(Birger Iversen) Hyperbolic Geometry (B-Ok - Xyz) PDF

LONDON MATHEMATICAL SOCIETY STUDENT TEXTS
Managing editor: Dr C.M. Series, Mathematics Institute

University of Warwick, Coventry CV4 7AL, United Kingdom
1 Introduction to combinators and A.-calculus, J.R. HINDLEY & J.P. SELDIN

2 Building models by games, WILFRID HODGES
3 Local fields, J.W.S. CASSELS
4 An introduction to twistor theory, S.A. HUGGETT & K.P. TOD
5 Introduction to general relativity, L.P. HUGHSTON & K.P. TOD
6 Lectures on stochastic analysis: diffusion theory, DANIEL W. STROOCK
7 The theory of evolution and dynamical systems, J. HOFBAUER & K. SIGMUND
8 Summing and nuclear norms in Banach space theory, G.J.O. JAMESON
9 Automorphisms of surfaces after Nielsen and Thurston, A. CASSON & S. BLEILER
10 Nonstandard analysis and its applications, N. CUTLAND (ed)
11 Spacetime and singularities, G. NABER
12 Undergraduate algebraic geometry, MILES REID
13 An introduction to Hankel operators, J.R. PARTINGTON
14 Combinatorial group theory: a topological approach, DANIEL E. COHEN
15 Presentations of groups, D.L. JOHNSON
16 An introduction to noncommutative Noetherian rings, K.R. GOODEARL &
R.B. WARFIELD, JR.
17 Aspects of quantum field theory in curved spacetime, S.A. FULLING
18 Braids and coverings: selected topics, VAGN LUNDSGAARD HANSEN
19 Steps in commutative algebra, R.Y. SHARP
20 Communication theory, C.M. GOLDIE & R.G.E. PINCH
21 Representations of finite groups of Lie type, FRAN4;OIS DIGNE & JEAN MICHEL
22 Designs, graphs, codes, and their links, P.J. CAMERON & J.H. VAN LINT
23 Complex algebraic curves, FRANCES KIRWAN
24 Lectures on elliptic curves, J.W.S. CASSELS
25 Hyperbolic geometry, BIRGER IVERSEN
26 An Introduction to the theory of L-functions and Eisenstein series, H. HIDA
27 Hilbert Space: Compact Operators and the Trace Theorem, J.R. RETHERFORD
London Mathematical Society Student Texts 25
Hyperbolic Geometry
Birger Iversen
Aarhus University
AMBRIDGE
UNIVERSITY PRESS
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo, Delhi
Cambridge University Press

The Edinburgh Building, Cambridge C132 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521435086
© Cambridge University Press 1992
This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 1992

Re-issued in this digitally printed version 2008
A catalogue record for this publication is available from the British Library
ISBN 978-0-521-43508-6 hardback

ISBN 978-0-521-43528-4 paperback
CONTENTS
INTRODUCTION ix
I QUADRATIC FORMS
I.1 ORTHOGONALITY 1
1.2 WITT'S THEOREM 6
1.3 SYLVESTER TYPES 9
1.4 EUCLIDEAN VECTOR SPACES 12
1.5 PARABOLIC FORMS 16
1.6 LORENTZ GROUP 23
1.7 MOBIUS TRANSFORMATIONS 32
1.8 INVERSIVE PRODUCTS OF SPHERES 39
1.9 RIEMANN SPHERE 44
1. EXERCISES 50
II GEOMETRIES
II.1 GEODESICS 57
11.2 EUCLIDEAN SPACE 59
11.3 SPHERES 65
II.4 HYPERBOLIC SPACE 70
11.5 KLEIN DISC 75
II.6 POINCARE DISC 76
11.7 POINCARE HALF-SPACE 78
11.8 POINCARE HALF-PLANE 80
H. EXERCISES 84
III HYPERBOLIC PLANE

III.1 VECTOR CALCULUS 88
111.2 PENCILS OF GEODESICS 92
111.3 CLASSIFICATION OF ISOMETRIES 99
111.4 THE SPECIAL LINEAR GROUP 104
111.5 TRIGONOMETRY 107
111.6 ANGLE OF PARALLELISM 111

vi CONTENTS
111.7 RIGHT ANGLED PENTAGONS 112
111.8 RIGHT ANGLED HEXAGONS 114

111.9 FROM POINCARE TO KLEIN 116
III.10 LIGHT CONE 118
III EXERCISES 121
IV FUCHSIAN GROUPS
IV.1 DISCRETE SUBGROUPS 125
APPENDIX: DISCRETE SUBSETS 129
IV.2 ELEMENTARY SUBGROUPS 131
IV.3 GEOMETRY OF COMMUTATORS 137

IV.4 JACOB NIELSEN'S THEOREM 141
IV.5 CUSPS 144

IV EXERCISES 148
V FUNDAMENTAL DOMAINS
V.1 THE MODULAR GROUP 150
V.2 LOCALLY FINITE DOMAINS 155
V.3 CONVEX DOMAINS 159

V.4 DIRICHLET DOMAINS 167
V EXERCISES 170
VI COVERINGS
VI.1 HYPERBOLIC SURFACES 171
VI.2 HOPF-RINOW THEOREM 175
VI.3 UNIFORMIZATION 179
VI.4 MONODROMY 182
VI.5 ORBIT SPACES 186
VI.6 CLASSIFICATION 189
VI.7 INSTANT JET SERVICE 191
VI.8 HOMOTOPY 196
VI EXERCISES 198
CONTENTS vu
VII POINCARE'S THEOREM
VII.1 COMPACT POLYGONS 200
VII.2 TRIANGLE GROUPS 209
VII.3 PARABOLIC CYCLE CONDITIONS 213
VII.4 CONJUGACY CLASSES 218
VII.5 SUBGROUPS OF THE MODULAR GROUP 221
VII.6 COMBINATORIAL TOPOLOGY 226
VII.7 FRICKE-KLEIN SIGNATURES 230
VII EXERCISES 235
VIII HYPERBOLIC 3-SPACE
VIII.1 EXTERIOR ALGEBRA 240
VIII.2 GRASSMANN RELATIONS 245
VIII.3 NORMAL VECTORS 248
VIII.4 PENCILS OF PLANES 251
VIII.5 BUNDLES OF GEODESICS 255
VIII.6 HERMITIAN MATRICES 260
VIII.7 DIRAC'S ALGEBRA 262
VIII.8 CLIFFORD GROUP 265
VIII.9 LIGHT CONE 268
VIII.10 QUATERNIONS 271
VIII.11 FIXED POINT THEOREMS 272
VIII EXERCISES 278
APPENDIX: AXIOMS FOR PLANE GEOMETRY 284
SOURCES 291
BIBLIOGRAPHY 292
SYMBOLS 295
INDEX 296
INTRODUCTION
What is hyperbolic geometry? Let me try to give an answer by telling the

story of the parallel axiom. I shall use modern language which will ruin part of
the story but highlight the basic points.
Axioms for plane geometry A simple set of axioms for plane

geometry can be presented in the framework of metric spaces. By a line in a
metric space X we understand the image of a distance preserving map
The three axioms of plane geometry are (the axioms are analysed in an appendix)
INCIDENCE AXIOM Through two distinct points of X there passes a unique

line. The space X has at least one point.
REFLECTION AXIOM The complement of a given line in X has two connected

components. There exists an isometry o of X which fixes the points of the line,
but interchanges the two connected components of its complement.
PARALLEL AXIOM Through a given point outside a given line there passes a
unique line which does not intersect the given line.
Investigations of the parallel axiom by among others J.Bolyai (1802 -

1860), C.F.Gauss (1777 - 1855), N.I.Lobachevsky (1793 - 1856) show that this
axiom is independent of the other axioms in the sense that there exists a plane, the
so called hyperbolic Wane H2, which satisfies the first two axioms of plane geome-
try but not the parallel axiom. H2 is unique in the following sense.
CLASSIFICATION THEOREM A metric space which satisfies the three axioms

of plane geometry is isometric to the Euclidean plane. A space which satisfies the
first two axioms but not the parallel axiom is isometric to H2 (after rescalingl).
x INTRODUCTION
The discovery of the hyperbolic plane began with a series of attempts to

prove the parallel axiom from the other axioms. Such attempts should inevitably
lead to the discovery of the hyperbolic plane since a system of axioms for the
hyperbolic plane consists in the incidence axiom, the reflection axiom and the
negation of the parallel axiom. G.Sacheri (1667- 1733) was the first to penetrate
deeper into this universe. He was followed by J.H.Lambert (1728 - 77) who made
some important speculations on the nature of a non-Euclidean geometry. By 1850
most of the properties of a new geometry were known to Bolyai, Gauss and
Lobachevsky. At that time the biggest problem was the existence of such a
geometry! Twenty years later this was firmly settled through the works of
E.Beltrami (1835 - 1900), A.Cayley (1821 - 95) and F.Klein (1849 - 1925). For
more information on the history of hyperbolic geometry see [Milnor] or [Greenberg]
and the references given there.
The Poincare half-plane In 1882 H.Poincare (1854 - 1912) discovered

a new model of H2, which I shall now describe. The model is the upper half-plane
of the complex plane equipped with the metric given by
cosh d(z,w) = 1 + Iz - w12lm[z]-1 Im[w]-1
2
The lines in the Poincare half-plane are traced by Euclidean circles with centres on
the x-axis or Euclidean lines perpendicular to the x-axis. It is quite obvious that
this model satisfies the incidence and reflection axioms but not the parallel axiom:
Euclidean inversions furnish the isometries required by the reflection axiom.
The usefulness of the Poincare half-plane model lies in the fact that the
group of orientation preserving isometries consists of analytic transformations of
the form
z H az+b E S12(R)
cz + d
In fact the full group of isometries of the Poincare half-plane is the group PG12(R).
In particular the group G12(Z) acts as isometries on H2, a fact which lies at the
root of the applications of hyperbolic geometry to number theory.
INTRODUCTION xi
The Poincare disc The open unit disc D is the second most commonly
used model for the hyperbolic plane. We shall not write down the metric here, but
mention that the lines in this model are traced by Euclidean circles and lines
orthogonal to the unit circle OD. The Poincare disc has certain aesthetic qualities,
as shown below by a one of the many beautiful illustrations from [Klein, Fricke].
Discrete groups Reflections in the of lines of the tesselation above

generate a discrete group of isometries of H2. The systematic study of such groups
was initiated by Poincare in his "Memoir stir les groupes fuchsienne" 1882.
Starting from a discrete group t of isometries of H2 he used a procedure due to
Dirichlet to construct a polygon 0 with a side pairing, i.e. instructions for
identification of the sides of the polygon, which allows us to reconstruct the space
H2/t. Then he took up the converse problem: given a hyperbolic polygon with a
side pairing, when does the side pairing generate a discrete group with the given
polygon as fundamental domain. Poincare laid down necessary and sufficient
conditions for a given polygon with side pairings to determine a discrete group.
The theorem has a bonus: from the geometry of the polygon we can read off a
xii INTRODUCTION
complete system of generators and relations for the group. In this way the topic
has had a big influence on what today is known as combinatorial group theory.
The theorem of Poincare is one of the main themes of this book. The proof of the
theorem is based on some geometric ideas which I shall outline next.
New geometries Let us for a moment go back to the discussion of the

parallel axiom and recall that the hyperbolic plane is quite unique. However, we
find numerous new geometries in the class of hyperbolic surfaces: spaces locally
isometric to H2. We shall see that complete hyperbolic surfaces can be classified
by discrete subgroups of PG12(R). The main point in the proof of Poincare's
theorem is the monodromy theorem: A local isometry f:X--+H2 from a complete
hyperbolic surface X to the hyperbolic plane H2 is an isomorphism.
Higher dimensions So far we have only talked about hyperbolic

geometry in two dimensions, but the hyperbolic space H" exists in all dimensions.
It is the second main theme of this book to construct these spaces and to identify
their isometry groups with the Lorentz group known from special relativity.
Hyperbolic n-space is constructed as one of the two sheets of the hyperbola
x02 - x12 - x32 - X42..... -xn2 = 1
It turns out that the most natural framework for this construction is the theory of
quadratic forms. This is actually where the books starts. We offer a general
introduction to quadratic forms in Chapter I as an opener to the construction of
geometries in Chapter H. It is natural to develop Euclidean and spherical
geometries in a parallel fashion in order to be able to draw on analogies with these
more familiar geometries. A special feature of hyperbolic geometry is the presence
of a natural boundary OHn. In terms of the hyperbola above, a boundary point
can be thought of as an asymptote to the hyperbola, i.e. line in the cone
X02
- x12 - - X42..... - Xn2 = 0
X32
It follows that the boundary OHn is a sphere. In the end we find that the Lorentz
group acts on the sphere as the group of Mobius transformations.
INTRODUCTION xiii
Three-dimensional geometry A specific feature of three-

dimensional hyperbolic geometry is that the group S12(C) operates on H3 (see
below) in such a way as to give an isomorphism between PG12(C) and the group of
orientation preserving isometries of 113, the special Lorentz group. This has the
consequence that a Kleinian rou , i.e. is a discrete subgroup F of S12(C), defines a
3-manifold H3/F. The work of W.P.Thurston on 3-manifolds [Thurston] has
generated new interest in hyperbolic geometry.
Exterior algebra Let us think of H3 as a sheet of the unit hyperbola in a

Minkowski space M, i.e. a four-dimensional vector space with a form of type
(-3,1). Once we pick an orientation of M, the space A 2M comes equipped with
the structure of a three-dimensional complex vector space with a complex
quadratic form Q. The assignment of normal vector n E A 2M to an oriented
geodesic line h in H3 creates a bijection between the complex quadric Q = - 1 and
the set of oriented geodesics in H3.
The Hermitian model As mentioned above, each Minkowski space M

comes equipped with its own copy of 113, a sheet of the unit hyperbola. It turns
out that there exist Minkowski spaces with more algebraic structures, in particular
a structure of a linear representation of S12(C). We shall be interested in the space
M of 2 x 2 Hermitian matrices, i.e. complex matrices of the form
;a,dER,bEC
The restriction of the determinant to M is a quadratic form of type (-3,1). The

sheet of the unit hyperbola (ad - bb = 1) consisting of positive definite forms
(a > 0) is called the Hermitian model of H3. The action of S12(C) on M is given
by the formula
oX=oXo* ;oES12(C),XEM
The key fact about the Herinitian model is the following trace formula: The
action of S E G12(C) on H3 can be decomposed into a,6 where a and 0 are half-
turns with respect to geodesics a and b. For any such decomposition we have that
xiv INTRODUCTION
tr2(S) = 4 <m,n>2
where m and n denote normal vectors for the geodesics a and b.
The realisation of Minkowski space through the space M of Hermitian
matrices has been used in physics for a long time. We shall introduce another tool
from physics namely the Dirac algebra, which is a concrete realisation of the
Clifford algebra of M. The use of Clifford algebras was suggested by the "strange
formulas" from the book of [Fenchel].
Prerequisites The general prerequisites are linear algebra and a modest

amount of point set topology, including compact subsets of a finite dimensional
real vector space. Familiarity with the concepts of group theory and, for the
understanding of Poincare's polygon theorem, a bare minimum of combinatorial
group theory ("generators and relations") is needed. Elementary topology is
needed for VI.8, VII.5, VII.7, while VII.6 is a good introduction to combinatorial
topology. The final chapter of the book requires familiarity with the exterior
algebra.
University of Aarhus, Denmark. July 1992 Birger Iversen
ACKNOWLEDGEMENT I would like to thank Jakob Bjerregaard, Elisabeth

Husum, Claus Jungersen and Pernille Pind for comments on the various notes on
which the book is built. A particular thank you goes to Gunvor Juul and editor
Caroline Series who helped the manuscript through its final stages.
I QUADRATIC FORMS
In the first two sections we lay down the general principles for quadratic
forms and the associated orthogonal groups. The third section will classify real
quadratic forms into Sylvester types, while the remaining sections will deal with
real forms of specific types.
Three types of real quadratic forms are of particular interest: positive
definite forms, parabolic forms and hyperbolic forms. The corresponding
orthogonal groups provide us with the isometry groups of the three basis
geometries: spherical geometry, Euclidean geometry and hyperbolic geometry.
For example, the orthogonal group of a hyperbolic form contains the
Lorentz group as a subgroup of index 2. The Lorentz group, on the one hand, is
the isometry group of hyperbolic geometry, while, on the other hand, it can be
identified with the somewhat older group of Mobius transformations. We shall see
that "the Mobius group is the boundary action of the Lorentz group".
1.1 ORTHOGONALITY
Let k denote a field' of characteristic # 2 and E a vector space over k of

finite dimension n. By a quadratic form on E we understand a function Q: E-+k
homogeneous of degree 2, i.e.
1.1 Q(\z)=A2Q(z) ;AEk,zEE
with the property that the symbol
1.2 <x,Y> = 2(Q(x + Y) - Q(x) - Q(y)) ;x,yEE
1 In this text we have applications only for the field R of real numbers,
the field C of complex numbers and occasionally the field Q of rational numbers.
2 QUADRATIC FORMS
is bilinear in x and y. The relation 1.2 between the bilinear form <x,y> and the
quadratic form Q(z) is called the formula for polarization. Observe that the
bilinear form <x,y> is symmetrical in x and y.
Proposition 1.3 Polarization gives a one to one correspondence between

quadratic forms and symmetrical bilinear forms over a field k with char(k) # 2.
Proof A quadratic form Q(z) satisfies the formula Q(2z) = 4Q(z) as a special
case of 1.1. We can now substitute z for x and y in formula 1.2 to get that
1.4 Q(z) _ <z,z> ;zEE
which recovers the function Q(z) from the symbol <x,y>. Conversely, if we start
with a symmetrical bilinear form <x,y> we can use formula 1.4 to define a k-
valued function Q on E which is homogeneous of degree 2. Bilinearity and
symmetry give us the formula
<x + y, x + y> = <x,x> + <y,y> + 2<x,y> x,y E E
which shows that Q(z) = <z,z> is a quadratic form.
The evaluation, Q(x), of the quadratic form at a vector x E E is often

referred to as the norm of x. Vectors x,y E E are called orthogonal if <x,y> = 0.
Linear subspaces U and V of E are called orthogonal if all vectors in U are
orthogonal to all vectors in V. For a given linear subspace K of E we define the
total orthogonal subspace K 1 by the formula
1.5 K1 = {eEEjVxEK<e,x>=0}
Definition 1.6 A quadratic form Q(z) on the finite dimensional vector

space E over k is called non-singular if the bilinear form <x,y> satisfies
1.1 ORTHOGONALITY 3
Vx E E (<x,y> = 0) y=0 ;yEE

Otherwise expressed, the form Q(z) is non-singular if and only if E 1 = 0.
Theorem 1.7 Let E be a finite dimensional vector space equipped with a

non-singular quadratic form Q. For any linear subspace K of E we have
dim K + dim K 1- = dim E
Proof Let us recall that a linear form on E is nothing but a k-linear map
:E--+k. Linear forms may be added and multiplied by a constant from k to form
a new vector space, the dual space E*. Let us prove that
1.8 dim E* = dim E
To this end we simply pick a basis el,...,ell for E and introduce the coordinate
forms xl,...,xn E E* by the formula
ei x1(e)e1 ;eEE
Let us verify that any linear form on E satisfies the formula
= 1: i (e;) xi E E*
To this end we simply evaluate both sides of the formula on the basis el,...,en.
This shows that xl,...,xn generates the vector space E*. To see that xl,...,xn are
linearly independent, consider a relation of the form E )i x1 = 0 , A1,...,An E k.
Evaluate the relation on ej, j = 1,...,n, and conclude that Ai = 0.
Let us introduce the map
1.9 q:E _ E*, q(e)=(x- <e,x>) ;eEE
Observe that the kernel for q is E 1 and use that Q is non-singular to conclude
that q is injective; formula 1.8 allows us to conclude that q is surjective as well.
Restriction of a form along the inclusion i:K- E defines a k-linear map
i*:E* K* ;fHoi,EE*
4 QUADRATIC FORMS
which is surjective: In the argument above pick the basis el,...,en for E such that
el,...,ek is a basis for K and observe that the coordinate forms for this basis are
xt o i,...,xk o i. We shall focus on the composite map
E q E* -, K*
Since this is surjective we conclude from the dimension formula of Grassmann2

dim E = dim K* + dim Ker i" q
We ask the reader to show that Ker i*q = K The result follows from the
formula above in combination with 1.8.
Let us consider a pair (E,Q) consisting of a finite dimensional vector space

E and a quadratic form Q on E. By an isomorphism from (E,Q) to a second such
pair (F,P) we understand a linear isomorphism o:E=*F such that Q = P o o. In
particular we may talk about automorphisms of (E,Q) which may be composed to
form a group, the orthogonal group of (E,Q) which is denoted 0(Q) or just O(E)
when no confusion is possible.
Let us quite generally consider a quadratic form (E,Q) and pick a basis
el,...,en for the vector space E. The Gram matrix G E MM(k) is given by
1.10 Gii = <ei,ei> ; i,j = 1,...,n
Proposition 1.11 Let Q be a non-singular quadratic form on E. An

orthogonal transformation o E O(Q) has determinant 1 or -1.
Proof Let us pick a basis el,...,e. for E and let A denote the matrix for o,
thus o(ej) = E; A;ie1 for j = 1,...,n. Direct calculation gives us
<o(ei)e'Y(el)> = < E kAkiek, > hAhleh> = E h,k TAik<ek,eh>Ahi
which shows that the Gram matrix for Q o o is TAGA. In particular we find
that o is an orthogonal transformation for Q if and only if the matrix A obeys
2 A linear map f.E-F between finite dimensional vector spaces satisfies

dim E = dim Im(f) + dim Ker(f)
1.1 ORTHOGONALITY 5
1.12 T
AGA=G
Let us observe that the Gram matrix G is the matrix for the map q:E _ E* with
respect to the basis e1,...,en for E and x1,...,x11 for E*. Since Q is non-singular we
conclude that del G # 0. The formula 1.12 gives us
del TA del G del A = del G
from which we conclude that det2A = 1, or del A = 1 or -1.
The orthogonal group for the standard quadratic form on kn
1.13 Q(x) = x12+x22+ ... + x112 n
is denoted On(k). The subgroup of transformations with determinant 1 is denoted

SOn(k). When k = C or any other algebraically closed field with char(k) # 2 any
non-singular quadratic form is isomorphic to the standard form 1.13.
Theorem 1.14 Let Q be a non-singular quadratic form on the vector

space E over C of dimension n. There exists a basis e1,...,e11 for E with
<e;,ej> = 6ii ; i,j = 1,...,n
Proof Let us observe that Q can't be identically zero since this implies that
the bilinear form <x,y> is identically zero. Thus we can find a vector v E E with
Q(v) # 0. Let us write Q(v) = A2, A E C, or Q(A-lv) = 1 to conclude that we can
find e E E with Q(e) = 1. Let K denote the line spanned by e and put F = K 1
and observe that E = K ® F (meaning that any vector x E E has a unique
decomposition x = f + k, where f E F and k E K). It is easily seen that the
restriction of Q to F is non-singular. It is left to the reader to complete the proof
by simple induction on dim(E).
6 QUADRATIC FORMS
1.2 WITT'S THEOREM
In this section we shall be concerned with a non-singular quadratic form Q

on a vector space E of dimension n over a field k with char(k) 2. The theme is
the flexibility of the orthogonal group 0(Q). The tool is reflections, an important
class of orthogonal transformations, which we proceed to introduce.
We say that a vector n E E is non-isotropic if Q(n) # 0. A non-isotropic
vector n E E defines a linear transformation rn of E by the formula
2.1 <x,n> ;xEE

rn(x) = x - <n,n> n
Direct computation gives us

<rn(x),rn(x)> _ <x,x> ;x E E
i.e. rn is an orthogonal transformation called reflection along n. The hyperplane
orthogonal to n is fixed by rn while 7n(n) n. It follows that rn is an
involution , i.e. rn2 = t but rn # t, and that
2.2 delrn=-1 ;nEE, <n,n>#0
Proposition 2.3 Let Q be a quadratic form on the vector space E. For

any A E k*, the orthogonal group 0(Q) acts transitively3 on the set
{eEEI Q(e)=A}
Proof Consider vectors e,f E E with Q(e) = Q(f) = A # 0. Observe, that

<e - f, e + f> = <e,e> - <f,f> = 0
It follows that we can't have Q(e + f) = 0 and Q(e - f) = 0 at the same time (the
non-isotropic vector e is a linear combination of e - f and e + f). If the vector e - f
is non-isotropic, reflection r along this vector fixes e + f. Thus
7(e -f) = f - e , r(e+f) = e+f
3 We say that the action of a group G on a set X is transitive, if for all

x E X and y E X there exists a E G with a(x) = y.
1.2 WITT'S THEOREM 7
Now, add these two formulas to see that r(e) = f. If e + f is isotropic, reflection p
along this vector interchanges e and - f. In conclusion the orthogonal trans-
formation rip maps e to f as required.
When Q is non-singular, it is also true that O(Q) acts transitively on the

set of isotropic lines, i.e. lines in E generated by isotropic vectors. In fact,
Witt's theorem 2.4 Let (E,Q) be a non-singular quadratic form. Any

isomorphism a: (U,Q)-=-(V,Q) , where U and V are linear subspaces of E, can be
extended to an orthogonal transformation of (E,Q).
Proof Let us first treat the case where U is non-singular. This will be done
by induction on dim(U). In case dim(U) = 1 let us pick a non-zero u E U and put
A = Q(u). Since Q(o (u)) = A, we can apply 2.3 to extend v to an orthogonal
transformation of E.
To accomplish the induction step let us pick a non-isotropic vector e E U.
Since Q(e) = Q(v(e)) we can use 2.3 to pick r E O(E) such that o(e) = r(e). It
will suffice to extend r-1a: U,E to an orthogonal transformation of E. To recapi-
tulate it suffices to extend v:U-E under the assumption that a fixes a non-
isotropic vector e E U. Let us introduce the total orthogonal space W to L = k.e
in U and the total orthogonal space F to L in E. Observe that W and F are non-
singular and that W C F and o -(W) C F. Thus we can use the induction
hypothesis to extend the restriction a:W-F to an orthogonal transformation of F.
Let us now assume that U is singular. Let there be given data consisting
of a non-zero vector e E U fl U 1 and a linear complement R to L in U. We are
going to show that there exists an isotropic vector fin E which is orthogonal to R
and such that <e,f> = 1. Observe that we can find ,Q E E" with /3(e) = 1 and
/i(R) = 0 ; the vector b E E with q(b) _ /3 is orthogonal to R and satisfies
<b,e> = 1. Let us show that the vector f = b - 2<b,b>e is isotropic
<f,f> = <b,b> - <b,b><e,b> = 0
8 QUADRATIC FORMS
The isotropic vector f is orthogonal to R and satisfies <e,f> = 1. Let us observe

that f 0 U since e is orthogonal to U but <e,f> = 1. The space W = U + Rf is the
direct sum of the two orthogonal subspaces R and ke + kf. Let us perform the
same construction on the data o(U), or(e), o(R) to find an isotropic vector g E E
which is orthogonal to o(R) and which satisfies <g,o(e)> = 1. Let us extend
o:U-+E to a linear map WE by the convention o(f) = g. It is easy to check that
the extended map preserves the inner product: use that o(W) is the direct sum of
the orthogonal subspaces o(R) and ko(e) + ko(f).
Let us finally consider the general subspace U of E. If U is non-singular
we can use the first part of the proof to extend o. If U is singular we can use the
middle part of the proof a number of times until U becomes non-singular, and then
apply the first part of the proof.
The orthogonal group 0(E) of a non-singular form is generated by

reflections along non-isotropic vectors as it follows from exercise 2.3. The number
of reflections needed can be bounded by the following theorem of Elie Cartan: An
orthogonal transformation of a non-singular form of dimension n can be written as
a product of no more that n reflections along non-isotropic vectors.
The proof of the theorem of E. Cartan is rather complicated, see [Berger)
and [Deheuvels]. Moreover, the theorem is not precise enough for our purposes.
For these reasons we shall not make use of the theorem in its generality, but prove
a number of more specific results along these lines: namely 4.5, 5.3, 6.11
1.3 SYLVESTER TYPES
In this section we shall deal exclusively with quadratic forms over the field
R of real numbers. By a Euclidean vector space we understand a finite dimen-
sional vector space E with a quadratic form Q which is positive definite, i.e.
Q(e) > 0 for all non-zero e E E. The form Q is called negative definite if Q(e) < 0
for all non-zero e E E. Our first objective is to generalise the concept of an
orthonormal basis.
Proposition 3.1 Let (E,Q) be a quadratic form on the vector space E

over R of dimension n. There exists an orthonormal basis for E, i.e. a basis
el,...,en with
r 0 ; 36j
j
-1, 0, +1 i =j
Proof Let us use induction with respect to dim(E) = n. If the form Q is identi-
cally 0, any basis for E is orthonormal. Thus we may assume the existence of an
e E E with Q(e) # 0. Let us write Q(e) = eat with e = ± 1 and A E R*. Put
el = A-le to get Q(e1)=± 1. Let us use the induction hypothesis to find an ortho-
normal basis e2,...,e. for (Re1) -1' . The basis e1,...,e1, meets our requirements. 0
Sylvester's theorem 3.2 Let e1,...,en be an orthonormal basis for the

quadratic form (E,Q) over R. The numbers
p=#{iI<e;,e;>=-1}, q=#{iI <e;,e;>=1}
are independent of the orthonormal basis considered.
Proof Let us fix the basis as in the statement of the proposition, and let E_
denote the subspace of E, generated by those elements e; from our basis for which
<e;,e;> = 0, - 1. Observe that Q(e) < 0 for all e E E_ and conclude that for any
Euclidean subspace F of E we have F fl E_ = 0. According to 3.4 we have that
10 QUADRATIC FORMS
dim(F) + dim(E_) = dim(F + E_) + dim(F fl E_)

This gives us dim(F) + n - q:5 n, i.e. dim(F) < q. It follows that
3.3 sup { dim(F) I Euclidean F C E } = q
This shows that q is independent of the basis chosen. Let us apply this to the
space (E,-Q) and conclude that p too is independent of the basis.
With the notation of the theorem we say that the quadratic form (E,Q)
has Sylvester tune (-p,q). Observe that two quadratic forms (E,Q) and (F,R)
with dim(E) = dim(F) of equal Sylvester type are isomorphic.
Dimension formula 3.4 For subspaces U and V of the finite dimen-

sional vector space E we have that
dim(U fl v) + diin(U+V) = dim(U) + dim(V)
Proof Let us apply the Grassmann dimension formula to the linear map
f:U®V,E, f(u,v)=u-v ;uEU,vEV
observing that Ker f U fl V and Ini f = U + V.
Discriminant inequality 3.5 Let (E,Q) be a non-singular quadratic

form of Sylvester type (-s,r). For any basis el,...,ell for E we have
sign detij <ei,ej> = (-1)s
Proof Let us consider a second basis related to the first basis

through the transition matrix B. This gives us
<fi,fj> = < >r B;rer, Es Bjses> rs Bir<er,es>TBsj
From this we conclude that
det j <f,,fj> = det B detrs<e,es> del TB = det2B delii<e1,ej>
This shows that sign detij<ei,ej> is independent of the basis considered. We leave
it to the reader to verify the formula in the case of an orthonormal basis.
Symmetric matrices From a symmetric matrix A E MM(R) we can

construct a quadratic form on the free vector space E with basis el,...,en
Q(E xiei) = E,. arsxrxs
By the Sylvester type of A we understand the Sylvester type of the form Q. The
Sylvester type of A can be evaluated by means of the principal minors
deiA1,..., delAn
here Ai is the i x i matrix obtained by deleting the n-i last rows and columns of A.
Lemma 3.6 The symmetrical matrix A E Mn(R) is positive definite if and

only if all principal minors deMA1,..., detAn are strictly positive.
Proof To begin let us just assume that the principal minors are different from
zero and let us show how to determine the signature of A. With the notation
above the subspace Ei of E generated by e1,...,ei is non-singular since del Ai # 0.
Let us pick a non-zero vector fi E Ei, orthogonal to Ei_1. The Gram matrices of the
two bases el,...,ei and el,.... ei_1,f1 have the same sign, thus
sign del Ai = sign <fi,fi> sign del Ai_1
In the case at hand we deduce that <fi,fi> > 0 for i = 1,...,n.
12 QUADRATIC FORMS
1.4 EUCLIDEAN VECTOR SPACES
Let us consider a Euclidean vector space E , i.e. a finite dimensional real

vector space equipped with a positive definite quadratic form. A vector e E E has
length IeI = <e,e>. Let us start with the inequality of
Cauchy-Schwarz 4.1 I <e,f> I < I e I IfI ; e,f E E

Addendum: The inequality is sharp when e and f are linearly independent
Proof The inequality is trivial if e and f are linearly dependent. If e and f are
linearly independent, they span a Euclidean plane and we get from 3.5 that
<e,e> <e,f>
det >0
<f,e> <f,f>
from which the sharp inequality follows. 0
Triangle inequality 4.2 I e + fI < IeI +IfI ; e,f E E

Addendum: The triangle inequality is sharp when e and f are linearly independent.
Proof A simple direct calculation gives us

le +f12 = IeI2+ IfI2+2IeIIfI +2(<e,f> - IelIfI)
and the result follows from 4.1. 0
On the basis of the Cauchy-Schwartz inequality we can introduce the

angle L(e,f) between two non-zero vectors e,f by the formula
4.3 cos L(e,f) = ee,f> ; L(e,f) E [0,ir]

I IIfI
In the rest of this chapter we shall be concerned with the orthogonal group O(E).
The basic example of an orthogonal transformation is reflection r in a linear
hyperplane H. If n E E is a unit normal vector for H we have that
4.4 r(x) = x - 2<x,n>n ;xEE
Theorem 4.5 An orthogonal transformation o of the Euclidean vector

space E of dimension n is the product of at most n reflections through hyperplanes.
Proof We shall use induction on n = dim(E). To accomplish the induction

step we pick a vector f E E of unit length. If o(f) = f observe that o stabilises
F = f 1 and use the induction hypothesis to decompose o into a product of at
most n-1 reflections. In case o(f) # f observe that reflection r along the vector
c(f) - f interchanges f and o(f). It follows that ro fixes the vector f, so we can
use induction to write ro as a product of at most n-1 reflections.
Let us recall from 1.11 that an orthogonal transformation has determinant

1 or -1. It follows that the special orthogonal arg ouL
4.6 SO(E) = { o E O(E) I det o = 1 }
has index 2 in 0(E). We have seen that O(E) operates transitively on the sphere
4.7 S(E)=f eEEJ lei =11
We ask the reader to verify that SO(E) acts transitively on S(E), dim E > 2.
14 QUADRATIC FORMS
Euclidean plane Let us assume that dim(E) = 2. An orthogonal transfor-

mation of E with determinant 1 is called a rotation. In order to analyse an
orthogonal transformation o of E in more detail, let us pick an orthonormal basis i
and j for E; we can express that o(i) lies on the circle S(E) by writing
o(i)=icos0+jsill 0 ;OER
It follows that o(j) is one of the unit vectors orthogonal to o(i), i.e. the vector -i
sinO+j cos 0 or its negative. Thus the matrix for o takes one of the forms
cos 0 -sin 0 cos 0 sin 0

4.8 ; OER
sin 0 cos 0 sin. 0 -cos 0
Evaluation of the determinants show that the first matrix defines a rotation, while
the second defines a reflection. It follows from the addition formulas for the
trigonometric functions that the first matrix in 4.8 defines an isomorphism
between the circle group SO(E) and the group R/2irZ.
Euler's proposition 4.9 An orthogonal transformation o E SO(E) of

Euclidean 3-space E is a rotation: there is a line L fixed by o and the
transformation induced by o in the plane L 1 is a rotation. An orthogonal
transformation o E 0(E) of determinant -1 has the form pr where p is a rotation
as above with axis L and r is reflection in the plane L 1 .
Proof Consider the characteristic polynomial for the transformation

x(t) = det( tt - o )
Let us analyse a real root A: pick a vector e # 0 with o(e) = Ae and take norms to
see that A2 = 1, i.e. A= 1 or A = -1.
If dec(o) = 1 we have X(0) = -1. Observe that y(t) tends to +oo as t
tends to +oo and conclude that 1 is a root for X(t). Consider a line L with
eigenvalue 1 and observe that o acts on the plane L 1 with determinant 1.
If dec(o) _ -1 we have x(0) = 1. Observe that X(t) tends to -oo as t

tends to -oo and conclude that -1 is a root for X(t). Consider a line L with
eigenvalue -1 and observe that or acts on the plane L 1 with determinant 1. 0
Theorem 4.10 Let o be an orthogonal transformation of the Euclidean

vector space E of dimension n. There exists a decomposition of E into an
orthogonal sum of lines and planes stable under o.
Proof According to lemma 4.11 below we can find a linear subspace R C E of

dimension 1 or 2 stable under o. It follows that R 1 is stable under o as well. A
simple induction on dim E concludes the proof.
Lemma 4.11 Let o:E--*E be an endomorphism of a finite dimensional real

vector space E. There exists a subspace R C E of dimension 1 or 2 stable under o.
Proof Let us present a construction known as complexification of o. We ask

the reader to turn EC = E ® E into a vector space over C through the convention
(x + iy) (e,f) = (xe - yf, xf + ye) ; x,y E R, e,f E E
Let us identify e E E with (e,0) E EC to get the formula
(e,f)=e+if ; e,fEE
With this notation, we can introduce the C-linear endomorphism ac: EC-'EC by
oC(e + if) = o(e) + io(f) ; e,f E E
the details are straightforward and left to the reader. From the fundamental
theorem of algebra we conclude that oC has an eigenvalue A =a+ ib. A non-
trivial eigenvector e+if with eigenvalue A satisfies the equation
o(e) + io(f) = (a + ib)(e + if)
From this it follows that e and f generate a subspace of E stable under o-.
16 QUADRATIC FORMS
1.5 PARABOLIC FORMS
In this section we shall investigate a finite dimensional vector space F over

R equipped with a positive quadratic form Q, which of course means that Q(f) > 0
for all f E F. The point of departure is the
Cauchy-Schwarz inequality 5.1

<e,f>2 < <e,e> <f,f> ;e,fEF
Proof The Cauchy-Schwarz inequality may be rewritten in determinant form
<e,e> <e,f>
det > 0
<f,e> <f,f>
This is trivial when e and f are linearly dependent. When e and f are linearly
independent we get from 1.12 that the sign of the determinant is unchanged if we
replace e and f by an orthonorinal basis for the plane they span. 0
Corollary 5.2 Let (F,Q) be a positive quadratic form. A vector f E F is

isotropic if and only if f E F 1 , i.e. <e,f> = 0 for all e E F.
We shall investigate the orthogonal group O(F), in particular the sub-

group generated by reflections. Let us observe that a reflection r fixes all isotropic
vectors as follows from 5.2 and formula 2.1.
Theorem 5.3 Let Q be a positive quadratic form on the m-dimensional

real vector space F. An isometry o E O(F) which is the identity on F 1 can be
written as a product of at most m reflections.
Proof We proceed by induction on m. Let us distinguish between two cases.

10 Im(o - t) F 1 . Choose f E F such that p = o(f) - f is non-isotropic.
Reflection a along p interchanges f and o(f), i.e. 7ra(f) = f. Observe that f is non-
isotropic since o(f) # f and apply the induction hypothesis to the restriction of Tro
to the hyperplane f 1 .
2° Im(o - t) C F 1 .Let us decompose F into a direct sum F = E ® N

where E is Euclidean and N = F 1 . Let us describe o in terms of a linear map
v:E-N
c(e+n)=e+v(e)-{-ii ; eEE,nEN
We can arrange for v:E-N to be injective: a non-zero vector e E E of Ker v is
non-isotropic and fixed by o, which allows us to use the induction hypothesis on
e 1 . We can now arrange for v:E-N to be bijective by replacing N by v(E).
From now on we may assume that
dint F=2dimF
Observe that o acts as the identity on F 1 and on F/F 1 and conclude that
det(o) = 1. Let us pick any reflection p in F and observe that we are back to case
1° with pa since det(po) = -1. Thus we can write
Po=pl...ps s<m
where pl,.... ps are reflections. From the facts that m is even and det(po) _ -1, it
follows that s < m as required. 0
Let us say that a space (F,Q) is parabolic if Q is positive and the space
F 1 is of dimension 1. The isotropic line F 1 is invariant under orthogonal
transformations. It follows that we have a morphism of groups
5.4 µ:0(F)-;R*
where the multiplier p(o) E R* denotes the eigenvalue of o E 0(F) on the isotropic
line. This allows us to introduce a subgroup of 0(F) of index 2
5.5 O.(F)_ {0' E0(F)I p(c)>0}

18 QUADRATIC FORMS
Let us observe that the antipodal map 1, belongs to the centre of O(F)
and that µ(i) = -1. This defines a decomposition of the orthogonal group 0(F) of
a parabolic space F
5.6 O(F) = 0,,(F) x 7L/(2)
The line at infinity We shall pursue the study of parabolic spaces in a

setting which is particular useful for Euclidean geometry. Let E be a Euclidean
space of dimension n and consider the following bilinear form on F = E ® R
5.7 <(e,a),(f,b)> = <e,f> e,f E E, a,b E R
This form is easily seen to be parabolic. The isotropic line is generated by the
vector (0,1) and is called the line at infinity. Let us use the inner product on F to
introduce the evaluation map ev: F x E . -> R
ev((x,l;),y) = <x,y> + ; x,y E E, l; E I$
The evaluation map identifies F with the space of affine functions on E.
Similarities Let us recall that a similarity of the Euclidean space E is an

affine transformation of E of the form
fi(x)=aj(x)+f ; 0E 0(E),fEE, AER,.A>0
Our main objective is to identify the group Siinl(E) of similarities with the
orthogonal group 0,,(E).
Proposition 5.8 For any orthogonal transformation or of parabolic space

F=E ®R, there exists a unique similarity & E Siml(E) such that
ev(o(f),o(e)) = p(o) ev(f,e) fEF ,eE E
The assignment o,-*o induces an isomorphism of groups
0,0(F) - Si1nl(E)
The similarity )4 + v, 0 E O(E), A > 0, v E E, corresponds to 1 E 0,,(F) given by

11(z,O = (4'(z), - <¢(z),v> + AO ; z E E, (E R
Proof Let us first observe that a point e E E defines a linear form f ev(f,e)
on F with evaluation 1 against the isotropic vector (0,1). It is easy to see that any
linear form 0 on F with 4'(0,1) = 1 can be represented in this way and that such a
representation is unique.
Let us return to 0 E O(F) and observe that p(o)-1o is a linear automor-
phism of F which fixes the isotropic vector (0,1). It follows that for a given e E E
we can find a uniquely determined vector 0"(e) in E such that
ev(o(f),e) = µ(o) ev(f,o'(e)) ;fEF
Variation of e E E defines a transformation a :E-E. It follows at once that
(Qr) = T-Q , t- = 6 ; o,r E 0(F)
We conclude that o' is invertible and that the transformation i7 = 0'-1 satisfies
the required formula. In order to show that o is a similarity, we shall work our
way through three special cases.
10
A vector v E E gives rise to a transformation Ov E O(F) given by
Ov(z,O = (z,( - <v,z>) ; z E E, E 02
Direct verification shows that Ov(e) = e + v, e E E.

2° A real scalar A E Jr gives rise to a transformation sA E O(F) given by
sA(z,() = (z,aS) ; z E E, r; E R
Direct verification gives us s'A(e) = A e, e E E.
3° An orthogonal transformation o E O(E) defines v E 0(F) given by
v(z,C) = (o(z),C) ; z E E, S E R
Another direct verification gives us o- = 0. We leave it to the reader to show
that any orthogonal transformation of F can be decomposed into a product of
transformations of the three types we have just considered.
20 QUADRATIC FORMS
Normal vectors Let there be given an oriented affine hyperplane (H,P)

in Eucliden n-space E. By this we understand that H is an affine hyperplane in E
and P is one of the open half-spaces of E bounded by H. The normal vector to
(H,P) is the unit vector n E F which is zero along H and positive on P. Let us
observe that reflection rn is given by
rn(z,O = (z,() - 2<n,z>(n,v) ; n = (n,v)
while the corresponding transformation ?n is given by the formula
rn(x) = x - 2(<x,n> + v)n ;xEE
which is a reflection in the affine hyperplane H, compare 11.2.
Let us observe that v E OA(F) transforms the normal vector n for (H,P)
into the normal vector for (5(H),5(P)) as follows from the formula
ev(v(n),5(x)) = p(v)ev(n,x) ; xEE
Coxeter matrix Let us consider an affine simplex D, i.e. the convex hull
of n+1 points eo,el,...,en, not contained in an affine hyperplane. Let
no, nt,..., nn
be the corresponding normal vectors: the evaluation of ns against ej is zero, i # s,
while the evaluation against es is positive. From this description it follows that
the normal vectors form a basis for F = E ® R. Let us write
5.9 XOn0+)lnl (0,1)
and observe that the A is are strictly positive which follows from evaluation of
formula 5.9 on the vertices of the simplex. The Coxeter matrix for D is
<nr+ns>rs
Proposition 5.10 A basis no,nl...... nn of unit vectors for E ®R such that

the isotropic vector (0,1) can be written in the form
,Ono + atnt + ... + A,,n = (0,1) Ai > 0 , i = 0,...,n
defines a Euclidean simplex in E.
Proof For s = 0,...,n let £s denote the linear form on E (D R with Qns) = as
and {$(nr) = 0 for s # r. The relation above ensures that s has evaluation 1 on
the isotropic vector (0,1). It follows from the remark made in the beginning of the
proof of 5.7 that we can find a point es E E with ev(f,es) = s(f) for all f E F. The
points eo,...,en define the simplex we are looking for.
Proposition 5.11 Two simplices D and P in Euclidean space E are

similar if and only if they have the same Coxeter matrix.
Proof We have already seen that two similar simplices have the same Coxeter
matrices. With the notation above let m0,mi,...,mn be the normal vectors for the
simplex P enumerated such that
<nl.,n,> = <mr,ms> ; r, s = 0,1,...,n
Let v E O(F) be such that o (nl.) = o(mr), for r = 0,...,n. It remains to prove that
p(Q) > 0. To this end let us apply o- to the relation 5.9 to get that
)0m0 + Alm1 +... +.nrnn = (0,h(o,))
Let us evaluate this against an interior point p E P to get that
.Dev(m0,p) +... + µ(o,)
which shows that u(o) > 0 as required. 0
In the rest of this section we shall be concerned with the existence of

simplices with given Coxeter matrices (ars). For application to groups generated
by reflections, see [Coxeter2] and [Bourbaki2], we shall be concerned with acute
angled simplices, i.e. simplices whose matrices satisfies J ars < 0 for r
22 QUADRATIC FORMS
Indecomposable matrices 5.12 Let A E M"(R) be a positive sym-

metric matrix with del A=O, which has 1 s along the diagonal and satisfies
a,.s < 0 , r # s, r,s = 0.... ,n ; A = (ars)
If the matrix is indecomposable, in the sense that no non-trivial subset J of 0,...,n
exists such that
a,,,=0 forrEJands0J
then there exists a Euclidean n-simplex with A as Coxeter matrix.
Proof Let us consider a real vector space F equipped with a basis eo,...,e. and
a quadratic form such that
<er,es> = a,,, ; r,s = 0,...,n
By assumption the form is positive but singular. We shall eventually prove that F
is parabolic. To this end we shall show that for a non-zero isotropic vector
n = E xsas we have as # 0 for all s = 0,1,...,n. From the inequality
Q(E xses) = E r,sarsxrxs >_ E r,s arslxr1Ixsl = Q(E Ix5Ies)
it follows that we can assume that all coefficients of n are positive. Let us for a
moment assume that the set J = { s I xs > 0) is different from 0,...,n. Let us recall
from 5.2 that the isotropic vector n is orthogonal to any vector:
0 = <n,es> = E r E J xr<el.,es> ; s = 0.... ,n
In particular we conclude that a1. = <er,es> = 0 for s 0 J, contradicting the
hypothesis that the matrix A is indecomposable.
Let us prove that F l has dimension 1. To this end let us introduce the
space N C R"+1 representing F 1
N = {(xo,...,x") E R"+1 I Q( Exses) = 0}
By the observations made above we conclude that the intersection between N and
the hyperplane x0 = 0 is zero; this implies dim(N) = 1. In conclusion if
Exses 0 is isotropic then all coefficients are non-zero of the same sign. Finally
we can use an isometry of F with E ® R to construct a basis for E ® R and apply
5.10.
1.6 LORENTZ GROUP
Let us consider a vector space F of dimension n+1 over R equipped with a

quadratic form of Sylvester type (-n,l). We shall be concerned with the action of
0(F) on the hvperboloid or pseudosphere
6.1 S(F)={fEFI <ff>=1}
Let us start our investigations by a Cauchy-Schwarz type inequality.
Lemma 6.2 Two points x and u of the hyperboloid satisfy the inequality
1 < I<x,u>I ; x,u E S(F)
Moreover, the inequality is sharp when x and u are linearly independent in F.
Proof When x and u are linearly dependent we have x = u or x = -u and the

inequality is trivial. When x and u are linearly independent, the plane R they
span is of Sylvester type (-1,1), (-1,0) or (-2,0), compare 6.3. From the fact
that R contains vectors of strictly positive norm it follows that the type is (-1,1).
The discriminant lemma 6.3 gives us <u,u><x,x> < <u,x>2 as required. 0
Discriminant lemma 6.3 Let R be a plane in F generated by two

vectors e and f. The Sylvester type of R is determined by the discriminant
A = <e,e> <f,f> - <e,f>2
according to the following table
0<0 A>0
(-2,0)
Proof Let us first prove that the plane R contains a vector of norm -1. Use
any orthonormal basis for F to exhibit a hyperplane E of Sylvester type (-n,0). It
follows from the dimension formula 3.4 that,
24 QUADRATIC FORMS
dim(R) + dim(E) = dim(R fl E) + dim(R + E)

Using dim(R + E) < n+1 we get that dim(R fl E) > 1. Thus R contains vectors
of strictly negative norm. From this we conclude that the only possible Sylvester
types for R are (-1,1), (-1,0) and (-2,0). It follows from the discriminant
inequality 3.5 that these three cases correspond to A < 0, A=O and A > 0
respectively. 0
A linear hyperplane in F must be of type (-n+1,1), (-n,0) or (-n+1,0)

as it follows from the argument used in the previous proof.
Proposition 6.4 The hyperboloid S(F) has two connected components.

