letters A, B, C have been assigned to vertices in coun-
terclockwise order.) Denote by sA , sB , and sC the reflections in the
bisectors of angles A, B, C of the triangle. Denote by r0 , r1 , r2 the
counterclockwise rotations about its center of gravity by 0, 120, 240
degrees, respectively. Thus r1 takes the vertex A to B, B to C, and C
to A. These six transformations are all called symmetries of triangle
ABC and the set that they constitute is denoted by Sym(). Thus
Sym() = {r0 , r1 , r2 , sA , sB , sC }.
There are no other isometries of . Indeed, any isometry takes

vertices to vertices, and each one-to-one correspondence between ver-
tices entirely determines the isometry. (For example, the correspon-
dence A → B, B → A, C → C determines the reflection sC .) But
there are only six different ways to assign the letters A, B, C to three
points, so there cannot be more than 6 isometries of .
In a certain sense, Sym() is the same thing as the family of all
permutations of the three letters A, B, C; this remark will be made
precise in the next chapter.
We will use the symbol ∗ to denote the composition (or product)
of isometries, in particular, of elements of Sym(), and understand
expressions such as r1 ∗ sA to mean that r1 is performed first, and
then followed by sA . Obviously, when we compose two elements of
Sym(), we always obtain an element of Sym().
What element is the composition of two given ones can be easily
seen by drawing a picture of the triangle ABC and observing what
happens to it when the given isometries are successively performed,
but this can also be done without any pictures: it suffices to follow
the “trajectory” of the vertices A, B, C. Thus, in the example r1 ∗ sA ,
the rotation r1 takes the vertex A to B, and then B is taken to C by
the symmetry sA ; similarly, B → C → B and C → A → A, so that
the vertices A, B, C are taken to C, B, A in that order, which means
that r1 ∗ sA = sB .
The order in which symmetries are composed is important, be-

cause the resulting symmetry may change if we inverse the order.
Thus, in our example, sA ∗ r1 = sC
= sB (as the reader will read-
ily check), so that r1 ∗ sA
= sA ∗ r1 . So for elements of Sym(),
composition is noncommutative.
The compositions of all possible pairs of symmetries of can be
conveniently shown in the following multiplication table:
∗ r0 r1 r2 sA sB sC
r0 r0 r1 r2 sA sB sC
r1 r1 r2 r0 sB sC sA
r2 r2 r0 r1 sC sA sB
sA sA sC sB r0 r1 r2
sB sB sA sC r2 r0 r1
sC sC sA sB r1 r2 r0
Here (for instance) the element sV at the intersection of the fifth

column and the third row is sB = r1 ∗ sA , the composition of r1 and
sA in that order (first the transformation r1 is performed, then sA ).
As we noted above, composition is noncommutative, and this is
clearly seen from the table (it is not symmetric with respect to its
main diagonal).
The composition operation ∗ in Sym() is (obviously) associa-
tive, i.e., (i∗j)∗k = i∗(j ∗k) for all i, j, k ∈ Sym(). The set Sym()
contains the identity transformation r0 (also denoted id or 1). Any
element i of Sym() has an inverse i−1 , i.e., an element such that
i ∗ i−1 = i−1 ∗ i =1.
The set Sym() supplied with the composition operation ∗ is
called the symmetry group of the equilateral triangle.
1.2.2. Symmetries of the square. Consider all the isometries of

the unit square = ABCD, i.e., all the distance-preserving mappings
of the square to itself.
Let us denote by sH , sV , and sac , sbd the reflections in the hor-
izontal and vertical midlines, and in the diagonals AC, BD, respec-
tively. Denote by r0 , r1 , r2 , r3 the rotations about the center of the
s
s s
s
Figure 1.1. Symmetries of the square.
square by 0, 90, 180, 270 degrees, respectively. These eight transfor-

mations are all called symmetries of the square. We write
Sym() = {r0 , r1 , r2 , r3 , sH , sV , sac , sbd }.
Just as in the case of the equilateral triangle, the composition of

any two symmetries of the square is a symmetry of the square, and a
multiplication table, indicating the result of all pairwise compositions,
can be drawn up:
∗ r0 r1 r2 r3 sH sV sac sbd
r0 r0 r1 r2 r3 sH sV sac sbd
r1 r1 r2 r3 r0 sac sbd sV sH
r2 r2 r3 r0 r1 sV sH sbd sac
r3 r3 r0 r1 r2 sbd sac sH sV
sH sH sbd sV sac r0 r2 r3 r1
sV sV sac sH sbd r2 r0 r1 r3
sac sac sH sbd sV r1 r3 r0 r2
sbd sbd sV sac sH r3 r1 r2 r0
Here (for instance) the element sV at the intersection of the sixth

column and the fourth row is sV = r2 ∗ sH , the composition of r2 and
sH in that order (first the transformation r2 is performed, then sV ).
Composition is noncommutative.
Obviously, composition is associative. The set Sym() contains
the identity transformation r0 (also denoted id or 1). Any element i
of Sym() has an inverse i−1 , i.e., an element such that

i ∗ i−1 = i−1 ∗ i = 1.
The set Sym() supplied with the composition operation is called

the symmetry group of the square.
1.2.3. Symmetries of the cube. Let

I 3 = {(x, y, z) ∈ R3 | 0 ≤ x ≤ 1, 0 ≤ y ≤ 1, 0 ≤ z ≤ 1}
be the unit cube. A symmetry of the cube is defined as any isometric
mapping of I 3 onto itself. The composition of two symmetries (of I 3 )
is a symmetry. How many are there?
Let us first count the orientation-preserving isometries of the cube
(other than the identity), i.e., all its rotations (about an axis) by
nonzero angles that take the cube onto itself.
Figure 1.2. Rotations of the cube.
There are three axes of rotation joining the centers of opposite

faces, and the rotation angles about each are π/2, π, 3π/2. There
are four axes of rotation joining opposite vertices, the rotation angles
for each being 2π/3 and 4π/3. There are six axes of rotation joining
midpoints of opposite edges, with only one nonzero rotation for each
(by π). This gives us a total of (3 × 3) + (4 × 2) + (6 × 1) = 23
orientation-preserving isometries, or 24 if we include the identity.
There are no other orientation-preserving isometries; at this point,
we could prove this fact by a tedious elementary geometric counting
argument, but we postpone the proof to Chapter 3, where it will be
the immediate result of a more general and sophisticated algebraic

method.
There are also 24 orientation-reversing isometries of the cube.
Listing them all is the task prescribed by Problem 1.2 (see the end
of the chapter), a task which requires little more than a bit of spatial
intuition.
Thus the cube has 48 isometries. All their pairwise compositions
constitute a multiplication table, which is a 49 by 49 array of symbols,
much too unwieldy to fit on a book page.
The set Sym(I 3 ) of all 48 symmetries of the cube supplied with
the composition operation is called the symmetry group of the cube; it
is associative, noncommutative, has an identity, and all its elements
have inverses, just as the symmetry groups in the two previous exam-
ples.
1.2.4. Symmetries of the circle. Let
:= {(x, y) ∈ R2 | x2 + y 2 = 1}
be the unit circle. Denote by Sym() the set of all its isometries.
The elements of Sym() are of two types: the rotations rϕ about
the origin by angles ϕ, ϕ ∈ [0, 2π), and the reflections in lines pass-
ing through the origin, sα , α ∈ [0, π), where α denotes the angle
from the x-axis to the line (in the counterclockwise direction). The
composition of rotations is given by the (obvious) formula
rφ ∗ rψ = r(φ+ψ)mod 2π ,
where mod 2π means that we subtract 2π from the sum φ + ψ if the

latter is greater than or equal to 2π.
The composition of two reflections sα and sβ is a rotation by the
angle |α − β|,
sα ∗ sβ = r2|α−β| .
The interested reader will readily verify this formula by drawing a

picture and comparing the angles that will appear when the two re-
flections are composed.
The set of all isometries of the circle supplied with the composi-
tion operation is called the symmetry group of the circle and is de-
noted by Sym(). The group Sym() has an infinite number of
elements. As before, this group is associative, noncommutative, has
an identity, and all its elements have inverses.
1.2.5. Symmetries of the sphere. Let

S2 := {(x, y) ∈ R3 | x2 + y 2 + z 2 = 1}
be the unit sphere. Denote by Sym(S3 ) the set of all its isometries and
by Rot(S3 ) the set of all its rotations (by different angles about differ-
ent axes passing through the center of the sphere). Besides rotations,
the transformation group Sym(S3 ) contains reflections in different
planes passing through the center of the sphere, its symmetry with
respect to its center, and the composition of these transformations
with rotations.
Reflections in planes, unlike rotations, reverse the orientation of
the sphere. This means that a little circle oriented clockwise on the
sphere (if we are looking at it from the outside) is transformed by
any reflection into a counterclockwise oriented circle, and the picture
of a left hand drawn on the sphere becomes that of a right hand.
Now a reflection in a line passing through the sphere’s center does
not reverse orientation (unlike reflections with respect to a line in the
plane!) because a reflection of the sphere in a line is exactly the same
transformation as a rotation about this line by 180◦ . On the other
hand, a reflection of the sphere with respect to its center reverses its
orientation (again, this is not the case for reflections of the plane with
respect to a point).
Note that the composition of two reflections in planes is a rotation
(see Problem 1.11), while the composition of two rotations is another
rotation (by what angle and about what axis is the question discussed
in Problem 1.12).
The set of all isometries of the sphere supplied with the com-
position operation is called the symmetry group of the sphere and is
denoted by Sym(S3 ). The group Sym(S3 ) has an infinite number of
elements. As before, this group is associative, noncommutative, has
an identity, and all its elements have inverses.
1.3. Transformation groups 41
1.2.6. A model of elliptic plane geometry. Consider the set

Ant(S2 ) of all pairs of antipodal points (i.e., points symmetric with
respect to the origin) on the unit sphere S2 ; thus elements of Ant(S2 )
are not ordinary points, but pairs of points. Now consider the family
(that we denote O(3)) of all isometries of the space R3 that do not
move the origin.1 Clearly, any such isometry takes pairs of antipodal
points to pairs of antipodal points, thus it maps the set X = Ant(S2 )
to itself.
The family O(3) of transformations of the set Ant(S2 ) is called
the isometry group of the Riemannian elliptic plane. This is a much
more complicated object than the previous “toy geometries”. We will
come back to its study in Chapter 6.
1.3. Transformation groups

1.3.1. Definitions and notation. Let X be a set (finite or infinite)
of arbitrary elements called points. By definition, a transformation
group G acting on X is a (nonempty) set G of bijections of X sup-
plied with the composition operation ∗ and satisfying the following
conditions:
(i) G is closed under composition, i.e., for any transformations
g, g ∈ G, the composition g ∗ g belongs to G;

(ii) G is closed under taking inverses, i.e., for any transformation

g ∈ G, its inverse g −1 belongs to G.
These conditions immediately imply that G contains the identity
transformation. Indeed, take any g ∈ G; by (ii), we have g −1 ∈ G;
by (i), we have g −1 ∗ g ∈ G; but g −1 ∗ g =id (by definition of inverse
element), and so id∈ G. Note also that composition in G is associative
(because the composition of mappings is always associative).
If x ∈ X and g ∈ G, then by xg we denote the image of the
point x under the transformation g. (The more usual notation g(x)
is not convenient: we have x(g ∗ h) = (xg)h, but (g ∗ h)(x) = h(g(x)),
with g and h appearing in reverse order in the right-hand side of this
equality.)
1
In linear algebra courses such transformations are called orthogonal and O(3) is
called the orthogonal group.
1.3.2. Examples. The five toy geometries considered in the pre-

vious section all give examples of transformation groups. The five
transformation groups Sym act (by isometries) on the equilateral tri-
angle, the square, the cube, the circle, and the sphere, respectively.
In the last example (1.2.5), the orthogonal group O(3) acts on pairs of
antipodal points on the sphere, these pairs being regarded as “points”
of the “elliptic plane”.
More examples are given by the transformation group consisting
of all the bijections Bij(X) of any set X. By definition of transfor-
mation groups, Bij(X) is the largest (by inclusion) transformation
group acting on the given set X. At the other extreme, any set X
has a transformation group consisting of a single element, the identity
transformation.
When the set X is finite and consists of n objects, the group
Bij(X) of all its bijections is called the permutation group on n objects
and is denoted by Sn . This is one of the most fundamental notions of
mathematics, and plays a key role in abstract algebra, linear algebra,
and, as we shall see already in the next chapter, in geometry.
1.3.3. Orbits, stabilizers, class formula. Let (X : G) be some

transformation group acting on a set X and let x ∈ X. Then the
orbit of x is defined as
Orb(x) := {xg | g ∈ G} ⊂ X,
and the stabilizer of x is
St(x) := {g ∈ G | xg = x} ⊂ G.
For example, if X = R2 and G is the rotation group of the plane

about the origin, then the set of orbits consists of the origin and all
concentric circles centered at the origin; the stabilizer of the origin is
the whole group G, and the stabilizers of all the other points of R2
are trivial (i.e., they consist of one element – the identity id ∈ G).
Suppose (X : G) is an action of a finite transformation group G
on a finite set X. Then the number of points of G is (obviously) given
by
(1.1) |G| = |Orb(x)| × |St(x)|
for any x ∈ X. Now let A ⊂ X be a set that intersects each orbit at

exactly one point. Then the number of points of X is given by the
formula
|G|
(1.2) |X| = ,
|St(x)|
x∈A
called the class formula. This formula, just as the previous one, fol-
lows immediately from definitions.
1.3.4. Fundamental domains. If X is a subset of Rn (e.g. Rn

itself) and G is a transformation group acting on X, then a subset
F ⊂ X is called a fundamental domain of the action of G on X if
• F is an open set in X;
• F ∩ F g = ∅ for any g ∈ G (except g = id);

• X = g∈G Clos(F g), where Clos(.) denotes the closure of a

set.
For example, in the case of the square, a fundamental domain of

the action of Sym() is the interior of the triangle AOM , where O is
the center of the square and M is the midpoint of side AB; of course
Sym() has many other fundamental domains. Thus fundamental
domains are not necessarily unique. Moreover, fundamental domains
don’t always exist: for instance, Sym(S 1 ) (and other “continuous”
geometries) does not have any fundamental domains.
1.3.5. Morphisms. According to one of the main principles of the

category approach to mathematics, as soon as an important class of
objects is defined, one must define their morphisms, i.e., the natural
class of relationships between them. Following this principle, we say
that a mapping of transformation groups α : G → H is a homomor-
phism if α respects the product (composition) structure, i.e.,
(1.3) α(g1 ∗ g2 ) = α(g1 ) ∗ α(g2 ) for all g1 , g2 ∈ G.

Let us look at a few examples of homomorphisms:
(i) the mapping μ : Sym() → Sym(I 3 ) obtained by placing the

square on top of the cube and extending its isometries to the whole
cube in the natural way (e.g. assigning the rotation by 90◦ about the
vertical axis passing through the centers of the horizontal faces of the
cube to the 90◦ rotation of the square);
(ii) the mapping ν : Sym() → Sym() assigning to each rota-

tion of the triangle the rotation of the circle by the same angle and, to
the reflections sA , sB , sC , the reflections s0 , s2π/3 , s4π/3 of the circle;
(iii) the mapping π : Rot(I 3 ) → Rot() induced by the projection

of the cube on its bottom horizontal face Φ, i.e., assigning the identity
element to all isometries of the cube that do not map Φ to Φ, and
assigning, to all the other isometries of the cube, their restriction to
Φ;
(iv) the mapping ι : S3 →Sym() assigning to each permutation

of the symbols A, B, C the isometry that performs that permutation
of the vertices A, B, C of the triangle.
The proof of the fact that these mappings are indeed homomor-
phisms, i.e., relation (1.3) holds, is a straightforward verification left
to the reader.
A homomorphism α of transformation groups is said to be a
monomorphism if the mapping α is injective (i.e., takes different el-
ements to different ones). Examples of monomorphisms are the ho-
momorphisms μ and ν above. A homomorphism α of transformation
groups is said to be an epimorphism if α is surjective (i.e., is an onto
map). An example is the mapping π above. A homomorphism α of
transformation groups is said to be an isomorphism if it is both a
monomorphism and an epimorphism, i.e., if the mapping α is bijec-
tive.
Two transformation groups G and H acting on two sets X and Y
(the case X = Y is not forbidden) are called isomorphic if there exists
an isomorphism φ : G → H. If two isomorphic groups are finite, then
they necessarily have the same number of elements (but the number
of points in the sets on which they act can differ, as for example in the
case of the isomorphic groups Sym() and S3 ). Note in this context

that Sym() and S4 are not isomorphic, because the first of these
groups consists of 8 elements, while the second has 4! = 24.
1.3.6. Order. The order of a transformation group G is, by defini-

tion, the number of its elements; we denote it by |G|. Thus
|Sym()| = 6, |Sym()| = 8, |Sym()| = ∞.
The order of an element g of a transformation group G is, by

definition, the least positive integer k such that the element g∗g∗· · ·∗g
(k factors) is the identity; this integer is denoted by ord(g); if there
is no such integer, then g is said to be of infinite order. For example,
◦
√ rotation by 30 in Sym() has order 12, while the rotation by
the
2π
√ is of infinite order. (The last fact follows from the irrationality
of 2.)
1.3.7. Subgroups. Many important classes of objects have natu-

rally defined “subobjects” (e.g. spaces and subspaces, manifolds and
submanifolds, algebras and subalgebras). Transformation groups are
no exception: if G is a transformation group and H is a subset of G,
then H is called a subgroup of G if H itself is a group with respect to
the composition operation ∗, i.e., if it satisfies the two conditions:
(i) H is closed under composition, i.e., g, g ∈ H =⇒ g ∗ g ∈ H;
(ii) H is closed under taking inverses, i.e., g ∈ G =⇒ g −1 ∈ G.
According to this definition, any transformation group G has at
least two subgroups: G itself and its one-element subgroup, i.e., the
group {id} consisting of the identity element. We will call these two
subgroups trivial, and all the others, nontrivial.
For example, the subset of all rotations of the group Sym() is
a (nontrivial) subgroup of Sym() (of order 4), the set consisting of
the identity element and a reflection sα is a subgroup of order 2 in
Sym(), while the set of all rotations of Sym() is a subgroup of
infinite order.
If g is an element of order k in a transformation group G, then
the set of k elements{g, g ∗ g, . . . , g ∗ g ∗ · · · ∗ g = id} is a subgroup
of G of order k; it is called the cyclic subgroup of G generated by g.
This terminology is also used when g is an element of infinite order,

but then the subgroup {id, g, g ∗ g, . . . , g ∗ g ∗ · · · ∗ g, . . . } is also of
infinite order.
1.4. The category of geometries

In this section, we present the main definition of this course (that of
a geometry) and define some related basic concepts.
1.4.1. Geometries in the sense of Klein. A pair (X : G), where

X is a set and G is a transformation group acting on X will be called
a geometry in the sense of Klein. The six examples in Sec. 1.2 define
the geometry of the equilateral triangle, the geometry of the square,
the geometry of the cube, the geometry of the circle, the geometry
of the sphere, and the geometry of Riemann’s elliptic plane. Another
example is the set Bij(X) of all bijections of any set X.
1.4.2. The Erlangen program. The idea that geometries are sets
of objects with transformation groups acting on them was first stated
by the German mathematician Felix Klein in 1872 in a famous lecture
at Erlangen. In that lecture (for an English translation, see [10]), he
enunciated his views on geometries in the framework of what became
known as the “Erlangen program”.
There is no doubt that all the geometries known in the times of
Klein satisfy the property that he gave in his lecture, and so do all the
geometries that were developed since then. However, this property
can hardly be said to characterize geometries: it is much too broad.
Thus, in the sense of the formal definition from the previous subsec-
tion, the permutation group is a geometry, and so is any topological
space, any abstract group, even any set.
Nevertheless, we will stick to the notion of geometry given in 1.4.1
for want of a more precise formal definition. Such a definition, if it
existed, would require supplying (X : G) with additional structures
(besides the action of G on X), but it is unclear at this time what these
structures ought to be. If one looks at such branches of mathematics
as global differential geometry, geometric topology, and differential
1.4. The category of geometries 47
topology, there appears to be no consensus among the experts about

where geometry ends and topology begins in those fields.
The definition in 1.4.1 may be too broad, but it has the advantage
of being extremely succinct and leading to the definition of a very
natural category.
1.4.3. Morphisms. According to the general philosophy underly-

ing the category language, a morphism from one geometry to another
should be defined as a mapping of the set of points of one geome-
try to the set of points of the other that respects the actions of the
corresponding transformation groups. More precisely, given two ge-
ometries (G : X) and (H : Y ), a morphism (or an equivariant map)
is any pair (α, f ) consisting of a homomorphism of transformation
groups α : G → H and a mapping of sets f : X → Y such that
(1.4) f (xg) = (f (x))(α(g))
for all x ∈ X and all g ∈ G. This definition is typical of the category

approach in mathematics: at first glance, the boxed formula makes no
sense at all (no wonder category theory is called abstract nonsense),
but actually the definition is perfectly natural.
To see this, let us take any point x ∈ X and let an arbitrary
transformation g ∈ G act on x, taking it to the point xg ∈ X. Under
the map f : X → Y , the point x is taken to the point f (x) ∈ Y and
the point xg is taken to the point f (xg) ∈ Y . How are these two points
related? What transformation (if any) takes f (x) to f (xg)? Clearly,
if the pair of maps (f, α) respects the action of the transformation
groups in X and Y , it must be none other than α(g), and this is
precisely what the boxed formula says.
To check that the reader has really understood this definition, we
suggest that she/he prove that α(1)=1 for any morphism (f, α).
1.4.4. Isomorphic geometries. In any mathematical theory, iso-

morphic objects are those which are equivalent, i.e., are not distin-
guished in the theory. Thus isomorphic linear spaces are not dis-
tinguished in linear algebra, sets of the same cardinality (i.e., sets
for which there exists a bijective map) are equivalent in set theory,
isomorphic fields are not distinguished in abstract algebra, congru-

ent triangles are the same in Euclidean plane geometry, and so on.
What geometries should be considered equivalent? We hope that the
following definition will seem natural to the reader.
Two geometries (X : G) and (Y : H) are called isomorphic, if
there exist a bijection f : X → Y and an isomorphism α : G → H
such that
f (xg) = (f (x))(α(g)) for all x ∈ X and all g ∈ G.
In the definition, the displayed formula is a repetition of relation (1.4),

so it expresses the requirement that an isomorphism be a morphism
(must satisfy the equivariance condition, i.e., respect the action of
the transformation groups), the conditions on α and f say that they
are equivalences, so what this definition is saying is that (X : G) and
(Y : H) are the same.
At this stage we have no meaningful examples of isomorphic ge-
ometries. They will abound in what follows. For instance, we will
see (in Chapter 10) that the Poincaré half-plane model of hyperbolic
geometry is isomorphic to the Cayley–Klein disk model.
1.4.5. Subgeometries. What are subobjects in the category of ge-

ometries? The reader who is acquiring a feel for the category language
should have no difficulty in coming up with the following definition. A
geometry (G : X) is said to be a subgeometry of the geometry (H : Y )
if X is a subset of Y , G is a subgroup of H, and the pair (idX , idG ),
where idG : g → g ∈ H and idX : x → x ∈ Y are the identities, is a
morphism of geometries.
A closely related definition is the following. An embedding (or
injective morphism) of the geometry (X : G) to the geometry (Y : H)
is a morphism (f, α) such that α : G → H is a monomorphism and
f : X → Y is injective.
Examples of subgeometries and embeddings of geometries can
easily be deduced from the examples of subgroups of transformation
groups in Subsection 1.3.4.
1.5. Some philosophical remarks 49
1.5. Some philosophical remarks

The examples in Section 1.2 (square, cube, circle) were taken from
elementary school geometry. This was done to motivate the choice of
the action of the corresponding transformation group. But now, in
the example of the cube, let us forget school geometry: instead of the
cube I 3 with its vertices, edges, faces, angles, interior points and other
structure, consider the abstract set of points {A, B, C, D, A , B , C , D }
and define the “isometries” of this “cube” as a set of 48 bijections;
for example, the “rotation by 270 degrees” about the vertical axis is
the bijection
A → B, B → C, C → D, D → A,
A → B , B → C , D → D , D → A ,

and the 47 other “isometries” are defined similarly. Then (still forget-
ting school geometry), we can define vertices, edges (AB is an edge,
but AC is not), faces, prove that all edges are congruent, all faces
are congruent, the “cube” can “rotate” about each vertex, etc.). The
result is the intrinsic geometry of the set of vertices of the cube.
This geometry is not the same as the geometry (I 3 : Sym(I 3 ))
of the cube described in Subsection 1.2.3. Of course the group G
acting in these two geometries is the same group of order 48, but it
acts on two different sets: the (infinite) set of points of the cube I 3
and the (finite) set of its 8 vertices A, B, C, D, A , B , C , D . Thus
the algebra of the two situations is the same, but the geometry is
different. The geometry of the solid cube I 3 is of course much richer
than the geometry of the vertex set of the cube. For example, we can
define line segments inside the cube, establish their congruence, etc.
Note also that the geometric properties of the cube I 3 regarded as
a subset of Euclidean space R3 are richer than its properties coming
from its own geometry (I 3 : Sym(I 3 )), e.g., segments of the same
length inside the cube, which are always congruent in the geometry
of R3 , don’t have to be congruent in the geometry of the cube!
Another example: the set of three points {A, B, C} with two
transformations, namely the identity and the “reflection”
A → A, B → C, C → B,
is of course a geometry in the sense of Klein. What should it be

called? An appropriate title, as the reader will no doubt agree, is
“the intrinsic geometry of the vertex set of the isosceles triangle”.
1.6. Problems
1.1. List all the elements (indicating their orders) of the symmetry
group (i.e., isometry group) of the equilateral triangle. List all its
subgroups. How many elements are there in the group of motions
(i.e., orientation-preserving isometries) of the equilateral triangle?
1.2. Answer the same questions as in Problem 1.1 for
(a) the regular pyramid with four lateral faces;
(b) the regular tetrahedron;
(c) the cube;
(d)* the dodecahedron; 2
(e)* the icosahedron;
(f) the regular n-gon (i.e., the regular polygon of n sides); consider
the cases of odd and even n separately.
1.3. Embed the geometry of the motion group of the square into the
geometry of the motion group of the cube, and the geometry of the
circle into the geometry of the sphere.
1.4. For what values of n and m can the geometry of the regular
n-gon be embedded in the geometry of the regular m-gon?
1.5. Let G be the symmetry group of the regular tetrahedron. Find
all its subgroups of order 2 and describe their action geometrically.
1.6. Let G+ be the group of motions of the cube. Indicate four
subsets of the cube on which G+ acts by all possible permutations.
1.7. Find a minimal system of generators (i.e., a set of elements such
that any element can be represented as the product of some elements
from this set) for the symmetry group of
(a) the regular tetrahedron;
(b) the cube.
2
Here and in what follows the asterisk after a problem number means that the
problem is not easy and should be regarded as a challenge.
1.6. Problems 51
1.8. Describe the fundamental domains of the symmetry group of

(a) the cube;
(b) the icosahedron;
(c) the regular tetrahedron.
1.9. Describe the Möbius band as a subset of RP 2 .
1.10. Show that the composition of two reflections of the sphere in
planes passing through its center is a rotation. Determine the axis
of rotation and, if the angle between the planes is given, the angle of
rotation.
1.11. Given two rotations of the sphere, describe their composition.
Chapter 2
Abstract Groups and

Group Presentations
In order to study geometries more complicated than the toy models

with which we played in the previous chapter, we need to know much
more about group theory. In this chapter we present the relevant
facts of this theory (they will constantly be used in what follows).
The theory of transformation groups began in the work of sev-
eral great mathematicians: Lagrange, Abel, Galois, Sophus Lie, Felix
Klein, Élie Cartan, Hermann Weyl. At the beginning of the 20th cen-
tury, algebraists decided to generalize this theory to the formal theory
of abstract groups. In this chapter, we will study this formal theory
and learn that it is not a generalization at all: Cayley’s Theorem
(which concludes this chapter) says that all abstract groups are ac-
tually transformation groups. We will also learn that two important
classes of groups (free groups and permutation groups) have certain
universality properties. Finally, we will learn about group presenta-
tions, which allow us to replace computations in groups by games
with words.
2.1. Abstract groups

2.1.1. Groups: definition and manipulation. By definition, an
(abstract) group is a set G of arbitrary elements supplied with a binary
53
54 2. Abstract Groups and Group Presentations
operation ∗ (usually called multiplication) if it obeys the following

rules:
• (neutral element axiom) there exists a unique element e ∈ G
called neutral such that g ∗ e = e ∗ g = g for any g ∈ G;
• (inverse element axiom) for any element g ∈ G there exists
a unique element in G, denoted g −1 and called inverse to g,
such that g ∗ g −1 = g −1 ∗ g = e;
• (associativity axiom) (g ∗h)∗k = g ∗(h∗k) for all g, h, k ∈ G.
A group (G, ∗) is called commutative or Abelian if g ∗ h = h ∗ g
for all g, h ∈ G (in that case the operation is usually called a sum and
the inverse element is usually denoted by −g instead of g −1 ).
Note that the elements of an abstract group can be objects of
any nature, they are not necessarily bijections of something and the
operation ∗ is not necessarily composition, while the notation g −1 for
inverse elements is purely formal, it does not mean that g −1 is the
inverse of a bijection.
The three axioms for groups listed above are much stronger than
necessary. For example, the uniqueness condition in the inverse el-
ement axiom can be omitted without changing the class of objects
defined by these axioms. The definition can be weakened further, but
this is not an important fact from the point of view of geometry, so
we do not dwell on it further.
The group axioms have some obvious consequences that are use-
ful when performing calculations with elements of groups. In these
calculations and further on, we omit the group operation symbol, i.e.,
we write gh instead of g ∗ h.
The first immediate consequence of the group axioms are the left
and right cancellation rules, which say that one can cancel equal terms
on the two sides of an equation, provided they both appear at the left
(or at the right) of the corresponding expression, i.e.,
∀g, h, k ∈ G gh = gk ⇐⇒ h = k, hg = kg ⇐⇒ h = k.
The implications in these formulas are two-sided; reading them from
right to left, we can say that one can multiply both sides of an equa-
tion by the same element from the same side. The phrase in italics is
2.1. Abstract groups 55
of course important, because for non-Abelian groups the cancellation

of equal terms on different sides of an equation can result in a false
statement.
Another simple but important consequence of the axioms is the
rule for solving linear equations, i.e.,
∀g, h, x ∈ G gx = h ⇐⇒ x = g −1 h, xg = h ⇐⇒ x = hg −1 ,
which are proved by multiplying both sides by the element g −1 (it

exists by the inverse element axiom) from the left and the right, re-
spectively, using associativity and the neutral element axiom.
These two rules are constantly used when performing manipula-
tions with equations in groups, as the reader will see in solving some
of the exercises at the end of this chapter.
2.1.2. Examples of groups. It is easy to see that any transforma-

tion group is a group. Indeed, the three axioms of abstract groups
listed above, although they do not appear explicitly in the definition
of transformation groups, hold automatically for the latter, because
their elements are not arbitrary objects, they are bijections, and the
multiplication operation is not arbitrary (it is composition): for them
associativity and the neutral element axiom hold automatically.
Here are some other important examples of groups.
(i) The standard numerical groups: the integers under addition

(Z, +), as well as the rational, real, and complex numbers under addi-
tion (Q, +), (R, +), and (C, +); the nonzero rational, real, and com-
plex numbers under multiplication (Q \ {0}, ×), (R \ {0}, ×), and
(C \ {0}, ×). Note that the nonzero integers under multiplication are
not a group (no inverse elements!), neither are the natural numbers N
under addition (for the same reason). Another nice numerical group
is formed by the unimodular complex numbers under multiplication
S1 := {z ∈ C : |z| = 1}.
(ii) The group of residues modulo m, (Zm , ⊕) (also known as the

m-element cyclic group); its elements are the m infinite sets of integers
that have the same remainder under division by the natural number
m; we denote these sets by 0, 1, . . . , m − 1; their sum ⊕ is defined by
i ⊕ j := (i + j)mod m,
where (·)mod m stands for the remainder under division by m. The

sum operation ⊕ is well defined, i.e., does not depend on the choice
of the representatives i and j in the classes i and j. Indeed, if we take
i + rm instead of i and j + sm instead of j, then
(i + rm) + (j + sm) = (i + j + (r + s)m) = (i + j).
(iii) The group of permutations of n objects Sn : its elements are

bijections of a set of n elements that we denote by natural numbers
({1, 2, . . . n}); we will write bijections s ∈ Sn in the form
s = [i1 , i2 , . . . , in ], where i1 = s(1), i2 = s(2), . . . , in = s(n);
multiplication in Sn is the composition of bijections. This group is

extremely important not only in geometry, but also in linear algebra,
combinatorics, representation theory, mathematical physics, etc. We
will come back to permutation groups later in this chapter.
(iv) The free group Fn = F(a1 , . . . , an ) on n generators; its el-

ements are equivalence classes of words and the group operation is
concatenation; a detailed definition of Fn appears in Subsection 2.6.2
below.
(v) The group GL(n) of nonsingular linear operators on Rn ; its

elements are n by n matrices with nonzero determinant and the group
operation is matrix multiplication (or, which is the same thing, com-
position of operators).
(vi) The groups of orthogonal and special orthogonal operators

on Rn , standardly denoted by O(n) and SO(n). We assume that
the reader is familiar with the groups GL(n), O(n), and SO(n) at
least for n = 2 and n = 3; if this is not the case, he/she is referred
to any introductory linear algebra course. In what follows, we will
only need the low-dimensional case (n = 2, 3), and when we do, the
corresponding definitions will be given.
2.2. Morphisms of Groups 57
2.1.3. Order of a group and of its elements, generators. The

notions of order (of elements of a group and of the group itself) and of
generator for abstract groups are defined exactly as for transformation
groups (see Section 1.3). In this book, |G| denotes the order of the
group G (i.e., the number of its elements), ord(g) denotes the order
of the element g ∈ G, i.e., the least natural number k such that
g k = e. For example: |Z5 | = 5; |Sym()| = ∞; for 3 ∈ Z15 , we have
ord(3) = 5; for any nonzero real number x in the additive group R,
we have ord(x) = ∞.
A family of generators of a group G is a (finite or infinite) set
of its elements {g1 , g2 , . . . } in terms of which any element g of G
can be expressed, i.e., written in the form g = g1ε1 . . . gkεk , where
the εi ’s equal ±1 and gi+1 stands for gi . For example, any nonzero
element of Zp , where p is prime, constitutes a (one-element) family
of generators for Zp , while Sym() does not have a finite family of
generators. If g is an element of order m of a group G, then the set
{g, g 2 , . . . , g m−1 , g m = e} is also a group (it is a “subgroup” of G, see
the definition in 2.3.1), and its order is m. This justifies the use of
the same term “order” for groups and their elements, i.e., for notions
that seem very different at first glance.
2.2. Morphisms of Groups

In accordance with the traditions of the category language, as soon
as we have defined an interesting class of objects, in this case groups,
we should define their morphisms.
2.2.1. Definitions. Suppose (G, ∗) and (H, ) are groups; a map-

ping φ : G → H is called a homomorphism (or a morphism of groups)
if it respects the operations, i.e.,
ϕ(g1 ∗ g2 ) = ϕ(g1 ) φ(g2 ).
Thus, the inclusion Z → R, n → n, is a morphism, while the inclusion

(Q \ {0}, ×) → (Q, +) is not (the operations are not respected, e.g.,
2 × 3
= 2 + 3).
By definition, a homomorphism ϕ is a monomorphism (respec-

tively, an epimorphism or an isomorphism) if the mapping ϕ is injec-
tive (resp., surjective or bijective). From the point of view of abstract
algebra, isomorphic groups are identical.
2.2.2. Examples. The group Sym() of isometries of the equilat-

eral triangle is isomorphic to the permutation group S3 , the group
Sym() is isomorphic to SO(2), the group of motions of the Eu-
clidean plane R2 preserving the origin; there are obvious monomor-
phisms of the rotation group Rot() into SO(2) and of Z3 into Z15 ;
there is an equally obvious epimorphism of Z onto Z17 .
2.3. Subgroups
Worthwhile mathematical objects should not only be related by mor-
phisms, they should have naturally defined subobjects.
2.3.1. Definitions and examples. A subgroup H of a group G is

a subset of G which satisfies the group axioms. Note that in order
to check that H is a subgroup of G, it is not necessary to verify
all the group axioms; it suffices to check that H is closed under the
group operation and under taking inverses. Any group G has at least
two subgroups: the one-element subgroup consisting of the neutral
element e ∈ G and the group G itself. These two subgroups are
sometimes called trivial, and of course in the study of the structure
of groups we are interested in nontrivial subgroups.
Examples: Rot() is a subgroup of the group Sym(), the set
{[1234], [2134]} is a subgroup of Sn , the set {0, 5, 10} is a subgroup of
Z15 , while {0, 5, 11} is not.
2.3.2. Partition of a group into cosets. If H is a subgroup of G,

then the (left) coset gH ⊂ G, for some g ∈ G, is the set of all elements
of the form gh for h ∈ H. Right cosets Hg are defined similarly. Right
cosets as well as left cosets form a partition of the set of elements of
a group, i.e., two cosets either do not intersect or coincide.
To prove this, it suffices to show that if two cosets have a common
element g ∈ g1 H ∩ g2 H, then any element of g1 H belongs to g2 H and
vice versa. So suppose that g ∈ g1 H (which means that g = g1 h
2.4. The Lagrange theorem 59
for some
h ∈ H); we must show that g ∈ g2 H, i.e., we must find an
hx ∈ H such that g = g2 hx .
Since g ∈ g1 H ∩ g2 H, there exist elements h1 , h2 ∈ H for which
we have g1 h1 = g = g2 h2 , which implies that g1 = g2 h2 (h1 )−1 . Now
we can write
−1
−1

g = g1
h = g2 h2 (h1 ) h = g2 h2 (h1 ) h = g2 h x ,
−1
where we have defined hx as h2 (h1 ) h, and since hx belongs to H
(as the product of elements of H), we have proved the implication g ∈
g1 H =⇒ g ∈ g2 H. The reverse implication is proved by a symmetric
argument (interchange the indices 1 and 2).
Thus we have obtained the partition of G into left cosets. The
partition into right cosets is obtained similarly.
Note also that all cosets have the same number of elements (finite
or infinite), because there is an obvious bijection between any coset
and the subgroup H. This bijection for left cosets is given by the rule
gH gh → h ∈ H.
2.4. The Lagrange theorem

The corollary to the elementary theorem proved below is the first
structure theorem about abstract groups. It was proved (for trans-
formation groups) almost two centuries ago by Lagrange.
Theorem 2.4.1. If H is a subgroup of a finite group G, then the

order of H divides the order of G.
Proof. The cosets of H in G form a partition of the set of elements

of G (see 2.3.2) and all have the same number of elements as H.
Corollary 2.4.2. Any group G of prime order p is isomorphic to Zp .
Proof. Let g ∈ G, g
= e. Let m be the smallest positive integer such
that g m = e. Then it is easy to see that H := {e, g, g 2 , . . . , g m−1 } is
a subgroup of G. By Theorem 2.4.1, m divides p. This is impossible
unless m = p, but then H = G is obviously isomorphic to Zp .
2.5. Quotient groups

Nice mathematical objects often have naturally defined “quotient ob-
jects” obtained by “dividing out” the given object by some subobject
(examples that may be familiar to the reader are quotient spaces in
linear algebra). The construction of “quotient groups” along those
lines works only when the subgroup used is in a sense “nice”, and we
begin by defining such subgroups.
2.5.1. Normal subgroups. A subgroup H ⊂ G is normal if we

have gHg −1 = H for any g ∈ G, i.e., for any h ∈ H and any g ∈ G
the inclusion g −1 hg ∈ H holds.
An example of a normal subgroup is the set {0, 5, 10} in Z15 . More
generally, any subgroup of an Abelian group is (obviously!) normal.
To see an example of a subgroup which is not normal, consider
the subset D := {e = [1, 2, 3, 4], [2, 1, 3, 4]} in the permutation group
S4 . The set D is obviously a subgroup (isomorphic to Z2 ) of S4 , but
it is not normal, because
[4, 1, 2, 3] [2, 1, 3, 4] [2, 3, 4, 1] = [1, 3, 2, 4] ∈
/ D.
2.5.2. Construction of quotient groups. If H is a normal sub-

group of G, there is a well-defined operation in the family of cosets:
the product of two cosets is the coset containing the product of any
two elements of these cosets. For left cosets this may be written as
(g1 )H (g2 )H := (g1 g2 )H.
To prove that this is an operation well-defined on cosets, we must
show that if we replace g1 by another element g 1 from Hg1 and re-
place g2 by another element g 2 from Hg2 , then g 1 H g 2 H = g1 H g2 H.
Without loss of generality, it suffices to consider the case in which only
one of the two elements is replaced, say g1 . Then we have g 1 = g1 h1
for some h1 ∈ H. We must prove that g 1 g2 ∈ g1 g2 H, i.e., that there
exists an hx such that g 1 g2 = g1 g2 hx . Replacing g 1 by its expression
g1 h1 (see above), we can rewrite the previous equation as
g1 h1 g2 = g1 g2 hx .
Solving this (linear) equation for hx , we obtain hx = g2−1 h1 g2 . Re-
call that H is normal, therefore the right-hand side of the previous
2.6. Free groups and permutations 61
equality is an element of H. Thus we have found the required element

hx ∈ H, thereby proving that the product of cosets is well defined.
The family of cosets supplied with this product operation is called
the quotient group of G by H and is denoted by G/H. It is easy to
show that G/H satisfies the axioms for groups.
Example: in the additive group of integers (Z, +), elements of
the form 5k, k ∈ Z, constitute a normal subgroup (of infinite order),
denoted 5Z; the corresponding quotient group Z/5Z is isomorphic to
the group Z5 .
2.6. Free groups and permutations

In this section, we study two classes of groups: the free groups (which
have the “least structure”) and the permutation groups (which have
the “most structure”).
2.6.1. Free groups. Let {a1 , . . . , ak } be a set of symbols. Then the

set of formal symbols (called letters)
A := {e, a1 , . . . , ak , a−1 −1
1 , . . . , ak }
will be our alphabet. A string of letters from our alphabet will be

called a word. Two words w1 and w2 are called equivalent, if one can
be obtained from the other by using the following trivial relations:
ai a−1
i = a−1
i ai = e for any i and ae = ea = a for any a ∈ A; for
example,
a1 a−1 −1 −1 −1 −1 −1
3 ∼ a1 a3 e ∼ a1 a3 a2 a2 ∼ a1 a3 a2 ea2 .
The product of two words is defined as their concatenation (i.e., the

result of writing them one after the other). The free group with
generators a1 , . . . , ak is defined as the set of equivalence classes of
words supplied with the product (concatenation) operation and is
denoted by Fk = F[a1 , . . . , ak ]. The fact that concatenation is well
defined on equivalence classes (i.e., the concatenation of equivalent
elements produces an element from the same equivalence class) is
obtained by an straightforward verification, which we omit.
For example, F[a] is isomorphic to (Z, +), while F[a1 , a2 ] is not
commutative.
2.6.2. Permutation groups. The permutation group Sn on n ob-

jects was defined in 2.1.2 as the family of all bijections of the set
{1, 2, . . . , n} supplied with the operation of composition; Sn consists
of n! = 1 · 2 · · · · · n elements denoted by [i1 , . . . , in ], where ik := β(k)
and β is the bijection defining the given permutation.
Geometrically, the permutation group S3 can be interpreted as
the isometry group Sym() of the equilateral triangle, while S4 is
isomorphic to the isometry group of the regular tetrahedron (as we
shall see in the next chapter).
2.6.3. Universality theorem. It turns out that permutation groups

and free groups have important “universality” properties.
Theorem 2.6.4. (i) For any finite group G there exists a monomor-
phism of G into Sn for some n.
(ii) For any group G with a finite number n of generators there
exists an epimorphism of the free group Fn onto G.
Proof. (i) Let |G| = n and g0 ∈ G; then the mapping

βg 0 : G → Sn given by G g → gg0 ∈ G
is a monomorphism. Indeed, it is obviously a homomorphism (indeed,
we have βg0 βg1 = βg0 g1 , because both maps are given by the rule
g → gg0 g1 ). The homomorphism βg0 is injective, because g0 g = g0 g
implies g = g by the cancellation rule.
(ii) Let g1 , . . . , gn be a set of generators of G. Then the mapping
α : F[a1 , . . . , an ] → G given by α(ai ) = gi , i = 1, . . . , n,
is obviously a homomorphism. It is also surjective, because to each el-
ement giε11 giε22 . . . giεmm ∈ G, where the εi ’s are equal to ±1, the mapping
α takes the element aεi11 aεi22 . . . aεim
m
∈ F[a1 , . . . , an ].
2.7. Group presentations

A presentation of a group is a way of defining the group by means of
equations (called defining relations) in the generators of the group.
This reduces concrete calculations in the group to the formal editing
of words according to simple rules. The formal definition of the notion
2.7. Group presentations 63
of group presentation is easy to state but perhaps difficult to grasp,

so we begin with some examples.
2.7.1. Examples of group presentations. (i) Consider all words

in the three-letter alphabet {e, a, a−1 }, i.e., expressions such as eaa−1 a,
a−1 aeaaa, etc. Let us say that two words are equivalent if one can
transform one word into another by means of the trivial relations
aa−1 = e = a−1 a and ae = ea = a and the relation a5 = 1 (as usual,
a5 stands for aaaaa). This is obviously an equivalence relation in
the technical sense, i.e., it is reflexive, symmetric, and transitive, so
that the set of all words splits into equivalence classes. Define the
product of two equivalence classes as the class containing the con-
catenation of any two elements of the given classes. It is easy to see
that this product is well defined, i.e., does not depend on the choice
of representatives in the classes. Obviously, there will be 5 equiv-
alence classes (determined by the elements a, a2 , a3 , a4 , a5 = e) and
they form a group under the product operation defined above. The
group obtained is clearly isomorphic to Z5 .
(ii) Now consider words in the five-letter alphabet {e, s± ±1
1 , s2 }.
Let us say that two words are equivalent if one can be transformed
into the other by means of the trivial relations (which we won’t write
out again) and the relations s21 = s22 = e and s1 s2 s1 = s2 s1 s2 (the
latter is known as the Artin relation) . Defining the product of the
corresponding equivalence classes as in the previous example, we ob-
tain a group which is isomorphic to S3 (see Problem 2.9 below).
2.7.2. Formal definition. The definition of a group presentation is

the following. An expression of the form
G = g1 , . . . , gn : R1 , . . . , Rk ,
where R1 , . . . , Rk are words in the alphabet
A = {g1 , . . . , gn g1−1 , . . . , gn−1 },
is called a presentation of the group G; the words Rj are called rela-

tors; the group G is defined by its presentation as the quotient group
F[g1 , . . . , gn ]/{R1 , . . . , Rk },
where {R1 , . . . , Rk } is the minimal (by inclusion) normal subgroup of

the free group F[g1 , . . . , gn ] containing the elements R1 , . . . , Rk .
This formal definition may be difficult to understand. But the
notion of group presentation is simple. The elements of the group G
that it defines are words in the alphabet A defined up to the trivial
relations (see 2.6.1) and up to the defining relations R1 = e, . . . , Rk =
e; the product is concatenation (and is well defined).
Here are some examples:
(i) Zm = a : am is the m-element cyclic group;
(ii) F[g1 , . . . , gn ] = g1 , . . . , gn : is the free group on n gener-
ators (nothing appears after the colon in the angle brackets because
the free group has no defining relations);
(iii) the permutation group on four elements can be presented as
(2.1) S4 = s1 , s2 , s3 : s21 , s22 , s23 , s1 s2 s−1 −1 −1 −1 −1