Points x,u E S(F) are in the same connected component if and only if <x,u> > 0.
Proof Let us fix the point u E S(F) and let E denote the hyperplane orthogo-
nal to u. Since E has type (-n,0) we find that E separates S(F) into two parts
H={xES(F)I <x,u> > 0}, H-={xES(F)I<x,u><0}
These two parts are interchanged by reflection along u. We are going to prove
that stereo ranhic projection g:H-+E with centre -u maps H homeomorphically
onto the open unit disc
D={yEEI-<y,y><1}
The affine line through x E H and -u is parametrised (x + u)t - u, t E R. Its inter-
section g(x) with E is given by <(x + u)t - u,u> = 0 , which gives us
6.5 x - <x,u>u
g (x) = 1 + <x,u>
<g(x),g(x)> = 1 -+<x,u>
1
<x,u>
From 6.2 we conclude that <x,u> > 1, so the formula above enables us to
conclude that g(x) E D as required. We ask the reader to work out that the inter-
section between H and the affine line through the points z E D and -u is given by
6.6 f(z) = 2
1+z <z,z>
+u
-u ; zED
These investigations show that f:D-+H is a homeomorphism.
Definition 6.7 By a Lorentz transformation4 of F we understand an

orthogonal transformation o E 0(F) which preserves the connected components
of the pseudosphere. The group of Lorentz transformations is called the Lorentz
rou and is denoted Lor(F).
It follows from 6.4 that a Lorentz transformation can be recognised by the

inequality
6.8 <x,o(x)> > 0 ; o E Lor(F), X E S(F)
The antipodal map x'-+-x is in the centre of O(F) but is not a Lorentz
transformation. It follows that Lor(F) has index 2 in O(F), more precisely
6.9 O(F) = Lor(F)x7L/(2)
The basic example of a Lorentz transformation is provided by reflection along a

vector c E F with <c,c> = -1
rc(x) = x+ 2<x,c>c ;xEF
Let us use 6.4 to check that this is a Lorentz transformation:
<rc(x),x> = 1 + 2<x,c> 2 ; x E S(F)
Lorentz transformations with determinant 1 are called even or special Lorentz
4 In literature with reference to physics such a transformation is called an

orthochroneous Lorentz transformation.
26 QUADRATIC FORMS
transformations. These transformations make up the special Lorentz rou

Lor+(F), which is a subgroup of Lor(F) of index 2.
Let us make a detailed examination of a Lorentz transformation a when
the space F has dimension 2. Pick an orthonormal basis e,f for F with <e,e> = 1
and <f,f> _ - 1. The point xe + tf lies on the hyperbola S(F) if and only if
x2 - t2 = 1. The branch of the hyperbola containing (1,0) can be parametrised
(coshs, sinks), s E R. Thus we can write a(e) = e coshs+fsinhs
cosh s, sinh s)
The vector has norm -1 and is orthogonal to or(e), so we conclude that

a(f) = (e sinks + f coshs). In the first case the matrix for o, is
coshs -sinks
sinks -coslcs
It follows that det(er) = -1 and lr(o-) = 0. This implies that a has eigenvalues 1
and -1. We ask the reader to verify that the basis e,f can be picked such that the
matrix for or corresponds to the case s = 0. This shows that a is a reflection along
a vector of norm -1. In the second case the matrix for o is
coshs sinks
L(s)_ ;sER
sinks coshs
From the addition formulas for the hyperbolic trigonometric functions it follows
that L(s) is a morphism of groups
6.10 L(s + t) = L(s) L(t) ; s,t E R

This shows that the special Lorentz group is isomorphic to R when dim(F) = 2.
Theorem 6.11 Let F be a vector space of dimension n+1 equipped with

a quadratic form of Sylvester type (-ii,l). Any Lorentz transformation o of F is
the product of at most n+1 reflections of the type rc where c E F with
<c,c> = -1.
Proof We shall proceed by induction on n. If n = 1 the result follows from

our investigation above. If n > 2 let us pick a hyperplane E of F of type (-n,0)
and focus on the map E F given by
eEE
If this map is not injective we can find a vector e with norm -1 fixed by o. The
space D = e 1 has Sylvester type (-n+1,1) and the restriction of o to D is a
Lorentz transformation as it follows from 6.8
If the map a i--+ o(e) - e, e E E, is injective, then the image space has
dimension n-1 > 2 and must have non-zero intersection with the hyperplane E. It
follows that the image space contains a vector c say of norm -1. Let us write
c = o(e) - e ; e E F, <e,e> < 0
Since o(e) and e have the same norm we conclude from the proof of 2.3 that
reflection r, along c interchanges e and o(e). It follows that rca fixes the vector e
of strictly negative norm. Apply the induction hypothesis to the space e -- to
conclude the proof.
Let us turn to the action of the Lorentz group on the isotropic cone C(F),
i.e. the set of vectors of norm 0. The multiplicative group IR* acts on C(F)-0; the
orbit space is called the proiectivised cone of F and is denoted PC(F). We may of
course think of PC(F) as the set of isotropic lines in F.
28 QUADRATIC FORMS
Proposition 6.12 Let F be a vector space of dimension n+1 equipped

with a quadratic form of Sylvester type (-n,l). The projectivised cone PC(F) is
homeomorphic to the sphere Sn-1
Proof Let us return to the proof of 6.3 and introduce the unit sphere Sn"1
= 8D
in the hyperplane E orthogonal to the base vector u. We shall make use of the
map 4:E-*F given by
6.13 m(z) = 2z + u - <z,z>u ;zEE
Direct calculation using that <u,z> = 0 gives us

<5(z),c(z)> = (1 + <z,z>)2 ;zEE
which shows that 0 induces a map D:8U,PC(F). Observe that u and z E aU
span a plane of Sylvester type (-1,1) and that q(z) and 0(-z) are two linearly
independent isotropic vectors in that plane. Observe also that this plane is
generated by ¢(z) and u. From these remarks it follows that 4D is injective. In
order to see that I is surjective note that an isotropic line L in F together with the
vector u span a plane R of type (-1,1): pick a non-zero vector v E L and let us
examine the sign of the discriminant
<u,u><v,v> - <u,v>2 = - <u,v>2 < 0
The intersection E n R is a line generated by a vector z say of norm -1. It follows
that L is generated by one of the two vectors O(z) and c5(-z). 0
Corollary 6.14 When dim(F) > 3, the Lorentz group Lor(F) acts
faithfully on PC(F).
Proof Let a be a Lorentz transformation of F which acts trivially on PC(F).

We shall analyse the problem in terms of the map': Sn-1- PC(F) introduced in
S°-1
the proof of 6.13. For z E let A (z) be the eigenvalue belonging to the isotropic
vector O(z). The function )t:Sn"1,R is continuous which follows from the formula
«¢(z),u> = \(z)<di(z),u> = 2)(z) ; z E Sn"1
The set of values of the function A is finite simply because it is a subset of the set
of eigenvalues of o. Since dim(F) = n+1 > 3, we can use the connectedness of S' 1
to conclude that A is constant.
Let us prove that F is generated by C(F). To this end we shall prove
that any vector f E F with non-zero norm is contained in a hyperbolic plane. If
<f,f> < 0, then f 1 is of type (-n+1,1), so we can select e orthogonal to f with
<e,e> > 0. When <f,f> > 0 the space f 1 has type (-n,0) so we will have no
trouble finding e orthogonal to f with <e,e> < 0. Now, observe that a hyperbolic
plane is generated by two isotropic vectors.
Let us observe that multiplication by a scalar A is not an orthogonal
transformation of F unless A = 1 or A = -1. We have already observed that
multiplication by -1 interchanges the two sheets of the hyperboloid S(F).
Proposition 6.15 The truncated isotropic cone C*(F) = C(F) - {0} has
two connected components provided that dim(F) > 3. The connected components
are preserved by the Lorentz group Lor(F).
Proof Let us fix a point u E S(F) and observe that C*(F) is the union of
C+(F)={ceFI <c,u> > 0 } , C (F)={ceFI <c,u><0)
It follows from the proof of the previous proposition that these two sets are the con-
nected components of C*(F).
Let us prove that the two connected components are unchanged if we
replace the point of reference u with an other point v of the same sheet of the
hyperboloid S(F). To this end let us focus on a point c E C*(F) and observe that
the function xH<x,c>, omits the value zero since the hyperplane
orthogonal to c has type (-n+1,0), compare the remark made after 6.4. It follows
that <u,c> and <v,c> have the same sign.
From the fact that o(u) and u belong to the same sheet of S(F) it follows
that <o(x),u> and <o(x),o(u)> have the same sign for x E C*(F). In particular
for x E C+(F) we have <o(x),o(u)> = <x,u> > 0, and we conclude that <o(x),u>
is positive; thus o(x) E C+(F).
30 QUADRATIC FORMS
Proposition 6.16 Let L denote an isotropic line in F, and let LorL(F)

denote the group of Lorentz transformations stabilising L. Restriction from F to
L l defines an isomorphism
LorL(F) -: 0,,(L 1 )
where 0,,(L 1) denotes the subgroup of 0(L 1) made up of orthogonal
transformations with positive eigenvalue on the isotropic line L, compare 5.5.
Proof Let OL(F) denote the subgroup of O(F) made up of transformations

which stabilises the line L. Let us prove that restriction defines an isomorphism
OL(F) =: 0(L l )
First, restriction is surjective as it follows from Witt's theorem 2.4. - Let us

analyse a o E OL(F) which restricts to the identity on L 1 . To this end we
introduce an n-1 dimensional subspace E of L 1 of type (-n+1,0). From the fact
that o has trivial restriction to E follows that we can write o = i ® a where a is
an orthogonal transformation of E 1 . The plane E 1 has type (-1,1) and can be
decomposed E 1 = L ® K where K is the second isotropic line in E -L . The
transformation a preserves K and L and the eigenvalues must have the form A,A-1,
A E R*, since the inner product is preserved. Remembering that o has trivial
restriction to L 1 we conclude first that A = I and second that p = e as required.
The final statement of the proposition is a consequence of 6.15 in case dim(F) > 3.
In case dim(F) = 2 the result follows from the investigations a few lines above.
Proposition 6.17 The special Lorentz group Lor+(F) acts transitively on

the space N(F) of vectors of norm -1, provided dim(F) > 3.
Proof The full orthogonal group O(F) acts transitively on N(F) as a

consequence of Witt's theorem 2.4. To see that Lor(F) acts transitively it suffices
to prove that the stabiliser in O(F) of a given N E N(F) is not contained in the
Lorentz group. As an example we take the transformation which fixes N but
transform xH-x on the orthogonal complement of N in F. We ask the reader to
check that the stabiliser of N in Lor(F) is not contained in Lor+(F).
The projectivised cone PC(F) is a subset of projective snare P(F), that

i.e. the set of all lines in F. Let PS(F) denote the subset of P(F) consisting of lines
in F generated by vectors of strictly positive norm. In the natural topology of
P(F) we have 8PS(F) = PC(F).
Let Hn denote any one of the two sheets of the hyperbola S(F). We have
a natural homeomorphism Hn_PS(F) which assigns to a point X E Hn the line L
in F generated by the vector X. Using this identification we can write
6.18 all- = PC(F)

32 QUADRATIC FORMS
1.7 MOBIUS TRANSFORMATIONS
Let E denote a Euclidean vector space of dimension n. A sphere C with

centre c and radius r > 0 gives rise to an involution o of E-{c} given by
7.1 o(x) = c + r2 x - c ; x E E - {c}

Ix-c12
The transformation o is called inversion in the sphere C. It is apparent from the

formula that o(x) lies on the ray through x emanating from c. The sphere C can
be recovered as the fixed point set for o.
In the following we shall consider the 1-point compactification E of E,
obtained from E by adding an extra point oo. Reflection o in the sphere C can be
extended to an involution & of t which interchanges a_ and the centre c of C. It is
easy to check directly that & is continuous.
Let us consider an affine hyperplane H in E through the point u E E.
Using a unit normal vector n E E for H we can describe write Euclidean reflection
in H by the formula
7.2 \(x) = x- 2<x - u , n>n ; xEE
We leave it to the reader to extend this to a continuous transformation A of E,

which preserves the point at infinity.
In order to unify the treatment let us introduce the word sphere C in E for
a Euclidean sphere in E or a subset of t of the form C = {oo} U H where H is an
affine hyperplane in E. The two types of transformation of t we have introduced
are called inversions in spheres in E. The subgroup of the group of homeo-
morphisms of t generated by inversions in spheres in t is called the M6bius group
of E and is denoted Mob(E).
The main purpose of this section is to identify the Mobius group with a
Lorentz group. To this end we introduce the auxiliary space E ®R2 equipped
with the bilinear form
7.3 <(e,a,b),(f,c,d)> <e,f> + 2(ad + bc) ; e,f E E, a,b,c,d E R
It is easily seen that this space has Sylvester type (-n-1,1). Let us consider the
transformation t: E-> C(E + R2) given by
7.4 Le = (e,<e,e>,1) ; eEE
This induces a map from E into PC(E+922) which identifies E with the com-
plement of the point of PC(E+R2) representing the line through the point (0,1,0).
Let us extend this to a bijection t: E-+PC(E + R2), where L(oo) is the line through
the point (0,1,0).
Theorem 7.5 The action of the Lorentz group Lor(E (D922) on the
projectivised cone PC(E + R2) and the homeomorphism a E - + PC(E (D R2)
identifies the Mobius group M"ob(E) with the Lorentz group Lor(E+R2).
Proof Let us return to the sphere C from 7.1 and form the vector
C = (c,<c,c> -r2,1)
We ask the reader to verify the formulas
<C,C> = - r2 , )<C,tx> = Ix - c12 - r2 xEE
From this we can deduce a commutative square
E PC(E(D R2)
& [ 7C
E PC(E(D R2)
More precisely we ask the reader to verify the formula

a t (x) = r2lx - cl-2 rC L (x) x E E - {c}
Next, we return to the reflection A from 7.2 and introduce the vector
N = (n,2<n,u>,0)
From this we can deduce a commutative square
34 QUADRATIC FORMS
E PC(E (D R 2)
la 1 TN
E PC(E ® 082)
More precisely, we ask the reader to verify that t rN = A t.

According to 6.11 we can write any Lorentz transformation as a product of
transformations of the form rN where N E E ® R2 with <N,N> < 0. Let us
observe that the locus
{xEEI <tx,N>=0}
is a sphere or an affine hyperplane: Let us write N = (e,a,b); upon multiplying N
by a non-zero scalar we may assume that b = 1 or b = 0.
Case N = (e,a,1). Let us introduce the number r = - <N,N> and
observe that a = <e,e> - r2 to get the formula
2<tx,N> = - 2<x,e> + (a + <x,x>) = Ix - e12 - r2
which shows that the locus is the circle with centre e and radius r.
Case N=(e,a,0). We have <N,N> = - <e,e>, in particular e # 0, so we
can multiply N by a constant to obtain that e is a unit vector. Choose u E E such
that <u,e> = 2a, and observe that
<tx,N> <x,e> + 2a = <u - x,e>
which shows that the locus is the affine hyperplane through u with normal vector
e. At this point it remains only to quote 6.14.
Corollary 7.6 A Mobius transformation a of the Euclidean space of

dimension n can be written as a product of at most n+2 inversions.
Proof Any Lorentz transformation of the space E ® 082 can be written as a

product of at most n+2 reflections, compare 6.11.
Let us say that a Mobius transformation a of E is odd or even depending

on the sign of the determinant of the corresponding Lorentz transformation. An
even Mobius transformation can be expressed as a product of an even number of

inversions, while an odd Mobius transformations is a product of an odd number of
inversions.
Corollary 7.7 The restriction to E of the group of Mobius transformations

of k which fixes oo is the full group Siml(E) of Euclidean similarities of E.
Proof Let us first check through three special cases. First observe that a A>0
gives rise to a Lorentz transformation
(x,a,b)'--' (x,Aa,A-ib)
which induces the dilatation of E. Next observe that a o E O(E) is induced
on E by the Lorentz transformation
(x,a,b) '-4 (o(x),a,b)
Let us finally observe that a vector v E E gives rise to the Lorentz transformation
(x,a,b) '- (x + by, a + b<v,v> + 2<x,v>, b)
which induces translation x F x + v on E. This shows that any similarity of E is
a Mobius transformation. In order to prove the converse, we must show that these
three types of transformation generate the group LorL(E ® 082) consisting of
Lorentz transformations which stabilise the line L generated by (0,1,0).
To this end we shall use the map (z,O -- (-z,2(,0) to identify E ®08 with
the subspace L -L of E ® R2. This allows us to express the evaluation map from
1.5 in terms of the inner product on E ® 082.
7.8 ev((z,O,x) = <(-z,2(,0) , ex> ; (z,() E E (D 08, x E E
We ask the reader to complete the proof by means of 6.16 and 5.8. 0
DISCS Recall also that N = N(E ® 082) denotes the set of vectors in E ® 082 of
norm -1. A vector U E N defines a sphere f = {x E E I <tx,U> = 0} and decom-
poses the complement of Y into the union of two connected open subsets
36 QUADRATIC FORMS
`.b={xEEI<tx,U> > 0}, `b ={xEEI<tx,U><0}

which are called the discs bounding the sphere 1. The set 5 is called the disc with
normal vector U.
Let us make a slight refinement of our earlier concepts by focusing on the
connected component of the truncated isotropic cone C(E (D 022) made up of points
whose inner product with (0,1,1) is positive:
7.9 C+(E e R2) = { (A,a,b) I - <A,A> + ab = 0, a + b > 0 }
Observe that the map t: M C factors through C+. It is useful to reinterpret the
projectivised cone as
PC(E ® R2) = C+(E e R2)/ R+ ; R+= ]0,+o0[
Corollary 7.10 A Mobius transformation v E M'ob(E) maps a sphere Y

into a sphere v(Y). Inversion in the sphere v(1) is given by vov"1 where o is
inversion in the sphere f.
Proof Let us agree to denote by v the Lorentz transformation of E ® R2

corresponding to v by theorem 7.5. Let U be the normal vector for one of the
discs ll bounding the sphere 1, we ask the reader to identify the set v(`.D) with the
disc with normal vector v(U). Observe that v = rU and use the general formula
rp(U) = vr0-1
to conclude that voi 1 is reflection in the sphere v(s).
Corollary 7.11 The Mobius group acts transitively on the set of

discs in E.
Proof We can identify the action of M'ob(E) on the set of discs in E with the
action of the Lorentz group Lor(E (DR2) on the space N(E (D R2) of vectors of norm
-1. Once this is done, the result follows from 6.16.
Corollary 7.12 Let 9G be a sphere in E. A Mobius transformation o which

fixes 9G pointwise is either the identity or inversion in JG (dim(E) > 2) .
Proof Let K be a normal vector for one of the discs `D bounding 9G and let k
denote inversion in the sphere 9G. From o(9G) = 9G we conclude that o(`D) = `D or
o(S) = tc(s). Upon replacing o by tco we may assume that o(`D) = `D or otherwise
expressed o(K) = K. Let us observe that restriction of v to F = K 1 acts trivially
on the PC(F); it follows from 6.14 that the restriction of a to F is t, the identity.
This gives O = t, which implies that o = t.
Definition 7.13 We say that two spheres I and T are orthogonal if the
normal vectors for the bounding discs are orthogonal5.
Proposition 7.14 Let o denote inversion in the sphere Y and r inversion

in the sphere 4T. If Y $ T then the following conditions are equivalent
10 1 and T are orthogonal
20 or = ro
30 r(Y) _ 1
40 o(C)_CT
Proof It is a consequence of 7.10 that the last three conditions are equivalent.
To proceed, introduce normal vectors H and K for two bounding spheres. The
correspondance 7.7 identifies the Mobius inversion r with the Lorentz reflection rK
along the vector K. From the formula
7K(H) = H + 2<H,K>K
we conclude that rKK(H) = ± H if and only if <H,K> = 0. This shows that
condition 1° is equivalent to condition 3°.
5 The geometric meaning of this concept will be given in next section.

38 QUADRATIC FORMS
Proposition 7.15 Let A and B be points of t conjugated with respect to

the sphere I (by this we mean that A and B are points outside Y interchanged by
inversion in Y). Any sphere through A and B is orthogonal to Y.
Proof Let a generate the isotropic line in E ® R2 represented by A and let n

denote a normal vector for T. The vector b = a+ 2<n,a>n generates the isotropic
line represented by B. Since A 0 Y we have <n,a> $ 0 and the formula for b
shows that n belongs to the plane spanned by a and b. It follows that the normal
vector for any sphere through A and B is orthogonal to n.
Proposition 7.16 Let there be given an open disc `h in E. The group

Mob(9) of Mobius transformations of E which leaves `D invariant is generated by
inversions in spheres orthogonal to Y = c99.
Proof Let N denote the normal vector for `D. A Mobius transformation
leaves `D invariant if and only if the corresponding Lorentz transformation fixes
N. It follows that we may identify M"ob(`D) with the Lorentz group of N 1 . This
group is generated by reflections rI< where <K,K> = -1 and <K,N> = 0.
Corollary 7.17 Let L denote a linear hyperplane in E and `ll one of the
half-spaces in E bounded by H. Restriction from k to L defines an isomorphism
Mob(9) M"ob(L)
Proof Specialise the proof of 7.16 to this case.
The composite of the inverse of 7.17 and the inclusion Mob(`D)-+M'ob(E) is

known as Poincare extension : M'ob(L)- M'ob(E).
1.8 INVERSIVE PRODUCT OF SPHERES 39
1.8 INVERSIVE PRODUCT OF SPHERES
Let E denote a Euclidean vector space of dimension n > 2. We shall make

a closer study of the intersection of two spheres I and T in E. This will be done
through a numerical character Y* J . Let us represent the spheres by their standard
equations
Y: - 2<x,f> + b<x,x> + a = 0 ; - <f,f> + ab < 0

8.1 `T: - 2<x,g> + d<x,x> + c = 0 ; - <g,g> + cd < 0
The inversive product of I and `T is given by
I <f,g> -lad - !be

8.2
<f,f> - ab <g,g> - cd
From the investigations made in the previous section it follows that the inversive
product of I and T is preserved under Mobius transformations:
8.3 o(Y)*o(T) = oj*gj ; o E M"ob(E)
The relative positions of two distinct spheres S and T can be read off T*`J :
Y*`T<1 Y*1f = 1 Y*T>1

8.4
#(inT)>1 #(inT)=1 #(in-f)=o
Proof Let H and K be normal vectors for discs bounding Y and T. These two
vectors span a plane R, and we can make the identification
InGT=PC(R1)
The geometry is ruled by the discriminant
0 = <H,H> <K,K> - <H,K>2 = 1 - (Y* r)2
From the discriminant lemma 6.3 we deduce that
t*T < 1 : 0 < 0 , the plane R has Sylvester type (-2,0) which implies that
40 QUADRATIC FORMS
R 1 has Sylvester type (-n+l,l). Thus PC(R 1) consists of at least two points.
Y*°J = 1 : A = 1 , the plane R has Sylvester type (-1,0) which implies that
R 1 has Sylvester type (-n,0). Thus #PC(R 1) = 1.
Y*°J > 1 : A < 0 , the plane R has Sylvester type (-1,0) which implies that
R 1 has Sylvester type (-n,0). Thus PC(R 1) = 0.
Proposition 8.5 The Mobius group M''b(E) acts transitively on pairs of

spheres (Y,r) with given inversive products.
Proof This follows easily from Witt's theorem 2.4.
Let us investigate an ordinary point x E E of intersection of the spheres Y

and IT, x E Y fl of. To this end we pick a disc 9 bounding Y and let nx(`D) E E
denote the outward pointing unit normal to Y at x. Similarly, let S be a disc boun-
ding T and let n.(S) E E denote the outward pointing unit normal to `J at x. We
have that
8.6 Y*T = j<nx(%,nx(9)>j
We shall prove more precisely that the inner product of the normal vectors
N('D),N(g) E N(E (D R2) is given by
8.7 <N(`D),N(g)> <nx(`D),nx(g)>
Proof Let us suppose that Y is an ordinary sphere in E and that `D is its

interior. In the equation 8.1 for Y we can take b = 1 which gives us that f is the
centre for Y, while the radius r is given by r2 = <f,f> - a. Thus
nx(`D) = r-1 (x - f) , N(s) = - r-1(f,a,1)
When `J is an ordinary sphere with radius s and centre g we can write
nx(S) = s '(x-g) , N(9) = -s't(g,c,1)
where c = <g,g> - s2. Direct calculation gives us

<N(5),N(9)> = r 1s1(- <f,g> + za +11c)
Then using 8.1 with b = 1 and d = 1 to eliminate a and c we get that
<N(°D),N(g)> = r 1s 1(<x,f> - <f,g> + <x,g> - <x,x>) = - <n.(''i),nx(g)>
When Gl is an affine hyperplane through x with nx(g) = n we have that
nx(g) = n , N(S) = (n,2<x,n>,O)
Maintaining that I is an ordinary sphere as above, we get that
<N(9),N(6)> = r l( - <f,n> + <x,n>) = - <n.(`D),n.(S)>
The case where both Y and T are affine hyperplanes is now obvious.
In the rest of this section we shall limit ourselves to the two-dimensional

case. First a rather amusing example due to Jacob Steiner.
Steiner's alternatives 8.8 Let A and % be circles in the Euclidean

plane of which °.B contains A in its interior. It is possible to find a kissing chain of
n circles each touching A on the outside and `.B on the inside, if and only if
A*% = 1 + 2 tan 2 n
Proof According to 8.5 we may assume that the circles A and `.B are concentric
with radii a and b, a < b. The condition for the existence of a chain of kissing
circles can easily be worked out by Euclidean trigonometry. The result is
0
sin 11= 1+B ' -b
42 QUADRATIC FORMS
From this formula we deduce without difficulty that
a - 1-sin il
b 1+sin n
On the other hand we find by direct computation
.A*`B=2(b+a)
A simple straightforward calculation concludes the proof. 0
Definition 8.9 By a of circles in the Euclidean plane E we under-

stand a two-dimensional subspace R of E R2. The actual circles in the pencil are
the circles in k with normal vector in R.
Proposition 8.10 Let R be a pencil of circles in E. The set of circles in

the pencil can be described as follows, depending on the Sylvester type
(-2,0) : circles through two given distinct points P and Q
(-1,0) : circles through a given point P orthogonal to a given circle N through P.
(-1,1) circles which conjugate two given distinct points P and Q.
Proof Let us check through the three possible Sylvester types.

Case (-2,0). The orthogonal space R 1 has type (-1,1) and contains
precisely two isotropic lines P and Q say. We can recover R as the set of vectors
orthogonal to P and Q. In geometric terms, a circle Y belongs to the pencil if and
only if it passes through P and Q, compare the proof of 7.15.
Case (-1,1). The space R contains exactly two isotropic lines. Let us
observe that a circle I with normal vector N conjugates P and Q if and only if
N E R. Alternatively, let us treat the case P = oo and Q = 0 directly. The circles
in k which conjugate these two points are the ordinary circles in E with centre 0.
Notice that through each point of t - {0,oo} passes a unique circle of this pencil.
Case (-1,0). Let us observe that the dual pencil R 1 has the same type
and that R fl R 1 = P is an isotropic line. Pick a circle N in the pencil R 1 and
observe that the circles in R are precisely the circles through P, orthogonal to N.
Example 8.11 Consider the following families of circles in R2

x2+y2-b+2Ax=0 ;AER
10 b > 0 . All circles pass through (0,4) and (0,-4g)
2° b = 0 . The equation for the circle can be written
(x-A)2+y2=A2
which shows that all circles pass through (0,0) and are tangent to the y-axis at this
point.
3° b < 0 . Put b = - k2 and rewrite the equation as follows
(x-A)2+y2 = A2-k2
For A = k , the point (k,0) is the only solution, and similarly A = -k gives (-k,0).
The system is, in fact, orthogonal to the pencil of circles through (k,0) and (-k,0).
44 QUADRATIC FORMS
1.9 RIEMANN SPHERE
Let us introduce the Riemann sphere t as the 1-point compactification of

of the complex plane C. This compactification can be realised in a number of
ways, but we shall use a construction from projective geometry which displays the
underlying algebra in a natural way.
Consider the scalar action of the complex multiplicative group C* on the
complex vector space C x C, and let t denote the orbit space for the induced action
on the complement of the origin. The equivalence class of a pair of complex
numbers (z,w) $ (0,0) is denoted [ w J. Thus
za
; aEC*
wa
1=
The group G12(C) acts on the Riemann sphere in virtue of the formula
a b z az+bw
9.1 ;ad-bc#0
c d w cz+dw
Let us remark that scalar matrices act trivially on C, thus we are dealing with an
action of PG12(C) = G12(C)/C* on C. In order to calculate the action of the
inverse matrix S-1 we can use the cofactor matrix S' given by
9.2 a b - d -b
c d -c a
This is justified by the well known formula S S" = S" S = t del S.
Proposition 9.3 The action of the group PG12(C) on t is triply transi-

tive: for any two triples A,B,C and P,Q,R, each consisting of three distinct points,
there is a unique v E PG12(C) which transforms the first triple into the second.
Proof Let the points A,B,C be represented by the vectors a, b, c in C x C. By

assumption, the first two vectors are linearly independent so we can write
c=aA+bit ; A,pEC
Since A and p are different from zero we can change a and b such that c = a + b.
Quite similarly, we can represent the points P,Q,R by vectors p, q, p + q. The
linear automorphism o of C x C with o(a) = p and o(b) = q induces the
required Mobius transformation.
It remains to investigate a linear automorphism r of C x C which induces
the identity on C. By assumption, all non-zero vectors of C x C are eigenvectors
for r. It follows easily that r is multiplication by a scalar it E C*. 0
Let us introduce the point at infinity oo = [ ] and observe that we can

o
identify the complement C - {oo} with C once we identify the complex number z
with the point [ i l of C. Let us investigate the transformation o from 9.1 in this
terminology. When o(oo) = oo we have c = 0. If c t_ 0 we find that o(oo) = and
c
o l(oo) _ -d To recapitulate
u(00) 0.(L) =
9.4 0 - ( _A )
_
00
,
cz + b
zEC,z#-a
Theorem 9.5 The group G12(C) acts on t as even Mobius transfor-

mations. More precisely, this action induces an isomorphism of groups
PG12(C) M1ob+(C)
Proof Let us first observe that the group of upper triangular matrices acts on
C fixing oo and induces the group of even similarities of C. In general, the matrix
identity (c # 0)
a
C
c d 0 1
leads us to consider the transformation z-z 1 of C. This transformation is the

46 QUADRATIC FORMS
composite of inversion in the unit circle and reflection z- in the x-axis as follows
from the basic formula
Z1 _ZIZI-2 ;zEC*
This concludes the proof of the fact that G12(C) acts through even Mobius trans-
formations. In order to conclude the proof, observe that PG12(C) acts transitively
on t and that the group of even Mobius transformations fixing oo induces the
group of even similarities on C, compare 7.7. 0
Cross ratio We shall introduce a fundamental invariant of four distinct

points (P,Q,R,S) of the Riemann sphere C. To this end let us represent each of
the four points by 2 x 1 matrices p, q, r, s and define the cross ratio as the point
of C given by
9.6 [P,Q,R,S] - rl det[p,s]

del[p,r] del[q,s] 1
det[q,r] ]
where the symbol [p,q] denotes the 2 x 2 matrix with first column p and second
column q. For a Mobius transformation v we have
del [o(p),o(q)] = del(a) del [p,q]
From this it follows that the cross ratio is invariant under Mobius transformations
9.7 [o-(P),a(Q),a(R),o(S)] = [P,Q,R,S] ; a E G12(C)
At this point let us remark that the cross ratio [P,Q,R,S] is well defined as long as
P,Q,R,S represent at least three distinct points of C. This allows us to fix three
distinct points P,Q,R of C and make a free variation of the fourth point S. The
result is a Mobius transformation:
Proposition 9.8 Let P,Q,R denote three distinct points of C. The map
S H [P,Q,R,S] ; SEC
is an even Mobius transformation, which transforms P,Q,R into 00,0,1.
Proof Let us rewrite formula 9.6 as follows
-g2det[p,r] gldel[p,r] st
[P,Q,R,S] = I
-p2det[q,r] pl det[q,r] S2
I
Observe that the 2 x 2 matrix has determinant det[p,r] det[q,r] det[p,q] $ 0. Direct
calculation reveals that this transformation maps P,Q,R into oo,0,1.
Proposition 9.8 in connection with 9.3 offers an implicit definition of the

cross product as the even Mobius transformation which maps P,Q,R into 00,0,1.
In particular we find that
9.9 [00,0,1,S]=S ; SEC
Proposition 9.10 Two 4-touples of distinct points of the Riemann

sphere P,Q,R,S and X,Y,Z,W can be transformed into another by an even Mobius
transformation if and only if
[P,Q,R,S] = [X,Y,Z,W]
Proof It follows from 9.7 that the condition is necessary. Conversely, if the
two cross ratios are identical, let o and r be the Mobius transformations given by
o(P,Q,R) = (oo,0,1) , r(X,Y,Z) = (c ,0,1)
According to formulas 9.7 and 9.9 we get that
[P,Q,R,S] = [oo,0,l,o(S)] = o(S)
and similarly, [X,Y,Z,W] = r(W). Thus the Mobius transformation 7-1a maps S
to W as required.
Proposition 9.11 An odd Mobius transformation Q E Mob(C) satisfies

[o(P),o(Q),o(R),o(S)] = [P,Q,R,S]
whenever P,Q,R,S represent four distinct points of C.
48 QUADRATIC FORMS
Proof In order to generate the full Mobius group we need PG12(C) and one
single odd Mobius transformation, for example complex conjugation
[wz , F'[ w]
The formula is easily verified for complex conjugation.
Let us record two symmetry properties of the cross ratio. The verification
is immediate and is left to the reader.
9.12 [P,Q,R,S] = [R,S,P,Q] , [P,Q,R,S] = [Q,P,S,R]
Proposition 9.13 Any a E PG12(C) is conjugated to a transformation

given by a matrix of the following form
1 1
; a E C*
0 1
I
Proof Let a be given by the matrix S E G12(C). Note that a non-zero vector
V E C X C is an eigenvector for S if and only if [v] E C is a fixed point for a. From
this we conclude that a $ e has one or two fixed points on C.
When a has two distinct fixed points A and B, let r be an even Mobius
transformation with r(oo) = A and r(0) = B. The transformation 7--1 ar will fix 0
and oo. It follows that a representing matrix must be diagonal. After multipli-
cation by a complex number we get a matrix of the first type. When a has only
one fixed point A, choose an auxiliary point C # A and let r be the even Mobius
transformation given by
r(0) = C , r(1) = o(C)
r(oo) = A ,
Observe that the transformation rlar fixes oo and maps 0 to 1. It follows that
this transformation can be represented by a matrix of the form
a 1
; aEC*
0 1
But the matrix can only have one eigenvalue so we must have a = 1. 0
Let us introduce an important invariant of a matrix S E G12(C)
9.14 +2c- (trS)2

(let S
; S E G12(C)
Observe that lr2 S depends only on the conjugacy class of S in PG12(C). In parti-
cular, we may introduce the invariant 1r2o of an even Mobius transformation o.
Proposition 9.15 Let o and r be even Mobius transformations, both

different from t. Then o is conjugated to r in PG12(C) if and only if tr2o = tr2r.
Proof With the notation of 9.14 we have the following explicit formulas
a 0
trz 1 1 ]=4 , tr2 1=2+a+a-1 aEC-{0,1}
0 1 0 1
Notice, that the value of a+ a-' is unchanged if a is replaced by a-1. On the

other hand we have the matrix formulas
0 -1 a 0 0 1 J=[ 1 0 ] _ [ a- 0
1 0 0 1 -1 0 0 a 0 1
The remaining details are left to the reader. 0

50 QUADRATIC FORMS
I EXERCISES
EXERCISE 1.1 Let E be a finite dimensional vector space equipped with a non-
singular quadratic form.
10 Let F be a non-singular subspace of E, show that E = F ® F 1 and that
(x,y)-(x,-y) ;xEF,yEF1
defines an isometry a E 0(E) with a2 = 1.
2° Conversely, let a E 0(E) be given with a2 = c. Show that E = E+ ®E_
where E+ is the eigenspace for 1 and E_ is the eigenspace for -1. Hint: Use that
x = z(x - a(x)) + z(x + a(x)) ;xEE
3° Given an isometry a E 0(E) with a2 = t. Show that the fixed point space
E. for a is a non-singular subspace of E and that a(x) = -x for x E E01 1.
EXERCISE 1.2 1° Let (E,R) and (F,S) be quadratic spaces. Show that the
function
Q(e,f) = R(e) + S(f) ; (e,f) E E ® F
defines a quadratic form on the direct sum E Fg F of E and F. The form Q on
E ®F is denoted (E,R) 1 (F,S)
2° Let (D,Q) be a quadratic form and E and F orthogonal subspaces of D
such that D = E + F and E fl F = 0. Show that (D,Q) is isomorphic to the form
(E,R) I (F,S) where R and S denote the restrictions of Q to E and F respectively.
EXERCISE 2.1 By a hyperbolic plane over k, we understand a quadratic form

(F,P) which has a basis consisting of two isotropic vectors e and f with <e,f> # 0.
1° Show that all hyperbolic planes over k are isomorphic.
2° Let (E,Q) be a non-singular quadratic form which contains an isotropic vector
e 36 0. Show that e is contained in a hyperbolic plane V C E.
Hint: Choose g E E with <e,g> = 1 and observe that f=2g-<g,g>e is isotropic
with <e,f> = 2.
3° Under the assumptions of 2°, show that the equation Q(x) = a has a solution
xin EforallaEE.
I EXERCISES 51
EXERCISE 2.2 Let us consider a non-singular quadratic form (E,Q). We say

that a subspace F of E is isotropic if the restriction of Q to F is identically zero.
1° Given maximal isotropic subspaces C and D of E. Show there exists a
o E 0(E) such that o(C) = D. Hint: Use Witt's theorem.
2° Show that an isotropic subspace F of E is contained in a maximal isotropic
subspace. Hint: Use Witt's theorem.
3° Let q denote the dimension of a maximal isotropic subspace of E. Show that
(E,Q) is isomorphic to a quadratic form of the type
Hi 1...1Hg1F
with 111,...,Hq hyperbolic planes and F non-singular without isotropic vectors.
EXERCISE 2.3 Let Q be a quadratic form on the vector space E, and let
OP(E) = {o E 0(E) I o(e) = e; e E E 11
1° Show that OP(E) is a normal subgroup of O(E) and that it contains all
reflections along non-isotropic vectors.
2° Given A E k*, show that OP(E) acts transitively on the set
{eEEI Q(e)=a}
Hint: Use the proof of proposition 2.3. 3° Show that any o E 0P(E) can be
written as a product of at most 2 dimE orthogonal reflections.
4° When dim E = 2 show that any or E OP(E) can be written as a product of
one or two orthogonal reflections.
EXERCISE 2.4 Consider the quadratic form on R 3 with Gram matrix G (below)
0 0 -1/s
r,s E R
G= A= R= -r 1 -r/s
s#0
-s 0 0
1° Show that any orthogonal reflection of R3 has a matrix of the form R

above.
20 Show that the matrix A represents an orthogonal transformation of R3
which can't be written as a product of two orthogonal reflections.
52 QUADRATIC FORMS
EXERCISE 3.1 1° Show that the determinant det is a quadratic form on M2(R)
and that its polarisation is given by
<A,B> = 2 it AB` ; A,B E M2(R)
where the symbol B' denotes the cofactor matrix of B given by
a b J=[ d -b
c d -c a
2° Show that the Sylvester type of det on M2(OI) is (-2,2).

3° Show that
<A,B> = it A tr B - Ir AB )
2(
EXERCISE 4.1 Let E denote a Euclidean space of dimension 3.

1° Given two planes P and Q, show that there exists a plane R whose normal
vector is orthogonal to the normal vectors of P and Q.
2° Show that an even isometry a E SO(E) can be written or = k,\ , where k
and ) are half-turns in lines K and L (reflections in K and L in the notation from
Exercise 1.1).
3° Show that any two half-turns are conjugated in SO(E).
EXERCISE 4.2 Show that S03(UR) is a simple group, in the sense that any
normal subgroup N 54 {t} equals S03(R). Hint: Use the previous exercise to see
that it suffices to show that N contains a half-turn. Use the next exercise actually
to find a half-turn in N.
EXERCISE 4.3 Let E be a Euclidean vector space of dimension 3 and let

a E 50(E) be a rotation with angle > 7r/2.
1° Show that there exists a line L through 0 such that L and a(L) are
orthogonal. Hint: Observe that the angle between L and a(L) is a continuous
function of L.
2° Let A be half-turn with respect to a line L with the property that L and
a(L) are orthogonal. Show that aAa A is a half-turn. Hint: Show that a(L) is
stable under a)o . with eigenvalue -1.
I EXERCISES 53
EXERCISE 4.4 1° Show that the Killing form

<X,Y> = ir(X Ty) ; X,Y E MM(R)
is a symmetrical bilinear form on Mn(R) and that the corresponding quadratic
form satisfies the Cauchy-Schwarz inequality
<X,Y>I < 1XI IYI ; X,Y E
2° Show that the Killing form is invariant under 0,,,(R) in the sense that
<gxg 1, gXg'> = <X,X> ; X E MM(OZ), g E OO(R)
EXERCISE 6.1 1° Show that Lor+(F) acts transitively on each of the sheets Hn
of the hyperboloid S(F).
2° Show that the special Lorentz group Lor+(F) is connected.
EXERCISE 6.2 Show that the orthogonal group 0(F) has four connected
components.
EXERCISE 7.1 Justify the classical construction by ruler and compass of

inversion B of the point A in the sphere with centre 0 and radius r
Hint: Note that SOAP is similar to AOPB and deduce that d(O,A)d(O,B) = r2.
Alternatively, observe that the circle with diameter BP is orthogonal to the initial
circle, and passes through the point A; compare 1.7.14 and 1.8.7.
54 QUADRATIC FORMS
EXERCISE 7.2 1° Show that dilatation with centre 0 and ratio r>0 can be
realised as the composite of reflections in the Euclidean spheres with centre 0 and
radius 1 and q-r-.
2° Show that parallel translation in E can be written as a composite of
Euclidean reflections in two parallel affine planes.
EXERCISE 7.3 (Ptolemy's theorem) Let E be an Euclidean vector space of dimen-

sion n. Show that n+2 points x1,...,x,, 2 of E lie on a sphere if and only if
detij d(x1,xj)2 = 0
where d denotes the Euclidean distance, compare 11.2.1. Hint: With the notation
of 1.7.4 show that <t x, t y> = 2 d(x,y)2 for x,y E E.
EXERCISE 8.1 Let us consider Euclidean discs 1.t and 11 with centres u and v
and radii r and s. Show that
v12
niar - in - - r2 - s2
2rs
EXERCISE 8.2 Let us consider a square matrix with n+2 rows and n+2 columns
-1 1 1 1
diagonal entries: -1
1 -1 1 1
entries elsewhere: 1
1 1 -1 1
1 1 1 -1
1° Show that this matrix defines a quadratic form of type (-n-1,1)

2° Let E be a Euclidean n-space. Show that there exists n+2 spheres in E
which are two and two tangent to each other. Hint: Pick a basis for E ® R2 with
the above matrix as Gram matrix.
I EXERCISES 55
EXERCISE 8.3 Let R denote a pencil of circles in the Euclidean plane E, and let
G denote the group generated by reflections in the circles of the pencil.
1° Show that the action of G on R identifies G with a subgroup of O(R), and
identify this subgroup depending on the type of the pencil.
2° Show that a product a/y of three reflections in circles of the pencil is itself
a reflection in a circle from the pencil.
EXERCISE 9.110 Show that A E G12(C) with tr A = 0 defines a Mobius

transformation on t, which is an involution.
2° Show that all involutions in Mob(C) can be represented in this way.
EXERCISE 9.2 Let P,Q,R denote three distinct points of C. Show that the point
S E C belongs to the circle through P,Q,R if and only if [P,Q,R,S] E R.
EXERCISE 9.3 Let P,Q,R,S be four distinct points on a circle 9G. Show that
[ P,Q,R,S] < 0
if and only if the pair {P,Q} separates the pair {R,S}; meaning that P and Q
belongs to different connected components of 9G - {R,S}.
EXERCISE 9.4 By a Half-turn we understand an even Mobius transformation

o#t with o,2=t.
10
Show that a half-turn has precisely two fixed points on C.
2° Show that there exists a unique half-turn fixing two given points of C.
3° Let there be given four distinct points P,Q,R,S of C. Show that there
exists a unique o E PG12(C) which transposes P and Q and transposes R and S. In
the following this half-turn will be denoted (P,Q)(R,S). Observe that this symbol
is still meaningful in case R = S.
4° Show that an even Mobius transformation of C can be written as a
product of two half-turns; otherwise expressed, show that a given triple A,B,C of
points of t can be transformed into another given triple P,Q,R by a product of
two half-turns. Hint: In case A 54 Q and B # P let D denote the image of C by
(A,Q) o (B,P) and consider the transformation
56 QUADRATIC FORMS
(R,D)(P,Q) ° (A,Q)(B,P)
When A = P and B = Q, use the transformation (A,B)(R,R) o (A,B)(C,R).
EXERCISE 9.5 1° Show that the cross ratio of four distinct points is a complex
number distinct from 0 and 1 which obeys the rules
[P,Q,R,S] = [Q,P,R,S]-1 = [P,Q,S,R]-1
2° Five distinct points P,Q,R,S,T of t satisfy the cancellation rule

[P,Q,R,S] [P,Q,S,T] = [P,Q,R,T]
EXERCISE 9.6 Fix three distinct points P,Q,R of C and consider the rational
function A on C given by A(S) = [P,Q,R,S]. Show that permutations of P,Q,R
give the six rational functions A, 1- A 1 1 A- 1 A
A-1
1,
EXERCISE 9.7 Let 0,7 E PG12(C) be transformations represented by matrices

S,T E G12(C). Show that tr[o,r] = lr STS-1T-1 is independent of S and T.
2° Show that lr[o,r] = 2 if and only if a and r have a common fixed point
on the Riemann sphere C. Hint: Use the normal forms for S from exercise 9.1.
3° Give an example of transformations o and r with lr[o,r] = - 2.
EXERCISE 9.8 For o E PG12(C) let Fix(o) denote the set of fixed points for o on
C. Show that if r E PG12(C) commutes with or then o(Fix(r)) = r(Fix(o)).
2° Show the converse of 1°. In particular, show that Fix(o) = Fix(r) implies
that o and r commute.
3° Find the condition for two half-turns to commute.
EXERCISE 9.9 Show that the evaluation map

PG12(C) Cxtxt ; o ,--, (-(00),o(0),o(1))
is a homeomorphisrn of PG12(C) onto the open subset W of C x C x C consisting of
triples of distinct points of C. Hint. With the notation of 11.9.6 show that the
following Mobius transformation maps (00,0,1) onto (P,Q,R)
pl del[q,r] -q1 det[Rr]

p2det[q,r] -g2det[p,r]
II GEOMETRIES
We are now ready to introduce the three basic geometries: Euclidean,

spherical and hyperbolic geometry. Using the algebra from the previous chapter
we shall work our way through the three geometries and identify their isometrics
and geodesic' curves. In the case of hyperbolic geometry we present three "non-
linear models" : the Klein disc model, the Poincare disc model and the Poincare
half-space model.
11.1 GEODESICS
Let us recall that a metric space is a set X equipped with a symmetrical

function d: X x X - IR with the property that d(A,B) = 0 if and only if A = B and
which satisfies the
Triangle inequality 1.1 d(A,C) < d(A,B) + d(B,C) ; A,B,C E X
A map o,: X--+Y from one metric space to another is distance preserving if
1.2 d(o(A),o(B)) = d(A,B) ; A,B E X
It is clear from the definition that a distance preserving map is injective. A
distance preserving map o: X--*Y which is surjective is called an isometry. Observe

that the inverse map of an isometry is itself an isometry. It follows that the set of
isometrics of a given metric space form a group. We shall say that metric spaces
X and Y are isometric if there exists an isometry of X onto Y.
'The geometries we are going to study all carry the structure of metric
space. The notion of a geodesic curve in a metric space, as introduced in section
11.1, is consistent with the notion of a geodesic curve in Riemannian geometry.
58 GEOMETRIES
The basic example of a metric space is the set R of real numbers equipped
with the standard metric d(x,y) y - x 1, x,y E R. It will be important for us to
know all isometries of R.
Proposition 1.3 An isometry o:R - R is either a translation x x+a or

a reflection x i-+ b-x.
Proof For a given isometry o of R, let r be the reflection or translation which

agrees with o at 0 and 1. We are going to prove that o = r. Suppose for a
moment that c E R is a point where r(c) # a(c): but this implies that the point
a = 0(0) = r(0) is the midpoint of o(c) and r(c) since
d(o(c),a) = d(o(c),o(0)) = d(c,0) = d(r(c),r(0)) = d(r(c),a)
Similarly, b = u(1) = r(l) is the midpoint of o(c) and r(c), contradicting a # b.
Definition 1.4 Let J C R be an interval and X a metric space. A curve

y:J-+X is said to be a geodesic curve if each point c E J has a neighbourhood
U C J such that the restriction of y:U-'X is distance preserving.
Lemma 1.5 Let u be a point of the open interval J of R and y:J->R a

geodesic curve with y(u)=0. We can find e = ± 1 such that
y(t)=e(t-u) ;tEJ
Proof Let us choose an open neighbourhood U C J of a given point u such

that the restriction of y to U is distance preserving. We conclude from the proof
of 1.3 that the restriction of y to U is an affine function. It follows that y is
differentiable on U with constant derivative. Using that R is connected we
conclude that y' is globally constant on J. At this point the result follows from
elementary calculus. 0
11.2 EUCLIDEAN SPACE
Let E denote an n-dimensional Euclidean space. The scalar product of

vectors e and f will be denoted < e,f > , while the length of a vector c is given by
Icl = <c,c>. The distance between two points A and B of E is defined by
2.1 d(A,B)=I A - BI ; A,BEE
This turns E into a metric space as follows from Cauchy-Schwarz, I.4.1.

Let us observe that a point v E E and a vector e of norm 1 defines a
geodesic curve t '-, to + v, t E R, whose image is an affine line. Conversely
Proposition 2.2 Any geodesic curve y : R - E has the form

y(t) = et + v ;tEIR
where v is a point of E and e is a vector of norm 1.
Proof Let us consider a point a E R and pick an open interval J around a such
that the restriction of y to J is distance preserving. It follows from the sharp
triangle inequality 1.4.2 that for any three distinct parameters r,s,t E J the points
y(r), y(s),y(t) lie on an affine line. If we fix two distinct points c,d E J we find
that y(J) is contained in the affine line through y(c) and y(d). It follows from 1.5
that we can find a vector e of norm 1 such that
y(t) = e(t - a) + ry(a) ;tEJ
This implies that y is a differentiable curve whose derivative y'(t) is locally
constant. Since IR is connected, we conclude that ye(t) is independent of t E R. It
follows from elementary calculus that y has the required form. 0
Let us find all isometries of the Euclidean space E. A basic example is

reflection r in an affine hyperplane H of E. In terms of a unit normal vector n to
60 GEOMETRIES
H and a point u E H, the action of r is given by
2.3 r(x)=x-2<x-u,n>n ;xEE
The isometry r fixes H while the image r(x) of a point x 0 H can be described by
observing that the affine line through x and r(x) is perpendicular to H and that H
intersects this line in the midpoint of x and r(x).
Lemma 2.4 Let there be given two sequences A1,...,Ap and B1,...,Bp of
points of E such that
d(Ai,Ai) = d(Bi,B3) ; i,j = 1,...,p
Then there exists an isometry a of E composed of at most p reflections such that
cT(Ai) = Bi ; i = 1,...,p
Proof Let us recall that the bisecting hvnernlane for two distinct points A
and B of E is given by
{ X E E I d(X,A) = d(X,B) }
It is easily seen that reflection in the bisecting hyperplane interchanges A and B.
We shall prove the lemma by induction on p. Observe that we have just
taken care of the case p = 1. To accomplish the induction step, suppose that an
isometry p composed of at most p-1 reflections has been found to transform
AI,...,Ap-1 into BI,...,Bp-1. When p(Ap) = Bp we can take a = p. When
p(Ap) $ Bp, observe that
d(p(Ap),Bi) = d(p(Ap),p(Ai)) = d(Ap,Ai) = d(Bp,Bi) i = 1,...,p-1
;
which means that the points BI,...,Bp_1 all lie on the hyperplane bisecting p(Ap)
and B. If r denotes reflection in this hyperplane then the isometry o = rp serves
our purpose. 0
By a Euclidean simplex in the n-dimensional space E we understand a

sequence A0,A1,...,An of points of E not contained in an affine hyperplane.
Lemma 2.5 Let there be given a Euclidean simplex A0,A1,...,An of E. If

two isometries a and ,3 of E agree on A0,...,A,, then a= 0.
Proof Put o = /34a and assume that P is a point of E where o(P) P. From
d(o(P),Ai) = d(o(P),o(A1)) = d(P,Ai) ; i = o,...,n
it follows that the points A0,...,A. all belong to the affine hyperplane bisecting P
and o(P). This contradicts that the points A0,...,An form a simplex.
Corollary 2.6 Let o be an isometry of E which fixes an affine hyperplane K

pointwise. Then, either o is reflection in K or o = t.
Proof Let us assume that o $ t and pick P E E with o(P) # P. It follows from
the proof of 2.5 that K equals the perpendicular bisector for o(P) and P. Let us
form the composite of r, reflection in K, and o to get an isometry ro which fixes
P and the points of K. We conclude from 2.5 that ro = t.
Theorem 2.7 Any isometry /3 of a Euclidean space of dimension n can be

written as the product of at most n+1 reflections.
Proof Let us pick a Euclidean simplex A0,...,A. in E. We can use lemma 2.4
to pick an isometry o which is the composite of at most n+1 reflections with
o(Ai) = /3(Ai) ; i = 0,...,n
It follows from lemma 2.5 that o = 0.
We are now in a position to describe all isometries of Euclidean space.

First of all a vector e E E defines translation re given by the formula
2.8 re(x)=e+x ;xEE

62 GEOMETRIES
The translations form a subgroup 7(E) of the group Isom(E) of all isometrics.
Lemma 2.9 Any isometry p of Euclidean space E can be written

p=Teo ;eEE,oEO(E)
The orthogonal transformation a is called the linearisation of p and is denoted p.
Proof Let Isom'(E) denote the set of isometrics of E which can be written in
the form TeU as above. We ask the reader to verify the formula
are °1=To.(e) ;e EE, 0EO(E)
We can now verify that Isom'(E) is closed under composition
TV a Te P = TV TQ(e) P = TV+O-(e) vP
Observe also that Isom'(E) contains all reflections in hyperplanes in E as follows
by inspection of the formula 2.3. Thus we get Isonz(E) = Isom'(E) from 2.7. 0
The linearisation pi-+p of an isometry plays an important role in Eucli-

dean geometry. Let us notice some of the basic features. First of all
2.10 (0 i/i) = $ o i' ; 04 E Isom(E)
Observe that for 0 E Isom(E) we have i = c if and only if 0 is a translation. This

fact can be recaptured by saying that the following sequence of groups is exact
2.11 0 -+ T(E) Isom(E) 0(E) -+ 0
As a final general remark on the structure of Isom(E), we ask the reader to verify
2.12 . TV O 1 =
T.i(V)
;oEIsom(E),vEE
We shall classify isometrics of Euclidean space of dimension 2 and 3. The

discussion is based on the following innocent looking theorem.
Classification theorem 2.13 An isometry ik of Euclidean space E can

be decomposed into a product of two commuting isometries
jG=TVOO =00Tv
where TV is a translation along a vector v E E and 0 is an isometry with a fixed
point. Such a decomposition of 0 is unique.
Proof Let us decompose Vi = Tea as in 2.9. The transformation 0, we are

looking for have a fixed point u, say, in E. This means that we are looking for a
transformation of the form .0 = ruar-11, u E E. From 2.12 we find that the
condition for 0 to commute with a translation r, v E E , is that q5(v) = v or
a(v) = v. Thus we are looking for two vectors u and v in E such that
Tea = T,,r crr 1 , a(v) = v
Using orU 1 = T-au a this may be rewritten as
e = v + (u - au) ; v E Ker(a-t), u E E
Such a decomposition of e E E is always possible by the following lemma. Unique-
ness of v comes from the same source.
Lemma 2.14 A transformation a E O(E) gives rise to an orthogonal

decomposition of E
E = Ker(a-t)®Im.(a-t)
Proof Let us first show that the two subspaces Ker(a-t) and Im(a-t) of E
are orthogonal to each other
<x,a(y)-y> = <x,a(y)> - <x,y> = <a(x),a(y)> - <x,y> = 0
This gives us Ker(a - t) n Im (a - t) = 0 . From Grassmann's dimension formula
applied to the linear map a-t we get that
dim Ker(o - t) + dim Im(a - t) = n
The result follows from the dimension formula 1.3.4.
Let us return to the classification theorem 2.13. The fact that rv and 0
commute can be written in the form
64 GEOMETRIES
O(x+v)=O(x)+v ; xEE
This shows that the fixed point set for 0 is stabilised by the translation rv.
Let us apply the classification theorem to dimensions 2 and 3. As a
consequence of our investigations of O(E) made in 1.4.8 and 1.4.9 we get
Euclidean plane 2.15 An even isometry of the plane is a rotation or a

translation. An odd isometry is a fide reflection, i.e. an isometry rlc composed of
a reflection is in a line k and a translation r along k. The special case where r = t
is a reflection.
Euclidean space 2.16 An even isometry is a screw composed of a

rotation and a translation along the axis of rotation. An odd isometry is either a
rotary reflection, composed of a reflection in a plane and a rotation with axis
perpendicular to the same plane, or a fide reflection , composed of a reflection in
a plane and a translation along that plane.
11.3 SPHERES 65
11.3 SPHERES
Let us consider a Euclidean space F of dimension n+l. We shall introduce

a metric on its unit sphere Sn = S(F). To begin, observe that the Cauchy-Schwarz
inequality 1.4.1 gives us
<P,Q> E [-1,1] ; P,Q E Sn
This allows us to define the spherical distance d(P,Q) between two points P and Q
of Sn through the formula
3.1 cos d(P,Q) = <P,Q> ; d(P,Q) E [0,a]
It is clear that the spherical distance d(P,Q) is symmetrical in P and Q. Let us

show that two distinct points P and Q have non-zero distance. If P and Q are
linearly independent, the Cauchy-Schwarz inequality 1.4.1 is sharp and we find
that <P,Q> E ]-1,1[. If P and Q are linearly dependent, we have P = - Q which
gives <P,Q> = -1 and d(P,Q) = 7r. The triangle inequality will be proved in 3.6
after an introduction to spherical trigonometry.
Definition 3.2 By a tan ent vector to a point A E S" we understand a

vector T E F with <T,A> = 0. A tangent vector of norm 1 is called a unit
tangent vector. The space of tangent vectors to S" at A forms a linear hyperplane
TA(Sn) of F called the tangent space to the sphere Sn at A.
Lemma 3.3 Let A and B be points of the sphere Sn. A unit tangent vector
U to S" at A can be chosen such that
B = A cos d(A,B) + U sin d(A,B)
Proof Let us at first suppose that A and B are linearly independent. Let U
be a unit tangent vector at A in the plane spanned by A and B, and let us write
66 GEOMETRIES
B =xA+yU ;x,yER
Using <A,U> = 0 we get x2 + y2 = 1. Thus we can determine s such that
B = A cos s + U sin s ; sE[-7r,7r]
Notice that the formula is unchanged under the substitution (s,U)H(-s,-U), and
conclude that we can write B as above with s E [0,7r]. Observe that
cos d(A,B) = <A,B> = cos s
and conclude that s = d(A,B). If A and B are linearly dependent we have B = A
or A = -B. It follows that sin d(A,B) = 0 and that any unit tangent vector U to
Sn at A will serve our purpose.
Let us consider a spherical triangle AABC. By this we understand three

points A,B,C of Sn linearly independent in the ambient space F. The following
notation is standard
a = d(B,C)
b = d(A,C)
c = d(A,B)
Let us use lemma 3.3 to pick tangent vectors U and V at A such that
B=Acosc+Usinc , C=Acosb+Vsinb
We can use this to introduce LA = a as the angle between U and V
3.4 cos a = <U,V> ; a E ]0,7r[
We are now in a position to prove the basic formula of spherical trigonometry.

111.3 SPHERES 67
Cosine relation 3.5
1
cos a = cos b cos c + sin b sin c cos a
Proof We ask the reader to verify the determinant formula
<A,A> <B,A> <A,A> <U,A>

= sin c sin b
<A,C> <B,C> <A,V> <U,V>
Hint: Observe that the left hand side is zero if B is replaced by A or if C is

replaced by A. Evaluation of the determinants using 3.4 gives us 3.5.
Triangle inequality 3.6 Three points A,B,C in Sn satisfy

d(A,B) < d(A,C) + d(C,B)
This inequality is sharp when A,B,C are linearly independent in F.
Proof When A,B,C are linearly independent we can use the cosine formula
with cosa < 1 to get the inequality
cosa < cos(c-b)
When c-b is positive we deduce that a > c-b which gives c < a+b. Observe
that this inequality is trivial when c-b is negative. This concludes the proof of
the sharp triangle inequality. When A,B,C lie in a plane we can modify the
trigonometry to maintain the cosine formula 3.5 allowing for a = O,ir. We can
deduce the triangle inequality from 3.5 using the inequality cosa < 1.
An alternative proof of the triangle inequality including the sharp triangle

inequality can be found in exercise 111.5
Let us observe that the metric we have introduced on the sphere Sn defines
the topology on Sn induced from the ambient Euclidean space F as follows from
3.7 1 P- Q I = 2- 2<P,Q> = 2- 2 cosd(P,Q)

68 GEOMETRIES
Proposition 3.8 Any geodesic curve can be written in the form

y(t)=Acost+Tsint ; tER
where A E S° and T is a unit tangent vector to Sn at A.
Proof Let us investigate y in a neighbourhood of a point u E R. To this end

we pick an open interval J around u of length less than a and such that the
restriction of y to J is distance preserving. It follows from 3.6 that for any three
distinct parameters r,s,t E J, the points y(r),y(s),y(t) are linearly dependent. In
particular if we fix two distinct points c,d E J such that 7(c) and y(d) are linearly
independent, we find that q'(J) is contained in the plane R spanned by y(c) and
7(d). Put A = y(u) and pick a unit tangent vector T E TA(S") contained in R
and consider the curve
B(s)=Acoss+Tsins ;sEJ-7r,7r[
It follows by a simple direct calculation using 3.3 that 9 induces an isometry
between ]-ir,a[ and the set of points X E R n Sn with d(A,X) < 7r. According to
1.5 we can find e = f 1 such that
-y(t) = A cos e(t-u) + T sin e(t-u) ; tEJ
We conclude that 7 is continuously differentiable on J with y'(u) = eT , thus
3.9 y(t) = y(u) cos(t-u) + y'(u) sin(t-u) ;tEJ
Let us show that two geodesic curves o,,-f: R- Sn which coincide in a neighbour-
hood of 0 are identical. To this end we introduce the set
{ t E R I o(t) = -y(t) , o'(t) = -y(t) }
Observe that the set is closed since 7,0,7',o' are continuous. The set is open as
well, as follows from formula 3.9. Since the set is non-empty, by assumption, we
get from the connectedness of R that y = o. Apply this to the given geodesic curve
7 and the geodesic curve o given by the right hand side of 3.9 with u = 0, and the
result follows.
11.3 SPHERES 69
Isometries of the sphere An orthogonal transformation a E O(F)

induces an isometry of the sphere S. We intend to show that any isometry of the
sphere has this form; but first a lemma.
Lemma 3.10 Let there be given two sequences A1,...,Ap and B1,...,BP of
points of S" such that
d(A1,Aj) = d(Bj,Bj) ; i,j = 1,...,p
Then there exists an isometry v composed of at most p hyperplane reflections with
o,(Ai) = B. ; i = 1,...,p
Proof Consider two distinct points A and B of S", let us analyse the set
{ P E S" I d(A,P) = d(B,P) }
From the definition of the spherical distance it follows that this is the intersection
between S" and the linear hyperplane K orthogonal to the vector A - B.
Orthogonal reflection in K will interchange A and B. We can now conclude the
proof by the method used in the Euclidean case, see the proof of 2.4.
Theorem 3.11 Any isometry 0 of the sphere S" is induced by an

orthogonal transformation of the ambient Euclidean space.
Proof Let us pick a spherical simplex in S", i.e. a sequence A0,...,An of points
of S" linearly independent in F. We can use lemma 3.10 to pick an isometry o
which is the composite of at most n+1 reflections in hyperplanes with
o(A;) = /3(Ai) ; i = 0,...,n
We can conclude the proof by the method used in the Euclidean case, see 2.5.
70 GEOMETRIES
11.4 HYPERBOLIC SPACE
Let us consider a real vector space F of dimension n+1 equipped with a

quadratic form of Sylvester type (-n,1). The hyperboloid S(F) consisting of all
vectors of norm 1 has two sheets as we have seen in 1.6.3. There is no way in
general to distinguish the two sheets of the hyperboloid, but let Hn denote any one
of them. We are going to introduce a metric on H".
The point of departure is the inequality <P,Q> > 1 obeyed by any two
points P and Q of H", compare 1.6.2 and 1.6.4. We define the hyperbolic distance
d(P,Q) between P and Q to be the positive real number d(P,Q) such that
4.1 cosh d(P,Q) = <P,Q> ; P,Q E H"
It is clear that the hyperbolic distance d(P,Q) is symmetrical in P and Q. Let us

verify that d(P,Q) > 0 for P $ Q : observe that the vectors P and Q are linearly
independent and conclude from 1.6.2 that <P,Q> > 1. Let us now turn to the
triangle inequality.
Triangle inequality 4.2 Any three points A,B,C in H" satisfy

d(A,B) < d(A,C)+ d(C,B)
The triangle inequality is sharp in case A,B,C are linearly independent in F.
Proof Consider points A,B,C on H" and put a = d(B,C), b = d(C,A) and
c = d(A,B). We shall calculate the determinant of the Gram-matrix 1.1.10 of
A,B,C using the standard abbreviations cha = cosha and sha = sinha.
1 cha chb
4.3 A = del cha 1 chc
chb chc 1
H.4 HYPERBOLIC SPACE 71
Direct expansion after the first row of -A gives us

A = (1 - ch2c) - cha(ch a - chb chc) + ch b(ch a chc - chb) _
1 - ch2a - ch2b - ch2c + 2 cha chb chc =
(ch2b-1)(ch2c-1)-(cbb chc-cha)2-
sh2b sh2c- (chb chc - cha)2 =
(cha- chbchc+shb shc)(chb chc+shb shc- cha) _
[cha - ch(b - c)] [ch(c + b) - ch(a)] =
4sh2(a+b-c) sh2(a+c-b) sh.'(a+b+c) sh2(c+b-a)
This gives us the following factorisation of the discriminant
4.4 O = 4 sh p sh(p - a) sh.(p - b) sh(p - c) ; p=2(a+b+c)
Let us assume that the vectors A,B,C are linearly independent in the ambient
space F. It follows that they generate a subspace of F of type (-2,1) and we
conclude from 1.3.5 that A > 0. If c > a and c > b we have that
p-a=2(c-a)+2b>0, p-b=2(c-b)+2a>0
Using A > 0 and p > 0 we conclude from the factorisation 4.4 that p - c > 0. It
follows that a + b - c = 2(p - c) = 0.
If the vectors A,B,C are linearly dependent in the ambient space F, then
A = 0. If A,B,C do not represent the same point then p $ 0 and we conclude from
the factorisation 4.4 that p - a = 0 , p - b = 0 or p - c = 0. The remaining details
are left to the reader.
Definition 4.5 By a tangent vector T to a point A E H" we understand a

T E F with <T,A> = 0. A tangent vector of norm -1 is called a unit tangent
vector. The space of tangent vectors to H" at A forms a hyperplane of F of type
(-n,0) which is called the tangent space to H" at A and is denoted TA(H").
The tangent space TA(I-I") inherits a quadratic form from the ambient
space F. It is easily seen that the tangent space has Sylvester type (-n,0)
72 GEOMETRIES
Lemma 4.6 Let A and B be points of H". It is possible to choose a unit

tangent vector U to H" at A such that
B = A cosh d(A,B) + U sinh d(A,B)
Proof If A = B we can use any unit tangent vector U at A. If A # B, the two

vectors are linearly independent and the plan R they span has Sylvester type
(-1,1): according to the discriminant lemma 1.6.3 only three types are possible
(-1,1), (-1,0), (-2,0), but only the first type contains vectors of norm 1. It
follows that we can pick a unit tangent vector U in R and write
B=xA+yU ;x,yER
Using <A,U> = 0 we conclude that x2 - y2 = 1. Thus we can find s such that
B=Acosh s+Usin.hs ;sER
Notice, that the formula is unchanged under the substitution (U,s) H (-U,-s) and
conclude that we can arrange for the formula to hold with s > 0. Observe that
cosh d(A,B) = <A,B> = cosh s
and conclude that s = d(A,B) as required.
Geodesics Starting from a point A E H" and a unit tangent vector T to H"
at A we can construct a curve 0: R - II" by the explicit formula
5(s) = A cosh s+ T sinh s ;sER
Let us calculate the hyperbolic distance between y(s) and y(t), s,t E R
cosh d(y(t),y(s)) = <y(t),y(s)> = cosh(t-s) = cosh It-si
From this we conclude that
d(y(t),y(s)) = I t - s ;s,tER
This shows that 0: R - H" is a geodesic curve. The image of the curve is contained
in the plane R spanned by A and T; in fact it follows from lemma 4.6 that the
image of y is H" fl R.
11.4 HYPERBOLIC SPACE 73
Proposition 4.7 Any geodesic curve 7:R--+H" can be written in the form
y(t) = A cosh t + T sinh t ; tER
where A E H" and T is a unit tangent vector to H" at A.
Proof Let us fix a point u E R and pick an open interval J around u such that
the restriction of y to J is distance preserving. It follows from the sharp triangle
inequality 4.2 that for any three distinct parameters r,s,t E J, the points
y(r),y(s),y(t) are linearly dependent. In particular if we fix two distinct points
c,d E J such that 7(c) and y(d) are linearly independent we find that O(J) is
contained in the plane R spanned by y(c) and y(d). Put A = y(u) and pick a unit
tangent vector T E TA(H') contained in R. It follows from the discussion
preceding 4.7 that we can use lemma 1.5 to find e = ± 1 such that
y(t) = A cosh c(t-u) + Tsinh E(t - u) ;tEJ
This implies that y is continuously differentiable on J with y'(u) = eT. Thus
y(t) = y(u) cosh(t-u) + yo(u) sinh(t-u) ;tEJ
The merits of this formula is to reconstruct y in a neighbourhood of u E R from
the values of 7(u) and yo(u). We ask the reader to conclude the proof by the
method used in the proof of 3.8.
In recapitulation, a geodesic curve y : R - H" is distance preserving and

not merely locally distance preserving. By a geodesic line or hyperbolic line in H"
we understand the image of a geodesic curve y: R -+ H". We have seen that
through two distinct points of H" there passes a unique geodesic line.
Isometries A Lorentz transformation o E Lor(F) preserves the two sheets of

the unit hyperbola S(F), in fact it induces an isometry of hyperbolic space H".
We intend to show that this procedure induces an isomorphism between the
Lorentz group and the group of isometrics of H".
74 GEOMETRIES
Lemma 4.8 Let there be given two sequences A1,...,Ap and B1,...,Bp of
points of Hn such that
d(A;,Ai) = d(Bi,Bj) ; i,j = 1,...,p
Then there exists an isometry a composed of at most p Lorentz reflections with
o-(Ai) = B; ; i = 1,...,p
Proof Consider two distinct points A and B of Hn and let us analyse the
perpendicular bisector for A and B
{ P E HIn I d(A,P) = d(B,P) }
From the definition of the hyperbolic distance it follows that this is the
intersection between H' and the linear hyperplane orthogonal to the vector
N = A - B. We can use 1.6.2 and 1.6.4 to conclude that
<N,N> =2-2<A,B> <0
Reflection along N will interchange A and B. We can now conclude the proof by
the method used in the proof of 2.4. 0
Theorem 4.9 A. isometry 13 of the hyperbolic space Hn is induced by a

Lorentz transformation of the ambient space F.
Proof Let us pick a hyperbolic simplex, i.e. a sequence A0,...,An of points of

H" linearly independent in F. We can use lemina 4.8 to pick an isometry a which
is the composite of at most n+1 Lorentz reflections such that
a(A1) = 13(A;) ; i = 0,...,n
It follows that a = /3, compare the proof of 2.5. 0
Definition 4.10 A. isometry a of the hyperbolic space Hn is called even

or odd if the corresponding Lorentz transformation is even or odd.
11.5 KLEIN DISC 75
11.5 KLEIN DISC
In this section we shall construct a model of hyperbolic n-space based on

the open unit disc D" of a Euclidean space E of dimension n. To begin we observe
that the auxiliary space E ® R carries a quite natural quadratic form of Sylvester
type (-n,l): the corresponding bilinear form is given by
5.1 <(A,a),(B,b)> = - <A,B> + ab ; A,B E E, a,b E R
For hyperbolic space H" we take the sheet of the unit hyperbola in E ® R
consisting of points (A,a) with -<A,A> +a 2 = 1 and a > 0. We shall make use of
the parametrisation p:D"-+H" given by the formula
(A,1)
5.2 ;AED"
P(A)
- 1 - <A,A>
Geometrically, the point p(A) is the intersection of H" and the line through (A,1)
and 0. The Klein model of hyperbolic n-space is D" equipped with the metric
obtained by transporting the hyperbolic metric along p: D" - H". It follows from
5.2 and 4.1 that the metric in the Klein model is given by
I - <A,B>
5.3 coshd(A,B) = ;A,BED"
4 (1 - <A,A>)(l - <B,B>)
The geodesic lines in the Klein model are precisely traces on D of ordinary affine
lines in E. This can be seen from the description of the geodesic lines H" given in
4.7 and the geometric origin of the parametrisation p:D" H".
The boundary 8H" (in the sense of 6.18) can be parametrised by the
Euclidean boundary of OD" by assigning to a point A E 8D" the isotropic line in
E ®R through the point (A,1).
76 GEOMETRIES
11.6 POINCARE DISC
Let us once again start with an n-dimensional Euclidean space E and its
open unit disc D. We shall make use of the auxiliary space E ® R equipped with
the hyperbolic form given in 5.1 and the sheet Hn of the unit hyperboloid in E ® R
passing through (0,1). This time we shall consider the parametrisation f: Dn - Hn
given by the formula
(2P,1 + <P,P>) ;PEDn

6.1 f(P) = 1- <P,P>
From the geometric point of view, this is the stereographic projection with centre
(0,-1) already considered in 1.6.6. The result of transporting the hyperbolic
metric along f is given by
6.2 cosh d(P,Q) = 1 + 2 1P2 Q12 ,P,QED"

(1-IP))(1-1Q7 2
The open unit disc Dn equipped with this metric is called the Poincare disc.
Theorem 6.3 The group M'o6(Dn) of A18bius transformations of E which

leaves Dn invariant acts as the group of isometries of the Poincare disc Dn. In
particular, any isometry of Dn extends to a Nl6bius transformation of E.
Proof Let us first record from 1.6.5 that f-1:Hn-*Dn is given by the formula
f-1(A,a) = A(l+a)-1 ; (A,a) E Hn
Let r denote reflection along a vector (N,n) in E 9 R of norm -1. Let us trans-
form r to Dn. We ask the reader to verify the formula
f-'7-f(P) - P + (n + n<P,P> - 2<P,N>) N ; PEDn

I N-nP 12
When n = 0, we have <N,N> = 1 and f-1rf is Euclidean reflection along N.

II.6 POINCARE DISC 77
When n 10, we have 1

+n2 = <N,N> and f-1rf is given by the formula
P+(InP-N12-1)Nii1
' rf(P) _
f1 nP-N 12
= Nri +ii 2 P-Nn1
1
; P E Dn
1 P-Nn-1 12
This shows that f-1rf is induced by inversion in the Euclidean sphere N with
centre Nn-1 and radius n-1. Let us observe that
(n-1 )2+ 1 = <Nn-1, Nii 1>
which shows that X is orthogonal to Sr-1 = DDn.
The result follows by combining

1.7.16 with 11.4.9 and 1.6.11.
Lemma 6.4 An isometry 0 of the Poincare disc Dn C E is induced by an

orthogonal transformation of E if and only if 6(0) = 0.
Proof An isometry of D can be represented by a Mobius transformation 0 of

E which commutes with reflection o in the unit sphere, 1.7.15. When ¢(0) = 0, let
us evaluate the formula oO = 0o at oo to get o(O(oo)) = 0(0) = 0 or 0(oo) = oo.
It follows from 1.7.7 that 0 is a Euclidean similarity of E. Using once again that
0(0) = 0 and ¢(D) = D, we conclude that 0 is orthogonal.
Proposition 6.5 An isometry 0 of the Poincare disc D° C E can be written

Sn"1
pp, where p E 0(E) and p is reflection in a sphere orthogonal to = DD'.
Proof Let us focus on the point P = 0(0) E Dn. If P = 0, the result follows
from 6.4. If P $ 0, put R = o(P) where a is inversion in the sphere Sn"1. Let
be a circle with centre R orthogonal to Sn"1. We ask the reader to verify that
* Sn 1 = 2i. I p(0) - o(R) I
Observe that o(R) = P and conclude that p(O) = P or p(P) = 0. It follows that
po fixes 0 and we conclude from the previous lemma that pO E 0(E).
78 GEOMETRIES
11.7 POINCARE HALF-SPACE
Let us start from a Euclidean vector space L of dimension n-I and form
the Euclidean space E = L ® R. We shall be concerned with the inner half-space
E+ consisting of points (p,h) of E with h > 0. Inversion o in a sphere with radius
and centre in the south pole S = (0,-1) will map E+ onto the unit disc Dn
Observe that o induces a stereographic projection of 8Dn onto the boundary L of

the upper half-space. The explicit formula 1.7.1 for o:E+ _a Dn may be written
7.1 (p , I+h)
o(p,h) = (0,-1) + 2 IP12 ;pEL,heR
+ (I +h)2
Let us use o to transport the metric of the Poincare disc onto the upper half-space
E+. Explicit calculation based on 6.2 and 7.1 gives us
2
7.2 coshd(P,Q) = 1+ 1P -Q P = (p,h), Q = (q,k)
ph
The open half-space E+ equipped with this metric is called the Poincare half-space.
Theorem 7.3 The group Mob(E+) of Mobius transformations of E which

leaves E+ invariant acts as the group of isometries of the Poincare half-space. In
particular, any isometry of E+ can be extended to a Mobius transformation of E.
11.7 POINCARE 11ALF-SPACE 79
Proof Let v denote the inversion considered in 7.1. Conjugation by o in

Mbb(E) transforms the subgroup Mob(Dn) into the subgroup Mbb(E+). This allows
us to transform theorem 6.3 into 7.3. 0
Corollary 7.4 The group of isometries of the Poincare half-space E+ is

isomorphic to the full Mobius group of L = 8E+.
Proof This results from a simple combination of 1.7.16 and 11.7.3.
The geodesics in the Poincare half-space E+ are traced by circles in E

orthogonal to the boundary aE+.
80 GEOMETRIES
11.8 POINCARE HALF-PLANE
The open upper half-plane H2 of C is a model of for the hyperbolic plane

when we specify the metric according to 7.2
_ 12
8.1 coshd(z,w) = 1 + 2 Im[w] II [z] ; z,wEH2
For the convenience of the reader we shall give two different formulas
Iw-zl I w -Y I
8.2 sinhd(z,w)
z _ Im[w] Im[z] '
cosh d(z , w) _
2 2 Im[w] Im[z]
In order to describe the isometries of H2 recall from 1.9 that G12(R) acts on C-R
through the formula
a b az+b
c d cz + d
We ask the reader to work out the following useful expression for the imaginary
part of this fraction.
IM[ az + b _ Im[z] bl
ra d]
8.3
cz+d] -Icz+d12 det Lc c a] E G12(R)
This leads us to define the following action of G12(R) on H2
az + b
cz + d
deter>0
8.4 a+b 0- = Lc db]
cz + d deto, <0
Proposition 8.5 The action of G12(R) on H2 identifies PG12(R) and the

group Isom(H2) of isometries of H2.
11.8 POINCARIs IIALF-PLANE 81
Proof The formula 8.4 in fact defines an action of G12(R) on t, which

preserves H2. By 7.3 it suffices to prove that any Mobius transformation of C
which preserves H2 has this form. By Poincare extension 1.7.16 it suffices to
identify PG12(R) and the Mobius group of the subspace R of C. This can be done
by modifying the construction in 1.9 to bring PG12(I8) to act on I8 in a triply
transitive manner.
Proposition 8.6 The geodesics in H2 are traced by circles in t orthogonal

to A. Expressed in another way, the geodesics are traced by Euclidean circles with
centres on the x-axis R and Euclidean lines perpendicular to the x-axis.
Proof The curve y: R- H2 given by y(s) = ies, s E R, is distance preserving:
t SZ
e)
cosh (s - t) =2t(e8-t + et-sl =
e2t,+e2s
2etes
_1+ (e
-
2etes
=cosh d(ies,iet)
It follows that dJ, the positive y-axis, is a geodesic. Let us observe that PG12(R)
acts transitively on the set of circles in C orthogonal to A as follows from the fact
that PG12(R) acts transitively on R. The action of PG12(R) on the set of geodesics
in H2 is transitive as well as follows from 8.5 and 4.9.
Let us amplify the fact that the isometries of H2 are induced by Mobius
transformations by asking the reader to verify that the distance can be written
8.7 cosh d(z,w) = I + 2 1 [z,`-V ',z] I ; z,w E H2, z 0 w
Alternatively, let us introduce the circle L through z and w orthogonal to h and

let z* and w* denote the points of intersection between the circle 3. and h, defined
such that z,w and z*,w* give the same orientation to the geodesic Hfl1..
82 GEOMETRIES
z w
z* w*
For the distance between the points z and w we have
8.8 d(z,w) = log [z,w,w*,z*] ; z,w E H2, z $ w
Proof Upon performing a Mobius transformation we may assume that z and w

are points of tlJ with Im[z] < Im[w]. This gives z* = 0 and w* = oo and
[z,w,x,0] - zw
Let us write z = ies and w = iet (s < t) and conclude from the proof of 8.6 that
d(z,w) = t - s. Let us put this together
log [z,w,oo,0] = log z = log et'S = t - s = d(z,w)
and the formula follows. 0
Let us take a closer look at the action of a given o E S12(R) on H2 using

the theory of pencils of circles developed in I.B. We shall make use of the
invariant tr2o = (lro)2. The discussion will be resumed in III.4 and IV.2.
Following Felix Klein we shall divide these transformations into three
classes depending on the invariant lr2o.
Elliptic (rotation): lr2o < 4. The transformation o has a fixed point A E H2

and fixes the pencil of circles orthogonal to the pencil of geodesics through A. A
circle from this pencil which is contained in H2 is called a hvnerbolic circle with
centre A. Hyperbolic circles with centre A can be characterised as the orbits for
the group of elliptic isometries with fixed point A.
11.8 POINCARE HALF-PLANE 83
Hyperbolic (translation) tr2o The isometry a has two distinct fixed

> 4.
points A and B on OH2 and leaves the geodesic h with ends A and B invariant.
The pencil of circles through A and B is left invariant by o. The trace on H2 of a
circle from this pencil is called a hvnercycle for the geodesic h through A and B.
The hypercycles can be identified with the orbits of the group of even isometries
which leave A and B invariant.
Parabolic (horolation) ir2v = 4. The transformation a has a unique fixed

point P on OH2. The pencil of circles through P and tangent to the x-axis is
invariant under c. A circle from this pencil which is contained in H2 is called a
horocycle with centre Q. The horocycles can be characterised as the orbits for the
group of parabolic transformations with fixed point P E 8H2.
84 GEOMETRIES
II EXERCISES
EXERCISE 1.1 Let X and Y be metric spaces. For points P = (x,y) and Q = ({,rl)
on X x Y let us define the distance by
d(P,Q) = d(yrl)2
Show that this defines a metric on X x Y.
EXERCISE 2.1 Give a direct proof of the fact that an isometry f of Euclidean
space E with f(O) = 0 is linear. Hint: Use polarisation to show that
<f(x),f(y)> = <x,y> ; x,y E E
Conclude from this that vectors of the form below have norm zero
f(x+y) - f(x) - f(y) , f(rx) - rf(x) ; x,y E E, r E R
EXERCISE 2.2 Let o and p be rotations of the Euclidean plane with distinct
centres A and B.
10
Justify the following procedure for the calculation of pa-: Write p = 07
and o = ya where y is reflection in the line through A and B and Q is reflection in
a line b through A and a is reflection in a line a through B. If a and b are parallel
we find that pa. = 6a is a translation. If a and b intersect in C we find that
po = ,3a is a rotation with centre C.
2° Suppose that p and a are rotations with the same angle -27r/3 but
distinct centres A and B. Show that pa is a rotation with angle 27r/3 and that its
centre C forms an equilateral triangle with A and B.
30 Let DABC be a triangle in the Euclidean plane. Let A* denote the point
in the plane such that ABC is equilateral and such A and A* lie on opposite
sides of the line through B and C. We let BA denote the rotation about the
circumcentre OA of DA*BC which transforms B to C. Define similar transfor-
mations BC and BB and show that 8C BB BA = c. Hint: Show that 0C BB BA is
rotation with centre B and angle 0.
4° With the notation above, show that the circumcentres OA, OB and Oc
form an equilateral triangle. Hint: Use the method from 1° to calculate 0B BA.
II EXERCISES 85
EXERCISE 3.1 Given two lines k and I in F through the origin. Let us define
the acute angle L(k,l) E [0,2] between k and I by the formula
cos L(k,l) = I<k,l>I
where k and I are unit vectors generating k and I.
10
Show that the acute angle defines a metric on projective space P(F), the
set of lines in F. Hint: Interpret P(F) as the orbit space for the action of the
antipodal map on the sphere S(F).

2° Show that through two distinct points of P(F) passes a unique geodesic.
EXERCISE 3.2 Let S2 denote the unit sphere in R3 and let the standard
spherical coordinates o': l 2 S2 be given by
o(9,0) = (silt 0 Cos 0, Sill 0 sill 0, Cos 0)
1° Show that a continuously differentiable curve y: [a,b] S2 which does not
pass through the north and the south pole can be written y = a(0,0) , where
0,0:[a,b] R are continuously differentiable functions.
2° With the notation above show that the length of the curve y is given by
l(7)= Jab 4 02 + siln2 0 ¢2 dt
and deduce from this that

1(y) ? 10(b) - 9(a) I
3° Let us consider the colatitude 0 as a continuous function 0:S2-*[O,a].
Show that a continuously differentiable curve y:[a,b]_S2 satisfies the inequality
1(y) ? I0(y(b)) - 0(7(a))I
Hint: If 0 < 0(y(a)) < 9(y(b)) < it, let c E [a,b] be the last time 0(7(t)) takes the
value 0(y(a)) and d E [a,b] the first time 0(y(t)) takes the value 0(y(b)) and apply
the inequality from 2° to the restriction of y to the subinterval [c,d].
4° Let P, Q be points on the sphere S2. Show that the spherical distance is
d(P,Q) = infy 1(7)
where y runs through the set of piecewise continuously differentiable curves y on
S2 connecting P with Q. I-Iint: Arrange for P and Q to have the same longitude.
86 GEOMETRIES
EXERCISE 3.3 In R3 let the standard cylinder be given by

2 2
Cy1 = { (x,y,z) I
x +y =1}
For p,q E Cyl define the distance d(p,q) to be
d(p,q) = infy 1(y)
where y runs through the set of piecewise continuously differentiable curves on the
cylinder Cyl.
10 Show that the distance defined above is a metric on Cyl.
2° Recall that the standard cylinder coordinates tc:R2-> Cyl are given by
rc(O,r) = (cos 0, sin 0, r)
Show that a continuously differentiable curve y:[a,b]-+Cyl can be factored y=4cp
where µ:0t-*R2 is a continuously differentiable curve.
3° With the notation above show that the length of the curve y on Cyl is
1(y) = 1(µ)
4° Show that the distance between points p and q on Cyl is given by
d(p,q) = inf d(P,Q)
P,Q; p=ti(P),q= c(Q)
5° Show that X:022_Cy1 is a local isometry, or more precisely that there

exists a constant k>0 such that for every P E R2 the disc with centre P and radius
k is mapped bijectively onto the metric disc with centre p = ic(P) and radius k.
6° Show that an affine line in R2 is mapped by n: onto a geodesic in Cyl and
that all geodesics on Cyl have this form.
7° Let there be given points p and q on Cyl with d(p,q) = 1. Show that we can
find a geodesic curve y:R-Cyl with y(0) = p and 7(l) = q.
EXERCISE 3.4 We shall be concerned with the torus Si x S1 considered as

metric space as described in exercise 1.1. The canonical map X:R--+S1 induces a
map
v: R2 S1 X S1, v(x,Y) = (1(x),X(Y)) ; x,Y E R
1° For points p and q on the torus S1 x S1, show that
d(p,q) = inf d(P,Q)
P,Q; p=v(P), q=v(Q)
2° Show that v:R2_S1 x S1 is a local isometry, or more precisely that there

II EXERCISES 87
exists a constant k > 0 such that for all P E R 2 the disc with centre P and radius k
is mapped bijectively onto the metric disc with centre p = v(P) and radius k.
Show that an affine line in R2 is mapped by v onto a geodesic in S1 x S1,
3°
and that all geodesics on S1 x S1 have this form.
4° Let there be given points p and q on S1 x S' with d(p,q) = d. Show that
we can find a geodesic curve y:R->S' x S' with y(0) = p and y(d) = q.
EXERCISE 3.5 For real numbers a,b,c consider the determinant g given by
1 cosa cosb
g = det cosa 1 cosc
cosb cosc 1
1° Show that
g = 4 sin(7r-p) sin(p-a) sin(p-b) sin(p-c) p = 2(a+b+c)

2° For real numbers a, b, c E ]O,ir[ show that g > 0 if and only if a < b + c,
b <a+c, c <a+b and a+b+c <2a.
3° Show that the inequality g > 0 occurs precisely when the remaining four
inequalities all are sharp.
4° Use this to give an alternative proof of the triangle inequality 11.3.6
EXERCISE 5.1 For d E D', d # 0, let d* denote the mirror image of d in the
unit sphere Si-1 and let 'cd denote reflection in the circle with centre d*
orthogonal to Sr"1. Let Pd denote Euclidean reflection in the hyperplane orthogonal
to d and put rd = t'd Pd-
1° Extend the definition of the symbol rd to all of Dn by the convention
ro = t , and show that any µ E Mob(D") can in a unique way be written
p=rdo ;dEDn,aEO(E)
20 Use the previous result to establish a homeomorphism
Mob(Dn) -Z D' x O(E)
The topology on Mob(Dn) is uniform convergence on Sn-1. (See [Beardon] §3.7.)
EXERCISE 8.1 Show that a E G12(R) acts on H2 as reflection in a geodesic if

and onlyifdeia<Oand trv=0.
III HYPERBOLIC PLANE
In this chapter we shall make a detailed study of the hyperbolic plane H2:
classification of isometries and trigonometry. The even isometries of H2 can be
divided into three classes: rotations, translations and horolations. This division
follows the natural occurrence of geodesics in three types of pencils. We shall
keep track of the pencils by assigning a normal vector to an oriented geodesic.
This will be dealt with in a specific model for H2 referred to as the s12(R)-model.
The big advantage of the s12(R)-model over say the Poincare models is a very rich
vector calculus 1.
111.1 VECTOR CALCULUS
In this section we shall present a very useful three-dimensional real vector

space with an inner product of Sylvester type (-2,1) with a good deal more
algebraic structure. The underlying real vector space is s12(R), the space of real
2 x 2 matrices with trace 0
s12(R) = { R E M2(Id) I it R = 0 }
The square of a matrix R E s12(P) is a scalar matrix, in fact
1.1 R2=-cdetR ; REs12(R)
In view of the importance we offer direct verification of 1.1
a b a b 0
_ (a2+ be)
C -a c -a 0 1
11
1 An interpretation of the vector calculus on s12(R) can be given in terms

of Clifford algebra. See the remarks at the end of section 111.1.
111.1 VECTOR CALCULUS 89
We shall be concerned with the symmetrical bilinear form
1.2 <R,S> = - 2 tr(RS) ; R,S E s12(R)
The corresponding quadratic form is the determinant as follows from 1.1
1.3 del R = <R,R> ; RE s12(R)
Otherwise expressed, the inner product 1.2 is obtained by polarisation of the

determinant, compare 1.1.2. Let us use this to "polarise" the identity 1.1
(R + S)2 - R2 - S2 = - 2<R,S> t
An elementary calculation gives us the formula
1.4 RS + SR = -2 <R,S> t ; R,S E s12(R)
It is time to show that the Sylvester type of our inner product is (-2,1). To this
end we ask the reader to verify that the following three matrices form an
orthonormal basis for s12(R)
I
The determinants of these matrices are -1, -1, +1.
It is now time to introduce an important tri-linear form vol on s12(R).
The evaluation of the form vol on three vectors K,L,M is given by
1.5 vol(K,L,M) = -1 tr(KLM) ; K,L,M E s12(R)
This is actually an alternating form, in the sense that the form annihilates any
triple where two of the vectors are equal: for example if K = M we get from
K2 = - t del K
tr(KLK) = tr(LK2) = - tr(L det K) = - del K lr L = 0

90 HYPERBOLIC PLANE
The alternating form 1.5 is a volume form for s12(R) in the sense that it
takes the value 1 or -1 whenever K,L,M form an orthonormal basis. To see this,
we ask the reader to verify that vol has evaluation 1 against the basis above.
Let us introduce a third opeation on s12(R), the wedge product of two
vectors K and L from s12(IR), given by the formula
1.6 KAL= (KL - LK) ; K,L E S12(R)

Z
The three elements of structure are related through the formula
1.7 <K A L,M> = vol(K,L,M) ; K,L,M E s12(R)
as follows from straightforward computation:

<K A L,M> !r(KLM - LKM) = 2(vol(K,L,M) - vol(L,K,M)) = vol(K,L,M)
4
We shall need one more formula relating the inner product and the wedge product
1.8 A A (B A C) = <A,C>B - <A,B>C ; A,B,C E S12(R)
This formula can be deduced directly from the definitions using formula 1.4:
4(<A,C>B - <A,B>C) =
- (AC + CA) B - B (AC + CA) + (AB + BA) C + C (AB + BA) _
ABC - ACB + CBA - BCA = 4 A A (B A C)
The inner product of two wedge products can be evaluated by the formula
1.9 <A A B,C A D> = <A,C><B,D> - <A,D><B,C> ; A,B,C,D E sl2(Id)
Let us verify this formula. From formula 1.8 we deduce that

<AAB,CAD> = vol(A,B,CAD) = vol(B,CAD,A) = <BA(CAD),A>
We find from 1.8 that B A (CAD) = <B,D>C - <B,C>D and 1.9 follows.
III.1 VECTOR CALCULUS 91
Let us present a very useful volume formula. For two triples of vectors
A1,A2,A3 and B1,B2,B3 in s12(R) we have
1.10 vol(A1,A2,A3) vol(B1,B2,B3) = detij <A;,Bj>
Proof Let us first fix A1,A2,A3. Both sides of the formula are alternating in
B11B21B3 so we may assume that B1,B2,B3 is a fixed positively oriented or-
thonormal basis. We can make a variation of A1,A2,A3 and observe, that both
sides are alternating in the A1,A2,A3. Thus it suffices to treat the case where the
A1,A2,A3 and B1,B2,B3 form the same orthonormal basis. In this case the result
follows from the fact that s12(R) has Sylvester type (-2,1).
Clifford algebra 1.11 Let us add some remarks for readers familiar
with the formalism of Clifford algebra. For reference see [Deheuvels] or the
opening chapter of [Gilbert,Murray], but it is our personal opinion that the
arguments of Clifford algebra are best learned through concrete examples.
The real algebra M2(C) is easily seen to be generated by the linear
subspace isl2(R). The square of an element X of this subspace is a real scalar. In
fact X2 is a real quadratic form of X E isl2(OR) as follows from 1.1:
(iS)2 = det S ; S E s12(R)
Observe that dimR M2(C) = 23 and conclude from general principles that M2(C) is
a concrete realisation of the Clifford algebra of the quadratic space s12(l) of
Sylvester type (-2,1). A good many of the definitions and arguments of this
section conform with the general notions of Clifford algebra. This is particularly
true for the definition of the wedge product 1.6 and the arguments presented in
section 111.4.
92 IIYPERBOLIC PLANE
111.2 PENCILS OF GEODESICS
Let us introduce hyperbolic 2-space H2 as a definite sheet of the unit

hyperboloid in the space s12(R). A point of the unit hyperboloid is represented by
a matrix of the form
A= ; a,b,c E R, detA = 1
Observe that the entry c $ 0 which follows from - a2 - be = 1. Thus the sign of
entry c distinguishes between the sheets of the hyperboloid. We let H2 denote the
sheet consisting of matrices A as above with c>0
Normal vectors Let us recall from 11.4.7 that a geodesic curve y:R-H2
has the explicit parametrisation
7(t) = A cosh t + T sinh t ;tER
where A = 7(0) and T = y'(0) is a vector orthogonal to A with <T,T> 1.
With this notation the normal vector to y is
2.1 N= y'(t) A y(t) ;tER
The normal vector N is independent of time t as follows from the parametrisation

of y given above. Observe that N is orthogonal to A and T and and that the
norm of N is -1 :
<T A A,T A A> = <T,T><A,A> - <A,T><A,T> = -1
The tangent vector y'(t) can be recovered from N by the formula
2.2 y'(t) = y(t) A N ; tER
This follows from 2.1 using the general formula 1.8. Let us remark that the
normal vector N depends on the orientation of the geodesic curve y. Thus the
normal vector of an unoriented geodesic is determined up to a sign only.
Proposition 2.3 The complement of a geodesic h in H2 has two

connected components. Reflection r along a normal vector N for h will fix h
pointwise but interchange the two connected components of its complement.
Proof Let N be a normal vector for h. The function X <X,N>, X E H2,

is zero along h but non-zero elsewhere. This divides the complement of h in H2
into the open sets U and V where this function is positive, resp. negative. The sets
U and V are interchanged by r since <X,N> = -<r(X),N> as follows from
r(X)=X+2<X,N>N ; XEH2
It remains to prove that U is connected. Consider two distinct points A and B of
U. We can use 11.4.6 to pick a unit tangent T to H2 at A such that
B = A coshd+T sinhd ; d = d(A,B)
Let us form the scalar product of N and y(t) = A cosh t + T sinh t, t E ld, to get
<y(t),N> = cosht (<A,N> + <T,N> tanht)
When <T,N> > 0 we find that y(t) E U for all t E R. When <T,N> < 0 we can
use the standard properties of the function tanht to find a number r E R such that
<y(t),N> < 0 for all t > r and <y(t),N> < 0 for all t < r. It follows that y(t) E U
for all t < r and y(t) E V for all t > r. This shows that the geodesic are [A,B] is
entirely contained in U. 0
The two connected components of the complement of h are called the sides
of h. Once the geodesic h is oriented by a normal vector N, we can distinguish
between the two sides of h: a geodesic curve through a point A E h with tangent
vector N at A passes from the negative side of h to the positive side of h (warning:
the function X F-+ <X,N> takes negative values on the positive side of h).
94 HYPERBOLIC PLANE
Oriented angles Let us recall that the tangent space of H2 at a point A is

the space TA(H2) of vectors in s12(R) orthogonal to the vector A. We say that a
pair of tangent vectors X,Y are positively oriented if vol(A,X,Y) > 0. Given a unit
tangent vector S E TA(H2), then S and S A A form a positively oriented
orthonormal basis for TA(H2) as follows from 1.7 and 1.9. Given two unit tangent
vectors S and T to H2 at the point A E H2, the oriented angle Lor(S,T) between S
and T is defined by
2.4 T = S cos Lor(S,T) + S A A sin Lor(S,T) ; Lor(S,T) E 02/2irl
In accordance with Euclidean geometry we shall make use of a second kind of

angle, namely the interior angle LitA E ]0,2a[ of a polygon A at the vertex A.
Intersecting geodesics We shall investigate the intersection between

two oriented geodesics h and k through the point A E H2. Let us introduce the
directional angle a from h to k as indicated below
or = Lor(V,-U)
-U=Vcosa+VAAsina
Using 1.7 and 1.8 we deduce that <U,V> = cosa and U A V = A sina. For the
corresponding normal vectors H = U A A and K = V A A we deduce from 1.8 that
2.5 <H,K> = cos a , H A K = A sin a
In particular, we find that I<H,K>I < 1. Conversely

Proposition 2.6 Two distinct geodesics h and k in H2 intersect if and

only if their normal vectors H and K satisfy I<H,K>I < 1.
Proof Suppose that the normal vectors H and K satisfy I<H,K>l < 1.
Consider the discriminant A for the plane E spanned by H and K
A = <H,H><K,K> - <H,K>2 = 1- <H,K>2
Observe that A>O and conclude from the discriminant lemma 1.6.3 that E has
Sylvester type (-2,0). It follows that the line E L has type (0,1) and that the
point A of intersection between H2 and E L lies on h and k. The converse
implication follows from 2.5.
We say that geodesics h and k are perpendicular if they meet at a point A

of H2 at an angle a.
Corollary 2.7 Geodesics h and k in H2 are perpendicular if and only if their

normal vectors H and K satisfy <H,K> = 0.
Proof Two geodesics which meet at a right angle will have <H,K> = 0 as it
follows from 2.5. Conversely, if <H,K> = 0 the two geodesics will meet according
to 2.6. The angle at which they meet must be a right angle, 2.5.
Geodesics with a common perpendicular Let us investigate

two geodesics h and k both perpendicular to the same geodesic 1. The points of
intersection with 1 will be denoted A and B respectively. Let L be the normal
vector for I corresponding to the orientation of 1 from A towards B. We shall
make use of the normal vector H = A A L for h and the normal vector K = -B A L
for k, compare 2.2.
96 HYPERBOLIC PLANE
H=AAL
K= -BAL
Let us evaluate the inner product between the normal vectors H and K
<H,K> = <A,L><L,B> - <A,B><L,L> = <A,B>
This gives us the first of the following two formulas
2.8 <II,K> = cosh d(A,B) , vol(H,L,K) = - sinh d(A,B)
In order to prove the second formula, observe first that L A K = B as follows from
1.8 using K = L A B. Next, observe that vol(H,L,K) = <H,B> to get that
B = A cosh d(A,B) + H sinh d(A,B)
This gives <H,B> = <H,H> sinh d(A,B) = - sinh d(A,B) as required.
Theorem 2.9 Two distinct geodesics h and k are perpendicular to the same
geodesic if and only if their normal vectors H and K satisfy I<H,K>l > 1. The
common perpendicular geodesic is unique whenever it exists.
Proof Observe that H and K span a plane E, and form the discriminant
A = <H,H><K,K> - <H,K>2 = 1 - <H,K>2
It follows that A<O if and only if I<H,K>l > 1. If A < 0 we conclude from the
discriminant lemma I.6.3 that E has Sylvester type (-1,1). It follows that E -1-
contains a vector L of norm -1. The geodesic 1 with normal vector L is perpen-
dicular to h and k.
Conversely, consider a geodesic I perpendicular to both h and k. It follows
IH.2 PENCILS OF GEODESICS 97
from 3.3 that a normal vector L for 1 is orthogonal to both H and K, i.e. L E E 1
This makes L unique up to a sign.
Lines with a common end A geodesic h in H2 spans a linear plane

of type (-1,1). The two isotropic lines in this plane are called the ends of h. Let
us observe that a normal vector H for h is orthogonal to the the two ends of h.
Proposition 2.10 Two distinct geodesics h and k in H2 have a common

end if and only if their normal vectors H and K satisfy I<H,K>l = 1.
Proof Observe that H and K span a plane E, and form the discriminant
A = <H,H><K,K> - <H,K>2 = 1 - <H,K>2
If h and k have a common end S we must have S = E 1 . In particular, we
conclude that the line E 1- is isotropic. It follows from the discriminant lemma
1.6.3 that the plane E has type (-1,0) and that A = 0, i.e. I<H,K>l = 1.
Conversely, if I<H,K>l = 1 we get A = 0 and we get from 1.6.3 that E 1 is an
isotropic line, in fact a common end for h and k.
Pencils of geodesics A set `.P of geodesics in H2 is called a en ncil if

there exists a linear plane P in s12(R) such that `.P equals the set of geodesics with
normal vectors in P. Let us recall from 1.6.4 that a linear plane P in s12(Ot) is of
Sylvester type (-2,0), (-1,1) or (-1,0). It follows by inspection that P is
generated by its vectors of norm -1. In other words, the plane P is generated by
the normal vectors of geodesics in the pencil it determines. Let us examine the
three Sylvester types separately.
Type (-2,0). The line P 1 has type +1 and intersects H2 at a point A,
say. The pencil 9 is the set of geodesics through the point A.
Type (-1,1). The line P 1 has type -1 and is generated by the normal
vector of a geodesic h. The pencil 9 is the set of all geodesics perpendicular to h.
Type (-1,0). The line S = P 1 is isotropic. The pencil 9 is the set of

geodesics with end S.
Let us examine the intersection of two pencils. It follows from the

discussion above that two distinct pencils can't have more than one geodesic in
common. The following is a list of true statements about the intersection of
pencils.
2.11 Through two given points A and B passes a unique geodesic h.
2.12 Through a given point A passes a unique geodesic perpendicular to

a given geodesic 1.
2.13 Through a point A passes a unique geodesic with a given end.
2.14 There is a unique geodesic with a given end S perpendicular to a

given geodesic h provided the ends of h are different from S.
2.15 There is a unique geodesic with given ends R and S.
Proof Let A be a point of H2 and A the pencil of geodesics through A.