1 s2 , s1 s2 s1 s2 s1 s2 ,
s2 s3 s2 s−1 −1 −1
3 s2 s3 .
More details and examples appear in the problem section of this

chapter.
2.8. Cayley’s theorem

The following theorem (due to the British mathematician Arthur Cay-
ley) shows that the notion of abstract group is not a real generaliza-
tion: all groups are in fact transformation groups!
Theorem 2.8.1. Any group G is a transformation group acting on

the set G by right multiplication: g → gg0 for any g0 ∈ G.
Proof. First, we must show that the assignment g → gg0 is a bi-

jection for any g0 ∈ G. But this is obvious: it is injective (by the
cancellation rule) and surjective (to any element h ∈ G the element
g0 assigns the element hg0−1 ). Further, we must verify the transfor-
mation group axioms (see 1.3.1). This verification is also obvious: the
transformations defined by elements of G are closed under composi-
tion (because so are elements of G) and under taking inverse elements
2.9. Problems 65
(the transformation inverse to the one given by g0 is the one given by

g0−1 ).
Corollary 2.8.2. Any group is a geometry in the sense of Klein (i.e.,

in the sense of formal definition given in 1.4.1).
This corollary shows (as we mentioned previously) that the def-

inition of geometry given in 1.4.1 is of course too general; additional
restrictions on the set of elements and the transformation group are
needed to obtain an object about which most mathematicians will
agree that it is a bona fide geometry. However, there seems to be no
formal agreement on this subject, so that the “additional restrictions”
to be imposed are a matter of opinion, and we will not specify any
(at least on the formal level) in this course.
2.9. Problems
2.1. Describe all the finite groups of order 6 or less and supply each
with a geometric interpretation.
2.2. Describe all the (nontrivial) normal subgroups and the corre-
sponding quotient groups of
(a) the isometry group of the equilateral triangle;
(b) the isometry group of the regular tetrahedron.
2.3. Let G be the motion group of the plane, P its subgroup of

parallel translations, and R its subgroup of rotations with fixed center
O. Prove that the subgroup P is normal and the quotient group G/P
is isomorphic to R.
2.4. Prove that if the order of a subgroup is equal to half the order
of the group (i.e., the subgroup is of index 2), then the subgroup is
normal.
2.5. Find all the orbits and stabilizers of all the points of the group
G ⊂ S10 generated by the permutation
[5, 8, 3, 9, 4, 10, 6, 2, 1, 7] ∈ S10
acting on the set {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}.
2.6. Find the maximum order of elements in the group (a) S5 ; (b)
S13 .
2.7. Find the least natural number n such that the group S13 has no
elements of order n.
2.8. Prove that the permutation group Sn is generated by the trans-

position
(1, 2) := [2, 1, 3, 4, . . . , n]
and the cycle
(1, 2, . . . , n) := [2, 3, . . . , n, 1].
2.9. Present the symmetry group of the equilateral triangle by gen-

erators and relations in two different ways.
2.10. How many homomorphisms of the free group in two generators

into the permutation group S3 are there? How many of them are
epimorphisms?
2.11. Prove that the group presented as

a, b | a2 = bn = a−1 bab = 1
is isomorphic to the dihedral group Dn (defined in Chapter 3).
2.12. Show that if the elements a and b of a group satisfy the relations
a5 = b3 = 1 and b−1 ab = a2 , then a = 1.
Chapter 3
Finite Subgroups of
SO(3) and the Platonic
Bodies
This chapter is devoted to the classification of regular polyhedra (the

five “Platonic bodies”) pictured below:
The proof of the classification theorem given here is based on

group theory, more precisely on the study of finite subgroups of the
isometry group of the two-dimensional sphere.
67
68 3. Finite Subgroups of SO(3) and the Platonic Bodies
3.1. The Platonic bodies in art, philosophy, and

science
The perfection of the shape of regular polyhedra attracted the great
artist and thinker Leonardo da Vinci, who pictured them in various
media. Figure 3.1 reproduces his engravings of two of them.
Figure 3.1. Da Vinci engravings: the icosahedron and dodecahedron.
Some philosophers and scientists felt an almost mystical attrac-

tion to these amazingly symmetric shapes. Thus the great astronomer
Kepler believed that the distances from the planets to the Sun could
be calculated from a system of nested inscribed Platonic bodies (see
his weird engraving reproduced in Figure 3.2).
The engraving shows a cube inscribed in a sphere, then a smaller
sphere inscribed in the cube, a tetrahedron inscribed in that second
sphere, a third sphere inscribed in the tetrahedron, followed by suc-
cessively inscribed sphere, dodecahedron, sphere, octahedron, sphere,
icosahedron. Kepler claimed that the distances from the five planets
to the Sun were proportional to the distances from the vertices of
the five nested polyhedra to their common center of symmetry. He
regarded this “discovery” as his main scientific achievement, far more
important than the three fundamental astronomical laws that bear
his name. Fortunately for his self-esteem, he did not live to see the
3.1. The Platonic bodies in art, philosophy, and science 69
Figure 3.2. Kepler’s theory of planetary orbits.
day when more exact measurements of the distances between the Sun
and the planets showed that Kepler’s theory was erroneous.
The five regular polyhedra were known to the ancient Greeks,

in particular, to the philosopher Plato, who expressed his admira-
tion for their unique perfection so beautifully that today they are
often called “Platonic bodies”. Of course Plato cannot be credited
with their discovery (they were known before his time), but who the
actual discoverers were is not clear. It is also unclear whether the
ancient Greeks had a proof of the fact that there are no other reg-
ular polyhedra, or indeed felt that such a proof was necessary. We
can only conjecture that Archimedes had such a proof, or that it was
possibly known to the Pythagorean school.
We do know whether Pythagoras was interested in the regular
polyhedra in connection with his theory of the “singing spheres”. In
the 20th century, his theory was revived in the work of the German
physicist Heisenberg, but the relevant ideas lie outside the scope of a
mathematical textbook.
3.2. Finite subgroups of SO(3)

As we mentioned above, the main goal of this chapter is to prove that
the only regular three-dimensional polyhedra are the five Platonic
bodies. The proof that we give here is essentially group-theoretic (we
reduce the classification problem of regular polyhedra to classifying
finite subgroups of the special orthogonal group SO(3), or, which is
the same thing, the group of motions of the sphere S2 ). This proof is
quite natural and more geometric, in a deeper sense, than the tedious
and eclectic space geometry proof anterior to the appearance of the
notion of transformation group in mathematics.
Let us return to the geometry (briefly studied in Chapter 1, see
1.2.5) of the two-dimensional sphere
X = S2 := {(x, y, z) ∈ R2 | x2 + y 2 + z 2 = 1}
defined by the action of its isometry group Sym (S2 ). (In linear alge-
bra courses this group is defined in a different (but equivalent) way,
it is called the orthogonal group, and usually denoted by O(3).) Here
we will be dealing with the subgroup of O(3) = Sym (S2 ) consisting of
rotations, namely the group Rot(S2 ) each element of which is a rota-
tion of the sphere about some axis passing through the origin by some
angle φ, 0 ≤ φ < 2π. In linear algebra courses this group is defined in
a different (but equivalent) way, is called the special orthogonal group,
and is usually denoted by SO(3).
Our goal is to find the finite subgroups of SO(3). We begin with
some examples of finite subgroups of O(3) and SO(3).
3.2. Finite subgroups of SO(3) 71
3.2.1. The monohedral group Zn for any n ≥ 2. Its n elements

are rotations about an axis by angles of 2kπ/n, where k = 0, . . . , n−1.
Figure 3.3. The dihedral group Dn for n = 6 and n = 5.
3.2.2. The dihedral group Dn for any n ≥ 2. This 2n-element

group is the isometry group of the regular n-gon (lying in the hori-
zontal plane Oxy and inscribed in the sphere S2 ); Dn consists of n
rotations (by angles of 2kπ/n, k = 0, 1, . . . , n − 1) and n reflections
in the horizontal lines passing through the center of the sphere, the
vertices, and the midpoints of the sides (be careful: these lines are
different when n is even or odd – look at the figure!). Note that the
reflections of Dn in the horizontal lines are actually rotations by 180◦
in space about these lines.
3.2.3. The isometry group of the regular tetrahedron. It con-

sists of 24 elements, it is denoted by Sym(Δ3 ), and its (12 element)
rotation subgroup is:
Rot(Δ3 ) = Sym+ (Δ3 ) ⊂ Sym(Δ3 );
Sym(Δ3 ) consists of 8 rotations about 4 axes (containing one vertex)

by angles of 2π/3 and 4π/3, of three rotations by π about axes joining
the midpoints of opposite edges and of the identity. It is easy to see
that Sym(Δ3 ) is isomorphic to the permutation group S4 . But here
we think of this group geometrically, regarding the tetrahedron as
inscribed in the sphere S2 and the elements of Sym(Δ3 as acting on
the sphere as well.
3.2.4. The isometry group Sym(I 3 ) of the cube. It has 48 ele-

ments (see 1.2.3); its rotation subgroup consists of 24 elements:
Rot(I 3 ) = Sym+ (I 3 ) ⊂ Sym(I 3 ).
If we join the center of each of the 6 faces of the cube by segments to
the four neighboring centers, we obtain the carcass of the octahedron
dual to the cube (see Figure 3.4). The octahedron has 6 vertices and
8 triangular faces; its isometry group is obviously the same as that of
the cube.
3.2.5. The isometry group Sym(Dod) of the dodecahedron. It

has 120 elements and possesses a (60 element) rotation subgroup:
Rot(Dod) = Sym+ (Dod) ⊂ Sym(Dod).
The dodecahedron is the (regular) polyhedron (inscribed in the sphere
S2 ) with 12 faces (congruent regular pentagons), 30 edges, and 20
vertices (see Figure 3.4). The existence of such a polyhedron will be
proved at the end of this chapter. Joining the centers of the faces of
the dodecahedron having a common edge (look at Figure 3.4 again),
we get the icosahedron dual to the dodecahedron; it has 20 faces, 30
edges, and 12 vertices. Its transformation group is the same as that
of the dodecahedron.
Figure 3.4. Dual pairs of regular polyhedra.
The following theorem states that SO(3) has no finite subgroups

other than those listed above.
Theorem 3.2.6. Any finite nontrivial subgroup G+ of Sym+ (S2 ) =
SO(3) is isomorphic to one of the following groups:
(i) Zn , n ≥ 2, (ii) Dn , n ≥ 2, (iii) Rot(Δ3 ), (iv) Sym+ (I 3 ),

(v) Sym+ (Dod).
Proof. We know that any element of SO(3) (and hence of G+ ) is a

rotation about a diameter of the sphere S2 and has two fixed points
(the ends of the diameter). Let F be the set of fixed points of the
group G+ :

F = x ∈ S2 | ∃g ∈ G+ \ id, xg = x .
For example, for the group Zn , F consists of two points, while for
the rotation group Rot(Δ3 ) of the tetrahedron it has 14, namely the
4 vertices, the 4 intersection points of the axes of rotations of the
faces with the sphere, and the 6 intersection points of the three axes
of rotations passing through the midpoints of opposite edges of the
tetrahedron.
Consider the (finite) geometry (F : G+ ) and let A be a set con-
taining one point in each orbit of G+ in F . First we claim that the
number of points in F is
|F | = |A||G+ | − 2(|G+ | − 1).
The proof of this fact is the object of Problem 3.3 at the end of the
present chapter. Using the class formula (1.2) from Chapter 1, we
can write
|G+ |
|F | = , where v(x) := |St(x)|.
v(x)
x∈A
Note that v(x) is the order of the rotation subgroup of G+ generated

by the rotations about the axis containing x and its antipodal point.
Replacing |F | by its value found above and dividing by |G+ |, we
obtain
2 1
(3.1) 2− + = 1− ,
|G | v(x)
x∈A
or solving for |G |,
+
1 1 −1
(3.2) |G+ | = 1 − · 1− .
2 v(x)
x∈A
The left-hand side of the boxed formula is less than 2, but greater than
or equal to 1; hence so is the sum in the right-hand side, and thus
the summation over A cannot contain 4 summands or more (because

v(x) ≥ 2); therefore there can be only 2 or 3 orbits of the action of
G+ on F .
First let us consider the case in which |F | = 2, i.e., when there
is only one rotation axis (with intersection points x1 and x2 with the
sphere). In that case there are two orbits in F , each consisting of one
point, namely {x1 } and {x2 }. Is such a situation possible? Of course
it is, but only if G+ consists of rotations about the unique axis x1 x2 .
But then it follows that G+ is isomorphic to Zn for some n ≥ 2. So
the theorem is proved for the case |F | = 2. Note that in this case
v(x1 ) = v(x2 ) = n = |G+ |.
It is easy to see that if the action of G+ on F produces only two
orbits, then the stabilizers of points from these two orbits have the
same number of elements and we are in the case |F | = 2 considered
above. Thus for the rest of the proof, we can assume that there are
three orbits.
Denote by x1 , x2 , x3 points of these three orbits, so that A =
{x1 , x2 , x3 }, and denote by v1 , v2 , v3 the values of v(x) (the number
of elements in the stabilizers, or which is the same thing, the order
of the corresponding rotation axis) at these points, numbered so that
v1 ≤ v2 ≤ v3 .
We can assume that |F | > 2 (the case |F | = 2 was considered
above), i.e., there are two rotation axes or more; but then the com-
position of the two rotations gives a rotation about a third axis and
so |G+ | ≥ 6. We now claim that there is an orbit with stabilizer equal
to 2.
Indeed, if, in contradiction with our claim, all the vi were greater
than 2, the right-hand side of formula (3.1) would be greater than or
equal to 2, which we know is impossible.
Thus it remains to consider the situation in which v1 = 2 and
there are three orbits of the action of G+ on F .
The rest of the proof is a case-by-case analysis of this situation
depending on the possible values of the vi . These values must satisfy
relation (3.1), whose right-hand side is, as we remember, less than 2.
Thus we must have the inequality

1 1 1
(3.3) 3− − − < 2.
2 v2 v3
When is this inequality possible? Since v2 and v3 are both integers
greater than or equal to 2, this can happen only in the cases 2–5
indicated in the following table (in it, the column for the number of
elements of G+ was filled by using formula (3.2)):
v1 v2 v3 |G+ |
case 1 n n - n
case 2 2 2 n 2n
case 3 2 3 3 12
case 4 2 3 4 24
case 5 2 3 5 60
In the rest of the proof, we consider each case separately and

distinguish (among the points of F ) the vertices of a (possibly degen-
erate) polyhedron on which G+ acts. We then show that this action
is one of those listed in the claim of the theorem, i.e., the distin-
guished polyhedron either degenerates into a regular polygon or is
the tetrahedron, or the cube, or the dodecahedron.
Case 1: This is the case in which |F | = 2; it was considered above,
and we showed that it yields the group Zn , n ≥ 2.
Case 2: Assume that v2 = 2. Then we have two rotation axes
l1 , l2 of order 2, i.e., such that the rotation angle is 180◦ . Consider
the line l3 perpendicular to these two axes. One of its intersection
points with the sphere will be x3 . Let n be the order of the axis
l3 . It follows from formula (3.2) that the number of elements of G+
is equal to 2n. We can now specify the three orbits in F : the n-
point orbit containing x1 , which lies in the plane perpendicular to l3
passing through the center of the sphere, the n-point orbit containing
x2 , lying in the same plane, and the 2-point orbit consisting of x3 and
its antipodal point. It is now clear that in our case G+ is isomorphic
to the dihedral group Dn .
Case 3: v2 = v3 = 3. Then the number of elements of our group
can be computed from formula (3.2), and is equal to 12. Consider
the axis of rotation l3 passing through x2 ; it is of order 3. Let x3 and

x3 be the two points to which the rotations about l2 take the point
x3 . The rotation about the axis l1 containing the point x1 is of order
2, hence at least one of the three points x3 , x3 , x3 must be taken to
a point (which we denote by x 3 ) that does not coincide with one of
those three. Thus we obtain a tetrahedron x3 , x3 , x3 , x3 , which, as
we will soon see, turns out to be regular. Taking the composition
of the rotation about l3 and the rotation about l1 , we get another
rotation of order 3, from which we conclude that another face of the
tetrahedron is an equilateral triangle, and therefore the tetrahedron
x3 , x3 , x3 , x
3 is regular. Taking the composition of two order three
rotations, we obtain another order two rotation and, continuing in
the same vein, we describe all 12 rotations of G+ and can conclude
that G+ is isomorphic to Rot(Δ3 ).
Case 4: Assume that v2 = 2 and v3 = 4. Here the strategy

of proof is similar to the one in Case 3, except that now we find
the 8 vertices of a cube (rather than those of a tetrahedron) among
the points of F . To do this, we begin with the order four rotation,
obtaining two squares inscribed in the sphere, then use the other
rotations to show that the two squares are actually opposite faces of a
cube, and finally verify that the 24 elements of G+ are the symmetries
of this cube, so that G+ is isomorphic to Sym+ (I 3 ).
Case 5: Assume that v2 = 3 and v3 = 5. Here the strategy of

proof is similar to that used in Cases 3 and 4, except that now we con-
struct a dodecahedron from points of F and obtain an isomorphism
between G+ and Sym+ (Dod). We relegate the details to Problem
3.10.
Thus we see that the five cases correspond to the groups (i)–(v),
respectively. The theorem is proved.
3.2.7. Let us denote by D n the subgroup of SO(3) generated by the

elements of Dn and the reflection ρ in the plane passing through the
rotation axis of order n and one of the axes of order 2 in Dn . Obviously
the subgroup D n has 4n elements (because the compositions of ρ with
different elements of Dn are all different from each other).
3.3. The five regular polyhedra 77
Note also that the subgroup of SO(3) generated by the elements

of Zn (interpreted as the motion group of the regular n-gon lying
in the equatorial plane of the sphere and inscribed in it) and the
reflection in a vertical plane passing through a vertex of the n-gon
and the center of the sphere is Dn .
Corollary 3.2.8. Any finite subgroup G of O(3) is either isomorphic

to one of the groups listed in Theorem 3.2.6 or to one of the following
groups:
n , (ii) S4 , (iii) Sym(I 3 ), (iv) Sym(Dod).
(i) D
Proof. Let G be a finite subgroup of SO(3) and let G+ be its rotation

subgroup. By Theorem 3.2.6, G+ must be one of the five groups listed
in the theorem. The whole group G is generated by the elements of
G+ and one reflection in a plane passing through the origin, so it must
be one of the five groups listed in the statement of the corollary.
3.3. The five regular polyhedra

A regular polyhedron is defined as a convex polyhedron inscribed in
the sphere S2 such that
(i) all its faces are congruent regular polygons of k sides for some
k > 2;
(ii) the endpoints of all the edges issuing from each vertex lie in
one plane and form a regular l-gon for some l > 2.
Theorem 3.3.1. There are exactly five different regular polyhedra:

the tetrahedron, the cube, the octahedron, the dodecahedron, and the
icosahedron.
Proof. This theorem follows from the corollary to Theorem 3.2.6.

Indeed, the definition implies that the isometry group of a regular
polyhedron is finite and therefore must be one of the groups listed in
Theorem 3.2.6. The two “series” (i) and (ii) do not give any (non-
degenerate) polyhedra (why?). In case (iii), we get the tetrahedron
(because its symmetry group is isomorphic to the permutation group
S4 ). In case (iv), we get the cube and its dual, the octahedron, and
in case (v), the dodecahedron and its dual, the icosahedron.
Thus we obtain five geometries with three different group actions

(tetrahedron, cube ∼ octahedron, dodecahedron ∼ icosahedron). To
understand the group actions in these geometries, it is useful to have
a look at their fundamental domains (Figure 3.5).
Figure 3.5. Fundamental domains of Platonic bodies.
In all five cases, their fundamental domains are pyramids with

vertex the center of the body and base the fundamental domain (a
right triangle in all five cases) of the isometry group of a face. These
triangles have acute angles 30◦ (tetrahedron, octahedron, icosahe-
dron), 45◦ (cube), 54◦ (dodecahedron).
3.4. The five Kepler cubes

Kepler observed that the cube can be inscribed in five different ways
into the dodecahedron. Here we will perform the opposite construc-
tion: starting from the cube, we will construct a dodecahedron cir-
cumscribed to the cube. This will prove the existence of the dodeca-
hedron.
Consider two copies ABCDE and A B C D E of the regular
pentagon with diagonals of length 1. Place these pentagons in the
plane of the unit square P QRS so that the diagonals BE and B E
are identified with P S and QR, respectively, and CD is parallel to
C D . By rotating the pentagons in space about P S and QR, identify
the sides CD and C D above the square P QRS.
Now let P QRS be the top face of the unit cube P QRSP Q R S .
Place two more pentagons on the face SRR S of the cube the same
way as before, so that their parallel sides are parallel to SR. Now
rotate these two pentagons until these parallel sides are identified.
3.5. Regular polyhedra in higher dimensions 79
Figure 3.6. Constructing the dodecahedron.
Then it is not hard to prove that the upper endpoint of the identified
segment will coincide with one of the endpoints of the common (iden-
tified) segment of the first two pentagons. Perform similar construc-
tions on the other faces of the cube. The polyhedron thus obtained
will be the dodecahedron.
3.5. Regular polyhedra in higher dimensions

In any dimensions n > 3, there is a classification theorem for regular
n-dimensional polyhedra similar to that in three dimensions. Surpris-
ingly, the number of types of polyhedra decreases with the increase
of n, changing from five (in dim=3) and six (dim=4) to three (when
dim≥ 5). Thus, instead of the increased variety of regular bodies that
might be expected in high dimensions, there are basically only three:
the analogs of the tetrahedron, the cube, and the polyhedron dual to
the cube.
In this section, after presenting the necessary definitions, we state
the corresponding classification theorems without proof.
3.5.1. Examples and definitions. We begin with a simple exam-

ple: the four-dimensional cube. In the Euclidean space R4 , consider
the 16 points (±1, ±1, ±1, ±1); their convex hull is by definition the
4-cube. A projection of the four-dimensional cube on the plane ap-
pears in Figure 3.7.
Even simpler (as its name indicates) is the regular n-dimensional
simplex Δn , which is the n-dimensional analog of the tetrahedron,
and is defined inductively: given the (n − 1)-dimensional (regular)
simplex Δn−1 lying in Rn−1 , we construct a perpendicular from its
center of gravity into the nth dimension (i.e., a line parallel to the
basis vector (0, . . . , 0, 1) ∈ Rn ⊃ Rn−1 ) and take for the n+1st vertex
of our simplex the point whose distance from the n vertices of Δn−1
is equal to the length of the edges of Δn−1 . It is easy to see that the
transformation group of Δn is the permutation group Σn+1 .
Regular n-dimensional polyhedra are defined recursively. The
recursion begins for n = 3 and is that of a Platonic body (see Sec. 3.3
above). If regular (n − 1)-dimensional polyhedra have been defined,
we define a regular n-dimensional polyhedron as a convex polyhedron
(inscribed in the sphere Sn−1 := {(x1 , . . . , xn ) ∈ Rn | x21 +· · ·+x2n = 1})
such that
(i) all of its faces are congruent regular (n − 1)-dimensional poly-
hedra;
(ii) the endpoints of all the edges issuing from each vertex lie in
one hyperplane and form a regular (n − 1)-dimensional polyhedron;
all such polyhedra are congruent (but are not necessarily the same as
those from item (i)).
To each regular polyhedron P , one can assign its symbol, defined
(inductively) as the n-tuple of integers (r1 , r2 , . . . , rn−1 ) in which r1
is the number of edges of any one of the 2-dimensional faces Q of P ,
while (r2 , . . . , rn−1 ) is the symbol of Q. For example, (4, 3, 3) is the
symbol of the four-dimensional cube, (5, 3) is that of the dodecahe-
dron, (3, 3, 3, 3) that of the five-dimensional regular simplex.
One can define the dual to any regular polyhedron in the natural
way (similarly to the way it is done in dimension 3). For example, the
5-simplex is dual to itself, while the dual to the 4-cube is the so-called
cocube, which has the symbol (3, 3, 4).
Theorem 3.5.2. There are (up to homothety) six different regular

polyhedra in dimension 4; their symbols are
(3, 3, 3), (4, 3, 3), (3, 3, 4), (3, 4, 3), (5, 3, 3), (3, 3, 5).
The reader who wishes to find a proof of this theorem is referred

to Problem 3.13, in which hints about the mysterious polyhedra with
symbols (3, 4, 3), (5, 3, 3), (3, 3, 5) appear, or to Figure 3.7, where their
projections on the plane are shown.
3.6. Problems 81
Figure 3.7. Regular 4-dimensional polyhedra.
Theorem 3.5.3. In dimension n ≥ 5 there are (up to homothety)

three different regular polyhedra: the n-simplex, the n-cube, and the
n-cocube; their symbols are
(3, 3, . . . , 3, 3), (4, 3, . . . , 3, 3), (3, 3, . . . , 3, 4).
We omit the proof (see [2]); the reader is also referred to Prob-
lem 3.14.
3.6. Problems
3.1. A regular pyramid of six lateral sides is inscribed in the sphere
S2 . Find its symmetry (i.e., isometry) group and its group of motions.
How does your answer relate to the theorem on finite subgroups of
SO(3)?
3.2. Answer the same questions as in Problem 3.1 for
(a) the regular prism of six lateral sides;
(b) the regular truncated pyramid of five lateral sides;
(c) the double regular pyramid of six lateral sides (i.e., the union
of two regular pyramids of six lateral sides with common base and
vertices at the poles of the sphere).
3.3. Let G+ be a finite subgroup of SO(3) acting on the sphere S2

and F the set of all the points fixed by nontrivial elements of G+ ;
prove that F is invariant with respect to the action of G+ and
|F | = |G+ | · |A| − 2(|G+ | − 1),
where A ⊂ F is a set containing exactly one point from each orbit of
the action of G+ on the set F .
3.4. Does the motion group of the cube have a subgroup isomorphic
to the motion group of the regular tetrahedron?
3.5. Does the motion group of the dodecahedron have a subgroup

isomorphic to the motion group of the cube?
3.6. In the motion group of the cube, find all groups isomorphic to
Zn and Dn for various values of n. Does it have any other subgroups?
3.7. Prove the existence of the dodecahedron in detail.
3.8. Given a cube inscribed in the sphere, let the set F consist of
all the vertices of the cube, all the intersection points of the lines
joining the centers of its opposite faces, and of the lines joining the
midpoints of opposite edges, and let G+ be the motion group of the
cube. Prove that G+ acts on F , find all the orbits of this action and
the stabilizers of all the points of F . Compare your findings with the
proof of Theorem 3.2.6 in Case 4.
3.9. Given a regular tetrahedron inscribed in the sphere, let the set F
consist of all its vertices and of the lines joining the midpoints of the
edges, and let G+ be the motion group of the tetrahedron. Prove that
G+ acts on F , find all the orbits of this action and the stabilizers of
all the points of F . Compare your findings with the proof of Theorem
3.2.6 in Case 3.
3.10. Given a dodecahedron inscribed in the sphere, let the set F

consist of all the vertices of the dodecahedron, all the intersection
points of the lines joining the centers of its opposite faces and of the
3.6. Problems 83
Figure 3.8. Projection of the edges of the 5-dimensional cube.
lines joining the midpoints of the edges, and let G+ be the motion
group of the dodecahedron. Prove that G+ acts on F and complete
the proof of Theorem 3.2.6 in Case 5.
3.11. Prove Theorem 3.2.6 in Case 4 by constructing an octahedron

(instead of a cube) from the points of F .
3.12. Use your computer to produce a picture of the projection on

an appropriately chosen two-dimensional plane of the five-dimensional
cube. Compare with Figure 3.8.
3.13∗ . Prove the classification theorem for regular polyhedra in di-

mension four.
3.14∗ . Prove the classification theorem for regular polyhedra in di-

mension five.
Chapter 4
Discrete Subgroups of
the Isometry Group of
the Plane and Tilings
This chapter, just as the previous one, deals with a classification of

objects, the original interest in which was perhaps more aesthetic than
scientific, and goes back many centuries ago. The objects in question
are regular tilings (also called tessellations), i.e., configurations of
identical figures that fill up the plane in a regular way. Each regular
tiling defines a geometry in the sense of Klein; it turns out that, up
to isomorphism, there are 17 such geometries; their classification will
be obtained by studying the corresponding transformation groups,
which are discrete subgroups (see the definition in Section 4.3) of the
isometry group of the Euclidean plane.
4.1. Tilings in architecture, art, and science

In architecture, regular tilings appear, in particular, as decorative
mosaics (Figure 4.1) in the famous Alhambra palace (14th century
Spain). According to M. Berger [2] and B. Grünbaum [7], part or all
the 17 geometries mentioned above are realized by Alhambra mosaics.
The reader can easily find beautiful color reproductions in the web
by Googling “Alhambra mosaics”.
85
86 4. Discrete Subgroups of Sym(R2 ) and Tilings
Figure 4.1. Two Alhambra mosaics.
In art, the famous Dutch artist M.C. Escher, known for his “im-
possible” paintings, used regular tilings as the geometric basis of his
wonderful “periodic” watercolors, which can be easily found in the
Internet, but not copied and reproduced. (Apparently, the Escher
Foundation is more interested in making money than in popularizing
the artist’s work.) So we can show only the regular tilings underlying
two of Escher’s periodic watercolors (Figure 4.2).
Figure 4.2. Two periodic tilings.
From the scientific viewpoint, not only regular tilings are impor-
tant: it is possible to tile the plane by copies of one tile (or two) in an
irregular (nonperiodic) way. It is easy to fill the plane with rectan-
gular tiles of size say 10cm by 20cm in many nonperiodic ways. But
the fact that R2 can be filled irregularly by nonconvex 9-gons is not
obvious. Such an amazing construction, due to Vorderberg (1936),
is shown in Figure 4.3. The figure indicates how to fill the plane by
4.2. Tilings and crystallography 87
copies of two tiles (their enlarged copies are shown separately; they
are actually mirror images of each other) by fitting them together to
form two spiraling curved strips covering the whole plane.
Figure 4.3. The Vorderberg tiling.
Somewhat later, in the 1960s, interest in irregular tilings was

revived by the nonperiodic tilings due to the British mathematical
physicist Roger Penrose, which are related to statistical models and
the study of quasi-crystals. A version of a very famous Penrose tiling
is shown in Figure 4.4 on the next page. The reader can find Penrose’s
original version by searching the web for “Penrose tiling” or (for a
more artistic and amusing version), “Penrose chickens”.
4.2. Tilings and crystallography

The first proof of the classification theorem of regular tilings (defined
below, see 4.5.1) was obtained by the Russian crystallographer Fe-
dorov in 1891. Mathematically, they are given by special discrete
subgroups, called the Fedorov groups, of the isometry group Sym(R2 )
of the plane. As we mentioned above, there are 17 of them (up to
isomorphism). The Fedorov groups act on the Euclidean plane, form-
ing 17 different (i.e., nonisomorphic) geometries in the sense of Klein,
which we call tiling geometries.
Figure 4.4. A Penrose tiling.
The proof given here, just as the one in the previous chapter,
is group-theoretic, and is based on the study of discrete subgroups
of the isometry group of the plane. In fact, the actual classification
principle cannot be stated without using transformation groups, and
at first glance it is difficult to understand how it came about that the
architects of the Alhambra palace, five centuries before the notion
of group appeared in mathematics, actually found most or all the 17
regular tilings (in this connection, see the article [7] by B. Grünbaum).
Actually, this is not surprising: a deep understanding of symmetry
suffices to obtain answers to an intuitively clear question, even if
one is unable to state the question in the terminology of modern
mathematics.
Less visual, but more important for the applications (crystallo-
graphy) is the generalization of the notion of regular tiling to three
dimensions: configurations of identical polyhedra filling R3 in a reg-
ular way. Mathematically, they are also defined by means of discrete
subgroups called crystallographic groups of the isometry group of R3
and have been classified: there are 230 of them. Their study is beyond
the scope of this book.
We are concerned here with the two-dimensional situation, and

accordingly we begin by recalling some facts from elementary plane
geometry, namely facts concerning the structure of isometries of the
plane R2 .
4.3. Isometries of the plane

Recall that by Sym(R2 ) we denote the group of isometries (i.e., dis-
tance-preserving transformations) of the plane R2 , and by Sym+ (R2 )
its group of motions (i.e., isometries preserving orientation). Exam-
ples of the latter are parallel translations and rotations, while reflec-
tions in a line are examples of isometries which are not motions (they
reverse orientation).
(We consider an isometry orientation-reversing if it transforms a
clockwise oriented circle into a counterclockwise oriented one. This is
not a mathematical definition, since it appeals to the physical notion
of “clockwise rotation”, but there is a simple and rigorous mathe-
matical definition of orientation-reversing (-preserving) isometry in
linear algebra courses, based on the sign (±) of the determinant of
the corresponding linear map.)
Below we list some well-known facts about isometries of the plane;
their proofs are relegated to exercises appearing at the end of the
present chapter. These facts are also discussed in Section 0.9
of Chapter 0.
4.3.1. A classical theorem of elementary plane geometry says that

any motion is either a parallel translation or a rotation (see Problem
4.1).
4.3.2. A less popular but equally important fact is that any orienta-
tion-reversing isometry is a glide reflection, i.e., the composition of a
reflection in some line and a parallel translation by a vector collinear
to that line (Problem 4.2).
4.3.3. The composition of two rotations is a rotation (except for the

particular case in which the two angles of rotation are equal but
opposite: then their composition is a parallel translation). In the
general case, there is a simple construction that yields the center and
angle of rotation of the composition of two rotations (see Problem

4.3). This important fact plays the key role in the proof of the theorem
on the classification of regular tilings.
4.3.4. The composition of a rotation and a parallel translation is a

rotation by the same angle about a point obtained by shifting the
center of the given rotation by the given translation vector (Problem
4.4).
4.3.5. The composition of two reflections in lines l1 and l2 is a ro-

tation about the intersection point of the lines l1 and l2 by an angle
equal to twice the angle from l1 to l2 (Problem 4.5).
4.4. Discrete groups and discrete geometries

The action of a group G on a space X is called discrete if none of
its orbits possesses accumulation points, i.e., there are no points x ∈
X such that any neighborhood of x contains infinitely many points
belonging to one orbit. Here the word “space” can be understood as
Euclidean space Rn (or as a subset of Rn ), but the definition remains
valid for arbitrary metric and topological spaces.
A simple example of a discrete group acting on R2 is the group
consisting of all translations of the form k v , where v is a fixed nonzero
vector and k ∈ Z. The set of all rotations
√ about the origin of R by
2
angles which are integer multiples

√ of 2 π is a group, but its action
on R2 is not discrete (since 2 is irrational, orbits are dense subsets
of circles centered at the origin).
4.5. The seventeen regular tilings

4.5.1. Formal definition. By definition, a tiling or tessellation of
the plane R2 by a polygon T0 , the tile, is an infinite family {T1 , T2 , . . . }
of pairwise nonoverlapping (i.e., no two distinct tiles have common

∞
interior points) copies of T0 filling the plane, i.e., R2 = i=1 Ti .
For example, it is easy to tile the plane by any rectangle in dif-
ferent ways, e.g., as a rectangular lattice as well as in many irregular,
4.5. The seventeen regular tilings 91
nonperiodic ways. Another familiar tiling of the plane is the hon-

eycomb lattice, where the plane is filled with identical copies of the
regular hexagon.
A polygon T0 ⊂ R2 , called the fundamental tile, determines a
regular tiling of the plane R2 if there is a subgroup G (called the
tiling group) of the isometry group Sym(R2 ) of the plane such that
(i) G acts discretely on R2 , i.e., none of the orbits of G has accu-
mulation points;
(ii) the images of T0 under the action of G fill the plane, i.e.,

g(T0 ) = R2 ;
g∈G
(iii) the action of G is transitive, i.e., for g, h ∈ G the images

g(T0 ), h(T0 ) of the fundamental tile coincide if and only if g = h.
Actually, (ii) and (iii) imply (i), but we will not prove this (see
the first volume of Berger’s book [2], pp. 37–38 of the French edition).
The action of a tiling group G ⊂ Sym(R2 ) on the plane R2 is,
of course, a geometry in the sense of Klein that we call the tiling
geometry (or Fedorov geometry) of the group G.
4.5.2. Examples of regular tilings. Six examples of regular tilings

are shown in Figure 4.5.
Given two tiles, there is one element of the transformation group
that takes one to the other. The question marks show how the tiles
are mapped to each other. (Without the question marks, the action of
the transformation group would not be specified; see Problem 4.16.)
The first five tilings (a–e) are positive, i.e., they correspond to
subgroups of the group Sym+ (R2 ) of motions (generated by all ro-
tations and translations) of the plane (one-sided tiles slide along the
plane). The sixth tiling (f) allows “turning over” the (two-sided) tiles.
Let us look at the corresponding tiling groups in more detail.
Theorem 4.5.3 (Fedorov, 1891). Up to isomorphism, there are ex-

actly five different one-sided tiling geometries of the plane R2 . They
are shown in Figure 4.5(a)–(e).
Figure 4.5. Six regular tilings of the plane.
Proof. Let G be a group of positive tilings. Consider its subgroup

GT of all the parallel translations in G.
Lemma 4.5.4. The subgroup GT is generated by two noncollinear

vectors v and u.
Proof. Arguing by contradiction, suppose that GT is trivial (there

are no parallel translations except the identity). Let r, s be any two
(nonidentical) rotations with different centers. Then rsr −1 s−1 is a
nonidentical translation (to prove this, draw a picture). A contradic-
tion.
4.5. The seventeen regular tilings 93
Now suppose that all the elements of GT are translations gener-

ated by (i.e., proportional to) one vector v. Then it is not difficult
to obtain a contradiction with item (ii) of the definition of regular
tilings.
Now if G contains no rotations, i.e., G = GT , then we get the

tiling (a). Further, if G contains only rotations of order 2, then it is
easy to see that we get the tiling (b).
Lemma 4.5.5. If G contains a rotation of order α ≥ 3, then it

contains two more rotations (of some orders β and γ) such that
1 1 1
+ + = 1.
α β γ
Sketch of the proof. Let A be the center of a rotation of order α.

Let B and C be the nearest (from A) centers of rotation not obtainable
from A by translations. Then the boxed formula follows from the fact
that the sum of angles of triangle ABC is π. The detailed proof of
this lemma is the topic of one of the exercises.
Since the three rotations are of order greater or equal to 3, it

follows from the boxed formula that only three cases are possible.
1/α 1/β 1/γ

case 1 1/3 1/3 1/3
case 2 1/2 1/4 1/4
case 3 1/2 1/3 1/6
Studying these cases one by one, it is easy to establish that they

correspond to the tilings (d), (c), (e) of Figure 4.5, respectively.
This concludes the proof of Theorem 4.5.3.
In the general case (all tilings, including those by two-sided tiles),

there are exactly 17 nonequivalent tilings. This was also proved by
Fedorov. The 12 two-sided ones are shown on the next page.
We will not prove the second part of the classification theorem for
regular plane tilings (it consists in finding the remaining 12 regular
Figure 4.6. Two-sided regular tilings.

4.7. Problems 95
tilings, for which two-sided tiles are required). However, the reader
can study examples of these 12 tilings by doing some of the exercises.
Note that there is a nice website with beautiful examples of decorative
patterns corresponding to the 17 regular tilings:
http://fac-web.spsu.edu/math/tile/symm/ident17.htm
One can also visit the Escher website.
4.6. The 230 crystallographic groups

The crystallographic groups are the analogs in R3 of the tiling groups
in Euclidean space R2 . The corresponding periodically repeated poly-
hedra are not only more beautiful than tilings, they are more impor-
tant: the shapes of most of these polyhedra correspond to the shapes
of real-life crystals. There are 230 crystallographic groups. The proof
is very tedious: there are 230 cases to consider, in fact more, because
many logically arising cases turn out to be geometrically impossible,
and it lies, as we mentioned above, outside the scope of this book.
Those of you who would like to see some nontrivial examples
of geometries corresponding to some of the crystallographic groups
should look at Problem 4.5 and postpone their curiosity to the next
chapter, where 4 examples of actual crystals will appear in the guise
of Coxeter geometries. Another possibility is to consult the website
http://webmineral.com/crystall.shtml or to google the words “crys-
tallographic group”.
4.7. Problems
4.1. Prove that any motion of the plane is either a translation by

some vector v, |v| ≥ 0, or a rotation rA about some point A by a
nonzero angle.
4.2. Prove that any orientation-reversing isometry of the plane is a

glide reflection in some line L with glide vector u, |u| ≥ 0, u||L.
4.3. Justify the following construction of the composition of two

rotations r = (a, ϕ) and (b, ψ). Join the points a and b, rotate the ray
[a, b around a by the angle ϕ/2, rotate the ray [b, a around b by the
angle −ψ/2, and denote by c the intersection point of the two obtained
rays; then c is the center of rotation of the composition rs and its angle
of rotation is 2(π − ϕ/2 − ψ/2). Show that this construction fails in
the particular case in which the two angles of rotation are equal but
opposite, and then their composition is a parallel translation.
4.4. Prove that the composition of a rotation and a parallel transla-

tion is a rotation by the same angle and find its center of rotation.
4.5. Prove that the composition of two reflections in lines l1 and l2

is a rotation about the intersection point of the lines l1 and l2 by an
angle equal to twice the angle from l1 to l2 .
4.6. Indicate a finite system of generators for the transformation

groups corresponding to each of the tilings shown in Figure 4.5 (a),
(b), (c), (d), (e),(f).
4.7. Is it true that the transformation group of the tiling shown in

Figure 4.5(b) is a subgroup of the one of Figure 4.5(c)?
4.8. Indicate the points that are the centers of the rotation subgroups
of the transformation group of the tiling shown in Figure 4.5(c).
4.9. Write out a presentation of the isometry group of the plane

preserving
(a) the regular triangular lattice;
(b) the square lattice;
(c) the hexagonal (i.e., honeycomb) lattice.
4.10. For which of the five Platonic bodies can a (countable) collec-
tion of copies of the body fill Euclidean 3-space (without overlaps)?
4.11. In the Internet, find the two Escher pictures schematically

shown in Figure 4.2 and indicate to which of the 17 Fedorov groups
they correspond.
4.12. Exactly one of the 17 Fedorov groups contains a glide reflection

but no reflections. Which one?
4.13. Which two of the 17 Fedorov groups contain rotations by π/6?