Observe that the plane A 1 has type (-2,0) and conclude that the intersection
between A 1- and any other plane P is a line of type -1. This proves the first
three statements.
To prove 2.14, pick a normal vector N to h and observe that N 1 has type
(-1,1) while S -L has type (-1,0). It follows that the intersection of these two
planes has type -1 or 0. Type 0 can be ruled out since S is not an end of h.
To prove 2.15, observe that the planes R 1- and S 1- have type (-1,0).
The intersection is a line of type -1 or 0. Use the fact that our form is non-
singular to rule out 0.
III.3 CLASSIFICATION OF ISOMETRIES 99
111.3 CLASSIFICATION OF ISOMETRIES
An isometry of the hyperbolic plane H2 is the product of at most three

reflections in geodesics in H2 as we have seen in 11.4. In particular we can write an
even reflection as a product a,6 of reflections in geodesics a and b. A finer
classification will depend on the relative position of the two geodesics a and b. In
other words, the classification of even isometries is tied to the three types of
pencils of geodesics. The classification of odd isometries turns out to be simpler.
An odd isometry is a glide reflection, i.e. the composite of a translation along a
geodesic and a reflection in the same geodesic.
Theorem on three reflections 3.1 Let be reflections in

three geodesics a,b,c belonging to the same pencil 9. The product a/3y is itself a
reflection in a geodesic from the pencil 9.
Proof Let S # 0 be a vector orthogonal to the plane generated by the normal

vectors of the geodesics from 9. Let us show that a Lorentz transformation o of
s12(R) which fixes S is the product of one or two reflections in geodesics from 9.
Note that a induces an orthogonal transformation of the plane P = S 1 . When
<S,S> > 0 , the plane P has type (-2,0) and the result follows from 1.4.5. When
<S,S> = 0 , the plane P is parabolic and the result follows from 1.5.3 and 1.6.16.
When <S,S> < 0 , the plane P is hyperbolic and the result follows from 1.6.11.
Observe that the vector S is fixed by the three reflections a, Q and y and
conclude that a$y can be written as a product of at most two reflections. On the
basis of parity II.4.10, we conclude that a13y is a reflection. 0
The theorem on three reflections has an application to the geometry of a

triangle AABC. Let rna, mb, in,, be the perpendicular bisectors for AB, BC and
CA, compare the proof of 11.4.8.
100 HYPERBOLIC PLANE
Corollary 3.2 The perpendicular bisectors ma, mb, m, for triangle DABC
lie in a pencil.
Proof Let 9 denote the pencil containing mb and ma and let a E 9 be the
geodesic through A. Reflections in the geodesics ma, mb, a are denoted µa, Pb and
a. It follows from 3.1 that µaµba is a reflection in a geodesic h E 9. Observe that
µaµba transforms A into B and conclude that h = mc. Thus me E 9.
Horolations Let S be an isotropic line in s12(R). A horolation with centre S

is an isometry of the form a,Q where a and 6 are reflections in geodesics a and b
with end S. The set of horolations with centre S form a group, 3.1.
Corollary 3.3 The group of horolations with centre S is abelian.
Proof For three reflections p,a,r in geodesics with end S we have par = rap,
as follows from par = (par)-1 = 7_ 1
0`
1P- 1 . For four such reflections, we get that
(a1)(7) = (aQy)b = (7)3a)b = y(3ab) = y(ba,Q) = (yb)(a/3)
and the result follows.
A closer look at the proof of 3.1 reveals that the group of horolations with
centre S is isomorphic to the additive group of R, see also IV.2 The orbits on H2
for the group of horolations with centre S are called horocycles with centre S.
Corollary 3.4 Two distinct points A and B belong to the same horocycle
with centre S if and only if the perpendicular bisector m for A and B has end S.
Proof Suppose that the perpendicular bisector in for A and B has end S. Let
It be a reflection in m and let a be a reflection in the geodesic through A with end
S. With this notation, the horolation µa transforms A into B. Conversely, let a
be a horolation with B. It follows from the theorem on three reflections
that as is a reflection in a geodesic it with end S. Note that o-a(A) = B and
conclude that n is the perpendicular bisector for A and B.
Proposition 3.5 The group of horolations with centre S acts on the set of
geodesics with end S in a simply transitive manner.
Proof Let a and b be geodesics with end S and let M and N be normal
vectors for a and 6 respectively. The linear automorphism a of the plane P = S 1
with o,(S) = S and a(M) = N is orthogonal. It follows from 1.5.3 that o can be
written as a product of one or two reflections. Upon replacing N by -N we can
assume that a is the product of two reflections. It is immediate to extend a to a
horolation with centre S.
Let us assume that the horolation cr fixes the geodesic a. For A E a we
must have a(A) = A: otherwise, the perpendicular bisector for A and has end
S, contradicting the fact that two perpendicular lines can't have the same end.
Corollary 3.6 A horocycle 3G with centre S and a geodesic a with end S

intersect in precisely one point.
Proof If X and a have two distinct points A and B in common, then the
perpendicular bisector m for A,B has end S, contradicting the fact that two perpen-
dicular geodesics can't have a common end. Let us choose a point B E )G and a
horolation or which transforms the geodesic b through B with end S into the
geodesic a. It follows that a(B) is a point of intersection between 7G and a.
We shall present the remaining two pencils leaving the detailed treatment
to the reader. This can be done by algebra as above, but elementary geometry may
be applied as well. See also 11.8.
Rotations By a rotation around a point A of H2 we understand an isometry

of the form a13 where a and /3 are reflections in geodesics a and 6 passing through
A. The rotations form a group as it follows from the theorem on three reflections.
The orbits for the action of the rotation group on H2 are the circles with centre A.
The group of rotations around A is isomorphic to R/2irZ: The rotation angle B of
a rotation o is given by
3.7 0 = Lor(t,o(t)) ; t E TA(H2) , <t,t> = -1
The rotation with angle a is called a half-turn or symmetry with respect to A. A

half-turn can be expressed K A where r and A are reflections in a pair of
perpendicular geodesics through A.
Translations Let k denote a geodesic in H2. By a translation along k we

understand an isometry of the form a/0 where a and /Q are reflections in geodesics
a and b perpendicular to k. The translations along k form a group as follows from
the theorem on three reflections 3.1. The group of translations along k acts on k
in a simply transitive manner. The orbits for the group of translations along k are
called hypercycles .
For a translation o along k the geodesic k is called the translation axis of
o. The translation length T of o is given by T = d(A,o(A)), A E k. It is worth
noticing that the translation axis has a natural orientation.
If we let w denote reflection in k then we can decompose the translation
above as a# = (alc)(K#). In other words, a translation along k can be written as
the composite of two half-turns with respect to points of k. Conversely, the
composite of two half-turns around points of k is a translation along k.
Odd isometries The odd isometries of H2 are easy to classify by virtue of

the following proposition.
Proposition 3.8 An odd isometry 0 of H2 is a lg ide reflection , i.e. of the

form 0 = roc where r is translation along a geodesic k and K is a reflection in k.
Proof Let us fix A E H2 and focus on the point O(A). If A = O(A), we find
that 0 is a reflection in a geodesic through A. If A # O(A), let in denote the
geodesic through A and q5(A) and let l denote the perpendicular bisector for A and
O(A). The point of intersection between in and I is denoted M while µ and A
denote reflections in these geodesics. Observe that \q is even and fixes the point
A. Thus a4 is a rotation around A, and we can use the theorem on three
reflections to write a¢ = ACV where v is a reflection in a geodesic n through A. Let
us introduce reflection ti in the geodesic k through M perpendicular to n and write
Note that \p and inc are half-turns around points of k. It follows that the isome-
try r = () p)(vic) is a translation along k.
111.4 THE SPECIAL LINEAR GROUP
The group of isometries of the hyperbolic plane H2 is the Lorentz group of

the ambient space s12(R). We shall use this to identify the group of isometries of
H2 with the group PG12(R) given by
4.1 PG12(R) = G12(R)/R*
Let us first bring the whole group G12(R) to act on sl2(R). To this end we shall
make use of the sign of the determinant
4.2 sign: G12(R) - {-1,1}
With this notation, the group G12(R) acts on s12(R) through the formula
4.3 o X = sign(o) oXo-1

; o E G12(R), X E s12(R)
It is clear that o acts as an orthogonal transformation of s12(R). We proceed to

verify that this is a Lorentz transformation. To this end we ask the reader to verify
the following formula
<of 10 01
],y-', [ ° 1]> = 2(deto)-1(a2+b2+c2+d2)
where a,b,c,d are the entries of o. It follows from this and 1.6.4 that conjugation
with o interchanges the two sheets of the hyperbola if and only if sign(o) < 0. It
follows that the formula 4.3 defines a Lorentz transformation of s12(UR).
Theorem 4.4 The action of G12(R) on sl2(Od) induces an isomorphism

PG12(R) - Lor(s12(R))
111.4 THE SPECIAL LINEAR GROUP 105
Proof The main point is to realise any Lorentz transformation of s12(R) by an

element of Gl2(R). Let us start with reflection rI< along a vector K E s12(R) of
norm -1. We shall involve the "strange formula" 1.4
KX + XK = -2 <K,X> t ; X E s12(R)
Multiply this identity by the matrix K = K-1 and deduce the formula
4.5 -KXK-1 = X + 2<K,X>K ; X E s12(R)
From the fact 1.6.11, that Lor(sl2(R)) is generated by reflections, it follows that
the map G12(R) -> Lor(s12(R)) is surjective.
Let us show that the kernel of the map G12(R) -; Lor(sl2(R)) consists of
scalar matrices. Consider a o E G12(R) with dec(o) > 0 which acts trivially on
s12(R). Observe that the vector space M2(R) is generated by t and s12(R) and
conclude that o commutes with all matrices X E M2(R). From this it follows that
o is a scalar matrix. Next, consider a a E G12(R) with dec(o) < 0 which acts
trivially on s12(R). This means that o anti-commutes with s12(R). We ask the
reader to verify the general fact that a matrix o E M2(R) which anti-commutes
with s12(R) is the zero matrix.
Corollary 4.6 The action of S12(R) on s12(R) identifies the group PS12(R)
with the special Lorentz group Lor+(s12(R)).
Proof We shall show that the action of a o E G12(R) defined by 4.3 is an even
Lorentz transformation on s12(R) if and only if dec(o) > 0. To this end, we pick a
decomposition of the action of o on s12(R) into a product r1...rs of reflections
along vectors K1,...,Ks of norm -1. It follows from 4.5 that the matrices o and
K1...Kg E G12(R) have the same action on s12(R). Thus we can use 4.4 to pick a
scalar r E R* such that
o = rK1..K5
Calculate the determinant of both sides to get that dec(o) = r2(-1)$. This shows
that dec(o) > 0 if and only if s is even.
Theorem 4.7 Let the action of o E S12(U8) on H2 be decomposed as a,3

where a and ,6 are reflections in geodesics h and k. If H and K denote normal
vectors for h and k, then
lr2(o) = 4<H,K>2
Proof From the end of the proof of the previous proposition follows that we
can write o = eHK, where e = ± 1. This gives us
ir o = e lr(HK) = - 2e<H,K>
The result follows by taking the square of this relation. 0
Let us summarise the discussion in a small table, compare also 11.8.
4.8 Position of h and k tr2(0.)

Klein notation geometric notation
intersecting elliptic rotation
common perpendicular hyperbolic translation
common end parabolic horolation
tr o)2
(det
For a general a E G12(R) we shall make use of the invariant tr2(o) =
)
introduced in 1.9.14.
Proposition 4.9 Let o E G12(R) be a non-scalar matrix. The conjugacy

class of a in PG12(R) is entirely determined by tr2(a) and sign deco.
Proof It is known from linear algebra that any conjugacy class in G12(R)
contains a matrix from the following list
A 0 a -,3 A 1
0 µ (3 a 0 A
The remaining details are left to the reader. 0

III.5 TRIGONOMETRY 107
111.5 TRIGONOMETRY
In the next four sections we shall study the trigonometry of the hyperbolic
plane and apply this to a number of existence and classification problems.
Our treatment is based on the vector calculus from section III.1 and the
concept of normal vectors from 111.2. Let us use the convention that the natural
orientation of a closed polygon is such that the polygon lies on the positive side of
its edges. In other words, the corresponding normal vectors are inward directed, see
the remarks made after 2.3. Let us start with the hyperbolic triangle AABC.
Cosine and sine relations 5.1
cosh a = cosh b cosh c - sinh b sink c cos a
sinh a - sinh b
sin a - sin Q
sinh c
sin y
c
Proof Let us first prove the cosine relation. Recall from 2.5 that
B = A cosh c + V sinh c , C=Acoshb - Usinkb
Let us use this and <U,V> = cosa to calculate the inner product of B and C
<B,C> = cosh b cosh c - sink b sink c cos a
Observe that <B,C> = cosh a and the cosine relation follows.
In order to prove the sine realations we observe that 1vol(A,B,C)I is
symmetrical in A,B,C. Let us evaluate this in an asymmetrical manner: with the
notation from 2.5 we have
vol(A,B,C) = vol(A,U,V) sink b sinh c
A few lines before 2.5 we found that U A V = A sin a, thus
vol(A,B,C) = sin a sinh b sink c = sin a sinh a sinh b sinh c
sinh a
and the result follows since I vol(A,B,C) I is symmetrical in A,B,C. 0
Alternative cosine relation 5.2

cos ,3 cos y + cos a
cosh a = sin ,Q sin y
Proof Let H,K,L be the inward directed normal vectors of the edges of the
triangle
Formula 1.9 in combination with formula 2.5 gives us that

<H A K,L A H> = <K,H><H,L> - <H,H><K,L> = cos /3 cosy + cos a
From the second formula in 2.5 we get that

HAK=Csiny , LAH=Bsin,O
Combine this with the definition of hyperbolic distance to get that
<H A K,L A H> = <B,C> sin y sin 0 = cosh a sin y sin ,Q
and the alternative cosine relation follows. 0
Corollary 5.3 The sum of the angles in ABC satisfies

a+Q+y< 7r
Proof Let us assume that a > ,3 and > y. We ask the reader to use
formula 5.2 to rewrite the inequality cosh a > 1 in the form
cos(7r - a) < cos(,3 +'Y)
When ,8 + y < it we conclude that 7r - a > 13 + y or a + a + y < 7r as required.
When 7r <,3+y we can rewrite the inequality above as cos(a+7r) < cos(,13+y);
this gives a + 7r < 0 + y, which contradicts a > / and a > y.
111.5 TRIGONOMETRY 109
Theorem 5.4 Let a,Q,y E ]0,lr[ be real numbers with a +,3 + y < a, then
there exists a triangle DABC in H2 with LA = a, LB =,3, LC = y.
Proof Let us rewrite the basic assumption 0 + y < Tr - a and conclude that
cos (f3+y)>cos(ir-a)
Using the trigonometric addition formulas we find that
sin 0 sin y < cos,3 cos y + cos a
It follows that we can determine a E ]0,+Oo[ such that
sin / sin y cosh a = cos 0 cos y + cos a
Let us now pick two points B and C with d(B,C) = a , and draw a geodesic I
through B forming an angle ,3 with BC and a geodesic k through C forming an
angle y with the geodesic h through BC. We let H,K,L denote inward directed
normal vectors for these geodesics
Proceeding as in the proof of 5.2 we find that
5.5 cosh a sin ,3 sin y = cos ,3 cos y + <K,L>
This gives us cos a = <K,L>, which shows that I<K,L>I < 1. We can use 2.6 to
conclude that the geodesics I and k intersect. It follows from 5.2 that the angle of
intersection is a. 0
Let us investigate a quadrangle OABCD with three right angles. Such a

quadrangle is called a Lambert quadrangle after J.H.Lambert (1728-74). It is
known since the early days of hyperbolic geometry that the fourth angle of a
Lambert quadrangle is acute. We can deduce this from the second of the two
trigonometric formulas to follow.
Lambert quadrangle 5.6
cosh d(A,D) = cosh d(B,C) sin y

sinh d(A,B) sinh d(D,A) = cos y
Proof Let H,K,L,M be inward directed normal vectors as indicated in the

illustration below
Let us substitute /3 = 2 in formula 5.5 and deduce that

cosh d(B,C) sin y = <K,L>
Formula 2.8 gives us <K,L> = cosk d(A,D), which takes care of the first formula.
To prove the second formula, use 1.10 to get that
0 <K,L> <K,H>
5.7 vol(K,M,L) vol(M,L,H) = det -1 0 <M,H> = <K,H>
0 -1 0
and the result follows from formulas 2.8 and 2.5. 0

III.6 ANGLE OF PARALLELISM 111
111.6 ANGLE OF PARALLELISM
In this section we shall study the trigonometry of a triangle DABC in

which the vertices B and C are ordinary points of H2 while A is a point at infinity
(an end). For the history of this problem and its relations to astronomy, the
reader is referred to "Hyperbolic geometry : The first 150 years" [Milnor].
We are going to prove a relation for AABC which formally looks like the
alternative cosine relation 5.2 of an ordinary triangle with LA = a = 0.
Proof Let us first observe that formula 5.5 remains valid in this context. It
follows from 2.10 that <K,L> = ± 1. In order to decide between +1 and -1,
observe that the left hand side of formula 5.5 is positive. . 0
In particular, if LC is a right angle we find that
cosh a sin /3 = 1
6.2 sinh a tan,6=I
lanha sec(3=1
The last two formulas are elementary consequences of the first. In the third
formula, we are using the old convention sec,3 = 1/cos /3.
111.7 RIGHT ANGLED PENTAGONS
In this section we shall investigate a number of pentagons on which there

are no analogues in Euclidean geometry. Let us start with the trigonometry of the
right angled pentagon. The existence of such a thing will be demonstrated in 7.6.
P4
7.1
cosh as = sinha2 sinha3
cosha3 = cotha2 cotha4
Proof We shall assume that the enumeration of the vertices is consistent with
the natural orientation of the sides of the polygon, 11.5. For i = 1,...,5 we let Ni be
the inward directed normal vector for the edge P;Pi+1. Let us at once record that
<Ni,N;+1> = 0 ; i = 1,...,5 , N6 = Ni
From the general formula 1.10 (or the more specific formula 5.7), we get that
7.2 vol(N1,N2,N3) vol(N2,N3,N4) = <N1,N4>
Evaluate this by means of 2.8 to get the first of the two formulas in 7.1.
Let us now turn to the second formula, which we, by the way, intend to
generalise beyond the case where LP1 is a right angle. We shall first establish the
following formula
7.3 vol(N1,N3,R) = vol(N1,N2,N3) <N2,R>
Observe that this formula is linear in R and check the formula directly in the case
where R=N1,N2,N3. We ask the reader to use 1.10 to get that
vol(N1,N3,N4) vol(N3,N4,N5) = <N1,N5> + <N1,N3><N3,N5>
Combine this formula with 7.3 with the specification R = N4 to get that
111.7 RIGHT ANGLED PENTAGONS 113
7.4 vol(N1,N2,N3)<N2,N4>vol(N3,N4,N5) = <N1,N5> +<N1,N3><N3,N5>
Let us take advantage of formula 2.8 and deduce that
7.5 sinha2 sinha4 cosh a3 = <N1,N5> + cosh a2 cosh a4
In case LPI = 2 we have <N1,N5> = 0 , which concludes the proof.
Proposition 7.6 Let a1 > 0 and a2 > 0 be real numbers. There exists a
right angled pentagon with two consecutive sides of length a1 and a2 if and only if
sinhat sinha2 > 1
Proof The condition is necessary as follows from the first formula in 7.1. To
see that the condition is sufficient, start from a right angle LP1P2P3 with
d(PI,P2) = a1 and d(P2,P3) = a2. Draw the geodesic k3 through P3 perpendicular
to P2P3 and the geodesic k5 through P1 perpendicular to P1P2
Formula 5.7, used in the analysis of Lambert quadrangles, at once gives that
sinhat sinha2 = <N3,N5>
It follows that <N3,N5> > 1 which implies that k3 and k5 have a common
perpendicular geodesic, compare 2.9.
111.8 RIGHT ANGLED HEXAGONS
Let us now turn to the right angled hexagon. The method we have used
in the analysis of the right angled pentagon applies to the hexagon as well. We
shall show that right angled hexagons are classified by the length of three
alternating sides a2,a4,a6.
For geometric applications see [Fathi, Laudenbach, Poenaru], in parti-
cular, "expose 3 et appendice aux expose 8". The point is that the basic building
blocks of hyperbolic geometry, hyperbolic pants, can be sewed out of two isometric
right angled hexagons, see VII.2.6.
P4
Sine relations 8.1
sinha2 sinha6 _ sinha4

=
sinha5 sinha3 - sinhal
P1
Proof With a notation similar to the one applied in the previous section
we get from formula 7.2 that
vol(N4,N5,N6) vol(N5,N6,N1) = <N4,N1>
vol(N1,N2,N3) vol(N2,N3,N4) _ <N1,N4>
Using 2.8 we deduce equality between the two first ratios. The remaining formulas
follow by symmetry.
HI.8 RIGHT ANGLED HEXAGONS 115
Cosine relations 8.2

sinh a2 sinh a4 cosh a3 = cosh a6 + cosh a2 cosh a4
Proof Let us observe that formula 7.5 remains valid in this context. Thus
the result follows after the specification <N1,N5> = cosh a6 , compare 2.8. 0
Classification theorem 8.3 Let a2,a4,a6 be three strictly positive real

numbers. Then, there exists a right angled hexagon in H2 such that three alter-
nating sides have lengths a2,a4,a6. The hexagon is unique up to isometry.
Proof Let the number a3 > 0 be determined by the cosine relation 8.2, and
start the construction of the hexagon on the basis of the numbers a2,a3,a4
Let N1 and N5 denote the inward directed normal vectors for h1 and h5 and deduce
from the relations 8.2 and 7.5 that
<N1,N5> = cosh a6
We conclude from this that the geodesics h1 and h5 have a common perpendicular
h6 say. We leave the remaining details to the reader.
I would like to conclude this chapter with a reference to [Fenchel] which

contains a wealth of trigonometric formulas. In particular, [Fenchel] p.84 gives
information on "improper" right angled hexagons.
111.9 FROM POINCARE TO KLEIN
In this section we shall describe an isometry from the Poincare half-plane

H2 to the s12(R)-model of the hyperbolic plane discussed in this chapter. Let us
recall that the metric on the Poincare half-plane is given by
z-wI2
9.1
I
coshd(z,w) = 1 + z,wEH2
2Im[z]Im[w]
To a point z E H2 we assign the matrix F(z) E s12(R) given by
x _IzI2
9.2 F(z) = 1
Y ;z=x+iyEH2
1 -x
We ask the reader to verify that F is an isometry of hyperbolic planes. We also

leave it to the reader to show that F is G12(R)-equivariant in the sense that it
relates the action from 11.8.4 to the action of 111.4.3 as follows
9.3 F(o(z)) = sign(o) oF(z)o"1 ; z E H2, o E G12(IR)
Proposition 9.4 The transformation F from the Poincare half-plane H2

to the s12(R)-model preserves oriented angles.
Proof Let z vary along a smooth curve in H2 and differentiate F(z) to get
x -I z2 I
x/ -2(xx'+yy')
F(z)I 32
1 -x 0 -x/
Let us rewrite this in terms of the tangent map of F
1 -2x I x -x2+y2
_y1
TZF(u) = u v ; u,v E R
0 -1 y2 1 -x
The norm of the tangent vector TZF(u) at the point F(z) is given by
111.9 FROM POINCARE TO KLEIN
9.5 < TF(°), TF(U)> = del TF(u) =-y2 (u2+v2)
From this follows that TF(U) preserves angles.
It is also worth making the calculation
9.6 vol (F(z), TZF(s), TZF(u)) = (rv-su) ; u,v,r,s E R

2
y
Let us end this section with some remarks for readers familiar with differential
forms. The formula 9.6 may be written
9.7 F*vol = 2 dx A dy
y
Formula 9.5 gives us the Riemannian metric on the Poincare half-plane
9.8 y2(dx2+dy2) x+iy E C, y>O
Hyperbolic area Let us define the hyperbolic area of an ordinary

hyperbolic polygon A on n vertices by the formula
9.9 Area A = (n-2)a - E LintP

P
where the sum is over all vertices P of A. We ask the reader to check that the
area is additive under simple subdivision of polygons. Once this is done, it follows
from 5.3 that Area A> 0. In the Poincare half-plane model the area can be
calculated from the area-form 9.7
9.10 Area A = 12 dxdy

J A
Y
There are hints on how to verify this in the exercises. In general, formula 9.8 may
be taken as the definition of area.
111.10 LIGHT CONE
Let us investigate the space PC of isotropic lines in s12(R). This means

that we are investigating lines generated by matrices of determinant 0. To a non-
zero vector e E K2 we shall assign the matrix
zw -Z
10.1 L(e)
w2 - zw
e- z
w
E R2
'
The matrix L(e) has determinant 0, more precisely the vector e belongs to the
kernel of the linear map with matrix L(e). From this it easily follows that our
construction defines a bijection L: R- + PC(s12(K)). This map obeys the trans-
formation rule (see 1.9.2 for notation)
10.2 L(o(e)) = oL(e) o' ; o E G12(K), e E K2
Proof Let us first perform the decomposition

z z 0 -1
L(c) = ;e=l zIEK2
w
w w 1 0
IT I
Next, observe the matrix formula
T
a b 0 -1 ]=[ 0 -1
c d 1 0 1 0
and the transformation rule follows rather easily.
To an ordered pair (Z,U) of distinct points of K we assign the vector

L(Z,U) E s12(R) of determinant -1 given by
111.10 LIGIIT CONE 119
zv+wu - 2zu
10.3 L(Z,U) = wu 1
zv Z=[z],
w U=[v]
2wv - zv - w u
Alternatively, this can be described by means of the wedge product 1.5
10.4 L(Z,U) = 22 L( w) n L(v) ; D = det [W vl
It follows from 1.6 that the geodesic in H2 with normal vector L(Z,U) has ends
(Z,U). Moreover, we conclude from 10.2 and 10.4 that
10.5 o,L(Z,U)o' = L(o(Z),o(U)) ; a E G12(R)
We shall now evaluate the inner product of these vectors in terms of the cross ratio
on A. For four distinct points A,B,P,Q we shall prove that
10.6 < L(A,B),L(P,Q)> = [A,B,P,Q] -

1
Proof According to 10.5 and 1.9.7 both sides of the formula are invariant
under S12(R). From the fact, 1.9.3, that the action of S12(R) on U2 is triply
transitive it follows that it suffices to verify the formula in the case A = oo, B = 0,
P = 1, Q = q, q E R. Let us recall from 1.9.9 that [oo,0,1,q] = q. Thus it suffices
to prove that
<L(oo,0),L(l,q)> = 9 + l qER
This is a straightforward verification, which we leave to the reader. 0
We ask the reader to establish the following supplement to the general

formula displayed in 10.6
10.7 <L(A,B),L(B,C)> = I ;A$B$C

Let us consider two distinct geodesics h and k in H2 with normal vectors

H and K and ends (A,B) and (P,Q). We can summarise our investigations in the
following table
10.8
Relative position of h and k < H,K > Sylvester type [A,B,P,Q]
intersecting ]-1,1[ (-2,0) ]-oo,0[

perpendicular 0 (-2,0) -1
common perpendicular R-[-1,1] (-1,1) ]0,1[ U ] 1,+oo[
common end 1,-1 (-1,1) 1
The column entitled Sylvester type refers to the pencil generated by h and k, see
the end of 111.2.
III EXERCISES 121
III EXERCISES
EXERCISE 1.1 Determine the eigenspace decomposition for the operator X H X°

on M2(R) . For notation see 1.9.2.
EXERCISE 1.2 Let E denote a two-dimensional real vector space. By a complex

structure on E we understand a linear endomorphism J of E with J2 = -I.
10
tr J = 0 and del J = 1.
Show that J E End(E) is a complex structure 4*
2° Show that two complex structures on E belong to the same sheet of the
hyperbola del = 1 in sl(E) R E End(E) I tr R = 0 } if and only if they define
the same orientation of E.
EXERCISE 2.1 1° Show that the vertices of an equilateral triangle lie on a circle.
2° Show that this is not true for a general triangle in H2.
EXERCISE 2.2 Let h be a geodesic in H2. For a point P E H2 let Ph denote the
perpendicular projection of P onto h. Show that the map P --, Ph decreases
distance, in the sense that d(P,Q) > d(Ph,Qh) for all points P and Q.
EXERCISE 3.1 Let OABCD be a Parallelogram, i.e. a convex quadrangle in

which opposite sides have equal length. Show that the diagonals bisect each other
and that the geodesics through opposite sides have a common perpendicular. Hint:
Show that half-turn µ with respect to the midpoint M of the diagonal AC takes
the quadrangle into itself.
EXERCISE 3.2 Let S be an isotropic line in s12(R). Show that the horocycles
with centre S are cut by the affine planes in s12(R) parallel to TS(H2) = S 1.
EXERCISE 3.3 1°Given a triangle DABC. Let PA denote the rotation with
centre A which transforms the ray AB into the ray AC. Show that PA2 = -YO
where y is reflection in the geodesic AB and ,Q is reflection in the line AC.
2° Define rotations PB and pC as above and prove that PA2 ° PB2 o pC2 =t.
3° If DABC is equilateral show that PA,PB,pC are conjugated in Isom+.
EXERCISE 4.1 Show that an involution in PG12(R) can be represented by a

matrix R E Gl2(ld) of trace 0 and determinant 1 or -1. Hint: Show that a matrix
R E G12(R) which represents an involution in PG12(0R) is an eigenvector for the
operator X H X.
EXERCISE 4.2 Let A E s12(R) represent a point of H2. Show that the
transformation X r» AXA"1 of s12(R) is half-turn with respect to the point A.
EXERCISE 4.3 1° Show that the set of vectors in s12(R) of norm -1 is

connected.
2° Prove that S12(R) is connected. Hint: Any o E Sl2(I) can be written in the
form o = XY, where X,Y E s12(R) are vectors of norm -1.
EXERCISE 4.4 1° Show that an even Mobius transformation a of t with matrix
a b
A= _ a,bEC, Ja12-lb12>0
b a
leaves the unit disc D invariant.

2° Show that any even isometry of D has this form. Hint: Find all even
Mobius transformations which commute with reflection in 8D.
3° Find the connection between lr2A and the "type" of the isometry a.
40 Show that an even isometry of D can be written
exp2aiO1+Cz ;cED,0EIR
EXERCISE 5.1 Consider a triangle DABC with LC a right angle. Show that
sin A = lim cos A= lim b
a-40 c b-->o c
III EXERCISES 123
EXERCISE 5.2 Show that a triangle LABC has an inscribed circle with radius
cos2A + cos2B + cos2C + 2 cos A cos B cos C

r
2 (1 + sin A)(1 + sin B)(1 + sin C)
EXERCISE 6.1 Let S be an isotropic line in s12(R). Show that the set of vectors
of norm -1 in S 1 has two connected components. Use this to replace the second
half of the proof of 6.1 by a "deformation argument".
EXERCISE 8.1 1° Show that the normal vectors of a right angled hexagon
satisfy the formula
vol(N6,N1,N2) <N1,N3> vol(N3,N4,N5) <N2,N4> =
<N21N5><N2,N6> + <N31N5><N3,N6>
2° Prove the following cotangent relation for a right angled hexagon
cosh a2 cosh a3 = coth a1 sinh a3 + coth a4 sinh a2
3° Investigate an improper hexagon with five right angles in which the sixth
vertex P6 lies at infinity. Hint: Derive a formula for the normal vectors similar to
the formula above (the term <N5,N6> must be added to the right hand side).
EXERCISE 9.1 Let DABC be a triangle in the hyperbolic plane H2, and let
a,Q,y be translation from B to C, C to A resp. A to B.
1° Show that /3ay is a rotation around A of angle -Area A relative to the
orientation defined by the triangle. Hint: For two distinct points P and Q we let
tpQ denote the unit tangent vector at P to the geodesic from P to Q. Justify the
following "mod 27r" calculation
Lor(tCA, aytAB) = Lor(tCA, tCB) + Lor(tCB, aytAB) = LintA - Lor(tCB, atBA)
LintA - Lor(a 1tCB, tBA) = LintA + Lor(tBC, tBA) = LintA + LintB
Let 9 denote the rotation angle in accordance with 3.7. Justify the calculation
Lor(QtCA, QaytAB) = Lor(QtCA, tAB) + Lor(tAB, /3aytAB) = - LintA + 9
Finally, combine the two results.
2° Generalise the result to an arbitrary polygon in H2.
30 Prove a similar result for the sphere S2.
EXERCISE 9.2 Show that a disc D in H2 with radius r has

Area(D) = 2a (cosh r -1)
EXERCISE 9.3 Let the hyperbolic area of say a compact subset . of the
Poincare half-plane H2 be defined by formula 9.8. Show that
Area(o(.)) = Area(.) a E Isom(H2)
2° Given r>O and v E [0,2]. For s>r let E(s) denote the subset of H2 bounded by
the y-axis, x = rcosv, y = s and the circle with centre 0 and radius r. Show that
lims_;00 Area E(s) = 2 - v
3° Prove that the area of .ABC is 7r - LA - LB - LC. Hint: Move the triangle
such that two vertices are located on the y-axis and apply 2°.
IV FUCHSIAN GROUPS
We shall introduce the central topic of this book, discrete groups of

isometries of the hyperbolic plane H2. A discrete group of even isometries is called
a Fuchsian rg oun. The set of fixed points for the elliptic elements of a discrete
group r' form a discrete subset of H2. We shall prove after Jacob Nielsen a
converse to this: A non-elementary group r with discrete elliptic fixed point set is
itself discrete.
It is a common feature of the Poincare half-plane and the s12(IR)-model of
H2 that these models provide natural isomorphisms between Isom(H2) and
PG12(R), compare 111.9.3. The results of this chapter can be interpreted in either
of the two models. The s12(R)-model is used throughout in the more technical
parts of the proofs.
IV.1 DISCRETE SUBGROUPS
The group PG12(R), being the factor group of the topological group G12(R)
with respect to a its centre R*, carries the structure of a topological group: the
open subsets of PG12(IR) correspond to open subsets of G12(R) stable under R*. We
shall use the action of PG12(R) on H2 to study discrete' subgroups of PG12(R).
But first, we shall exhibit some convenient compact neighbourhoods of the identity
t in PG12(R).
Theorem 1.1 Acnpoint z E H2 and an c > 0 determine a subset of PG12(UR)

.N(z;() = for E PG12(R) I d(z,o'z) < c }
which is a compact neighbourhood of the identity i. in PG12(R).
'The notion of a discrete subset is discussed in an appendix to this section.

126 FUCIISIAN GROUPS
Proof Let us introduce the norm II a II of a matrix a E G12(R) as
a2+b2+c2+d2 a b
1.2 hall = I det or I
a
c d
Observe the formula II as II = II a II , a E 13*, and conclude that the norm induces a
continuous function on PG12(IR). It follows that a set of the form
1.3 { or E PG12(13) 111 all<r } ; r>F2-
is a neighbourhood of a in PGI2(IR). Let us show that a set of the form 1.3 is

compact. Observe first that the set 1.3 is the image by a continuous function of
the set
{ o EG12(13) 111o1<r,IdetaI =1}
It follows from 1.2 that this set is a closed and bounded subset of M2(R). But a
closed and bounded subset of M2(R) 134 is compact.
Let us recall that the action of PG12(IR) on H2 is transitive. It follows that
it suffices to treat the point i E H2 of the s12(R)-model given by
i=
Let us recall from 111.4.3 that the action of a E PG12(IR) on i is
1 a b 0 -1 d -b
a( : ) lad - bcl c d 1 0 -c a
Simple calculation gives us
2<a(i),i> tr[a(i) i] = lad - bcl-1(a2 + b2 + c2 + d2)

With the notation above this may be written
1.4 2 cosh d(i,ai) = 110"11 ; a E PGI2(13)
The result follows from the fact that sets of the form 1.3 are compact
neighbourhoods of the identity c in PG12(R).
IV.1 DISCRETE SUBGROUPS 127
Corollary 1.5 Let r denote a subgroup of PG12(Id). If r is discrete, then

for all z E H2 and c > 0 the following set is finite
{ o E F I d(o(z),z) < c}
Conversely, if for one c > 0 and one point z E H2 this set is finite, then r is a
discrete subgroup of PG12(R).
Proof With the notation above observe that

{ o E F I d(o(z),z) < c`R,(z;c) fl r
and the result follows from 1.1 and the general lemma 1.13. 0
The modular group 1.6 The group G12(71) of 2 x 2 matrices with

entries from Z and determinant ± 1 is a discrete subgroup of G12(R) as follows
from the elementary fact that any of the subsets
{o E G12(Z) I Ior i < n n>
is finite. The image of G12(Z) in PGl2(Id) is discrete: in fact we ask the reader to
show that any discrete subgroup r C G12(R) consisting of matrices with
determinant ± 1 maps onto a discrete subgroup of PG12(I).
Proposition 1.7 A discrete subgroup F of PGI2(R) operates on the

hyperbolic plane H2 with discrete orbits.
Proof Consider an orbit A for r and pick a. base point z E A. A disc D with
centre z E A will contain finitely many points only from A as follows from 1.5.
Thus we can find e > 0 such that D(z;c) fl .A={z}. Let us show that in fact
1.8 d(u,v) > c ; u,v E A, u # v
To see this choose o E F with o(z) = u. For a second point v # u of the orbit A
we have o(v) 0 D(z;c) and consequently d(u,v) = d(o(u),o(v)) > c. In order to
see that A is closed, consider a point w 0 A. The open disc D(w;c/2) contains at
most two points of A as follows from 1.8. It is now easy to find a disc with centre
w not meeting A.
Theorem 1.9 Let F be a discrete subgroup of PG12(R). For compact

subsets K and L of the hyperbolic plane 112, the sets a(K) and L are disjoint for all
but finitely many a E F.
Proof Given a compact set K, a point z E H2 and a number r > 0, let us show
that D(z;r) and a(K) are disjoint for all but finitely many a E F. Choose s > 0
with K C D(z;r+s) and conclude from 1.5 that S = {y E F Iy(z) E D(z;r+s)} is
finite. Observe that yD(z;r) = D(y(z);r) and conclude that K fl yD(z;r) = 0 for all
y E F - S. Given a second compact set L, choose z E L and r > 0 such that
L C D(z;r). It follows that a(K) and L meet for at most finitely many a E F.
Corollary 1.10 Let r be a discrete subgroup of PG12(R) and K a compact

subset of H2. Only finitely many elements of r have fixed points on K.
Proof For all but finitely many a E F the sets K and a(K) are disjoint. It
follows that only a finite number of elements from r have fixed points on K. 0
By an elliptic fixedt oint for r on H2 we understand a point z E H2 which

is fixed by some proper elliptic element a c F.
Corollary 1.11 Let F be a discrete subgroup of PG12(R). The set P of

elliptic fixed points for F on H2 is a discrete subset of H2.
IV.! DISCRETE SUBGROUPS 129
APPENDIX: DISCRETE SUBSETS
Let us consider a locally compact space X, i.e. a topological space X which

is Hausdorff and in which every point has a compact neighbourhood. We say that
a subset S of X is discrete if S is locally finite in X, meaning that every point
x E X has a neighbourhood which meets at most finitely many points from S.
Observe that a discrete subset S of X is closed and that every point v E S
is an isolated point of S in the sense that v has a neighbourhood V in X with
V fl S={v}. Conversely, a closed subset S of X consisting entirely of isolated
points is discrete.
Let us say that a point w E X is an accumulation point for S if every
neighbourhood of w contains infinitely many points of S. A discrete set in S has
no accumulation points in X and conversely, a subset S of X without accumulation
points in X is a discrete subset of X. Discreteness of a subset S can be expressed
geometrically as follows
Proposition 1.12 A subset S of a locally compact space X is discrete if

and only if S intersects every compact subset of X in a finite set.
Proof This follows immediately from the Borel-Heine theorem. 0
By a discrete subgroup r of a locally compact group G we understand a

subgroup I' of G which is a discrete subset of G. Discreteness of subgroups can be
detected as follows
Proposition 1.13 Let r be a subgroup of the locally compact topological

group G. If there exists a neighbourhood V oft in G such that V n r = {r} then
r is a discrete subgroup of G.
Proof Let V be an open neighbourhood oft in G with V n r = {t}. Let us

consider a point o E G with the intention of constructing an open neighbourhood
of o in G which meets F in at most one point. Consider the alternatives
cV-'nr=0 , oV-'nF00
The first alternative gives us the open neighbourhood oV-' of o disjoint from r.
The second alternative allows us to pick a y E oV-' n F. From o E yV and y E F
we conclude that yV n F = {y}:
y'( yVnr) = vny'r = Vn r = {t}
In conclusion the open neighbourhood yV of o meets r in the point y only.
Corollary 1.14 Let the locally compact group G act continuously on the
topological space X, and let r be subgroup of G. If there exists a point x E X,
isolated in its orbit rx, and such that the stabiliser rx is discrete in G, then IF is a
discrete subgroup of G.
Proof Let U be an open neighbourhood of x in X such that U n rx = {x}.

Then the inverse image of U along the map r,(:G->X, gHgx, satisfies
TX'(U) n r = rX
Let us pick an open neighbourhood of V oft in G such that V n F. = {t}. The
open neighbourhood V n rX'(U) isolates t in r, and 1.13 applies.
IV.2 ELEMENTARY SUBGROUPS
Let us start the study of subgroups of PG12(R) with a list of its so called
elementary subgroups. By this we mean a subgroup G which fixes a point of H2,
fixes a point of 8H2 or stabilises a geodesic in H2. We shall see that a subgroup
G of PG12(R) is elementary if and only if the group G is solvable.
Proposition 2.1 The stabiliser of a point of H2 under the action of the

group G12(R) is conjugated to a group of matrices of the form
a b a b
;a2+b2#0
-b a b -a
PG12(R) acts on H2 with stabilisers conjugated to P02(R) = 02(R)/{ - t}.
Proof It is easily seen that the matrices from the list above stabilises the point
i E H2 considered in the proof of 1.1. From formula 1.4 follows that matrix [a d]
stabilises i if and only if
a2+b2+c2+d2 = 2Iad-bcl
When ad - be > 0, this can be untangled via the identity
(a - d)2 + (b + c)2 =a2+b2+c2+d2-2(ad-bc)
The case ad - be < 0 is handled in a similar manner. It follows that the stabiliser
of i in G12*(R) = {o E G12(R) ( dela = ± 1} is 02(R). This shows that the
stabiliser of i in PG12(R) is 02(R)/{-t}.
Proposition 2.2 The stabiliser in G12(R) of a given geodesic k in H2 is

conjugated to the subgroup of G12(R) represented by matrices
Ia 0 0 b
;ad$0,be960
0 d c 0
132 FUCHSIAN GROUPS
Proof Let y be the geodesic with normal vector N =[o 01. Let us show that
the matrices we have listed above make up the set of transformations which acts
on s12(R) with N as eigenvector. Direct calculation gives us
0 11 d -b ad+bc -2ab
-1 -c a 2cd -ad-be
The details are left with the reader.
Proposition 2.3 The stabiliser in G12(R) of an end of H2 is conjugated to

the subgroup
a b
;ad#0
0 d
Proof it is cleare that a matrix of the above form stabilises the special end
00 = Eo of Conversely, let the matrix a, E G 12( R) stabilise the en d oo. Direct
calculation give s us
a b 0 1 d -b -ac a2
c d 0 0 -c a -c2 ac
I
From this we conclude that c = 0, compare III.4.3.
Discrete groups Let us find the discrete elementary groups. From 2.1 we
find that a group I' which fixes a point is discrete if and only if I' is finite. From
2.2 we find that a discrete group I' which stabilises a geodesic k is discrete if and
only if the translation subgroup is cyclic. In the case of a fixed point on 8H2 we
have the following result.
Proposition 2.4 Let r be a subgroup of PS12(R) which fixes a point of

OH2. The group is discrete if and only if it is cyclic.
Proof Let us consider a group generated by the two matrices

u r v s
i
T = ;u>l,v>1
0 u-1 0 v1
We ask the reader to verify that after conjugation by a matrix of the form [1 i]
we may assume that r = 0. Let us study the group generated by o and the
commutator y = orO 1T 1
1 c
;u>1
0
I
1
Direct computation yields the following family of elements of I'
1 c
; nE?L
0
which shows that I' is a non-discrete group unless u = 1 or c = 0.
Let us present an important class of examples of non-discrete groups. The

material is presented as a lemma and will serve as such in section IV.4.
Lemma 2.5 Let r be a non-elementary subgroup of PS12(Ot) containing

translations o and r whose translation axes h and k are distinct but have a
common end oo. Then the rotations in I' accumulate at t.
Proof Let us first show that r contains a. translation p which doesn't fix oo:
pick p E I' which doesn't fix oo. The geodesics p(h) and p(k) have p(oo) as an
end. The second end of one of these geodesics must be different from oo. It
follows that one of the translations pap-1 and µrµ 1 doesn't fix oo.
The group r of horolations in r with centre oo is non-discrete as we saw
during the proof of 2.4. Let us investigate the commutator 71Pil 1p 1 as r/ varies
through I'... To this end we put
a b 1 e
P=
c d 11 0 1
We ask the reader to verify that

tr(rlp1i1 p-1) = 2 - c2e2 ;eER
By assumption we have c # 0 and we conclude that tr2(qpr 1p 1) < 4 for e $ 0
sufficiently small.
Solvable groups We proceed to make some investigations of group theore-

tical nature based on the following fundamental principle.
Normaliser Principle 2.6 Let X be a set on which a group G acts.

For any normal subgroup N of G the fixed point set XN is stable under G.
Proof For x E XN and y E G we have that

v(.Yx) = y(7 1v7)x = "Yx
which shows that yx is a fixed point for N.
Proposition 2.7 A subgroup of PG12(R) is elementary if and only if it is

a solvable group.
Proof The elementary groups we have displayed in the three first propositions
of this section are easily seen to be solvable. Conversely, let us consider a solvable
subgroup G # 0 and form the derived series of commutator groups
G=D°GDD1GJ ...Di-1GJDT'GD D°+1G={t} DG {t}
Let us first assume that G consists of even transformations. We shall make use of
the classification of even isometries and proceed accordingly.
10 D"G contains a rotation o with centre A. This means that the fixed
point set for o on H2 consists of A only. Let us prove by decreasing induction that
the fixed point set for D'G on H2 is {A}, i = 0,...,n. For i = n observe that the
group D"G is abelian and use the "normaliser principle" 2.6. The induction step
is likewise conducted by means of 2.6.
2° DG contains a horolation o with centre P E (9H2. It follows that the
fixed point set for o on OH2 is {P}. We can proceed as above and conclude from
2.6 that the fixed point set for G on OH2 is {P}.
3° DG contains a translation o $ t along the geodesic k. We shall utilise
the action of G on the set g of geodesics in H2; let us show that gjo = {k}. Let 0
stabilise the geodesic h which means that the set {A,B} of ends for h is stabilised
by o,. It follows that o2 fixes A and B. Observe that v2 $ t has precisely two
fixed points on OH2 and conclude that A and B must be the ends for k, thus h = k.
We conclude as above that the fixed point set for G on fj is {k}.
Let us now return to the general solvable group G and observe that G+ is
solvable as well. When G+ # {t} we can assume that G+ has precisely one fixed
point on either one of the following three sets H2, OH2 or g and we can use the
"normaliser principle" 2.6 to conclude that G has a fixed point as well. 0
Let us investigate various types of groups which lack hyperbolic elements:

groups of elliptic elements and parabolic elements.
Proposition 2.8 A non-elementary subgroup G of PGI2(R) contains a

translation.
Proof Let us make the general remark that the commutator of two rotations o
and r is always a translation. To see this, let their centres be A and B and let /3
denote reflection in the geodesic through A and B. Let us use the theorem on
three reflections 111.3.1 to write o = aO and r = /37 where a and y are reflections
in geodesics through A and B. Direct calculation gives us
orU 1T 1 = a,(3,3"Y(3a"y0 = (a7)33)2
Using that ay,(i that is a glide reflection, compare 111.3.8, we conclude that its
136 FUCHSIAN GROUPS
square orc 1r 1 is a translation. Thus oro'r = t. 1
The same argument shows that the commutator of two horolations is a

translation and that the commutator of a rotation and a horolation is a
translation.
Let us now assume that G doesn't contain translations. This implies that
that G+ is abelian and that G is solvable. It follows from proposition 2.7 that G
is an elementary subgroup.
Proposition 2.9 Let G be a non-elementary subgroup of PG12(!R). If

G : G+ then G contains a glide reflection.
Proof Let us assume that all odd transformations from G are reflections.
Pick a non-trivial transformation o E G+. If o is a rotation around the point A,
then we conclude from 2.10 that all reflection axes of elements of G pass through
A, and we conclude that A is fixed by G. If o is a horolation with centre S, then
we conclude from 2.10 that all reflection axes from G pass through S, which
implies that G fixes S. If o is a translation then we conclude from 2.10 that all
reflection axes from G are perpendicular to the translation axis s of o. This
implies that s is stabilised by the group G.
Lemma 2.10 Let n; be a reflection in a geodesic k and o an even isometry of

H2. If tco is a reflection, then o is a rotation around a point of k or a horolation
with a centre at one of the ends of k or a translation with axis perpendicular to k.
Proof Observe that tr2(n;o) = 0, exercise 11.8.1, and use explicit calculation in
the Poincare model taking for k the y-axis. See also exercise IV.3.1.
IV.3 GEOMETRY OF COMMUTATORS
In this section we shall work out a number of trace formulas for the
composition of two even isometries and for their conmmutator.
Proposition 3.1 Let a and ,Q be translations whose axes a and b have a

common perpendicular. If a and Q have the same translation length T, but are
oppositely directed, then with d = d(a,b)
ir2(a,Q) - 4 = 16 sinh2(IT) sinh2(1-2d) [Sinh2(ZT) sinh2(-12d) - 1]
Proof Let us represent a by the matrix A E S12(R) with it A > 0. This allows
us to assign to a the transformation vector U= 2(A' - A). Observe that
it U = 0, otherwise expressed, U E s12(R). The axis a of the translation a has a
natural orientation and M is defined as the corresponding normal vector. We shall
prove that U and M and the translation length T are related through
3.2 U = sinh(1T)
2
M 12 it A = cosh(21T)
To see this, let us pick points P and Q on a such that Q is the midpoint of P and
a(P). Let H and K be positively directed unit tangent vectors to a at P and Q
respectively
138 FUCHSIAN GROUPS
From the parametrisation of a given by Pcosh(s) + Hsinh(s), s E R, we get that

Q = Pcosh(2T) + Hsinh(IT) , K = Psinh(2T) + Hcosh(IT)
The second of these two formulas gives us
H A K = H A P sinh(2T) , <H,K> cosh(ZT)
Let us use 111.4.5 to represent a by the the matrix A = KH. This gives us
2(A- - A) = z(HK - KH) = H A K , tr A = -<H,K> = cosh(ZT)
2
Recall from 111.2.1 that M = H A P, which concludes the verification of 3.2.
Similarly, let us represent ,0 by a B E S12(R) with tr B = cosh(2T) and
2
V= 2(B' - B). The normal vector for the natural orientation of the axis b for ,0 is
denoted N. The fact that M and N represent opposite directions can be expressed
by <M,N> > 0. As a consequence of II1.2.8 we have that
<M,N> = cosh d
Let us substitute this information into lemma 3.3 below to get
z tr(AB) = cosh2(IT) - sin/t2(ZT) cosh d =
1 + sinh2(ZT) - sinh2(ZT)(1 + 2 sinh2(Zd)) _
1 - 2sinh2(2T) sinh2(d)
Finally, square this relation and the announced formula follows. 0
Given matrices A,B E M2(R). The transformation vectors U = Z(A- - A)

and V = Z(B- - B) satisfy
3.3 2tr(AB)=2 IrA 1IrB-<U,V>

Proof The elementary relation A + A' = t trA gives us U = 2(t irA - 2A) and
similarly V = 2(t trB - 2B). Straightforward calculation yields
-2<U,V> = 1r(UV) = a tr[(t trA - 2A)(t trB - 2B)] _
Irt trA trB + trAB - 4trA 1r2B - 4Ir2A trB
from which formula 3.3 follows. 0
Corollary 3.4 Let a and )3 be translations with the same translation

length T whose axes a and b have a common perpendicular. If neither a/3 nor
aa-' is a rotation, then
sinh sinh. >1 ; d = d(a,b)
zT zd
Let us now turn to commutators. It is useful to remark that the symbol

tr(a,6a 1,(3-1) E R is a well defined function of E PS12(IR) since ir(ABA-1B-1)
is independent of representatives A and B for a and 13.
Lemma 3.5 Any two transformations a,Q E PS12(R) satisfy

2 tr(6aQ 1a 1) = q tr2a +
where the vector U is given by U = z(A--A) for A E S12(R) representing a.
Proof Let us represent a and /3 by matrices A and B. The transformation

vector for A is U = !(A` - A), while the transformation vector for BAB-1 is
2((BAB")-- BAB-) = z BA-B-- BA') = BUB'1
and the transformation vector for A"1 is -U. Let us substitute this into 3.3
Ztr(BAB"1A-1) = 2 irBAB'1 IrA-1 + <BUB-1,U>
which concludes the proof.
We shall now investigate the commutator of two translations a and 0

with the same translation length. This situation occours for example when a and
,Q are conjugated.
140 FUCHSIAN GROUPS
Proposition 3.6 Let a and /3 be two translations with the same

translation length T. If the axes a and b for a and /3 intersect at an angle
0 E ]O,ZaJ then
1r2 (/3a/3"la"1) - 4 = 16 sinh4(IT)
2
sin20 [sinh4(IT)
2
sin 20 - 1 ]
Proof From the formulas 3.2 and lemma 3.5 we deduce that
2 tr(,3a/3"la 1) = cosh 2(IT)
2
+ sinh2(IT)
2
where M is a normal vector for a. Observe that /3(M) is a normal vector for /3(a)
From 111.2.8 we get with d(a,)3(a)) = d that

- <M,/3(M)> = cosh d(a,/3(a)) = 1 + 2 sinh2(d)
The trigonometry of the right angled triangle, 111.5.1, gives us
sinh(2) = sin 0 sinh.(ZT)
Substitute this into the formula above to get that
z tr (aaa" la 1) = cosh2(2)
1 1T -sinh2(z1T)(1+ sin20
? sinh2(z1T))= 1-2sin20 sinh4 (z1T)
Finally, square this formula to obtain the desired result. 0
Corollary 3.7 Let a and /3 be two translations with the same translation
length T, whose axes a and b intersect at an angle 0 E ]O,Iir]. The commutator
/3a/3-la' is not a rotation if and only if
sinh2(2T) sin 0 > 1
IV.4 JACOB NIELSEN'S THEOREM
We have seen that a discrete subgroup r of PG12(Il;) acts on H2 with a

discrete set of elliptic fixed points. We shall prove the converse for a non-
elementary group.
Jacob Nielsen's theorem 4.1 A non-elementary subgroup r of

PG12(R) is discrete if and only if the set P of elliptic fixed points on H2 is discrete.
Proof It follows from 2.7 that a subgroup r of PG12(Id) is elementary if and

only if r+ is elementary. Thus we may assume that r is Fuchsian.
Let us assume that P is non empty. Pick a point z E P and observe that
its orbit under r is discrete since it is a subset of the discrete set P. According to
1.5 it suffices to prove that FZ is finite. To see this, use that r is non-elementary
to pick a second point w E P, w $ z. The orbit of w under FZ is finite since it is a
discrete subset of the circle S' with centre z through w. We can use that the
action of F. on S1 is faithful to conclude that FZ is finite as required.
In the rest of the proof we assume that F is non-elementary and contains
no rotations. It follows from 2.8 that r must contain proper translations. Let us
pick a geodesic s such that the subgroup F3 made up of translations along s is non-
trivial. Since r is non-elementary we have r # F3. For r l FS we have r(s) $ s
since we can rule out half-turns. It follows from 2.5 that r(s) and s will intersect
or have a common perpendicular; we shall examine the two cases separately.
When s and r(s) have a common perpendicular, let us introduce the dis-
tance d between s and r(s). For a non-trivial c E FS let us apply corollary 3.4 to v
and rar 1 and deduce the inequality
4.2 sin.h(2T0.) sinh(2d) > 1 ; v E F3 , a 54 c
When s and r(s) intersect at an angle 0 E ]0,2] say, we can apply corollary 3.7 to
o and 7-or-1 and deduce the inequality
4.3 sinh 2(ZTo) sin(g) > 1 o E r, , o # t
Let us fix r for a moment. We can then use 4.2 and 4.3 to conclude that the
numbers T. are bounded away from 0 as o runs through r,-{t). It follows that
IF, is discrete, in fact, r, is cyclic generated by an element o of translation length
T, say. For any r E r such that s and r(s) have a common perpendicular we get
from 4.2 that
4.4 sinh(2T) sinh(2 1) > 1 ; d = d(s,r(s))
For any r E I' such that s and r(s) intersect we get from 4.3 that
4.5 sin.h2(2T) sin(O) > 1 ; 0 = L(s,r(s))
We shall now study the action of r on N = N(s12(68)), the space of vectors in sl2(lt)
of norm -1. Let us pick a normal vector N for s and observe that the stabiliser
for N is r,. Let us recall from 111.2.5 and 111.2.8 that for the inner product
between N and rN we have respectively
(<N,rN>l = cosh d , I<N,rN>l = cos 0
We conclude from 4.4 that d is bounded away from 0 and from 4.5 that 0 is
bounded away from 0. It follows that we can find a constant c>0 such that
I<N,rN>I 0 ]1 -c,l +c[ ; r o rN
From this we conclude that N is isolated in its orbit rN in N. Recall once again
that rN is discrete and conclude from 1.14 that r is discrete.
A slight modification of the proof of Jacob Nielsen's theorem yields the

following variation due to C.L. Siegel.
Theorem 4.6 A non-elementary subgroup I' of PG12(R) is discrete if and

only if the elliptic elements of r do not accumulate at t.
Proof Let us assume that r is non-elementary and that the elliptic elements do
not accumulate at t. Let us use 2.8 to pick a proper translation o E I' with axis s,
say, and introduce the group rs made up of the translations in I' with axis s. Let
us now pick a neighbourhood V of t in PG12(R) such that
<N,7-N> < 0 and TUT 1u 1 is not a proper rotation ; o,T E V
Let us analyse a pair of elements T E V-rs. o E rs fl V, o # t. Observe that T is
not a half-turn around a point P of a half-turn 1r around P satisfies
s:
<N,irN> = 1. We conclude that T(s) # s. Let us remark that s and T(s) cannot
have a common end as follows from 2.5 applied to o and 7017 -1. If T(s) intersects
s, then we can observe that
U(TU lr1)U 1(TQ 1r1)-1 = (TUT 1o 1)-2
is not a rotation and we conclude from 3.4 that inequality 4.3 holds. If T(s) and s
have a common perpendicular then we conclude from <N,7-N> < 0 that c 1 and
Tor 1 translate in opposite directions, compare the proof of 3.2. We conclude from
3.1 that inequality 4.2 holds. This is sufficient to conclude that rs is discrete, so
let us pick a definite generator o E rs and change the neighbourhood V of t such
that
<N,TN> < 0 and To7'1o 1 is not a proper rotation ;TEV
At this point we leave the remaining details to the reader.
IV.5 CUSPS
In this section we shall be concerned with the action of a discrete group r

on the boundary of the hyperbolic plane H2. We say that a point S E 8H2 is a
cusp for H2 if it is fixed by some proper parabolic transformation 63 E r. The
stabiliser I'S of a cusp S E OH2 consists of horolations and reflections in geodesics
through S, 2.4. By a horodisc we understand a region in H2 bounded by a
horocycle.
Proposition 5.1 Given a cusp S E 8II2 for the discrete group I' of
isometries of H2. There exists a horodisc D with centre S E H2 such that
y(D) n D= O ; y E t -Fs
Proof We shall work with the point oo of the Poincare upper half-plane.
Moreover, we shall assume that the group 1 s+ is generated by A = [o i]. Let
B = [ a] E G12(R) with det B = f 1 represent 13 E FF. According to lemma 5.3
we have jcj > 1. An elementary estimate based on 11.8.3 and 11.8.4 gives us
Im[/3(x + iy)] = y Ic(x ± iy) + d 1-2 < y c 2y 2 < y-t
Otherwise expressed
5.2 Im[z] Im[9(z)] < 1 ; z c H2, ,3 E t-I oo
It follows that the horodisc Im[z] > 1 will serve our purpose.
Lemma 5.3 Let us consider two matrices A and B from G12(R)
B= c#0, detB= ± 1
If A and B generate a discrete subgroup of PG12(R) then Ici > 1.

IV.5 CUSPS 145
Proof Let us proceed under the assumption that Icl < 0. In order to derive a
contradiction we consider the sequence of matrices (Bn)n E N given by BO = B and
1
Bn+1 = BnABn ; nEN
We ask the reader to establish the recursion formula
an+1 bn+1 1-ancn an 2

2
;n>1
Cn+1 do+1 -cn 1+ancn
A simple induction on n E N gives us that
icnl = Ic012n and n + 1aol nEN
Using co = c and Icl < 1 we conclude from this that c11-+0 and ancn-+0 for n-+oo.
Put this information into the right hand side of the recursion formula and
conclude that Bn-+A for n-+oo. Let us assume that the corresponding classes a
and (3 generate a discrete subgroup of PG12(R). Then we get that fl. = a for n
large and deduce, by descending induction on n, that oo is fixed by
Since 0 = Qo this contradicts the assumption c t- 0.
With the notation of 5.1, let us observe that the orbit of a point A E 8D
does not meet D: for y E rs we have y(OD) = OD and for y E F-I s we find that
y(8D) and D are disjoint.
Proposition 5.4 Let S E 8H2 be a cusp for the discrete group r of iso-
metries of H2. For any compact subset K of H2 there exists a horodisc D with
centre S E H2 such that y(D) fl K = 0 for ally E F.
Proof Let us return to the proof of 5.1 and choose r>1 such that K lies in the
vertical strip of the Poincare half-plane bounded by Im[z] = r and lm[z] = r"1. For
y E FS we have y(D) = D and for y E f -I s we have y(D) fl K = 0 as a
consequence of the inequality 5.2.
146 FUCHSIAN GROUPS
Corollary 5.5 Given cusps S,T E 49H2 in different 17-orbits, for any horo-
disc E with centre T there exists a horodisc D with centre S such that
y(D)f1E=0for all yEr.
Proof Let us pick a compact subset K of dE such that FTK = E. Use 5.4 to
choose a horodisc D with centre S such that D fl rK = 0. This gives rD f OE = 0.
For y E r we conclude from the connectedness of yD that yD C E or yD fl E = 0.
The first option can be ruled out on the grounds that 7(S) # T.
Corollary 5.6 If H2/17 is compact, then the group r contains no proper

horolations.
Proof Let us first observe that the projection H2-H2/r maps open subsets to
open subset. Next, cover H2 with open discs and apply the Borel-Heine theorem
to the images of the discs in the compact space H2/r. The upshot is that we can
find a finite set of discs in H2 which meet all r-orbits. It follows that we can find
a compact set K in H2 which meets all r-orbits. We conclude from 5.4 that F
contains no horolations.
Horocyclic Topology For a discrete group r of isometrics of H2 let us

consider the subset Y of H2 consisting of all ordinary points of H2 and the set of
cusps for F. Let us introduce the horocyclic topology on Y: a subset W of Y is
open if for all points S E W there exists a disc/horodisc with centre S entirely
contained in W. The horodiscs with centres at a cusp S E OH2 form a fundamental
system of neighbourhoods for S, while the ordinary discs with a given centre
A E H2 form a fundamental system of neighbourhoods of A E Y.
The group r acts continuously on Y, which gives the orbit space X = Y/r
a structure of topological space. This new space is a two-dimensional topological
manifold with boundary:
IV.5 CUSPS 147
Proposition 5.7 With the notation above, the topological space X = Y/r
is a Hausdorff space in which every point has a neighbourhood homeomorphic to a
neighbourhood of 0 in R2 or a neighbourhood of 0 in R x [0,oo[.
Proof The space x = Y/r is Hausdorff as follows from 5.4 and 5.5. In order
to construct a neighbourhood of a cusp point, it suffices to treat the point oo of the
Poincare half-plane. It follows from 5.1 that it suffices to treat the case where
r = rte. Let us assume that r is generated by zi--+z+k, k>O. The space Y/r is
homeomorphic to the open unit disc D in the complex plane
Y/r - + D , z i-+ exp(21rik-1z) , z E H2
In the general case we can arrange for r to be generated by zF + - i and zF--*z + k
as above. In this case we find that the orbit space X is homeomorphic to the orbit
space for the action of complex conjugation on the open unit disc D in C.
In order to investigate a finite point, it suffices to treat the origin 0 in the
Poincare disc D. According to 1.5 (for more details see VI.5.8) we may assume
that the full group r fixes 0. Let us assume that r+ is generated by zHBz, where
0 is a primitive nth root of unity. Using the well known fact that z-*z" maps open
set to open set, we can make the identification
H2/r+=, D
A non-trivial element of r/r+ corresponds to reflection of D in a line through 0.
Corollary 5.8 The space X = Y/r is compact if and only if there exists a
compact subset K of H2 and a finite number of horocycles D1,...,Dr whose centres
S1.... ,Sr are cusps for r such that any r-orbit in H2 meets K U D1 U ... U Dr.
In the case of a Fuchsian group r, the proof of 5.7 provides a structure of
Riemann surface [Farkas, Kra] for the space X = Y/F.

When the space X is compact, it is called the horocyclic compactification
of H2/r, compare VII.5. For general information see [Lang] and [Shimura].
IV EXERCISES
EXERCISE 1.1 Let 1' be a subgroup of G12(R) and put A = 1' fl S12(R). Let P1'
and PO denote images of I' and A in PGI2(R) and assume that [PI':PD] is finite.
Show that PI' is discrete in PG12(R) if and only if PA is discrete in PG12(R).
EXERCISE 1.2 The additive group U8 of real numbers acts on R/27rZ. Show that
the i-orbit of 1 mod 2a is not discrete.
EXERCISE 2.1 1° Show that the group of rotations around a point A E H2

equals its normaliser in PS12(IR).
2° Show that group of even isometries fixing a given geodesic line k equals its
own normaliser in PS12(IR).
3° Show that the group of even isometries fixing a given end S equals its own
normaliser in PGl2(E).
EXERCISE 2.2 Let a and ,3 be translations whose axes have a common perpen-
dicular. Show that the commutator a/3a-1/3-1 is a translation.
EXERCISE 2.3 1° Show that if a subgroup F of PSI2(E) has a finite orbit on H2

of order n, then n = 1. Hint: If n > 2, show that yn! = t for all y E G.
2° Show that if a subgroup r of PS12(IR) has a finite orbit of order n on OH2,
yni
then n = 1,2. Hint: If n > 3 show that = t for ally E I'.
EXERCISE 2.4 Let n. E PG12(R) be a proper translation with fixed points A,B.
1° Show that for any point Q 0 {A,B} the set {t:n(Q) I n E Z } accumulates
IV EXERCISES 149
1rof
at A and B. Hint: Take A=O and B=oo, in which case K =
110, 1
20 Show that a non-empty ic-stable closed subset Z of R contains A and B.
EXERCISE 3.1 Let S E G12(R) and put U = 2(S' - S). Show that U E s12(R) is
an eigenvector for or and use detS = (2 trS)2 - trS2 (exercise 1.3.1) to prove that
<U,U> = a det S (4 - tr2 S)
2° For a second matrix K E s12(R) prove that
<U,K> = IrSK -1 IrS trK
3° Use this to give an alternative proof of lemma 2.10.
EXERCISE 4.1 1° For A E H2 and r > 0 fixed, prove that the set below is
compact
{ N E N(s12(R)) (I<N,A>l < r }
Hint: Pick a basis S,T for the tangent space of H2 at A and use that any point of
F can be written in the form N = xA + yS + zT , x,y,z E R.
2° Let `ll be a discrete subset of N(42(R)). Given a point A E H2 and r > 0,
show that only finitely many geodesics in H2 with normal vectors in 9 will meet
the hyperbolic disc with centre A and radius r.
V FUNDAMENTAL DOMAINS
The most important way to get information about a discrete group r of

isometries of H2 is through fundamental domains. The art of reading information
about discrete group from a fundamental domain dates from the last century. A
good example for this is the "modular figure" which served Gauss as a
fundamental domain for G12(Z). The modular figure can be constructed by a
general method due to Dirichlet to be presented in V.4.
The finer analysis of a fundamental domain involves a decomposition of
its boundary into linear pieces which are paired together through side trans-
formations. A good fundamental domain leads to a representation of the group by
generators and relations, compare Poincare's theorem in chapter VII.
V.1 THE MODULAR GROUP
By a fundamental domain U for a group t of isometries of H2 we

understand an open subset U of H2 such that U and 7(U) are disjoint for ally # r,
and such that each t-orbit meets U. As a basic example, let us construct a
fundamental domain for the modular group PS12(7L).
= erp2ai/3
V.1 TIIE MODULAR GROUP 151
The modular figure 1.1 The three geodesics in the upper half-plane H2
Re[z]=2, Re[z]=-2, zj = 1
bound an open set M which is a fundamental domain for the group PSl2(7).
Proof Let us fix a point w = yi, y > 1 and introduce the three transformations
o(z) = -z 1, T(Z) = z + 1 , W(Z) = z - 1
We ask the reader to verify that the three geodesics bounding M are the per-
pendicular bisectors for w,w(w), w,o(w) and w,T(w). In order to find a point of M
in the orbit through a given point z E H2 we shall minimise the hyperbolic distance
d(w,p(z)), p E PS12(7L). Precisely if v E S12(7L) is chosen such that
d(w,v(z)) < d(w,p(z)) E S12(7)
Then v(z) E M as follows from the inequality
d(w,v(z)) < d(w,c,i 1(z)) = d(w(w),v(z))
and similar inequalities with w replaced by o and r.
Let us consider an orbit 0 of S12(Z) in H2 and let us show that Im[z] is a
bounded function of z E 0 which assumes its upper bound whenever z E M. In
concrete terms, let us consider a z E M and p E r represented by
Let us note the inequality

Icz +d F = c21zI2 + 2Re[z] cd + d2 > c2 + d2 - jcdj = (Icl - Id 1)2 + IcdI > 1
and draw the conclusion from the elementary formula
f
Im [ bz + d] = Im.[z] Icz + d1-2 del I
c b
When jzi > 1 and c # 0, then the inequality above is strict w rich gives us that
Im[,u(z)] < Im[z]. When c = 0, the transformation µ takes the form zHZ+b and
Im[z] is preserved. We conclude that the orbit 0 meets M in at least one point
and M in at most one point as required. Taking into account that M is open, we
find that an orbit which meets 8M will not meet M. We ask the reader to verify
that the figure M, which is the union of M and the part of 8M which lies in the
half-plane Re[z] > 0, meets each S12(71)-orbit in precisely one point. The identi-
fication of 8M can be realised by the transformations o,T introduced above.
152 FUNDAMENTAL DOMAINS
1.2
We shall see in the next, section that the modular group is generated by o and T.
The transformation p = ro is a rotation of order 3. Thus we have the relations
1.3 2=t , p3
I
0" i p=To-
It follows from Poincare's theorem VII.3.7 that this set of relations is complete.
The hyperbolic triangle A with vertices i, C, oo is a fundamental domain
for the extended modular group PG12(7L) as follows from the fact that reflection 6
in the y-axis divides the fundamental domain for the modular group into two
halves. Let us introduce reflections a,/3,ry in the sides of A as given by the figure
Let us construct a fundamental domain for some subgroups of PS12(7)

using the general lemma
V.1 THE MODULAR GROUP 153
Lemma 1.4 Let I' denote a discrete group of isometries of H2 and II a

subgroup of F of finite index. If D is a fundamental domain for I' and S C F is a
full set of representatives for II\F, then the interior U of
F= UaESaD
is a fundamental domain for H.
Proof Since S is finite, we conclude that F is the closure of Ua ES

aD.
Since this set is open, it must be contained in U (the interior of F), and it follows
that F = U. Let us show that the II-orbit of a given point z E H2 meets F. To
this end choose y E F with -yz E D and write y 1 = 7r-1a with a E S and 7r E II to
get that a-1az E D or irz E A.
Let us analyse a 7r E II and a point u E F with 7ru E U. Pick a E S such
that u E aD and let us observe that a-1F is a neighbourhood of u. It follows that
a 1F meets aD or F fl iraD 0 0. Since F is the closure of U a,OD , we conclude
that 7raD meets ,6D for some Q E S. This means Ira = 0 which implies a = t.
The level 2 modular group Let us describe a fundamental domain

for the group G(2) given by the exact sequence
1.5 0 G(2) - PG12(7) -f G12(F2) ' 0
The group G12(F2) is isomorphic to the permutation group S3 through the action
of G12(F2) on the three lines in F2 ®12. This group is lifted back to PG12(7L) by
the dihedral group D3 generated by a and y whose matrices are
1 -1 1 , [ 0 1
0 1 1 0
According to lemma 1.4 the union of the six translates of z by D3 is a

fundamental domain for G(2). It follows from the illustration above that this
union is the hyperbolic triangle 0,-1,c .
The level 2 modular group I'(2) = G(2)+ has fundamental domain the
quadrangle 1,0,-l,oo as follows from 1.4 applied to the reflection 0. We shall see
in the next section that r(2) is generated by
z
v(z) =
µ(z) = z + 2 ,
-2z+1
In fact we shall see as a consequence of Poincare's theorem VII.3.7 that the level 2
modular group I'(2) is free on these two generators.
I'(2) = G(2)+
-1 0 1
V.2 LOCALLY FINITE DOMAINS
Let I' denote a discrete group of isometries of the hyperbolic plane H2 and
let us consider a fundamental domain P C HZ which is locally finite in the sense
that any point z of H2 has a neighbourhood which meets o(P) for at most finitely
many a E F. We shall relate the group theory of I' and the topology of H2/F to
the geometry of P.
Proposition 2.1 Let P C H2 be a locally finite fundamental domain for

the discrete group r . The canonical projection is a homeomorphism
p:P/r=; H2/r
Proof Since our map p is continuous, we must prove that p maps open subsets
of P/r to open subsets of H2/r. To this end it suffices to show that for any given
(relative) open subset B of P the set
V=UyEry(B)
is an open subset of H2. It suffices to consider a point b E B and construct an
open disc with centre b entirely contained in V. Let us use the fact that P is
locally finite to see that the following subset S of r is finite
S={aEIF Ivl(b)EP}
Since P is locally finite we can choose a>O so small that
D(b;c)na(P)=0 ; o¢S
The disc D(b;c) is contained in the union of all IF-translates of P, thus
D(b;) C Uo E S o(P)
Let us take advantage of the fact that S is finite to decrease e > 0 further to get
D(Q 1(b);e) (l PC B ; or E S
Let us combine these two pieces of information to get
D(b;e) = Uo E S o(P) n D(b;e) = Uo E
So(P n D(v 1(b);e)) C Ua E So(B)
This concludes the construction of the required disc. 0
Corollary 2.2 Let P C H2 be a locally finite fundamental domain for the

discrete group IF. Then H2/I' is compact if and only if P is bounded.
Proof A closer look at the proof of 2.1 reveals that the projection
P onto open subsets of H2/I'. If H2/F is compact
then we can use Borel-Heine to find a finite set of open discs D1,...,Dn with
centres z1,...,zn in P such that each r-orbit in H2 meets one of the sets
P fl D1,...,P fl D. It follows that P is contained in the union of these discs.
Proposition 2.3 Let P C H2 be a locally finite fundamental domain for

the discrete group r of isometries of H2. The group r is generated by
{yErI7(P)flPs0)
Proof Let H denote the subgroup of r generated by this subset. For z E H2

choose o E IF with z E o(P) and observe that the class [o] E 17/11 is independent
of o. In other words we have defined a map
p : H2 -> r/H
This map is surjective since the I'-translates of P cover H2. The map a is locally
constant: at a given point z E H2 we can use the fact that P is locally finite to
find a finite subset S of F such that z E cr(P) for all o E S and such that
N = Uo E So(P) is a neighbourhood of z in H2; it follows that p takes the value
[o], o E S. We can use the fact that H2 is connected to conclude that p is
constant. Taking into account that p is surjective, we conclude that H = F.
Let us conclude this section with some results on unbounded fundamental

domains. The problem is to relate cusp points for r to "boundary" points of a
locally finite fundamental domain P. The horocvclic boundary 8hP is the set of
points R E OH2 such that any horodisc with center R meets P. We ask the reader
to compare this material with section IV.5 on cusps.
Proposition 2.4 Let P be a locally finite fundamental domain for 1' and
A E OH2 for which there exists a point L E P such that the geodesic segment [L,A[
is contained in P. A transformation y E r+ with y(A) = A is parabolic.
Proof The transformation y is elliptic, hyperbolic or parabolic. A proper

elliptic transformation has no fixed points on 8H2. Let us assume that y is proper
hyperbolic with translation axis h. Upon replacing y by 7-1 we may assume that
y translates towards A. Let the horocycle with centre A through L intersect h in
W and put r = d(L,W). Let us contradict that P is locally finite by showing that
D(W;r) n 7 n(P) i4o ; nEN,n>1
For n > 1, let Ln E [L,A[ denote the point of intersection between [L,A[ and the
horocycle with centre A through yn(W). From the inequality
d(Ln,yn(W)) < d(L,W)
it follows that y n(Ln) E D(W;r) as required. 0
Example 2.5 Let I' denote the group of isometries of the Poincare half-
plane generated by 7(z) = 2z, z E H2. The region P bounded by the two curves
x+i(21xl +1) , x+i(21xl+2) xER
is a locally finite fundamental domain. Observe that y(oo) = oo and that any horo-
disc with centre oo meets P. Thus oo E al,(P) but oo is not a cusp point for I.
Proposition 2.6 Let A E 0112 be a cusp for the discrete group r and D a
horodisc with centre A. A locally finite fundamental domain P for IF meets (at
least one but) at most finitely many F-translates of D.
Proof Let us assume that the horodisc D is closed and use the remarks fol-
lowing IV.5.3 to pick an open horodisc U contained in D such that there exists a
full F-orbit outside U. Let y E F be a parabolic element with fixed point A.
Moreover, we pick a geodesic It emanating from A and let C be the compact subset
of D - U bounded by h and y(h). Observe that D-U is union of translates of C
by powers of y. Let us enumerate the F-translates of P which meet C
ol(P),...,on(P) ; ol...... n E F
Let us now assume that o-1(D) meets P or what amounts to the same thing that
o(P) meets D. Assume for a moment that o(P) C U; this implies o(P) C U which
contradicts the existence of a full F-orbit outside U. In conclusion o(P) meets
D - U. It follows that we can find s E Z such that yso(P) meets C. This proves
that yso = of for some i = 1,...,n, and we conclude that o(D) = o1-y--s(D) = o1(D).
It follows that P meets at most finitely F-translates of D.
Corollary 2.7 Let F denote a discrete group and P a locally finite funda-
mental domain for F. The f-orbit of a cusp meets the horocyclic boundary 01iP.
Proof Let us pick a horodisc K with its centre at a cusp A and let
K, a1(K)...... on(K) ; 0"i,---,0'n E IF
be a complete enumeration of the F-translates of K which meet P. For any
horodisc L contained in K we find that at least one of the horodiscs
L, ol(L)....... n(L) ; 01,...,on E F
will meet P since P meets any F-orbit. From this it follows that at least one of
the points A,o1(A),...,on(A) belongs to 01iP.
V.3 CONVEX DOMAINS
In this section we shall study a group IF of isometries of H2 and a

fundamental domain P, which is convex and locally finite. Recall that a subset R
of H2 is convex' if for any two points A,B E R the geodesic arc [A,B] C R. The
main objective is to relate the geometry of 8P to r. It turns out that although no
polygonal structure is supposed from the outset, the boundary comes out as a
geodesic polygon. Let us start with a basic result on open convex subsets of H2.
Hahn-Banach principle 3.1 Let U be an open convex subset H2.

Through any point A E H2 outside U passes a geodesic not meeting U.
Proof Let S' denote the unit circle in the tangent space TA(H2). Let us define
a map p:U-S' as follows: for Q E U draw the oriented geodesic from Q to A and
let p(Q) E S' denote its unit tangent vector. The map p is continuous and the
image V = p(U) is an open subset of S' which does not contain antipodal points.
Let .A:S'-+S' denote the antipodal map, and observe that V and A(V) are non-
empty open disjoint subsets of S'. Since S' is connected we can find a point s E S'
not in V U A(V). It follows that neither s nor A(s) belongs to V. It follows that
the geodesic through A with tangent vector s does not meet U.
'In terms of Klein's disc model 11.5, the set R is convex in the hyperbolic
sense if and only if R is a convex subset of the ambient Euclidean plane.
Corollary 3.2 Let U be an open convex subset of H2. For points A E U

and B E 8U the geodesic segment [A,B[ is contained in U.
Proof If Q E [A,B[ is not in U we can draw a geodesic h through Q not

meeting U. The closed half-plane H with 8H = h and A E H will contain U but not
B. This makes B an exterior point for U.
Corollary 3.3 Let U be a convex open subset of H2. The closure U is a

convex subset of H2.
Proof Let us first show that we can write

U=UII,HjUH
where the union is taken over the set of all closed half-planes H in H2 containing
U. To this end consider a point w 0 U with the intention of constructing a closed
half-plane H 7 U with w 0 H. Pick a base point u E U and a point z E ]u,w[
outside U and use the Ilahn-Banach principle to draw a geodesic h through z not
meeting U. The closed half-plane H with Oil = h and u E H does the job. The
representation of U as a union of closed half-spaces shows immediately that U is a
convex subset of 112.
Corollary 3.4 Let U be a convex open subset of H2. The interior of the
closure U equals U.
Proof We must show that any neighbourhood of a given point z E SU meets

the complement of U. To see this, use the Hahn-Banach principle to draw a
geodesic h through z not meeting U. Since U is connected, it is contained in a
closed half-plane, say, iI hounding h. It follows that H contains U, but any
neighbourhood of z meets the complement of H.
Lemma 3.5 Let P be a locally finite convex fundamental domain for the
discrete group y. A set of the form
Pflcr(P) 0- 0t,0' Er
is either empty or a closed geodesic segment. A set of the form
Pflo(P)flr(P) ; t#o#r#t, o,rEI
is empty or reduced to a point.
Proof From c(P) fl P = 0 it follows that aP fl o(P) = 0 and o i(P) fl aP = 0.

Write P = P U OP and o(P) = o(P) U o(OP) to get that
P fl c(P) = aP fl o(aP) ;oEr,o
We conclude that P fl c(P) is a convex subset of OP. Let us show that P fl o(P) is
contained in a geodesic line. If not, we can find three points of P n o(P) not on a
geodesic. An inner point z of the triangle based on these points furnishes an
interior point of aP , contradicting 3.4.
Let us investigate an interior point z of s = P fl o(P). Pick points x,y E s
such that z E ]x,y[ and consider a point u E P. The interior On of the triangle
with base [x,y] and vertex u is contained in P, 3.4. Similarly, the interior Dou of
the triangle with base [x,y] and vertex o(u) is contained in o(P). It follows that
we can find an open disc D with centre z such that D fl s is a diameter which
divides D into two open half-discs D' and D" entirely contained in P and o(P)
respectively. The half-discs D' and D" are disjoint from r(P). From D C D' U D"
we conclude that D fl r(P) = 0. We can now conclude the proof by recalling that
the set P fl o(P) fl r(P) is convex. 0
Definition 3.6 Let P be a locally finite convex fundamental domain for the
discrete group I'. By a side of P we understand a closed geodesic segment, not
reduced to a point, of the form P fl o(P) with c E r-{t}. By a vertex of P we
understand a point which can be presented in the form P n o(P) fl r(P),
o,r E r-{t}, o 54 r.
Proposition 3.7 The boundary 8P of a locally finite convex fundamental

domain is the union of its sides.
Proof Let z E OP and consider the following subset S of r

S = {rEr-{t}IzEo(P)}
Since P is locally finite we can find an open disc D with centre z such that
D C P U U or E S°(P)
Observe that S:0: the alternative gives D C P which according to 3.4 gives z E P,
contradicting z E 8P. As a consequence we have that
3.8 aP C UO'E r-{1}°(P)
Returning to the situation at hand, we cannot have P n a (P) = {z} for all o E S
since D-{z} is connected. Thus we have found a side of P through z E 8P. 0
A side s of P generates a geodesic h. The boundary points for s in h are

called the the end points for the side s. The number of end points for s is 0,1 or 2.
Proposition 3.9 Let s be a side of the locally finite convex fundamental

domain P. A point z E s is a vertex for P if and only if z is an end point for s.
Proof We have already shown during the proof of 3.5 that an interior point
of s is not a vertex of P. Let us investigate a point z of s = P n a(P), which is not
a vertex of P. From the previous proof we find an open disc D with centre z such
that D C P U o(P). This gives us a partition
D = DnP U Dno(P) U Dns
Since D - D n s is disconnected and D n s is a geodesic arc through the centre z of
D, we conclude that D n s is a diameter of D. 0
Corollary 3.10 A vertex A of P is contained in exactly two sides of P.
Proof Consider a open disc D with centre A which meets only those sides of
P which have vertex Az. Let these sides be enumerated cyclically $1,s2,.... sr. Note
also that D fl P is connected since it is a convex set. Let us change the
enumeration above such that D fl P C L(sl,s2). Observe that s fl D C P for
i = 1,...,r and conclude that r = 2. 0
A side s of P can be put in the form s = P fl o(P) where o E r-{t}.

According to 3.6 the transformation o is unique and will be called the side
transformation os generated by s. Let us put *s = P fl o *-'(P) and observe that
os(*s) = s. The assignment sN*s is an involution on the set of sides of P which
is called the side pairina of P.
By an edge of P we understand an oriented side. The operator * extends

in a natural way to an operator on the set of S of edges of P. We shall in
particular be interested in the subset S+ of edges whose initial vertex is finite. For
an edge s with initial vertex A we let is denote the second edge with initial vertex
A, compare 3.8. Let us combine these two operators to obtain a new operator
%Fs=J*s ; sES+
The operator %k is a bijection since it is composed of two involutions.
Lemma 3.11 The operator W: g+->g+ has finite orbits.
Proof Let us consider a sequence of iterates of ' on s E g

S17...'Sri... si=%P's,sE7L
Observe that the side transformation of = on maps the initial vertex Ai of si onto
the initial vertex A1_1 of si_1. It follows that the sequence A = A1,...,Ar,... of
initial points all belong to the finite set P fl TA. 0
Key lemma 3.12 Let s1 be an edge of P with initial vertex A and let
st,...,Sri... ol,...,Orr,...
be the sequence of *-iterates of st and corresponding side transformations. There
exists n > 2 with t and such that the first of the two unions
n n
U 0"1o2...o1(P) U 0-10-2...oi(P)
i=1 i=1
is neighbourhood of A while the second union is a disjoint union.
Proof Let us make use of the orientation of H2 which has P on the positive
side of is1. Let us define the sign of an edge s of P by sign(s) = +1 when P is on
the positive side of s and sign(s) = -1 otherwise. Observe that
sign(*s) = - det(os) sign(s) ;sEg
Using sign( is) sign(s) we get for s = s1_1 that
sign([si) = det(o1-1) sign(Tsi_1) ; i = 2,...
Let us investigate the angle Lor(jsi,sl) with vertex A1. We have that
Lor(lsi,si) = sign (Jsi)L1ncA1 mod 29r
Observe that o1...oi_1P has vertex A = o1o2...o1_1(A1) and angle
Lor(0*1 ...a1_1 f 1,01...oi_lsi) = det(o1...0i-1)Lor(jsi,si)
From the above formulas and the normalisation sign(js1) = 1 we get that
Lor(ot...oi_1 jsi,ot...oi_tsi) = Lint Al mod 27r
Let us also observe that
ot...oi_t Jsi = ol...oi_1*si_1 = Q1...oi_25i_1
It is now clear that the integer n for which 21r does the job.
Let sl,...,s,, be a full edge cycle, i.e. a full orbit for $ on C. The
corresponding sequence Al,...,Ar of initial points is called a vertex cycle. The
second vertex cycle which starts out with Al is A1,Ar.,Ar_1,...,A2, as follows by
observing that is1,1sl.,Jsi._1,...,Js2 form a full edge cycle.
Theorem 3.13 An interior point z of P is r-equivalent to no other point

of P. An interior point z of a side s of P is F-equivalent to the point O's -'(Z) of P,
but to no other point of P. Two vertices of P are equivalent under r if and only if
they belong to the same vertex cycle.
Proof Let us investigate an interior point z of a side s=P fl o(P). A point

w E P equivalent to z can be written in the form w=r(z), r E I'. This implies
z E r"1(P) fl P fl o(P) and we conclude that r = t or r = 0"1 = os 1 Let us now
turn to the case where A is a vertex of P and let us consider y E r such that
y(A) E P. This means that A belongs to the closure of 7 1(P) and we conclude
from 3.12 that we can find i = 1,...,n such that
'V 1(P)fo1...o1(P) 0
This gives -y- I = 0102...0j Evaluation of this formula at the vertex Al+1 gives
1(A;+1) = A or Ai = y(A).
y
Corollary 3.14 Let P be a locally finite convex fundamental domain for F.