4.7. Problems 97
4.14. Which three of the 17 Fedorov groups contain rotations by

π/2?
4.15. Which five of the 17 Fedorov groups contain rotations by π

only?
4.16. Rearrange the question marks in the tiling (c) of Figure 4.5
so as to make the corresponding geometry isomorphic to that of the
tiling (a).
Chapter 5
Reflection Groups and

Coxeter Geometries
In this chapter, as in the previous one, we study geometries defined

by certain discrete subgroups of the isometry group of the plane (and,
more generally, of n-dimensional space), namely the subgroups gen-
erated by reflections (called Coxeter groups after the 20th century
Canadian mathematician who invented them). These geometries are
perhaps not as beautiful as those studied in the previous two chapters,
but are more important in the applications (in algebra and topology).
On the other hand, they do have an aesthetic origin: what one sees
in a kaleidoscope (a child’s toy very popular before the computer
era) is an instance of such a geometry. Following E.B. Vinberg, we
call these geometries (in the two-dimensional case) kaleidoscopes. We
prove the classification theorem for them in dimension 2 and state its
generalization to higher dimensions without proof (using the notion
of Coxeter scheme).
5.1. An example: the kaleidoscope

The kaleidoscope is a children’s toy: bright little pieces of glass are
placed inside a regular triangular prism and are multiply reflected by
three mirrors forming the lateral faces of the prism.
99
100 5. Reflection Groups and Coxeter Geometries
Figure 5.1. Geometry of the kaleidoscope.
Looking into the prism, you see a colorful repeated pattern: the pic-
ture in the triangle and its mirror images alternate, forming a hexagon
(the union of six equilateral triangles), see Figure 5.1(a), surrounded
by more equilateral triangles ad infinitum.
Mathematically, this is a two-dimensional phenomenon: the equi-
lateral triangle forming the base of the prism is the fundamental do-
main of a discrete group acting on the plane of the base.
Now if the kaleidoscope is deformed (e.g., the angles between the
faces of the prism are slightly changed), then the picture becomes
fuzzy, no pattern can be seen. In such a situation, the images of
the base triangle overlap infinitely many times (see Figure 5.1(b), the
transformation group acting on the triangle is not discrete; we will
not study this “bad” case: we only study the case of the “nice” kalei-
doscope in dimension two and then generalize it to any dimension.
5.2. Coxeter polygons and polyhedra

Consider a dihedral angle α < π/2 formed by two plane two-sided
mirrors Π1 , Π2 . What will the observer O see? Any picture inside the
angle will be reflected by Π1 ; its image will in turn be reflected by the
image of Π1 by Π2 , and so on. At the same time, the picture inside
the angle will be reflected by Π2 ; its image will in turn be reflected
by the image of Π2 by Π1 , etc. Two cases are possible: either the
5.3. Coxeter geometries on the plane 101
reflections coming from different sides will overlap (Figure 5.1(b)) or

the reflected pictures will coincide (Figure 5.1(a)). Obviously, the
pictures will coincide if (and only if) the angle α is of the form π/k,
where k = 2, 3, . . . .
Mathematically, this situation is the following. On the Euclidean
plane, we take two straight lines forming the angle α and consider
the group G of all transformations of the plane generated by the
reflections in these two lines. Let F be the plane region bounded by
the two rays forming the angle α. Obviously, no two regions g(F )
and h(F ), g, h ∈ G, g
= h, overlap iff α = π/k, where k = 2, 3, . . . .
In that case, G is the dihedral group Dk .
Now suppose we are given a convex polygon F in the plane with
vertex angles less than or equal to π/2. Consider the group GF of
transformations of the plane generated by reflections in the lines con-
taining the sides of F . We say that GF acts transitively on F if the
images g(F ), g ∈ GF , never overlap. A necessary condition for the
transitive action of GF on F is that all the vertex angles of F be of
the form π/k for various values of k; this follows from the argument
in the previous paragraph. Obviously, this condition is not sufficient.
The previous arguments are the motivation for the following defi-
nition. A convex polygon F is called a Coxeter polygon if all its vertex
angles are of the form π/k for various values of k = 2, 3, . . . and it
generates a transitive action of the group GF . Coxeter polygons will
be classified below – there are only four.
The above can be generalized to three-dimensional space. The
corresponding definition is the following: a convex polyhedron is
called a Coxeter polyhedron P if all its dihedral angles are of the
form π/k for various values of k = 2, 3, . . . and it generates a tran-
sitive action of GP , where GP is the transformation group generated
by the reflections in the planes containing the faces of P . Coxeter
polyhedra will be classified below (there are seven).
5.3. Coxeter geometries on the plane

Let F be a Coxeter polygon in the plane R2 . The Coxeter geometry
with fundamental domain F is the geometry (R2 : GF ), where GF is
the group of transformations of the plane generated by the reflections

in the lines containing the sides of the polygon F . The goal of this
section is to classify all Coxeter geometries on the plane.
Figure 5.2. The four plane Coxeter geometries.
Theorem 5.3.1. Up to isomorphism, there are four Coxeter geome-

tries in the plane; their fundamental polygons are the rectangle, the
equilateral triangle, the isosceles right triangle, and the right triangle
with angles π/3 and π/6 (see Figure 5.2).
Proof. Let F be the fundamental polygon of a Coxeter geometry. If

it has n sides, then the sum of its angles is π(n−2) and so the average
value of its angles is π(1 − 2/n). Now n cannot be greater than 4,
because F would then have an obtuse angle (and this contradicts the
definition of Coxeter polygon). If n = 4, then all angles of F are
π(1 − 2/4) = π/2 and F is a rectangle. Finally, if n = 3, and the
angles of the fundamental triangle are π/k, π/l, π/m, then (since their
sum is π) we obtain a Diophantine equation for k, l, m:
1 1 1
+ + =1.
k l m
5.4. Coxeter geometries in Euclidean space R3 103
This equation has three solutions: (3, 3, 3), (2, 4, 4), (2, 3, 6). These
solutions correspond to the three triangles listed in the theorem.
5.4. Coxeter geometries in Euclidean space R3

5.4.1. In this section we study the Coxeter geometries in R3 . A
Coxeter polyhedron F ⊂ R3 is a convex polyhedron (i.e., the bounded
intersection of a finite number of half-spaces in R3 ) with dihedral
angles of the form π/k for various values of k = 2, 3, . . . .
A Coxeter geometry in Rd with fundamental polyhedron F is
defined just as in the case d = 2 (see Section 5.3).
Theorem 5.4.2. There are seven Coxeter geometries in three-dimen-
sional space; their fundamental polyhedra are the four right prisms
over the rectangle, the equilateral triangle, the isosceles right triangle,
and the right triangle with acute angles π/3 and π/6, and the three
(nonregular) tetrahedra shown in Figure 5.3.
It is not very difficult to prove that the seven polyhedra (listed in

the theorem) indeed define Coxeter geometries. To prove that there
are no other geometries, nontrivial information from linear algebra
(in particular, the notion of Gramm matrix) is needed. Therefore, we
omit the proof (see the book [4] or, for readers of Russian, a series of
articles in Matematicheskoe Prosveshchenie, Ser. 3, no. 7, 2003).
A remark about terminology. The term “Coxeter geometry” is not
a standard term. E.B. Vinberg uses the term “kaleidoscope” instead.
Also, we do not use the term “Coxeter group” for the transformation
group of a Coxeter geometry. This is because the expression “Cox-
eter group” is standardly used in the literature in a sense somewhat
different from “transformation group of a Coxeter geometry”.
Coxeter geometries are not only abstract mathematical objects,
they are also important models in crystallography. For example, the
polyhedron in Figure 5.3(b) is the crystal of ordinary salt, while the
polyhedron in Figure 5.3(a) is a diamond crystal. Thus these poly-
hedra also appear as fundamental domains of the crystallographic
groups mentioned in the previous chapter.
Figure 5.3. The seven Coxeter polyhedra in 3-space.

5.5. Coxeter schemes and the classification theorem 105
5.5. Coxeter schemes and the classification

theorem
5.5.1. In this section we study the general case of a Coxeter geom-
etry in Rd for an arbitrary positive integer d. A Coxeter polyhedron
F ⊂ Rd is a convex polyhedron (i.e., the bounded intersection of a
finite number of half-spaces in Rd ) with dihedral angles of the form
π/k for various values of k = 2, 3, . . . such that the reflections in
the (d − 1)-dimensionsal hyperplanes containing its faces generate a
transitively acting group GF . (The definition of the measure of a
dihedral angle in Euclidean space of arbitrary dimension d appears in
the linear algebra course.) A Coxeter geometry in Rd with fundamen-
tal polyhedron F is defined exactly as in the cases d = 2 and d = 3
(see Sections 5.3 and 5.4).
5.5.2. A Coxeter scheme is a graph (with integer weights on the

edges) encoding a Coxeter polyhedron (in particular, polygons) in
any dimension d. The scheme of a given Coxeter polyhedron is con-
structed as follows: its vertices correspond to the faces of the polyhe-
dron, two vertices whose corresponding faces form an angle of π/m,
m ≥ 3, are joined by an edge with the weight m − 2; if two faces are
parallel, the corresponding vertices are joined by an edge with weight
∞. (Note that vertices corresponding to perpendicular edges are not
joined by an edge.)
Graphically, instead of writing the weights 2, 3, 4 on the edges of
a scheme, we draw double, triple, quadruple edges; instead of writing
∞ on an edge, we draw a very thick edge.
For example, the Coxeter scheme of the rectangle consists of two
components, each of which has two vertices joined by an edge with
weight ∞, while the scheme of an equilateral triangle has three ver-
tices joined cyclically by three edges with weights 1.
Theorem 5.5.3. The Coxeter geometries in all dimensions are clas-
sified by the connected components of their Coxeter schemes listed in
Figure 5.4.
We omit the proof (see the book [4] or, for readers of Russian, the
articles in the issue of Matematicheskoe Prosveshchenie cited above).
Many ideas of this (geometric) theory are similar to the very

important (algebraic) theory of group representation (studied in ad-
vanced algebra courses and used in modern physics). In particular,
the famous Dynkin diagrams are similar to Coxeter schemes and very
similar notation (An , Bn , Cn , E6 , etc.) is used there.
Name Coxeter scheme dim #(faces) view in R3
1
A 1 2
n
A n−1 n
n
B n−1 n
n
C n−1 n
n
D n−1 n≥5 none!
4
D 4 5 none!
F4 4 5 none!
2
G 2 3
6
E none!
7
E none!
8
E none!
Figure 5.4. Coxeter schemes for the Coxeter geometries.

5.6. Problems 107
5.6. Problems
5.1. Three planes P1 , P2 , P3 passing through the z-axis of Euclidean
space R3 are given. The angles between P1 and P2 , P2 and P3 are α
and β, respectively.
(a) Under what conditions on α and β will the group generated
by reflections with respect to the three planes be finite?
(b) If these conditions are satisfied, how can one find the funda-
mental domain of this action?
5.2. Three straight lines L1 , L2 , L3 in the Euclidean plane form a

triangle with interior angles α, β, and γ.
(a) Under what conditions on α, β, γ will the group generated by
reflections with respect to the three lines be discrete?
(b) If these conditions are satisfied, how can one find the funda-
mental domain of this action?
5.3. Consider the six lines L1 , . . . , L6 containing the six sides of

a regular plane hexagon and denote by G the group generated by
reflections with respect to these lines. Does this group determine a
Coxeter geometry?
5.4. Let F be a Coxeter triangle, let s1 , s2 , s3 be the reflections with

respect to its sides, and GF the corresponding transformation group.
(a) Give a geometric description and a description by means of
words in the alphabet s1 , s2 , s3 of all the elements of GF that leave a
chosen vertex of F fixed.
(b) Give a geometric description and a description by means of
words in the alphabet s1 , s2 , s3 of all the elements of GF which are
parallel translations.
Consider the three cases of different Coxeter triangles separately.
5.5. Draw the Coxeter schemes of

(a) all the Coxeter triangles;
(b) all the three-dimensional Coxeter polyhedra.
5.6. Prove that all the edges at each vertex of any three-dimensional
Coxeter polyhedron lie on three straight lines passing through that
vertex.
5.7. Let (F : GF ) be a Coxeter geometry of arbitrary dimension.

Prove that
(a) if s ∈ GF is the reflection in a hyperplane P , then, for any
g ∈ GF , gsg −1 is the reflection in the hyperplane gP ;
(b) any reflection from the group GF is conjugate to the reflection
in one of the faces of the polyhedron F ,
5.8. Describe some four-dimensional Coxeter polyhedron other than

the four-dimensional cube.
5.9. (a) Does the transformation group generated by the reflections

in the faces of regular tetrahedron define a Coxeter geometry?
(b) Same question for the cube.
(c) Same question for the octahedron.
(d) Same question for the dodecahedron.
Chapter 6
Spherical Geometry
So far we have studied finite and discrete geometries, i.e., geometries

in which the main transformation group is either finite or discrete. In
this chapter, we begin our study of infinite continuous geometries with
spherical geometry, the geometry (S2 :O(3)) of the isometry group of
the two-dimensional sphere, which is in fact the subgroup of all isome-
tries of R3 that map the origin to itself; O(3) is called the orthogonal
group in linear algebra courses.
But first we list the classical continuous geometries that will be
studied in this course. Some of them may be familiar to the reader,
others will be new.
6.1. A list of classical continuous geometries

Here we merely list, for future reference, several very classical ge-
ometries whose transformation groups are “continuous” rather than
finite or discrete. We will not make the intuitively clear notion of
continuous transformation group precise (this would involve defining
the so-called topological groups or even Lie groups), because we will
not study this notion in the general case: it is not needed in this
introductory course. The material of this section is not used in the
rest of the present chapter, so the reader who wants to learn about
109
110 6. Spherical Geometry
spherical geometry without delay can immediately go on to Section

6.3.
6.1.1. Finite-dimensional vector spaces over the field of real numbers

are actually geometries in the sense of Klein (the main definition of
Chapter 1). From that point of view, they can be written as
(Vn : GL(n)) ,
where Vn denotes the n-dimensional vector space over R and GL(n)

is the general linear group, i.e., the group of all nonsingular linear
transformations of Vn to itself.
The subgeometry of (Vn : GL(n)) obtained by replacing the group
GL(n) by its subgroup O(n) (consisting of orthogonal transforma-
tions) is called the n-dimensional orthonormal vector space and de-
noted
(Vn : O(n)) .
These “geometries” are rather algebraic and are usually studied in lin-
ear algebra courses. We assume that the reader has some background
in linear algebra and remembers the first basic definitions and facts
of the theory.
6.1.2. Affine spaces are, informally speaking, finite-dimensional vec-

tor spaces “without a fixed origin”. This means that their transfor-
mation groups Aff(n) contain, besides GL(n), all parallel translations
of the space (i.e., transformations of the space obtained by adding a
fixed vector to all its elements). We denote the corresponding geom-
etry by
(Vn : Aff(n)) or (Rn : Aff(n)) ,
the latter notation indicating that the elements of the space are now
regarded as points, i.e., the endpoints of the vectors (issuing from the
origin) rather than the vectors themselves. This is a more geometric
notion than that of vector space, but is also usually studied in linear
algebra courses.
6.1. A list of classical continuous geometries 111
6.1.3. Euclidean spaces are geometries that we denote
(Rn : Sym(Rn )) ;
here Sym(Rn ) is the isometry group of Euclidean space Rn , i.e., the

group of distance-preserving transformations of Rn . This group has,
as a subgroup, the orthogonal group O(n) that consists of isome-
tries leaving the origin fixed (the group O(n) should be familiar from
the linear algebra course), but also contains the subgroup of parallel
translations.
We assume that, for n = 2, 3, the reader knows Euclidean ge-

ometry from school (of course it was introduced differently, usually
via some modification of Euclid’s axioms) and is familiar with the
structure of the isometry groups of Euclidean space for n = 2, 3.
The group Sym(R2 ) of isometries of the plane is generated by
parallel translations, rotations, and mirror reflections (symmetries
with respect to a line); it contains as a subgroup the group of motions
of the plane (denoted by Sym+ (R2 ) and generated by rotations and
translations).
The group Sym+ (R3 ) is generated by parallel translations, rota-
tions about lines, and mirror reflections (symmetries with respect to
planes); it contains as a subgroup the group of motions of 3-space
(denoted by Sym+ (R3 ) and generated by rotations and translations).
Elements of Sym(R3 ) not contained in Sym+ (R3 ), e.g., mirror reflec-
tions, are orientation-reversing, i.e., they transform a right hand into
a left hand.
The reader who feels uncomfortable with elementary Euclidean
plane and space geometry can consult Chapter 0. A rigorous ax-
iomatic approach to Euclidean geometry in dimensions d = 2, 3 (based
on Hilbert’s axioms) appears in Appendix B.
Note that the transformation groups of these three geometries

(vector spaces, affine and Euclidean spaces) act on the same space
(Rn and Vn can be naturally identified), but the geometries that
they determine are different, because the four groups GL(n), O(n),
Aff(n), Sym(R3 ) are different. The corresponding geometries will not
be studied in this course; traditionally, this is done in linear algebra
courses, and we have listed them here only to draw a complete picture
of classical geometries.
Our list continues with three more classical geometries that we
will study, at least in small dimensions (mostly in dimension 2).
6.1.4. Hyperbolic spaces Hn (called Lobachevsky spaces in Russia)

are “spaces of constant negative curvature” (you will learn what this
means much later, in differential geometry courses) with transforma-
tion group the isometry group of the hyperbolic space (i.e., the group
of transformations preserving the “hyperbolic distance”). We will
only study the hyperbolic space of dimension n = 2, i.e., the hyper-
bolic plane. Three models of H2 will be studied, in particular, the
Poincaré disk model,
(H2 : M) ;
here H2 := {(x, y) ∈ R2 | x2 + y 2 < 1} is the open unit disk and

M is the group of Möbius transformations (the definition appears in
Chapter 7), which take the disk to itself.
We will also study two other models of hyperbolic plane geometry
(the half-plane model, also due to Poincaré, and the Cayley–Klein
model). A special chapter describes how attempts to prove Euclid’s
Fifth Postulate led to the appearance of hyperbolic plane geometry
and the dramatic history of its creation by Gauss, Lobachevsky, and
Bolyai.
6.1.5. Elliptic spaces Elln are “spaces of constant positive curvature”

(what this means is explained in differential geometry courses). We
will only study the two-dimensional case, i.e., the elliptic plane, in
the present chapter after we are done with spherical geometry, which
is the main topic of this chapter, but can also be regarded as the
principal building block of elliptic plane geometry.
6.1.6. Projective spaces RP n are obtained from affine spaces by

“adding points at infinity” in a certain way, and taking, for the trans-
formation group, a group of linear transformations on the so-called
“homogeneous coordinates” of points (x1 : · · · : xn : xn+1 ) ∈ RP n .
6.2. Some basic facts from Euclidean plane geometry 113
We can write this geometry as
(RP n : Proj(n)) .
For arbitrary n, projective geometry is usually studied in linear alge-

bra courses. We will study the projective plane RP 2 in this course,
and only have a quick glance at projective space RP 3 (see Chapter
12).
6.2. Some basic facts from Euclidean plane

geometry
Here we list several fundamental facts of Euclidean plane geometry
(including modern formulations of some of Euclid’s postulates) in
order to compare and contrast them with the corresponding facts of
spherical, elliptic, and hyperbolic geometry.
I. There exists a unique (straight) line passing through two given
distinct points.
II. There exists a unique perpendicular to a given line passing
through a given point. (A perpendicular to a given line is a line forming
four equal angles, called right angles, with the given one.)
III. There exists a unique circle of given center and given radius.
IV. Given a point on a line and any positive number, there exist
exactly two points on the line whose distance from the given point is
equal to the given number.
V. There exists a unique parallel to a given line passing through
a given point not on the given line. (A parallel to a given line is a
line without common points with the given one.) This is the modern
version of Euclid’s Fifth Postulate, sometimes described as the single
most important and controversial scientific statement of all time.
VI. The parameters of a triangle ABC, namely the angles α, β, γ
at the vertices A, B, C and the sides a, b, c opposite to these vertices,
satisfy the following formulas.
(i) Angle sum formula: α + β + γ = π.
(ii) Sine formula:

a b c
= = .
(iii) Cosine formula: c2 = a2 + b2 − 2ab cos γ.
6.3. Lines, distances, angles, polars, and

perpendiculars
Let S2 be the unit sphere in R3 :
S2 := {(x, y, z) ∈ R3 | x2 + y 2 + z 2 = 1};
our present aim is to study the geometry (S2 : O(3)), where O(3) is
the orthogonal group (i.e., the group of isometries of R3 leaving the
origin in place), which obviously takes S2 to itself.
6.3.1. Basic definitions. By a line on the sphere we mean a great

circle, i.e., the intersection of S2 with a plane passing through the
sphere’s center. For example, the equator of the sphere, as well as
any meridian, is a line. The angle between two lines is defined as the
dihedral angle (measured in radians) between the two planes contain-
ing the lines. For example, the angle between the equator and any
meridian is π/2. The distance between two points A and B is defined
as the measure (in radians) of the angle AOB. Thus the distance be-
tween the north and south Poles is π, the distance between the south
pole and any point on the equator is π/2.
Obviously, the transformation group O(3) preserves distances be-
tween points. It can also be shown (we omit the proof) that, con-
versely, O(3) can be characterized as the group of distance-preserving
transformations of the sphere (distance being understood in the spher-
ical sense, i.e., as explained above).
6.3.2. Poles, polars, perpendiculars, circles. Let us look at the

analogs in spherical geometry of the Euclidean postulates.
IS . There exist a unique line passing through two given distinct
points, except when the two points are antipodal, in which case there
are infinitely many. All the meridians joining the two poles give an
example of this exceptional situation.
6.3. Lines, distances, angles, polars, and perpendiculars 115
IIS . There exists a unique perpendicular to a given line passing

through a given point, except when the point lies at the intersection of
the perpendicular constructed from the center O of the sphere to the
plane in which the line lies, in which case there are infinitely many
such perpendiculars. The exceptional situation is exemplified by the
equator and, say, the north pole: all the meridians (which all pass
through the pole) are perpendicular to the equator (Figure 6.1).
More generally, the polar of a point P is the (spherical) line ob-
tained by cutting the sphere by the plane passing through O and per-
pendicular to the (Euclidean!) straight line P O. Conversely, given a
(spherical) line l, the poles of that line are the two antipodal points
Pl and Pl for which the (Euclidean) line Pl Pl is perpendicular to the
plane determined by l. The assertion IIS may now be restated as
follows: there exists a unique perpendicular to a given line passing
through a given point, except when the point is a pole of that line, in
which case all the lines passing through the pole are perpendicular to
the given line.
IIIS . There exists a unique circle of given center C and given

radius ρ, provided 0 < ρ < π. It is defined as the set of points
whose (spherical) distance from C is equal to ρ. It is easy to see that
any (spherical) circle is actually a Euclidean circle, namely the one
obtained as the intersection of the sphere with the plane perpendicular
to the Euclidean line OC and passing through the point I on that line
Figure 6.1. Perpendiculars, poles, and polars.

such that OI = cos ρ. Note that the radius of the Euclidean circle
will be less than ρ.
Given a spherical circle of center C and radius ρ, note that it can
be regarded as the circle of radius π − ρ and center C , where C is
the antipode of C. Further, note that the longest circle centered at
C is the polar of the point C; its radius is π/2.
IVS . Given a point on a line and any positive number, there
exist exactly two points on the line whose distance from the given
point equals the given number, provided the number is less than π.
VS . Any two lines intersect in two antipodal points, i.e., in two
points symmetric with respect to the center of the sphere S2 . There-
fore there are no parallel lines in spherical geometry. If two points
A, B are not antipodal, then there is only one line joining them and
one shortest line segment with endpoints at A and B. For opposite
points, there is an infinity of lines joining them (for the north and
south poles, these lines are the meridians).
6.3.3. Lines as shortest paths. It is proved in differential geome-

try courses that spherical lines are geodesics, i.e., they are the shortest
paths between two points. To do this, one defines the length of a curve
as a curvilinear integral and uses the calculus of variations to show
that the curve (on the sphere) of minimum length joining two given
points is indeed the arc of the great circle containing these points.
6.4. Biangles and triangles in S2

6.4.1. Biangles. Two lines l and m on the sphere intersect in two
(antipodal) points P and P and divide the sphere into four domains;
each of them is called a biangle, it is bounded by two halves of the
lines l and m, called its sides, and has two vertices (the points P and
P ). The four domains form two congruent pairs; two biangles from a
congruent pair touch each other at the common vertices P and P , and
have the same angle at P and P . The main parameter of a biangle
is the measure α of the angle between the lines that determine it; if
α
= π/2, the two biangles not congruent to the biangle of measure α
are called complementary, their angle is π − α. Note that the angle
6.4. Biangles and triangles in S2 117
measure α determines the corresponding biangle up to an isometry of

the sphere.
6.4.2. Areas of figures on the sphere. In order to correctly mea-

sure areas of figures on the plane, on the sphere, or on other surfaces,
one must define what an area is, specify what figures are measur-
able (i.e., possess an area), and devise methods for computing areas.
For the Euclidean plane, there are several approaches to area: many
readers have probably heard of the theory of Jordan measure; more
advanced readers may have studied Lebesgue measure; readers who
have taken multivariable calculus courses know that areas may be
computed by means of double integrals.
In this book, we will not develop a rigorous measure theory for
the geometries that we study. In this subsection, we merely sketch
an axiomatic approach for determining areas of spherical figures; this
approach is similar to Jordan measure theory in the Euclidean plane.
The theory says that there is a family of sets in S2 , called measurable,
satisfying the following axioms.
(i) Invariance. Two congruent measurable figures have the same
area.
(ii) Normalization. The whole sphere is measurable and its area
is 4π.
(iii) Countable additivity. If a measurable figure F is the union
of a countable family of measurable figures {Fi } without common
interior points, then its area is equal to the sum of areas of the figures
Fi .
An obvious consequence of these axioms is that the area of the
northern hemisphere is 2π, while each of the triangles obtained by
dividing the hemisphere into four equal parts is of area π/2.
6.4.3. Area of the biangle. From the axioms formulated in the

previous subsection, it is easy to deduce that the area Sπ/2 of a biangle
with angle measure π/2 is π. Indeed, the sphere is covered by four
such nonoverlapping biangles, which are congruent to each other; they
have the same area by (i), the sum of their areas is that of the sphere
by (iii), and the latter is 4π by (ii), whence Sπ/2 = (4π)/4 = π.
For the case in which the angle measure α of a biangle is a rational

multiple of π, a similar argument shows that
(6.1) Sα = 2α .
This formula is actually true for any α, but for the case in which
π/α is irrational, its proof requires a passage to the limit based on an
additional “continuity axiom” that we have not explicitly stated. We
therefore omit the proof, but will use the above formula for all values
of α in what follows.
6.4.4. Area of the triangle. Let A, B, C be three distinct points

of S2 , no two of which are opposite. The union of the shortest line
segments joining the points A and B, B and C, C and A is called
the triangle ABC. For a triangle ABC, we always denote by α, β, γ
the measure of the angles at A, B, C, respectively, and by a, b, c the
lengths of the sides opposite to A, B, C (recall that the length a of
BC is equal to the measure of the angle BOC in R3 ).
Theorem 6.4.5. The area SABC of a spherical triangle with angles

α, β, γ is equal to
SABC = α + β + γ − π .
Proof. There are 12 spherical biangles formed by pairs of lines AB,

BC, CA. Choose six of them, namely those that contain triangle
ABC or the antipodal triangle A1 B1 C1 formed by the three points
antipodal to A, B, C. Denote their areas by
SI , SI , SII , SII

, SIII , SIII .
Each point of the triangles ABC and A1 B1 C1 is covered by exactly
three of the chosen six biangles, while the other points of the sphere
are covered by exactly one such biangle (we ignore the points on the
lines). Therefore, using relation (6.1), we can write
4π = SI + SI + SII + SII

+ SIII + SIII − 2SABC − 2SA1 B1 C1
= 2α + 2β + 2γ + 2α + 2β + 2γ − 2SABC − 2SA1 B1 C1
= 4(α + β + γ) − 4SABC ,
6.4. Biangles and triangles in S2 119
Figure 6.2. Area and sine theorem for the triangle.
because the two triangles ABC and A1 B1 C1 have the same area (since
they are congruent). Clearly, the previous formula implies the re-
quired equality.
This theorem has the following fundamental consequence.

Corollary 6.4.6. The sum of angles of any spherical triangle is more
than π.
The analog of the sine formula for the Euclidean triangle is the
following statement about spherical triangles.
Theorem 6.4.7 (The spherical sine theorem).
sin a sin b sin c
= = .
In order to establish this formula, we will use the following state-

ment, sometimes called the “theorem of the three perpendiculars”.
Lemma 6.4.8. Let A ∈ R3 be a point outside a plane P, let K be its
perpendicular projection on P and let L be its perpendicular projection
on a line l contained in P. Then KL is perpendicular to l.
Proof of the lemma. The line l is perpendicular to the plane AKL

because it is perpendicular to two nonparallel lines of AKL, namely
to AL and AK (to the latter since AK is orthogonal to any line in

P). Therefore l is perpendicular to any line of the plane AKL, and
in particular, to LK.
Proof of the theorem. Let H be the projection of A on the plane

BOC, let L and M be the projections of A on the lines OB and OC.
Then by the lemma, L and M coincide with the projections of H on
OB and OC. Therefore,
AH = LA sin β = sin c sin β, AH = M A sin γ = sin b sin γ.
Thus, sin b : sin β = sin c : sin γ. Similarily, by projecting C on the

plane AOB and arguing as above, we obtain sin b : sin β = sin a :
sin α. This is the required equality.
6.5. Other theorems about triangles

In this section, we state a few more theorems about spherical trian-
gles. Their proofs are relegated to the problems appearing at the end
of this chapter.
Theorem 6.5.1 (First cosine theorem).
cos a = cos b cos c + sin b sin c cos α .
Theorem 6.5.2 (Second cosine theorem).
cos α + cos β cos γ = sin β sin γ cos α .
Corollary 6.5.3 (Analog of the Pythagorean theorem). If triangle

ABC has a right angle at C, then
cos c = cos a cos b .
Theorem 6.5.4. The medians of any triangle intersect at a single

point.
Theorem 6.5.5. The altitudes of any triangle intersect at a single

point.
6.7. Two-dimensional elliptic geometry 121
6.6. Coxeter triangles on the sphere S2

We will not develop the theory of tilings on the sphere S2 and Coxeter
geometry on the sphere in full generality, but only consider Coxeter
triangles, i.e., spherical triangles all of whose angles are of the form
π/m, m = 2, 3, . . . . It follows from Theorem 6.4.5 that any spherical
Coxeter triangle (π/p, π/q, π/r), N copies of which cover the sphere,
must satisfy the Diophantine equation
N/p + N/q + N/r = N + 4.
The transformation group of the corresponding Coxeter geometry is
finite, and so Theorem 3.2.6 tells us what group it has to be: it must
be either one of the dihedral groups, or the tetrahedral, hexahedral,
or dodecahedral group. The dihedral groups yield an obvious infinite
series of tilings, one of which is shown in Figure 6.3.
The three other groups yield three possibilities for N , namely
N = 24, 48, 120, and we easily find the corresponding values of
(p, q, r) in each of the three cases. Finally, the solutions of our Dio-
phantine equation are:
(2, 3, 3), (2, 3, 4), (2, 3, 5), (2, 2, n) for n = 2, 3, . . . .
The corresponding tilings of the sphere (and their Coxeter

schemes) are shown in Figure 6.3.
6.7. Two-dimensional elliptic geometry

6.7.1. Spherical geometry is closely related to the elliptic geometry
invented by Riemann. Elliptic geometry is obtained from spherical
geometry by “identifying opposite points of S2 ”. The precise defini-
tion can be stated as follows. Consider the set Ell2 whose elements
are pairs of antipodal points (x, −x) on the unit sphere S2 ⊂ R3 . The
group O(3) acts on this set (because isometries of S2 take antipodal
pairs of points to antipodal pairs), thus defining a geometry in the
sense of Klein (Ell2 : O(3)), which we call two-dimensional elliptic
geometry.
Lines in elliptic geometry are defined as great circles of the sphere
S2 , angles and distances are defined as in spherical geometry, and
Figure 6.3. Four Coxeter tilings of the sphere.
the trigonometry of triangles in elliptic geometry is the same as in

spherical geometry. More generally, one can say that elliptic geometry
is locally the same as spherical, but these geometries are drastically
different globally. In particular, in elliptic geometry
• one and only line passes through any two distinct points;
• for a given line and any given point (except one, called the pole
of that line) there exists a unique perpendicular to that line passing
through the point.
6.8. Problems 123
The relationship between the two geometries is best expressed by

the following statement, which yields simple proofs of the statements
about elliptic geometry made above.
Theorem 6.7.2. There exists a surjective morphism
D : (S2 : O(3)) → (Ell2 : O(3))
of spherical geometry onto elliptic geometry which is a local isomor-
phism (in the sense that any domain contained in a half-sphere is
mapped bijectively and isometrically onto its image).
Proof. The map D is the obvious one: D : x → (x, −x), while the
homomorphism of the transformation groups is the identity isomor-
phism. All the assertions of the theorem are immediate.
As we noted before, globally the two geometries are very different.

Being metric spaces, they are topological spaces (in the metric topol-
ogy) which are not even homeomorphic: one is a two-sided surface
(S2 ), the other (RP 2 ) is one-sided (it contains a Möbius strip).
6.8. Problems
In all the problems below a, b, c are the sides and α, β, γ are the oppo-
site angles of a spherical triangle. The radius of the sphere is R = 1.
6.1. Prove the first cosine theorem on the sphere S2 :

cos a = cos b cos c + sin b sin c cos α.
6.2. Prove the second cosine theorem on the sphere S2 :

cos α + cos β cos γ = sin β sin γ cos a.
6.3. Prove that a + b + c < 2π.
6.4. Does the Pythagorean theorem hold in spherical geometry?

Prove the analog of that theorem stated in Corollary 6.5.3.
6.5. Does the Moscow–New York flight fly over Spain? Over Green-
land? Check your answer by stretching a thin string between Moscow
and New York on a globe.
6.6. Find the infimum and the supremum of the sum of the angles of
an equilateral triangle on the sphere.
6.7. The city A is located at the distance 1000km from the cities B
and C; the trajectories of the flights from A to B and from A to C
are perpendicular to each other. Estimate the distance between B
and C. (You can take the radius of the Earth equal to 6400km.)
6.8*. Find the area of the spherical disk of radius r (i.e., the domain
bounded by a spherical circle of radius r).
6.9. Find fundamental domains for the actions of the isometry groups
of the tetrahedron, the cube, the dodecahedron, and the icosahedron
on the 2-sphere and indicate the number of their images under the
corresponding group action.
6.10. Prove that any spherical triangle has a circumscribed and an

inscribed circle.
6.11. Prove that the medians of a spherical triangle intersect at one

point.
6.12. Prove that the altitudes of a spherical triangle always intersect

at one point.
6.13. Suppose that the medians and the altitudes of a spherical

triangle interest at the points M and A, respectively. Can it happen
that M = A?
Chapter 7
The Poincaré Disk

Model of Hyperbolic
Geometry
In this chapter, we begin our study of the most popular of the non-
Euclidean geometries – hyperbolic geometry, concentrating on the
case of dimension two. We avoid the intricacies of the axiomatic
approach and define hyperbolic plane geometry via the beautiful
Poincaré disk model, which is the geometry of the disk determined
by the action of a certain transformation group acting on the disk
(namely, the group generated by reflections in circles orthogonal to
the boundary of the disk).
In order to describe the model, we need some facts from Eu-
clidean plane geometry, which should be studied in high school, but,
unfortunately, in most cases, are not. So we begin by recalling some
properties of inversion (which will be the main ingredient of the trans-
formation group of our geometry) and some constructions related to
orthogonal circles in the Euclidean plane. We then establish the ba-
sic facts of hyperbolic plane geometry and finally digress, following
Poincaré’s argumentation from his book Science et Hypothèse (for the
English version, see [12]) about epistomological questions relating this
geometry (and other geometries) to the physical world.
125
126 7. The Poincaré Disk Model of Hyperbolic Geometry
7.1. Inversion and orthogonal circles

7.1.1. Inversion and its properties. The main tool that we will
need in this chapter is inversion, a classical transformation from ele-
mentary plane geometry. Denote by R the plane R2 with an added
extra point (called the point at infinity and denoted by ∞). The set
R := R2 ∪ ∞ can also be interpreted as the complex numbers C with
the “point at infinity” added; it is then called the Riemann sphere
and denoted by C.
An inversion with center O ∈ R2 and radius r > 0 is the trans-
formation of R that maps each point M to the point N on the ray
OM so that
(7.1) |OM | · |ON | = r 2
and interchanges the points O and ∞. Sometimes inversions are called

reflections with respect to the circle of inversion, i.e., the circle of
radius r centered at O.
Figure 7.1. Inversion |OM | · |ON | = r 2 .
There is a simple geometric way of constructing the image of a

point M under an inversion with center O and radius r: draw the
circle of inversion, draw the perpendicular to OM from M to its
intersection point T with the circle and construct the tangent to the
7.1. Inversion and orthogonal circles 127
circle at T to its intersection point N with the ray OM ; then N will

be the image of M under the given inversion. Indeed, the two right
triangles OM T and OT N are similar (they have a common acute
angle at O), and therefore
|OM | |OT |
= ,
|OT | |ON |
and since |OT | = r, we obtain (7.1).
If the extended plane R is interpreted as the Riemann sphere C,
then an example of an inversion (with center O and radius 1) is the
map z → 1/z, where the bar over z denotes complex conjugation.
It follows immediately from the definition that inversions are bi-
jections of R = C that leave the points of the circle of inversion in
place, “turn the circle inside out” in the sense that points inside the
circle are taken to points outside it (and vice versa), and are involu-
tions (i.e., the composition of an inversion with itself is the identity).
Further, inversions possess the following important properties.
(i) Inversions map any circle or straight line orthogonal to the
circle of inversion into itself. Look at Figure 7.2, which shows two
orthogonal circles CO and CI of centers O and I, respectively.
CO
CI
Figure 7.2. Orthogonal circles.
It follows from the definition of orthogonality that the tangent

from the center O of CO to the other circle CI passes through the
intersection point T of the two circles. Now let us consider the in-
version with center O and radius r = |OT |. According to property
(i) above, it takes the circle CI to itself; in particular, the point M is
mapped to N , the point T (as well as the other intersection point of

the two circles) stays in place, and the two arcs of CI cut out by CO
are interchanged. Note further that, vice versa, the inversion in the
circle CI transforms CO in an analogous way.
(ii) Inversions map any circle or straight line into a circle or

straight line. In particular, lines passing through the center of inver-
sion are mapped to themselves (but are “turned inside out” in the
sense that O goes to ∞ and vice versa, while the part of the line
inside the circle of inversion goes to the outside part and vice versa);
circles passing through the center of inversion are taken to straight
lines, while straight lines not passing through the center of inversion
are taken to circles passing through that center (see Figure 7.3).
Figure 7.3. Images of circles and lines under inversion.
(iii) Inversions preserve (the measure of ) angles; here by the mea-

sure of an angle formed by two intersecting curves we mean the or-
dinary (Euclidean) measure of the angle formed by their tangents at
the intersection point.
7.1. Inversion and orthogonal circles 129
The (elementary) proofs of properties (i)–(iii) are left to the

reader (see Problems 7.1–7.3).
7.1.2. Construction of orthogonal circles. We have already

noted the important role that orthogonal circles play in inversion
(see 7.1.1(i)). Here we will describe several constructions of orthogo-
nal circles that will be used in subsequent sections.
Lemma 7.1.3. Let A be a point inside a circle C centered at some
point O; then there exists a circle orthogonal to C such that the re-
flection in this circle takes A to O.
Proof. From A draw the perpendicular to line OA to its intersection

T with the circle C (see Figure 7.4).
Figure 7.4. Inversion taking an arbitrary point A to O.
Draw the tangent to C at T to its intersection at I with OA.

Then the circle of radius IT centered at I is the one we need. Indeed,
the similar right triangles IAT and IT O yield |IA|/|IT | = |IT |/|IO|,
whence we obtain |IA| · |IO| = |IT |2 , which means that O is the
reflection of A in the circle of radius |IT | centered at I, as required.

Corollary 7.1.4. (i) Let A and B be points inside a circle CO not ly-
ing on the same diameter; then there exists a unique circle orthogonal
to CO and passing through A and B.
(ii) Let A be a point inside a circle CO and P a point on CO , with

A and P not lying on the same diameter; then there exists a unique
circle orthogonal to CO passing through A and P .
(iii) Let P and Q be points on a circle CO of center O such that
P Q is not a diameter; then there exists a unique circle C orthogonal
to CO and passing through P and Q.
(iv) Let A be a point inside a circle CO of center O and D a
circle orthogonal to CO ; then there exists a unique circle C orthogonal
to both CO and D and passing through A.
Proof. To prove (i), we describe an effective step-by-step construc-

tion, which can be carried out by ruler and compass, yielding the
required circle. The construction is shown in Figure 7.5, with the
numbers in parentheses near each point indicating at which step the
point was obtained.
Figure 7.5. Circle orthogonal to CO containing A, B.
First, we apply Lemma 7.1.3, to define an inversion ϕ taking A to

the center O of the given circle; to do this, we draw a perpendicular
from A to OA to its intersection T (1) with C, then draw the perpen-
dicular to OT from T to its intersection I(2) with OA; the required
7.2. Definition of the disk model 131
inversion is centered at I and is of radius |IT |. Joining B and I, we

construct the tangent BS(3) to the circle of the inversion ϕ and find
the image B (4) of B under ϕ by dropping a perpendicular from S to
IB.
Next, we draw the line B O and obtain the intersection points
M, N of this line with the circle of the inversion ϕ. Finally, we draw
the circle C passing through the points M, N, I. Then C “miracu-
lously” passes through A and B and is orthogonal to CO ! Of course,
there is no miracle in this: C passes through A and B because it is
the inverse image under ϕ of the line OB (see 7.1.1(ii)), and it is
orthogonal to CO since so is OB (see 7.1.1(iii)).
Uniqueness is obvious in the case A = O and follows in the general
case by 7.1.1(ii)–(iii).
The proof of (ii) is analogous: we send A to O by an inversion ϕ,
join O and ϕ(P ) and continue the argument as above.
To prove (iii), construct lines OP and OQ, draw perpendiculars
to these lines from P and Q, respectively, and denote by I their
intersection point. Then the circle of radius |IP | centered at I is the
required one. Its uniqueness is easily proved by contradiction.
To prove (iv), we again use Lemma 7.1.3 to construct an inversion
ϕ that takes C0 to itself and sends A to O. From the point O, we
draw the (unique) ray R orthogonal to ϕ(L). Then the circle ϕ−1 (R)
is the required one.
7.2. Definition of the disk model

7.2.1. The disk model of the hyperbolic plane is the geometry
(H2 : M) whose points are the points of the open disk

H2 := (x, y) ∈ R2 | x2 + y 2 < 1 ,
and whose transformation group M is the group generated by reflec-
tions in all the circles orthogonal to the boundary circle
A := {(x, y) : x2 + y 2 = 1}
of H2 , and by reflections in all the diameters of the circle A. Now
M is indeed a transformation group of H2 : the discussion in 7.1.1
implies that a reflection of the type considered takes points of H2

to points of H2 and, being its own inverse, we have the implication
ϕ ∈ M =⇒ ϕ−1 ∈ M.
We will often call H2 the hyperbolic plane. The boundary circle
A (which is not part of the hyperbolic plane) is called the absolute.
7.2.2. We will see later that M is actually the isometry group of

hyperbolic geometry with respect to the hyperbolic distance, which
will be defined in the next chapter. We will see that although the
Euclidean distance between points of H2 is always less than 2, the
hyperbolic plane is unbounded with respect to the hyperbolic dis-
tance. Endpoints of a short segment (in the Euclidean sense!) near
the absolute are very far away from each other in the sense of hyper-
bolic distance.
Figure 7.6 gives an idea of what an isometric transformation (the
simplest one – a reflection in a line) does to a picture. Note that
from our Euclidean point of view, the reflection changes the size and
the shape of the picture, whereas from the hyperbolic point of view,
the size and shape of the image are exactly the same as those of the
original. It should also be clear that hyperbolic reflections reverse
orientation, e.g., the image of a right hand under reflection will look
like a left hand, but of somewhat different size and shape.
Figure 7.6. An isometry in the hyperbolic plane.

7.3. Points and lines in the hyperbolic plane 133
7.3. Points and lines in the hyperbolic plane

7.3.1. First we define points of the hyperbolic plane simply as points
of the open disk H2 . We then define the lines on the hyperbolic plane
as the intersections with H2 of the (Euclidean) circles orthogonal to
the absolute as well as the diameters (without endpoints) of the ab-
solute (see Figure 7.7).
Figure 7.7. Lines on the hyperbolic plane.
Note that the endpoints of the arcs and the diameters do not
belong to the hyperbolic plane: they lie in the absolute, whose points
are not points of our geometry.
Thus the hyperbolic plane, as well as the lines in it, is not com-
2
pact. Its compactification (or closure) is the compact disk H .
Figure 7.7 shows that some lines intersect in one point, others
have no common points, and none have two common points (unlike
lines in spherical geometry). This is not surprising, because we have
the following statement.
Theorem 7.3.2. One and only one line passes through any pair of
distinct points of the hyperbolic plane.
Proof. The theorem immediately follows from Corollary 7.1.4(i).

7.4. Perpendiculars
7.4.1. Two lines in H2 are called perpendicular if they are orthogonal
in the sense of elementary Euclidean geometry. When both are diam-
eters, they are perpendicular in the usual sense, when both are arcs
of circles, they have perpendicular tangents at the intersection point,
and when one is an arc and the other, a diameter, then the diameter
is perpendicular to the tangent to the arc at the intersection point.
Theorem 7.4.2. There is one and only one line passing through a
given point and perpendicular to a given line.
Proof. The theorem immediately follows from Corollary 7.1.4(iv).

7.5. Parallels and nonintersecting lines

7.5.1. Let l be a line and P a point of the hyperbolic plane H2 not
contained in the line l. Denote by A and B the points at which l
intersects the absolute. Consider the lines k = P A and m = P B and
denote their second intersection points with the absolute by A and
B . Clearly, the lines k and m do not intersect l. Moreover, any line
passing through P between k and m (i.e., any line containing P and
joining the arcs AB and BA ) does not intersect l. The lines AP A
and BP B are called parallels to l through P , and the lines between
them are called nonintersecting lines with l.
Figure 7.8. Perpendiculars and parallels.