The group IF is generated by the side transformations us, as s runs through the
sides of P.
Proof According to 2.3 it suffices to show that the subgroup of r generated by

the side transformations contains any y E r for which there exists a z E P with
y(z) E P. But this problem has already been dealt with during the proof of 3.13.
Corollary 3.15 Let sl,...,sr denote a full edge cycle for the operator 'I,
and let ol,...,or denote the corresponding side transformations. The cycle man
o = ol...or
is a rotation around the initial vertex Al of s1. The order of o is given by
2a = ord(o) (LintA1 + ... + LintAr)
where Al,...,Ar are the initial vertices of the edge cycle.
Proof Let us return to the proof of 3.12. Evaluation of the formula al.... on =
on the edge sn+1 gives us sl = sn+1 Taking into account that r is the length of
the cycle, we find that q = n/r is an integer. From the proof of 3.12 we get that
sign1(s,.+1) = del(o1...o,.) sign(sl)
Since sr+1 = s1, we get that del or = 1 which makes or a rotation. Observe that
27r = E i "1 LintAl = cl E i r l LintAl
and the result follows.
Proposition 3.16 Let P be a locally finite convex fundamental domain for

F and A E 0hP a cusp for r. The number of sides of P with end A is two.
Proof Let us first observe that for any point Q E P and any point A E 8hP the
geodesic segment [Q,A[ is contained in P as follows from 3.1.
Let us assume that A E 8hP is a cusp and let us pick a horocycle D with
center D. Recall from 2.6 that D meets at most finitely many translates of P and
conclude that D meets at most finitely many sides of P. This allows us to shrink
D such that all sides of P which meets D have ends at A. Let us observe that
D fl P # 0 and D T= P and conclude from the connectedness of D that D f OP # 0.
It follows from 3.7 that D f OP is traced on D by a number of geodesics hl,...,hr
through A. It follows from this that D fl P is fibred by traces of geodesics.
Observe also P fl yP = 0 for a generator y of r A. The remaining details are left of
the reader.
V.4 DIRICIILET DOMAINS 167
V.4 DIRICHLET DOMAINS
We shall present a method due to Dirichlet for construction of fundamen-

tal domains for a discrete group r of isometrics of H2. Let us recall from IV.1.9
that for a compact subset K of H2 we have that o(K) meets K for finitely many o
only. This allows us to fix a point w E H2 with r,, = {t}.
For a non-trivial o E IF let Lo(w) denote the perpendicular bisector for w,
o(w) and let Ho(w) denote the open half-plane of the complement of Lo(w)
containing w.
Lo(w)
4.1
The Dirichlet domain with centre w is the set P(w) given by
4.2 P(w) =
CEr,o#t
fl Ho(w)
Proposition 4.3 The Dirichlet domain P(w) is a fundamental domain for

IF in H2. Its closure P(w) is given by
P(w) = I I
Ho(w)
or E r,C
Proof Let us verify that the intersection P(w) fl D(w;r) is open for all r > 0.
Observe that the set S = {o E r-{t} I o(w) E D(w;2r)} is finite and that
P(w) fl D(w;r) = n a E S Ho(w)
We find that P(w) is open, since P(w) fl D(w;r) is open for all r>O. Next, we ask
the reader to verify the formula
7(P(w)) = P(7(w)) Er
From this it follows immediately that P(w) and y(P(w)) are disjoint for 7 54 t.
It is now time to verify the formula for P(w). Observe that the left hand
side is included in the right hand side since the intersection of a family of closed
sets is closed. To verify the opposite inclusion, consider a point z E H2 which
belongs to Ho(w) for all o E r-{t). Notice that [w,z] belongs to Ho(w) for all o
and conclude that [w,z[ belongs to P(w); it follows that z E P(w).
Let us show that any orbit A meets P(w). To this end pick a z E A with
the shortest possible distance from w. This gives
d(z,w) < d(o 1(z),w) ; o E IF, o 0 t
or d(z,w) < d(z,o(w)) for all o t, which shows that z E H. for all o E F -1t). It
follows that z E P(w), since P(w) is the intersection of such sets.
Proposition 4.4 Let r be a discrete group of isometries of H2 and let w be

a point of H2 such that rN, = {t}. The Dirichlet domain P with centre w is a
locally finite convex fundamental domain.
Proof Let us fix a number r > 0 and show that the open disc D(w;r) with
centre w and radius r > 0 meets o(P) for at most finitely many o E 17. If o(P)
meets P, we can find z E P with o(z) E D(w;r); this gives us
d(w,o(w)) <d(w,o(z))+d(o(z),o(w)) < r + d(z,w)
On the other hand, from the fact that z E P it follows that
d(z,w) < d(z,a 1(w)) = d(o(z),w) < r
But indeed, only finitely many o E I satisfy d(o(w),w) < 2r.
The modular figure from 1.1 is in fact a Dirichlet domain for the modular
group as follows from the following lemma.
V.4 DIRICIILE'F DOMAINS 169
Lemma 4.5 With the notation of 4.2 let S be a subset of r-{r} such that
Ps(w) - fl Hb(w)
b E S
meets each F-orbit in at most one point. Then PS(w) = P(w).
Proof Let us consider a r E IF and investigate two alternatives. If Lr(w) does

not meet PS(w) then PS(w) C Hr(w) as follows from the fact that PS(w) is
connected and contains w. If Lr(w) fl PS(w) $ 0 then we can find a point z which
is an interior point of PS(w) - P(w). Since z E o(P(w)) for some o E r, o # c, we
can find a point of o(P(w)) in PS(w) - P(w) contradicting the assumption that no
orbit meets PS(w) twice. 0
The reader will find much more information about Dirichlet domains in
[Beardon]. In particular, the change of base point for the Dirichlet domain is
discussed and many applications are given. See also [Macbeath, Hoare].
V EXERCISES
EXERCISE 1.1 The group PG12(Z) contains the congruence group G(2) as a
normal subgroup.
10 Show that PG12(Z) has a unique subgroup N of index 2 containing G(2).
2° Show that the hyperbolic quadrangle i, 1 + i, exp(7ri/3), 00 is a
fundamental domain for N. Show that the sides of the quadrangle can be paired
by a rotation about exp(3ai) and a glide reflection fixing oo, both belonging to N.
EXERCISE 1.2 Show that PS12(Z) contains three subgroups of index 3 which
contain I'(2). Find fundamental domains for these groups.
VI COVERINGS
In this chapter we shall introduce a number of geometric concepts which

come up in the following problem, central for the proof of Poincare's theorem in
chapter VII: Let there be given a connected Hausdorff space X and a local homeo-
morphism f:X-+H2; give conditions for this to be a global homeomorphism.
The first thing to do is to introduce a metric on X. This makes X into a
hyperbolic surface with its own intrinsic geometry. The main result is the
monodromy theorem, which asserts that f:X-H2 is a global homeomorphism if
and only if X is a complete hyperbolic surface.
The geometric concepts we introduce create their own universe with its
own life. We are led to consider the problem of space forms : to classify all
geometries, which locally look like the hyperbolic plane. This will be done in
terms of conjugacy classes of discrete subgroups of PGI2(R).
VI.1 HYPERBOLIC SURFACES
For the convenience of the reader we shall recall the definition of the arc
length of a curve y:[a,b]--X in a metric space X. To this end we consider a
subdivision a = a0 < a1... < an = b of our interval and form the sum
n
E d(y(ai-1),-y(ai)) ; a=a0<al... <an=b
i=1
We say that the curve y:[a,b]-%X is rectifiable if y is continuous and sums of the
above form are bounded. The length of a rectifiable curve y:[a,b]-:X is defined to
be the supremum over sums of this form taken over all subdivisions of [a,b]
1.1 1(y) = sup d(-y(ai-1),y(ai))

i=1
It follows from the triangle inequality that d(x,y) < l(y). We shall mostly be
concerned with metric spaces which have the following property.
172 COVERINGS
Definition 1.2 We say that a metric space X has the shortest length
property if, first of all, any two points x and y of X can be joined by a rectifiable
curve y. Moreover, it is required that
d(x,y) = inf7 1(y)
where the infimum is taken over all rectifiable curves in X joining x to y.
The hyperbolic plane Hz has the shortest length property: the distance
d(x,y) between x,y E H2 is realized as the arc length of the geodesic arc [x,y]. We
shall be concerned with metric spaces X with the shortest length property which
locally look like H2.
Lemma 1.3 Let X be a metric space with the shortest length property and
D an open disc in H2 with centre a and radius r. For any isometry o:D-+X of D
onto an open neighbourhood of o(a) in X we have d(x,o(a)) > r for all points
x E X - o(D).
Proof Let there be given a rectifiable curve y in X which joins o(a) to the
point x outside o(D). For a number p E ]O,r[ let K denote the compact disc in H2
with radius p and centre a and let us prove that y meets o(OK). Observe that
o(K) is a compact subset of X and conclude that V = X - o(K) is an open subset
since X is Ilausdorff. This gives us a partition of X into the closed subset o(8K)
and the two open subsets V and U = o(K - OK). Observe that y meets V and U
and conclude from the connectedness of the image of y that y meets o(8K) in y
say. This gives us 1(y) > d(o(a),y) = p. Take supremuwn over p E ]O,r[ to get
that 1(y) > r. Finally, take infimum over all y and conclude that d(o(a),x) > r.
Definition 1.4 By a hyperbolic surface we understand a metric space Y

with the shortest length property, for which every point has an open neighbour-
hood isometric to an open disc in the hyperbolic plane H2.
VI.1 HYPERBOLIC SURFACES 173
For a point y of a hyperbolic surface Y and a real number r > 0, D(y;r)

denotes the metric disc of points v E Y with d(y,v) < r. For r > 0 sufficiently
small, D(y;r) is isometric to a hyperbolic disc of radius r as follows from 1.3.
Local homeomorphisms. Let X and Y be topological spaces. A con-

tinuous map f:X-+Y is called a local homeomorphism if each point x E X has an
open neighbourhood which is mapped by f homeomorphically onto an open
neighbourhood of f(x). When X and Y are metric spaces, a map f:X-+Y is called a
local isometry if any point x E X has an open neighbourhood which is mapped
isometrically onto an open neighbourhood of f(x).
We shall consider a connected Hausdorff space X and a hyperbolic surface
Y and demonstrate that a local homeomorphism f:X- Y induces a natural metric
on X, which makes X into a hyperbolic surface and f:X-+Y into a local isometry.
Let us first show that any two points u,v E X can be joined by a
continuous curve ry such that fy is rectifiable in Y. To see this, fix the point u and
separate the space X into the set U of points v E X which can be joined to u by
such a curve and the set V of points v E X which can't. The assumption that X is
connected ensures that V = 0. Let us put
1.5 d(u,v) _ Mf7 1(f7) ; u,v E X
where the infimum is taken over all continuous curves -y in X joining x to y such
that fy is rectifiable in Y. We shall establish, during the proof of 1.6, that this
defines a metric on X.
Proposition 1.6 Let X be a connected Hausdorff space and Y a hyperbolic

surface and f:X-+Y a local homeomorphism. The distance function given in 1.5
defines a metric on X, which turns X into a hyperbolic surface and f:X-+Y into a
local isometry.
174 COVERINGS
Proof The distance d(u,v) between u,v E X, as defined in 1.5, is obviously

symmetrical in u and v and satisfies the triangle inequality. We shall show below
that distinct points have non-zero distance. Let us first record that
d(u,v) > d(f(u),f(v)) ; u,v E X
Let us fix a point x E X and let W be an open neighbourhood of x which is
mapped by f homeomorphically onto a hyperbolic disc D(o(x);r) in Y (r > 0). For
points u and v in W let ) be the curve in W such that fA is the geodesic arc from
f(u) to f(v). This gives d(u,v) < 1(fa) = d(f(u),f(v)) and we conclude that
d(u,v) = d(f(u),f(v)) ; u,v E W
Let us show that d(x,z) > 0 for any point z # x. If z E W, this follows from the
formula above. For z 0 W, let y be any continuous curve in X with fy rectifiable
which joins x and z. Pick P E ]0,r[ and let us show that fy meets the circle S in Y
with centre y = o(x) and radius p. To see this, let us prove that y meets the
inverse image S' of S by f:W-*Y (we shall give an outline only since a similar
argument has been used in the proof of 1.3). Observe that the complement of S'
in X is partitioned into two open sets: the inverse image D' of D(y;p) by f:W->Y
and the complement in X of the inverse image K' of D(y,p) by f:W-;Y. Finally
use that the image of 7 is connected to conclude that y meets S'. It follows that
1(fy) > p. Let us make a variation of y and conclude that d(x,z) > p.
We claim that a continuous curve y:[a,b]-,X is rectifiable in X if and only
if fy is rectifiable in Y. To see this, use the Borel-Heine theorem to introduce a
subdivision a = ao < al... < an = b of [a,b] such that the restriction y:[a1-l,ai]--,X
takes values in a hyperbolic disc W as considered above. It follows that
1(7) = E; 1(yi) = Ei 1(fy;) = 1(fy)
This reveals that X has the shortest length property. D
VI.2 IIOPF-RINOW THEOREM 175
VI.2 HOPF-RINOW THEOREM
In this section we shall be concerned with the geodesics on a hyperbolic

surface X. The basic problem is to join two points x and y of X by a geodesic
curve of arc length equal to the distance between x and y. Substantial results will
only be obtained when X is complete in the sense that any Cauchy sequence on X
is convergent.
To begin let us start with a geodesic curve y:J-*X defined on the interval
J, compare 11.1, and let us show that the following inequality is satisfied
2.1 d(7(s),7(t)) <- It-sl ; s,tEJ
This follows from a straightforward subdivision argument based on the theorem of

Borel-Heine and the triangle inequality.
Proposition 2.2. Let X be a hyperbolic surface. If two geodesic curves

a,Q: R--*X agree on a non-empty open interval J C IR then they agree on all of R.
Proof Let us assume that 0 E J and let us introduce the following set
E = Is E ]0,+00[ l a and 0 agree on [0,s]}
From the very definition of E it follows that E is a non-empty interval with 0 as
its left end point. Let us show that E is an open subset of R. To this end we
consider a point s E E and a small open disc D in X with centre a(s) =,3(s). We
conclude from our study (made during the proof of 11.4.8) of geodesics in H2 that
a and ,6 agree in a neighbourhood of s E It
Let us show that E is unbounded. To this end we must rule out the
alternative E=]O,p[ for some p E It by the definition of E we find that a and Q
agree on [O,p[. By continuity of or and 3, this implies that a and 3 agree on [O,p],
and we get p E E, a contradiction. This argument shows that a and ,Q agree on
[0,+oo[. The interval ]-00,0] is handled in a similar manner. 0
176 COVERINGS
Proposition 2.3 Let X be a complete hyperbolic surface. A geodesic curve

y:J--+X defined on an open interval J can be extended to a geodesic R -> X.
Proof Let us first remark that the result is true for X = H2 as follows from the
proof of 11.4.8. In the general case let us assume that 0 E J and try to extend y to
the right of J. To this end put
j = { x E ]0,+co[ 1 y extends to ]0,x[ }
The set is a subinterval of ]0,+oo[. Let us prove that a E I implies a+e E j for
some a>O. From the inequality 2.1 we conclude that y:[O,a[-+ X is uniformly conti-
nuous. From the completeness of X it follows1 that y can be extended to a conti-
nuous curve on [O,a]. In order to extend y to the right of the point a consider a
hyperbolic disc in X with centre y(a) and use the result for H2. It follows that
I = ]0,+oo[. Extension of y to the left of J can be done in a similar way.
Let us now consider two fixed points x and y of a hyperbolic surface X.

The distance d = d(x,y) can almost (1.2) be realized as a the length of a curve
connecting x to y, but not quite in general. Consider for example the punctured
Poincare disc D-{0} and the points x = -1/2 and y = 1/2.
On a complete surface, however, we can realize the distance as the length
of a geodesic connecting x to y:
Hopf- Rinow theorem 2.4 Let X be a complete hyperbolic surface.

For any two points x and y on X of distance d = d(x,y), there exists a geodesic
curve o:I2-+X with a(0) = x and o(d) = y.
1 We are referring to the following general fact: let X be a complete

metric space, any uniformly continuous curve y:[0,a[-+X can be extended to a
continuous curve on [0,a]. A proof can be based on the observation that a uni-
formly continuous map y preserves Cauchy sequences.
VI.2 IIOPF-RINOW THEOREM 177
Proof Let us introduce an abbreviation which will facilitate the exposition and
underline the strategy of the proof
x and z satisfy the Hopf-Rinow theorem

%%(x'z'y)
: and d(x,z) + d(z,y) = d(x,y) J
Let us first show that for any two distinct points x and y of X, we can find z # x
such that %%(x,z,y). To this end choose an r E ]O,d[ such that the disc D(x;p) is
hyperbolic for some p > r. The circle S(x;r) with centre x and radius r is compact
as follows from 1.3. Let the continuous function
z ,--, d(z,y) ; z E S(x;r)
take its minimum at the point z E S(x;r). Consider a rectifiable curve y:[a,c]--, X
with y(a) = x and y(c) = y. Choose a point b E ]a,c[ with d(x,y(b)) = r. We have
1(y) _ k(-Y(a,h)) + 1(y[b,e]) >_ d(x,7(b)) +d(7(b),Y) > r+d(z,Y)
Take infimum over all rectifiable curves y from x to y to get
d(x,y) > d(x,z) + d(z,y)
The opposite inequality is the triangle inequality and we have verified ]b,(x,z,y).
Let us establish the basic relation
2.5 J R (x,Z,Y) & Jt%(Z,w,Y) N%(x,w,Y) ; x,Y,z,w E X
Simple manipulations of the two identities and the triangle inequality yield
d(x,w) + d(w,y) < d(x,z) + d(z,w) + d(w,y) = d(x,z) + d(z,y) = d(x,y)
Another application of the triangle inequality gives us
d(x,w) + d(w,y) = d(x,y)
Put this information back into the first inequality and deduce that in fact
d(x,z) + d(z,w) = d(x,w)
It follows from this and lemma 2.6 that x and w satisfy the Hopf-Rinow theorem.
To conclude the proof, let x and y be points with d = d(x,y). Pick a point
v $ x satisfying N (x,v,y), put e = d(x,v) and use 2.3 to construct a geodesic
curve o:R- X with o(0) = x and o(e) = v. We shall study the set
J = { s E [O,d] I d(x,o(s)) = s & d(o(s),y) = d - s }
It is easily seen that the set J is a closed subinterval of [O,d] containing 0. Let us
178 COVERINGS
show that s E ]O,d[ fl J implies t E J for some t > s. Observe that z = a(s) satisfies
3G%(x,z,y). Pick a point w $ z such that 3G%(z,w,y) and conclude from 2.8 that
36,(z,w,y). It follows from 2.6 and 2.3 that w belongs to the image of o. The
parameter t E R with o(t) = w satisfies our requirements. It follows from this that
J=[O,d]. In particular d(o(d),y) = 0, i.e. o(d) = y. 0
Lemma 2.6 Let -r :[a,c]-+X be a curve in the metric space X and let there be
given b E [a,c] such that
d(7(a),7(b)) + d(7(b),7(c)) = d(7(a),7(c))
If the restrictions y:[a,b]-+X and y:[b,c]-X are distance preserving, then the curve
y:[a,c]--->X is distance preserving.
Proof Let us first prove the following simple inequality

d(y(s),y(t)) < t - s ; s,t E [a,c], s < t
It suffices to treat the case s E [a,b] and t E [b,c]; the triangle inequality yields
d(y(s),y(t)) < d(7(s),y(b)) + d(y(b),7(t)) = b - s + t - b = t - s
which completes the proof of the inequality. We need to prove the opposite
inequality as well. To this end let us observe that
c - a = (c - b) + (b - a) = d(y(a),y(c)) < d(y(a),y(s)) + d(y(s),y(t)) + d(y(t),y(c))
Let us combine this with the first inequality to get that
c-a<s-a+d(y(s),y(t))+c-t
which may be rewritten t - s < d(y(s),y(t)). 0
VI.3 UNIFORMIZATION
In this section we shall show that a complete hyperbolic surface X admits

a local isometry of the form f.H2-+X , where H2 is the hyperbolic plane. But first
a result to the effect that plane figures cannot be isometric without being
congruent. Such figures can't be deformed: they are rigid.
Rigidity 3.1 Let D be a subset of H2 not contained in a hyperbolic line.

Any isometry o:D-JH2 has a unique extension to an isometry of H2.
Proof Let us pick three points A,B,C of D not on a geodesic. We can use
lemma 11.4.8 to pick an isometry r:H2_H2 which agrees with o at the three points
A,B,C. It follows from the proof of 11.4.9 that the restriction of r to D is actually
equal to a. 0
Uniformization theorem 3.2 Let D be an open disc in the

hyperbolic plane H2. Any isometry ¢:D-'X of D into a complete hyperbolic
surface X has a unique extension to a local isometry H2- X.
Proof Let 0 denote the centre of D and let us use 2.3 to extend the
isometry D-+X to a map 0:H2-.X whose restriction to each hyperbolic line
through 0 is a geodesic. Let us fix a point A of H2 and proceed to find an open
disc with centre A which is mapped isometrically into X by 0. To this end use the
Borel-Heine theorem to find an r > 0 such that for all C E [0,A] the metric disc
D(O(C);2r) is a hyperbolic disc. Let us first prove that
d(,A(A),O(B)) = d(A,B) ; B E D(A;r)
To this end consider a subdivision of the segment [0,A]
0 = A0, A1,.... Ai_1, A = A d(Ai_1)A;) < r , i = 1,...,n
and let B0,B1,...,Bn be the perpendicular projections of the sequence of As onto
180 COVERINGS
the line through 0 and B. Let us prove2 that for i = 1,...,n

LAi-1Bi-1A1 ,,, z (Ai-1)4(Bi-1)0(Ai)
OB1-1A1Bi Odi(Bi-1)0(Ai)0(Bi)
where means that the triangles are congruent in the sense that distances between
pairs of corresponding vertices are equal.
This is done by induction on i. To accomplish the induction step, let us
focus on the hyperbolic discs Di = D(Ai,2r) and Si = (O(Ai);2r). Observe at first
that the restriction 0:[Ai,Ai+1]-'X has image in `11i and conclude that
d(Ai,Ai+1) = d(0(Ai),0(Ai+1))
From the induction hypothesis we get that
d(0(Ai),0(Bi)) = d(Ai,Bi) < d(An,Bn) < r
Let us combine this with the inequality
d(0(Bi),q(T)) < d(Bi,T) < r ; T E [B1,Bi+1]
to conclude that the restriction m:[Bi,Bi+1],X has image in `Di. This gives
d(0(Bi),0(Bi+1)) = d(B1.Bi+1)
Let us now focus on the following configuration in Di
and the image configuration in `.bi. From the induction hypothesis we conclude that
LAi-1A;B1-1 = L0(Ai-1)0(Ai)0(Bi-1)
LBi-iA;Bi = L0(Bi-1)c5(Ai)c5(B;)
with the conclusion that
2The details of the proof are elementary and can be administered in a

number of ways. The reader is advised to work out his own version.
LA;_1A;B; L0(Ai-1)0(Ai)0(Bi)
From this we draw the same conclusion for the complementary angle
LAi+1A;Bi =L0(Ai+1)0(Ai)0(Bi)
We can now use elementary geometry to conclude that
AAi+IAiBi = A0(Ai+1)0(Ai)0(Bi)
In a similar way we can first deduce that
LBi-1BiAi = Ld'(Bi-1)0(Bi)0(Ai)
LA,B;AJ+1 =
with the conclusion that
LAi+1BiBi+1 =
Another reference to elementary geometry gives us
AAi+1B;Bi+1 A0(Ai+1)0(Bi)0(Bi+1)
which concludes the induction. Let us now prove that
AAnB11B OO(An)O(Bn)O(B)
Observe that LBn=LO(Bn) is a right angle, and conclude as above that the
corresponding sides have equal length. In conclusion d(A,B) = d(O(A),0(B)). Let
us remark that at the same time we have proved that the following two angles are
equal LAn_1AB = L0(Ai-1)0(A)0(B).
It remains to prove that the distance between any two points B and C of
D(A;r) is preserved. To see this, observe that the proof above shows that the
angle LOAB is preserved by 0. A similar remark applies to the point C and the
result follows by hyperbolic trigonometry, 111.5.1.
The proof of the uniformization theorem is adapted from the Euclidean

case given in [Nikulin, Shafarevich] II, §10. The reader is advised to consult this
source for further comments.
182 COVERINGS
VI.4 MONODROMY
In this section we shall be concerned with general properties of a local

isometry f.-X--+Y between hyperbolic surfaces. It follows from rigidity 3.1 that a
local isometry is an open map in the sense that any open subset U of X is mapped
by f onto an open subset f(U) of Y. Let us establish an important inequality
4.1 d(f(x),f(y)) < d(x,y) ; x,y E X
Proof Consider a point x E X and an r > 0 such that the metric disc D(x;r) is a
hyperbolic disc and such that the restriction f:D(x;r)-Y is an isometry. From the
fact that f is open it follows that D(x;r) is mapped homeomorphically onto an
open neighbourhood of f(x) in Y. We conclude from lemma 1.3 that
f(D(x;r)) = D(f(x);r)
Let there be given points x,y E X and a continuous curve y:[a,b]-X with y(a) = x
and y(b) = y. We ask the reader to use the Borel-Heine theorem to find an r > 0
such that for all s E [a,b] the point y(s) is the centre of a hyperbolic disc with
radius r such that f maps D(y(s);r) isometrically onto D(fy(s);r). A simple
subdivision argument shows that y is rectifiable in X if and only if fy is rectifiable
in Y. For a rectifiable curve -y in X we find that
1(fy) = 1(y)
The formula 4.1 follows from variation of y. 0
Geodesic lifting property 4.2 Let f: X--+Y be a local isometry

between complete hyperbolic surfaces and consider points x E X and y E Y with
f(x) = y. For any geodesic curve 71:R-:Y with ij(0) = y, there exists a unique
geodesic curve £: R-X with f{ = q and (0) = x.
Proof Choose r>0 such that f induces an isometry D(x;r)- D(y;r). Next,
choose an open interval J around 0 in R such that 71(J) C D(y;r). Let f:J-D(x;r)
VI.4 MONODROMY 183
be given by ff =,9. Let us use 2.3 to extend to all of O and use 2.2 to conclude
that the identity ff _ 17 is generally valid. 0
Proposition 4.3 A local isometry f:X--*Y between complete hyperbolic sur-

faces is surjective. Moreover, let there be given a point y E Y and a p > 0 such
that D(y;p) is hyperbolic, then for any x E X with f(x) = y the disc D(x;p) is
mapped isometrically onto D(y;p) by f.
Proof We have already observed that f(X) is open. Let us show that f(X) is
closed. To this end consider a point w in the closure of f(X). Pick a hyperbolic
disc D with centre w and a point y E D fl f(X). Let us join the centre w of D with
y along the radius of D and extend this to a geodesic ij:l ->Y. According to 4.2 we
can find a geodesic x:R,X with p = From this it follows that all points of y
belong to f(X). Since f(X) is open and closed we can use the fact that Y is
connected to conclude that f(X) = Y.
Let us now investigate the point x. For small values of r > 0, the map f
induces an isometry D(x;r)-+D(y;r). The inverse of this induces an isometry
s:D(y;r)-+X with fs = t and s(y) = x. Let us use the uniformization theorem 3.2 to
extend s to a local isometry s:D(y;p)--X. The extension will still satisfy s(y) = x
and fs = c as follows from 2.2 applied to the various radii in the disc D(y;p). In
fact s is an isometry: for u,v E D(y;p) we get from the inequality 4.1 applied twice
d(u,v) = d(f(s(u)),f(s(v)) < d(s(u),s(v)) < d(u,v)
which shows that d(u,v) = d(s(u),s(v)). We conclude from the proof of 4.1 that
s(D(y;p)) = D(x;p). Let us recall the relation fs = t and it follows that D(x;p) is
mapped by f isometrically onto D(y,p).
We shall see that a local isometry f:X->Y between complete hyperbolic

surfaces X and Y is a covering projection in the sense of homotopy theory. For
more topological results and references see VI.8.
184 COVERINGS
Corollary 4.4 Let f:X- Y denote a local isometry between complete

hyperbolic surfaces and let y E Y and p > 0 be given such that the metric disc
D(y;p) is hyperbolic. For distinct points x and z of the fibre f-'({y})
D(x;p) fl D(z;p) = 0 ; x,z E f-1{y} , x # z
Moreover, any point w E X with f(w) E D(y;p) is contained in the disc D(x;p) for
some x E f-1{y}.
Proof Let us put D = D(y;p) and make a general remark on two continuous
sections r,s of f:X-}Y over D C Y. Let us show that
r(D)fls(D)54 0 r=s
We may write a common point u of r(D) and s(D) in the form u=r(vl)=s(v1),
where v1,v2 E D. This, however, gives f(u) = v1 = v2. Thus we have found a
v E D with r(v) = s(v). Observe that the connected set D is partitioned into the
two sets
{wEDIr(w)0s(w)}, {vEDIr(v)=s(v))
The first set is open since r and s are continuous and X is Hausdorff. To see that
the second set is open consider a point v E D with r(v) = s(v); choose a neigh-
bourhood U of r(v) = s(v) such that the restriction of f:X-'Y to U is injective and
choose a neighbourhood V of v in D with r(V) C U and s(V) C U. We have that
w= f(r(w)) = f(s(w)) ;wEV
which gives r(w) = s(w) for all w E V. In conclusion we have partitioned the
connected set D into two open subsets of which the second is non-empty. It
follows that r = s.
To conclude the proof of the first statement, observe that at a point of the
fibre x E f-1{y} we can use the restriction of f to the disc D(x;p) to get an
isometry ,: D(x;p) D(y;p). The inverse s,, = f,-' defines a section s, of
f:X--*Y over D with s,(D) = D(x;p).
Consider a point w E X with d(f(w),y) = r < p. Let q:R-4Y denote a
geodesic with q(O) = f(w) and il(r) = y. Let us use 4.2 to find a geodesic e:R-+X
with f. = p and (O) = w. This gives f(i(r)) = zl(r) = y. In conclusion we have
found a geodesic {: R-X with (0) = w and fi(r) = x E f-1 {y} where r E ]0,p[. It
follows that w E D(x;p) as required.
VI.4 MONODROMY 185
Proposition 4.5 Let f:X-Y be a local isometry of complete hyperbolic

surfaces. The metric on Y can be recovered from that of X by the formula
d(y,w) = inf { d(x,v) I x,v E X, f(x) = y, f(v) = w }
Proof Use the Hopf-Rhiniw theorem 2.4 to pick a geodesic curve il:[O,d]->Y
with 9(0) = y and q(d) = w, where d = d(y,w). Let :[O,d]-*X be a lifting of rl to
X, compare 4.2. This gives points x = (O) and v = (d) with f(x) = y and
f(v) = w. It follows from the inequality 2.1 that d(x,v) < d. The required identity
is now a consequence of the general inequality 4.1.
A refinement of the argument above gives us
4.6 f(D(x;r)) = D(f(x);r) ;xEX, r>0
Proposition 4.7 Let Z be a complete hyperbolic surface. Any closed and

bounded subset of Z is compact.
Proof By 3.2 we can find a local isometry f

from that all closed metric discs in Z are compact.
Monodromy theorem 4.8 Let X be a complete hyperbolic surface.

Any local isometry is a bijection.
Proof Let us first show that any local isometry o:H2-+H2 is an isometry. To
this end choose an open disc D C H2 such that the restriction o': D-+H2 is an
isometry. Next use 3.1 to extend o' to a global isometry o": H2-*H2. Observe
that or = o" as follows by restriction to hyperbolic lines through the centre of D.
In the general case, let us use the uniformization theorem 3.2 to construct
a local isometry ir:H2-+X. The composite Oir:H2-H2 is a local isometry and
consequently a bijection. We conclude from this that it is injective and from 4.3
that rr is surjective. This shows that 0 is a bijection as well.
186 COVERINGS
VI.5 ORBIT SPACES
Let us consider a hyperbolic surface X and a group G of isometries of X

which acts discontinuously on X. By this we mean that any point x E X has an
open neighbourhood U such that
5.1 o(x) 0 U for all but finitely many o E G
For the G-orbit A through x we conclude the existence of r > 0 such that
D(x;r) fl A = {x}
For two distinct points p,q of A we can choose o E G with o(p) = x and conclude
that d(o(p),o(q)) > r. Since o is an isometry we can conclude that
5.2 d(p,q) ? r ;p,gEA,p q
Proposition 5.3 A discontinuous group G of isometries of the hyperbolic

surface X acts with closed orbits on X.
Proof Let A be a G-orbit on X and z a point of X outside A. With the

notation of 5.2 the disc D(z; r/2) intersects A in at most one point. From this it is
easy to find a disc with centre z not meeting A. 0
We intend to put a metric on the space X/G of orbits. To get started let
us introduce the distance between two G-orbits A and `.9 by the formula
5.4 d(A,%) = inf{d(x,y) I xEA,yE`.B}
Let us fix points x E A and Y E S. We can rewrite the definition above as

d(A,`$) = inf { d(o(x),r(y)) I o,r E G }
Since we can write d(o(x),r(y)) = d(x,o-ir(y)), we conclude that
VI.5 ORBIT SPACES 187
5.5 d(A,`.B) = inf { d(x,y) I y E ` } ;xEA
Proposition 5.6 The distance function defines a structure of metric space

on the orbit space X/G. This metric has the shortest length property.
Proof The distance between two orbits is symmetrical by 5.4. In order to

prove the triangle inequality for orbits .A,`B,C let us fix a point y E S. This gives
d(A,C) < d(x,z) < d(x,y) + d(y,z) ; x E .A, z E C
Fix x E .A and take infimum over all z E C and conclude from 5.5 that
d(A,C) < d(x,y) + d(%,C) ; XEA
Take infimum over all x E A and apply 5.5 to get
d(A,C) < d(A,B) + d(B,C)
If A 54 `9 we conclude from 5.5 and 5.3 that d(A,`.B) > 0. Let f:X-X/G denote
the projection. Observe that f decreases distances to conclude that a rectifiable
curve -y in X joining x to v maps to a rectifiable curve f7 in X/G with
1(y) > 1(fy) > d(f(x),f(v))
For points y,w E Y choose x,v E X with f(x) = y, f(v) = w and d(x,v) = d(y,w).
For c > 0 given, choose a rectifiable curve 'y in X joining x to v which satisfies
1(y) < d(x,z) + e. From the inequality above we get that
1(f7) < d(y,w) + e
which shows that X/G has the shortest length property. 0
It follows from 5.1 that the canonical projection f:X-+X/G satisfies
5.7 f(D(x;r)) = D(f(x);r) ; x E X , r>0
Let us make a closer analysis of this in terms of the stabiliser of x given by

Gx={uEGIv(x)=x}
Let us notice that Gx acts on any disc D(x;r) with centre x and radius r > 0. Thus
we can form the orbit space D(x;r)/G.
188 COVERINGS
Proposition 5.8 Let G be a group of isometries of X acting discon-

tinuously on X. For any point x of X there exists r > 0 such that f induces an
isometry of D(x;r)/Gx onto D(y;r).
Proof Let us use 5.1 to choose r > 0 such that

d(x,o(x)) > 4r ; o E G - Gx
Using the triangle inequality we get for y E D(x;r) and a E G-Gx that
4r < d(x,o(x)) < d(x,o(y)) + d(o(x),o(y)) < d(x,a(y)) + r
which allows us to conclude that d(x,o(y)) > 3r for y E D(x;r) and a E G - G.
From this it follows that for z,y E D(x;r) we have that
inf{ d(z,o(y)) I o E G}= inf{ d(z,o(y)) I a E GX }
which is the required result. 0
Proposition 5.9 Let G be a discontinuous group of isometries of the hyper-

bolic surface X. The orbit space X/G is complete if and only if X is complete.
Proof When X is complete, we get from 5.7 that any closed metric disc in X is
compact. It follows from 5.4 that any closed metric disc in X/G is compact. To
prove that a Cauchy sequence {z11) in X/G is convergent we can refer to the
elementary fact that a Cauchy sequence is bounded, i.e. contained in some closed
metric disc in X/G. Let us finally recall the fact that a Cauchy sequence in a
compact metric space is convergent.
Suppose conversely that X/G is complete and let us investigate a Cauchy
sequence {x17} in X. The sequence {p(x11)} is a Cauchy sequence in X/G and
converges towards a point y E X/G, say. Let 9J be the corresponding G-orbit in X
and choose r>0 such that the discs D(x;3r)x E c are disjoint. Let us choose an
n E N such that d(p(xn),y) < r and such that d(xn,xp) < r for all p > n. Let the
point x E qJ be given by the condition d(x1l,x) < r. It follows that
xp E D(x;2r) ; p>n
Upon shrinking r we can obtain that the closure of D(x;2r) is compact, compare
1.3. It follows that the sequence {x11} is convergent in X. 0
VI-6 CLASSIFICATION 189
VI.6 CLASSIFICATION
In this section we bring our theory of complete hyperbolic surfaces to

culmination by classifying all such surfaces in terms of torsion free3 discrete
subgroups of PG12(R). Let us observe that a torsion free discrete subgroup I' of
PG12(R) acts discontinuously on H2 with trivial stabilisers. It follows from 5.8
that the orbit space H2/I is a complete hyperbolic surface.
It is useful to observe that the canonical projection a:H2-+H2/I' has the
following mapping property: a map f:H2/F-+Y into a hyperbolic surface Y is a
local isometry if and only if fir:H2-+Y is a local isometry.
Theorem 6.1 Any complete hyperbolic surface X is isometric to a surface

of the form H2/F where r is a torsion free discrete subgroup of PGl2(03). Two
such subgroups r and E define isometric surfaces H2/F and H2/E if and only if r
and E are conjugated subgroups of PG1.2(R).
Proof Given a hyperbolic surface X. Use the uniformization theorem 3.2 to

pick a local isometry a:H2-+X. Let us introduce the subgroup r of PG12(IR) by
r = { o E PGL2(R) I ro = it }
Let us investigate the action of r in a neighbourhood of a point z E H2. To this
end pick an open disc D with centre z such that the restriction of a:D->X is
injective. Let us show that
o(z)0D for all oEF - {t}
Assume for a moment that o E f satisfies o(z) E D. From a(o(z)) = a(z) we
conclude that o(z) = z, which implies o(D) = D. This gives ir(o(x)) = a(x) for all
x E D and therefore o(x) = x for all x E D; thus 3.1 gives us or = t. This shows
that r acts discontinuously on H2 without fixed points. We can now conclude that
F is a discrete subgroup of PGl2(Od) which is torsion free: elements of finite order
of PG12(R) are rotations and reflections, but both types have fixed points.
3 A group r is torsion free if all non-trivial elements have infinite order.

190 COVERINGS
Consider points z,w E 112 with x = a(z) = a(w). Let us show that we can
find a transformation y E r with y(z) = w. The restrictions it and
?r,,:D(w;r)->D(x;r) are isornetries for small values of r > 0. Introduce the isometry
0 = aw 11r : D(z;r) -' D(w;r)
and use rigidity 3.1 to extend 0 to a global isometry o of H2 into itself. The
identity ao = n holds in D(z;r), but it extends to all of H2 by rigidity 3.1.
We can now use 7r:H2->X to produce a bijection 0: H2/r X. But 4.5
allows us to conclude that 0 is an isometry.
Let us now study local isometries from H2/r to H2/E. Let us first show
that a o E PG12(U8) with arc r-1 C E induces a map
6.2 H2/r H2/E ; ll H or (h) ; orf 1 C E
In other words, we must verify that " x - y mod r o(x) u(y) mod E ".
To this end we write y = 7(x) with y E r to get o(y) = oy(x)) = (uyo1)o(x) and
observe that oyo1 E E to get o(x) = o(y) mod E as required. It follows from
the remarks from beginning of this section that the map 6.2 is a local isometry.
Let us finally verify that the transformation 6.2 is an isometry if and only
if oro' = E. To see this make the factorization of 6.2
H2/r H2/oro 1 , H2/E
where the first map is the isomorphism induced by o and the second map is simple
projection. This second map is bijective if and only if oro 1 = E. 0
Corollary 6.3 Let r c PG12(R) denote a discrete torsion free subgroup.

The automorphism group of H2/r is isomorphic to Nr/r where the normaliser
AT of r is given by
AT ={oEPGl2(R)Ioro1=r}
VU INSTANT JET SERVICE
Let us consider a group I' of isometrics of a hyperbolic surface X and a

closed subset Z of X with the property that the family (QZ)a E I. forms a locally
finite covering of X. This assumption implies that r acts discontinuously on X.
The action of r induces an equivalence relation Q on Z and the projection
f:X- X/I' induces a homeomorphism Z/gj X/1', as follows from the proof of
V.2.1. We are interested in transporting the metric from X/1' to Z/9.
Let us illustrate this by the analogous example of a cylinder realized as the
orbit space of the Euclidean plane X under the group r generated by a single
parallel translation r; the set Z is a plane strip perpendicular to the translation
direction and of width equal to the length of the translation. The corresponding
examples in hyperbolic geometry are treated in 7.7. These examples are basic for
the proof given in VII.3 of the Poincare theorem in the case of unbounded
polygons.
Definition 7.1 Let fj be an equivalence relation on the metric space Z.

The distance between two equivalence classes X; and W is defined by
n
d(%,4V) = 111f E d(zl-1,wl)
1=1
where the infimum is taken over finite sequences in Z of the form
ZQ,...,Zn-1, W I ,..., Wn ;zoE%,wnE`W
where zi =- wi mod 9 for all i = 1,...,n-1.
This "instant jet service" distance on Z/gj is symmetrical and satisfies

the triangle inequality. Two distinct equivalence classes may have distance 0:
Example 7.2 Consider the closed interval Z=[0,1] with the identifications
t = 1 -t t E ]0,1[. The "instant jet service" distance between {0} and {1} is 0.
192 COVERINGS
Lemma 7.3 Let Z be a metric space equipped with an equivalence relation

(j, and let f:Z-+W be a distance decreasing map into a second metric space. If f is
constant on the equivalence classes modulo g, then f induces a distance decreasing
map f:Z/g-+W.
Proof Let us first observe that a sum of the type which occurs in 7.1 gives rise
to the following inequality, where we have put z11 = wn
n n n
E d(zi-1,wi) ? E d(f(zi-1),f(wi)) ? E d(f(zi-1),f(zi)) ? d(f(zo),f(wn))
i=1 i=1 i=1
Take infimum over all sequences of the type considered and conclude that the map
f:X/g->W is distance decreasing.
Proposition 7.4 Let r be a group of isometries of a hyperbolic surface

X and Z a closed subset of X whose I'-translates form a locally finite covering of X.
The projection f:X- X/I' induces an isornetry f:Z/gj =+ X/1' where gj is the
equivalence relation on Z induced by the action of r on x and the metric on Z/fj
is instant jet service.
Proof A simple application of lemma 7.3 shows that f:Z/gj X/I decreases
distances. Let us prove that the inverse map f-1 decreases distances. To this end
we consider points z,w E Z with equivalence classes Z,W E Z/gj and show that
d(z,u(w)) > d(%,°W) ; o E 1'
Let us fix c E I' and consider a rectifiable curve y:[a,b]-X with y(a) = z and
y(b) = c(w). We shall show below that there exists a subdivision of [a,b]
a = to < t1... < to = b and elements i = c1,...,cn = c of r such that
7.5 7(ti) E ci(Z) no-i+1(z) ; i = 1,..,n-1
Let us use 7.5 to pick a sequence of points z = w of Z with

y(ti) = ki(wi) = -i+1(zi) ; i = 1,..,n-1
This gives us the estimate
n n
1(7) ? . d(y(ti-1),7(ti)) _ d(o;(zi_1),oi(wi)) _ . d(z;_1,wi) ? d(`%,`W)
Finally, make a variation of y and deduce that d(x,o(w)) > d(Z,` V) as required.
Let us start the construction of a subdivision of the interval [a,b] subject
to 7.5 by observing that y([a,b]) meets only finitely many I'-translates of Z as
follows from Borel-Heine. When y(b) E Z we can use n = 2 and ti = b. Let us
assume that o(w) 0 Z and let tl E [a,b] be the last time -y leaves Z:
t1 = sup{t E [a,b] I y(t) E Z)
Let us observe that any union of I'-translates of Z is a closed set as follows from
the fact that the covering of X with F-translates of Z is locally finite. We can
now conclude that
1(t1) E U , 54 , o(Z)
since the alternative would make y(t1) an interior point of Z, contradicting its
definition. Choose 0-2 of = c such that y(tl) E o2(Z). When y(b) E o2(Z) we
can use n = 3 with t2 = b. If y(b) ¢ o2(Z), let t2 E [a,b] be the last time y leaves
o2(Z) and observe that
y(t2) E U o 0 al p2 o(Z)
since the alternative makes y(t2) an interior point of ol(Z) U o2(Z) contradicting
the definition of t2 and y(b) 0 o2(Z), that is t2 < b. This procedure terminates
since y meets only finitely many F-translates of Z.
We are going to stydy some examples which are usefull for the
understanding of Poincares theorem in VII.3 in case of unbounded polygones.
Example 7.6 Let h and k be the geodesics in the Poincare half-plane H2

given by Re[z] = 1 and Re[z] = a where a > 1. Observe that z-+az transforms h
into k. Let `R, denote the equivalence relation in the strip Z in H2 bounded by h
and k based on the identification z " az , z E h. We are going to show that Z/`J
is not complete. (For a different approach see exercise 5.2).
194 COVERINGS
h k
zi=awi
wi=zi 1-a+1 zi-1
Consider the sequence {zi}i E N in Z given by the recursion formulas above starting
from zo=a+ib, b > 0. It will be useful to observe that Im[zi] = a'b, i E N. Let us
recall from 11.8.2
d(z,w) < 2 sinh 2 (z, w) = 1w - zi (Ir[w]Im[z])-1/2
Let us use this to estimate the distance in Z/%

1)b-1a1-i i>1
d([z]i,[zi-1]) = d([wi],[z1_1]) < d(wi,zi-1) < (a - ;
Recall that a > 1 to see that the infinite series Ed([zi_1],[zi]) is convergent and
conclude that [zi]i E N is a Cauchy sequence in Z/R. In order to show that our
sequence is not convergent, consider the function f:Z-HIS given by f(z) = y/x. The
function f induces a continuous function on Z/% (see exercise 5.2). If our sequence
was convergent, then f would be bounded on the sequence, which is not the case.
Proposition 7.7 Let Z denote the closed subset of H2 bounded by two

non-intersecting geodesics h and k with ends A,B and C,D in the relative positions
shown below. Let o be an even isometry with o(A) = D and o(B) = C. The
equivalence relation given by the identification z - o(z), z E h, is denoted %. If
the ends of h and k are distinct, then Z/% is complete. If h and k have a common
end A = D, then Z/% is complete if and only if o is a horolation.
Proof Suppose that the ends of h and k are distinct. Consider a sequence
(wn)n E N of points of Z representing a Cauchy sequence in Z/`3G. We shall prove
that the sequence is bounded in H2. Once this is done, we conclude that
{wn}n
EN
is contained in a compact subset of Z. It follows that our Cauchy
sequence is contained in a compact subset of Z/%, which makes it convergent.
Let us introduce the geodesic s through A and D and consider the function
f,:H2-+R which is zero on the half-plane containing B and C, while its value at a
point z on the other half-plane is the hyperbolic distance from z to s. We ask the
reader to show that f,:X;R is distance decreasing and we conclude from lemma
7.3 that the sequence f,(zn)n E N is a Cauchy sequence. This implies that we can
find a constant c such that
fs(z1l) < c ;nEN
The same conclusion is available for the geodesic through B and C.
In order to show that our initial sequence is bounded away from the points
A and D, pick a horocycle A with centre A and define the polar function
pA:H2--+R which is zero outside the horodisc bounded by A while the value at a
point z of the horodisc is the distance from z to A measured along the geodesic
through z and A. Let us pick a second horocycle 9 with centre D disjoint from A
and such that c(A fl h) = 9 fl k. The function p = 2(pA + pct) is distance
decreasing and constant on the equivalence classes modulo `J,. It follows from
lemma 7.3 applied to p:Z-*R that p is bounded on our sequence. The same
conclusion is available for the points B and C. Thus we have proved that our
sequence is bounded in H2.
Let us now assume that A = D and that a is a horolation with centre A.
Pick a horocycle A with centre A and observe that the function PA is constant on
the equivalence classes of !. As above we conclude that Z/`R is complete.
When A = D and a is a translation we can use example 7.6 to conclude
that Z/% is not complete. 0
196 COVERINGS
V1.8 HOMOTOPY
In this section we shall study homotopy classes of paths on a complete

hyperbolic surface X. The key fact of topological nature is 4.4 according to which
a local isometry f:X-+Y between complete hyperbolic surfaces is a covering
projection in the sense of 8.1 below. For proofs of 8.2 and 8.3 see [Massey].
Definition 8.1 A continuous map f:X=Y is called a covering projection if

every y E Y has an open neighbourhood D such that for each x E f-1{y} there
exists a continuous section sX: D--+X with s`(y) = x inducing a homeomorphism
f 1{y} x D f1(D) ; (x,v)'-+sx(v)
Unique lifting lemma 8.2 Let f:X-+Y denote a covering projection

and r,s:W--+X continuous maps with fs = fr. If W is connected and X is Hausdorff,
then s(w) = r(w) for some w E W implies that s = r.
Homotopy lifting lemma 8.3 Let f:X-+Y be a covering projection.

For any topological space T, any solid diagram of continuous maps
T x [0] -+0 X
I I f
T x [0,1] -, Y
can be completed with a dashed arrow to a commutative diagram in the category
of continuous maps.
Theorem 8.4 Let x and y be distinct points of the complete hyperbolic

surface X. Any homotopy class of continuous curves joining x to y can in a unique
way be represented by a geodesic curve joining x to y.
VI.8 IIOMOTOPY 197
Proof Let us use the uniformization theorem 3.2 to find a local isometry
f:H2-+X and let us use 4.3 to pick a base point z E H2 with f(z) = x. For a curve
7:[a,b]-+X with y(a) = x and y(b) = y there is a unique curve ,6:[a,b]-+H2 with
,6(a) = z and f,6 = y. It follows from 8.2 and 8.3 that the homotopy type of ,6 is
determined by the point ,3(b) E f-1 {y}. The image in Y of the geodesic line
segment [,6(a) 6(b)] represents the homotopy class of ,6. The uniqueness in the
statement follows from the geodesic lifting property 4.2.
Definition 8.5 By a closed geodesic we understand a geodesic curve

y:R-+X such that there exists a > 0 with 7(x + a = y(x) for all x E R.
Let us recall that two continuous curves a,/3:S1-X are freely homotopic if there
exists a continuous map W:S1 x [0,1]-X such that
'(s,0) = a(s) , 'Y(s,1) = Q(s) s E S'
Theorem 8.6 Let X be a compact hyperbolic surface. A closed curve in X

not homotopic to zero is freely homotopic to a unique closed geodesic in X.
Proof With the notation from the proof of 8.1, let us consider a curve
g:[0,1]-X with g(0) = g(1) = x. Let h:[0,1]-i[2 be a lifting of g with h(0) = z and
let y E IF be determined by h(1) = yz. Since g is not homotopic to zero we have
y # t. We conclude from IV.5.6 that y is a translation with translation axis say k.
We intend to show that the restriction of the projection f:H2-X to k defines a
closed geodesic L---+X which is freely homotopic to g. A free homotopy is formed by
the restriction of f:II2--*X to the hyperbolic quadrangle given below
7z
The horizontal coordinate lines in the quadrangle are formed by y-hypercycles.

198 COVERINGS
VI EXERCISES
EXERCISE 1.1 Let y:[a,c]--*X be a continuous curve in a metric space X.

10 Show that for any point b between a and c the restrictions yl:[a,b]---*X and
y2:[b,c]-aX are rectifiable if and only if y:[a,c]-*X is rectifiable.
2° With the notation above assume that y is rectifiable and show that
1(7) = 1(71) + 1(72)
30 Show that a rectifiable curve y:[a,c]-YX which satisfies
1(7) = d(7(a),y(b))
is a geodesic curve. Hint: Show that y is distance preserving.
EXERCISE 1.2 Let us fix a point u+iv of the Poincare half-plane and consider
the real function V)(x,y) in two variables given by
1G(x,Y) = d(x+iy,u+iv)2
10 Show that the Hessian of this function at the point u+iv is
= v(dx2 + dy2)
d2.0(x,Y)
2° Show that a continuously differentiable curve y:[a,b]-+H2 is rectifiable

with curve length b
1(7) _ (x'(t)2 + Y'(t)2)/Y(t) dt
where 7(t) = (x(t),y(t)) , t E [a,b]. a
EXERCISE 1.3 Show that a circle in H2 with radius r has arc length 27r sink r,
compare VI.1. Show for example first that the length k11 of the edge of an inscribed
n-gon is given by the formula
sinh(k,i) = sinh(r) 2 sin(7f/n)
and use this to calculate the length of the circle as lien nkn
n--;00
EXERCISE 1.4 Consider a fixed real number y > 0 and let Clf denote the
Euclidean line in the upper half-plane through iy and parallel to the x-axis.
VI EXERCISES 199
10 Show that 'J is a horocycle, and that the are length is given by
length (iy,x+iy) = ; x>0
Y
2° Verify the formula
length (iy,x+iy) = 2 sinh 1-2d(iy,x+iy) ; x>0
EXERCISE 2.1 Let X be a metric space and 7:[a,b]-#X a geodesic curve. Show
that the curve y is rectifiable with arc length 1(y) = b - a.
EXERCISE 5.1 Let X be a hyperbolic surface and G a group of isometries of X

which acts discontinuously on X and let f:X- X/G denote the canonical projection.
Show that a subset U of X/G is open (in the sense of the metric on X/G) if and
only if f"1(U) is an open subset of X.
EXERCISE 5.2 Let a be the isometry of the Poincare half-plane given by zp- az,
where a > 1 is a fixed real number.
10 Show that the group IF generated by a acts discontinuously on
X = {zEH2IRe[z]>0).
2° Let f:X-Q be the function given by f(x+iy) = y/x, x+iy E X. Show that
f induces a continuous function on the surface X/I'.
3° Show that the surface X/I' is not complete. Hint: Use 5.9.
EXERCISE 7.1 With the notations and assumptions of proposition 7.7. Show
that the set X = U EZ
onZ is open and connected. Prove that Z/' is
complete if and only if X = H2. Hint: Apply propositions 5.9 and 7.4.
EXERCISE 8.1 Let I' be a discrete group of isometries of H2 and y E r a

horolation. Explain how to construct from y a closed curve in H2/17 unique up to
free homotopy. Hint: Model the proof of 8.6.
VII POINCARE`S THEOREM
We have arrived at the central theme of the book, Poincare's theorem,

which allows us to construct a discrete group from a given hyperbolic polygon with
side pairing and to write down generators and relations for the group.
The proof of the theorem is a beautiful application of the monodromy
theorem VI.4.8. In order to display the geometric ideas of the proof, we have first
treated the case where the polygon is compact. The general case requires a more
technical argument based on completeness and is given in section 3. Let me
mention that Poincare's original proof [Poincare] is incomplete in the non-
compact case, compare [de Rham].
VII.1 COMPACT POLYGONS
Let us consider an ordinary polygon A in the hyperbolic plane H2 with a

side airin . By this we understand an involution si-+*s on the set of edges g of A
such that s and *s have the same hyperbolic length and such that
1.1 -1
;sEg
Recall that an edge s is an oriented side and that s-1 denotes the result of changing
the orientation. For s E 6 we let ors denote the isometry of H2 which maps *s to s
in such a way, that locally, the half-plane bounded by *s containing A is mapped
to the half-plane bounded by s but opposite A. The isometry as is called the it
transformation determined by s. Let us at once observe that we have the
Reflection relations 1.2

or *s = as
1
;sEg
VII.! COMPACT POLYGONS 201
Given an edge s of A, we let Is denote the edge with the same initial vertex as s
but different final vertex. This defines for us a second involution on 6, su- s. The
combination of these two operators defines the edge operator T:9-9 given by
1.3 IQs= f*s ;sES
The edge operator I on g is bijective and g is finite so we conclude that W has

finite order. The cycles for ' on g are called edge cycles. The sequence of initial
points of an edge cycle is called a vertex cycle (a vertex cycle may contain
repetitions).
Lemma 1.4 Let sl,...,s, be an edge cycle with vertex cycle P1,...,Pr and side
transformations (oi = o-S , i = 1,..,r). The cycle ma o = 03 r is a
rotation around P1 of angle (measured in the direction determined by A and sl)
congruent mod 27r to the sum of the interior angles
LintPi + LintP2 + ... LintPr
Proof Let us choose the orientation of 112 which places A on the positive side
of s1 near P1. With the notation of 111.2.4 we have
Lor(si,1si) = sign(si) LintP1 mod 27r ; i = 1,..,r
where sign(si) = ei equals +1 (resp.-1) in case z lies on the positive(resp.
negative) side of si. From the very construction of the side transformations it
follows that
sign(sm) = sign(s,u+1) m = 1,...,r
Let us multiply these equations together for m = 1,...,r to get that
sign(s1) = det(ol...or) sign(sr+1)
Since sr+1 = s1 we conclude that the cycle map is even. Let us proceed to
calculate the rotation angle of the cycle rrta.p. Let us prove a general formula
Ei=m...r.int Pi = emLor (sm, om...or(sl)) m = 1,...,r
To this end we recall that
si = oi(*si) , °i+1 = *si , (Si = 0i(si+1)
202 POINCARE'S THEOREM
This will be done by decreasing induction on in. The case m = 1 runs as follows
LintPr = Er' or(Sr'Tsr) = ErLor(Sr,0r(Sl))
The induction step m+1 follows from the formulas we have just derived
Ei=m.... rLint Pi = emLor(sm'lsm) + em+1Lor(Sm+1, 47m+1"'0r(Sl))
emLor(Sm,lsm)+Em+l det(om)Lor(lsnl, am"'Qr(Si)) _
emLor(Sm,lsm) +EmLor(1Sm,Om...Or(S1)) = EmLor(sm, om...or(sl))
This concludes the proof of our formula. Finally, put m = 1 in our formula and
recall that el = 1 to get the announced rotation angle. 0
Examples Consider an 8-gon with side pairing as indicated in the drawing.
P3
a=0a, 0 =0-b,'Y=0-c1 b=od.
Edge cycle: a*b"1*a 1b c *d"1 *c-1 d

Vertex cycle: P1P4P3P2P5P8P7P6
Cycle map: a/3 1a'113yb"ly"1b
The identification space is a sphere

with 2 handles (oriented surface of genus 2)
Our next example is an isoceles triangle with the side pairing indicated on the
drawing, the base edge c is paired with itself *c=c. The side transformation a=oa
is a rotation with centre C, while -y=ac is a reflection in the geodesic through c.
C
Edge cycles a *a 1 , a c *a c
Vertex cycles C C , BAAB
Cycle maps a, a 1
, ayay
The identification space is a disc.

Let us consider a different side pairing on the same triangle given below: again
*c = c. The transformation y = o, is a reflection as before, while the
transformation a = oa is a glide reflection.
Edge cycle: a a 1 c *a 1 *a c'1

Cycle map: a2ya 2y
Vertex cycle: BCAACB
The identification space is a Mobius band.
Remark 1.5 Let us consider an edge cycle SI S2,...Sr. From sj+1 = 1*sj,
i = 1,...,r, we conclude that Jsj+1 = *si. This gives us
I'(lsi+1) = ls, ; i = 1,...
It follows that lsl,Jsr,lsr_1,...,1s2 form an edge cycle. Notice that the corresponding
vertex cycle is P1,P,,Pi_1,...,P1, which is nothing but the reverse cycle of the
original vertex cycle. It follows that a vertex cycle makes up a full equivalence
class of A under the equivalence relation generated by the side pairing. Let us
observe that
1S111Sr,1sr_11 ...,152 = *Sr,*Sr_1,...,*S1
and conclude that the cycle map is o,. 10,, 2 1...01 1, which is the inverse of the
cycle map of the original cycle.
Cycle condition 1.6 We say that an edge cycle s11 ...,sr satisfies the
cycle condition if the angle sum along the corresponding vertex cycle P1,..-,Pr has
the form
i=1...r Li.,t P = 2'r

nc ; nc = 1,2,3...
The cycle map o = al-or of an edge cycle which satisfies the cycle condition is a
rotation about P1 of order no.1
'Cycles c with nc = 1 are sometimes called accidental cycles

Poineare's Theorem 1.7 Let A be a compact convex polygon A with

a side pairing which satisfies the cycle conditions 1.6. Then, the group G genera-
ted by the side transformations as, s E g, is a discrete group with A as a
fundamental domain. A complete set of relations for these generators is the
reflection relations 1.2 and the cycle relations
o.cnc = d
one for each edge cycle c.
Proof Let I' denote the universal group on the generators as , s E g, subject to
the reflection relations 1.2 and the cycle relations. The canonical map F-+G is
denoted On the basis of these data we shall construct a topological space X
and a local homeornorphism f: X- H2, compare the introduction to chapter VI.
The set underlying X is the quotient space for the equivalence relation on r x A
generated by relations of the form
1.8 (7QS,h)'.,('Y,vs(h)) ; 'YEr,sEfi,hE*s
The equivalence class of the point (y,k) is denoted y. k E X. Let us equip I' x A
with the product topology: an open subset has the form U = U-7 E r {0.) x U0.
where U. is an open subset of A, u E r. The topology on X is the quotient
topology: open subsets of X correspond to open, - stable subsets of IF x A.
The evaluation map (y,h)i-- (h), I' x A-+H2 is continuous and compatible
with our equivalence relation, so it will induce a continuous map f.X-*H2
y(h) ; 7EI, heA
The natural action of r on r x A induces an action of r on X through the formula
a(13. s) = (a13). s I',sEA
Let us make the observation that f:X- H2 is F-equivariant in the sense that
f(rx)=rf(x) ;rEI ,xEX
Let us prove that X is connected. To this end let us note that sr-+t.s induces a
continuous section of f:X-H2 over A and conclude that t.A is a connected subset
of X. For x E X given, let us construct a connected subset of X containing x and
meeting t.A. It is clear from the construction that X is the union of all 17-translates
VII.1 COMPACT POLYGONS 205
of t.0. So let us pick r E r with x E r.4 and let us write r = al...a. where the cis
are side transformations with respect to the edges s1,...,sn of A. For i = 1,...,n we
get from 1.8 that t.0 meets oi.A, which implies that Oi...Oi_i.0 meets
The chain
t.0 U o 1.0 U QiQ2.0 U ... 0, 10'2...On_1.0 U T.0
is a connected set of the required sort.
We proceed to construct some special neighbourhoods of a given point
x E X. For convenience we introduce the following terminology concerning the
equivalence class of r x A represented by x. But first let us agree that for z E A
and c > 0 we shall put A(z;c) = {x E i d(x,z) < (}.
Local tesselation property 1.9 We say that an equivalence class

C of F x A has the local tesselation property if the following three conditions are
satisfied. 1° C is a finite subset of r x A
C : (i 1,k1),...,(ilm,kn1) ; t)i E r , ki E A
2° For e>0 sufficiently small, the set
{,71) x A(k1;e) U ... U {77m) x A(km;f)
is a - stable subset of r x A and 3° the corresponding open subset of X
171.0(kl;r) U ... U
is mapped by f bijectively onto the open disc D(z;c), z = 1(k1) _ '
We are going to show that all equivalence classes of r x A have the local
tesselation property. Observe that if the equivalence class C has the local
tesselation property, then the same is true for the class yC for all y E F. Thus it
suffices to verify that for each z E A, the equivalence class generated by (t,z) has
the local tesselation property. Let us investigate a number of special cases.
The full equivalence class of a point of the form (t,z), z E Into, consists of
that point alone. Any subset of {t} x A is stable in r x, and the open subset
t.IntA of X is mapped by f:X-+H2 onto InI A.
The full equivalence class in r x A generated by a point of the form (t,a),
where a is an interior point of an edge s, consists of two points
(t,a) , (os,b) ; b = vs 1(a)
For e > 0 sufficiently small, the following subset of I' x A

{t} x A(a;e) U {as} x O(b;()
is open and stable. The image in X of this set
t.A(a;e) U vs.O(b;c)
is mapped by f:X-'H2 onto D(a;e).
Let us now consider a vertex P1 of A. Pick an edge sl with initial point
Pl and iterate the edge operator T to get a sequence of edges sl,...,sr,...,sd, where r
is the length of the edge cycle and d = nr where n is defines in 1.6 such that
n1_1 ,...,r Lint Pi = 2ir
We let P1,...,Pd denote the corresponding sequence of initial vertices and
the sequence of side transformations. Let me stress that both sequences have
period r. Observe that P1}1 is the initial vertex of s1+1 and (si+1 = *si, this gives
us vi(Pi+1) = Pi for i = 1,...,r. We conclude that the full equivalence of class of
the point (1,P1) in I' x A is
(t'P1) , (o1,P2) , (C1O-21P3) ,..., (Q1O2...Ud-1'Pd) ; d = rn
Let us prove that this equivalence class of r x L satisfies the local tesselation
property 1.9. The stability condition follows basically by observing that the
boundary of O(Pi;c) is located on the edges si and f si and that
Ili-10Si) = oi-1(*si-1) = Si-1 ' Si = o1(*Si) = o1(Ts1+1)
from which we deduce that
0'i-1.1si = t. si_1 , Si = 0'i- 1si+1
It is now time to calculate the image of in H2, that is
vlo2...vi_1O(Pi;e)
The sector A(Pi;e) is bounded by L(si,f s1). The image is bounded by the circle
with centre P1 and radius a and the two edges
0"1"',7i-1(Si) ,
.71...o1-1(jSi) = 01...0i_2(si-1)
Let us use the notation from the proof of lemma 1.4 to calculate the oriented angle
between these two edges
L0r(ol...oi-1(si)'Q1...Q1-1(lsi)) = det(o1)...det(ai-1)Lor(sPlsi)
det(ol)...det(v1_1)sign(si)LintP1 = sign(s1) Lint Pi
Observe that if these angles are added from i = 1 to i = d = nr we get sign(sl)21r.
From this it follows that our class satisfies the local tesselation property 1.9.
Let us show that the open subsets of X exhibited in 1.9 form a basis for
the topology on X: with the notation above, an open subset U of X which contains
the point x representing the equivalence class C can be thought of as a - stable
open subset V of r x A containing (rj1,k1),...,(r11,,,k1 ). This implies that V contains
{r1j} x A(k;;e) for i = 1,...,m and e > 0 sufficiently small as required. As a conse-
quence of the local tesselation property, we find that f:X-*H2 maps open subsets of
X to open subsets of H2 . Indeed f is a local homeomorphism.
It is time to show that X is a Hausdorff space: let 10 C be equivalence
classes where C is given by 1.9 and `D is given by
; j E I' , hl E A
Let us consider a fixed pair ij and show that we can find (ij > 0, such that
{t1i} x 0(k,;E) n x A(hj;E) = 0 ; E E }O,Ejj]
If rh # tj any eij > 0 will do. If T1, = j then k, # hj and it suffices to make sure
that A(k1;e) and A(hj;e) are disjoint.
As another consequence of the local tesselation property, we conclude that
the r-translates of t.0 form a locally finite covering of X: let us consider the
point x E X given by the equivalence class C from 1.9. If we let U denote an open
neighbourhood of x in X of the form displayed in 1.9, we have that
o(t.-A) n u = 0 a7 711,...,17m
In particular, we have ax E U for finitely many o E I only; that is the action of r

on X is discontinuous.
Let us prove that X/I' is complete. The continuous map A-X, hp-+t.h,
induces a surjective map ._A->X/1' which implies that X/I is compact and therefore
complete. We conclude from VI.5.9 that X is complete.
We can now use the monodromy theorem VI.4.8 to show that f:X-?H2 is
bijective. Surjectivity of f means that
H2 =UgEG g(-A)
From the very definition of the equivalence relation follows that
o.0nt.In(L =0 ;0El,a}tt
The injective map f:X-+H2 transforms disjoint sets into disjoint sets, thus we get
0(A) n InlZ = 0 ; a E r-{t}
This shows that the kernel of r-G is trivial. It follows that the canonical map
from the universal group r to the group G is a bijection and that A is a

fundamental domain for the action of G on H2. 0
Parallelogram groups 1.10 Let us consider a parallelogram ABCD

in H2. By this we mean a convex quadrangle ABCD where the opposite sides
have equal length. The elementary geometry of the parallelogram tells us that a
half-turn with respect to the midpoint of AC transforms DABC into OCDA.
Otherwise expressed, the diagonals have a common midpoint M and a half-turn
with respect to M transforms the parallelogram into itself. The angle sum of the
quadrangle is less that 27r, 111.9.7. Let us assume that this has the form
LA+LB+LC+LD = 2n ; n= 2,3,...
We can apply Poincare's theorem to the side pairing *AD = BC and *AB = DC.
The corresponding side transformations are denoted or and 0.
Edge cycle : a b *a 1*b-1

Cycle map : a/3a 10-1
Vertex cycle : ABCD
The cycle transformation a,3a-10-1 is a rotation around A of order n. We find

that the parallelogram group G is the universal group on a and )3 subject to
(a)3a1,3-1)11=c
VII.2 TRIANGLE GROUPS
Let us consider DABC in the hyperbolic plane H2. Reflection in the

geodesic BC is denoted a, reflection in CA is denoted ,3, while reflection in AB is
denoted y. We shall assume that the angles have measures as follows
2.1 LA=p, LB=E, LC=r p,q,rEN
Since the angle sum of ABC is < 7r we must of course assume that
2.2 ; p,q,r E N
Conversely, for any three integers p,q,r satisfying this condition, we can find
DABC with angles of these measures; this triangle is unique up to congruence,
compare 111.5. The subgroup of Isom(H2) generated by the reflections a,(3,y is
denoted G(p,q,r).
The isometry yQ is a rotation around A of angle 2a/p which makes y/3 a
rotation of order p, similar remarks apply to the vertices B and C. We can now
write down three reflection relations and three cycle relations for our three
generators
2.3 a2 = t, l32 = i, y2 = t, (7/)p = t, (ay)q = , (a)3)r = t
Let us apply Poincare's theorem to L ABC with the trivial pairing *a = a, *b = b,

*c = c. The edge cycles are
c b-1 , a c-1, a1 b
This realises the relations 2.3 as reflection relations and cycle relations.
A picture of the tesselation of the Poincare disc by the fundamental
domain for the triangle group D(2,3,7) is given in the introduction.
Let us work out generators and relations for the even part G+(p,q,r) of the
triangle group. It is quite clear that G+(p,q,r) is generated by
P2A=N7, P2B=rya ,P2C=a13
From 2.3 we derive the following relations for the ps
2.4 P2A P2B P2C=' , P2AP=1, P2Bq=t , P2Cr=,
Let us show that the list 2.4 is a complete set of relations for G+(p,q,r). To this
end we apply Poincare's theorem to the quadrangle ABCD where the point D is
reflection of B in the geodesic through A and C.
Edge cycles: a*c 1, c 1*a, a-, *a 1, c, *c. The side transformations are oa=P2C
and o*c = P2A as follows from
P2C(D) = aa(D) = a(B) = B , P2A(B) = /37(B) =,3(B) = D
Thus Poincare's theorem gives us the following complete set of relations
(P2CP2A)q = P2C' = P2AP =
which is the same as 2.4.
We shall continue to give some applications of Poincare's theorem to
compact polygons. The following notation is useful and will be exploited in VII.6.
Definition 2.5 By a boundary cycle of a polygon A we understand a

cyclic enumeration of its edges a1,...,a1, a1) where the final vertex of ai
equals the initial vertex of a1+1 for i = 1,...,n. Once the hyperbolic plane is
oriented, the positive boundary cycle is defined by requiring that locally A is
situated on the positive side of the edges of the boundary.
Hyperbolic pants Consider a rectangular hexagon `A in H2 with boun-

dary cycle a1,...,a6 and side pairing *ai = ai, i=1,...,6. Poincare's theorem tells us
that reflections p1,...,p6 in the sides of `3v generate a discrete group H with
fundamental domain %. A complete set of twelve relations is given below
pie
=t , P,Pi+l = Pi+1Pi
; i = 1,...,6 (P7 = Pl)
Consider a second copy of the same hexagon and sew the two hexagons together
along the seems a2,a4,a6. The result is a pair of hyperbolic pants. In order to
realise this as an orbit space, consider the rectangular octagon A = U p4(A)
with the side pairing given below (*a = a, *c = c, *e = e, *f = f).
The boundary cycle is a b c *b-1

d e f *e-1. The two side transformations are
ob = p2p4 and oe = p4p6. We leave it to the reader to apply Poincare's theorem
and conclude that the group I' generated by
P11 P3, P5, P64' P42' P26 Pij = PiPj
is a discrete subgroup of index 2 in II with fundamental domain A. The cycle and
reflection relations may be rewritten in the following form
2.6 P12 = P32 = t, P52 = t, P64 42P26 = t

P1P26 = P26P1 , P3P42 = P42P3 , P5P64 = P605
Torus with a hole Let us start with a right angled hexagon with
boundary cycle a1a2a3a,1a5a6 as before. This time we require a2 and a6 to have
the same length; formula III.8.1 implies that a3 and a5 have the same length as
well. Reflect this in the side a1 to get an octagon with the side pairing as
indicated
Edge cycles: ab*a1*c, a*c1f1c1
The group is generated by the side transformations a (3 y r) 0 subject to
2.7 a,3a"1 =7,712=t, 2=I,'Y?11'y1=0
It follows that the group is generated by a, 3 q subject to the sole relation q2 = 1.
Example 2.8 Let us fix a pentagon A with angle sum inn and consider the
side pairing with boundary cycle
a*b"1*a1 be
in particular the edge c is reflected in itself. The edge cycle is

ab*a1*b"1cb'1a1*b*ac1
With the convention a = oa... we find a complete set of relations
(a,3a"1)3-ly fa,6-la 1y)" = c, y2 = t
It is convenient to introduce e = [a,3]"1
and C1-y( = 6 to get
[a,3]e,6=c-1'Yc,(6Y)n=e,62=y2=t
The orbit space is a torus with a hole and a marked point on the boundary.
VII.3 PARABOLIC CYCLE CONDITION 213
VII.3 PARABOLIC CYCLE CONDITION
Let us consider a polygon A in H2 with some of its vertices on OH2. To

be sure the sides of the polygon are geodesic arcs in H2 or arcs of OH2. The edges
supported by arcs on the boundary are called free edges. The set of edges
including the free ones is denoted 9. Let there be given a side pairing that is an
involution si *s on g with
3.1 =(*s)-1 ;sE9
The free edges are supposed to be fixed points under * while the non-free edges are
subject to the following conditions: if the initial, resp. final vertex of the edge s is
finite we require the initial, resp. final vertex of *s to be finite as well ; if both
ends of s are finite, we require s and *s to have the same length.
We shall assume that for each edge s E 9 there is given an isometry as of
H2 subject to the conditions: for a non-free edge s the map as maps the edge *s
onto s in such a way that locally the half-plane determined by *s and A is mapped
to the half-plane bounded by s but opposite A. For a free edge s we require as = t.
Observe that unless both ends of the non-free edge s E S are infinite then
the isometry as is uniquely determined by these requirements. In general the
precise determination of as is part of the structure.
The isometry as is called the side transformation determined by s. It is
required that these data satisfy the
Reflection relations 3.2

als = as , aas *s = ;sE9
For an edge s of A we let is denote edge with the same initial vertex as s but
different final vertex. This defines for us a second involution on g, s I f s. The
composite of these two operators defines the edge operator 41: given by
3.3 1Ys=1*s ;sE9
Let us introduce the set 1900 of edges with initial point on OH2. Observe that goo
is stable under * and 1 and conclude that goo is stable under the edge operator W.
The cycles sl,.... sr of 'Y on go, are of two sorts: free cycles which contain free
edges and parabolic cycles which do not contain free edges. The cycle man
a= a parabolic cycle is even, as follows from the proof of 1.4. Let us
remark that o is a parabolic transformation if the side transformations generate a
discrete group with fundamental domain .A, compare V.2.4. The cycle map of a
free cycle is a reflection or a as follows from the general
Lemma 3.4 Let sl,...,sr be an edge cycle with side transformations o1,...,0r
and cycle map o = 0102... or. If none of the edges of the cycle are fixed by * then
the vertex cycle P1,...,Pr has no repetitions. Conversely, if the vertex cycle
P1,...,Pr contains a repetition, P1 = P11 say, then the edge cycle s1,...,Sr contains
exactly two edges fixed by *, and the "half cycle maps"
01...0n-1 , 0n...or
are conjugates of the side transformations relative to the two edges fixed by *.
Proof From P1 = Pn and s1 $ sn we get s1 = Isn, and we conclude that

s1 = 11*sn_1 = *sn_1, S2 = 1*s1 = 1Sn.1 = 11*sn-2 = *Sn-2
A simple induction on gives us
Let us assume for a second that n is odd, say n = 21n+1. Put i = n1 in the formula
above to get that sin = *snl+1 or *sm = Sm+1 = !*s., which is absurd. In
conclusion the number in must be even, n = 2m. From the boxed formula with
n = 2m and i = m we get that sm = *sm. From the same source we get that
sl,...,sm,...,s2m-1 = 511 ...1sii-1'sm'*sin-1,...,*Sl
and we conclude that
VII.3 PARABOLIC CYCLE CONDITION 215
0'1 ...on_1 - 1
which shows that a,...on_1 is a conjugate of o,u

Let us proceed to apply this result to the opposite edge cycle
From sl = sn we get Isl = f sn or *Sr = *sii_1. Thus we can conclude that the
sequence sn,...,sr consists of an odd number of edges and that the middle edge sf is
fixed, *sf = sf. Moreover, we find that on...or is a conjugate of of.
Conversely, let us suppose that our cycle sl'...'sr contains and edge c with
*c = c. Let us consider the full infinite sequence of edges defined by
ci='Y'c ;iEZL
Let us make a simple observation that *c = c = j*c_1 which gives c1=*c_1. By
continuation of this process we obtain the formula
*c-m=cm ;mEZ
or cl_m = J,cm. Thus we can rewrite our sequence as
Ic4, Jc3, Jc2, Icl, c1, c2, C3 ...
It follows that the sequence P1,...,Pr. of initial points of sl,...,sr contains a
repetition, PI = P. say. With the notation above we must have c = sm or c = sf.
Example 3.5 Let us consider a pentagon with all vertices on OH2
Boundary cycle :
a*b"l*al
be
Edge cycle :
a b *al*b-lc b-la l*b *a c-1
Cycle transformation :
aQa la-1y )3a,6-'al'Y
C
The vertex cycle is ADCBEEBCDA. The "half-cycles" relative to a point P are

reflections, providing a decomposition of the cycle map into a product of two
reflections in geodesics through P. It follows that the cycle map is parabolic.
Example 3.6 Let us consider a quadrangle with a free edge cycle.
Edge cycle : a c *a b
Vertex cycle: ACCA
Cycle map: reflection fi in b
Theorem 3.7 The conclusion of Poincare's theorem 1.7 remains valid for
polygons with vertices at infinity, provided that each parabolic cycle s1,...,sr
generates a parabolic cycle map
Proof Let us return to the proof of 1.7 and observe that it suffices to prove
that X/r is complete. According to VI.7.4 this space is isometric to the space
t.0/g where g is the equivalence relation on t.0 induced by the action of I' on X.
Projection t.0--* decreases distances2 and induces a distance decreasing bijection
t.0/g -A/%
where % denotes the equivalence relation on A generated by the side pairing. On
the other hand, the presence of the continuous section, zHt.z , 0->t.A shows that
the bijection above has a continuous inverse. Thus it suffices to prove that A/96
is complete.
Let us consider a sequence (wn)n E N
of points of A representing a Cauchy
sequence in A/%. We intend to show that this sequence is bounded in H2. Once
this is done, we conclude that our Cauchy sequence is contained in a compact
subset of A/`3U, which makes it convergent in -A/%.
Consider first a free edge AB of A. Let us pick a hypercycle C with ends
A and B, which does not meet OA. The function f f:H2--*R introduced in the proof
of VI.7.7 is distance decreasing and induces a distance decreasing function on
2 When A is convex, the projection t.-A= A is an isometry, a priori.

VIL3 PARABOLIC CYCLE CONDITION 217
A/%, VI.7.3. It follows that fe is bounded on the sequence (wn)n E N. As a

consequence we can find a hypercycle C as above such that none of the points of
our sequence belongs to the region bounded by C and the free edge AB.
Let us consider a parabolic cycle s1,...,sr. Pick a horocycle A at the initial
point P1 of s1 and let Al,...,Ar be defined recursively by
Ai = a1Ai+t , Ar _ o,.Al
Notice that .A = Al since o = Q1.... O-r is either a horolation or t. If repetitions
occur in the sequence P1,...,Pr, the same repetitions occur in the sequence Al,...,Ar
as follows from 3.4 : "half-cycle maps" are reflections.
With the notation from the proof of VI.7.7 let pi denote the radial
function defined by A1. Recall that pi is zero outside the horodisc bounded by Ai
and that pi is distance decreasing. Let us form the distance decreasing function
p:A-*R
p = r (P1+... +
and observe that this function is constant on the equivalence classes modulo `Jb
when the horocycle A is sufficiently small. We conclude from VI.7.3 that p defines
a distance decreasing function on A/%. It follows that the sequence (p(wn))n E N
is convergent and we conclude that p is bounded on (wn)n E N' In conclusion we
can choose the horocycle A such that the sequence (wn)n E N does not meet any
of the horodiscs bounded by -411...)A,.
Let us consider a free cycle s1,...,sr. It follows from 3.4 that the cycle
contains one or two free edges. In the case of two free edges we get from 3.4 that
all "half-cycle maps" are t. If the number of free edges is 1 we get from 3.4 that
the "half-cycle maps" are i and reflection in a geodesic. It follows that the cycle
map is a reflection in a geodesic with end P1. Thus a free cycle can be treated just
like a parabolic cycle with parabolic cycle map. Our journey has ended !
VII.4 CONJUGACY CLASSES
In this section we shall show how to use a fundamental domain to deter-

mine the conjugacy classes of parabolic and elliptic elements in a discrete group r.
More precisely we are given a polygon -A with a side pairing satisfying the
hypothesis of Poincare's theorem 3.7 and we look for conjugacy classes in the
group r generated by the side transformations of A.
Let us start by the observation that an elliptic element is conjugated to an
an element which stabilises a point of 8 while a parabolic element is conjugated
to an element which stabilises a point of (9h-A, V.2.7. This makes it relevant to
determine the stabiliser IPp of point of an edge s of A.
Proposition 4.1 Let P be an interior point of the edge s. The stabiliser rp

is trivial except in the following two cases
*s = s : rp is generated by the reflection os.
*s = s-1 and P is the midpoint of s : rp is generated by os.
Proof Let us assume that P E 112 is an inner point of the edge s. Observe that
A U arA is a neighbourhood of P and conclude that a y E I'p must equal t or os.
When y = os, we conclude that P is a fixed point for os and we get that *s = s or
*s =s-1. In the first case as is a reflection in s, in the second case os is a half-turn
in P.
Let us assume that P E 0h0. Observe first that an even y E rp is
parabolic as follows from V.2.4. It follows that an odd y E rp must be a reflection
in a geodesic through P since y2 can't be proper hyperbolic. It also follows that
horocycles with centre P are stable under r. When P is an interior point of a free
side s of A we conclude that A contains a full horocycle A with centre P. For
y E rp we conclude that hat A meets y hat A, thus y = t. 11
VILA CONJUGACY CLASSES 219
Theorem 4.2 Let P be a vertex of the polygon A and let sl,.,.,sr be a full
edge cycle with vertex cycle P = P11 ...,Pr. If the vertex cycle is without repetitions
then F is generated by the cycle map a,...a,. If the vertex cycle has a repetition,
P1 = Pn say, then 1'p is generated by the two "half-cycle maps"
Q1...an-i +
an...61.
Proof Let us first consider the case where P is a point of H2. Let us inves-
tigate the following union of translates of I'
3=AU0_1AUa10_2AU ...Ua1a2...ar-1L
From the formula ai(Pi+1) = Pi it follows that each of these translates passes
through the point P1; in fact P1 = °1...oi(Pi+1) Observe that
Q1...ai-10 fl ul...oi(A) D o1...vi-1(si)
as follows from si = vi(*si). Let us consider a small circle A with centre Pt and
observe that 3, = j fl A is a connected with J, fl vJ, # 0. More precise information
is available: the arc 2. subtends an angle 27r/n where n = ord(o ). It follows that
translates of J of the form Qsj, s E Z, cover an open neighbourhood of P. For
'Y E 1'p we conclude that y Int A meets one of these translates ; in fact y must have
the form
y=as or yasQ1...Qn.1 ;n=2,...,r,sEZ
In the second case, we must have PI = u1...an-1(P1), i.e. P1 = P. It follows that
0" is a "half-cycle map".
Let us consider a parabolic cycle sl,...,sr. The main point is to show that
the cycle map Q1...an is non-trivial. Pick a 1lorocycle A at the initial point P1 of
sl and let A1,...,A,. be defined recursively by
Ai = °iAi+1 ,
A,. = O- A 1,...,n
Notice that A = Al since o = al.... ar is either a horolation or L. If repetitions
occur in the sequence P1,...,Pr, the same repetitions occur in the sequence Al,...,Ar
as follows from 3.4: "half cycle maps" are reflections. We can now measure the
the distance from st to a(s1) along the horocycle A and show that this equals the
sum E i=1 r 0i where ©i is the length of the horocyclic arc cut by (si,1si) in A.
The detailed argument is like the proof of 1.4
We can introduce the horocycle arc L = .A n J and observe that I. n o3. 0 0 implies
that .A is covered by translate o-5L, s E Z. We can proceed as above taking the
second half of the proof of 4.2 into account.
Let us now turn to a free cycle. Let us treat the case where *sr = Is1 is a
free edge. The full cycle can be written, compare the proof of 3.4
s1...s11_1sn*sii_1...*sls1- ; r = 2n
with *sn = s, . In the case where the edge sn is non-free we have found that sn is a
reflection and it follows that o = 01...oi_1 is a reflection. Observe that the
horocyclic arc L _ .A n J contains .A n A and .A n o0. Since L is connected we
find that 3, = .A and we can proceed as before. In the case where the edge sn is
free, let us observe that Int 0 and o1...on-1Int 0 are disjoint. It follows that A and
cover the two ends of A and we can conclude that k = A. We can
finally conclude the proof as above. 0
VII.5 SUBGROUPS OF THE MODULAR GROUP
In this section we shall apply our theory to some important subgroups of

the modular group introduced in V.1. Let us take our point of departure at the
fact that the hyperbolic triangle 0 with vertices i, (, oo is a fundamental domain
for the extended modular group G = PGI2(7L). For unexplained notation see V.1.
Straightforward application of Poincare's theorem to z gives us the
generators a,#,y for G and the complete set of relations
5.1
-1/2 1/2
Commutator group From the relations for G it follows by pure group

theory that its commutator subgroup [G,G] has index 4 in G with its factor group
generated by Q and y (observe that a = y mod [G,G]). It follows from V.1.4
applied to that we can use the hyperbolic quadrangle -(, 0, C, oo as
the fundamental domain for [G,G]. Let us apply Poincare's theorem to get:
of=p=TQ= 3a7O f
0'r=ay=a)30-y=o-p-10"
Boundary cycle : r*r 1*flf -C
Elliptic edge cycles : r'1 , f

Parabolic edge cycle : r *f1, *r fl
It follows that [G,G] is generated by the third order rotations p and opo and that
there is no relation between these two generators. For further reference let us note
that the parabolic cycle transformations are
(oP 1.Y) P 1 = r-2 , (op 10) 1P = OT2o
It follows from our investigations that G has three subgroups of index 2.
The modular group r is generated by o and p with the complete set of

relations p3 = t and a2 = t follows by applying Poincare's theorem to the
it
modular figure from V.I. From these relations it follows that [r,r] has index 6 in
F with a factor group generated by ap = T. In particular, we conclude that r has
a unique subgroup F2 of index 2, in fact r2 = [G,G].
From our investigations above it follows that r has a unique normal sub-
group r3 of index 3. The group r3 contains or = 03 and its conjugates ror-1 and
r for. Let us show that r3 is generated by these three elements of order two
subject to no further conditions. From p = ro it follows that r/r3 is cyclic of
order 3 generated by T. We can use V.1.4 to conclude that r- 14 U 4 U T.d repre-
sents a fundamental domain.
3
7
Side transformations :
r3, 7 _1C 7 , U, 7-?r 1
Cycle relation :
(r1oT) o (ror1)T3=t
T
1cr a TOT-1
The cycle relation, which may be read 7-2p37 2 = t , eliminates the generator T3.
The remaining cycle conditions simply express the fact that r1or, 0-, TUT- I all
have order 2. We may express this by saying that r3 is the free product of three
cyclic groups of order 2.
Proposition 5.2 The commutator group [r,r] has index 6 in r and the
group mr,r] is cyclic. The group [F,I'] is a free group on the two generators
?T IOT , QrUT 1
Proof The group has index 2 in r3 with a factor group generated by o.

It follows from the discussion above that [r,r] has the fundamental domain
U
*k
*n
n m
1 R fk
Two of the side pairings can at once be written down

p = 0T 1QT , V = 0ro.r 1
taking into account that 11(r"1(()) = -( and v(T(-O) _ (. Observe that
µ(oo) = 1, µ(-1) = 0, v(oo) = -1, v(1) = 0
and conclude that the remaining side transformations are A = vp and is = µv.
The counter clockwise boundary cycle, starting at oo, can now be written as
*1 *m n 1"1k m-1*n 1 *k-1
The edge cycles are n-1 m *1-1 and r n-1 n *k-1. The cycle relations are
vµ.1-'=t , µvtil=t
For later reference let us notice that the parabolic cycle *k 1 gives us the parabolic
V1/11- I vp=r6
It follows that the stabiliser [F,r],, has index 6 in r,,.
Let us conclude this section with a general formula for the genus of the
compact Riemann surface, IV.5.8, associated with a normal subgroup II of finite
index in the modular group F. The important invariants are the indices [Fx:IIx] of
the stabiliser IIx of the point x, where x = i, (, oo (( = exp(27ri/3)).
Genus formula 5.3 Let II denote a normal subgroup of finite index of

the modular group F. The horocyclic compactification X of H2/H is a Riemann
surface with Euler characteristic
X(X) = 2[r:II] - [r:II] E (1 - [rx:llx]-1)
x=i,(,eo
Proof Let us consider the triangulation `r of the space X induced by the

triangulation of H2 formed by the hyperbolic triangle A with vertices i, (, oo and
its translates by the extended modular group. The Euler characteristic is given by
X(X)=F-E+V
where the number of vertices in `r is V, the number of unoriented edges is E and
the number of unoriented faces is F. We have F = 2[F:II] and E = 2 where e is
the number of edges. For a vertex v of T let ev denote the number of edges with
an initial point at v to get
X(X)=2[r:II]- Ev(2 v-1)
The number e, can also be calculated as the number of triangles from `r with a
vertex at v. From the fact that H is a normal subgroup follows that F acts on X
and T. It also follows that we can separate the vertices of 'r into the three orbits
of v = [x], x = i, C, oo. This identifies the orbit of [x] and the set r/fl.Fx and
gives us the formula e, = 2(Fx:IIj. The standard exact sequence of groups
o - rx/IIx r/fi r/II.F, -, o
gives us the numerical formula
[r:II] = [r:II.rx][rx:flx]
Simple elimination between the three formulas completes the proof.

The genus g of the compact Riemann surface X, given by the formula

X = 2 - 2g , can be computed from the branch scheme
([rt,j ], [r(,nkl,
In particular we find that the group [r,r] has a branch scheme (2,3,6) and the
genus formula gives g = 1. Observe also that the branch scheme (2,3,6) gives
genus 1 independent of the index The group r2 has a branch scheme (2,1,2)
while the group r3 has a branch scheme (1,3,3).
Let us conclude this section with the introduction of a series of important
normal subgroups of the modular group r. Let us consider a positive integer n
and introduce F(n), modular rou of level n, by the exact sequence
5.4 0 r(n) PS12(7) PS12(7L/n) 0
We leave it to the reader to prove that the reduction map S12(7L)-+S12(7L/n) is

surjective. This is a non-trivial exercise in elementary arithmetic. It is also left to
the reader to show that the branch scheme of r(n) is (2,3,n).
For more information on this topic refer the reader to the books of
[Magnus] and [Newman] and the references given there. In particular [Newman]
p.143: apart from the three exceptions r, r2 and r3, every normal subgroup of r
is free.
VII.6 COMBINATORIAL TOPOLOGY
In this section we shall consider a discrete group of isometries I' of H2 for

which H2/1' is compact. From the Dirichlet construction V.4 it follows that r has
a compact fundamental polygon A, compare V.2.2. We shall introduce a simple
cut and paste technique to modify A in such a way that the topological type of
H2/F can be read off the boundary cycle of A.
In order to justify the cut and paste technique we must generalise the
concept of the fundamental polygon to allow for the sides of the polygon to be
composed of finitely many geodesic arcs. This generalisation does not destroy the
validity of Poincare's theorem as follows from its proof.
The edges of the polygon are paired by the symbol ai *a and we insist
on the condition *a # a-'. Let us enumerate one half of the edges (one edge from
each side of the polygon)
el,...,eel,*el,...*e,,,fl,...,fm
its understood that f. _ *f1, i = 1,...,m. The f1s are referred to as the unpaired
letters. We can write the boundary cycle as a word of these 2n + m symbols and
their inverses.
Let us assume that the boundary cycle of A has the form ApQB*p"1W
(see the illustration on the next page), where some of the words A,B,Q,W may be
empty. Let us introduce a curve p in the interior of A with the same initial point
as p and the same final point as Q. Let me stress that a curve is oriented, without
double points, and composed of finitely many hyperbolic arcs.
Let the side
transformation o'1 = oP 1 map the polygon pQp 1 into *pQ*p 1 and let us
construct a new fundamental polygon
A,= (A-PQp1) U *PQ*p1
The boundary cycle of A' is ApBQ*p1W. This justifies the first of the four rules
given below. The remaining rules are left to the reader to justify.
6.1
I ApQB*p'1W i--. ApBQ*p1W
6.2
[A p QB*pW H ApB*pQ-1W
6.3
IAQPB*p'1W ti ApB*p'1QW
6.4
AQpB*pW H ApBQ-1*pW
Theorem 6.5 The fundamental polygon A for r can be chosen in such a

way that the boundary cycle for A has the form
fI ldk*dk-1) A (albl*al'1*bl I)
fI (ci*ci) (fI ehUh*eh 1) V
k 1=1 i=1 h=1
where V and U1,...,U1. are words of the unpaired letters.
Proof Let us depart from the boundary cycle of any fundamental polygon and
apply the rules 6.1 through 6.4. Let us first observe that if the boundary cycle
contains p and *p, then we can use 6.2 with B = 0 to bring these symbols together
in such a way that the boundary cycle contains the syllable p*p. This syllable can
be moved freely around as follows from 6.4 and 6.2 with B = 0
Qp*p '-+ PQ-1*P '-+ P*PQ
Let us now assume that the boundary cycle doesn't contain pairs of the form p,*p.
Observe that a syllable of the form a*a 1 can be brought up front in virtue of 6.3
with B = 0
Qa*a 1 . a*a 1Q
Let us suppose that the boundary cycle contains the letters a,*a 1,b,*b-1. The
heart of the matter is the relative position of these four symbols. To get the
general idea we ask the reader to work out the two alternative reduction steps
aXbY*a1 aZYXb*a1*b-1Hab*a1*b"1ZYX
aXbY*b"1Z*a1F-+ aZXbY*b"1*a1 aZX*a"1 by*b"1
We trust the reader to complete the proof.
Remark 6.6 The standard form of the boundary cycle as given in 1.5 is
not unique. The following reductions can easily be deduced from our four rules
ab*a1*b"1c*c u-+ ab*a1c*b*c H abc*b*a*c u-+
ac1b*b*a*c i-+ a*c"1c1b*b*a I- *c"1c1b*ba*a

The number 2g + s is unique, being the epos of the surface H2/I'. Let us also
mention that the surface is orientable if s = 0 and unorientable if s > 0. The
number of non-empty words on the list U1,...,Ur,V equals the number of boundary
components of H2/17. These facts follow from Reidemeister's theory of surfaces: a
boundary cycle for the surface H2/1' is obtained by deleting the symbol * in the
boundary cycle 6.5 for A, see for example [Zieschang, Vogt, Coldewey] p.114.
Let us conclude this section with some standard notations, see [Griffiths].
Sphere a*a1b*1)1
Torus a *b-1 *a-1 b
Mobius band a *a c
Projective plane a *a b *b-1
Klein bottle a *a b* b , a b *a1 *b
The first presentation of the Klein bottle is the result of gluing two Mobius bands
together along their boundaries. The second presentation alludes to a cylinder
whose boundary circles have been given different orientations and glued together
accordingly.
VII.7 FRICKE-KLEIN SIGNATURES
In this section we consider a Fuchsian group r of H2 with compact orbit

space. We try to gather enough geometric information about r to be able to
reconstruct r as an abstract group. The signature itself is a collection of geometric
data sufficient to provide a reconstruction of the group by generators and relations.
As a general principle, two discrete groups should have the same signature if and
only if they are isomorphic as abstract groups.
The basic part of the signature is the genus g of the compact Riemann
surface H2/r given through the Euler characteristic
7.1 x(H2/r)=2-22g
Moreover, we let nl,...,nr, denote the order of the conjugacy classes of elliptic
elements of r. The fact that the number of conjugacy classes is finite follows from
the presence of a compact Dirichlet polygon. The signature of f is
7.2 sign r = ( g; n1,...,ni,)
Thus the signature is collection of integers with g > 0 and n1 > 2 , i = 1,...,r. In
order to investigate which data can occur as the signature we consider a Dirichlet
polygon A for r. We shall show that
i=1
7.3 -
Area(O) = 2a[ x(H2/r) + E (1-
Proof Let the polygon A have v vertices. By the definition of area, 111.9.9
Area(s) = 7r (v - 2) - E p L;ncP
where the sum is over all vertices P of A. Let us recall from VII.3.5 that the
vertex cycles of a Fuchsian group are without repetitions. This allows us to
partition the sum above into vertex cycles and use V.3.15 to rewrite the formula
VII.7 FRJCKE-KLEIN SIGNATURES 231
Area(A) = 7r(v - 2) - r 21r

=1 '
The polygon A modulo identification of sides defines a complex on H2/F with

x(H2/F)=V-E+F
where V = r , E = 2v and F = 1. Elimination of v and r gives us
Area(A)=27r(E-1)-V+27rE(1-n)
i=1 '
which is the required result, once its recalled that F = 1. 0
As an interesting application of the area formula 7.3, we find a lower

bound for the area of a Dirichlet polygon A for a Fuchsian group
7.4 Area(A) > 21
Proof According to 7.3, it amounts to the same to prove the estimate
r
2g-2+n.)?42
Let us first remark that this is clear for g > 1. For g = 1 we must have r > 1 and
the result follows since (1 -
j ) > 2. When g = 0 we have r > 3 since the area is
positive. When r > 5, the estimate follows by observing that each term in the sum
is at least 2. When r = 4, the minimum is taken for signature (0;2,2,2,3), but
0-2+2+2+2+1-3-G
It remains to treat the case g = 0 and r = 3. We must prove the inequality
f(k,m,n) = I 42 ;2<k<m<n
for all strictly positive values of f(k,m,n). Observe that f(3,3,4) = i2 and conclude
that we only have to worry about the case k = 2. Observe that
f(2,2,n) < 0 , f(2,4,4) = 0 , f(2,4,5) = 120
and conclude that we only have to worry about k = 2 and m = 3. We have
f(2,3,n) = s -
In the end, the result follows from the evaluation f(2,3,7) _ 2.
Theorem 7.5 The signature (g ; n1) ...,n1.) of a Fuchsian group F satisfies

r
2g-2+ E(1 -.) > 0
i=1
Conversely, any sequence of integers (g; n1,...,nr) with g > 0 and ni > 2, i = 1,...,r,
which satisfies this inequality can be realised as the signature of a Fuchsian group.
Proof The first part of the theorem follows from 7.3. Conversely, for any
data as above, we shall construct a suitable polygon and apply Poincare's theorem.
We fix a point 0 of H2 and consider a circle with centre 0 and radius r, say. Let
us introduce the angle
0- 8g+2i.
21r
and a sequence of vertices Al,...,Ag,B1,.... B1. (B1 = Ag+11 Br+1 = A1) on the circle
subject to the conditions
LA;OAi+1 = 80 , LBiOBi+1 = 20 ; i = 1,...,g, j = 1,...,r
For the remaining specifications we refer to the drawing
VII.7 FRICKE-KLEIN SIGNATURES 233
The boundary cycle for the polygon is given by
"1*b1"1... agbg*ag 1*bg 1cl*c1"1... cr*cr 1
The edge cycles are

Cl-1*C11 C2 1*C2 , ... , Cr. 1*C7.
a1*b1 1*a1"1b1... ag*bg1*aglbeclc2... Cr
The sum of the inner angles for the cycle of initial vertices is
r
f(r) = 8ga + 2 /3i
If f(r) = 2r, then Poincare's theorem assures us that the side transformations
generate a Fuchsian group with the given signature. Let vary r through the
interval ]O,oo[ to see that this can be achieved. The angles a and (3j, j = 1,...,r can
be analysed through the "alternative cosine formula" 111.5.2
cos X31 cos 0 + cos r/ni

cosh r =
5777 /3j S177.
cosh r = cot a cot 0
From this it follows rather immediately that

lin7r-oo a = 0 , li7n7.-,oo /31 = 0 ; j = 1,..,r
which gives limr_,._,,,f(r) = 0. In order to investigate the limit for r-+0 let us
rewrite the two cosine formulas as follows
cosh(© +,6j) = cos(r - r/m3) + sin 0 sin ,61 (cosh r - 1)
cosh a = cosh r cot(lr/2 - 0)
It is now rather obvious that
Ii7nr-O a = 2 - 0 , limr-+0 ,(3i = r - n,
We conclude from the formula for f(r) that
limr-O f(r) = 2r[ 2g - 2 - (1 - tom.)] + 2r > 21r
This concludes the construction of our polygon. 0
We shall now complete our investigations by showing that the abstract

group structure of our Fuchsian group can be recovered from the signature.
Theorem 7.6 Let I' be a Fuchsian group with H2/I' a compact surface of
genus g. If n,,...,nr denotes the orders of the conjugacy classes of elliptic
subgroups of r, then I' admits a presentation with generators
0(l,...,a91/31,...,0g161,...,6r
and relations ( [a 6] = a0a-1,0-1)
b1...br11
rf [ai,,3i] = t , bn1 = 1 ,..., bn' = c
i=1
Proof Let us use 6.5 to chooses fundamental domain with boundary cycle
11 (dk*dk 1) fI (ai*bi 1*a; 1 bi)
k=2 i=1
The edge cycles are
g
d2....d,. fl (aibi*a; 1*b;'1) , d2 1,..., dr 1
Let us use Greek letters for the corresponding side transformations to get the
presentation of I' with a complete set of relations
g
[ 13 ])111
112_ nr-
1=1
Upon introducing the auxiliary transformation bl given by
g
bl = 62...b,, n [-i,/3 ]
i=1
we can bring the presentation of 1' into the announced form 0
Analogous results can be derived by the same method for the general
group I' with H2/1' compact, see [Wilkie] and [Macbeath]. For a different
approach see [Zieschang, Vogt, Caldeway].
VII EXERCISES 235
VII EXERCISES
EXERCISE 2.1 Let p and q be ordinary integers which satisfy the relation
(p-2)(q-2) > 4
1° Show that there exists a regular p-gon 0 with angles 2ir/q.
2° Show that there exists a tesselation `JI, q of H2 consisting of regular p-gons
with angles of measure 2ir/q, such that any two of the p-gons have empty
intersection or have an edge in common. Hint: Consider the triangle group
G(2,p,q) and its canonical tesselation.
3° Let p denote a rotation of order p about the centre of 0 and let o denote a
rotation of order q with respect to a vertex of A. Show, that the group of even
symmetries of the tesselation Tp q is generated by p and o, and that
(PT)2
Pp = oq = = t
is a complete set of relations for the group of even symmetries of Tp q.