7.6. Sum of the angles of a triangle 135
We have proved the following statement.

Theorem 7.5.2. There are infinitely many lines passing through a
given point P not intersecting a given line l if P ∈
/ l. These lines are
all located between the two parallels to l.
This theorem contradicts Euclid’s famous Fifth Postulate, which,

in its modern formulation, says that one and only one parallel to a
given line passes through a given point. For more than two thou-
sand years, many attempts to prove that the Fifth Postulate follows
from Euclid’s other postulates (which, unlike the Fifth Postulate,
were simple and intuitively obvious) were made by mathematicians
and philosophers. Had such a proof been found, Euclidean geome-
try could have been declared to be an absolute truth both from the
physical and the philosophical points of view; it would have been
an example of a set of facts that the German philosopher Kant in-
cluded in the category of synthetic a priori. For two thousand years,
the naive belief among scientists in the absolute truth of Euclidean
geometry made it difficult for the would be discoverers of other ge-
ometries to realize that they had found something worthwhile. Thus
the appearance of a consistent geometry in which the Fifth Postu-
late does not hold was not only a crucial development in the history
of mathematics, but one of the turning points in the philosophy of
science. In this connection, see the discussion in Chapter 11.
7.6. Sum of the angles of a triangle

7.6.1. Consider three points A, B, C not on one line. The three
segments AB, BC, CA (called sides) form a triangle with vertices
A, B, C. The angles of the triangle, measured in radians, are de-
fined as equal to the (Euclidean measure of the) angles between the
tangents to the sides at the vertices.
Theorem 7.6.2. The sum of the angles α, β, γ of a triangle ABC is
less than two right angles:
α+β+γ <π .
Proof. In view of Lemma 7.1.3, we can assume, without loss of gen-

erality, that A is O (the center of H2 ). But then if we compare the
hyperbolic triangle OBC with the Euclidean triangle OBC, we see

that they have the same angle at O, but the Euclidean angles at B and
C are larger than their hyperbolic counterparts (look at Figure 7.9),
which implies the claim of the theorem.
Figure 7.9. Sum of the angles of a hyperbolic triangle.
It is easy to see that very small triangles have angle sums very
close to π; in fact, the least upper bound of the angle sum of hy-
perbolic triangles is exactly π. Further, the greatest lower bound of
these sums is 0. To see this, divide the absolute into three equal arcs
by three points P, Q, R and construct three circles orthogonal to the
absolute passing through the pairs of points P and Q, Q and R, R
and P . These circles exist by Corollary 7.1.4(iii). Then all the angles
of the “triangle” P QR are zero, so its angle sum is zero. Of course,
P QR is not a real triangle in our geometry (its vertices, being on the
absolute, are not points of H2 ), but if we take three points P , Q , R
close enough to P, Q, R, then the angle sum of triangle P Q R will
be less than any prescribed ε > 0.
7.7. Rotations and circles in the hyperbolic

plane
We mentioned previously that distance between points of the hyper-
bolic plane will be defined later. Recall that the hyperbolic plane is
7.7. Rotations and circles in the hyperbolic plane 137
Figure 7.10. Ordinary triangle and “triangle” with angle

sum 0.
the geometry (H2 : M), in which, by definition, M is the transfor-

mation group generated by all reflections in all the lines of H2 . If
we take the composition of two reflections in two intersecting lines,
then what we get should be a “rotation”, but we can’t assert that at
this point, because we don’t have any definition of rotation: the usual
(Euclidean) definition of a rotation or even that of a circle cannot be
given until distance is defined.
But the notions of rotation and of circle can be defined without
appealing to distance in the following natural way: a rotation about a
point P ∈ H2 is, by definition, the composition of any two reflections
in lines passing through P . If I and A are distinct points of H2 , then
the (hyperbolic) circle of center I and radius IA is the set of images
of A under all rotations about I.
Theorem 7.7.1. A (hyperbolic) circle in the Poincaré disk model is

a Euclidean circle, and vice versa, any Euclidean circle inside H2 is
a hyperbolic circle in the geometry (H2 : M).
Proof. Let C be a circle of center I and radius IA in the geometry

(H2 : M). Using Lemma 7.1.3, we can send I to the center O of H2
by a reflection ϕ. Let ρ be a rotation about I determined by two lines
l1 and l2 . Then the lines d1 := ϕ(l1 ) and d2 := ϕ(l2 ) are diameters
of the absolute and the composition of reflections in these diameters
is a Euclidean rotation about O (and simultaneously a hyperbolic

one). This rotation takes the point ϕ(A) to a point on the circle C
of center O and radius Oϕ(A), which is simultaneously a hyperbolic
and Euclidean circle. Now by Corollary 7.1.4(i), the inverse image
ϕ−1 (C ) will be a (Euclidean!) circle. But ϕ−1 (C ) coincides with C
by construction, so C is indeed a Euclidean circle in our model.
The proof of the converse assertion is similar and is left to the
reader (see Problem 7.7).
7.8. Hyperbolic geometry and the physical

world
In his famous book Science et Hypothèse, Henri Poincaré describes
the physics of a small “universe” and the physical theories that its
inhabitants would create. The universe considered by Poincaré is
Euclidean, plane (two-dimensional), and has the form of an open unit
disk. Its temperature is 100◦ Farenheit at the center of the disk and
decreases linearly to absolute zero at its boundary. The lengths of
objects (including living creatures) are proportional to temperature.
How will a little flat creature endowed with reason and living
in this disk describe the main physical laws of his universe? The
first question he/she may ask could be: Is the world bounded or
infinite? To answer this question, an expedition is organized; but as
the expedition moves towards the boundary of the disk, the legs of
the explorers become smaller, their steps shorter – they will never
reach the boundary, and conclude that the world is infinite.
The next question may be: Does the temperature in the uni-
verse vary? Having constructed a thermometer (based on different
expansion coefficients of various materials), scientists carry it around
the universe and take measurements. However, since the lengths of
all objects change similarly with temperature, the thermometer gives
the same measurement all over the universe – the scientists conclude
that the temperature is constant.
Then the scientists might study straight lines, i.e., investigate
what is the shortest path between two points. They will discover
that the shortest path is what we perceive to be the arc of the circle
7.9. Problems 139
containing the two points and orthogonal to the boundary disk (this
is because such a circular path brings the investigator nearer to the
center of the disk, and thus increases the length of his steps). Further,
they will find that the shortest path is unique and regard such paths
as “straight lines”.
Continuing to develop geometry, the inhabitants of Poincaré’s
little flat universe will decide that there is more than one parallel to a
given line passing through a given point, the sum of angles of triangles
is less than π, and obtain other statements of hyperbolic geometry.
Thus they will come to the conclusion that they live in an infi-
nite flat universe with constant temperature governed by the laws of
hyperbolic geometry. But this is not true – their universe is a finite
disk, its temperature is variable (tends to zero towards the boundary)
and the underlying geometry is Euclidean, not hyperbolic!
The philosophical conclusion of Poincaré’s argument is not ag-
nosticism – he goes further. The physical model described above,
according to Poincaré, shows not only that the truth about the uni-
verse cannot be discovered, but that it makes no sense to speak of
any “truth” or approximation of truth in science – pragmatically, the
inhabitants of his physical model are perfectly right to use hyperbolic
geometry as the foundation of their physics because it is convenient,
and it is counterproductive to search for any abstract Truth which
has no practical meaning anyway.
This conclusion has been challenged by other thinkers, but we
will not get involved in this philosophical discussion.
7.9. Problems
7.1. Prove that inversion maps circles and straight lines to circles or
straight lines.
7.2. Prove that inversion maps any circle orthogonal to the circle of
inversion into itself.
7.3. Prove that inversion is conformal (i.e., it preserves the measure

of angles).
7.4. Prove that if P is a point lying outside a circle γ and A, B are

the intersection points with the circle of a line l passing through P ,
then the product |P A| · |P B| (also called the power of P with respect
to γ) does not depend on the choice of l.
7.5. Prove that if P is a point lying inside a circle γ and A, B are

the intersection points with the circle of a line l passing through P ,
then the product |P A| · |P B| (also called the power of P with respect
to γ) does not depend on the choice of l.
7.6. Prove that inversion with respect to a circle orthogonal to a

given circle C maps the disk bounded by C bijectively onto itself.
7.7. Prove that any Euclidean circle inside the disk model is also a
hyperbolic circle. Does the ordinary (Euclidean) center coincide with
its “hyperbolic center”?
Figure 7.11. A pattern of lines in H2 .

7.9. Problems 141
7.8. Study Figure 7.11. Does it demonstrate any tilings of H2 by

regular polygons? Of how many sides? Do you discern a Coxeter
geometry in this picture with “hyperbolic Coxeter triangles” as fun-
damental domains? What are their angles?
7.9. Prove that any inversion of C preserves the cross ratio of four
points:
z3 − z1 z4 − z1
z1 , z2 , z3 , z4 := : .
z3 − z2 z4 − z2
7.10*. Using complex numbers, invent a formula for the distance

between points on the Poincaré disk model and prove that “symmetry
with respect to straight lines” (i.e., inversion) preserves this distance.
7.11. Prove that hyperbolic geometry is homogeneous in the sense

that for any two flags (i.e., half-planes with a marked point on the
boundary) there exists an isometry taking one flag to the other.
7.12. Prove that the hyperbolic plane (as defined via the Poincaré
disk model) can be tiled by regular pentagons.
7.13. Define inversion (together with the center and the sphere of
inversion) in Euclidean space R3 , state and prove its main properties:
inversion takes planes and spheres to planes or spheres, any sphere or-
thogonal to the sphere of inversion to itself, any plane passing through
the center of inversion to itself.
7.14. Using the previous problem, prove that any inversion in R3

takes circles and straight lines to circles or straight lines.
7.15. Prove that any inversion in R3 is conformal (preserves the

measure of angles).
7.16. Construct a model of hyperbolic space geometry on the open

unit ball (use Problem 7.13).
7.17. Prove that there is a unique common perpendicular joining any

two nonintersecting lines.
7.18. Let A∞ P and A∞ P be two parallel lines (with A∞ a point on

the absolute). Given a point M on A∞ P , we say that M ∈ A∞ P is
the corresponding point to M if the angles A∞ M M and A∞ M M are

equal. Prove that any point M ∈ A∞ P has a unique corresponding
point on the line A∞ P .
7.19. The locus of all points corresponding to a point M on A∞ P

and lying on all the parallels to A∞ P is known as a horocycle. What
do horocycles look like in the Poincaré disk model?
Chapter 8
The Poincaré Half-Plane

Model
In this chapter, we will present another model of the hyperbolic plane,

also due to Poincaré. This model is also a geometry in the sense of
Klein, and we will learn in subsequent chapters that it is actually
isomorphic (as a geometry) to the disk model studied in Chapter 7.
The points of the half-plane model are simply complex numbers
with positive imaginary part (the part of the complex numbers that
lies “above” the real axis). Such a configuration of points does not
appear to be as symmetric as that of the disk, but the half-plane
model has the advantage that the elements of its transformation group
(which is a concrete subgroup of the Möbius group of linear fractional
transformations, see the definition below) are defined by simple ex-
plicit formulas and there is a neat formula for the distance between
two points.
It will turn out that the isometry group with respect to this dis-
tance is actually the transformation group of the model, so that this
model shows that hyperbolic geometry is a geometry in the traditional
sense: its structure is defined by a distance function. This will allow
us to study “hyperbolic trigonometry”, and understand the meaning
of certain mysterious “absolute constants” that arise in hyperbolic
plane geometry.
143
144 8. The Poincaré Half-Plane Model
In order to define the half-plane model, we will need to specify cer-

tain transformation groups acting on the Riemann sphere
C = C ∪ ∞, and we begin this chapter by studying these transforma-
tions.
8.1. Affine and linear-fractional transformations

of C
In this section, we will be studying various linear-fractional groups
acting on the Riemann sphere C. An efficient tool in our constructions
will be the notion of cross ratio, with which we begin.
8.1.1. Cross ratio of four complex numbers. The cross ratio of

four complex numbers z1 , z2 , z3 , z4 ∈ C is defined as the number
z3 − z1 z4 − z1
(8.1) z1 , z2 , z3 , z4 := : .
z3 − z2 z4 − z2
8.1.2. Affine transformations. A transformation of C onto itself

of the form z → az + b, ∞ → ∞, where a, b ∈ C and a
= 0, is called
affine. In particular, if a = 1, the corresponding affine transformation
is a parallel translation (by the vector OB, where B is the point of
the complex plane corresponding to the complex number b).
Theorem 8.1.3. Affine transformations take straight lines to straight
lines, circles to circles, and preserve angles and cross ratios.
Proof. Denoting a = reiϕ , r > 0, we can write

z → eiϕ z → r(eiϕ z) → (reiϕ z) + b = az + b,
which shows that any affine transformation is the composition of a
rotation (by the angle ϕ), a homothety (with coefficient r), and a
parallel translation (by the vector b). This implies the theorem, be-
cause rotations, homotheties, and translations obviously possess all
four of the properties asserted by the theorem. The least obvious of
these facts is that homotheties preserve cross ratio, but this follows
immediately from the fact that homothety in the plane of the complex
variable is multiplication by a real number (which will cancel out in
each of the fractions of the cross ratio).
8.1. Affine and linear-fractional transformations of C 145
8.1.4. Linear-fractional transformations. A transformation of

C given on C \ {−d/c} by
az + b
(8.2) z → , where cb − ad
= 0 ,
cz + d
which takes the point −d/c to ∞ and ∞ to a/c, is called linear-
fractional.
The set of all linear-fractional transformations forms a group,
called the Möbius group and denoted by Möb.
Indeed, the fact that the composition of two linear-fractional
transformations is a linear-fractional transformation can be shown
as follows: substitute (a1 z + b1 )/(c1 z + d1 ) for z in the expression
(az + b)/(cz + d), which yields (after some manipulations)
(aa1 + bc1 )z + (ab1 + bd1 )
(8.3) ;
(ca1 + dc1 )z + (cb1 + dd1 )
but this expression is of the same form as (8.2), so the composition
is indeed linear-fractional.
The fact that the inverse of any linear-fractional transformation
is a linear-fractional transformation is also easy to prove. To do that,
it suffices to find values of a1 , b1 , c1 , d1 (in terms of a, b, c, d) so that
the expression (8.3) reduces to (1 · z + 0)/(0 · z + 1); such values must
satisfy the system of four linear equations in four unknowns,
aa1 + bc1 = 1, ab1 + bd1 = 0, ca1 + dc1 = 0, cb1 + dd1 = 1,
but this system obviously has a nonzero solution.
The following property of linear-fractional transformations gives
an insight in the geometric meaning of this class of transformations
and turns out to be extremely useful in constructing and analyzing
them.
Lemma 8.1.5. Let z1 , z2 , z3 and w1 , w2 , w3 be two triplets of distinct
points of the Riemann sphere. Then there exists a unique linear-
fractional transformation taking zi to wi , i = 1, 2, 3.
Theorem 8.1.6. Linear-fractional transformations take straight lines
and circles to straight lines or circles, and preserve angles and cross
ratios.
Proof. As can be easily checked, the image of the point z under the
linear-fractional transformation (8.1) may be rewritten as
az + b a bc − ad
= + ,
cz + d c c(cz + d)
and therefore can be regarded as the composition
z → cz + d =: z1 → cz1 =: z2 → 1/z2 =: z3 → (bc − ad)z3 =: z4
a a bc − ad az + b
→ + z4 = + =
c c c(cz + d) cz + d
of an affine transformation, a homothety, a transformation taking z
to 1/z, another homothety, and a parallel translation. Concerning all
of these transformations, except z → 1/z, we know that they take
straight lines to straight lines, circles to circles, and preserve angles
and cross ratios.
Concerning the transformation z → 1/z, a straightforward if
somewhat tedious calculation shows that it preserves cross ratios (one
replaces zi by 1/zi , i = 1, 2, 3, 4, and the obtained rather cumbersome
fractions, after cancellations, reacquire the exact form of the original
ratio). Further, since 1/z = 1/z, the transformation z → 1/z is the
composition of a reflection, an inversion, and another reflection. But
we know that inversion takes straight lines or circles to straight lines
or circles and preserves angles (see 7.1.1(i)–(iii)), which proves the
theorem.
8.1.7. Two examples of linear-fractional transformations. Li-

near-fractional transformations are the subject matter of an impor-
tant chapter of the theory of a complex variable; in it, one studies
what types of domains can be mapped into each other by such trans-
formations. We will not need the general theory of this study, but
the following two examples of linear-fractional transformations will
be very important for what follows.
Example 8.1. The linear-fractional transformation

1+z
Ω : z → i ·
1−z
maps the unit disk D2 := {z ∈ C : |z|2 ≤ 1} to the upper half-plane
C+ := {z ∈ C : Im z > 0}. Indeed, it is easy to verify that the
8.2. The Poincaré half-plane model 147
points −1, i, 1 are mapped to 0, −1, ∞, respectively, which means (by

Theorem 8.1.3) that the boundary circle of the disk D2 is mapped to
the real axis. A simple computation shows that |z| < 1 implies that
Im(Ω(z)) > 0, as required.
Example 8.2. The linear-fractional transformations

az + b a(−z) + b
(8.4) z → and z → ,
cz + d c(−z) + d
where a, b, c, d ∈ R and bc − ad > 0, take the upper half-plane to itself,
the first of them preserving, the second, reversing the orientation of
the half-plane.
For the first of these formulas, it is obvious that points of the real
axis are taken to points of the real axis; further, if z, Im z > 0, is any
point in the upper half-plane, then
az + b (az + b)(cz + d) adz + bcz (ad − bc)Im z
Im = Im = Im = ,
cz + d |cz + d|2 |cz + d|2 |cz + d|2
which is positive iff bc − ad > 0.
The second formula differs from the first by a transformation of
the form z → −z, which obviously takes the upper half-plane to itself,
but reverses the orientation.
The set of all linear-fractional transformations (8.4) constitutes
a group under composition, which we denote by RMöb. Indeed, this
follows from the fact that the set of all linear-fractional transforma-
tions of the form (8.2) is a group and a composition of transformations
taking the half-plane to itself. The group RMöb will be the transfor-
mation group of the half-plane model.
8.2. The Poincaré half-plane model

The Poincaré half-plane model is the geometry consisting of the points
z ∈ C such that Im z > 0, supplied with the transformation group
RMöb. In this geometry, straight lines are defined either as open half-
circles (in the upper half-plane) perpendicular to the line Im z = 0
(which is called the absolute) or as the open rays
{z ∈ C : Re z = x0 ∈ R, Im z > 0}.
Figure 8.1. “Straight lines” in the half-plane model.
8.3. Perpendiculars and parallels

The situation with perpendiculars and parallels in the half-plane
model is quite similar to that for the disk model, except that the
corresponding pictures look very different.
Theorem 8.3.1. Given a point P and a line l in the half-plane model,

there exists a unique perpendicular to l passing through P .
Proof. There are two cases to consider (depending on whether l is a

half-circle or a half-line); see Figure 8.2.
In the first case, there are two different subcases: in the first
subcase, the given point P lies on the vertical line passing through
the center of the half-circle l, and the foot of the desired perpendicular
is obviously the intersection point K of that vertical line with the
half-circle l; in the second subcase, when the given point P is not on
that vertical line, the construction of the foot K of the perpendicular
reduces to a nice problem in the Euclidean geometry of circles: to
find a circle orthogonal to a given one, centered on a given diameter
of the given circle and passing through a given point; this problem is
left to the reader as Problem 8.2.
In the second case, the construction is obvious: one considers the
circle passing through P and centered at the intersection point of l
and the absolute and, for the desired perpendicular, one takes the arc
P K, where K is the intersection point of that circle with l.
8.3. Perpendiculars and parallels 149
Figure 8.2. Perpendiculars in the half-plane model.
Theorem 8.3.2. Given a point P and a line l in the half-plane model,

there exist infinitely many lines passing through P and not intersect-
ing l. All these nonintersecting lines lie between the two parallels to l
from P .
Proof. There are two cases to consider (depending on whether l is a

half-line or a half-circle); see Figure 8.3.
In the first case, the two parallels are easily constructed as follows:
each of them is a half-circle centered on the absolute, passing through
the given point P and through one of the two intersection points of
the half-circle l with the absolute. Their uniqueness is obvious.
In the second, the construction of the parallels is even simpler:
for one of them, we must take the straight line passing through P and
perpendicular to the absolute and for the other, the circle centered
on the absolute, tangent to l, and passing through P . Uniqueness is
also obvious in this case.
Figure 8.3. Parallels in the half-plane model.
8.4. Isometries w.r.t. Möbius distance

Let us define the Möbius distance μ(A, B) between two points A, B
of the upper half-plane by setting
μ(A, B) := | log(|A, B, X, Y |)| ,
where X and Y are the intersection points of the line (AB) with
the absolute if the points A, B have different real parts (note that
A, B, X, Y ∈ R because the four points lie on a circle, so that the
natural logarithm is well defined); if Re(A) = Re(B) = x0 , we set
μ(A, B) := | ln(A, B, ∞, X)| ,
where X is the point with coordinates (x0 , 0).
Theorem 8.4.1. The isometry group of the upper half-plane with

respect to the distance μ coincides with the group RMöb described in
Example 8.1.8.
The proof is a tedious verification that we omit.

8.5. Problems 151
8.5. Problems
8.1. Prove that
(a) linear-fractional transformations preserve the cross ratio of
four points on the Riemann sphere C;
(b) a linear-fractional transformation is uniquely determined by
three points and their images.
8.2. Let l be a straight line in the Euclidean plane, γ a circle with

center O on l, P a point not on l and not on the perpendicular to l
from O. Prove that there exists a unique circle passing through P ,
orthogonal to γ, and centered on l.
8.3. Let l be a straight line in the Euclidean plane, γ a circle with

diameter AB on l, P a point not on l and not in γ. Prove that there
exists a unique circle passing through P and A with center on l, and
a unique circle passing through P and B with center on l.
8.4. Prove that all motions (i.e., orientation-preserving isometries)

of the Poincaré disk model are of the form
az + b
z → ,
bz + a
where a and b are complex numbers such that |a|2 = |b|2 = 1.
8.5. Show that there exists an isometry of the half-plane model that
takes any flag to any other flag (a flag is a triple consisting of a line in
the hyperbolic plane, one of the two half-planes that the line bounds,
and a point on that line).
8.6*. Find a formula for the area of a triangle in hyperbolic geometry.

Chapter 9
The Cayley–Klein
Model
In this chapter, we study one more model of hyperbolic plane geom-

etry – the Cayley–Klein model. Its set of points consists of all the
points of the open disk (just as in the case of the Poincaré disk model)
and its transformation group is isomorphic to M (the transformation
group of the Poincaré model), but the action of M in the two models
is not the same. As a result, the lines in the two models look very
different: instead of arcs of circles as in the Poincaré model, in the
second model lines are open chords of the disk.
Another essential difference between our study of the two models
is in the approach to the definition of the Cayley–Klein model as a
geometry (in the sense of Klein), i.e., the definition of its transforma-
tion group. This is done in a more traditional way: we will begin by
defining the distance between points and then introduce the transfor-
mation group of the geometry as the isometry group of this distance,
i.e., the group of all distance-preserving bijections of its set of points.
9.1. Isometry and the Cayley–Klein model

9.1.1. The distance function. Let H2 be the interior of the unit
disk on the Euclidean plane and let A and B be points of H2 . Suppose
2
the (Euclidean) line AB intersects the boundary of the disk H at the
153
154 9. The Cayley–Klein Model
points X and Y , the points Y, A, B, X appearing on the line AB in

that order (see Figure 9.1).
Figure 9.1. Line in the Cayley–Klein model.
Then the distance d between the points A and B is defined as

1 |AX| |AY |
(9.1) d(A, B) := log : .
2 |BX| |BY |
The coefficient 1/2 in the right-hand side of (9.1) can be replaced

by any other positive real number c – all such distances define the
same geometry (up to isomorphism, but not up to isometry). The
reason for this strange choice (c = 1/2 rather than the more natural
c = 1) is that the coefficient c = 1/2 leads to more elegant formulas
than c = 1 and gives a metric compatible with the one in the Poincaré
model.
Note that if the points Y, A, B, X are ordered on the line AB
as shown in the figure (and A
= B), then the expression under the
logarithm sign is greater than 1 and therefore the distance between
A and B is positive. Note further that if we introduce coordinates on
the line AB, placing the origin “to the left” of Y and assigning the
real numbers y, a, b, x to the points Y, A, B, X, respectively, then the
expression under the logarithm sign can be rewritten as the following
cross ratio:
x−a y−a
: = a, b, x, y.
x−b y−b
9.1. Isometry and the Cayley–Klein model 155
This cross ratio looks very similar to the one we used to define the
distance in the half-plane model, but it should be stressed that here
we are dealing with real numbers rather than complex ones.
9.1.2. Properties of the distance function. The distance func-

tion d given by (9.1) defines a metric on the open disk H2 , i.e.,
(i) d(A, B) ≥ 0, and d(A, B) = 0 if and only if A = B;
(ii) d(A, B) = d(B, A);
(iii) d(A, B) + d(B, C) ≥ d(A, C).
Proof. Item (i) obviously holds: the distance d(A, B) between dis-
tinct points A and B is positive (as we have shown above), while
if A = B, then the denominators in (9.1) cancel, leaving us with
log(1) = 0.
Item (ii) follows from the obvious formula
x − a y − a x − b y − b −1
: = : .
x−b y−b x−a y−a
Finally, item (iii) can be proved by using projective transforma-
tions. Since we won’t be using (iii) in what follows, we postpone its
proof to Chapter 12 (see Problem 12.13).
9.1.3. Definition of the Cayley–Klein model. As explained

above, we will define the geometry (in the sense of definition 1.4.1)
of the Cayley–Klein model by taking for its transformation group
the isometry group of the distance d, i.e., the group of all distance-
preserving bijections of H2 , which we denote by N . (We will prove
later that N is isomorphic to M, the transformation group of the
Poincaré disk model, but this fact does not concern us now.)
Thus we define
the Cayley–Klein model of the hyperbolic plane
as the geometry H2 : N , where N is the isometry group of the open
unit disk H2 with respect to the distance (9.1).
9.1.4. Lines and points in the Cayley–Klein model. The points

of the Cayley–Klein model, as explained above, are simply the points
of the open unit disk H2 in R2 . The boundary of the disk is tradition-
ally called the absolute, and its points do not belong to our geometry.
The lines of our geometry are defined as the chords of the absolute
(without their endpoints). This definition immediately implies the
fundamental facts that one and only one line passes through any two
distinct points and that two noncoinciding lines either don’t intersect
or have exactly one common point.
In the following two sections, just as in the corresponding sec-

tions in the previous two chapters, we shall derive the basic facts of
hyperbolic geometry in the case of the model under consideration.
9.2. Parallels in the Cayley–Klein model

The situation with parallelism in this model is similar to that in the
Poincaré disk model, except that the picture looks slightly different
(rectilinear chords instead of arcs of circles).
Figure 9.2. Parallels and nonintersecting lines.
9.2.1. Definitions. Given a line l = AB and a point P not on this

line, it is easy to describe the lines that pass through P and do not
intersect l. Indeed, denoting by k and m the lines passing through
P and through the intersection points X, Y of the line l with the
9.2. Parallels in the Cayley–Klein model 157
absolute, we see that any line passing through P and lying between k
and m does not intersect line l; these lines are called nonintersecting
lines w.r.t. AB, while the lines k and m are the parallels to AB passing
through P (see Figure 9.2).
More generally, two lines (i.e., open chords of the disk) are parallel
if they have no common points in H2 and one common point on the
absolute; if two lines (chords) have no common points at all (in the
closed disk H2 ), then they are called nonintersecting.
We have shown that there are infinitely many lines passing through
a given point P not intersecting a given line l = AB if P ∈ / l; these
lines are all located between the two parallels to l passing through P .
9.2.2. Remark. Note that the set of all lines passing through a fixed
point of the absolute fills the entire hyperbolic plane H2 (see Figure
9.3, where both disk models are pictured).
Figure 9.3. Parallels filling the hyperbolic plane.
This means that, by using the metric on each of these lines, we can
try to define the notion of “parallel translation” and therefore that
of a “free vector” of sorts in hyperbolic geometry. This might lead
one to think that one can associate a linear space with our geometry.
Unfortunately, this is not the case (see the discussion in 9.4.1 and in
Problem 9.7).
9.3. Perpendiculars in the Cayley–Klein model

9.3.1. What they look like. Unlike perpendiculars in the Poincaré
disk model, perpendicular lines in the Cayley–Klein model do not
form right angles in the Euclidean sense. An exactly constructed
example is shown in Figure 9.4 (the nontrivial geometric construction
by means of which this “hyperbolically perpendicular straight line”
was drawn does not appear in the figure, and will be discussed in the
next chapter, in Subsection 10.1.5, and shown in Figure 10.4).
Figure 9.4. Strange looking perpendicular.
9.3.2. Definitions. Before discussing perpendicularity, we must de-

fine what perpendicular lines are. To do that, we first define a re-
flection with respect to a given line as the nonidentical isometry of
H2 that takes each point of the given line to itself. Now we can de-
fine two lines as perpendicular if the reflection with respect to one of
them takes the other line to itself. It is true that there exists one and
only one perpendicular to a given line passing through a given point,
but the proof of this fact directly in the Cayley–Klein model is quite
difficult and is omitted.
9.4. The hyperbolic line and relativity 159
9.3.3. Remark. It should be stressed that in our model the “hy-

perbolic measure” of angles is in general not equal to their Euclidean
measure. In particular, triangles in the Cayley–Klein model, which
look like rectilinear Euclidean triangles, have angle sums less than π
(although visually this does not seem to be the case).
9.4. The hyperbolic line and relativity

In this section, we digress about the distance function on hyperbolic
straight lines and point out a remarkable relationship between the
composition of shifts on such a line and the additivity of velocities
in the Special Relativity Theory of Einstein. But we begin with a
general remark concerning vectors in hyperbolic geometry.
9.4.1. Remark about free vectors. The notion of free vector

in Euclidean geometry, defined as an equivalence class of equal at-
tached vectors, allows us to associate to the Euclidean plane a two-
dimensional real vector space whose elements are precisely the free
vectors of the Euclidean plane. Any free vector also defines parallel
shifts of the entire plane in a natural way. All this is possible be-
cause at each point of the Euclidean plane there is one and only one
(attached) vector pointing in the same direction and having the same
length as a given (attached) vector. On the hyperbolic plane supplied
with a metric, we can say when two vectors have the same length, but
the expression “point in the same direction” is meaningless (compare
with Remark 9.2.2), so that there is no well-defined notion of parallel
shift. However, the notion of parallel shift along a fixed hyperbolic
straight line makes sense, and we discuss it in the next subsection.
9.4.2. Adding shifts and velocities. Let us distinguish some hy-

perbolic straight line in the Cayley–Klein model (i.e., an open chord
of the open disk H2 ) and parametrize it by an appropriate Euclidean
parameter x so that it is isometric to the open interval (−1, 1). Let
v be a real number of absolute value less than 1. Consider the map
x+v
Tv : [−1, 1] → [−1, 1], x → .
xv + 1
It is easy to prove that Tv is a bijection of the closed interval [−1, 1]
to itself leaving its endpoints in place and its restriction to the open
interval (−1, 1) is an isometry with respect to the hyperbolic distance

(for the details, see Problem 9.9). This isometry can therefore be
regarded as the parallel shift along the given hyperbolic line by the
vector v.
Let us calculate the composition of two parallel shifts by the
vectors v1 and v2 :
x + v1 x+v x + v
2 2
x → → + v1 v1 + 1
xv1 + 1 xv2 + 1 xv2 + 1
v1 + v2 v1 + v2
= x+ x +1 ;
1 + v1 v2 1 + v1 v2
we see that the composition Tv2 ◦ Tv1 is exactly the parallel shift Tv ,
where v is defined by the formula
v1 + v2
(9.1) v := .
1 + v1 v2
Thus we have proved that the composition of two parallel shifts by
vectors v1 and v2 is a parallel shift by the vector v given by formula
(9.1).
The reader will surely have noticed that this formula is the analog
of the famous Einstein formula for the addition of velocities:
v1 + v2
v := ,
(1 + v1 v2 )/c2
where c is the speed of light. The two formulas differ only in the choice
of the scale of velocity, and if in our hyperbolic scale the “speed of
light” is set equal to 1, they coincide. Note that in both situations,
if the “velocity vectors” v1 and v2 are very small as compared to the
constant c (or 1 in our case), then v is approximately equal to v1 + v2 .
The above observation shows the deep relationship existing be-
tween special relativity and hyperbolic geometry. Is our universe
hyperbolic rather than Euclidean? Actually, most physicists believe
it is neither.
9.5. Problems
9.1. Prove that for three points A, B, C on one line, where B is
between A and C, one has d(A, B) + d(B, C) = d(A, C).
9.5. Problems 161
9.2. Prove that the equality d(A, B) + d(B, C) = d(A, C) implies

that the points A, B, C lie on one line and B is between A and C.
9.3. Prove that the reflection in a line in the Cayley-Klein model is

an involution.
9.4. Show that the notion of perpendicular lines in the Cayley–Klein

model (as introduced in 9.3.2) is well defined (i.e., does not depend
on the order of the two lines).
9.5. Prove that the four angles formed at the intersection point of
two perpendiculars are congruent.
9.6∗ . Prove that the sum of angles of a triangle in the Cayley–Klein

model is less than π directly from the definitions pertaining to the
model.
9.7. Having defined the notion of free vector in hyperbolic geometry

as suggested in 9.2.2, try to define the sum of two vectors and inves-
tigate the possibility of associating a two-dimensional vector space
with hyperbolic plane geometry.
9.8. Construct a triangle in the Cayley–Klein model with angle sum

less than a given positive ε.
9.9. Prove that the parallel shift Tv defined in 9.4.2 does take (−1, 1)
to itself and find the appropriate hyperbolic distance for which it is
an isometry.
Chapter 10
Hyperbolic
Trigonometry and
Absolute Constants
We begin this chapter by showing that the three models of the hy-
perbolic plane are, in fact, isomorphic geometries. In continuing and
concluding our study of hyperbolic plane geometry, we will then feel
free to use whichever model is more convenient in the given context.
This study includes the main formulas of hyperbolic trigonometry,
which we obtain after having recalled the definitions of the hyper-
bolic functions, usually studied in complex analysis. In conclusion of
the chapter, we learn that in hyperbolic geometry, unlike Euclidean
geometry, there are inherent absolute constants.
10.1. Isomorphism between the two disk models

As we mentioned in the previous chapter, the Cayley–Klein model
and the Poincaré disk model are isomorphic. This means that there
is a bijection between their sets of points and an isomorphism of their
transformation groups which are compatible in the sense specified
in 1.4.4. To prove this, we will need a classical construction from
Euclidean space geometry.
163
164 10. Hyperbolic Trigonometry and Absolute Constants
10.1.1. Stereographic projection. Let S2 be the unit sphere, let

Π be the equatorial plane of the sphere, and N its north pole. The
stereographic projection σ : Π → S2 is the map that takes each point
M ∈ S2 \ N to the intersection point M of the ray N M with Π.
Figure 10.1. Stereographic projection.
Obviously, σ is a bijection of S2 \{N } onto Π. It is also not hard to

prove that stereographic projection is conformal (see Problem 10.1).
10.1.2. Bijection between the sets of points of the two disk

models. We regard the intersection of the open unit ball with the
equatorial plane Π as the set H2 of points of both disk models. In
order to prove that the two models are isomorphic, we begin by es-
tablishing a bijection β between their point sets. This bijection is not
the identity map, and can be described as follows.
Let A be an arbitrary point of H2 and let XY be the chord (of
the absolute) perpendicular to the radius OA (Figure 10.2). Consider
the vertical plane containing XY ; it intersects the unit sphere along
a circle. Denote by A1 the intersection of the downward vertical ray
passing through A with this circle. Now join the points A1 and N
and denote by A the intersection of A1 N and the equatorial plane.
The correspondence A → A defines a map from H2 to H2 that we
denote by β.
It is not hard to prove that the map β is a bijection of H2 onto

itself (for the details, see Problem 10.2).
10.1. Isomorphism between the two disk models 165
Figure 10.2. Bijection between the two disk models.
10.1.3. Isomorphism between the transformation groups.

The next step in the proof of the fact that the two disk models are
isomorphic geometries is the construction of an isomorphism between
their transformation groups N and M that would be compatible with
β. But that construction is in a sense automatic, because, as we
shall see, the compatibility condition actually prescribes the choice of
isomorphism.
Our aim is to construct an isomorphism ϕ : N → M, where
N and M are the transformation groups of the Cayley–Klein and
the Poincaré disk models, respectively. Let g ∈ N be an arbitrary
element and A an arbitrary point of the Poincaré disk. We define the
element ϕ(g) by setting

ϕ(g) (A) := β g β −1 (A) ,
where β is the bijection defined in the previous subsection.

This
formula says that in order to obtain the image B := ϕ(g) (A) under
ϕ(g) of an arbitrary point A, we perform the only possible natural
actions: pull back the point A from the Poincaré disk model to the
Cayley–Klein disk via β −1 , obtaining A := β −1 (A), act on A by g,

and return the obtained point g β −1 (A) to the Poincaré disk via β
(look at Figure 10.3).
Figure 10.3. Isomorphism of the two disk models.
The fact that ϕ is a group homomorphism is obvious by construc-

tion, the fact that it is bijective is also easy to prove (see Problem
10.3), while the fact that the pair (β, ϕ) is an isomorphism of geome-
tries is also immediate from the construction. We have proved the
following theorem.
Theorem 10.1.4. The map β from 10.1.3 defines an isomorphism
of the geometry (H2 : N ) (the Cayley–Klein model) and the geometry
(H2 : M) (the Poincaré disk model) if we define the corresponding
isomorphism (which we denote by ϕ) of the groups N and M by
setting
ϕ(g) (A) := β g β −1 (A) ,
where A is any point of the Poincaré disk and g ∈ N .
10.1.5. Construction of perpendiculars in the Cayley–Klein

model. The fact that we have a concrete isomorphism between the
two disk models can be used to construct the “strange looking per-
pendiculars” (look at Figure 9.4 again) in the Cayley–Klein model.
To do that, we use the bijection β from 10.1.2 to pass to the Poincaré
10.1. Isomorphism between the two disk models 167
disk model, where we know how to construct perpendiculars (see The-

orem 7.4.1) and, having performed that construction, we return to the
Cayley–Klein model via β −1 , obtaining the required perpendiculars.
Figure 10.4. Constructing perpendiculars in the Cayley–

Klein model.
In more detail, the construction is as follows (Figure 10.4). We

are given a line l = XY and a point P in the Cayley–Klein model H2 .
First we construct the chord W Z containing P and perpendicular to
the radius OP . Next, we construct the two arcs of circles perpendic-
ular to the absolute and passing through the points X, Y and W, Z
and denote by P the intersection point of the arc subtending W Z
with the radius OP . Note that the two arcs are the images of the
Cayley–Klein lines XY and W Z under the bijection β (see 10.1.2)
and are therefore lines in the Poincaré disk model.
From the point P , we draw the arc orthogonal to the absolute
and orthogonal to the arc l subtending XY (see 7.4.1) and denote by
H the intersection point of these two arcs. Note that H is the foot
of the perpendicular drawn from P to l in the sense of the Poincaré
disk model. Now if we construct the ray OH , its intersection point
H with the line l is the foot of the required perpendicular drawn from
P to l, because the map β −1 transforms the Poincaré perpendicular
P H to the Cayley–Klein perpendicular P H.
10.2. Isomorphism between the two Poincaré

models
In this section, we show that the Poincaré disk model (Chapter 7)
is isomorphic to the half-plane model studied in Chapter 8. To do
that, we will need the linear-fractional transformation Ω defined (in
Example 8.1) by the formula
1+z
Ω : z → i · ;
1−z
Ω maps the unit disk D2 := {z ∈ C : |z|2 ≤ 1} to the upper half-
plane C+ := {z ∈ C : Im z > 0}. The transformation Ω, together
with the compatibility (equivariance) condition determines the iso-
morphism between the two geometries. More precisely, we have the
following result.
Theorem 10.2.1. The map Ω from Example 8.1 defines an isomor-

phism of the geometry (H2 : M) (the Poincaré disk model from
Chapter 7) and the geometry (C+ : RMöb) (the Poincaré half-plane
model) if we define the corresponding isomorphism (which we denote
by Δ) of the groups RMöb and M by setting
M g → Ω ◦ g ◦ Ω−1 ∈ R Möb.
Proof. The map Ω is one-to-one because it has the obvious inverse

given by the rule w → (i − w)/(i + w). The isomorphism Δ is com-
patible with the group actions by definition.
Now let us define the Lobachevsky distance λ between two points

A, B of the open disk H2 (in the framework of the Poincaré disk
model) by setting
1
λ(A, B) := | log(A, B, X, Y )| ,
2
where X and Y are the intersection points of the line (AB) with the
absolute and log stands for the natural logarithm.
Now Theorem 8.1.3, Lemma 8.1.5, and Theorem 10.2.1 immedi-
ately imply the following result:
10.3. Hyperbolic functions 169
Corollary 10.2.2. The group of isometric transformations of the

disk with respect to the distance λ coincides with the group M gener-
ated by all reflections in the “straight lines” of the disk model.
Since isomorphism of geometries is a transitive relation, we have

the following.
Corollary 10.2.3. The three models of hyperbolic geometry, namely

the Poincaré disk and half-plane models and the Cayley–Klein model,
are isomorphic as geometries in the sense of Klein.
10.3. Hyperbolic functions

The complex exponent ez , z ∈ C, is related to the ordinary trigono-
metric functions by the beautiful Euler formula:
eiϕ = cos ϕ + i sin ϕ,
whose proof is obvious if we consider the unit circle centered at the
origin of the plane C. The real exponent ex , x ∈ R, is related to
the “trigonometric functions” of hyperbolic geometry, known as the
hyperbolic functions sinh, cosh, tanh, coth (hyperbolic sine, hyperbolic
cosine, hyperbolic tangent, hyperbolic cotangent, respectively) and de-
fined by the formulas:
ex − e−x ex + e−x
sinh x := , cosh x := ,
2 2
ex − e−x ex + e−x
tanh x := , coth x := .
ex + e−x ex − e−x
These functions satisfy formulas similar to the main formulas for or-
dinary trigonometric functions. Here are some examples:
cosh2 x − sinh2 x = 1, tanh x coth x = 1,
s sinh(x ± y) = sinh x cosh y ± cosh x sinh y,

sinh 2x = 2 sinh x cosh x, cosh 2x = sinh2 x + cosh2 x,
cosh(x ± y) = cosh x cosh y ± sinh x sinh y.
The proofs are obtained by plugging the definitions into the for-
mulas and performing simple calculations.
10.4. Trigonometry on the hyperbolic plane

Because of Corollary 10.2.3, the elementary trigonometric formulas
for hyperbolic triangles are exactly the same for the half-plane and
the disk model. Their proof is quite straightforward (perhaps a lit-
tle simpler in the case of the half-plane) and are relegated to the
problems. We state them in the form of theorems. Below ABC is a
triangle, α, β, γ are the angles A, B, C, respectively, and a, b, c the
sides opposite to A, B, C, respectively.
Theorem 10.4.1 (Hyperbolic sine theorem).
sinh a sinh b sinh c
= = .
Theorem 10.4.2 (Hyperbolic cosine theorem).
cosh a = cosh b cosh c − sinh c sinh b cos α .
10.5. Angle of parallelism and Schweikart

constant
10.5.1. Let (AB) be a line in hyperbolic geometry (we can use either
one of the two models here) and C a point not on (AB); let X and Y
be the intersection points of the line (AB) with the absolute, so that
the rays [CX) and [CY ) are the parallels to (AB) passing through
C; let [CH], H ∈ (AB), be the perpendicular dropped from C to
(AB); let d := λ(C, H) be the Lobachevsky distance between C and
H; finally, let α be the measure of the angle XCH (or, which is the
same, of Y CH). (See Figure 10.5, left.)
Then it is not difficult to prove that α depends only on d (see
Problem 10.11); α is called the angle of parallelism.
Theorem 10.5.2. The angle of parallelism α is given by the formula:
tanh d = cos α .
For the proof, see Problem 10.9.

This formula shows, in particular, that when d is very small, the
angle of parallelism is close to π/2, while for large values of d, α
becomes very small.
10.5. Angle of parallelism and Schweikart constant 171
10.5.3. Now let O be the center of the disk model and let [OA)
and [OB) be perpendicular rays issuing from O; let X and Y be
the intersection points of the rays [OA) and [OB) with the absolute;
let (CD) be the line intersecting the absolute at X and Y ; finally,
let [OH], H ∈ (CD), be the perpendicular from O to (CD); let
σ := λ(O, H) be the hyperbolic distance between O and H (Figure
10.5, right).
The number σ is called the Schweikart constant ; it is an absolute
constant of the hyperbolic plane. If we think of hyperbolic geometry
as a model of physical reality, then we must conclude that there is an
absolute unit of length in our universe (no such unit appears in the
Euclidean model of space).
Figure 10.5. The angle of parallelism and the Schweikart constant.
10.5.4. Another absolute constant of hyperbolic geometry comes

from the measure of a standard area, namely that of a special infi-
nite “triangle”. To construct this triangle, consider three rays issuing
from the center (actually, any other point will do) of the disk model
and forming angles of 2π/3. Denote by X, Y, Z their intersection
points with the absolute, and consider the lines XY, Y Z , ZX. They
form an “infinite equilateral triangle” with all three angles equal to
zero. Then its area can be computed by the formula for the area of a
triangle in hyperbolic geometry
S =π−α−β−γ =⇒ S=π
(see Chapter 8 and Problem 8.6).
Figure 10.6. Infinite triangle of area π.
The above argument was not very rigorous, since the formula
used is applicable only to finite triangles, but it can be made rigorous
by approximating triangle XY Z by finite triangles and passing to the
limit.
Thus we have obtained a third absolute constant, namely π, the
area of the figure bounded by three lines joining three points of the
absolute.
10.5.5. Remark. We noted above (see Section 9.4) that the formula
for adding vectors on the hyperbolic line is very similar to Einstein’s
formula for adding the velocities of inertial frames. In this section,
we have obtained three absolute constants – this is another trait of
hyperbolic geometry that is similar to the properties of Einstein’s the-
ory of the physical world, in which absolute constants (e.g. the speed
of light) appear. In this connection, one should not be misled by the
word “relativity”: Einstein’s theory doesn’t say that “everything is
relative”; on the contrary, it supplies us with physically meaningful
absolute constants, something that a Euclidean model of the universe
cannot do. On the other hand, a physical model entirely based on hy-
perbolic space geometry and an independent “time axis” is not viable
either: our universe is more complicated than that, time and space
are not independent, according to Einstein, they “mingle together”
in a certain sense.
10.6. Problems 173
10.6. Problems
10.1. Prove that stereographic projection is conformal (i.e., it pre-
serves the measure of angles).
10.2. Prove that the map β constructed in 10.1.2 is bijective and

show that any chord of H2 (i.e., any line in the Cayley–Klein model)
is taken by β to the arc of the circle passing through X and Y and
orthogonal to the absolute (i.e., to a line in the Poincaré disk model).
10.3. Prove the main relations between the hyperbolic functions

indicated in Section 10.3.
10.4. Prove the hyperbolic sine theorem.
10.5. Prove the hyperbolic cosine theorem.
10.6. Prove that two triangles with equal sides are congruent in
hyperbolic geometry.
10.7. Prove that in hyperbolic geometry two triangles having an

equal angle and equal sides forming this angle are congruent.
10.8. Show that homothety is not conformal in hyperbolic geometry.
10.9. (a) Prove the formula for the angle of parallelism α for a point
A and a line l:
tanh d = cos α,
where d is the distance from A to l (thereby showing that the angle of
parallelism depends only on the distance from the point to the line).
(b) Prove that the previous formula is equivalent to the following
one (obtained independently by Bolyai and Lobachevsky):
α
tan = e−d .
2
10.10. Prove that in a triangle with right angle γ the sides a, b, c and
their opposite angles α, β, γ = π/2 satisfy the following relations:
sinh a = sinh c sin α, tanh b = tanh c cos α,
cot α cot β = cosh c, cos α = cosh a sin β.
What do these relations tend to as a, b, c become very small?
10.11. Prove that the sides a, b, c and opposite angles α, β, γ of any

triangle on the hyperbolic plane satisfy the following relations:
(a) cosh a sin β = cosh b sin α cos β + cos α sin γ,
cos α + cos β cos γ
(b) cosh a = .
sin β sin γ
10.12. Prove that if the respective angles of two triangles are equal,
then the triangles are congruent.
10.13. Prove that all the points of the (Euclidean) straight line
y = kx that lie in the upper half-plane y > 0 are equidistant from the
(hyperbolic) straight line Oy.
10.14. (a) Prove that any hyperbolic circle contained in any one of
the Poincaré models of hyperbolic geometry is actually a Euclidean
circle.
(b) For the Poincaré upper half-plane model, find the Euclidean
center and radius of the hyperbolic circle of radius r centered at the
point (a, b).
(c) For the Poincaré model in the unit disk D, find the relation-
ship between the radii of the Euclidean and the hyperbolic circles
centered at the center of D.
10.15. Prove the triangle inequality for the distance in the Poincaré
half-plane model.
10.16. Prove that the three (a) bisectors, (b) medians, (c) altitudes
of any hyperbolic triangle intersect at one point.
10.17. (The hyperbolic Menelaus Theorem ) The line l intersects the

lines BC, CA, AB (containing the sides) of triangle ABC at the points
A1 , B1 , C1 respectively; then
sinh AC1 sinh BA1 sinh CB1
= 1.
sinh C1 B sinh A1 C sinh B1 A
10.18. (The hyperbolic Ceva Theorem.) The points A1 , B1 , C1 are

chosen on the sides BC, CA, AB of triangle ABC. Prove that the
10.6. Problems 175
segments AA1 , BB1 , CC1 intersect at one point if and only if one of
the following two equivalent conditions holds:
sin ACC1 sin BAA1 sin CBB1
· · = 1,
sin C1 CB sin A1 AC sin B1 BA
sinh AC1 sinh BA1 sinh CB1
· · = 1.
sinh C1 B sinh A1 C sinh B1 A
Chapter 11
History of
Non-Euclidean
Geometry
In this chapter, we will retrace the history of the creation of non-

Euclidean geometry by Gauss, Lobachevsky, and Bolyai (and their
predecessors and followers) and discuss the traditional axiomatic ap-
proach to the foundations of geometry. The story begins with Euclid’s
Elements, the brilliant first attempt to construct mathematics as a
deductive science (see [8] and Appendix A).
11.1. Euclid’s Fifth Postulate

The Ancient Greeks realized that, in a deductive science, in order to
deduce (prove) facts from other facts by logical reasoning, it is nec-
essary to start from some facts which are not proved. Euclid called
these facts postulates (we call them axioms) and explicitly formulated
five of them. He also used several other axioms implicitly (without
formulating them). Apparently, Euclid (and other Greek mathemati-
cians) thought that the postulates should be self-evident (simple and
so obvious that no doubt about their truth could arise).
Euclid’s last axiom, the Fifth Postulate, however, is not simple
and not obvious. Its modern equivalent can be stated as follows.
177
178 11. History of Non-Euclidean Geometry
(V+) For any straight line and any point not on this line there is
a unique parallel to this line passing through the given point.
Here by a parallel to a given line one means a straight line that
has no common points with the given line. In Euclid’s formulation,
the statement was more complicated and less obvious.
(V) If a straight line falling on two straight lines makes the sum
of the interior angles on one side less than two right angles, then the
two straight lines, if extended indefinitely, meet on that side on which
the angles with sums less than two right angles exist.
Presumably, Greek mathematicians (perhaps Euclid himself) tried
to deduce the Fifth Postulate from the other axioms. In any case, in
Euclid’s Elements, the application of the Fifth Postulate is postponed
as much as possible: it occurs for the first time in the proof of Propo-
sition 27 of Book 1 (there are 48 propositions, i.e., theorems in our
terminology, in that book). The interested reader may want to look
at the postulates and theorems in Book 1 of Euclid’s Elements: they
appear in Appendix A of the present book.
After Euclid, for more than two thousand years, many scientists
tried to prove the Fifth Postulate, and many “succeeded”, usually by
proving statements equivalent to (V) by means of arguments based
on additional axioms which were not explicitly formulated.
11.2. Statements equivalent to the Fifth

Postulate
We have already mentioned one such statement, namely (V+). Here
are some more (in square brackets [ ], we indicate the mathematician
who used this approach to “prove” the Fifth Postulate).
(1) The sum of the three angles of any triangle is equal to π (to
two right angles, in Euclid’s terminology). [This statement appears
in Euclid’s Elements as Proposition 32, and was proved by using the
Fifth Postulate; Legendre gave a “proof ” in 1805 without the Fifth
Postulate.]
(2) A line intersecting one of two parallel lines intersects the other
[Proclus, 5th century].
11.3. Gauss 179
(3) Similar but not congruent triangles exist [John Wallis, 1663].
(4) The fourth angle of a quadrilateral with three right angles is

also a right angle [Nasiraddin, 13th century; Saccheri, 1773; Lambert,
1766] . Such a quadrilateral was later called a Saccheri quadrilateral.
Trying to prove the Fifth Postulate, most mathematicians (in-

cluding those mentioned above) argued by contradiction. As a rule,
they considered two cases, assuming that the sum of angles of a tri-
angle is (a) more than π or (b) less than π (equivalently, that the
fourth angle of the Saccheri quadrilateral is more (less) than π/2,
or that there are no parallels, respectively, more than one parallel,
through a given point to a given line). In the first case, it is possible
to correctly obtain a contradiction using the Euclidean axioms. In the
second case, a contradiction does not follow, but the desire to prove
the Fifth Postulate was so strong that the mathematicians working
on the problem usually produced what they claimed to be a proof,
but which was actually flawed.
11.3. Gauss
Carl Friedrich Gauss (1777–1855) first began working on the Fifth
Postulate in 1796, at the age of nineteen, and argued by contradiction,
like his predecessors, but went much further in developing the theory
in case (b). It is not clear when he came to the conclusion that no
contradiction would arise. In a famous letter (1824) to his friend
F.A. Taurinus, he explained that in the case α + β + γ < π one
obtains a “thoroughly consistent curious geometry”, which he called
“non-Euclidean”. He concluded his letter by asking Taurinus not to
tell anyone about his “private communication”, which he was thinking
of publishing at “some future time”.
Later, in 1832, he learned from his friend Farkas Bolyai that
the latter’s son, Janos, had arrived at the same conclusions. Later
still, in 1841, he found out that Lobachevsky had done the same.
Gauss even learned Russian (to read Lobachevsky’s early work?), but
never directly communicated with either Janos Bolyai or Lobachevsky
about these questions.
Carl Friedrich Gauss
The most amazing thing, however, is that Gauss, when he was

not thinking about number theory or the Fifth Postulate, had con-
structed the differential geometry of surfaces, including surfaces of
constant negative curvature, which are, in fact, a model (at least lo-
cally) of hyperbolic geometry. All these years, he had this model
before his eyes, but never made the obvious connection with non-
Euclidean geometry. He died without suspecting that a proof of the
consistency of hyperbolic geometry was at his finger tips!
11.4. Lobachevsky
Nikolai Ivanovich Lobachevsky (1793–1856), like everybody else, tried
to prove the Fifth Postulate by contradiction. As he progressed fur-
ther in the case α + β + γ < π, he became convinced that the theory
11.4. Lobachevsky 181
was consistent. In an unpublished textbook, written in 1823, he men-

tions that all attempts to prove the Fifth Postulate were erroneous.
In 1826, Lobachevsky published a memoir about a new geometry
(which he called imaginary) in the Kazan Bulletin, but this publi-
cation (written in Russian) went unnoticed abroad. Trying to gain
recognition, he published his work in German (Geometrische Unter-
suchugen, 1840) and in French (Pangéométrie, 1855), but without
success (for an English translation of his work, see [11]).
Nikolay Ivanovich Lobachevsky
N.I. Lobachevsky was not only the President (Rector, in the Rus-
sian terminology) of Kazan University, but also its Head Librarian.
The Kazan library received many scientific periodicals, including the
most famous mathematical journal of the time, Crelle’s Journal. Li-

brary cards (which have come down to us) show that Lobachevsky
read every issue of Crelle’s Journal that reached Kazan, except two
successive issues in the 1830s. These two issues contained two papers
by Mindling, in which the latter obtained, on surfaces of constant neg-
ative curvature, trigonometric formulas identical to the trigonometric
formulas previously obtained by Lobachevsky on the hyperbolic plane.
Had Lobachevsky seen one of these papers, it is very likely that he
would have observed that they could be used to obtain a proof of the
consistency of hyperbolic geometry!
11.5. Bolyai
Janos Bolyai (1802–1860) was the son of a mathematician, Farkas
Bolyai, who had “proved” the Fifth Postulate (his friend Gauss had
pointed out his error). Janos first followed in his father’s footsteps by
trying to prove the Fifth Postulate by contradiction, but soon realized
that he was obtaining a consistent geometry. In 1823 he wrote to
his father: “Out of nothing I have created a strange new universe.”
But it was only in 1832 (three years after Lobachevsky) that his
investigations were published in an Appendix to his father’s book
Tentamen (both were written in Latin; for the German translation,
see [15], the English translation of the Appendix appeared in [17] and
in [14], p. 375).
Farkas sent the book to Gauss, asking to comment on the Ap-
pendix. Instead of praising and encouraging Janos, Gauss wrote that
this would be “praising myself ”, since he had discovered the same
things thirty years before, and the Appendix “spared him the effort”
of writing up his discovery. Discouraged, Janos Bolyai stopped work-
ing for several years, but then started working on a book that would
contain a detailed exposition of his results.
When Gauss had learned about Lobachevsky’s results, he “kindly”
communicated this fact to Janos Bolyai via the latter’s father. For a
while, Janos thought that Lobachevsky did not exist, that he was a
creation of Gauss, who used “Lobachevsky” as a pen name to publish
results stolen from J. Bolyai’s Appendix! Fortunately, Janos Bolyai
finally understood that this was not the case, but he never finished
11.6. Beltrami, Helmholtz, Lie, Cayley, Klein, Poincaré 183
Janos Bolyai
his book; in fact, he published nothing more. He died fairly young,

unrecognized by his contemporaries. . .
11.6. Beltrami, Helmholtz, Lie, Cayley, Klein,

Poincaré
The first proof of the consistency of hyperbolic geometry is due to Bel-
trami, who showed (1868) that its axioms and theorems hold (at least
locally) on surfaces of constant negative curvature. Recent research
in the history of the subject shows that Beltrami was also aware of
what is commonly known as the Poincaré and Cayley–Klein models.
The physicist Helmholtz was probably the first to understand how one
can prove the consistency of hyperbolic geometry, but his arguments
were regarded as insufficiently rigorous by mathematicians. Sophus
Lie improved the arguments of Helmhotz and was the first to stress
the role of transformation groups in mathematics. Klein gave the
definition of geometry that we introduced in Chapter 1, and, simulta-
neously with Cayley (but independently of him), gave the elementary
global model of hyperbolic geometry described in Chapter 9; he also
coined the terms hyperbolic, parabolic, elliptic for the three geome-
tries. Poincaré constructed the two models of hyperbolic geometry
that we discussed in Chapters 7 and 8.
11.7. Hilbert
David Hilbert made the first successful attempt to give an axiomatic
exposition of Euclidean (space) geometry, rigorous in the modern
sense of the word. It consists of 21 axioms, three undefined concepts
(point, line, plane), and several undefined relations. Hilbert’s axioms
for plane geometry are presented and discussed in Appendix B of the
present book.
Hilbert’s axiomatic approach is rarely used in teaching geometry
in our time, because Euclidean geometry can be introduced in a much
simpler way: it can easily be constructed as a branch of linear algebra
over the real numbers (based on the fact that the straight line is
“isomorphic” to the real numbers R). This fact can be deduced from
Hilbert’s axioms by using the axiomatic definition of the real numbers
and checking that these algebraic axioms are satisfied by the points of
any line, provided the product and sum operation are appropriately
defined on it.
Chapter 12
Projective Geometry
In this chapter, we introduce the main ideas of projective geometry

for the particular case of RP 2 , the projective plane, and we only
take a brief look at the projective space RP 3 . The general theory of
d-dimensional projective spaces (RP d , d ≥ 1) is traditionally studied
in linear algebra courses by means of the so-called homogeneous coor-
dinate model, but we do not go beyond the dimension d = 3. We use
a more geometric approach, which may seem strange at first, because
in our model “points” of RP 2 will be lines in Euclidean space R3 , but
ultimately we will also appeal to the homogeneous coordinate model.
12.1. The projective plane as a geometry

12.1.1. Main definition. The projective plane RP 2 is defined as
the geometry (RP 2 : Proj(2)), whose elements (called projective points)
are straight lines in R3 passing through the origin O and whose trans-
formation group Proj(2) is defined as follows. We start with the gen-
eral linear group GL(3) and identify any two linear transformations of
R3 whose matrices can be obtained from each other by multiplication
by nonzero constants; the composition of matrices is well defined on
such equivalence classes of transformations, and Proj(2) is defined as
the group whose elements are these classes and the group operation
is composition (i.e., multiplication of matrices).
185
186 12. Projective Geometry
12.1.2. Points and lines. The elements of RP 2 (projective points)

are Euclidean lines; nevertheless, we will often simply call them points
(of our geometry). The straight lines (of our geometry) are defined as
the (Euclidean) planes passing through the origin. These definitions
immediately imply the following two assertions.
I. One and only one “line” passes through any two distinct “points”.
II. Any two distinct “lines” intersect in one and only one “point”.
Thus there are no parallel lines in our geometry, just as in spheri-
cal geometry. But we will see that the two geometries are very differ-
ent; in particular, there is no natural metric in projective geometry
(and hence no measure of angles, no perpendiculars, no areas, and
so on). Unlike spherical geometry, in which “straight lines” intersect
in two points, in projective geometry lines intersect in one point, not
two.
12.1.3. Intuitive description. You can imagine the projective

plane as a Euclidean plane to which a “line at infinity” Λ∞ has been
added. When you move along a Euclidean line L to infinity in some
direction, you intersect the line at infinity at some point P = L ∩ Λ∞ ;
if you move along L in the opposite direction, you will reach Λ∞ and
intersect it at the same point P . Parallels (in the Euclidean sense) in-
tersect on the infinite line. Thus lines in RP 2 are some kind of cycles
(like “infinite circles”). The line at infinity, however, should not be
regarded as a “special” line, because most projective transformations
transform it into an “ordinary” line. The informal description of RP 2
given here will be made rigorous in Subsection 12.2.4.
12.2. Homogeneous coordinates

12.2.1. Returning to our geometry (RP 2 : Proj(2)), let us introduce
coordinates for our points. Each point L (i.e., each Euclidean line
passing through the origin) is uniquely determined by its direction
vector, i.e., by three coordinates (x1 , x2 , x3 ), in the standard basis of
R3 , namely in the basis
e1 = (1, 0, 0), e2 = (0, 1, 0), e3 = (0, 0, 1).

12.2. Homogeneous coordinates 187
Conversely, however, points do not uniquely determine the coordi-

nates: if λ is a nonzero real number, then (λx1 , λx2 , λx3 ) determines
the same point as (x1 , x2 , x3 ). In this situation, we call the two sets of
coordinates equivalent, denote the corresponding equivalence class by
(x1 : x2 : x3 ), and refer to χ(L) = (x1 : x2 : x3 ) as the homogeneous
coordinates of the point L.
12.2.2. Homogeneous coordinates make the computation of the ac-

tion of elements g ∈ Proj(2) on points L ∈ RP 2 very easy: the
transformation g is given by a 3 × 3 matrix Ag ∈ GL(3) (defined up
to a constant), and
⎛ ⎞⎛ ⎞
a11 a12 a13 x1
g(L) = Ag ((x1 : x2 : x3 )) = ⎝ a21 a22 a23 ⎠ ⎝ x2 ⎠ .
a31 a32 a33 x3
The geometric meaning of the transformation with matrix Ag is that
its column vectors are the images of the standard basis vectors under
that transformation, but since Ag is defined up to a nonzero scalar,
these images are also defined up to a nonzero scalar multiple.
12.2.3. Projective spaces of higher dimensions. In linear alge-

bra courses, the projective space RP d , for any value of d, is defined in
a similar way: its elements are homogeneous coordinates
(x0 : x1 : · · · : xd ),
i.e., equivalence classes of (d + 1)-tuples (x0 : x1 : · · · : xd ) of real
numbers (not all equal to zero) up to multiplication by a nonzero con-
stant. The group Proj(d + 1) acts on each element by multiplication
by (d + 1) × (d + 1) matrices corresponding to linear operators in Rd+1
(defined up to a constant). We will not study higher-dimensional pro-
jective spaces RP d , d > 3, in this course. A detailed account can be
found in most linear algebra courses. However, we will look at pro-
jective space RP 3 briefly in Section 12.8 below.
12.2.4. Now let us describe a rigorous model of RP 2 that will explain

why RP 2 is called the projective plane. In R3 consider the plane Π
given by the equation x3 = 1. Points of this plane have coordinates
of the form (x1 , x2 , 1). To the plane Π add the line at infinity Λ∞
whose points are equivalence classes of Euclidean points (x1 , x2 , 0) up
to multiplication by a nonzero constant (notation (x1 : x2 : 0)). The
set Π ∪ Λ∞ is the set of points of the projective plane.
Figure 12.1. The projective plane.
Note that the “points at infinity” (x1 : x2 : 0) ∈ Λ∞ determine

Euclidean straight lines in the plane x3 = 0. Intuitively, you should
think of these lines as “pointing to infinity” in a certain direction, so
that the set Λ∞ “surrounds” the plane Π. More precisely, these lines
are not rays, they are ordinary “two-sided” lines, and so they point to
infinity in two opposite directions, but they intersect the projective
line Λ∞ at only one point (you should think of this point as being the
identification of two diametrically opposite points at infinity).
The lines in this model of RP 2 are the ordinary (Euclidean) lines
in Π plus the “line” Λ∞ . There is an obvious bijection between the
points and lines of RP 2 (as defined in the previous section) and those
in the model Π ∪ Λ∞ ; in particular, the line Λ∞ corresponds to the
(Euclidean) plane x3 = 0. Using this bijection, it is easy to define the
action of Proj(2) in this model.
12.3. Projective transformations

12.3.1. One may want to ask: Why is our geometry called “projec-
tive”, when it is defined by a group of linear operators in R3 ? Let us
12.3. Projective transformations 189
try to answer this question. Let Π1 and Π2 be two planes in R3 and

let P ∈ R3 be a point. The projection of Π1 to Π2 from P is the map
π that to each point A ∈ Π1 assigns the point A ∈ Π2 at which the
line P A intersects Π2 . This assignment is not necessarily bijective: π
will be undefined at some points X (if P X is parallel to Π2 ) and not
onto (some points of Π2 will not be covered); see Figure 12.2.
Figure 12.2. Projective transformations of planes.
However, if we supply Π1 and Π2 with lines at infinity Λ1∞ and

Λ2∞ , and appropriately define the projection, then we obtain a bijec-
tion between the projective planes Π1 ∪ Λ1∞ and Π2 ∪ Λ1∞ . The details
are left to the reader.
12.3.2. A set of points A1 , . . . , An , n ≥ 3, of the projective plane

RP 2 (interpreted as the model described in Subsection 12.2.4) are
said to be in general position if for any three of them, Ak , Al , Am , the
−→ −→ −→
vectors OAk , OAl , OAm constitute a basis of R3 . If one of the points,
−→
say Ai , lies on the line at infinity, the vector OA1 is well defined, in
coordinates it has the form (a : b : 0). If three points or more from
our collection lie on the infinite line, then, of course, the collection
will not be in general position.
Another way of defining a collection of points in general position
is to say that no three of them lie on the same line.
Theorem 12.3.3. There exists one and only one projective transfor-
mation that takes four points A, B, C, D ∈ RP 2 in general position to
four other points A , B , C , D ∈ RP 2 in general position.
Proof. In accordance with our model of the projective plane, we

can think of the points A, B, C and A , B , C as lying in the plane
−→ −−→ −−→
x3 = 1. By assumption, the vectors OA, OB, OC constitute a basis
of R3 . Let (a1 , a2 , a3 ), (b1 , b2 , b3 ), (c1 , c2 , c3 ) be the coordinates of the
−→ −−→ −−→
vectors OA , OB , OC in that basis. Then the matrix
⎛ ⎞
a1 b1 c1
M = ⎝ a2 b2 c2 ⎠
a3 b3 c3
can be regarded as a linear transformation of R3 taking A, B, C to

A , B , C . Now let us multiply the columns of this matrix by scalar
constants, obtaining the matrix
⎛ ⎞
λa1 μb1 νc1
Ag = ⎝ λa2 μb2 νc2 ⎠
λa3 μb3 νc3
which we now regard as defining an element g of Proj(2). Clearly,

Ag takes the points A, B, C ∈ RP 2 to the points A , B , C ∈ RP 2 ,
although the same matrix regarded as acting in R3 does not take
A, B, C ∈ R3 to A , B , C ∈ R3 (when not all three of the scalars
λ, μ, ν are equal to 1).
Now let us denote by (d1 , d2 , d3 ) the coordinates of the point D
−→ −−→ −−→
in the basis OA, OB, OC and by (d1 , d2 , d3 ) the coordinates of the
point D in the same basis. We claim that it is possible to choose the
scalar parameters λ, μ, ν so that Ag will take D ∈ RP 2 to D ∈ RP 2 .
Indeed, this will be the case if the matrix Ag applied to the vector
(d1 , d2 , d3 ) will give the vector (d1 , d2 , d3 ), or, which is the same thing,
the system of equations,
⎧
⎪ a d λ + b1 d2 μ + c1 d3 ν = d1 ,
⎨ 1 1
a2 d1 λ + b2 d2 μ + c2 d3 ν = d2 ,
⎪
⎩
a3 d1 λ + b3 d2 μ + c3 d3 ν = d3 ,
12.4. Cross ratio of collinear points 191
in the unknowns λ, μ, ν will have a solution. But the determinant Δ

of this system can be expressed as Δ = d1 d2 d3 det(M ) and so, it is
nonzero. Hence our system of equations has a nonzero solution in
λ, μ, ν. Thus we have shown that Ag (D) = D (if we choose for the
values of λ, μ, ν the solution of our system) and proved the existence
of the required projective transformation.
Its uniqueness follows by working out the construction of Ag in
reverse order, which will bring us back to the same matrix (up to
multiplication by a scalar).
12.4. Cross ratio of collinear points

12.4.1. Main definitions. We mentioned above that there is no
natural metric on the projective plane, and no affine structure (the
ratio of the two segments determined by three collinear points of RP 2
is not well defined). Nevertheless, the affine structure in R3 allows us
to define the cross ratio of any four ordered collinear points of RP 2 .
The definition is the following. Let k, l, m, n be collinear points in
RP 2 , i.e., four coplanar lines of R3 passing through the origin; suppose
a line s cuts our four lines at the points A, B, C, D, respectively.
−→ −−→ −→ −−→
Then the vectors AC and BC are proportional, i.e., AC = λBC; the
real number λ (which may be negative) is denoted by A, B, C; the
number A, B, D is defined similarity. We now put
A, B, C
A, B, C, D := ;
A, B, D
the number thus obtained is called the cross ratio of four collinear
points A, B, C, D. It is not difficult to show that it is well defined, i.e.,
does not depend on the choice of the secant line s. Now if one of the
points, say B, lies on the infinite line Λ∞ , then we put A, B, C, D :=
C, D, A (similarly for the other cases).
12.4.2. Coordinate expressions. The cross ratio is easy to com-

pute in coordinates. To this end, we return to the model
Π = {(x, y, z) ∈ R3 | z = 1} ⊂ RP 2 = Π ∪ Λ∞
and suppose that the collinear points A, B, C, D have the coordinates:
(xA , yA , 1), (xB , yB , 1), (xC , yC , 1), (xD , yD , 1).
Then, obviously,
x C − xA yC − yA x D − xA yD − yA
A, B, C = = , A, B, D = =
xC − xB yC − yB xD − xB yD − yB
and therefore
x C − xA xD − xA yC − yA yD − yA
A, B, C, D = : = : .
xC − xB xD − xB yC − yB yD − yB
If one of the points, say B, is on the infinite line (at its intersec-
tion with the line containing the points A, C, D), then the cross ratio
reduces to the ordinary ratio. What happens in this case may be
described by saying that “the infinities cancel”:
xC − xA x D − xA x C − xA
: = = C, D, A.
xC − ∞ xD − ∞ xD − xA
In the case when all four points A, B, C, D lie on the infinite line,
their cross ratio is also a well-defined real number. Its calculation is
the object of Problem 12.3.
Theorem 12.4.3. The cross ratio of four collinear points is invariant
under projective transformations.
Proof. The proof is a problem in linear algebra; see Problem 12.4.

12.5. Projective duality

12.5.1. Points and lines on the projective plane (RP 2 : Proj(2)) play,
in a certain sense, symmetric roles. This will be easier to see if we
introduce the notion of incidence: we will say that two lines a and b
are incident at the point P if P is the intersection point of the lines
a and b, and that the two points P and Q are incident to the line a
if a passes through P and Q. Also, together with the standard term
collinear (used for points all lying on one line) we will use the term
copunctal for lines all passing through one and the same point.
Given an assertion of projective geometry formulated in this ter-
minology, we can translate it into another statement, called dual, by
replacing the word “line” by the word “point” (and “collinear” by
“‘copunctal”) and vice versa. For example, statement I from Sec-
tion 12.1 can be expressed as: “One and only one line is incident to
12.5. Projective duality 193
two distinct points”; its translation (i.e., the dual statement) will be
“One and only one point is incident to two distinct lines”, which is
exactly the assertion II (see Section 12.1). Another example: “Any
projective transformation takes collinear points to collinear points”
translates to “Any projective transformation takes copunctal lines to
copunctal lines”.
What is remarkable is that this kind of translation always trans-
lates true statements to true statements. To prove this, we will de-
fine the dual geometry to the geometry of RP 2 : it is the geometry
(DRP 2 : Proj(2)) whose points are planes of R3 passing through the
origin under the action of the group of linear nonsingular transforma-
tions of R3 . In DRP 2 , the intersection of two points (i.e., Euclidean
planes) will be called the line passing through the points (it is actually
a Euclidean line in Euclidean 3-space).
Theorem 12.5.2. The two geometries (DRP 2 : Proj(2)) and
(RP 2 : Proj(2)) are isomorphic: there is a bijection, called duality
and denoted by D, between the sets of points of the two geometries
compatible with an isomorphism of GL(3) onto itself.
Proof. To each “point” Π of DRP 2 , i.e., to each plane of R3 given

by the equation a1 x1 + a2 x2 + a3 x3 = 0, we assign the point of RP 2
with homogeneous coordinates (a1 : a2 : a3 ) (which is of course the
Euclidean line passing through the origin and perpendicular to the
plane). If an element g ∈ Proj(2) takes the point (a1 : a2 : a3 ) to
some point (b1 : b2 : b3 ), then the same element will take the plane
Π to the plane given by b1 x1 + b2 x2 + b3 x3 = 0. Thus the duality
map D : RP 2 → DRP 2 (which is obviously bijective) is compatible
with the action of Proj(2), so that we have constructed the required
isomorphism.
Note that the duality correspondence is an involution, i.e., D ◦ D

identically maps RP 2 onto itself. Further, note that the isomorphism
constructed above preserves incidence: if two points A, B of RP 2
(i.e., two Euclidean lines passing through the origin O of R3 ) are
incident to the line l (i.e., are contained in a Euclidean plane Πl ),
then the two lines D(A), D(B) in DRP 2 intersect in the point (of
DRP 2 ) D(l) = Πl . Thus we have the following statement.
12.5.3. Corollary: Duality Principle. There is a bijection be-

tween the set of lines and the set of points of RP 2 that preserves
incidence and takes any theorem of projective geometry to a theorem
of projective geometry.
12.6. Conics in RP 2
The nondegenerate conic sections (or conics for short) in the Eu-
clidean plane are, as is well known, the ellipse, the hyperbola and the
parabola. In RP 2 , these three curves are projectively equivalent, so
there exists only one nondegenerate conic in RP 2 (up to projective
equivalence).
A conic in RP 2 can be defined as any set of points obtained from
the curve C given by (x1 )2 + (x2 )2 = 1 (in the plane-with-line-at-
infinity model described in Section 12.2, this curve is the Euclidean
circle) by a projective transformation. Any projective transformation
under which the image of C does not intersect the line at infinity Λ∞
transforms C into an ellipse; a projective transformation that takes
one point of C to Λ∞ transforms C into a parabola, and a projective
transformation that takes two points of C to Λ∞ transforms C into
a hyperbola.
12.7. The Desargues, Pappus, and Pascal

theorems
We conclude our study of RP 2 with three beautiful classical theorems.
All three can be regarded as theorems about points and lines either in
the projective plane or in the affine (in particular Euclidean) plane.
12.7.1. Desargues’ Theorem. Suppose that the lines joining the

corresponding vertices of triangles A1 A2 A3 and B1 B2 B3 intersect
at one point S. Then the intersection points P1 , P2 , P3 of the lines
A2 A3 and B2 B3 , A3 A1 and B3 B1 , A1 A2 and B1 B2 , respectively, are
collinear.
Proof. We begin by passing from the plane to 3-space and prove the
three-dimensional analog of theorem Desargues’. (The proof of the
3D theorem turns out to be unexpectedly simple, but the argument
12.7. The Desargues, Pappus, and Pascal theorems 195
Figure 12.3. Desargues’ theorem.
used in it doesn’t work in the plane!) We then use the 3D theorem

to prove Desargues’ theorem in the plane by means of a continuous
deformation of the spatial picture to the planar one.
Suppose we are given two triangles A1 A2 A3 and B1 B 2 B3 in Eu-

clidean space R such that the three lines A1 B1 , A2 B2 , A3 B3 intersect
2
at one point S. (The reader should think of the points A1 , B1 , A3 , B3 ,

S as being the same as in the planar version of the theorem, while
the points A2 , B2 have been “lifted out” of the plane.) Then the lines
2 , SB3 define a trihedral angle in R3 (see Figure 12.4).
SB1 , S B
2 A3 and B
Consider the three pairs of lines A 2 B3 , A
2 A1 and B
2 B1 ,
A1 A3 and B1 B3 . We claim that each of these pairs has a common
point (in space!) and these three points are collinear.
Indeed, the (Euclidean) planes
2 A3 )
Π1 := (A1 A 2 B3 )
and Π2 := (B1 B
Figure 12.4. Desargues’ theorem in space.
intersect in a line Λ. Obviously, the lines A2 A3 and B 2 B3 intersect

at a point (denoted Q1 ) of Λ, and so do the lines A 2 A1 and B 2 B1
(the intersection point is denoted by Q3 ) as well as the lines A1 A3
and B1 B3 (at Q2 ). Since the points Q1 , Q2 , Q3 all lie on Λ, they are
collinear, as claimed.
Let us pass to the proof of the planar version of the theorem.
Consider the plane B1 SB3 (which we think of as being “horizon-
tal”), construct a plane perpendicular to it through the line SB2 , in
that plane choose a point O “below” the horizontal plane, and choose
points A2 and B 2 so that S, A
2 , B
2 are collinear by projecting the
points A2 , B2 from O (see Figure 12.5).
Using the 3D version of the theorem, we can now construct the
2 about S
three collinear points Q1 , Q2 , Q3 . Now rotate the line S B
downward in the vertical plane until it coincides with SB2 . Since the
12.7. The Desargues, Pappus, and Pascal theorems 197
Figure 12.5. Proof of Desargues’ theorem.
mobile points Q1 , Q2 , Q3 will always be collinear and, when they reach

the horizontal plane, they will coincide with the points P1 , P2 , P3 , it
follows that these three points are collinear. This proves the theorem.

12.7.2. Pappus’ Theorem. Suppose the points A1 , A2 , A3 are col-

linear, and the points B1 , B2 , B3 are collinear. Let P1 , P2 , P3 be the
intersection points of the lines A2 B1 and A1 B2 , A1 B3 and A3 B1 ,
A2 B3 and A3 B2 , respectively. Then the points P1 , P2 , P3 are collinear.
Sketch of the proof. By Theorem 12.3.3, we can assume that

−−→
A1 A2 B1 B2 is a square. Using the coordinate system with basis A1 A2 ,
Figure 12.6. Pappus’ theorem.
−−→
A1 B 1 , it is an easy exercise to prove that the points P1 , P2 , P3 are
collinear.
12.7.3. Pascal’s Theorem. Suppose the points A, B, C, D, E, F lie

on a conic. Let P1 , P2 , P3 be the intersection points of the lines AB
and ED, AF and CD, CB and EF , respectively. Then the points
P1 , P2 , P3 are collinear.
Figure 12.7. Pascal’s theorem.

12.8. Projective space RP 3 199
The theorem is illustrated by Figure 12.7, in which the conic is a

circle. In fact, Pascal actually proved the theorem in this particular
case without any loss of generality – he knew all conics are projectively
equivalent to the circle. Here we do not present the (not very difficult)
proof of his theorem.
Remark 12.7.4. Note that the theorem is true in RP 2 as well as in
R2 . To formulate it in full generality as a Euclidean theorem, one has
to consider several singular cases (which arise when one of the points
Pi “goes to infinity”); in these cases the proof differs somewhat from
the proof in the generic case. Note also that the Euclidean versions
have metric proofs (see Problem 12.14), but the projective proof is,
in a sense, more natural. Similar remarks hold for the Pappus and
the Desargues theorems.
12.8. Projective space RP 3

In this section we very briefly describe three-dimensional projective
geometry.
12.8.1. Definition of projective space. The projective space RP 3

can be defined in terms of homogeneous coordinates as explained in
Subsection 12.2.3, but here we adopt a more geometric approach.
Namely, we consider four-dimensional Euclidean space R4 and for the
points of RP 3 take the straight lines passing through the origin O
of R4 and define the transformation group Proj(3) of RP 3 as in the
two-dimensional case (using GL(4) instead of GL(3)). We then define
the lines of RP 3 as the planes passing through the origin O and its
planes as the three-dimensional hyperplanes of R4 passing through O.
The following basic statements immediately follow from the above
definitions.
I. One and only one “line” passes through any two distinct “points”.
II. Any two distinct “planes” intersect in one and only one “line”.
Thus there are no parallel lines or parallel planes in this geome-
try. Moreover, there is no natural distance function in RP 3 compat-
ible with its geometry, and so no measure of areas or angles, and no
perpendiculars.
12.8.2. Properties of projective transformations. Without go-

ing into detail, let us just mention that there is a “five point theorem”
similar to the “four point theorem” 12.3.3 and that the cross ratio of
four collinear points is invariant under projective transformations.
There is a neat theory of quadrics (surfaces given by second degree
equations) in which, for example, the hyperboloid of two sheets is
(projectively) equivalent to the hyperboloid of one sheet and to the
ellipsoid.
12.8.3. Projective duality in space. Just as in RP 2 , in RP 3 there

is a duality principle, but a somewhat more sophisticated one: it in-
volves not only points and lines, but also planes. After replacing the
expressions “passing through”, “intersecting in”, etc. by appropriate
versions of the notion of incidence and using the expressions “copunc-
tal” and “coplanar” in the formulation of a theorem, we obtain the
dual theorem simply by interchanging the words “point” and “plane”
(and not changing the word “line”, since lines are self-dual). The dual
theorem will also be correct, since its proof can be obtained by “dual-
izing” the proof of the original theorem. For example, the properties
I and II are dual to each other.
12.9. Problems
12.1. Five distinct collinear points A, B, C, D, E are given. Prove
that
A, B, C, D · A, B, D, E · A, B, E, C = 1.
12.2. How many different values does the cross ratio of four points
on a line take when the order of the points is changed?
12.3. Calculate the cross ratio of the following four points lying on
the infinite line Λ∞ : (xi : yi ; 0), i = 1, 2, 3, 4.
12.4. Prove Theorem 12.4.3.
12.5. Four planes pass through a common line l, while the line m
intersects all four planes. Prove that the cross ratio of the intersection
points of m with the planes does not depend on the choice of m.
12.9. Problems 201
12.6. State and prove the theorem dual to the Pappus theorem.
Draw the corresponding picture.
12.7. State and prove the theorem dual to Desargues’ theorem. Draw
the corresponding picture.
12.8*. Prove that under projective duality any point on a conic is

taken to a line tangent to the dual conic.
12.9. Using Problem 12.8, state and prove the theorem dual to Pas-
cal’s theorem (the dual theorem is known as Brianchon’s Theorem).
Draw the corresponding picture.
12.10. Three skew lines l, l1 , l2 in R3 are given. To a point A1 ∈ l1

let us assign the point A2 at which the line l2 intersects the plane
determined by A1 and l. Prove that the assignment A1 → A2 is a
projective map of l1 onto l2 .
12.11. The lines l1 , . . . , ln−1 and l are given on the plane. The
points O1 , . . . , On are chosen on l. The lines containing the sides of
a polygon A1 , . . . , An pass through the points O1 , . . . , On while its
vertices A1 , . . . , An−1 move along the lines l1 , . . . , ln−1 . Prove that
the vertex An also moves along a straight line.
12.12. Prove the triangle inequality for the hyperbolic metric by

using appropriate projective transformations.
12.13. Prove the Euclidean version of Pascal’s theorem for the case
of the circle.
Chapter 13
“Projective Geometry Is
All Geometry”
The title of this chapter is a quotation from Arthur Cayley, the out-
standing 19th century British mathematician, one of the founders
of projective geometry. The aim of this chapter is to give a pre-
cise mathematical meaning to these words, namely to show that the
three principal continuous geometries, parabolic (Euclid), hyperbolic
(Lobachevsky), and elliptic (Riemann), are subgeometries of projec-
tive geometry. We will prove this in dimension two, i.e., show that
the projective plane “contains” (in a certain precise sense) the hy-
perbolic plane, the elliptic plane, and the Euclidean plane. Since the
discrete geometries that we also studied in this book are, in turn,
subgeometries of the three principal continuous ones, this means that
all the geometries studied so far in this course are parts of projective
geometry.
But first we recall the notion of subgeometry, which appeared
briefly in Chapter 1.
13.1. Subgeometries
13.1.1. Recall that two geometries (X : G) and (Y : H) are isomor-
phic if there is an equivariant bijection between them, i.e., a bijection
203
204 13. “Projective Geometry Is All Geometry”
between their sets of points and an isomorphism between their trans-

formation groups which are compatible (for the detailed definition,
see Chapter 1). Further, the geometry (X : G) is a subgeometry of
(Y : H) if there is an injective map i : X → Y and a monomor-
phism γ : G → H compatible with the group actions, i.e., satisfying
(i(x))(γ(g)) = i(xg). (In this formula, we use the notation xg for
the result of the action of the element g ∈ G on the point x ∈ X;
thus (i(x))(γ(g)) stands for the result of the action of the element
γ(g) ∈ H on the point i(x) ∈ Y .)
Of course any geometry isomorphic to the given one is its subge-
ometry, but we are interested in the case when it is a proper subge-
ometry, i.e., when i is not a bijection, or γ is not an isomorphism, or
both.
13.1.2. Here are two toy examples of proper subgeometries:
• the motion group of the regular dodecagon (regular polygon of

12 sides) is a subgeometry of the dodecahedron with dihedral group
D12 acting on it;
• the dihedral group D6 acting on the regular dodecahedron de-

fines a subgeometry of Euclidean plane geometry (R2 : Ismtrd (R2 )).
(Here and below, instead of Sym(R2 ) we use the notation Ismtrd (R2 )
or Ismtr(R2 ) for the isometry group of Euclidean space w.r.t. the
standard metric.)
13.2. The Euclidean plane as a subgeometry of

the projective plane RP 2
13.2.1. The fact that the Euclidean plane (R2 : Isom(R2 )) is a sub-
geometry of the projective plane (RP 2 : Proj(2)) is rather obvious if
we interpret RP 2 (in the homogeneous coordinate model, see Section
11.2) as the plane
Π = {(x1 , x2 , x3 ) ∈ R3 | x3 = 1}
supplied with the “line at infinity” Λ∞ = {(x1 : x2 : x3 ) | x3 = 0},

i.e., if we take RP 2 = Π ∪ Λ∞ .
13.3. The hyperbolic plane as a subgeometry of RP 2 205
Indeed, let us define i : R2 → RP 2 = Π ∪ Λ∞ in the obvious way,

i.e., by setting i((x1 , x2 )) := (x1 , x2 , 1), and define γ : Ismtr(R2 ) →
−−
→ −→
GL(3) as follows. Let g ∈ Ismtr(R2 ), let (AB, AC) be an orthonormal
−−− → −− →
frame in R2 and (A B , A C ) its image under g. For γ(g) we take
the element of Proj(2) that takes the three lines OA, OB, OC to the
three lines OA , OB , OC . This construction is shown in the figure.
Figure 13.1. The Euclidean plane as a subgeometry of RP 2 .
Theorem 13.2.2. The construction described above shows that the

Euclidean plane is a subgeometry of the projective plane.
Proof. The theorem is obvious: clearly, i is injective, γ is a monomor-

phism, and the fact that compatibility holds is also immediate.
13.3. The hyperbolic plane as a subgeometry of

the projective plane RP 2
13.3.1. The fact that the hyperbolic plane (H2 : M ) is a subgeometry
of the projective plane (RP 2 : Proj(2)) is best seen by using the
Cayley–Klein model and interpreting RP 2 (as in Section 12.2 of the
preceding chapter) as the plane

Π = {(x1 , x2 , x3 ) ∈ R3 | x3 = 1}
supplied with the “line at infinity” Λ∞ = {(x1 : x2 : x3 ) | x3 = 0},
i.e., by taking RP 2 = Π ∪ Λ∞ .
We recall that the Cayley–Klein model was defined as
(H2 : Ismtrλ (H2 )), where H2 is the unit open disk and λ is the met-
ric given by the formula λ(A, B) = (1/2)| ln(A, B, X, Y )| (for the
details, see Section 9.2).
Figure 13.2. The Cayley–Klein model as a subgeometry of RP 2 .
Now let us define i : H2 → RP 2 = Π ∪ Λ∞ in the obvious way,

i.e., by setting i((x1 , x2 )) := (x1 , x2 , 1), and define γ : Ismtrλ (H2 ) →
Proj(2) as follows. Let g ∈ Ismtrλ (H 2 ). Take four points A, B, C, D ∈
H2 in general position and consider their images Ag, Bg, Cg, Dg ∈ H2
under g. Denote
A1 = i(A), B1 = i(B), C1 = i(C), D1 = i(D) ∈ H2 ,
A2 = i(Ag), B2 = i(Bg), C2 = i(Cg), D2 = i(Dg) ∈ H2 .
The two quadruples of points Ai , Bi , Ci , Di ∈ Π, i = 1, 2, are in gen-
eral position, and so by Theorem 12.3.3 there exists a unique projec-
tive transformation taking A1 , B1 , C1 , D1 to A2 , B2 , C2 , D2 ; we take
this transformation to be γ(g). The construction is shown in the
figure.
13.4. The elliptic plane as a subgeometry of RP 2 207
Basically, this construction is simply the natural extension of the

action of g from the open unit disk
{(x1 , x2 , 1) | x21 + x22 < 1} = i(H2 )
to the entire projective plane. To any “straight line” of H2 (i.e., any
chord XY of the unit circle) corresponds the straight line joining the
points i(X), i(Y ) in the projective plane; to parallel or nonintersecting
lines in H2 (chords of the unit circle) correspond straight lines in RP 2
that actually intersect (at a point outside the disk i(H2 ), possibly on
the “infinite line” Λ∞ ).
hyperbolic plane is a subgeometry of the projective plane.
Proof. The map i is obviously injective, so that it remains to show

that the restriction of γ(g) to the open disk
{(x1 , x2 , 1) | x21 + x22 < 1} = i(H2 )
coincides with γ. This is a consequence of the fact that projective
transformations preserve the cross ratio of any four collinear points,
and therefore preserve the distance λ between points inside i(H2 ) (λ
being the absolute value of the logarithm of the appropriate cross
ratio). But g is an isometry (with respect to λ), it coincides with
the restriction of γ(g) to i(H2 ) on three noncollinear points, therefore
it coincides with this restriction on all of i(H2 ). This proves the
theorem.
Remark 13.3.3. It can be proved that the subgroup of Proj(2) that

takes the circle {(x1 , x2 , 1) | x21 +x22 = 1} to itself is actually isomorphic
to Ismtrλ (H2 ), and this isomorphism is often used to establish various
formulations expressing the fact that the hyperbolic plane is “a part
of ” the projective plane. One can also define (a model of) hyperbolic
geometry by using this fact. We do not need this fact in our approach
to this topic, so we omit its proof here.
13.4. The elliptic plane as a subgeometry of RP 2

13.4.1. As in the previous two sections, we regard RP 2 as the plane
Π with the line at infinity Λ∞ added to it. Our model of Riemannian
two-dimensional elliptic geometry Ell2 will be the standard one, i.e.,

the unit sphere with its antipodal points identified, namely: Ell2 =
(S2 /Ant : O(3)). We think of this sphere as lying on the plane Π,
touching it at the point (0, 0, 1).
We first construct the inclusion (which will actually be a bijec-
tion) of S2 /Ant to RP 2 by simply projecting it from the center of the
sphere onto Π ∪ Λ∞ . Note that “straight lines” in S2 /Ant (i.e., great
circles of the sphere with diametrically opposed points identified) will
be mapped to straight lines of the projective plane; in particular, the
equator of the sphere will be mapped to the “infinite line” Λ∞ . Note
also that spherical triangles (not intersecting the equator) will be pro-
jected to ordinary rectilinear triangles in Π, but their angles will not
be preserved.
Figure 13.3. Bijection between the elliptic plane and RP 2 .
To construct the monomorphism γ : O(3) → Proj(2), we choose

two perpendicular arcs AB and AC (that do not intersect the equa-
tor) and denote by BCD the triangle symmetric to triangle ABC
with respect to the line BC. Denote by A1 , B1 , C1 , D1 the central
projections of the points A, B, C, D to the plane Π. Now suppose
g ∈ O(3) takes the points A, B, C, D to A , B , C , D , and denote by
13.5. Problems 209
A1 , B1 , C1 , D1 their projections to Π. We define γ(g) as the projec-
tive transformation that takes A , B , C , D to A1 , B1 , C1 , D1 (such a
projection exists and is unique by Theorem 12.3.3). The construction
is shown in Figure 13.3.
Riemannian elliptic plane is a subgeometry of the projective plane.
Proof. The theorem is an easy consequence of the following lemma,

whose proof is the object of Problem 13.3.
Lemma 13.4.3. The map γ described above is a monomorphism of
O(3) to Proj(2).
Indeed, the monomorphism γ is compatible with the map i by

construction, so the theorem follows.
13.5. Problems
13.1. Prove that any projective transformation of the projective
plane RP 2 preserves the cross ratio of collinear points
13.2. Prove that, conversely, any transformation of the projective

plane that preserves the cross ratio of all collinear points is projective.
13.3. Prove Lemma 13.4.3.
13.4. Give an example of a spherical triangle whose angle sum is close

to 2π and describe its image under the central projection defined in
Section 12.4.
13.5. Show that for any ε > 0 and any positive number S, there
exists a spherical triangle of area less than ε whose image under the
central projection defined in Section 13.4 is of area greater than S.
13.7. Prove that the subgroup of projective transformations that

take the unit circle centered at the origin to itself is isomorphic to the
isometry group of the hyperbolic plane.
13.8. Generalize and solve the previous problem by replacing the

circle by an arbitrary oval (nondegenerate second degree curve).
Chapter 14
Finite Geometries
A finite geometry is geometry whose set of points is finite. In that

situation, the possibilities for the transformation group are extremely
varied, and Klein’s definition of geometry is too general to single out
those finite geometries that actually deserve to be called geometries.
Thus one must impose restrictions on the group actions involved, and
this can be done by using coordinates from linear spaces over finite
fields. Another approach involves introducing the notion of “straight
line” and imposing conditions (axioms) which make the geometries
“projective” or “affine” in a certain sense.
Unfortunately, the two approaches are not equivalent, the ax-
iomatic approach yielding a wider class of finite planes than the alge-
braic coordinate one. However, it turns out that the two approaches
are equivalent if and only if Desargues’ theorem holds in the finite
geometry considered.
It should be noted that some basic natural questions about finite
geometries are at present unanswered and that these geometries are
the object of active ongoing research. Some of these questions and
related conjectures are mentioned in Section 14.11.
211
212 14. Finite Geometries
14.1. Small finite geometries

In this section, we try to classify all the geometries with a “small”
number of points. By classifying we mean listing (without repetitions)
all the geometries with a given number of points k := |X| up to
isomorphism. Recall that two geometries are isomorphic if there is an
equivariant bijection between them, i.e., a bijection between their sets
of points and an isomorphism between their transformation groups
which is compatible with the bijection (for the detailed definition, see
Chapter 1).
There is of course only one geometry with one point. For |X| = 2
there are two geometries (with |G| = 2 and |G| = 1). For |X| = 3
there are four: the symmetries (= isometries) of the vertices of the
equilateral triangle (G = S3 ), the motions of the vertices of the equi-
lateral triangle (G = Z3 ), the symmetries of the vertices of the isosce-
les triangle (G = Z2 ). For |X| = 4 there are ten: the symmetries of
the regular tetrahedron, its motions, the symmetries of the square,
its motions, the rotations of the square by 0 and π, the symmetries
of the rhombus, and four more geometries obtained when the trans-
formation group has a fixed point (the same one for each element).
For |X| ≥ 5 the situation becomes too complicated to handle,
while for |X| ≥ 10, even a supercomputer is powerless.
To continue our study, we need to specify some reasonable nar-
rower classes of finite geometries. To do that, we need some algebra.
14.2. Finite fields

The modern logical foundation of ordinary Euclidean affine geometry
is the notion of vector space over the real number field. To construct
something similar in the finite case, we need finite fields.
Theorem 14.2.1. For any q = pm , where p is prime and m is a
positive integer, there exists exactly one (up to isomorphism) field
consisting of q elements, called the finite field of order q and denoted
by F(q). There are no other finite fields.
14.3. Example: the finite affine plane over F(5) 213
We will not prove this theorem (the proof belongs to algebra

courses) and only present the simplest nontrivial example, the field
F(4) = {0, 1, 2, 3}, by showing its addition and multiplication tables:
+ 0 1 2 3 × 0 1 2 3
0 0 1 2 3 0 0 0 0 0
1 1 0 3 2 1 0 1 2 3
2 2 3 0 1 2 0 2 3 1
3 3 2 1 0 3 0 3 1 2
In order to get a feeling for the structure of the fields F(q), we

invite the reader to construct the addition and multiplication tables
for, say, F(32 ).
14.3. Example: the finite affine plane over F(5)

In this section we will construct a finite affine plane geometry starting
from the finite field F(p), where p is a prime number (i.e., in the case
m = 1). To make the construction more concrete, we will carry it out
for p = 5, although it works for any prime p.
14.3.1. Let us define the affine plane AF(5) of order 5 as the set
{(x, y) | x ∈ F(5), y ∈ F(5)} of pairs (coordinates of points). As
in ordinary Euclidean geometry, two points T = (a, b), S = (c, d)
−→
determine a vector T S = {c − a, d − b}. We will define straight lines
as in analytic geometry, i.e., by setting A(t) = A0 + t− →
v , where A0 is
→
−
a point, v is a vector, and t runs over F(5). For example, if we take
A0 = (0, 0) and −→
v = (1, 2), we obtain the “straight line”

(0, 0), (1, 2), (2, 4), (3, 1), (4, 3) .
Thus we obtain a total of 30 straight lines, 25 points, 5 points on

each line, and 6 lines passing through each point. In Figure 14.1, we
have shown the six lines passing through the point (0, 0).
Figure 14.1. Six lines in AF(5).
Arguing in the same way in the general case, we obtain p2 + p

straight lines, p2 points, p points on each line, and p + 1 lines passing
through each point.
14.3.2. The same result can be obtained by using the orbit space
of an appropriate geometry. Let Z ⊕ Z ⊂ R2 be the integer lattice
on the plane and let (Z ⊕ Z : G) be the geometry defined by the
transformation group G, isomorphic to Z ⊕ Z, acting by coordinate
shifts by 5, i.e.,
G (k, l) : (m, n) → (m + 5k, n + 5l).
The orbit space of this action consists of 25 “points”. We identify

them with the 25 points of the lattice with nonnegative coordinates
less than 5. The “straight line” passing through two points of this 5
by 5 square is defined as follows: construct the Euclidean line joining
these two points in R2 , take all the integer points on this line and
reduce both of their coordinates mod 5, obtaining three more points
in the square; together with the two given points, they constitute a
“straight line”.
Geometrically, you can visualize this as the covering of the torus
by the plane: under this map the points of the square lattice are
“wrapped around” the 25 points on the torus.
14.4. Example: the finite affine plane over F(22 ) 215
14.4. Example: the finite affine plane over F(22 )

We now start our constructions with the field F(22 ). Define the affine
plane over F(4) as the set {(x, y) | x ∈ F(4), y ∈ F(4)} of pairs
(coordinates of points). Using the same approach as in Section 14.3
(including the “vector definition” of straight lines), let us consider the
line passing through the point (0, 0) with direction vector {1, 1}. This
“line” has only two points (0, 0) and (1, 1) because (1, 1) + {1, 1} =
(0, 0). Thus the coordinate approach does not work over F(4).
Nevertheless, a reasonable affine geometry with 4 points on each
line can be constructed on the set of points P by defining straight
lines in a different way. In particular, the “straight line” that passes
through the points (0, 0) and (1, 1) also contains the points (2, 2)
and (3, 3), while the line passing through (0, 0), (1, 2) contains the
points (2, 3) and (3, 1). In this geometry, there are 16 = q 2 points,
20 = q 2 + q straight lines, 4 = q points on each line, and 5 = q + 1
lines pass through each point. The five lines passing through the
point (0, 0) are shown in Figure 14.2.
Figure 14.2. Five lines in AF(4).
The result is a geometry called the finite affine plane over the field
F(4) and is denoted by AF(4). The set AF(4) is indeed a geometry
in the sense of Klein, (AF(4) : Γ), if for the transformation group Γ
we take the set of all bijections of AF(4) that map lines into lines.
In the general case, i.e., when F = F(q), q = pm , m > 1, with
prime p, one can also construct the finite affine plane AF(q), but the
direct construction is rather tedious, and we omit it. However, we will

present a neat indirect construction via finite projective geometries
in Section 14.8.
First, we give an example of a finite projective geometry.
14.5. Example of a finite projective plane

14.5.1. Let AF(4) be the finite affine plane for q = 22 . We say that
two lines of AF(4) are parallel if they coincide or have no common
points. Parallelism is an equivalence relation, and so all lines are
partitioned into equivalence classes of parallel lines. It is easy to see
that there are 5 such classes. To AF(4) let us add 5 points (called
points at infinity) and agree that they all lie on one straight line (the
line at infinity). The set thus obtained is called the projectivization
of the affine plane AF(4) and is denoted by P F(4); it has 21 points,
21 straight lines, 5 points on each line, 5 lines passing through each
point, and any two distinct lines have exactly one common point. The
projective plane P F(4) is shown in Figure 14.3.
Figure 14.3. Projectivization of AF(4).
14.5.2. The construction described above for q = 4 actually works

for any q = pm with prime p. One obtains the projective geometry
P F(q); it has q 2 + q + 1 points, q 2 + q + 1 straight lines, q + 1 points
on each line, q + 1 lines passing through each point.
14.6. Axioms for finite affine planes 217
14.6. Axioms for finite affine planes

14.6.1. A more traditional approach to finite geometries is the ax-
iomatic approach. A finite affine plane is a nonempty finite set of
elements P (called points) with a family L of subsets (called lines)
that satisfy the axioms:
Aff.1. There is exactly one line passing through any two distinct
points.
Aff.2. There is exactly one line parallel to a given line and pass-
ing through a given point. (Two lines are called parallel if they have
no common points or if they coincide.)
Aff.3. There exists a generic triangle (three points not belonging

to one and the same line).
Here the second axiom ensures that the dimension of the set of
points is less than or equal to 2. The third axiom ensures that its
dimension is greater than or equal to 2. Thus the dimension of the
set of points is two, and this set can be regarded as a “plane”. The
construction of the two simplest affine planes (with 4 and 9 points)
is the object of Problem 14.1.
Theorem 14.6.2. (i) For every q = pm , where p is prime and m is

a positive integer, there exists an affine geometry P = AF(q) with q
points on a line.
(ii) The geometry P = AF(q) has q 2 points, a family of q 2 + q
subsets L that satisfies the axioms Aff.1–Aff.3.
(iii) If Γq is the group of bijections of P that map lines (i.e.,
elements of L) into lines, then (P, Γq ) is a geometry in the sense of
Klein called an affine Galois plane of order q.
The existence of AF(q) (item (i) of the theorem) will be proved in

14.8.3 below. The proof of items (ii)–(iii) is a series of moderately dif-
ficult problems for the reader (14.2–14.6) that appear in the problem
section.
14.7. Axioms for finite projective planes

14.7.1. A finite projective plane is a nonempty finite set of elements
P (called points) with a family L of subsets (called lines) that satisfy
the following axioms:
Proj.1. There is exactly one line passing through a given pair of
distinct points.
Proj.2. There is exactly one point contained in a given pair of dis-
tinct lines.
Proj.3. There exist four points that determine six distinct lines.
Proj.4. There exist four lines that determine six distinct points.
Actually the fourth axiom is redundant (it follows from the first
three), we include it for the sake of symmetry.
The simplest finite projective plane (called the Fano plane) is
shown in Figure 14.4. It has 7 points, 7 lines, 3 points on each line,
and 3 lines passing through each point. The four points in the mid-
dle of the picture satisfy the axiom Proj.3. The Fano plane can be
constructed from the four point affine plane by adding the “line at
infinity”, as explained in 14.5.1.
Figure 14.4. The Fano plane.
14.7.2. Projective duality. Just as in the case of the real projec-

tive plane RP 2 , the finite projective plane satisfies the Duality Prin-
ciple: Interchanging the words “point” and “line” in the statement of
any theorem and accordingly modifying the wording of the incidence
relations, one obtains another theorem. This principle follows from
14.7. Axioms for finite projective planes 219
the fact that the four axioms split into two pairs dual to each other.
However, the finite projective plane obtained from a given one by
duality is not necessarily isomorphic to the given one. Questions of
duality are rather delicate in the finite case, and we do not discuss
them here.
Theorem 14.7.3. If (P, L) is a finite projective plane, then there

exists a natural number n, called the order of the plane, such that:
(i) each line contains n + 1 points;
(ii) each point is contained in n + 1 lines;
(iii) the number of points is equal to the number of lines and equal
to n2 + n + 1.
Remark 14.7.4. The theorem does not assert the existence of finite
projective planes: in it, it is assumed that a finite projective plane
is given, and thus it only asserts that if a plane satisfying axioms
Proj.1–Proj.3 exists, then its number of lines and points satisfies the
constraints (i)–(iii).
14.7.5. Proof of Theorem 14.7.3. Suppose (P, L) is a finite pro-

jective plane, l, m ∈ L, and let a ∈ P be a point not lying on l nor
on m (such a point exists by axiom Proj.3). Consider the map f of
the set of points of the line l to the set of points of m that assigns to
each point x ∈ l the intersection point of the lines xa and m. Axioms
Proj.1–Proj.2 imply that f is well defined and bijective. Denoting the
number of points on l by n+1, we see that item (i) is proved. Item (ii)
follows by the duality principle. To prove (iii), fix some point a ∈ P.
Each line passing through a passes through n other points, and so
|P| = (n + 1)n + 1 = n2 + n + 1. By duality we have |L| = n2 + n + 1,
which concludes the proof.
14.7.6. Remarks. (1) One can pass from the finite affine plane to
the projective plane by adding q+1 “points at infinity” (corresponding
to each class of parallel lines) and one new line (the line of all points
at infinity). Conversely, one can pass from a projective plane to an
affine plane by removing one line (with all its points). Unfortunately,
the result is not well defined: it may depend on the choice of the line!
(2) There is no uniqueness theorem for projective planes of or-

der pm for m > 1 (for example, there is are several nonisomorphic
projective planes of order 9, see Problems 14.9).
(3) It is not known at present for what values of q there exist
projective planes of order q. Specifically, this question is unanswered
already for q = 12. This question, and other open questions, as well
as related conjectures, are briefly discussed in Section 14.11.
14.8. Constructing projective planes

over finite fields
In this section, we give a constructive definition of the finite projective
planes based on linear spaces over finite fields, similar to the definition
of the real projective plane RP 2 (cf. 12.1).
14.8.1. Main construction. Consider the three-dimensional vec-

tor space V over the finite field F = F(pm ), where p is prime. Denote
by P the set of one-dimensional subspaces of V , which we now call
points, and by L the set of two-dimensional subspaces, which we now
call lines; we say that a line l ∈ L passes through a point p ∈ P (or
p is contained in l, or l contains p) if we have the inclusion of linear
spaces p ⊂ l.
Theorem 14.8.2. (i) The construction described above yields a finite

projective plane (P, L) of order q = pm .
(ii) If we define the transformation group of P as the set of bijec-
tions Γ of P that take lines to lines, then (P, Γ) is a geometry in the
sense of Klein.
Proof. All four axioms Proj.1–Proj.4 are immediate consequences of

the main construction. Item (ii) is the object of Problem 14.6.
The geometry thus constructed is called the finite projective space

over the field F(pm ) and is denoted by P F(pm ).
Corollary 14.8.3. The finite affine plane of order q = pm , where p

is any prime and m is any natural number, exists.
14.9. The Desargues theorem 221
Proof. To construct the required plane, it suffices to remove one line

(and all its points) from the finite projective plane of order q.
14.9. The Desargues theorem

The Desargues theorem, which we proved for the real projective plane
RP 2 , is not true for arbitrary finite projective planes. However, we
have the following statement.
Theorem 14.9.1. The Desargues theorem holds for the finite projec-
tive planes P F(pm ) = (P, L), i.e., three lines x1 y1 , x2 y2 , x3 y3 ∈ L in-
tersect at one point if and only if the intersection points z1 , z2 , z3 ∈ P
of the pairs of lines x2 x3 and y2 y3 , x3 x1 and y3 y1 , x1 x2 and y1 y2 ,
respectively, are collinear.
Proof. In the proof, we will use the model of P F(pm ) given by the
construction 14.8.1, i.e., we regard points as one-dimensional linear
subspaces of the vector space over F(pm ) and lines as two-dimensional
subspaces.
First let us note that the Desargues theorem is self-dual, and
therefore it suffices to prove the “only if ” part, i.e., assuming that
the lines A1 B1 , A2 B2 , A3 B3 intersect at one point (which we denote
by S), to show that the intersection points P1 , P2 , P3 are collinear.
If the point S lies in each of the three lines P1 P2 , P2 P3 , P3 P1 , then
there is nothing to prove, so we can assume that S ∈ / P2 P3 .
In our model the points S, Ai , Bj , Pk are actually one-dimensional
linear spaces, and we will use the same lower case letters s, ai , bj , pk
to denote nonzero vectors belonging to (and therefore determining)
the corresponding linear spaces.
Now since the vectors s, a1 , b1 belong to the same two-dimensional
space, they are linearly dependent, and (by an appropriate choice of
these vectors in their linear spaces) we can write b1 = a1 + s. It is
easy to see that the vectors a1 , p2 , p3 are linearly independent, and
therefore we can put
b1 = α1 a1 + α2 p2 + α3 p3 ,
Figure 14.5. Desargues’ theorem.
where α1 , α2 , α3 ∈ F . Consider the linear operator ϕ on V given by
ϕ(a1 ) = b1 , ϕ(p2 ) = p2 , ϕ(p3 ) = p3 .
Then we have
(14.1) ϕ(s) = ϕ(b1 − a1 ) = (α1 − 1)b1 + α2 p2 + α3 p3

= (α1 − 1)b1 + b1 − α1 a1 = α1 s.
The linear operator ϕ is nondsingular, so it takes linear subspaces to

linear subspaces of the same dimension. In particular, we have
ϕ(A1 ) = B1 , ϕ(P2 ) = P2 , ϕ(P3 ) = P3 , ϕ(S) = S.
The vectors p2 , p3 form a basis of the line P2 P3 (regarded as a

two-dimensional vector space), and so the operator ϕ is the identity
on this line. Now if Λ is any line passing through S, then, since ϕ
14.10. Algebraic structures in finite projective planes 223
leaves S in place as well as the intersection point of the lines Λ and

P2 P3 , it follows that ϕ(Λ) = Λ.
Now the point A2 lies on the lines SA2 and B1 P3 , and therefore
ϕ(A2 ) is the intersection point of the lines SA2 and B1 P3 , and so
ϕ(A2 ) = B2 . Similarly, ϕ(A3 ) = B3 . Thus ϕ(A2 A3 ) = B2 B3 . Now
let P be the intersection point of the lines A2 A3 and P2 P3 . Then the
point ϕ(P ) lies on the line B2 B3 and, at the same time, ϕ(P ) = P .
Therefore, P = P1 and P1 lies on the line P2 P3 , which was to be
proved.
14.9.2. Remark. Note that this proof (like the proof given in 12.7.1)
is, in a certain sense, “three-dimensional”: when we replaced points
by vectors in the above proof, we were essentially adding a point (the
origin of coordinates in the three-dimensional space over F(pm )) lying
outside of the plane containing all the given points.
14.10. Algebraic structures in finite projective

planes
Until now, we have been using algebra (finite fields) to construct geo-
metric objects (finite affine and projective planes). Now we will try
to move in the opposite direction, i.e., analyze what the geometric
axioms for the finite projective plane imply concerning the algebraic
structure of the projective line. Unfortunately, it will turn out that
the natural and optimistic expectation that axioms Proj.1–Proj.4 im-
ply that there are pm + 1 points on each line (for some prime p and
natural number m) and that these points can be added and multi-
plied in a natural way, thereby forming a field isomorphic to F(pm ),
does not come true. The situation is much more complicated; in the
general case one can obtain an algebraic structure from the axioms,
but it is not that of a field: its multiplication is not commutative and
there is only one distributive law (see 14.10.3 below).
14.10.1. Introducing coordinates. Let (P, L) be a finite projec-

tive plane of order n ≥ 2. (Recall that this means that (P, L) satisfies
axioms Proj.1–Proj.4) and one of its lines (and therefore all lines)
contains n points. Denote by F a set of n elements; we stress that F
is a set of arbitrary symbols, it is not a field; in fact, at first it has no
algebraic operations defined on it. Our aim is to supply F with an

algebraic structure (hopefully that of a field) and use it to introduce
coordinates in our finite projective plane (P, L).
We begin by choosing two arbitrary elements of F that we denote
by 0 and 1. By ∞ we denote a symbol that does not belong to F .
Using axiom Proj.3, let us choose an initial quadrilateral in our plane,
i.e., four points, no three of which lie on one line. Denote these points
by (0, 0), (0), (∞), (1, 1) and denote the six lines passing through these
points as follows:
[0, 0] : = (0, 0)(0), [0] := (0, 0)(∞), [∞] := (0)(∞),

[1] : = (1, 1)(∞), [0, 1] := (1, 1)(0), [1, 0] := (0, 0)(1, 1).
These six lines intersect in seven points, four of which belong to the
initial quadrilateral, and we denote the other three as follows:
(1, 0) := [1][0, 0], (0, 1) := [0][0, 1], (1) := [∞][1, 0],
where the juxtaposition of two lines determines their intersection,

e.g., the formula (1, 0) = [1][0, 0] means that (0, 1) is the intersection
point of the lines [1] and [0, 0].
If there are no other points in P, then n = 2 and it is easy to
see that we have obtained the Fano plane. The reader will profit by
looking at Figure 14.4 and supplying its points with coordinates as
indicated in the construction described above.
If there are other points left, then n > 2 and we denote by a an
arbitrary element of F other than 0 or 1. For any such a, we define
new points and lines by setting
[a, 0] := (0, 0)(a), (1, a) := [1][a, 0], [0, a] := (0)(1, a),

(a, a) := [0, a][1, 0], [a] := (a, a)(∞), (a, 0) := [a][0, 0], (0, a) := [0, a][0].
If there are any other elements b in F other than 0, 1, a, we set
(a, b) := [a][0, b], [a, b] := (a)(0, b).
Thus we have supplied all the points of our finite projective plane
with coordinates, and we know what the intersection points of any
two lines are.
14.10. Algebraic structures in finite projective planes 225
14.10.2. Addition and multiplication. Now we can define the

sum and product of two arbitrary elements a, b ∈ F by setting
(a, a + b) := [a][1, b], (a, a · b) := [a][b, 0].
The motivation behind this definition is that it is compatible with the
addition and multiplication induced on points of the projective line in
the case of the finite projective plane over the field F(pm ). The reader
is invited to return to the definition of finite projective planes over
a field, check that they can be supplied with coordinates as specified
above and that the operations defined above coincide with the ones
induced by the field F(pm ).
As we noted before, it is not always true that these operations
supply F with a field structure. They satisfy axioms of a structure
weaker than that of a field, which we now define.
14.10.3. Almost fields. An almost field is a set F with two binary

operations, called addition and multiplication, such that under addi-
tion F is an Abelian group with neutral element 0, the set F \ 0 is
a group (not necessarily Abelian) under multiplication and the right
distributive law is satisfied, i.e., (a + b)c = ac + bc.
When the left distributive law is not satisfied (such examples of
almost fields exist), the almost field is not even a ring. We will not
describe examples of this type or study almost fields in detail: they
are complicated and rather ugly, and we will limit our exposition
to the statements (without proofs) of two beautiful theorems and of
some open problems.
Theorem 14.10.4. (i) Given any finite almost field F , a projective

plane over F can be determined by using the construction from Section
14.8 with F replacing the field F(pm ).
(ii) Given any finite projective plane of order n, there is an al-
most field F (of order n − 1) using which the projective plane can be
constructed as indicated in (i).
The proof of (i) is similar to that in Section 14.8, while (ii) can be
proved by a tedious series of geometric constructions needed to verify
the numerous axioms of almost fields.
Theorem 14.10.5. A finite projective plane is a projective plane

over the field F(pm ) if and only Desargues’ theorem holds in it.
The “only if ” part was proved above (see 14.8.1), while the “if ”
part is another complicated series of artificial geometric constructions
ensuring the required algebraic axioms.
14.11. Open problems and conjectures

The main open problem here is the following: For what values of
q does there exist a finite projective plane of order q, and for what
values of q is the finite projective plane of order q unique?
We know that there exists one and only one projective plane of
the orders 2, 3, 4, 5, 7, 8 (see Problems 14.10–14.11). We also know
certain number-theoretic constraints forbidding projective planes of
certain orders.
Theorem 14.11.1 (Brack–Raiser). Let q ≡ 1 or 2 (mod 4). If there
exists a projective plane of order q, then q can be presented as the
sum of squares of two natural numbers.
The proof appearing in the original article by Bruck–Ryser1 is

not easy, and we omit it. This theorem forbids projective planes of
orders 6, 14, 21, 22, 30, etc.
Conjecture 14.11.2. The order q of any finite projective plane is a
prime number q = p or a power of a prime q = pm .
The first natural number q which does not meet the assumptions
of the conjecture is 6, and indeed one can prove (see Problem 14.9)
that there is no finite projective plane of order 6. The next such
number is 10, and it is only in 1991 that it was established, with the
aid of a supercomputer, that the conjecture holds there also. But
already for q = 12 the existence of a projective plane of order q is an
open question.
Conjecture 14.11.3. All the projective planes of prime order p are
Desarguian.
1
R.H. Bruck, H.J. Ryser, The non-existence of certain finite projective planes,
Canadian J. Math., Vol. 1 (1949), 88–93.
14.12. Problems 227
There are non-Desarguian projective planes of nonprime order.

The “smallest” one is of order 9 (Problem 14.16).
14.12. Problems
14.1. Construct an affine geometry having 4 points, and a finite affine
geometry having 9 points.
14.2. Suppose that one of the lines of the affine plane (P, L) from
Corollary 14.8.3 consists of q points. Prove that the plane P consists
of q 2 points.
Corollary 14.8.3 consists of q points. Prove that all other lines consist
of q points.
Corollary 14.8.3 consists of q points. Prove that L consists of q 2 = q
lines.
14.5. Suppose that one of the lines of the affine plane (P, L) consists
of q points. Prove that q + 1 lines pass through each point.
14.6. Prove that the finite affine plane AF(pm ) is a geometry in the
sense of Klein.
14.7. In the affine plane consisting of q 2 points for q = 3, construct

the system of lines passing through one of the points.
14.8. Describe the projectivization of the affine plane from Problem

14.5.
14.9*. Prove that there does not exist a finite projective plane of
order q = 6.
14.10. Prove that the projective planes of order 2, 3, 4, 5 are unique.
14.11*. Prove that the projective planes of order 7 and 8 are unique.
14.12*. Does there exist a finite affine plane of order q = 6?
14.13*. Find two nonisomorphic finite affine planes of order q = 9.

14.14. By adding “points at infinity” to the affine geometries of

orders 3, 4, 5, construct the corresponding finite projective planes.
14.15**. Give an example of a finite projective plane from which

one can obtain nonisomorphic affine planes by removing one line.
14.16*. Construct a non-Desarguian projective plane of order 9.

Chapter 15
The Hierarchy of
Geometries
This chapter is, in a sense, an overview of the book: in it, we try

to put some order in the category of geometries by summarizing in
a systematic way the relationships between the various geometries
studied in the previous chapters.
We do this in the order of increasing dimension, beginning with
lines (dimension one, see Section 15.1), then considering various kinds
of planes (dimension two, see Section 15.2), then three-dimensional
spaces (Section 15.4). (Geometries of dimension higher than three
are not considered because, in my opinion, the proper place for them
is in a linear algebra course.)
In Section 15.3, we give a systematic comparison of metric, affine,
and projective geometries, stressing the role of distance, ratio, and
cross ratio as invariants of the corresponding transformation groups.
In Section 15.5, we recall geometries with finite and discrete trans-
formation groups and indicate their positions in the “hierarchy” of
geometries described in this chapter.
The contents of this chapter (and of the whole book) are summa-
rized in Section 15.6 and, very succintly, in Figure 15.1.
229
230 15. The Hierarchy of Geometries
15.1. Dimension one: lines

In this section, we bring together some very simple and basic facts
about the classical one-dimensional geometries. Logically, we should
have presented this trivial material at the very beginning of this book,
but didn’t, because that would have been a very uninteresting way to
start the study of geometries.
15.1.1. The Euclidean line. The Euclidean line is modeled by the

real numbers R with transformation group Ismtr(R) consisting of par-
allel shifts (x → x + v, v ∈ R) and of reflections with respect to any
point. The subset of all parallel shifts is a subgroup of Isom(R), the
motion group Ismtr+ (R) of the Euclidean line. To specify an element
of Ismtr+ (R), it suffices to indicate any point of the line and its image.
The distance between two points x, y ∈ R, defined in the standard
way by d(x, y) := |x − y|, is an invariant of Ismtr(R). Given a point
x and its image y, there are two elements of Ismtr(R) that take x to
y: one is a parallel shift, the other a reflection in the midpoint of the
segment joining x and y.
15.1.2. The hyperbolic line. The hyperbolic line can be modeled

by the open interval (−1, 1) with the distance function
1
d(x, y) := | log(1, −1, x, y)|,
2
where 1, −1, x, , y is the cross ratio of the points 1, −1, x, y, i.e.,
|x − 1| |y + 1|
1, −1, x, y = · .
|x + 1| |y − 1|
Isometries of the hyperbolic line include shifts (the one-dimensional
analogs of parallel translations), given by the formula
x+v
Tv : [−1, 1] → [−1, 1], x → ,
xv + 1
where v is a real number of absolute value less than 1. The compo-
sition of any two shifts is a shift, and this operation has a physical
interpretation in the theory of relativity (see Section 9.4).
Shifts are not the only isometries of the hyperbolic line, there
are also symmetries in a point; details about them are relegated to
Problem 15.1.
15.1. Dimension one: lines 231
15.1.3. The circle. The geometry of the circle S1 is the one-dimen-

sional analog of spherical geometry. Its transformation group, O(2),
consists of rotations (by angles 0 ≤ ϕ < 2π) and symmetries; O(2)
contains SO(2), the group of all rotations, as a subgroup. To specify
an element r ∈ SO(2), it suffices to indicate a single point x ∈ S1 and
its image r(x).
If one defines the distance between two points on the circle in the
natural way (as the angle between the radii passing through them),
one does not obtain a metric space (the triangle inequality does not
hold). However, this distance function supplies the circle with a local
metric space structure.
There is a natural morphism between the Euclidean line and the
circle, namely the exponential covering map given by
R ϕ → eiϕ ∈ S1 .
This covering plays a key role in elementary topology (e.g., for defin-
ing the degree of circle maps), and its visualization shown in Figure
16.1 in the next chapter is one of the classical icons in mathematics.
The morphism exp is obviously a local (but not global) isometry.
15.1.4. The elliptic line. The elliptic line is the one-dimensional

analog of Riemann’s elliptic plane. It can be modeled as the cir-
cle with diametrically opposite points identified. Its transformation

group O(2) consists of rotations (by angles 0 ≤ ϕ < π) and sym-
metries (any symmetry has two fixed points whose diameters form

a ninety degree angle). However, the group O(2) is actually isomor-
phic to O(2) (rotations by ϕ corresponding to rotations by 2ϕ and
symmetries acting in the same way), and it is easy to construct an
isomorphism of geometries between the elliptic line and the circle.
Thus the elliptic line is just another model of the geometry of the
circle.
15.1.5. The affine line. For the points of the affine line Aff1 , we
can take the real numbers (just as for the Euclidean line), but its
transformation group, the affine group Aff(1), is “much bigger” than
Ismtr(R); in fact, it contains Isomtr(R) as a subgroup. The group
Aff(1) consists of all transformations of the form x → ax + b, where

a, b ∈ R and a
= 0.
To specify an affine transformation, it suffices to indicate an or-
dered pair of points and their images (cf. Problem 15.2).
15.1.6. The projective line. The projective line RP 1 is obtained

from the real line R by adding the point at infinity ∞ so that, topo-
logically, it is the circle. Its transformation group consists of those
bijections of R ∪ ∞ that preserve the cross ratio of any four points
a, b, c, d, i.e.,
[(a − c)/(b − c)] : [(a − d)/(b − d)]
(for the definition of the cross ratio when one of the points a, b, c, d is
∞, see Problem 15.4 ).
Note that RP 1 can also be defined as the set of lines passing
through the origin of the plane R2 ; its transformation group is the
group of nonsingular two-by-two matrices considered up to multipli-
cation of the columns of the matrix by nonzero constants.
These two definitions of the projective line are equivalent, i.e.,
they determine isomorphic geometries; the proof is the object of Prob-
lem 15.5.
15.2. Dimension two: planes

Here we summarize some properties of the main objects of study in
this book – the classical two-dimensional geometries. We will use the
following (not very standard) terminology concerning transformation
groups: we say that a transformation group possesses two degrees of
freedom if any of its elements is determined by two points and their
images. (Note that this notion does not coincide with the dimension
of the group when the latter has a Lie group structure.)
15.2.1. The Euclidean plane. We denote the Euclidean plane by

R2 and assume that it is familiar to the reader (For the basic prop-
erties of R2 , the reader is referred to Chapter 0.) The plane is non-
compact and orientable. Its transformation group Ismtr(R2 ) contains
the subgroup of motions Ismtr+ (R2 ) which possesses two degrees of
freedom in the sense that any motion is determined by two points
15.2. Dimension two: planes 233
and their images. The basic invariant of this geometry is the distance
between two points.
15.2.2. The sphere. We denote the sphere by S2 . It is compact and

orientable. Its transformation group, denoted by Ismtr(S2 ) = O(3),
contains the subgroup Ismtr+ (S2 ) = SO(3) of orientation-preserving
isometries, which has two degrees of freedom. The basic invariant
of this geometry is the distance between two points. Note that this
distance is not the distance induced from Euclidean space by the
standard embedding of the sphere; it is the angular distance (or,
which is the same thing, the geodesic distance) between points. A
detailed exposition of spherical geometry appears in Chapter 6.
15.2.3. The hyperbolic plane. We denote the hyperbolic plane

by H2 . It is noncompact and orientable. Its transformation group,
denoted Ismtr(H2 ), contains the subgroup Ismtr+ (H2 ) of orientation-
preserving isometries (motions), which possesses two degrees of free-
dom. The basic invariant is the distance between two points. A
detailed exposition of hyperbolic geometry appears in Chapters 7–10.
15.2.4. The elliptic plane. We denote the elliptic plane by Ell2 .

It is compact and nonorientable. Its transformation group, denoted
Ismtr(Ell2 ), possesses two degrees of freedom. The basic invariant of
this geometry is the distance between two points. A somewhat more
detailed exposition of elliptic geometry appears in Section 6.7.
15.2.5. The affine plane. We denote the affine plane by Aff2 . It

is noncompact and orientable. Its transformation group, denoted by
Aff(2), contains the subgroup Aff+ (2) of orientation-preserving affine
transformations, which has three degrees of freedom. The basic in-
variant of this geometry is the ratio of three collinear points.
15.2.6. The projective plane. We denote the projective plane by

RP 2 . It is compact and nonorientable. Its transformation group,
denoted by Proj(2), has four degrees of freedom. The basic invariant
of this geometry is the cross ratio of four collinear points. A detailed
exposition of projective geometry appears in Chapter 12.
15.3. From metric to affine to projective

Before passing to dimension three (in the next section), we glance at
what happens when we move from two-dimensional metric geometries
to the projective plane via the affine plane.
15.3.1. Metric. Many geometries (but not all) possess a natural

distance function (in another terminology – a metric). This is the case
for Euclidean (parabolic) geometry, as well as hyperbolic, elliptic, and
spherical geometries. For such metric geometries, the corresponding
transformation group (which gives them the structure of a geometry in
the sense of Klein) can be defined as the group of distance-preserving
transformations (isometries) with composition as the group operation.
15.3.2. Affine. Euclidean geometry in dimension two (R2 ) is a sub-

geometry of affine plane geometry (Aff2 ): the latter is obtained from
R2 by keeping the same set of points but increasing the transforma-
tion group. Namely, choosing any three noncollinear points O, X, Y ,
we fix a coordinate system in which we write affine transformations
in the form

x = ax + by + k,
(15.1)
y = cx + dy + l.
where a, b, c, d, k, l ∈ R, ad − bc
= 0, and (x, y), (x , y ) are the coordi-
nates of the preimage and the image points in the basis OXY . Affine
transformations either preserve or reverse orientation, according to
the sign of the determinant ad − bc.
Unlike Euclidean geometries, hyperbolic, elliptic, and spherical
geometries have no “affine counterpart”, e.g., the transformation group
of the hyperbolic plane cannot be increased to a larger group (like the
affine group) acting on the same set of points; a more precise formu-
lation appears in Problem 16.7.
15.3.3. Projective. The projective plane is obtained from the affine

plane by adding the “line at infinity” and considerably increasing the
transformation group (so that the line at infinity is not special, it is
“just as good” as all the other lines, i.e., it intersects them all and can
be transformed into any other line by a projective transformation).
15.4. Three-dimensional space geometries 235
The group Proj(2) of projective transformations can be defined

via the homogeneous coordinate model or via the cross ratio of four
collinear points. The group of projective transformations has an extra
degree of freedom as compared to the affine group (see 15.2.6 above).
The projective plane contains, as subgeometries, not only the
affine plane, but also Euclidean, hyperbolic, and elliptic geometries.
It does not contain spherical geometry (see Problem 15.8).
This section is summarized by the following table.
Table: Properties of the two-dimensional geometries
compactness orientability degrees of freedom invariant

R2 − + 2 distance
S2 + + 2 distance
H2 − + 2 distance
Ell2 + − 2 distance
Aff2 − + 3 ratio
RP 2 − − 4 cross ratio
15.4. Three-dimensional space geometries

In this brief section, and throughout this book, we do not study three-
dimensional geometries in any detail. We have, however, mentioned
and even defined three-dimensional hyperbolic, elliptic, and projective
geometries (H3 , Ell3 , and RP 3 ); also, we assume that the reader is
familiar with Euclidean space geometry R3 and, possibly, affine space
geometry Aff3 .
The only goals of this brief section is to point out how the five
three-dimensional geometries mentioned above are related and to con-
nect 3D projective geometry with 2D spherical geometry, which (as
mentioned above) is not a subgeometry of the projective plane, un-
like two-dimensional elliptic, hyperbolic, and parabolic (Euclidean)
geometry.
The relationship between the five 3D geometries in question is
the same as that of their two-dimensional counterparts, namely: Eu-
clidean space geometry is a subgeometry of affine space geometry,
which in turn is a subgeometry of projective space geometry; further,

elliptic space geometry is a subgeometry of projective space geome-
try, and, finally, hyperbolic space geometry is also a subgeometry of
projective space geometry.
Now the two-dimensional sphere lies in Euclidean space, and of
course S2 is a subgeometry of the geometry R3 , i.e., it is a subset of
the space R3 and its transformation group O(3) is a subgroup of the
group Isom(R3 ). By transitivity, S2 is a subgeometry of RP 3 .
15.5. Finite and discrete geometries

In this very brief section, we only list the finite and discrete geometries
appearing in the first half of this book and indicate their relationships.
The two discrete geometries studied in Chapters 4 and 5, namely
the geometries of regular tilings (Fedorov geometries) and the geome-
tries of reflections (Coxeter geometries), are both discrete subgeome-
tries the Euclidean plane R2 . The finite geometries from Chapter 3,
i.e., those of the regular polyhedra (Platonic geometries), are subge-
ometries of the geometry (S2 , O(3)); in turn, they contain many of the
“toy geometries” studied in Chapter 1; other toy geometries are sub-
geometries of R2 . Another series of discrete geometries is formed by
the subgeometries of hyperbolic geometry briefly mentioned in Chap-
ter 7. These geometries are, in a sense, both Fedorov and Coxeter;
they are based on regular n-gons filling the hyperbolic plane. There
are actually many other discrete subgeometries of hyperbolic plane
geometry, but they are not studied in this book. Another class of
finite geometries is constituted by the triangular Coxeter tilings of
the sphere (see Section 6.6); unlike the Platonic geometries, they are
subgeometries of two-dimensional spherical geometry.
Note that we do not mention the finite geometries from Chapter
14 in the present chapter because they don’t “fit in” – they are not
subgeometries of projective geometry.
15.6. The hierarchy of geometries

15.6.1. The hierarchical tree of geometries. This section is es-
sentially a commentary on Figure 15.1. In it, all the main geometries
15.6. The hierarchy of geometries 237
studied in this book (except the finite geometries from Chapter 14)
are placed on five levels; from bottom to top, these levels are: pro-
jective, affine, metric, discrete, finite. The geometries all appear at
the appropriate levels; they are joined by arrows, which stand for in-
jective morphisms. If we regard the geometries as vertices and the
arrows as edges, we obtain a directed graph; this graph is a rooted
tree: starting from any vertex (geometry) the arrows lead us to the
root RP 3 . This means that all the geometries we studied (except
those in Chapter 14) are subgeometries of projective space geometry
RP 3 .
This substantiates Cayley’s famous utterance “Projective geom-
etry is all geometry.” A more correct, but less striking, formulation
would be “Most geometries are parts of projective geometry.”
Figure 15.1. Hierarchical tree of geometries.

15.6.2. Let us climb up this “hierarchical tree of geometries”. The

most interesting climb begins if we first move to the vertex RP 2
(still on the projective level); then we have the choice of three main
itineraries: via affine plane geometry to the Euclidean plane (with all
its beautiful discrete and finite subgeometries), or, skipping the affine
level, directly to the elliptic or to the hyperbolic plane, the latter also
rich in discrete geometries. (These discrete geometries, often called
Fuchsian, were mentioned in Chapter 7 and in the Problems to that
Chapter, but were not studied systematically in this book.)
From the root RP 3 , we could have chosen the itinerary leading us
to spherical geometry and ultimately to its finite Coxeter subgeome-
tries, or, along another limb, to the finite geometry level, reaching the
Platonic geometries, which in turn contain most of the “toy geome-
tries” – thus returning us to Chapter 1. Which of these two climbs is
more beautiful, is a matter of taste.
15.7. Problems
15.1. Write out a formula allowing to find the image Sa (x) of an
arbitrary point x ∈ (0, 1) of the hyperbolic line under the symmetry
in the point a ∈ (0, 1).
15.2. Express the parameters a and b of the affine transformation

given by x → ax + b in terms of the coordinates x1 , x2 of two points
and of their images y1 , y2 .
15.3. Express the parameters a, b, c, d, k, l of the affine transformation

(15.1) in terms of the coordinates of three points and of their images.
15.4. Define the cross ratio of four points on the projective line when
one of the points is ∞.
15.5. Show that there is a bijection between the point sets in the
two definitions of the projective line given in 15.1.6 and that the
corresponding transformation groups are isomorphic; prove that the
two definitions yield isomorphic geometries.
15.7. Problems 239
15.6. Show that there is no extension of the transformation group of

hyperbolic geometry and no definition of a vector that would endow
the set of vectors with a vector space structure over R.
15.7. Prove that two-dimensional spherical geometry is not a subge-

ometry of 2D-projective geometry.
Chapter 16
Morphisms of
Geometries
In this chapter, we describe concrete examples and some classes of

morphisms of geometries. They are particular cases of classes of
maps normally studied in algebraic topology or Lie group courses,
namely covering spaces, vector bundles, and principal G-bundles. The
formal definitions of the particular cases that we consider (and call
“geometric”) endow the morphisms with “more structure” than that
appearing in the definitions of the corresponding maps from topol-
ogy courses, but, surprisingly, practically all the examples considered
by topologists actually possess these structures (although topologists
usually ignore them).
The chapter begins with four sections containing concrete exam-
ples of morphisms of the four types listed above. These examples
include such beautiful constructions as the Hopf bundle, the Grass-
mannian, the Stiefel-over-Grassmann bundle, and the Milnor univer-
sal G-bundle. Their descriptions are quite elementary (although a
little linear algebra and some elementary topology is needed for some
of the examples), but the formal general definitions need more than
the elementary prerequisites required for the previous chapters of
this book. Thus, Section 4 (on Lie groups) requires the notion of
241
242 16. Morphisms of Geometries
smooth manifold, a good understanding of basic linear algebra and

basic topology.
The remaining sections contain the main definitions and a little
theory, including two universality theorems about geometric vector
bundles and geometric principal G-bundles, which yield effective con-
structions for obtaining all geometric vector bundles and all geometric
principal bundles over a given base. Here, as in Section 4, some non-
elementary mathematics is needed, but none of the standard tools of
algebraic topology (the fundamental group, homotopy and homology
groups) are used, the main tools being the transformation groups of
the source and target geometries.
16.1. Examples of geometric covering spaces

16.1.1. Covering of the elliptic plane by the sphere. There
is an obvious morphism of the geometry of the sphere (S2 : SO(3))
onto the elliptic plane Ell2 (see Section 6.7) obtained by identifying
antipodal points of the sphere. Another way of saying this is that we
consider the subgroup Z2 ⊂ SO(3) acting on the sphere by symmetry
w.r.t. the sphere’s center, and take the quotient of S2 by the two-
point orbits of this action. Thus we obtain a morphism of geometries
S2 → Ell2 such that the inverse image (which we call the fiber) of any
point p ∈ Ell2 consists of two points.
16.1.2. The exponential map. The exponential function x → eix ,

studied by the reader in calculus courses, is actually a morphism of
geometries, namely the morphism exp : R → S1 given by the rule
ϕ → eiϕ , where the circle S1 is understood as the geometry of the set
of unimodular complex numbers, i.e.,
S1 = {z ∈ C : |z| = 1}
(acting upon itself by multiplication), while the R is the geometry of
the set of real numbers (acting upon itself by addition).
The classical picture of the exponential map (as it usually appears
in elementary topology textbooks) is shown in Figure 16.1.
Note that the inverse image of any point ϕ of R, called the fiber of
the morphism exp, is in one-to-one correspondence with the integers;
16.1. Examples of geometric covering spaces 243
Figure 16.1. The exponential map.
in particular, the inverse image of the point 1 ∈ S1 is the subgroup

Z ⊂ R. Each fiber is also a geometry (with transformation group Z
acting by addition).
16.1.3. The winding map wn : ϕ → einϕ . This map is a morphism

of geometries S1 → S1 with fiber Zn , the set of integers modulo n with
the natural action of Zn on itself by addition.
16.1.4. The partial order in winding maps of the circle. Let

us say that the morphism wn is higher than the morphism wm or
wm is lower than wn (denoted wn wm ) if there exists a morphism
of geometries γ : S1 → S1 such that γ ◦ wm = wn (note that here
γ ◦ w means that w is performed first, then comes γ). For example,
w6 w3 , because if we take γ := w2 , then obviously γ ◦ w3 = w6 .
Clearly, is a partial order relation in the set of surjective morphisms
onto the circle (not only in the case of the winding maps wn , where
n = 1, 2, 3, . . . ).
There is an interesting connection between the relation for the
morphisms wn and the divisibility of the numbers n; see Problem
16.1.
16.1.5. The exponential map as the highest covering of the

circle. The exponential map is universal for the winding maps of
the circle in the sense that it is higher than any winding map of the
circle, i.e., for any morphism wn : S1 → S1 there exists a morphism
γ : R → S1 such that wn ◦ γ = exp. The proof is relegated to Problem
16.2.
16.1.6. The flat torus. The flat torus is the unit square
{(x, y) : 0 ≤ x ≤ 1, 0 ≤ y ≤ 1} ⊂ R2
with opposite sides identified ((x, 0) ∼ (x, 1) and (y, 0) ∼ (y, 1)). It is
supplied with the natural metric (for example, the distance between
the points (1/2, 1 − ε) and (1/2, ε), where ε < 1/4, is equal to 2ε).
This metric (see Problem 16.6) gives rise to the corresponding group
of isometries, which determines the structure of this geometry, that
of the flat torus. The geometry of the flat torus is quite different from
that of the torus embedded in R3 in the usual way, e.g., by means of
the equation
2
x2 + y 2 − 2 + z 2 = 1.
The metric of the above embedded torus (defined via geodesics) differs
from the metric of the flat torus, so that the two tori have different
isometry groups and thus are different geometries.
16.1.7. Remark. Actually, neither of these two geometries on the

torus is the “immanent” one. The true geometry of the torus is the
one given by the standard embedding of the torus in complex space
C2 , i.e.,

C2 ⊃ T = (r1 eiϕ1 , r2 eiϕ2 ) ∈ C2 : r1 = 1, r2 = 1 .
16.1.8. The universal covering of the flat torus. There is a

natural morphism of the plane R2 onto the flat torus T2F L given by
the rule
R2 (x, y) → (x mod 1, y mod 1) ∈ T2F L ,
which topologists call the universal covering of the torus. This mor-
phism is obviously a local isometry. Its fiber (i.e., the inverse image of
points of the torus) is the lattice Z ⊕ Z. An algebraist would describe
16.2. Examples of geometric G-bundles 245
this morphism as the quotient map taking the group R ⊕ R to its

quotient by its (normal) subgroup Z ⊕ Z.
16.1.9. Covering of the torus by the cylinder. There is an ob-

vious morphism of the flat cylinder R × S1 onto the flat torus T2F L
with fiber Z. The details are left to the reader (see Problem 16.7).
16.1.10. Covering of the flat torus by itself. There are many

geometric morphisms of the flat torus onto itself. Their investigation
is left to the reader (see Problem 16.8).
16.2. Examples of geometric G-bundles

We begin this section with a description of the famous Hopf bundle,
then study some other (less intricate) morphisms of geometries onto
the two-sphere S2 (regarded as the geometry of the rotation group
SO(3) acting on the sphere) with fiber (i.e., inverse image of points)
the circle. The section continues with more examples of morphisms
in which the main protagonists are some of the so-called “classical
groups”.
16.2.1. The Hopf bundle. This is one of the most intricately beau-
tiful geometric constructions in mathematics; it has a very simple
analytic description. Consider the sphere S3 as the subset of the
two-dimensional complex space C2 given by the formula

(z1 , z2 ) ∈ C2 : |z1 |2 + |z2 |2 = 1 .
We regard it as the geometry (S3 : SO(4)).
The group S1 = {eiϕ } acts on S3 by multiplication of coordinates
(z1 , z2 ) → (eiϕ z1 , eiϕ z2 ).
The quotient of S3 by this action is the complex projective line CP 1 ,
which is another name for the sphere S2 (see Problem 16.3).
Thus we obtain a surjective map h : S3 → S2 , called the Hopf
bundle, whose fiber (i.e., the inverse image of any point of S2 ) is the
circle S1 . The orbits of the S1 action on the 3-sphere S3 are circles
parametrized by the 2-sphere and filling up S3 . These circles are
linked together pairwise, as links of a chain. An attempt to show

what this looks like appears in Figure 16.2.
The figure represents S3 as Euclidean space with “the point at
infinity” added (it lives at “both extremities” of the vertical axis of R3 ,
transforming the axis into one of the orbit circles). The figure shows
only two other linked orbit circles; they lie on the shaded torus; the
other curves in the picture represent the sections of other concentric
tori by the vertical plane of the figure.
To really visualize the Hopf bundle, the reader is referred to Eti-
enne Ghys’s home page and his “Dimensions” – a series of beautiful
animations (two of them show the Hopf bundle in motion).
Figure 16.2. The Hopf bundle.
16.2.2. The natural morphism S2 × S1 → S2 . The Cartesian

product S2 × S1 possesses a natural geometry structure (Problem
16.3. Lie groups 247
16.4) so that the projection on the first factor pr1 : (s, ϕ) → s is a

morphism of geometries with fiber S1 .
16.2.3. A nontrivial morphism onto S2 with fiber S1 . Consider

the Cartesian product S2 × [0, 1]; let α be the involution of the sphere
S2 obtained by reflection in the equatorial plane; identify all pairs of
points of the form (s, 0) and (α(s), 1). The obtained space is a three-
dimensional manifold which may be supplied with the structure of a
geometry (Problem 16.5). This geometry may be mapped onto the
2-sphere via the assignment
S2 × [0, 1) (s, t) → s ∈ S2 ,
and this map will be a morphism of geometries with fiber S1 . In topol-
ogy courses, the morphism in Subsection 16.2.2 is said to be a trivial
S1 -bundle (because it is the projection along the factor S1 of a Carte-
sian product), whereas the morphism from the present subsection is
a nontrivial bundle.
16.3. Lie groups

This section is not an introduction to the (very rich) theory of Lie
groups. It only contains some basic definitions and a few examples
that will be used in subsequent sections. To understand the present
section, the reader must know some linear algebra and be familiar
with the notions of smooth manifold, smooth map, and diffeomor-
phism.
16.3.1. Main definitions. A Lie group is a smooth manifold G

which is also a group such that the maps G × G → G and G → G
given by (g, h) → gh and g → g −1 are smooth. In other words, it is
the geometry of a group G whose elements are the points of a smooth
manifold and which acts on itself smoothly by multiplication from the
right. Thus Lie groups are geometries, while morphisms of Lie groups
are defined as morphisms of the corresponding geometries.
Suppose E and B are Lie groups, p : E → B is a surjective
morphism, and B1 ⊂ B is a Lie subgroup; then one defines the re-
striction of p to B1 , denoted p|B1 , in the natural way. More generally,
if f : A → B is a morphism of Lie groups, then the pullback of p to

A, denoted f ∗ p, is defined by setting

E1 := (a, e) ∈ A × E : f (a) = p(e) and f ∗ p((a, e)) := f (a).
In that situation, there is a canonical morphism E1 → E given by the
formulas ϕ(a) = f (a), Φ((a, e)) = e.
Actually, the definition of pullback can be given in a much more
general context than that of Lie groups. Thus if p : E → B is any
surjective map and f : A → B is any map, then the pullback f ∗ p is
defined exactly as above.
16.3.2. Examples of Lie groups. (1) The simplest examples of

Lie groups are the groups R, C, Rn , Cn acting upon themselves by
addition.
(2) The classical groups GL(n) (nonsingular linear transforma-
tions of Rn ), O(n) and SO(n) (orthogonal and orientation-preserving
orthogonal transformations of Rn ), U(n) (Hermitian transformations
of Cn ), well known from linear algebra, possess obvious Lie group
structures.
(3) Such well-known groups from various parts of mathematics
as the Weyl group, the group SL(n), the group of upper triangular
matrices, all have obvious Lie group structures.
16.4. Examples of geometric vector bundles

Grassmannians (or Grassmann manifolds) Gnk are geometries which
generalize projective spaces in a natural way: their points are k-
dimensional subspaces of Rn , which in the case k = 1 is just the
definition of RP n−1 . These geometries possess a beautiful theory
(involving such neat objects as Plücker coordinates, Schubert cells,
etc.), but we do not go into this theory here: our aim is to describe
certain morphisms of geometries in which Grassmannians play the
key role.
16.4.1. Definition of Grassmann and Stiefel manifolds. The

Grassmann manifold Gnk is the set whose points are the k-dimensional
linear subspaces L of the n-dimensional vector space over R. This set
16.4. Examples of geometric vector bundles 249
has a natural topological structure, a smooth manifold structure, and

a geometric structure (Problem 16.9).
The Stiefel manifold Vkn is the set whose points are the k-dimen-
sional orthonormal frames of Euclidean space Rn . This set has a
natural topological structure, a smooth manifold structure, and a
geometric structure (Problem 16.10).
16.4.2. Some important morphisms. (1) The canonical Grass-

mann bundle γkm : Ekm → Gm m
k , where Gk is the Grassmann manifold,

Ekm := (L, r) ∈ Gmk ×R :r ∈L ,
m
and γkm is the natural projection (L, r) → L, is a morphism of geome-

tries, provided Ekm is supplied with the natural geometric structure
(Problem 16.11). Its fiber (γkm )−1 (pt) is the k-dimensional vector
space over R.
k ⊂ Gk and Ekm ⊂ Ekm+1 ,

m+1
(2) There are obvious inclusions Gm
which allow us to define G∞ ∞ ∞ ∞ ∞
k , Ek , γk : Ek → Gk by passing to
∞
the inductive limit. The mapping γk thus obtained is a morphism of
geometries that we will call the infinite canonical Grassmann bundle.
(3) The Stiefel-over-Grassmann bundle σkm : Vkm → Gm k , which
assigns to each frame of Vkm the linear subspace that it spans, is a
morphism of geometries.
16.4.3. Universality of the canonical Grassmannian. For a

certain class of geometric morphisms with fiber Rk (that we do not
specify here, see [5], Vol. II), the infinite canonical Grassmann bun-
dle possesses the following remarkable universality property: for any
morphism ξ k : E → B of this class, there exists a morphism of ge-
ometries p : B → G∞ ∗ ∞
k such that the pullback p γk is isomorphic
k
to ξ .
Thus the infinite canonical Grassmann bundle contains, so to
speak, all the information needed to construct all morphisms from
a vast class of geometric morphisms with fiber Rk . Note that the
pullback construction is an effective one, so the above theorem yields
an effective method for finding new examples of morphisms of geome-
tries.
16.5. Geometric G-bundles

In topology, a G-bundle is the quotient map from a topological space
supplied with a right action of a topological group G onto the orbit
space of this action. Here we study a particular case of the notion
of G-bundle, in which we consider geometries rather than topological
spaces and Lie groups rather than topological groups.
16.5.1. Main definitions. Let (E : Γ) be geometry, Γ a Lie group,

and G a subgroup of Γ; by a geometric G-bundle p : E → B we mean
the projection of the space E onto the orbit space B = E/G of the
action of G on E.
A morphism (ϕ, Φ) of two geometric G-bundles pi : Ei → Bi ,
i = 1, 2, is defined as a commutative diagram
Φ
E1 −−−−→ E2
⏐ ⏐
⏐p1 ⏐p2
! !
ϕ
B1 −−−−→ B2 ,
in which ϕ and Φ are morphisms of geometries. A morphism of geo-
metric G-bundles is an isomorphism if ϕ and Φ are isomorphisms of
geometries.
A geometric G-bundle p : E → B is said to be principal if the
group G acts transitively on E. Since the action of G is transitive,
each orbit is isomorphic to G; in other words, the fiber of the bundle
p is G. The class of all principal G-bundles also forms a category.
16.5.2. Examples. (1) The identification of antipodal points on the

n-dimensional sphere is a principal-Z2 bundle over RP 2 .
(2) The Hopf bundle (see 16.3.1) is a principal S1 -bundle.
(3) The natural projection of the Stiefel manifold Vkn onto the
Grassmann manifold Gnk (see 16.4.2(3)) is a principal O(k)-bundle.
(4) The canonical Grassmann bundle γkm : Ekm → Gm k is a geo-

metric GL(k)-bundle, but it is not principal. Its fiber is Rk .
16.6. The Milnor construction 251
16.6. The Milnor construction

The Milnor construction associates with any Lie group G (in fact,
any topological group) a certain geometric morphism ωG that allows
us to classify all geometric G-bundles over a given geometry. The
construction is based on the notion of join (see, e.g., [5], Vol. I). The
join of two topological spaces X and Y , denoted X ∗Y , is the quotient
space of X × [0, 1] × Y by the equivalence relation
(x, 0, y) ∼ (x , y), (x, 1, y) ∼ (x, 1, y ).
For example, [0, 1] ∗ [0, 1] is the 3-simplex, while S1 ∗ S1 is the sphere
S3 . The last example is a beautiful geometric fact that every math-
ematician should understand, along with the (related) fact that the
3-sphere can be glued together from two “linked solid tori”.
16.6.1. The construction. Let G be any Lie group; denote by

EG (n) := G ∗ G ∗ · · · ∗ G (n factors)
the n-fold iterated join of the group G with itself. Obviously,
G ⊂ EG (2) ⊂ EG (3) ⊂ · · · ⊂ EG (n) ⊂ · · · ⊂ EG ,
where EG is the inductive limit of EG (n) as n → 0.
Consider the action of G in EG by right shifts. The corresponding
bundle
ωG : EG → BG = EG /G
is called the universal geometric G-bundle, its base is called the clas-
sifying space of the Lie group G. Similarly, one defines the bundle
ωGn
: EGn
→ BG n
called (briefly) the n-universal G-bundle, while its
base is called the n-classifying space of the group G (in less shortened
form, the classifying space up to dimension n).
16.6.2. Examples. (1) The classifying space of the group S1 is CP ∞

and the corresponding universal bundle is
ωS1 : ES1 = S∞ → CP ∞ .
The k-classifying space of S1 is CP k and ESk1 = S2k+1 .
(2) The classifying space of the group Z2 is RP ∞ and EZ2 = S∞ .
The k-classifying space of Z2 is RP k and ESk1 = Sk .
16.6.3. The universality property. For a certain class of geo-

metric principal G-bundles (that we do not specify here, see [5],
Vol. II), the universal geometric G-bundle possesses the following re-
markable universality property: for any geometric principal G-bundle
ξG : E → B of this class, there exists a morphism of geometries
p : B → G∞ ∗ ∞
k such that the pullback p (γk ) is isomorphic to ξG .
Thus the universal geometric G-bundle contains, so to speak, all
the information needed to construct all G-bundles from a vast class of
geometric principal G-bundles. Since the pullback construction is an
effective one, the above theorem yields an effective method for finding
new examples of geometric principal G-bundles.
16.7. Problems
16.1. Given two winding maps of the circle, wk and wl (see 16.1.3),
find the lowest winding map wn which is higher than both wk and
wl .
16.2. Prove the universality property of the exponential map, i.e.,
show that it is higher than any winding map (see 16.1.5).
16.3. Prove that the complex projective line CP 1 is isomorphic as a
geometry to the sphere S2 .
16.4. Indicate the transformation group which supplies the Cartesian
product S1 × S2 with its natural geometric structure.
16.5. Indicate the transformation group which supplies the manifold
constructed in 16.2.3 with its natural geometric structure.
16.6. Give a precise definition of the metric of the flat torus.
16.7. Define the geometry of the infinite flat cylinder S1 × R. Show
that there is an uncountable infinity of different geometric morphisms
with fiber Z of the flat cylinder onto the flat torus. How many non-
isomorphic morphisms of that type are there?
16.8*. Show that the covering of the flat torus by the plane has a
universality property similar to that of the exponential.
16.9. On the Grassmann manifold Gnk , define the structure of a
topological space, of a smooth manifold, and of a geometry.
16.7. Problems 253
16.10. On the Stiefel manifold Vkn , define the structure of a topolog-

ical space, of a smooth manifold, and of a geometry.
16.11. Define the natural geometric structure on the set Ekm (see
16.4.2).
16.12*. Describe a map f : S1 → CP ∞ such that the pullback f ∗ ωS1
is the winding map wn .
Appendix A
Excerpts from Euclid’s

“Elements”
In this appendix, we present some excerpts from Book I (which deals

with plane geometry) of Euclid’s “Elements” (for an English transla-
tion, see [8]), interspersed by comments. I feel that some comments
are absolutely necessary here, because many of the formulations ap-
pearing in the “Elements” are so worded that they sound quite weird
to the contemporary reader’s ear, conditioned as it is by the modern
approach to mathematics. Actually, as we shall see, Euclid’s wording
is completely natural, once one understands the underlying meaning
of geometry for the Ancient Greeks.
Let us first recall that the root “geo” stands for earth, and the
root “metr” means measurement, so that the plane geometry of Eu-
clid is an abstract version, we would say a model, of the activity of
the land surveyor, working on a piece of paper (more precisely, a pa-
pyrus) rather than within the real-life landscape. Second, Euclid’s
contemporaries did not distinguish geometry and physics as we do;
in fact, there was no science called physics at the time, and plane ge-
ometry was just the two-dimensional physical model of our universe.
So Books XI–XIII of the “Elements”, which treat space geometry,
were nothing more for the Greeks than the theory of their (three-
dimensional) physical space.
255
256 A. Excerpts from Euclid’s “Elements”
Postulates of Book I
The Postulates (we would call them axioms) are not abstract state-
ments about points, straight lines, and circles; they simply prescribe
(except Postulate IV) what the geometer (i.e., land surveyor) can
draw on his papyrus. Euclid writes:
Let the following be postulated:
I. To draw a straight line from any point to any point.
II. To produce a straight line continuously in a straight line.
III. To describe a circle with any center and distance.
IV. That all right angles are equal to one another.
V. That, if a straight line falling on two straight lines makes the
interior angles on the same side less than two right angles, the two
straight lines, if produced indefinitely, meet on that side on which are
the angles less than the two right angles.
Of course, our land surveyor cannot draw an infinite line, and
the expression “straight line” in Postulates I, II, V means “line seg-
ment” in our terminology, and this is what makes Postulate II neces-
sary. (Note that this postulate cannot be adequately translated into
a meaningful statement in the modern terminology.)
Note, further, that no uniqueness statements appear in Postulates
I, II, III. This is quite natural, because all three are prescriptions
for admissible well-defined actions of the land surveyor, and it goes
without saying that, for Euclid’s contemporaries, there was only one
way to perform this action.
Postulate III sounds unnecessary to our ear; we would simply
define the circle as the locus of all points lying at the given distance
(the radius) from a given point (the center of the circle). But it
is, for the Greeks, an informative assertion, saying, as it does, that
drawing a circle is an admissible and well-defined action. We must
also remember that for the Greeks the circle, like the line, is not a set
of points, but a certain geometric entity.
Postulate IV is different from the other four in that it is an ob-
servation rather than a guide to action. In it, the word “equal” is
The Common Notions 257
used. What does it mean? Should it be understood as “congruent”?

We shall return to this question below.
Finally, a few comments on the Fifth Postulate (once described
as “the most important utterance in the history of science”). What
we regard as a better formulation of this axiom, namely one and only
one parallel to a given line passes through a given point, is absolutely
inadmissible for Euclid. Indeed, it involves the notion of infinite line
(which the geometer cannot draw) and is, essentially, a negative state-
ment, saying that there does not exist a common point to two specific
(infinite!) straight lines. On the contrary, Euclid’s version is very
constructive: it says that if a geometer performs certain actions, he
will obtain the intersection point of two lines, and even indicates the
whereabouts of this point.
The Common Notions

For us Euclid’s “common notions” sound like axioms specifying the
meaning of the word “equal”, but what did Euclid have in mind?
This is what he says:
1. Things which are equal to the same thing are also equal to
one another.
2. If equals be added to equals, the wholes are equal.
3. If equals be subtracted from equals, the remainders are equal.
4. Things which coincide with one another are equal to one an-
other.
5. The whole is greater than the part.
From the modern point of view, the first and fourth common
notions are the transitivity and the reflexivity of the equality relation,
and (if one adds symmetry, which for the Greeks was undoubtedly
implicitly assumed), we get the definition of an equivalence relation
in the set-theoretical sense.
But what are the “things” that appear (explicitly and implicitly)
in all five of the above statements? Are they geometric entities or
are they numbers (e.g. lengths of segments) as one is led to suspect
upon reading the second and third common notions? Actually, they
are both: we must remember that the Ancient Greeks had no theory
of real numbers, the (lengths of) line segments were the real numbers.
√
Thus, when the Pythagoreans discovered the irrationality of 2, this
fact was stated as the incommensurability of two line segments: the
diagonal of the unit square and its side. Further, “things” are not
only line segments, they can also be (the measure of) an angle, the
area of something, etc.
Finally, what does the word “equal” mean? Congruent? What
does the word “greater” in the fifth common notion mean? These are
difficult questions, but one should have in mind that nowhere does
Euclid mention any transformations of the plane, there are no parallel
translations, no rotations, no superpositions of triangles in Euclid’s
Book I, so that there are no hints whatsoever in it about the Klein
approach to geometry based on transformations. Thus equal triangles
are those which have three equal sides and whose corresponding an-
gles are also equal (however, a kind of superposition argument does
appear in the proof). It is less clear what the words “added” and
“subtracted” in 1 and 2 mean, but judging from some of the proofs
in Book I, they refer to the union and set difference of geometric ob-
jects, although of course such entities as angles and line segments are
not defined as sets of points.
The Definitions of Book I
We mentioned in Chapter 11 that the Ancient Greeks realized that

to develop geometry as a deductive science based on rigorous proofs
without any logical vicious circles, one had to start by formulating
certain statements without proof – the postulates (axioms in our ter-
minology). However, they did not use the same type of argument for
definitions, and so did not believe, as we do, that in a rigorous math-
ematical theory you cannot define concepts in terms of each other
without vicious circles, unless you start with some basic undefined
notions which are only named and not specified in any way.
But we should not regard the absence of undefined concepts in
Euclid as a logical defect of his exposition. Indeed, the geometric
entities considered were, for Euclid, objects of the physical world and
could be meaningfully described as such. Let us see how he describes
them.
1. The point is that which has no part.
2. A line is breadthless length.
3. The extremities of a line are points.
4. A straight line is a line which lies evenly with the points on
itself.
5. A surface is that which has length and breadth only.
6. The extremities of a surface are lines.
7. A plane surface is a surface which lies evenly with the straight
lines on itself.
I have always been struck by the beauty and the depth of these
descriptions. Perhaps the most striking fact is that the first concepts
defined (except the straight line in Definition 4 and the plane in Defi-
nition 7) are basic purely topological notions, namely points, lines (we
would say curves), surfaces, and extremities (we would say boundary
operators).
The notion of point (defined as an irreducible entity) is close to
physics (points are defined similarly to atoms). The description of
straight lines is beautifully mysterious. Is it a poetic surrogate of the
idea of geodesic? Or of translational invariance along itself?
Euclid continues with a description of angles formed by curves
and by straight lines, which in the English translation is rather poetic
because of the double meaning of the word “inclination”:
8. A plane angle is the inclination to one another of two lines in
a plane which meet one another and do not lie in a straight line.
9. And when the lines containing the angle are straight, the angle
is called rectilineal.
Here let us note first that Euclid begins with the most general
case of an angle between curves (rather than straight lines). It should
also be noted that the definition of rectilineal angle in Definition 9 is
the first characterization of a basic notion by Euclid that we can

regard as a mathematical definition in the modern sense, as opposed
to the eight previous ones, which are merely intuitive descriptions.
Note also that what an angle is (say a rectilineal one), actually, is
not clear from the definition: is it a pair of straight lines, or a pair of
rays, or part of the plane bounded by them?
The next three definitions (which also concern angles) can also
be regarded as mathematical in the modern sense, provided that the
use of the common notions “equal”, “greater”, “less” are allowed.
10. When a straight line set up on a straight line makes the

adjacent angles equal to one another, each of the equal angles is right,
and the straight line standing on the other is called a perpendicular
to that on which it stands.
11. An obtuse angle is an angle greater than a right angle.
12. An acute angle is an angle less than a right angle.
Note that Definition 10 actually defines two different (although

closely related) concepts: right angles and perpendiculars.
The next two definitions are remarkable for their topological gen-
erality.
13. A boundary is that which is an extremity of anything.
14. A figure is that which is contained by any boundary or bound-

aries.
The word “extremity” previously appears in Definitions 3 and

6, and apparently was easier to grasp for Euclid’s contemporaries
that the synonymous word “boundary”, hence Euclid uses it to give
a descriptive definition of the notion of boundary in Definition 13.
Note also that it is implicit in Definition 14 that a boundary is (in
our terminology) a connected set. In that definition, the use of the
expression “contained by” rather than “contained in” is crucial; we
would say “bounded by” in this context, but that expression can
hardly be used in the formulation of Definition 14, which would then
obviously sound like a tautology.
The next four definitions concern circles. More precisely, what

Euclid calls a circle would be called a disk by the modern mathemati-
cian; today the word circle stands for the boundary of the disk.
15. The circle is a plane figure contained by one line such that
all the straight lines falling upon it from one point among those lying
within the figure are equal to one another.
16. And that point is called the center of the circle.
17. A diameter of the circle is any straight line drawn through
the center and terminated in both directions by the circumference of
the circle, and such a straight line bisects the circle.
18. A semicircle is the figure contained by the diameter and the
circumference cut off by it. The center of the semicircle is the same
as that of the circle.
Of course one should keep in mind that, as before, “straight line”
means “line segment” (in modern terminology) and for “semicircle”
we would say “half-disk”.
The next four definitions are about various polygons, including
triangles, squares, rectangles (called “oblongs” by Euclid), and other
types of quadrilaterals.
19. Rectilineal figures are those which are contained by straight
lines, trilateral figures being those contained by three, quadrilateral
those contained by four, and multilateral those contained by more than
four straight lines.
20. Of trilateral figures, an equilateral triangle is that which has
its three sides equal, an isosceles triangle that which has two of its
sides equal, and a scalene triangle that which has its three sides un-
equal.
21. Further, of trilateral figures, a right-angled triangle is that
which has a right angle, an obtuse-angled triangle that which has an
obtuse angle, and an acute-angled triangle that which has its three
angles acute.
22. Of quadrilateral figures, a square is that which is both equi-
lateral and right-angled; an oblong that which is right-angled but not
equilateral; a rhombus that which is equilateral but not right-angled;

a rhomboid that which has its opposite sides and angles equal to one
another, but is neither equilateral nor right angled. And let quadri-
laterals other than these be called trapezia.
Definitions 19–22 are organized with esthetic flair, their symmet-
ric repetition is almost poetic, and pleases even the contemporary ear.
The reader has surely noted the use of some unusual terms: “recti-
lineal figure” for “polygon”, “rhomboid” for “parallelogram” (more
precisely, a rhomboid is a generic parallelogram, i.e., one which is nei-
ther a rhombus nor a rectangle). It is interesting that all the objects
defined in 19–22 are generic: in contrast with modern elementary
geometric terminology, for Euclid the square is not a particular case
of the rectangle (oblong), the rectangle is not a particular case of the
rhombus, etc.
Euclid concludes his list of definitions with the crucial definition
of parallel lines.
23. Parallel lines are straight lines which, being in the same plane
and being produced indefinitely in both directions, do not meet each
other in any direction.
Notice that Definition 23 is a perfectly rigorous mathematical
definition, but it does not assert the existence of parallel lines.
Why has Euclid postponed this definition to the very end of his
list? Clearly, it does not use any terms appearing in Definitions 7–
22, so why didn’t he place it earlier, say right after Definition 7? I
believe the reason for that is his dislike of the Fifth Postulate, which
he tries to avoid using as long as he can. Thus the word “parallels”
first appears in the formulations of the propositions of Book I only
in Proposition 27, and the construction (existence) of parallels is not
asserted until Proposition 31.
The Propositions of Book I

Here we state the theorems (propositions) of Euclid’s treatment of
plane geometry in the order of their appearance, without the proofs
and figures and with a minimum of comments. The key point here
is the order in which the propositions are proved. The reader should
note that many of the propositions assert the possibility of performing
a certain specific geometric construction (so that they can be under-
stood as existence theorems); however, it can be argued that in many
cases, when the possibility of a construction is claimed, it is tacitly
assumed that the construction is well defined, which means that often
such propositions were understood by the Ancient Greeks as existence
and uniqueness theorems.
1. On a given straight line to construct an equilateral triangle.
Of course the construction was carried out by means of a compass
(so that Postulate 3 was used twice), and the fact the two constructed
circles intersect was considered obvious (which indeed it is).
2. To place at a given point (as an extremity) a straight line equal
to a given straight line.
3. Given two unequal straight lines, to cut off from the greater a
straight line equal to the lesser.
Of course in these three statements, and in the subsequent ones,
the words “straight line” mean “line segment” (in our terminology).
4. If two triangles have two sides equal to two sides respectively,
and have the angles contained by the equal straight lines equal, they
will have the base equal to the base, the triangle will be equal to the tri-
angle, and the remaining angles will be equal to the remaining angles
respectively, namely those which the equal sides subtend.
This “first test of congruence of triangles” is familiar to high
school students all over the world and denoted by SAS in North Amer-
ica. Note that there is no mention of congruence or of some kind of
motion or superposition in the proposition. I don’t think that Euclid,
saying “the triangle will be equal to the triangle” implied some sort
of congruence, and the end of the proposition simply explains what is
meant by equal triangles, namely the equality of all the corresponding
elements (sides and angles) of the triangles.
5. In isosceles triangles the angles at the base are equal to one
another, and, if the equal straight lines be produced further, the angles
under the base will be equal to one another.
The reader will have noted the use of the word “base” in the
last two propositions and the word “under” in the last one, which of
course means that Euclid accompanied the propositions with pictures,
in which the “bases” of triangles were drawn as horizontal lines.
6. If in a triangle two angles be equal to one another, the sides

which subtend the equal angles will also be equal to one another.
Proposition 6 is the reciprocal of Proposition 5. In many high

school geometry courses, the two statements are united in an if and
only if theorem.
7. Given two straight lines constructed on a straight line (from

its extremities) and meeting in a point, there cannot be constructed
on the same straight line (from its extremities), and on the same side
of it, two other straight lines meeting in another point and equal to
the former two respectively, namely each to that which has the same
extremity with it.
Now this is a statement leading up to the “third test of con-

gruence of triangles” (SSS), and two cases must be considered in its
proof because of possible orientation reversal. Remarkably, Euclid
first states this as a uniqueness theorem, using the expression “on
the same side of it” to get rid of the nonuniqueness. Note that the
expression in quotes first appears in another context, namely in the
statement of the Fifth Postulate.
8. If two triangles have the two sides equal to two sides respec-
tively, and also have the base equal to the base, they will also have the
angles equal which are contained by the equal straight lines.
The reader should have recognized the conditions as the assump-

tions of the “third test of congruence of triangles” (SSS) taught in
high school geometry courses. However, the assertion of this proposi-
tion is weaker (than the one in the “test”); it only claims the equality
of one angle rather than three. The appearance of the term “base”
is rather curious – it is not a rigorous mathematical term, it simply
shows that Euclid and his followers would draw one of the sides of
a triangle horizontally, and that side would be called the base of the
triangle.
9. To bisect a given rectilineal angle.

10. To bisect a given finite straight line.
11. To draw a straight line at right angles to a given straight line
from a point given on it.
The three theorems 9–11 are all proved by simple compass and
straight edge constructions.
12. To a given infinite straight line, from a given point which is
not on it, to draw a perpendicular straight line.
This key assertion (with the explicit stipulation of uniqueness
added) is often taken to be one of the axioms in 20th century ax-
iomatic expositions of plane geometry. Note the appearance of the
expression “infinite straight line”, which used here instead of “straight
line” (in the sense of line segment), because otherwise the statement
would be false.
13. If a straight line set up on a straight line make angles, it will
make either two right angles or angles equal to two right angles.
Of course, by “angles equal to two right angles”, Euclid means
angles whose geometric sum equals two right angles.
14. If with any straight line, and at a point of it, two straight
lines not lying on the same side make adjacent angles equal to two
right angles, the two straight lines will be in a straight line with one
another.
This theorem may be difficult to understand for the reader. In
modern terminology, it says that if two line segments with an extrem-
ity at the same point A of a line l form angles with the line l whose
sum is 180◦ , then the two segments lie in the same straight line.
15. If two straight lines cut one another, they make the vertical
angles equal to one another.
16. In any triangle, if one of the sides be produced, the exterior
angle is greater than either of the interior and opposite angles.
In the English translation of Euclid’s text used here, the logical
meaning of the construction “either of ... and” may not be clear, but
of course Euclid and his followers understood the statement of this

theorem correctly.
17. In any triangle two angles taken together in any manner are
less than two right angles.
Note that this theorem shows that we are not in elliptic geometry,
but it does not contradict hyperbolic geometry (Euclid has not used
the Fifth Postulate in any of the proofs so far).
18. In any triangle the greater side subtends the greater angle.
19. In any triangle the greater angle is subtended by the greater
side.
20. In any triangle two sides taken together in any manner are
greater than the remaining one.
In its modern formulation, this fundamental statement is known
as the “triangle inequality”, the key axiom in the definition of metric
space.
21. If on one of the sides of a triangle, from its extremities,
there be constructed two straight lines meeting within the triangle, the
straight lines so constructed will be less than the remaining two sides
of the triangle, but will contain a greater angle.
22. Out of three straight lines, which are equal to three given
straight lines, to construct a triangle: thus it is necessary that two of
the straight lines taken together in any manner should be greater than
the remaining one.
This sounds like an existence theorem for triangles (related to
the SSS test), but it also contains a clear formulation of the triangle
inequality.
23. From a given straight line and a point on it to construct a
right angle equal to a given right angle.
24. If two triangles have the two sides equal to two sides respec-
tively, but have the base greater than the base, they will also have the
one of the angles contained by the equal straight lines greater than the
other.
25. If two triangles have the two sides equal to two sides, respec-
tively, but have the one of the angles contained by the equal straight
lines greater than the other, they will also have the base greater than
the base.
Concerning the term “base”, see the remark after Proposition 8.
26. If two triangles have the two angles equal to two angles and
one side equal to one side, namely, either the side adjoining the equal
angles, or the side subtending one of the angles, they will also have
the remaining sides equal to the remaining sides and the remaining
angle equal to the remaining angle.
This is the “second test of congruence of triangles” (ASA) in its
full generality, and it is used in the proof of the next proposition.
27. If a straight line falling on two straight lines makes the al-
ternate angles equal to one another, the straight lines will be parallel
to one another.
It is only at this point that Euclid makes use of the Fifth Postu-
late, and it is in this proposition that the word parallels first appears
(after the definitions).
28. If a straight line falling on two straight lines make the exterior
angle equal to the interior and opposite angle on the same side, or the
interior angles on the same side equal to two right angles, the straight
lines will be parallel to one another.
29. If a straight line falling on parallel straight lines makes the
alternate angles equal to one another, the exterior angle equal to the
interior and opposite angle, and the interior angles on the same side
equal to two right angles.
30. Straight lines parallel to the same straight line are also par-
allel to one another.
31. Through a given point to draw a straight line parallel to a
given straight line.
Here Euclid resorts (after a long interlude) to a statement as-
serting that our “land surveyor” can perform a certain construction.
Note that this proposition is close to the formulation of the “axiom
of parallels” as it usually appears in high school geometry courses.

Euclid does not explicitly state that the construction is unique; how-
ever, as we explained above, the uniqueness of the construction was
usually implicit in statements of this type.
32. In any triangle, if one of the sides be produced, the exterior
angle is equal to the two interior and opposite angles, and the three
interior angles of the triangle are equal to two right angles.
This is one of the key theorems of Euclidean geometry (which
differentiates it from the elliptic and hyperbolic geometries). It is
proved by using Proposition 31 to draw a parallel to the base through
the opposite vertex and applying Proposition 29 to compare the angles
formed at that vertex with the interior angles at the base.
33. The straight lines joining equal and parallel straight lines (at
the extremities which are) in the same directions (respectively) are
themselves also equal and parallel.
34–45. These twelve propositions have to do with constructions
of parallelograms and triangles which involve parallels. When speak-
ing of the equality of triangles and parallelograms in these proposi-
tions, Euclid means equality of areas (and not what we would call
isometry or superposition by a motion). Thus the phrase (from
Proposition 41) “the parallelogram is double the triangle” means that
the area of a certain parallelogram is twice that of a certain triangle.
46. On a given straight line describe a square.
Recall that the very first proposition asserted the possibility of
constructing the “perfect triangle” (an equilateral one) and now, al-
most at the end of Book I, Euclid shows that it is possible to construct
the “perfect quadrilateral”, i.e., the square. This search for perfection
is characteristic not only of Greek mathematics (recall the Platonic
bodies), but of Greek art and culture in general.
47. In right angled triangles, the square on the side subtending
the right angle is equal to the squares on the sides containing the right
angle.
This is the famous Pythagorean theorem. Note that it is a purely
geometric statement, it does not say that a2 + b2 = c2 , it asserts that
Conclusion 269
the area of the square constructed on the hypotenuse is the sum of

areas of the two squares constructed on the other two sides. Euclid’s
proof is not the familiar one (called “Pythagoras’ pants” in some
countries and obtained by cutting the two squares constructed on the
shorter sides of the right triangle into triangles that fit together to
form the square constructed on the hypotenuse) – he cuts the square
constructed on the hypotenuse into two rectangles by the prolongation
of the altitude issuing from the right angle and shows that each of
these rectangles has the same area as the corresponding square.
48. If in a triangle the square on one of the sides be equal to the
squares on the remaining two sides of the triangle, the angle contained
by the remaining two sides of the triangle is right.
Thus Euclid concludes Book I with the reciprocal statement to
the Pythagoras theorem.
Conclusion
It is customary among mathematicians to look down on Euclid and
criticize the lack of rigor of his development of geometry. To my mind,
this attitude only demonstrates the narrow-mindedness of such critics,
their absolutization of what was regarded as rigor in the 19th century.
Here I have tried to show that Euclid “Elements” have their own in-
ternal logic, and, unlike modern axiomatic theories, possess a striking
beauty which makes Euclid’s book one of the greatest achievements
of human culture.
Appendix B
Hilbert’s Axioms for

Plane Geometry
In this appendix, we present Hilbert’s axioms for plane geometry,

which he first developed in a series of lectures at the University of
Göttingen in 1898–1899. Following Hilbert (see the English transla-
tion of his celebrated Gründlagen in [9]), we then prove the consis-
tency of his theory (i.e., we show that there are no contradictions in
it, provided that there are none in the theory of algebraic numbers).
Hilbert’s axioms constitute the first rigorous (in our present un-
derstanding of the word) treatment of plane geometry in axiomatic
form. When Hilbert’s axiomatics was published, it was common
knowledge that a rigorous construction of plane geometry is possible
within the framework of the theory of real numbers by using Carte-
sian coordinates, thus transforming plane geometry into a particular
case of linear algebra over R. Moreover, by then Hermann Weyl
showed that, in the same framework, one could develop geometry in
coordinate-free form. Nevertheless, Hilbert’s work was a fundamen-
tal breakthrough in the understanding of geometry, and remains an
important milestone in the history of mathematics.
The main distinctive feature of Hilbert’s approach is that he un-
derstood and implemented the idea that in a rigorous exposition of
a mathematical theory some basic concepts and relations must be left
271
272 B. Hilbert’s Axioms for Plane Geometry
undefined, for otherwise a logical vicious circle necessarily appears in

the definitions. In his exposition, the undefined notions are “point”,
“straight line”, “plane”, “belongs to”, “lies between”, and “is con-
gruent to”. Thus the axioms give, in a sense, an implicit definition of
these undefined notions.
In his book [9], Hilbert not only lists the axioms, but also indi-
cates the basic facts of plane geometry in the order in which they
can be derived from the axioms. Also, in Hilbert’s exposition, the
axioms related to space geometry are not separated from those of
plane geometry; so in our exposition we simply omit the axioms (or
the parts of the axioms) dealing with space geometry (and modify
the numbering of the axioms accordingly).
In this appendix, as in the previous one, we look at and comment
on the axiomatics of plane geometry from the point of view of a his-
torian. As explained in the Preface, this is due to one of the author’s
biases: I believe that Euclidean plane geometry should not be taught
by means of axiom systems such as Euclid’s or Hilbert’s, its object
is a mathematical entity whose main protagonist is the real line, and
no rigorous study of plane geometry is possible unless we know what
the real numbers are. For that reason, although we reproduce the
wording of the axioms verbatim (in their English translation appear-
ing in the book [9]), we replace Hilbert’s mathematical commentary
concerning the axioms and their consequences by some remarks of
historical nature.
I. Axioms of connection
This first group of axioms establishes a relationship between the un-
defined concepts point and straight line expressed by means of the
(also undefined!) relation belongs to. (For the sake of brevity, in
what follows we often write “line” instead of “straight line”.) It is
very important to understand that in Hilbert’s exposition a line is a
line is a line (an undefined concept) – it is not a set of points as we
have been taught!
As to the relation “belongs to”, it has many synonymic versions:
instead of saying “the point A belongs to the straight line l” we can
say “l passes through A”, “l contains A”, “A is a point of l”, etc. If A
I. Axioms of connection 273
belongs to a line l and also belongs to another line m, we say that “A

is the common point of lines l and m”, or “the lines l and m intersect
at the point A”, etc. If two points A and B belong to the line l, we
can also say that “l joins A and B”, “A and B determine l”, etc.
I, 1. Two distinct points A and B always completely determine

a straight line a. We write AB = a or BA = a.
Hilbert explains what “A and B determine a” means in his com-

mentary, but what is meant by “completely determined” is not ex-
plained. Should it be understood as stipulating the uniqueness of the
line determined by A and B? Apparently not, because the next axiom
is a uniqueness axiom of sorts. If one wishes to be very formalistic,
it should be noted that at this stage we don’t know that two distinct
points actually exist. But it is reasonable to suppose that the non-
emptiness of the set of points and of the set of lines is tacitly assumed
by Hilbert.
I, 2. Any two distinct points of a straight line completely

determine that straight line; that is, if AB = a and AC = a, where
B
= C, then also BC = a.
Note that at this stage we do not know that there are three
distinct points A, B, C on any line, in fact, we do not even know
that there are two. This assumption appears later, in the seventh of
Hilbert’s axioms of connection. We reproduce only the part of the
latter axiom that concerns plane geometry and change its number
from seven to three. (Hilbert’s axioms of connection 3, 4, 5, 6 have
no bearing on plane geometry, they are about planes in space.)
I, 3. Upon any straight line there exist at least two points, and
there are three points not belonging to any straight line.
One of the main consequences of these axioms is that two distinct

lines have no more than one common point.
The axioms of connection are satisfied by the finite geometry
consisting of three “straight lines” each consisting of two points and
such that any two of the lines have one point in common.
II. Axioms of order

This group of axioms involves, besides the previously mentioned un-
defined notions (point, line) and the relation “belongs to”, a new
undefined relation lies between. Hilbert explains in his commentary
that this relation is a relation of order and even presents a picture
of point B lying between points A and C on a straight line as an
illustration for the first axiom of order, which reads:
II, 1. If A, B, C are points of a straight line and B lies between

A and C, then B lies also between C and A.
II, 2. If A and C are two points of a straight line, then there

exists at least one point B between A and C and at least one point
D so situated that C lies between D and B.
Note that the second axiom of order implies that there is an

infinite (at least countable) number of points on each line.
II, 3. Of any three points of a straight line, there is always one

and only one which lies between the other two.
II, 4. Any four points A, B, C, D of a straight line can always

be so arranged that B shall lie between A and C and also between A
and D, and, furthermore, that C shall lie between A and D and also
between B and D.
The rather awkward formulation of the fourth axiom (it doesn’t

sound any better in German) using the strange expression “so ar-
ranged” really means that any four points on a line may be denoted
by A, B, C, D in such a way that B lie between A and C, etc.
The first four axioms of order allow Hilbert to define the notion
of segment AB as the set of points lying between A and B; the points
A and B themselves are called the extremities of the segment AB.
Note that the extremities do not belong to the segment, so that we
would call AB an open interval.
The notion of order also allows us to define in the natural way
what Hilbert calls a half-ray (we would use the term ray instead): a
half-ray originating from the point A and passing through the point
III. Axiom of parallels 275
B different from A is the set of all points lying on the line AB on the
same side of A as B.
II, 5. Let A, B, C be three points not lying in the same straight
line and let a be a straight line not passing through any of the points
A, B, C. Then, if the straight line a passes through a point of the
segment BC, then it will also pass through either a point of the
segment AC or of the segment AB.
This statement is known as Pasch’s axiom. As worded by Hilbert,
it is incorrect, because it can happen that the line a passes through
the point A, which is neither a point of the “segment” AC nor of the
“segment” AB.
The axioms of order allow Hilbert to define the notions of half-
plane, although he never uses such a set-theoretical term, referring
to points lying on the same side (or on different sides) of a line in
the plane or on the same side (or on different sides) of a point on a
line. The reader will recall that in Euclid’s Elements the expression
“on the same side” was never defined and was apparently regarded
as obvious, in particular, in the Fifth Postulate.
Further, Hilbert defines the notions of broken line and of poly-
gon, and as particular cases of the latter, triangles, quadrangles, pen-
tagons,. . . , n-gons. He then states the Jordan Curve Theorem for
polygons, and claims that it can be obtained “without serious diffi-
culty.”
III. Axiom of parallels

Hilbert’s version of Euclid’s Fifth Postulate reads:
III, 1. In the plane there can be drawn through any point A
lying outside of a straight line a one and only one straight line which
does not intersect the line a. This straight line is called the parallel
to the line a through the given point A.
It is interesting to note that in this axiom Hilbert resorts to tradi-
tional terminology when he writes “there can be drawn” rather than
something more formal like “there exists”. Also note that for Hilbert
the relation of parallelism is not an equivalence relation (it is not
reflexive) and its definition is included in the formulation of the ax-

iom. Unlike Euclid, Hilbert does not separate the definitions and the
axioms, the definition of parallelism appearing in Axiom III, 1.
IV. Axioms of congruence

In this group of axioms, another undefined concept appears, that of
congruence. In modern expositions of plane geometry, it is usually
explicitly defined: two figures are congruent if there exists an isome-
try (i.e., a distance-preserving transformation of the plane onto itself)
that takes one of the figures to the other. This approach is of course
unacceptable to Hilbert, because it is based on the notion of distance,
which never appears in Hilbert’s axiomatization. In Hilbert’s geome-
try, as well as in Euclid’s, there is no fixed unit of measure.
IV, 1. If A and B are two points on a straight line a, and if
A is a point upon the same or another straight line a , then, upon a
given side of A on the straight line a , we can always find one and
only one point B so that the segment AB (or BA) is congruent to
the segment A B . We indicate this relation by writing
AB ≡ A B .
Every segment is congruent to itself; that is, we always have
AB ≡ AB.
This axiom may be described briefly as saying that any segment

can be laid off upon a given side of a given straight line in one and
only one way. Note that the straight line a is not really needed in
the statement of the axiom. Note also the expression “we can always
find”, used here instead of the more formal “there exists”, and the fact
that congruence (at least for segments) is a reflexive relation (which is
explicitly specified in the axiom); however, it is not specified that this
relation is symmetric and transitive. The symmetry of the congruence
relation will be proved later. As to transitivity, it is explicitly required
in the next axiom.
IV, 2. If a segment AB is congruent to the segment A B and
also to the segment A B , then the segment A B is congruent to
IV. Axioms of congruence 277
the segment A B ; that is, if AB ≡ A B and AB ≡ A B , then

A B ≡ A B .
The reader will readily show that Axiom IV, 3 (together with
IV, 1) implies the reflexivity of the congruence relation.
The next axiom says that adding congruent segments end to end
one gets congruent segments. It reads:
IV, 3. Let AB and BC be two segments of a straight line a which
have no points in common aside from the point B, and, furthermore,
let A B and B C be two segments of the same or another straight
line a having, likewise, no point other than B in common. Then, if
AB ≡ A B and BC ≡ B C , we have AC ≡ A C .
The first three axioms of congruence allow us to give several rig-
orous definitions related to the notion of angle. That of angle as two
half-rays (called the sides of the angle) drawn from the same point
(called the vertex of the angle) and lying on two different straight
lines, and of the interior of an angle as the set all points of the plane
not lying on the sides of the angle and such that any two points of
the set can be joined by a segment not intersecting the sides.
In the next axiom the (undefined) congruence relation is applied
to angles; the axiom says that one can “lay off” an angle congruent to
a given one upon a given side of a half-ray. In Hilbert’s formulation:
IV, 4. Let an angle (h, k) and a line a be given in the plane.
Suppose also that a definite side of the line a is given. Denote by h
a half-ray of the line a and emanating from a point O on that line.
Then there is one and only one half-ray k such that the angle (h, k),
or (k, h), is congruent to the angle (h , k ) and at the same time all
interior points of the angle (h , k ) lie upon the given side of a. We
express this relation by means of the notation
∠(h, k) ≡ ∠(h , k ),
Every angle is congruent to itself; that is,
∠(h, k) ≡ ∠(h, k)
or
∠(h, k) ≡ ∠(k, h).
Note that Hilbert takes great pains to explicitly state that the
equality of angles is reflexive and stipulates that an angle does not
depend (up to congruence) on the order in which its sides are indi-
cated. Just as for segments, the symmetry of the congruence relation
for angles is not explicitly stipulated, but is proved later.
The final axiom of this group has to do with two triangles; it is
similar to a theorem known as the “first test of congruence of trian-
gles” in the high school geometry textbooks of many countries, but
its conclusion does not claim that the two triangles are congruent, for
the excellent reason that the notion of congruence of triangles is an
undefined one, and that its definition will be given later.
IV, 5. If, in two triangles ABC and A B C , the congruences
AB ≡ A B , AC ≡ A C , ∠BAC ≡ ∠B A C
hold, then the congruences
∠ABC ≡ ∠A B C and ∠ACB ≡ ∠A C B
also hold.
The reader will have noticed that this is the first triangle congru-
ence test (SAS).
Once all the axioms of congruence have been stated, Hilbert gives
a few more definitions and some important consequences of all the ax-
ioms, but at first without using the axiom of parallels. The definitions
include the notions of supplementary and vertical angles and of a right
angle (the latter is defined as an angle congruent to its supplemen-
tary angle). This is followed by an explicit definition of congruent
triangles as those having all their sides and all corresponding angles
congruent.
After this Hilbert proves the First, Second, and Third Theorem
of Congruence of Triangles. This is followed by two theorems on the
congruence of angles (in particular, the theorem asserting that if two
angles are congruent, then so are their supplementary angles). These
theorems are used to prove that any two right angles are congruent.
At this point Hilbert points out that Euclid held this last fact as one
of the postulates, “although it seems to me wrongly”, he comments.
V. Axiom of continuity 279
Then, using the word “figure” to mean a finite set of points, he

states what he calls “the most general theorem relating to congru-
ence”. It asserts that if (A, B, C, . . . ) and (A , B , C , . . . ) are congru-
ent figures and P is any point, then it is always possible to find a point
P such that the figures (A, B, C, . . . , P ) and (A , B , C , . . . , P ) are
congruent; furthermore, if the two figures have three points not lying
in a straight line, then the point P is unique.
At this point Hilbert finally makes use of his Axiom of Paral-
lels and states the theorem on the congruence of the appropriate
(alternate-interior and exterior-interior) angles obtained by cutting
two parallel lines by a third line. It is interesting to note that Hilbert
ignores considering the logical possibility of the third line cutting one
of the two parallels but not the other. The fact that this possibility
cannot actually occur is an obvious consequence of Hilbert’s Axiom
of Parallels, and I suppose Hilbert felt that this would be obvious to
any reader.
The last theorem that Hilbert states after the axioms of the first
four groups reads: The sum of angles of a triangle is two right angles.
He does not explain what is meant by the “sum” of angles, but of
course the sum is understood in the geometric sense: the theorem
does not mean that the sum (of measures) of the three angles is 180◦ ,
it means that if the angles are laid off from some straight line in
succession, then the second half-ray of the third angle will lie on the
straight line from which we began laying off the angles of the given
triangle.
It is of course no accident that Hilbert stops at that point: un-
doubtly, he understands the crucial role of this theorem in the context
of Euclidean and non-Euclidean geometries.
V. Axiom of continuity
This is the last axiom in Hilbert’s axiomatics. It is commonly known
as the axiom of Archimedes and ordinarily appears in the axiomatic
definition of the real numbers. In its geometric form, it says that by
laying off any segment along a line a sufficient number of times we
will eventually reach any given point on the line. Hilbert words it as
follows.
V. Let A1 be any point upon a straight line between the arbi-

trarily chosen points A and B. Take the points A2 , A3 , A4 , . . . so that
A1 lies between A and A2 , A2 between A1 and A3 , A3 between A2
and A4 , etc. Moreover, let the segments
AA1 , A1 A2 , A2 A3 , A3 A4 , . . .
be equal to one another. Then, among this series of points, there
always exists a certain point An such that B lies between A and An .
The reader may have noticed the word “equal” in the formulation
of the axiom. This is a significant slip of the tongue – Hilbert means
“congruent” here, since he is dealing with segments, and not lengths
of segments (real numbers).
Although in Hilbert’s first publication of his axiomatics the above
axiom is the last one, in the French translation of his book he added
one more axiom, which he called the axiom of completeness. This
axiom is, as Hilbert says, “not of purely geometric nature” and its
raison d’être is to ensure that the points of any straight line be in one-
to-one correspondence with the real numbers, thus making Hilbert’s
axiomatics categorical. We postpone our discussion of these questions
to the end of the next subsection.
Consistency of Hilbert’s axioms

Hilbert establishes the consistency of his axiomatics (i.e., shows that
his axioms are noncontradictory) by constructing a model of his plane
geometry.
To do this Hilbert denotes by Ω the set of algebraic numbers ob-
tained from √ the number 1 by the four arithmetical operations and the
operation 1 + ω 2 , where ω is one of the previously defined numbers.
The set Ω consists of real numbers and is obviously countably infinite.
In the model, a pair (x, y), where x, y ∈ Ω, is called a point, a triple
(u : v : w), where u, v, w ∈ Ω and u2 + v 2
= 0, is called a straight line,
and we say that the point (x, y) belongs to the line (u : v : w) if
ux + vy + w = 0.
Then Hilbert notes that Axioms I and III are obviously fulfilled.
Further, using the usual order relation (x < v) for the numbers in Ω,
Conclusion 281
he defines the relation lies between for three points on a line in the
natural way. It is easy to verify that then the axioms of group II also
hold.
Defining the notion of congruence as it is usually done in analytic
geometry, Hilbert points out that the axioms of group IV (including
those about laying off segments and angles) are fulfilled as well. This
leaves the only axiom of group V (that of Archimedes), which of
course holds for the algebraic numbers Ω ⊂ R.
Thus Hilbert has constructed a model of Euclidean plane geome-
try in the arithmetic related to the algebraic numbers; in this model,
the undefined terms “point”, “straight line”, “belong to”, “lie be-
tween”, “be congruent” acquire a concrete interpretation so that all
the axioms of Euclidean plane geometry are fulfilled.
This means that any contradiction resulting from Hilbert’s system
of axioms must also appear in the arithmetic related to Ω. Thus we
can say that Hilbert’s axiom system is consistent provided that there
are no contradictions in the theory of algebraic numbers.
Conclusion
It should be noted, however, that in Hilbert’s model of plane geometry
the number of points, as well as the number of lines, is countable. Of
course it is possible to construct a noncountable model of this geom-
etry, with the set of points having the cardinality of the continuum,
simply as it is done in analytic geometry courses based on Cartesian
coordinates. Therefore, Hilbert’s axiomatics is, as logicians say, non-
categorical, which means that the axiomatics can have nonisomorphic
models.
Regarding this as a drawback, Hilbert added, in later editions of
his work, a Completeness Axiom which asserts, roughly speaking, that
no extra points may be added to any line without contradicting the
other axioms. This axiom, together with the axiom of Archimedes,
readily implies that there is an order-preserving correspondence be-
tween the set of points belonging to a straight line and the real num-
bers.
Why didn’t Hilbert make things much easier for himself and his
readers by coming right out and stating this simple and fundamental
fact as an axiom? For one thing, he undoubtedly wanted his geometry
to be formally independent of the theory of real numbers, being a
kind of “pure geometry”. However, there is no way of going around
the fact that any rigorous treatment of what we regard as Euclidean
plane geometry will result in implicitly constructing the theory of real
numbers as the set of all points on a line, addition being defined by
putting segments end-to-end, order by the betweenness relation, and
multiplication via homothety. I tend to believe that Hilbert purposely
went to great pains to hide the fundamental role of the field R in his
constructions, because he was aware that Euclidean geometry could
be rigorously constructed in a much simpler way by using R from the
very beginning, rather than as an afterthought.
This is why I don’t believe that geometry should be taught on
the basis of Hilbert’s axioms or some other improvement of Euclid’s
approach. But this point of view in no way denigrates the historical
importance of Hilbert’s work on the foundations of geometry. Not
only did Hilbert show that the axiomatic approach due to Euclid and
his contemporaries could be improved to suit the criteria of rigor of
the 20th century mathematics, but he was the first to put in practice
the fundamental idea of axiomatic mathematics as the science that
studies. . . undefined objects!
Answers & Hints
Here the reader should not expect to find solutions to the exercises,
only answers (without any comments) to those where it is required to
compute or find something, and hints for those (the majority) that
begin with the words “Prove that. . . ”. The hints are never detailed
and are written in a very informal style (not to be imitated by the
student in his/her written homework assignments, if such are required
by the instructor!); they usually only indicate the strategy of the
proof, or some of its ingredients, or vaguely indicate what sort of
proof can be obtained. In the latter case we write “Not very helpful
Hint ”. Note also that practically all the exercises in the first nine
chapters are supplied with answers or hints, whereas starting from
Chapter 10, most of the exercises are not. This is because the author
feels that by then, when the student has worked through half the
book, the time has come to throw him/her into the pool to see if
he/she is able to swim unassisted.
Chapter 1
1.1. There are three rotations (of orders 1, 3, 3) and three re-
flections in the altitudes of the triangle (all of order 2), and four
nontrivial subgroups (three groups of order 2 and one of order 3); the
group of motions has 3 elements.
283
284 Answers & Hints
1.2. (a) The symmetry group of such a pyramid is isomorphic

to that of the square. It has 8 elements (five of order 2, two of order
4, and one of order 1), 6 nontrivial subgroups (five of order 2, one of
order 4); there are 4 elements in the group of motions.
(b) The group is isomorphic to the permutation group S4 (see

Chapter 2). There are 24 elements of orders 1, 2, 3, of which 12
are rotations, 12 are reflections w.r.t. planes (of two different types),
there are 19 nontrivial subgroups (fifteen of order 2 and four of order
3), and the motion group has 12 elements.
(c) There are 48 elements of orders 1, 2, 3, 4, of which 24 are

rotations (see 1.2.3) and 24 are orientation-reversing transformations;
the motion group has 24 elements.
(d)* There are 120 elements of orders 1, 2, 3, 4, 5, of which 60

are rotations and 60 are orientation-reversing transformations; the
motion group has 60 elements.
(e)* The symmetry group of the icosahedron is isomorphic to that

of the dodecahedron, and so the answers are the same as for (d).
(f) There are n rotations, n reflections (of two different types

if n is even); the nontrivial subgroups are the rotation subgroup of
order n (which has nontrivial subgroups whose orders divide n), the
subgroups isomorphic to the dihedral group Dm (see Chapter 3) for
any m that divides n, and n reflection subgroups (of order 2); the
group of motions has n elements.
1.3. It suffices to place the square onto one of the faces of the
cube (respectively, the circle on the equator of the sphere) and check
that the different motions of that face (resp., of the equator) can be
extended to different motions of the cube (resp., of the sphere).
1.4. Whenever n divides m.
1.5. There are 7 such subgroups, of which 4 are rotations.
1.6. The four main diagonals.

Chapter 2 285
1.7. (a) One reflection in the plane passing through an edge and
the midpoint of the opposite edge and one rotation about the line
joining the midpoints of two opposite edges.
(b) A reflection in a plane passing through the parallel diagonals
of opposite faces and the composition of another such reflection with
the central symmetry w.r.t. the center of the cube.
1.8. (a, b, c) For example, any tetrahedron whose vertices are: a

vertex A of the polyhedron, the midpoint of an edge AB, the center
of a face containing AB, and the center of the polyhedron itself.
1.9. From the unit sphere in R3 , cut off two small symmetric

caps by planes parallel to the coordinate plane Oxy; then the set of
lines passing through the origin and through what is left of the sphere
is the Möbius strip.
1.10. The axis of rotation is the intersection of the planes, while

the angle of rotation is twice the angle between the planes.
1.11. It is the rotation whose axis of rotation is the intersection

of the two planes passing through each of the two given axes of rota-
tion and making angles equal to half the angle of the corresponding
rotation with the plane containing both axes of rotation.
Chapter 2
2.1. Z2 (symmetries of the unit interval), Z3 (rotations of the

equilateral triangle), Z4 (rotations of the square), the Klein group
Z2 ⊕ Z2 (symmetries of the rectangle), Z5 (rotations of the regular
pentagon), S3 (symmetries of the equilateral triangle), Z6 (rotations
of the regular hexagon).
2.2. (a) The only normal subgroup of Sym() is its rotation
subgroup, and the corresponding quotient group is isomorphic to Z2 .
(b) The subgroup of all the motions and the 4-element subgroup
(isomorphic to the Klein group) consisting of the identity and the
three rotations by π about the three lines joining the midpoints of
two opposite edges.
286 Answers & Hints
2.3. Not very helpful Hint. The proof is a straightforward verifi-

cation of definitions.
2.4. Hint. If H is a subgroup of G, we must prove that for
all g ∈ G we have g −1 Hg ⊂ H; take any h ∈ H, consider two cases
(hg ∈ H and hg ∈ / H), and prove the required inclusion – it is obvious
in the first case and follows in the second case because there are only
two cosets mod H.
2.5. The orbits are the cycles
(1 → 5 → 4 → 9 → 1), (2 → 8 → 2), (3), (6 → 10 → 7 → 6),
while the stabilizers of elements of the first orbit are {1, a4 , a8 }, of the
second one, {1, a2 , . . . , }, of the third, the whole group, of the fourth,
{1, a3 , a6 , a9 }, where a is the given permutation.
2.6. (a) 6; (b) 60.
2.7. 16.
2.8. Hint. First prove that Sn is generated by permutations σi of
successive elements (i and i+1) and then that any σi can be obtained
by conjugating (1 2) by an appropriate power of (1, 2, . . . , n).
2.9. s1 , s2 : s1 s2 s1 = s2 s1 s2 , s21 = 1, s22 = 1; another possibility
is p, q : p2 = 1, q 3 = 1, (pq)2 = 1; there are several more.
2.10. 36, including 18 epimorphisms.
2.11. An isomorphism between the given group and Dn may be
constructed by assigning a to one of the rotations by π and assigning
b to the rotation by 2π/n.
2.12. Not very useful Hint. There are many computations show-
ing that a = 1: one should start with the given formula b−1 ab = a2
and appropriately use the trivial relations and the relations a5 = b3 = 1.
Chapter 3
3.1. The symmetry group is the dihedral group D6 , the motion

group is Z6 , both are finite subgroups of O(3) in accordance with
Corollary 3.2.8.
Chapter 3 287
6 and Z6 ;
3.2. (a) D
(b) D5 and Z5 ;
6 and Z6 .
(c) D
3.3. Hint. Try to understand the meaning of the two terms on

the right-hand side in terms of the number of points in the orbits and
the number of elements in the stabilizers in the case of the motion
group of the cube. The meaning is the same in the general case and
you will see that the equation is practically a tautology – it expresses
two ways to compute the number of elements in F .
3.4. Yes, any group of isometries leaving in place an inscribed

regular tetrahedron.
3.5. No, because 60 is not divisible by 24.
3.6. There are 6 subgroups isomorphic to Z2 , 4 to Z3 , and 3 to

D4 .
3.7. Let M1 N1 and V2 N2 be the sides of the pentagons placed

on top of the cube (see Figure 3.6) and let Π be the vertical plane
parallel to M1 N1 and V2 N2 that cuts the cube into two congruent
halves. Now rotate the two pentagons around the sides AD and BC
until M1 N1 and M2 N2 meet Π. From considerations of symmetry it is
easy to see that the sides M1 N1 and V2 N2 will merge: M1 = M2 =: M
and N1 = N2 =: N . Denote by P the midpoint of M N , by S the
midpoint of BC, by R the midpoint of CD. Extend the line M R by
the segment RQ of length equal to |P S| and let K be the projection
of Q on the front face of the cube. Let H be the projection of P on
the top face of the cube. It now suffices to prove that the triangles
RKQ and SHP are congruent, but this is easy.
3.8. A straightforward verification shows that G+ does act on

F . There are three orbits (one for each of the three different types
of points of F , which correspond to faces, edges, and vertices), the
stabilizers of the points corresponding to the faces, edges, and ver-
tices consist of 3, 1, and 2 elements, respectively; these elements are
rotations by angles that are integer multiples of π/2, π, and 2π/3,
288 Answers & Hints
respectively. The vertices of the cube are those elements of F whose

stabilizers have exactly two elements.
3.9. This exercise is similar to Problem 3.8, but somewhat sim-

pler.
3.10. This exercise is similar to Problem 3.8, but a bit more

complicated.
3.11. Hint. The proof is similar to that sketched in Problem

3.8, except that for the vertices of the octahedron one chooses those
elements of F whose stabilizers have exactly three elements.
Chapter 4
4.1. Hint. Let A B be the image of the segment AB under the

given motion. If the lines AB and A B are parallel, then it is easy to
−−→
show that the motion is a translation by the vector AA . Otherwise
it is a rotation whose center can be easily constructed.
4.2. Hint. Let A B be the image of AB under the given isometry.

By parallel translation, move A B to A B , where A ≡ A. The glide
symmetry line will then be a line that passes through the midpoint
of AA and is parallel to the bisector of angle BAB .
4.3. Not very helpful Hint. The validity of the construction ba-
sically follows from the fact that the sum of the angles of a Euclidean
triangle is equal to π. When ϕ = ψ, one obviously obtains the parallel
translation by the vector joining the the first center of rotation to the
second.
4.4. Hint. The center is obtained from the center of the given
rotation by shifting it by minus the given translation vector.
4.5. Not very helpful Hint. The composition is obviously the

rotation indicated in the statement of the exercise.
4.6. (a) Two parallel translations.

(b) Two parallel translations and one rotation by the angle π.
Chapter 4 289
(c) Two perpendicular parallel translations by two squares and

one rotation by π.
(d) Two rotations by 2π/3 and two parallel translations.
(e) Two rotations by 2π/3 and two parallel translations.
(f) Two reflections and two translations (by the doubled sides of
the rectangles).
4.7. No.
4.8. There are two types: first, the vertices of the squares near
which two dots of the question marks are located (the corresponding
subgroups are rotations by π), and, second, those in the vicinity of
which there are no dots (the corresponding subgroups are rotations
by π/2).
4.9. For example, the group preserving the square lattice can be
presented as follows: h, v, r : hvh−1 v −1 = 1, rvr −1 v −1=1 , r 4 = 1.
4.10. Cubes only.
4.11. The picture on the left corresponds to the discrete geome-

try shown in Figure 4.5(a), the one on the right, to the middle picture
in the second row of Figure 4.6.
4.12. The first from the left in the second row of Figure 4.6.
4.13. The one pictured in Figure 4.5(c) and the middle one in
the last row of Figure 4.6.
4.14. The one pictured in Figure 4.5(c) and the middle ones in
the first and third rows of Figure 4.6.
4.15. The ones pictured in Figure 4.5(b), (f), the ones to the
right in the first and second rows of Figure 4.6, and the one to the
left in the third row.
4.16. Rotate all the question marks in (c) so that they lie hori-
zontally with the dot to the right.
290 Answers & Hints
Chapter 5
5.1. When both angles are rational multiples of π. In that

case, let α = kπ/p and β = lπ/q, where k and p, as well as l
and q, are coprime. Then for the fundamental domain one may
take any dihedral angle containing the z-axis whose measure equals
(gcd(k, l)/lcm(p, q))·π.
5.2. (a) If and only if the triangle is one of the three Coxeter
triangles.
(b) The triangle itself will be a fundamental domain.
5.3. Yes, it does determine a Coxeter geometry, namely the one

with fundamental domain, the equilateral triangle.
5.4. For example, in the case of the equilateral triangle and of

the vertex s1 ∩ s2 , they are the two reflections s1 and s2 and the two
rotations by 2π/3 and 4π/3, i.e., s1 ◦ s2 and (s1 ◦ s2 )2 .
5.6. Hint. This fact can be proved by inspection of the seven

Coxeter polyhedra or directly, by using the definition of the Coxeter
dihedral angles.
5.7. Not very helpful Hint. (a) Obvious. (b) Follows from (a).
5.8. For example, B 5 is an irregular polytope with 5 faces; one

face forms dihedral angles of π/4 with two others, one of these two
forms dihedral angles of π/3 with the remaining two faces, which are
orthogonal to each other.
5.9. (a) No. (b) Yes. (c) No. (d) No.
Chapter 6
6.1. Not very helpful Hint. The proof is an exercise in space

geometry of the same type as the proof of the spherical sine theorem.
6.2. Not very helpful Hint. The proof is an exercise in space

geometry of the same type as the proof of the spherical sine theorem.
Chapter 7 291
6.3. Hint. If ABC is the given triangle, consider the tetrahedron

OABC, where O is the center of the sphere and make use of the fact
that the three points A, B, C all lie in one half-sphere.
6.4. It doesn’t. The analog immediately follows from the corre-

sponding cosine theorems.
6.5. The geodesic between Moscow and New York intersects

Greenland but not Spain.
6.6. π and 3π.
6.7. Hint. Use the spherical analog of the Pythagorean theorem.
6.8. Hint. Use r to compute the Euclidean radius of the circle

and then plug it into the classical formula for the area of a spherical
cap.
6.9. Hint. For example, for the cube, a fundamental domain is

given by the pyramid OIAH, where O is the center of symmetry of
the cube, I is the center of the bottom square ABCD, and H is the
foot of the perpendicular drawn from I to AB. A total of 48 copies
of the fundamental domain fill the cube – this can be established
without counting the number of pyramids that actually yield such a
filling.
6.10. Hint. Construct the inscribed and circumscribed circles to

triangle ABC regarded as a Euclidean triangle in the plane ABC and
then project them on the sphere from its center O.
6.11. Hint. Read the hint to the previous exercise.
Chapter 7
7.1. Hint. One can compute the equation of the image of the
given circle regarded as lying in C by plugging in z = 1/z into its
equation. There are also purely geometric solutions, mostly using the
congruence of right triangles.
292 Answers & Hints
7.2. Hint. Let M be an arbitrary point of the given circle C

orthogonal to the circle of inversion CO of radius 1 centered at O.
Let S be the intersection point of the perpendicular to OM with CO
and N the intersection of the line OM with C. Then the three right
triangles OSM , SN M , and ON S are similar, and one readily shows
that |OM | · |ON | = 1.
7.3. Hint. The proof uses the similitude of right triangles in the
spirit of the previous exercise.
spirit of the two previous exercises.
spirit of the three previous exercises.
7.6. Hint. Let P be a point inside the given circle C, and M , N

the intersection points of line OP with C. Then |OM | · |ON | = 1 and
this fact easily yields the required bijectivity.
7.7. Hint. Consider any hyperbolic isometry α that takes the

(Euclidean!) center of the given circle C to the center O of the disk
model. Then α(C) is both a hyperbolic and a Euclidean circle and
therefore so is α−1 (α(C)). No, in general the centers of the two types
do not coincide (they do only if the circle is centered at O).
7.8. Hint. Here H2 is tiled both by regular heptagons and equilat-

eral triangles, each heptagon containing 7 triangles. Those triangles
determine a Coxeter geometry. The triangles have angles of 2π/7, the
angles at the vertices of the heptagons are of 4π/7.
7.9. Hint. This can be checked by a direct computation for the

inversion z → p/z, p > 0, and then for an arbitrary inversion by
conjugating it with the parallel shift taking the center of inversion to
the point (0, 0) ∈ C.
7.10. Hint. This is related to the cross ratio of four points and
is not easy. For the answer, see the next chapter.
7.11. Hint. One of the numerous ways of taking the flag (A, l, Π)
to the flag (A , l , Π ) is, first, to use a (hyperbolic) isometry α that
Chapter 7 293
takes A to A , then an isometry β that takes A to the center O of

the disk; then β(α(l)) and β(l ) will be diameters of the disk and a
symmetry γ w.r.t. a bisector of the angle formed by these two diam-
eters will interchange them: now the map φ := β −1 ◦ γ ◦ β ◦ α takes
A to A and l to l ; if φ(Π) coincides with the “half-plane” Π , we are
done, otherwise the composition of φ followed by the symmetry w.r.t.
l is the required isometry.
7.12. Hint. Consider a small regular pentagon with center of

symmetry at the center of the disk H2 . As the size of the pentagon
increases, its angles decrease and, by continuity, at some moment will
equal 2π/5. That pentagon will be the fundamental tile. The other
tiles are successively added so as to obtain 5 tiles at each vertex.
7.13. Not very helpful Hint. This theory is similar to, and not
much more complicated than, the two-dimensional theory of inversion
described at the beginning of Chapter 7.
7.14. Hint. Use the fact that any circle is the intersection of two
spheres and any straight line is the intersection of two planes.
7.15. Hint. Use the appropriate facts from the three previous
exercises and a construction similar to the one in Problem 7.3.
7.16. Hint. In this model, planes are the intersections of spheres

orthogonal to the absolute (i.e., the boundary sphere of the ball) with
the ball, the lines are circles orthogonal to the absolute, the isometries
are compositions of reflections with respect to the hyperbolic planes.
7.17. Not very helpful Hint. The problem easily reduces to

a mildly difficult problem in elementary Euclidean plane geometry
about orthogonal circles.
7.18. Hint. This is also a problem in the geometry of circles

with many possible solutions. One of them consists of constructing
the common perpendicular to the given two lines, taking its midpoint
N and drawing a third parallel n through N , then transforming n by
an isometry α to a line n0 passing through A∞ and M , and finally
using the symmetry of the whole picture w.r.t. n0 and using α−1 to
complete the construction.
294 Answers & Hints
7.19. Hint. Your first guess about the answer will probably be
correct!
Chapter 8
8.1. Not very helpful Hint. Both (a) and (b) can be proved by
direct calculations.
8.2. Hint. Let K be the intersection of the perpendicular to AP

through its midpoint M with the line l. One of the required circles
(namely, the one passing through A) is centered at K and |KP | is its
radius.
8.3. Hint. By using the transformation Ω−1 (see 8.1.6) one can
pass from the given situation to the Poincaré disk model, where this
problem reduces to constructing the two parallels through a given
point to a given line, and then return our situation via Ω. Of course
there is also a direct proof in the spirit of the previous exercise.
8.4. Hint. First check that the given transformation takes the
disk H2 to itself, then verify that the given formula holds whenever it
corresponds to an even number of reflections w.r.t. circles orthogonal
to the absolute.
8.5. Hint. The hint to the solution of 7.11 applies here as well.
8.6. π − α − β − γ, where α, β, γ are the angles of the triangle.
Chapter 9
9.1. Not very helpful Hint. Use the formula for the distance
function.
9.2. Not very helpful Hint. Argue by contradiction.
9.3. Hint. This follows immediately from the uniqueness of the

perpendicular from a given point to a given line and the definition of
reflections.
Chapter 12 295
9.4. Hint. Consider the composition of the reflections in the two

lines and use Problem 9.3.
9.5. Hint. Consider the reflection in one of the perpendiculars

and then in the other.
9.6. Hint. Argue by contradiction and, using the properties

of the model that are also true in Euclidean geometry (such as the
properties of the distance listed in 9.1.2 and the existence and unique-
ness of perpendiculars) show that Euclid’s Fifth Postulate would then
hold.
9.7. Hint. This is impossible: the existence of a linear space

structure in the model means that similar but noncongruent triangles
exist, but this, as is not hard to see, is equivalent to Euclid’s Fifth
Postulate.
9.8. Hint. Take an equilateral triangle (in the Euclidean sense)

with vertices on the absolute. Then another equilateral triangle in its
interior with vertices sufficiently close to the absolute will do.
Chapter 10
10.8. Hint. Take a very small equilateral triangle ABC (whose

angles are close to π/3) with center of symmetry at the center O of
the Poincaré disk model, consider the homothety with center O and
a huge coefficient, and look at the angles of the image of ABC.
10.9. Hint. (a) The proof is a nontrivial calculation; see [16],

pp. 149–151. (b) Use the previous formula and the formula
ed = sinh(d) + cosh(d),
similar to Euler’s famous formula for e−d and just as easy to prove.
Chapter 12
12.9. Brianchon’s theorem may be stateda as follows: Let a, b, c,

d, e, f be tangents to a conic. Let l, m, n be the lines passing through
296 Answers & Hints
the points L1 = a ∩ b and L2 = e ∩ d, M1 = a ∩ f and M2 = c ∩ d,

N1 = c ∩ b and N2 = e ∩ f , respectively. Then the three lines l, m, n
intersect at one point. The proof may be obtained by dualizing the
proof of Pascal’s Theorem.
Chapter 13
13.1. Not very helpful Hint. A proof can be obtained by a

straightforward calculation in coordinates.
13.5. Hint. By choosing a triangle of area less than ε in the

southern hemisphere close enough to the equator, we can ensure that
the area of its image under the central projection is as large as we
wish.
Chapter 14
14.1. The linear spaces AF(2) and AF(3) over the fields of 2
and 3 elements may be regarded as affine geometries with 4 and 9
elements, respectively; but they can be constructed without appealing
to finite fields. Thus the one with four points a, b, c, d consists of six
lines ab, bc, cd, da, ac, bd, three pairs of which (ab and cd, dc and da,
ac and bd) are parallel. It can also be described as the Fano projective
plane with the “points at infinity” removed. The one with 9 points
can also be constructed directly; the reader will profit by drawing a
picture of AF(3) in the style of Figure 14.1.
Chapter 15
15.7. If it was, spherical lines would be subsets of projective

lines, and any pair of spherical lines should have one common point
or less.
Chapter 16
16.1. wlcm(k,l) .
Bibliography
[1] A.F. Berdon, The Geometry of Discrete Groups, Springer, Berlin, 1983.
[2] M. Berger, Geometry, Springer, Berlin, 1987.
[3] A. Cayley, Collected Papers, Cambridge, 1889.
[4] H.S.M. Coxeter, Projective Geometry, Toronto, 1942.
[5] B.A. Dubrovin, S.P. Novikov, A.T. Fomenko, Modern Geometry.
Methods and Applications, Springer, Berlin, 1985.
[6] N.V. Efimov, Higher Geometry, Moscow, Mir, 1980.
[7] B. Grünbaum, What symmetry groups are present in the Alhambra,
Notices Amer. Math. Soc., 53, no. 6, 2006.
[8] T.L. Heath, The Thirteen Books of Euclid’s Elements, Cam-
bridge, 1926.
[9] D. Hilbert, Grundlagen der Geometrie, 7th Edition, Leipzig & Berlin,
1930; The Foundations of Geometry (Authorized translation by E.J.
Townsend), Chicago, 1902.
[10] F. Klein, A comparative review of recent researches in geometry, Bull.
New York Math. Soc., 2, 215–249, 1892–1893.
[11] N. Lobachevsky, Geometrical Researches on the Theory of Parallels,
Austin, Texas, 1891.
[12] H. Poincaré, Science and Hypothesis, London, 1905.
[13] B. Riemann, Gesammelte Mathematische Werke, Leipzig, 1892.
[14] D.E. Smith, A Source Book in Mathematics, New York, 1929.
[15] P. Stäckel, Wolfgang und Johann Bolyai, Geometrische Unter-
suchugen, Leipzig & Berlin, 1913.
[16] H.E. Wolfe, Non-Euclidean Geometry, Dryden Press, N.Y., 1945.
[17] J. Bolyai, Appendix, Edited by F. Kárteszi, North-Holland Mathemat-
ical Studies, Amsterdam, 1987.
297
Index
Abelian group, 54 coset, 58

absolute, 132, 147 Coxeter geometry, 101
adjacent angles, 4 Coxeter polygon, 101
affine space, 110 Coxeter polyhedron, 101
affine transformation, 144 Coxeter scheme, 105
angle, 4 cross ratio of collinear points, 191
angle of parallelism, 170 cross ratio of four complex
Artin relation, 63 numbers, 144
axial symmetry, 30 crystallographic group, 95
axiom of Archimedes, 279 cyclic group, 55
axiom of completeness, 280
axiom of continuity, 279 defining relations in a group
axioms of congruence, 276 presentation, 64
axioms of connection, 272 Desargues’ theorem, 194
axioms of order, 274 discrete group action, 90
distance function, 2
betweenness relation, 274 distance in the Cayley–Klein
biangle on the sphere, 116 model, 154
dodecahedron, 78
canonical Grassmann bundle, 249 duality in projective geometry, 193
Cayley–Klein model, 155 duality principle in projective
central symmetry, 8, 31 geometry, 193
circle, 20
class formula, 43 elliptic geometry, 121
classifying space, 251 embedding of a geometry, 48
common notions, 257 epimorphism, 44
conic section, 194 equivariant map, 47
consistency, 280 Erlangen program, 46
convex polyhedron, 30 Euclid, xiv, 10
convex set, 30 Euclidean geometry, 1
299
300 Index
Euclidean plane, 2 Lie group, 247

Euclidean space, 111 Lie, Sophus, xv
line at infinity, 188
Fano plane, 218 Lobachevsky distance, 168
Fedorov group, 87
Fifth Postulate, 177, 257 Möbius distance, 150
finite affine plane, 217 Möbius group, 145
over a field, 215 monomorphism, 44
finite field, 212 morphism of geometries, 47
flat cylinder, 245 morphism of transformation
flat torus, 244 groups, 43
free group, 56, 61 motion of the plane, 34
fundamental domain, 43
fundamental tile, 91 neutral element, 54
nonintersecting lines in the
general linear group GL(n), 56, 110 Caley–Klein model, 157
general position, 189 nonintersecting lines in the
generators of a group, 57 hyperbolic plane, 134
geometric G-bundle, 250 normal subgroup, 60
geometric sum, 4
geometry in the sense of Klein, 46 orbit, 42
glide symmetry, 24 order of a transformation group, 45
Grassmann manifold, 248 order of an element g of a
transformation group, 45
Hilbert, xiv order of an element of a group , 57
homomorphism, 57 orientation-reversing isometry of
of transformation groups, 43 the plane, 34
honeycomb lattice, 91 oriented angle, 7
Hopf bundle, 245 orthogonal group, 111
hyperbolic circle, 137 orthonormal vector space, 110
hyperbolic plane, 132
Pappus’ theorem, 198
incidence, 192 parallel lines, 2
inverse element, 54 in space, 28
inversion, 126 parallel planes, 28
involution, 127 parallel translation, 11
isometry, 3, 34 parallelogram, 11
isometry group of the Riemannian parallels in the Cayley–Klein
elliptic plane, 41 model, 157
isomorphic geometries, 48 parallels in the hyperbolic plane,
isomorphic transformation groups, 134
44 Pascal’s theorem, 199
isomorphism of transformation Penrose tiling, 87
groups, 44 permutation group, 42, 56, 62
perpendicular line to a plane, 29
join, 251
perpendiculars, 4
Klein, Felix, xiii, xv, 46 Poincaré disk model of the
hyperbolic plane, 131
Lagrange’s theorem, 59 point at infinity, 126
Index 301
polar of a point on the sphere, 115

poles of a line on the sphere, 115
presentation of a group, 63
principal G-bundle, 250
projective plane, 185
projective space, 199
of arbitrary dimension, 187
pullback, 248
quotient group, 61
reflection group, 99
reflection in a line, 5
reflection in a plane, 29
regular tiling, 91
residues modulo m, 55
Riemann sphere, 126
rotation, 7
rotation about an axis, 30
Schweikart constant, 171

stabilizer, 42
Stiefel manifold, 249
Stiefel-over-Grassmann bundle, 249
subgeometry, 48
subgroup, 58
of a transformation group, 45
symmetry group of the equilateral
triangle, 36
symmetry group of the square, 38
tessellation, 90
tiling, 90
tiling geometry, 91
tiling group, 91
transitive action, 91
triangle, 10
undefined notions, 272

universal geometric G-bundle, 251
vector space, 110

vector, attached, 11
vector, free, 12
vertical angles, 4
Vorderberg tiling, 86
Selected Published Titles in This Series
64 A. B. Sossinsky, Geometries, 2012
62 Rebecca Weber, Computability Theory, 2012
61 Anthony Bonato and Richard J. Nowakowski, The Game of Cops
and Robbers on Graphs, 2011
60 Richard Evan Schwartz, Mostly Surfaces, 2011
59 Pavel Etingof, Oleg Golberg, Sebastian Hensel, Tiankai Liu, Alex
Schwendner, Dmitry Vaintrob, and Elena Yudovina, Introduction to
Representation Theory, 2011
58 Álvaro Lozano-Robledo, Elliptic Curves, Modular Forms, and Their
L-functions, 2011
57 Charles M. Grinstead, William P. Peterson, and J. Laurie Snell,
Probability Tales, 2011
56 Julia Garibaldi, Alex Iosevich, and Steven Senger, The Erdős
Distance Problem, 2011
55 Gregory F. Lawler, Random Walk and the Heat Equation, 2010
54 Alex Kasman, Glimpses of Soliton Theory, 2010
53 Jiřı́ Matoušek, Thirty-three Miniatures, 2010
52 Yakov Pesin and Vaughn Climenhaga, Lectures on Fractal Geometry
and Dynamical Systems, 2009
51 Richard S. Palais and Robert A. Palais, Differential Equations,
Mechanics, and Computation, 2009
50 Mike Mesterton-Gibbons, A Primer on the Calculus of Variations and
Optimal Control Theory, 2009
49 Francis Bonahon, Low-Dimensional Geometry, 2009
48 John Franks, A (Terse) Introduction to Lebesgue Integration, 2009
47 L. D. Faddeev and O. A. Yakubovskiĭ, Lectures on Quantum
Mechanics for Mathematics Students, 2009
46 Anatole Katok and Vaughn Climenhaga, Lectures on Surfaces, 2008
45 Harold M. Edwards, Higher Arithmetic, 2008
44 Yitzhak Katznelson and Yonatan R. Katznelson, A (Terse)
Introduction to Linear Algebra, 2008
43 Ilka Agricola and Thomas Friedrich, Elementary Geometry, 2008
42 C. E. Silva, Invitation to Ergodic Theory, 2008
41 Gary L. Mullen and Carl Mummert, Finite Fields and Applications,
2007
40 Deguang Han, Keri Kornelson, David Larson, and Eric Weber,
Frames for Undergraduates, 2007
39 Alex Iosevich, A View from the Top, 2007
38 B. Fristedt, N. Jain, and N. Krylov, Filtering and Prediction: A
Primer, 2007
37 Svetlana Katok, p-adic Analysis Compared with Real, 2007
SELECTED PUBLISHED TITLES IN THIS SERIES
36 Mara D. Neusel, Invariant Theory, 2007

35 Jörg Bewersdorff, Galois Theory for Beginners, 2006
34 Bruce C. Berndt, Number Theory in the Spirit of Ramanujan, 2006
33 Rekha R. Thomas, Lectures in Geometric Combinatorics, 2006
32 Sheldon Katz, Enumerative Geometry and String Theory, 2006
31 John McCleary, A First Course in Topology, 2006
30 Serge Tabachnikov, Geometry and Billiards, 2005
29 Kristopher Tapp, Matrix Groups for Undergraduates, 2005
28 Emmanuel Lesigne, Heads or Tails, 2005
27 Reinhard Illner, C. Sean Bohun, Samantha McCollum, and Thea
van Roode, Mathematical Modelling, 2005
26 Steven J. Cox, Robin Forman, Frank Jones, Barbara Lee Keyfitz,
Frank Morgan, and Michael Wolf, Six Themes on Variation, 2004
25 S. V. Duzhin and B. D. Chebotarevsky, Transformation Groups for
Beginners, 2004
24 Bruce M. Landman and Aaron Robertson, Ramsey Theory on the
Integers, 2004
23 S. K. Lando, Lectures on Generating Functions, 2003
22 Andreas Arvanitoyeorgos, An Introduction to Lie Groups and the
Geometry of Homogeneous Spaces, 2003
21 W. J. Kaczor and M. T. Nowak, Problems in Mathematical Analysis
III, 2003
20 Klaus Hulek, Elementary Algebraic Geometry, 2003
19 A. Shen and N. K. Vereshchagin, Computable Functions, 2003
18 V. V. Yaschenko, Editor, Cryptography: An Introduction, 2002
17 A. Shen and N. K. Vereshchagin, Basic Set Theory, 2002
16 Wolfgang Kühnel, Differential Geometry: Curves – Surfaces – Manifolds,
Second Edition, 2006
15 Gerd Fischer, Plane Algebraic Curves, 2001
14 V. A. Vassiliev, Introduction to Topology, 2001
13 Frederick J. Almgren Jr., Plateau’s Problem: An Invitation to Varifold
Geometry, Revised Edition, 2001
12 W. J. Kaczor and M. T. Nowak, Problems in Mathematical Analysis
II, 2001
11 Mike Mesterton-Gibbons, An Introduction to Game-Theoretic
Modelling, Second Edition, 2001
For a complete list of titles in this series, visit the

AMS Bookstore at www.ams.org/bookstore/.
The book is an innovative modern exposition of geometry, or rather, of
geometries; it is the first textbook in which Felix Klein’s Erlangen Program
(the action of transformation groups) is systematically used as the basis for
defining various geometries. The course of study presented is dedicated to
the proposition that all geometries are created equal—although some, of
course, remain more equal than others. The author concentrates on several
of the more distinguished and beautiful ones, which include what he terms
“toy geometries”, the geometries of Platonic bodies, discrete geometries,
and classical continuous geometries.
The text is based on first-year semester course lectures delivered at the
Independent University of Moscow in 2003 and 2006. It is by no means
a formal algebraic or analytic treatment of geometric topics, but rather,
a highly visual exposition containing upwards of 200 illustrations. The
reader is expected to possess a familiarity with elementary Euclidean geom-
etry, albeit those lacking this knowledge may refer to a compendium in
Chapter 0. Per the author’s predilection, the book contains very little
regarding the axiomatic approach to geometry (save for a single chapter
on the history of non-Euclidean geometry), but two Appendices provide
a detailed treatment of Euclid’s and Hilbert’s axiomatics. Perhaps the most
important aspect of this course is the problems, which appear at the end
of each chapter and are supplemented with answers at the conclusion of
the text. By analyzing and solving these problems, the reader will become
capable of thinking and working geometrically, much more so than by
simply learning the theory.
Ultimately, the author makes the distinction between concrete mathemat-
ical objects called “geometries” and the singular “geometry”, which he
understands as a way of thinking about mathematics. Although the book
does not address branches of mathematics and mathematical physics such
as Riemannian and Kähler manifolds or, say, differentiable manifolds and
conformal field theories, the ideology of category language and transfor-
mation groups on which the book is based prepares the reader for the study
of, and eventually, research in these important and rapidly developing areas
of contemporary mathematics.
For additional information

and updates on this book, visit
www.ams.org/bookpages/stml-64
AMS on the Web

STML/64 www.ams.org

Geometries: A. B. Sossinsky

Uploaded by

Copyright:

Available Formats

Geometries: A. B. Sossinsky

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Geometries: A. B. Sossinsky

Uploaded by

Copyright:

Available Formats

S T U D E N T M AT H E M AT I C A L L I B R A RY

American Mathematical Society

2010 Mathematics Subject Classiﬁcation. Primary 51-01;

For additional information and updates on this book, visit

Library of Congress Cataloging-in-Publication Data

Copying and reprinting. Individual readers of this publication, and nonproﬁt

Chapter 0. About Euclidean Geometry 1

Chapter 1. Toy Geometries and Main Deﬁnitions 33

Chapter 2. Abstract Groups and Group Presentations 53

Chapter 3. Finite Subgroups of SO(3) and the Platonic Bodies 67

Chapter 4. Discrete Subgroups of the Isometry Group of the

Chapter 5. Reﬂection Groups and Coxeter Geometries 99

§5.2. Coxeter polygons and polyhedra 100

Chapter 6. Spherical Geometry 109

Chapter 7. The Poincaré Disk Model of Hyperbolic Geometry 125

Chapter 8. The Poincaré Half-Plane Model 143

Chapter 9. The Cayley–Klein Model 153

Chapter 10. Hyperbolic Trigonometry and Absolute Constants 163

Chapter 11. History of Non-Euclidean Geometry 177

Chapter 12. Projective Geometry 185

§12.9. Problems 200

Chapter 13. “Projective Geometry Is All Geometry” 203

Chapter 14. Finite Geometries 211

Chapter 15. The Hierarchy of Geometries 229

§15.7. Problems 238

Chapter 16. Morphisms of Geometries 241

Appendix A. Excerpts from Euclid’s “Elements” 255

Appendix B. Hilbert’s Axioms for Plane Geometry 271

Answers & Hints 283

in the geometric context. Thus, Cayley’s famous phrase: “projec-

Let no one enter who is not a geometer

should also be displayed on the gates leading to the world of mathe-

This book is based on lectures given in the framework of semester

The brevity of a short semester course (13 lectures) made me

The ﬁrst chapter of this book is Chapter 1 (see p. 33 below); the

and some experience in the kind of logical reasoning used in math-

0.1. The axioms of Euclidean plane geometry

I. The plane E2 is the set of all points.

II. The family L of all lines consists of nonempty subsets of the

VI. The distance function d possesses the following properties:

(iii) d(A, B) + d(B, C) ≥ d(A, C) for all A, B, C ∈ E2 ; the

A few more deﬁnitions are needed to state the last axioms.

d(β(P ), β(Q)) = d(P, Q) for all A, B ∈ E2 .

Following the traditional terminology, we will often call subsets of

Remark. In some expositions of Euclidean geometry (e.g., Hilbert’s,

VII. For any line l ∈ L and any point P ∈ E2 , there exists a

Figure 0.1. Dropping and raising perpendiculars.

Note that the axiom does not specify whether P lies on l or

One more deﬁnition, a very important one, is needed to formulate

VIII. The reﬂection in any line is an isometry.

If l is any given line and P is a point not on l, then the half-plane

0.2.2. Independence. The axioms I–VIII are not independent:

0.2.3. Consistency. The axiom system above is consistent (i.e., no

0.2.4. Categoricity. The axiom system above is categorical, i.e.,

0.2.5. Why the axiomatic approach? The reader may wonder

0.3.2. Deﬁnition of rotations. An oriented angle is an ordered

Figure 0.2. Rotation by 2α.

In the particular case in which the rotation is deﬁned by an ori-

Proposition 0.3.3. Rotations and central symmetries, being compo-