4° Let us consider integers p and q such that
(p-2)(q-2) > 4
Show that the universal group generated by R and S subject to the relations
RP = Sq=(RS)2=t
is infinite. Hint: When (p - 2)(q - 2) = 4 use Euclidean geometry.
EXERCISE 2.2 Let 0 denote a regular p-gon with angles of measure 2a/q with
the notation of the previous exercise.
1° Show that the group D(0) generated by reflections in the sides of 0 is
discrete if and only if q is an even integer.
2° Suppose q is even. Show that the group D(0) is a normal subgroup of the
symmetry group of `fp q, and identify D(0) with a normal subgroup of G(p,q,2) of
index 2p.
EXERCISE 2.3 Let us consider integers p and q such that

(p - 2)(q - 2) > 4
Show that the universal group generated by R and S subject to the relations
RP=Sq =(RS)2=c
is infinite. Hint: When (p - 2)(q - 2) > 4 use hyperbolic geometry, but when
(p - 2)(q - 2) use Euclidean geometry.
EXERCISE 2.4 Let A be an r-gon in the hyperbolic plane whose vertices has
angles of measures
LPi = 7r/n; ; n, E N, n, > 2, i=1,...,r
10 Show that reflections in the sides of A generate a discrete group D(0)
with A as fundamental domain.
2° Show that "Dyck's group" D+(0) can be generated by pl,...,pr subject to
nl = n1,
pl ... =P =P1P2...p1.=c
30 Show that the genus of the Riemann surface H2/D+(0) is zero.
EXERCISE 2.5 Let n > 2 be an integer and let A be a regular 2n-gon with
angles of measure a/n. Let us consider a side pairing on A with boundary cycle
al*al a2 *a2 ... an *an
For i = 1,...,n we let Mi resp. *M1 denote the midpoint of the edge a; resp. *Mi.
Show that the side transformation ai is glide transformation along M;*M4.
2° Show that al,...,an generate a discrete group G with fundamental domain
A and that a complete set of relations is
anon-1
2 2
... al2 =t
3° Identify the group G with a subgroup of index 4n in the triangle group
G(2,2n,2n).
EXERCISE 2.6 Let g > 2 be an integer and let A be a regular 4g-gon with
angles of measure 7r/2g. Let us consider the side pairing on .A which identifies
each side with the opposite side of the polygon by means of parallel translation
along the midpoints of the two sides. Let us enumerate the sides of A such that
the boundary cycle reads
al bl a2 b2 ... ag bg *al-1*bl... *ag 1*bg 1
1° Show that a discrete group G with fundamental
domain A and that a complete set of relations is
VII EXERCISES 237
1 _1 _1 _1
l(3» = Q» a» ... 01-1-1
2p2 ... an/3n_1:_
x101 a2)32-
2° Show that the Riemann surface H2/G has genus g.
3° Identify the group G with a subgroup of index 8g in the triangle group
G(2,4g,4g).
EXERCISE 2.7 Let 0 be the regular 8-gon with angles of measure it/4.
Consider the side pairing with boundary cycle
a b *a 1*c 1*d-1 c *b"1d
Show that the corresponding Riemann surface has genus 2 and a fundamental
group generated by a, 13, y, 6 subject to the relation
6 -Y60a-' 0-1 y1 a=t
EXERCISE 5.1 Show that the commutator subgroup [r,r] of the modular
group r has the fundamental domain A U r/, U r20 U r30 U r40 U r50 with side
transformations r6, r3o, r4or 1, r5or 2. Work out the edge cycles and apply
Poincare's theorem.
EXERCISE 5.2 1° Show that the subgroup r2 of index 2 in the modular group
1' has a branch scheme (2,1,2), and that this is the only normal subgroup of I'
with a branch scheme of the form (2,1,n).
2° Show that the normal subgroup r3 of r of index 3 has a branch scheme
(1,3,3) and that this is the only normal subgroup of finite index with a branch
scheme of the form (1,3,n).
30 Show that no normal subgroup of finite index of r has a branch scheme of
the form (1,1,n), except for 1' itself.
EXERCISE 5.3 Let i denote the hyperbolic triangle oo,0, l with side trans-
formations o, p-1 ap, pop 1 where p = ro. Show that Poincare's theorem applies
and deduce that r3 is the free product of the three cyclic subgroups generated by
0',p 1l7P, P7p1
EXERCISE 5.4 I 'Show that 1.3 is the smallest normal subgroup of r containing p.
2° Show that the subgroup I'2 of I' of index 2 is the smallest normal subgroup
of IF containing o.
3° Show that 1'2 and 1'3 are the only non-trivial normal subgroups of r which
contains elements of finite orders.
EXERCISE 5.5 Let 11 denote a normal subgroup of finite index p of I' with a
branch scheme of the the form (2,3,n).
10 Show that the corresponding compact Riemann surface has genus 0 for the
following values of (µ,n) only
(6,2) (12,3) (24,4) (60,5)
2° Show that the table above is realised by 1'(2), I'(3), r(4), 1'(5)
EXERCISE 5.6 Verify the following calculation of genus of F(n)

n 1 2 3 4 5 6 7 8 9 10 11 12
genus 0 0 0 0 0 1 3 5 10 13 26 25
VIII HYPERBOLIC 3-SPACE
In the first two sections of this chapter we return to the generality of

chapter I and study the exterior algebra over an oriented n-dimensional real
vector space E equipped with a non-singular quadratic form. The main issue is
parametrisation of subspaces of E of given Sylvester type by exterior vectors.
This construction is particularly satisfactory where our space is an oriented
Minkowski space M, i.e. is a four-dimensional space of Sylvester type (-3,1). In
this case the space A 2M has a natural complex structure and a complex inner
product familiar from special relativity. We find that a 2-vector z E A 2M is
decomposable (primitive) if and only if Im[<z,z>] = 0. A decomposable 2-vector
z # 0 determines a plane in M whose Sylvester type is determined by the sign of
the norm <z,z>. In particular we find a one to one correspondence between
oriented geodesics in H3 and 2-vectors z E A 2M with <z,z> 1. Our general
discussion of the geometry of H3 is based on this fact.
So far, our discussion has been based on a general Minkowski space. In
section VIII.6 we introduce a specific Minkowski space, namely the space M of
2 x 2 Hermitian matrices. This space is, in a natural way, a linear representation
of the group S12(C) and we deduce an isomorphism
PS12(C) Lor+(M)
where Lor+(M) is the group of special Lorentz transformations of M. We make a
definite choice of a sheet of the hyperbola in M and call this the Hermitian model
of H3, compare [Shafarevich p.149] and [Mania].
Beyond its own intrinsic interest, the Hermitian model has a trace formula
not available in general: Let S E S12(C) act on H3 as the product of half-turns a
and 0 in two geodesics h. and k with normal vectors H and K. Then
lr2(S) = 4 <H,K>2
This leads to a classification theorem: the conjugacy classes in the special Lorentz
group are in one to one correspondence with classes of pairs of geodesics in H3.
The Hernmitian model has its origin in physics. We shall make use of
another tool with its origin in physics, the Dira.c algebra. For readers familiar
with the formalism of Clifford algebra we can add that Dirac's algebra is a
240 HYPERBOLIC 3-SPACE
concrete realisation of the Clifford algebra of M.

The trace formula above makes it clear that geodesics have an important
role to play in the study of even isometries of H3. Geodesics occur in bundles as
we shall see in section VIII.5. In the final section of the book we shall illustrate
how the geometric notion of bundles of geodesics in H3 applies to fixed point
theorems.
VIII.1 EXTERIOR ALGEBRA
In this section we consider a vector space E of dimension n over R

equipped with a non-singular quadratic form. We shall be concerned with the
extra structure induced by the form on the exterior algebra A E. For p E N we can
introduce a symmetric bilinear form on APE by the formula
1.1 <u1 A ... A up,vl A ... A VP> = det1j<u,,vj>
A vector e E E gives rise to an operator Le on A E given by
1.2 u1 A ... A up Le = E .(-1)i-1 <u1,e> u1 A ... A uj... Au P
It is easily verified that our operation satisfies the general rule
1.3 (u A v)Le = (uLe) A v+(-1)ru A Me) ;u E A rE,v E ASE, e E E
Observe that the operator La on A E has square 0. It follows that we can extend
the operator L to all of A E making A E into a r ht module over the ring A E
1.4 u L(v A w) = (u Iv) Lw ; u,v,w E A E

VIII.! EXTERIOR ALGEBRA 241
It is now time to introduce the key formula
1.5 u L v = <u,v> ; u,v E APE
Proof Let us proceed by induction on p Consider two decomposable vectors

u = ul A ... A uP and v = vl A ... A vp. Using formulas 1.2 and 1.3 we get that
u L v = (E .(-1)i+l<ui,vl> ul A ... A iii... A up) L V2 A ... A vp =

(-1)1+1<ui,vl><u1 A ... A ui... A up, v2 A ... A vp> _ <u,v>
where we have used expansion of the determinant after the first row. 0
Let us combine the two previous formulas to get that
1.6 <u,v A w> = <u Lv, w> ; u E A P+qE, v E APE, w E n qE
Orientations Let us recall the algebra underlying the concept of orientation

of our n-dimensional real vector space E. By a frame in E we understand an
ordered basis el,...,en for E. The frames make up a subset Fr(E) of En on which
the group G1(E) acts transitively. The group GC+(E) of transformations with
positive determinant acts with two orbits on Fr(E), the orientations of E.
In the rest of this section we shall assume that an orientation of E has
been singled out. A frame el,...,en is said to be positively oriented if it represents
the preferred orientation. In particular, if we depart from a positively oriented
orthonormal basis el,...,en for E, the orientation vector or = el A ... A en is
independent of the positively oriented orthonormal basis considered. It is now
time to introduce the Node star operator * in the exterior algebra of the oriented
space E of Sylvester type (-s,t). To a p-vector v we shall associate the (n-p)-
vector *v given by
1.7 *v=(-1)5or[v ; vEAPE
By definition *1 = (-1)sor. Simple evaluation gives us
1.8 *or= 1
The * operator satisfies the formula
1.9 ** V = (-l)P(I1-P)+sv
; vE APE
Proof Let us observe that APE is spanned by p-vectors of the form

el h ... A ep where e1,...,ep are part of a positively oriented orthonormal basis
el,...,ep,...,e. for E. Simple direct calculation yields
1.10 (-l)'* el h ... h ep = 7P

11 <ei,ei> ep+1 A... A en
i=1
Let us remark that the operator * changes sign if the orientation is changed; the
orientation must be changed (n - p)p times to make eP+j.... jen,el,...ep positively
oriented. Thus we get from 1.7 that
(-1)n(n-p)
It
(-1)s* eP+1 A ... hen = fi <ei,ei> el h ... A ep
i=P+1
n
Apply 1.7 once more and observe that r[ <ei,ei> = (-1)5. 11
i=1
Consider an (n-p)-vector w and a p-vector u and let us write w h u = A or

with A E R. The constant A can be determined by formula 1.6 :
<or, w h u> = <orlw, u> = (-1)s<*w,u>
Taking into account that <or,or> = (-1)5 we get that A = <*w,u>. Thus
w h u = <*w,u> or
Let us apply the Hodge operator and use 1.8 to get that
VIII.1 EXTERIOR ALGEBRA 243
*(w A u) = <*w,u> ; w E An-PE, u E APE
In particular take w = *v , v E ARE, and use 1.9 to get that
1.12 <u,v> = (-1)5*(u A *v) ; u,v E APE
A combination of 1.11 and 1.12 gives that
1.13 <u,v> = (-l)5<*u,*v> ; u,v E APE
Complex structure Let M be an oriented Minkowski space, i.e. a space

of Sylvester type (-3,1). The operator * on A 2M satisfies *2 = -t and is self
adjoint as follows from 1.13
<u,*v> = <*u,v> ; u,v E A 2M
This allows us to introduce a complex structure on A 2E through the formula
1.14 (x+iy)v=xv+y*v ;x,yER, vE A2M
and a symmetrical C-valued bilinear form
1.15 <u,v>c = <u,v> - i <*u,v> ; u,v E A 2M
The subscript c will be omitted when no confusion is possible. Observe that

<iu,v>c = <*u,v>c = <*u,v> + i<u,v> = i(<u,v> - i<*u,v>) = i<u,v>c
In conclusion we have given A 2E the structure of a three-dimensional vector
space over C with a complex quadratic form. Let us look at this in terms of a
positively oriented orthonormal basis aO,o1,a2,o3 for our Minkowski space with
<0'0,UO> = 1 , <ai,oi> i = 1,2,3
We find by explicit calculations that
1.16 *(0-1 A a2) = a3 A ao , *(a2 A a3) = al A aO , *(o,3 A al) = a2 A aO

It is now straightforward to check that of A o-2, a2 A o3, 0.3 A of form a C-linear

basis for A 2E; in fact an orthonormal basis for the complex inner product.
Volume form For an oriented Minkowski space M as above we can

introduce the volume form
1.17 vol(a,b,c,d) = *(a A b A c A d) ; a,b,c,d E M
This is an alternating 4-form which takes the value 1 on a positively oriented

orthonormal basis as follows from 1.8. Conversely, an alternating 4-form which
takes the value 1 on some positively oriented orthonormal basis for M must be the
orientation form. We ask the reader to use 1.12 to show that
1.18 <a A b,c A d>c = <a,c><b,d> - <a,d><b,c> - i vol(a,b,c,d)
for any four vectors a,b,c,d E M.
The reader is referred to [Bourbaki] for the exterior algebra. Note that we
have used the sign conventions which makes A E into a right module over itself, as
opposed to later editions of Bourbaki.
VIII.2 GRASSMANN RELATIONS
Let us continue our study of an oriented vector space E of dimension n

over R equipped with a non-singular quadratic form. We shall try to decide when
a given p-vector v is nure, i.e. of the form v = e1 A e2... A ep, where el,...,ep are
linearly independent in E.
Quite generally, let us associate with a non-zero v E APE the smallest
subspace Ev of E such that v E A PE, Note that
2.1 dimEv > p ;vE A E,v#0
and that equality holds if and only if v is a pure p-vector.
Proposition 2.2 The subspace Ev of E associated with a p-vector v is

Ev = Ir(vL) , vL: A p-1E E ; x --* vLx
The subspace of E orthogonal to Ev is given by
Ev 1 = Ker(v() , vL:E A p-1E e H vle
Proof We shall take our point of departure in the identity

<vLx,Y> = 4LxLY = (-1)''-1vLYLx = (-1)p-1<x,vly> ; x E AP-1E, y E E
Using that the inner product on A p-1E is non-singular we get that
Im(vL) 1 = Ker(vL)
The inclusion Ev 1 C Ker(v[) follows by direct verification using 1.2. It remains
to prove Ev C Ker(vL) 1 or what amounts to the same thing, Ev C e 1 for all
e E Ker(vL), e 0. Choose a basis e1,...,e for E where e2,...,en E e l and write
v= EH aHeH H: [1,p]-'[1,n]
where H runs through all strictly increasing maps in the range indicated above and
eH = eH(1) A ... A eH(p). The relation vLe = 0 can be written
0= H aH <eH(1),e> eH(2) A ... A eH(p)
which gives aH = 0 whenever 11(1) = 1. Thus Ev C e 1 as required.
Proposition 2.3 If the vector w E APE is pure, then the vector

*w E A 'PE is pure and E*,, = EN, 1 .
Proof From 1.3 it follows that for u E A rE and v E ASE we have the formula
2.4 (u A v)Le1 A ... A ep = uL(e1 A ... A ep) A v ; ej E E, vLe1=0, i=1,...,p
Let us consider a pure vector w = el A ... A ep and let E, denote the space spanned
by el,...,ep. Pick a basis f1,...,fn for E with fp+l,...,fn E Eµ, 1 and apply 2.5 to get
(f1 A ... A fn)Le1 A... A ep = <f1 A ... A fp,el A ... A ep>fp+l A ... A fn
Observe that we can arrange for or = f1 A ... A fn and the result follows.
Corollary 2.5 Any non-zero vector w E An-1E is pure.
Theorem 2.6 A non-zero z E APE is pure if and only if the composite map
AP-1E zL E _zA Ap+1E
is zero. When z is pure, the sequence is exact.
Proof Let us write down the general formula
2.7 *(z A e) = or L(z A e) = (or Lz)Le = *zle ;eEE
to see that z A e = 0 if and only if e c Ker(*zL). We conclude from 2.2 that the
image of the first map is Mz while the kernel of the second map is M*Z 1 . Thus
(z A ) o (zL) = 0 a Ez is orthogonal to E*Z
If z is pure, we have seen in 2.3 that EZ is orthogonal to E*z. Conversely, if EZ is
orthogonal to E*,L we conclude from 1.1.7 that
dim Ez + dim E*Z < n
Since in general dim EZ > p and dim. E*z > n - p we get dim EZ = p. This means
that z is a pure p-vector.
Let us now specify our theory to a Minkowski space M. Our main concern
will be the classification of two-dimensional subspaces of M by 2-vectors.
Theorem 2.8 A non-zero 2-vector w E A 2M is pure if and only if

<w,w>c E R. A pure 2-vector w gives rise to an exact sequence
M
awl M -+*wl M
For a pure 2-vector w the plane Mw = Ker(w A: M-> A 3M) has Sylvester type
(-1,1), (-1,0),(-2,0) depending on whether <w,w> is negative, zero or positive.
Proof The composite of the two maps we have just written is zero if and only
if w is pure as follows from 2.6 and 2.7. The first statement will follow once we
have established the formula
2.9 *wl(wle) <*w,w> e = Im[<w,w>j e ;eEM

2 2
We shall be content with a particular case, namely, with the notation of 1.16
w_(x+iy)Q1Av2=xo'1Aa2+Ya3AQo, *w=iw= -yo1Aa2+xc3Aco
observe that Im[<w,w>,] = Im[(x + iy)2] = 2xy. Finally check through the
formula as e varies through ao,al,a2,a3. An alternative proof can be given using
the technique from VIII.7.
Let us investigate the general subspace E of M of dimension 2. Pick an
orthonormal basis e,f for E and calculate the norm of the 2-vector e A f
<e A f,e A f> = <e,e><f,f> - <e,f>2
The discriminant lemma 1.6.3 shows that the norm is negative, zero, positive
depending on the Sylvester type as stated. 0
Let us end this section by the observation that an operator on Minkowski

space M of the form zl, z E A 2M, is skew adioint
<zle,f> = zlelf = - zlfle = - <e,zlf>
248 HYPERBOLIC SPACE
VIII.3 NORMAL VECTORS
Let H3 be one of the sheets of the unit hyperbola in Minkowski space M.

We shall study the geometry of H3 by assigning to a geodesic in H3 a normal
vector N E A 2M. The relative position of two geodesics will be studied through
the complex inner product of their normal vectors.
Proposition 3.1 The points of a geodesic curve y:R-+H3 span a plane in M

of Sylvester type (-1,1). The normal vector
y(t)Ay'(t)E A2M ;tEIR
has norm -1 and is independent of t E R. This establishes a one to one cor-
respondence between oriented geodesics in H3 and 2-vectors in A 2M of norm -1.
Proof A geodesic y:R,H3 can be written in standard form

y(t) = A cosh t+ T sink t ; tER
where A and T are vectors from M of norm 1 and -1. Let us differentiate this to
get yl(t) = A sink t + T cosh t. A simple calculation gives
y(t) A y'(t) = A A T (cosh2t - sinh2t) = A A T
It follows from 3.6 that the norm of N = A A T is -1.
Conversely, let N E A 2M be a vector of norm -1. It follows from 2.8 that
we can find vectors A and T in M with A A T = N. The discriminant
-1 = <N,N> = <A A T,A A T> = <A,A><T,T> - <A,T>2
shows us that A and T span a plane of type (-1,1). We may assume that A and
T are orthogonal with A E H3 and <T,T> = -1. We can use the standard
formula above to write down a geodesic curve with normal vector N = A A T.
By a hyperbolic la aU we understand a subset of H3 traced by a linear

subspace of M of Sylvester type (-2,1). Let us observe that the linear span of a
hyperbolic plane in H3 is a linear subspace of M of type (-2,1)
VIH.3 NORMAL VECTORS 249
Proposition 3.2 Two geodesics h and k in H3 with normal vectors H and

K are coplanar, i.e. contained in a hyperbolic plane if and only if <H,K>c E R.
Proof Pick points A E h and B E k and unit tangent vectors S at A and T at B

such that H=A A S and K=BAT. We get from 1.18
Im[<H,K>c] = - vo1(A,S,B,T)
Thus <H,K>c E R if and only if A,S,B,T are linearly dependent. Let us observe
that a three-dimensional subspace E of M which meets H3 has type (-2,1), since
the two alternative types (-2,0) and (-1,0) are excluded on the grounds that they
do not contain vectors of norm 1, compare 1.6.3. In conclusion, if <H,K>c E R,
then the linear hyperplane E spanned by h. and k cuts H3 in a hyperbolic plane
which contains h and k.
Let us say that two geodesics h and k are perpendicular if they meet in a
point A and their tangent vectors at A are orthogonal.
Corollary 3.3 Two geodesics h and k are perpendicular if and only if

their normal vectors H and K are orthogonal <H,K>c = 0.
Proof If <H,K>c = 0 then h and k are coplanar by 3.2. Observe also that if h
and k are perpendicular then they are contained in a hyperbolic plane. Thus the
result follows from 111.2.
Theorem 3.4 The action of the special Lorentz group on A 2M defines an

isomorphism of groups
Lor+(M) - S03(C)
Proof The group S03(C) = SO( A 2M) acts transitively on orthonormal pairs
H,K in A 2M, while the group Lor+(M) acts transitively on pairs h,k of oriented
perpendicular geodesics.
Let OH3 denote the set of isotropic lines in M (on ints at infinity). We say
that P E OH3 is an end for a geodesic h in 113 if the line S is contained in the linear
span of h. It is clear that a geodesic h has two ends, and conversely, given two
distinct points A and B of 8H3, there exists a unique geodesic in H3 with ends A
and B. Observe also that two geodesics h and k with a common end are coplanar.
Theorem 3.5 Let h and k be two geodesics in H3. There exists a geode-
sic g perpendicular to h and k if and only if h and k don't have a common end.
Proof Let H and K be normal vectors for h and k respectively. The vectors H
and K are linearly independent over C: from H = AK, A E C, we get, taking
norms, that A2 = 1 and therefore H = K or H = -K, which means that H and K
represent the same unoriented geodesic. Let us introduce the complex vector space
E C A 2M spanned by H and K and let us form the discriminant
A = <H,H>c<K,K>c - <H,K>c2 = 1 - <H,K>c2
The complex line E 1 is isotropic if and only if A = 0.
If A # 0, we can find a 2-vector G with norm -1 spanning E 1 . The geo-
desic g with normal vector G is perpendicular to both h and k as follows from 3.3.
Conversely, if the geodesic g is perpendicular to h and k, then the normal vector G
for g spans the complex line E 1 and it follows that A # 0
If A = 0, then we have <H,K>,, = ± 1, and we conclude that h and k
coplanar. Thus the result follows from the discussion in 111.2.
VIII.4 PENCILS OF PLANES 251
VIII.4 PENCILS OF PLANES
Hyperbolic planes occur in certain families called pencils. In this section

we shall find all such pencils and demonstrate their significance for simple
geometric problems.
Let us recall that a hyperbolic plane P in H3 is the trace on H3 of a linear
hyperplane E of M of Sylvester type (-2,1): P = E n H3. Let there be given an
orientation of the supporting hyperplane E. A positively oriented orthonormal
basis K,L,M for E defines the normal vector for P
4.1 S=KALAME A3M
The hyperplane E can be recovered from S as the kernel of S A : M - A 4M.

Alternatively, the space E is the kernel of the composite map
M -sn A4M -* A°M
By 2.7 this map is (*SL): M R , so we find that E = (*S) 1 . In conclusion
4.2 P=H3 n (*S)1
Proposition 4.3 The normal vector S E A 3M of an oriented hyperbolic

plane P in H3 has norm 1. This construction establishes a one to one
correspondence between 3-vectors S E A 3M of norm 1 and oriented hyperbolic
planes P in II3.
Proof With the notation of 4.1, pick a positively oriented orthonormal basis
el,e2,e3 with norms -1,-1,+1. This gives
<S,S> = det1j<e1,ei> = 1
The main body of the proposition follows from VIII.2. 0
We shall investigate a number of incidence relations in terms of normal vectors.

Proposition 4.4 Let P be a hyperbolic plane with normal vector S E A 3M.

A geodesic k with normal vector K E A 2M is contained in P if and only if
(*K) A (*S) = 0
Proof Let us observe that MK is contained in P if and only if MK is

orthogonal to *S as follows from 4.2. This means *S E MK 1 or *S E M*K as
follows from 2.3. But *S E M*K means *S A *K = 0 by 2.8. 0
Definition 4.5 A geodesic n is said to be perpendicular to a hyperbolic

plane P in H3 if n intersects P in a point A and all geodesics h in P through A are
perpendicular to n..
Proposition 4.6 A geodesic it with normal vector N E A 2M is perpendicu-

lar to the hyperbolic plane P with normal vector S E A M3 if and only if
NA*S=0.
Proof Suppose that the geodesic n is perpendicular to P. Let us observe that a

unit tangent vector T to it at the point A of intersection between P and It is
orthogonal to any unit tangent S to P at A. Observe that <T,A> = 0 to conclude
that T is orthogonal to the linear span E of P. But the vector *S is orthogonal to
E too as follows from 4.2. Let us recall from 1.13 that
4.7 <*S,*S> = - <S,S> ; S E A3M
It follows that *S has norm -1. Thus T = ± S and

NA*S=AATA*S=0
Conversely, let us assume that the normal vector N to it satisfies N A *S = 0.
According to 2.8 this means that *S belongs to the linear span of n. From this we
can conclude the existence of a point A E n. with <A,*S> = 0. The point A is
common to n and P and the vector *S is tangent to it at A. It follows from 3.3
that the geodesic n is perpendicular to P. 0
VHI.4 PENCILS OF PLANES 253
Proposition 4.8 Let P and Q be two distinct hyperbolic planes with

normal vectors S and T. The plane MI< associated with the pure 2-vector
K = *(*S A *T)
is the intersection of the hyperplanes E and F in M spanned by P and Q.
Proof From the assumption that P $ Q it follows that *S and *T are linearly
independent which implies that *S A *T # 0. Observe that *K A *S=0 to conclude
that *S belongs to M.,<, compare 2.8. . Since M*I< = MI< 1 by 2.3, we conclude
that *S is orthogonal to MK or MK C (*S) 1 . Similarly, we find that
MI< C (*T) 1 . For dimension reasons we must have that MK = (*S) 1 n (*T) 1 .lJ
By a pencil P of hyperbolic planes in H3 we understand a two-dimensional

subspace of n 3M. The actual hyperbolic planes in the pencil are those hyperbolic
planes with normal vectors in 9. The Sylvester type of A 3M is (-1,3) as follows
from 4.7; thus the possible pencil types are (0,2), (0,1), (-1,1).
Let us consider a fixed pencil `P and pick two distinct hyperbolic planes P
and Q in the pencil with normal vectors S and T.
4.9 <S,T> Sylvester type Intersection P n Q #(aP n aQ)

I<S,T>I > 1 (-1,1) 0, common perp.l. 0
<S,T> I < 1 (0,2) geodesic 2
I<S,T>l = 1 (0,1) 0, common end 1
Proof With the notation of 4.8, let us calculate the norm of K

<K,K> = - <*K,*K> = - (<*S,*S><*T,*T> - <*S,*T>2) = <S,T>2 -1
If <K,K> > 0 we find that the discriminant A of 9 is negative and that
E n F = Mx has type (0,2). The plane G = 1 has type (1,-i) and defines a
geodesic perpendicular to P and Q.
If <K,K> < 0, we have A > 0 and the plane E n F has type (-1,1). It
follows that P fl Q is a geodesic.

If < K,K> = 0, we have A = 0, and the plane E fl F has type (-1,0). It
follows that P and Q have precisely one common end. 0
Corollary 4.10 Let n be a geodesic in H3. Through any point A E H3

there passes a unique geodesic perpendicular to n. There is a unique hyperbolic
plane perpendicular to n with a given end Q E 8H3.
Proof Let us introduce the following subset of A3M

A={SEA3MI<A,*S>=0}
This is a three-dimensional subspace of A 3M of type (0,3). The pencil X of
hyperbolic planes perpendicular to n has type (-1,1), and it follows that NIA
has dimension 1, spanned by a vector of norm 1. Introduce the set
0 ={SEA3MI<Q,*S>=0}
This is a three-dimensional subspace of A 3M of type (0,2) and it follows that
N n Q has dimension 1, spanned by a vector of norm 1.
Proposition 4.11 Let 9 be a pencil of hyperbolic planes. Through each

point A of II3 there passes at least one plane from the pencil. For pencils of types
(-1,1) or (0,1) there will pass precisely one plane from the pencil through A.
Proof Let us consider a pencil 9 of type (0,1). The condition for the
hyperbolic plane P with normal vector S to pass through the point A is
<*S,A> = 0. Thus we are lead to consider the intersection A 1 n *9. The plane
*9 in M has type (- 1,0) while A 1 has type ( - 3,0). It follows that the
intersection is one-dimensional, generated by a vector S of norm - I. We leave it
to the reader to treat the remaining two types of pencil.
VIII.5 BUNDLES OF GEODESICS 255
VIII.5 BUNDLES OF GEODESICS
By a en ncil 9 of geodesics in H3 we understand a two-dimensional linear

subspace of A 2M consisting of 2-vectors with real norm and spanned by 2-vectors
of norm -1. The actual geodesics in the pencil are those geodesics whose normal
vectors lie in 9.
Proposition 5.1 The geodesics in a pencil 9 are contained in a hyperbolic

plane. Conversely, the normal vectors of two geodesics span a pencil if and only if
the two geodesics are coplanar, i.e. is contained in a hyperbolic plane.
Proof Let us pick two distinct geodesics h and k from the pencil and let H and
K be normal vectors for these. Observe that the formula for polarisation 1.1.2
shows that <H,K> E R since 11, K and H + K all have real norms; it follows from
3.2 that h and k are contained in the same linear hyperplane N of M of type
(-2,1). It follows that 9 C A 2N.
Definition 5.2 By a bundle `9 of geodesics in H3 we understand a three-

dimensional linear subspace of A 2M consisting of 2-vectors with real norm, and
spanned by 2-vectors of norm -1. The actual geodesics in the bundle are the
geodesics with normal vectors in `.B.
Geodesics through a point Let us fix a point A of H3 and introduce

the following set of 2-vectors
A={XEA2MIAAX=O}
It is easy to check that exterior multiplication A A defines an exact sequence
5.3 0 -a A °M -* A 1 M A 2M -> A 3M - A 4M _ 0
From this it follows that .A = Im( A A: A 1M--> A 2M). This description reveals
that A has dimension 3 and consists of 2-vectors with real norm, 1.18. From
<A A X,A A Y> = <X,Y> ; X,Y E TA(H)
it follows that A is isometric to the tangent space TA(H3), in particular the
Sylvester type is (-3,0). An oriented geodesic curve k has normal vector K E A if
and only if k passes through A, compare 2.8. In resume, we have realised the set
of geodesics through A as the geodesics in the bundle A.
Geodesics with a given end Let us fix an isotropic line in M

represented by the isotropic vector Q and introduce the set Q of 2-vectors
Q= {XEA2MI XAQ=0}
From the exact sequence 5.3 it follows that Q = Im( A Q: A 1M-* A 2M). This
description reveals that A has dimension 3 and consists of 2-vectors with real
norm, 1.18. Let us pick three linearly independent vectors A,B,C in M with
<A,Q> = <B,Q> = 0 and <C,Q> = 1. From
<Q A X,Q A Y> _ - <Q,X><Q,Y> ; X,Y E M3
we find that Q A A, Q A B, Q A C form an orthonormal basis with norms 0, 0, -1.
An oriented geodesic k has normal vector K E Q if and only if Q belongs to the
plane spanned by k, compare 2.8. To summarise, we have realised the set of
geodesics with end RQ as the set of geodesics in the bundle Q.
Geodesics perpendicular to a plane Let us consider a hyperbolic

plane P in H3 with normal vector S E A 3M and let us introduce the following set
of 2-vectors
N={XEA2MI XA*S=0}
A look at the exact sequence 5.3 reveals that N consists of vectors of real norm
and that its dimension is 3. In order to calculate the Sylvester type , observe that
<*S,*S> = - 1 and that X r+ X A *S defines an anti-isometry between (*S) 1 and
N. In particular, the Sylvester type of X is (-1,2). It follows from 4.6 that a
geodesic n has normal vector N in N if and only if n is perpendicular to P.
V111I.5 BUNDLES OF GEODESICS 257
Geodesics in a plane Let us consider a hyperbolic plane P with given

normal vector S E A 3M and form the following set of 2-vectors
9_{XEA2MI*XA*S=0}
From the previous case we find that 9 consists of vectors of real norm and that its
dimension is 3 of Sylvester type (-2,1). A geodesic k has normal vector in 9 if
and only if k C P as follows from 4.4.
Let us summarise our findings in table 5.4 below. The heading "Pencils"
indicates the possible Sylvester types of pencil in the bundle.
5.2 Bundle Sylvester type Pencils

Geodesics through a point
Geodesics with a given end
Geodesics perpendicular to a plane
Geodesics in a plane
Let us notice that the Hodge * transforms spaces of type (-3,0) or (-1,0) into
spaces of type (0,3) or (0,1); however, such spaces will not contain 2-vectors of
norm -1. We have, in fact, exhausted the possible Sylvester types:
Lemma 5.5 A three-dimensional subspace B of A 2M consisting of 2-

vectors with real norm has Sylvester type from the following list
(0,3) (-3,0) (0,1) (-1,0) (-1,2) (1,-2)
Proof We must rule out the four types (0,2), (-2,0), (-1,1) and (0,0). In
the first three cases it is easily checked that an orthonormal real basis forms a
complex basis as well. These three types can now be ruled out on the grounds that
the complex inner product on A 2M is non-singular.
In order to rule out the fourth case, observe that a hypothetical three-
dimensional subspace consisting of 2-vectors of norm 0 would span a complex
subspace E isotropic for the complex inner product, i.e E C E 1 . Since the inner
product on A M2 is non-singular we get that dimCE = 1, 0 , which is a
contradiction.
Theorem 5.6 Table 5.4 is a complete list of bundles of geodesics in

hyperbolic 3-space H3.
Proof By theorem 3.4 it suffices to prove that SO3 =SO( A 2M) acts
transitively on the bundles of given type.
Let us treat type (-2,1) separately. A bundle is specified by a sequence
el,e2,e3 of orthogonal 2-vectors of norms -1,-1, 1. It is clear that 03 = O( A 2M)
acts transitively on such sequences. An orthogonal transformation has del = ± 1,
but the sign can be changed by permuting the first two vectors leaving the third
invariant. Types (-3,0) and (-1,2) can be treated in a similar fashion.
Let us consider a bundle E of type (-1,0) and let us prove that E fl iE # 0.
Pick an orthogonal basis el,e2,e3 with norms -1, 0, 0 to see that E n iE = 0 is
equivalent to E + iE = M, which implies that el,e2,e3 are linearly independent over
C; in particular we find that (Ce2 + Ce3) 1 = Ce2 + Ce3 contradicting that our
complex form is non-singular. Let us show that E fl iE C E 1 : starting with
<e,z>_-i<ie,z> ;zEE,e E EniE
and the observation ie E E we conclude that <e,z> = 0 as required. This shows
that we can construct a real basis for E as follows: pick any vector e E E with
norm -1 and a non-zero vector f E E fl iE to get the basis e, f, if. This shows
how to recover E from a pair of orthogonal vectors e,f of norms 1, 0. It follows
from Witt's theorem 1.2.4 that 03 acts transitively on such data. To see that S03
acts transitively, let us consider an orthonormal basis el, e2, e3 and put e = el,
f = e2 + ie3 ; observe that we can permute e2 and e3 keeping el fixed.
VIH.5 BUNDLES OF GEODESICS 259
Let us emphasise that the fact that geometrically different pencils have dif-
ferent Sylvester types amplifies statements like "through a given point there passes
at most one geodesic perpendicular to a given geodesic". As an illustration let us
apply this to the theorem of Whitehead, Smid and Van der Waerden, see
[Coxeterl) § 4.
Whitehead's theorem 5.7 If there are five distinct hyperbolic

planes containing pairs of four given geodesics, the remaining pair also consists of
two coplanar geodesics (a more precise statement is given in the proof).
Proof Let the four geodesics be a,b,c,d. In the graph below we have
represented each of the five hyperbolic planes by a bar. The problem is to show
that the geodesics b and d are coplanar.
F a-
b d
The assumptions may be restated: the triples a, b, c and a, c, d generate bundles

`.B and `D of Sylvester type # (-2,1). Observe that `9 and `.D have the pencil
generated by a,c in common. It follows from 5.8 below that `B = 5.
Lemma 5.8 Let `.B and `.b be bundles of geodesics both of type # (-2,1). If
`B and 9 have a common pencil, then `B = `D.
Proof Inspection of 5.4 reveals that `B and `D must be of the same type in
order to have a common pencil. We ask the reader to go through the three cases.
VIII.6 HERMITIAN MATRICES
In this section we shall introduce the space M of 2 x 2 Hermitian matrices

and show that the determinant is a quadratic form on M of type (-3,1). In fact,
this realisation of Minkowski space has been used in physics for a long time.
Let us fix the notation with respect to some basic operators on the algebra
M2(C) of 2 x 2 complex matrices. The cofactor matrix of A is denoted A`
d -b I
6.1 ; a,b,c,d E C
-c a
The basic formula for manipulations with this symbol is
6.2 AA'= A-A = t det A ; A E M2(C)
Our symbol is an anti-autornorphism in the sense that
(AB)"=B-A"
6.3 ;A,BEM2(C)
On the space M2(C) we shall consider the symmetric bilinear form
6.4 <A,B> = tr(A B") ; A, B E M2(C)

2
From the fact that A A" = t det A it follows at once that
6.5 <A,A> = del A ; A E M2(C)
Polarisation of formula 6.2 gives us the fundamental formula
6.6 AB'+BA-_ <A,B>2t ; A,B E M2(C)

VIII.6 HERMITIAN MATRICES 261
The second operator of importance is the star A* = TA or complex

transposed. A matrix which is invariant under the star operator is called
Hermitian. The space of Hermitian matrices is denoted
M= {XEM2(C)I X"=X}
Quite explicitly, a Iermitian matrix looks like this
; a,dER,bEC
From this description it follows that the inner product takes real values on M. In
particular, the determinant defines a real quadratic form on M. An orthonormal
basis for M is formed by the
6.7 Pauli matrices

1 0 0 1 0 -i 1 0
0 1
0.1
1 0 '
0'2
i 0
'73 = [ 0 -1
]
which have norms 1,-1,-1,-1 respectively. This shows that M is a Minkowski

space. The sheet of the unit hyperbola in M consisting of positive definite forms
(all > 0) will be our Hermitian model of H3. The group S12(C) acts on M:
6.8 0-.X= crXCT` ; aES12(C),XEM
These transformations are orthogonal and we conclude from the fact that S12(C) is
connected that a transformation of the form 6.8 preserves orientation and the
connected components of the unit hyperbola in M. Thus we deduce a morphisml
S12(C) - Lor+(M)
which we shall demonstrate to be surjective, see 8.5. It is this fine action of S12(C)
that makes the Hermitian model of H3 extremely useful.
'A different proof of this will be presented during the proof of 8.5. As it
happens, it follows from the same source that S12(C) is connected. See also 8.9.
262 IIYPERBOLIC 3-SPACE
VII1.7 DIRAC'S ALGEBRA
In this section we shall introduce an algebra which has been used in

physics by P.M.Dirac2. In mathematical terms it is a concrete realisation of the
Clifford algebra of the Minkowski space M, [Deheuvels, X.6]. However, the reader
is not assumed to be familiar with the formalism of Clifford algebras.
The Dirac algebra D is built over the real vector space M2(C) ® M2(C)
with multiplication given by means of the automorphism A i-- A = TA- of M2(C)
7.1 (A,B) (C,D) = (AC + BD,AD + BC) ; A,B,C,D E M2(C)
We ask the reader to verify that the Dirac algebra D is associative with multipli-
cative unit (i,0). Let us introduce the element %P = (O,t) and let us identify a
matrix A E M2(C) with the element (A,0) of the Dirac algebra. Then we can write
(A,B) = A + BAY ; A,B E M2(C)
The multiplication is ruled by the formulas
7.2 q,2=c , q'A=AP ;AEM2(C)
The principal involution is the autornorphism of D given by
7.3 (A + BWY)# = A - BAY ; A,B E M2(C)
Its eigenspaces relative to 1 and -1 are denoted D+ and D_ respectively. The

principal anti-involution X X" of the Dirac algebra is given by
7.4 (A + BW)- = A-+ WB` ; A,B E M2(C)
We ask the reader to check that indeed (XY)' = YX , X,Y E D. Our next
2See A.Messiah, Quantum Mechanics, North-Holland Publishing

Company, Amsterdam 1961.
VIII.7 DIRAC'S ALGEBRA 263
formula is a simple transcription of formula 6.6
7.5 RW SW + SAY RW = 2 <R,S> t ;R,SEM
The Minkowski space M is oriented by means of the Pauli matrices 6.7. This has
the effect of making A 2M into a complex vector space with a complex quadratic
form, 1.15. Let us introduce the D-valued alternating bilinear form on M
7.6 R x S= 2(RIYSW - SWRW) ; R,S E M
Observe that this symbol is invariant under the principal autornorphism but anti-
invariant under the principal anti-automorphism. As a consequence we conclude
that the symbol takes values in the following subspace of D+
s12(C) = {X E M2(C) I Ir X = 0 }
The space s12(C) has a natural complex structure and the determinant defines a
complex quadratic form on s12(C).
Theorem 7.7 The assignment A A B H A x B defines an isometry

A 2M - s12(C)
from the complex vector space A 2M equipped with the complex quadratic form
from 1.14 onto s12(C) equipped with the determinant form.
Proof The Pauli matrices Qo,v1,Q2,o3 form a positively oriented orthonormal

basis for M. It follows from 6.6 that
o,.W osW + osT orW = 0 ; r$s

7.8 (o0'h)2 =I , (osW)2 = -t s=1,2,3
It is also worth recording that the scalar matrix i = it can be written
7.9 i = o0W0r1WYo2WY0'3W = ouo10203

Let us prove that the map R A S f-+ R x S is C-linear. According to the formulas
1.16 this is equivalent to the formulas
i(o1XO2)=0.3Xo0, i(a2X0-3)=al xa0' t(o3xol)=0,2X00
These can be verified by means of 7.7 and 7.10. Let us be content with the first
00 471*aAo3" (0,1' 2W) = - 00"P02P03' A = 03*00%p
It remains to verify that of X o2 , a2 X 03, 0-3 x a1 form an orthogonal basis for
s12(C). Let us,/ verify that the first two vectors are orthogonal:
-2t'1(a1 X a2)(a2 X 03)) 21T(o1%Yo3'Y) = 0
The remaining details are left to the reader.
Let us end this section with some information for readers familiar with the
general formalism of Clifford algebras [Deheuvels].
Corollary 7.10 The Dirac algebra D is isomorphic to the Clifford algebra

of the quadratic space M of Hermitian matrices.
Proof Let us observe that we have the direct sum decompositions

D+=R sl2(C)®iR , D_=MW(B iMW
It follows from this that the R-algebra D is generated by MW. On the other hand
the real dimension of D is 16 and the dimension of the Clifford algebra is 24.
For the benefit of the reader we should mention that the Dirac matrices
refer to the following eight-dimensional real representation of Dirac's algebra
A B
A+BW ; D -' M4(C)
B A
It is known that D M2(H), compare execise 7.1. It follows that the eight-
dimensional representation above is simple.
VIII.8 CLIFFORD GROUP 265
VIII.8 CLIFFORD GROUP
In this section we see how a certain subgroup of the multiplicative group

U(D) of Dirac's algebra D acts on M. Since we have already identified A 2M with
a subspace of D, this will bring, for example, normal vectors of geodesics in H3 to
act on M. We shall apply this to the finer study of even Lorentz transformations,
and conclude the section with the main theorem of this chapter.
The map Ri-+R* identifies the space M with a linear subspace of the
Dirac algebra D. The Clifford group r is the subgroup of those G E U(D) which
have the following stability property relative to the subspace MAY of D
8.1 G# MW G-1 = MW
Thus the group r acts on M through the formula
8.2 (G.X)41 =G#XWG-1 ;XEM,GEF
Let us observe that G E r acts on M as an orthogonal transformation:

<G.X,G.X> t = (G.x)W(G.X)W = - (G.X)WY((G.X)W)# = <X,X>
As an example, let us consider a non-isotropic vector S E M and the corresponding

unit St E I. The action of S%P on M is reflection TS along S:
8.3 (SW).x = rs(x) ;XEM
Proof Let us recall from 7.5 the formulas

SWSW = <S,S> t , SWXW + XWSW = 2 <X,S> t ;XEM
Let us multiply the second formula from the right by (SW)-1 = S4/<S,S>-1 to get
- S1YXW(SW)-1 = XW - 2<X,S><S,S>-1SW = rs(X)W
Let us identify G12(C) with the group of multiplicative units in D+. For a
matrix R E G12(C) with real determinant we have i0 = R*(det R)'1 , thus
RXIR'1 = RXR-1 F
= RXR*,P(det R)-1 ;XEM
which shows that R E r+, where I'+ = D+ fl r. We ask the reader to show that
r+={REG12(C) I MR ER)
8.4
R.X=RXR''=RXR*(detR)-' ;REr+,XEM
Theorem 8.5 The action of S12(C) on M defines an isomorphism of groups

PS12(C) = Lor+(M)
Proof Let us show that R E r+ acts as a Lorentz transformation on M if and

only if del R is positive. To this end we ask the reader to verify the identity
it
a b 1 0
][ a c ][ 1 0
] =as+bb+cc+dd
c d
Let us also remark that R E r+ acts trivially on M if and only if R is a real scalar
matrix: this follows from the proof of 7.10. Consider RE r+ with del R > 0 and
decompose the action of R on M into a product of reflections T1T2...Td along
vectors Sl,...,Sd of norm -1. It follows that we can find k E R* such that
R = k SOP S2I'...Sd
Let us make use of the principal anti-involution 7.4 to get that
k2(-1)d
R R- = k2 S1S1- S2S2"... SdSd. =
Using that delR = RR' is positive we conclude that d is even.
An even Lorentz transformation of Al can be written TATBTCTD, where
A,B,C,D E M, each of determinant -1. According to 7.5 this is the action of
ABCD
This allows us to conclude that S12(C)-Lor+(Af) is surjective. 0
Vffl.8 CLIFFORD GROUP 267
Definition 8.6 By a half-turn in a geodesic k in H3 we understand the

orthogonal reflection in the linear span of k in the ambient Minkowski space.
Main theorem 8.7 Let S E S12(C) act on H3 as the product aQ of half-

turns a and i in geodesics h and k with normal vectors H and K. Then
tr2 S = 4 <H,K>2
Any even isometry of H3 can be decomposed into a product of two half-turns.
Proof Pick a point A E h and a unit tangent vector T to h at A such that

H = A X T = 2(AWTW - TWYAW) = A'YTWY
We conclude from formula 8.3 that H E I'+ acts on M as -a. Observe that the
scalar matrix i = it acts as the antipodal map x -x, x E M, to get that
8.8 (i H). X = a(X) ; XEM
It follows that (iH)(iK) E S12(C) acts as a/3 on M which gives S = f iH iK. Let
us recall that K" = -K, since K E s12(C). This gives us
lr(ill iK) = - tr(HK) = tr(HK-) = 2<H,K>
The required formula follows by squaring this relation. Conclusion by 8.9. 0
Lemma 8.9 Any S E S12(C) can be written S = HK where H,K E s12(C) with
detH = de1K = -1.
Proof We ask the reader to use the theory of quadratic forms to find vectors
H,K E s12(C) with <11,H> = <K,K> = -1 and 2<H,K> = tr(S). Observe that
1r2S = 1r2HK and conclude from 1.9.15 that S and IIK are conjugated in PSI2(C).
Upon changing the sign of H we may assume that S and HK are conjugated in
S12(C). 0
VIII.9 LIGHT CONE
In this section we shall work with the Hermitian model for H3 and identify
its boundary OH3 with t keeping track of the action of S12(C). Once this is done
we write down the normal vector L(A,B) for the oriented geodesic with ends
A,B E C and evaluate <L(A,B),L(P,Q)> in terms of the cross ratio [A,B,P,Q].
To begin let us assign to e E C2 the following Hermitian matrix
T
:]=[:]
Zz
9.1 L(e) _
w
;e=[]
Notice that L(Ae) = ).L(e) , A E C, and observe the pleasing transformation rule
9.2 L(v(e)) = o L(e)o * ; o E S12 (C), e E C2
which follows immediately from the decomposition

T
zZ Zw z Z
;z,wEC
wz ww w w
Proposition 9.3 The transformation L induces a S12(C) equivariant

bijection from t onto the space PC(M) of isotropic lines in M.
Proof The equivariant part of the statement is formula 9.2. The verification
of bijectivity is straightforward. Let me take the opportunity to point out that the
construction performed here is a special case of the general treatment in 1.7. It
suffices to identify C ® R2 with M through the assignment
; zEC,x,yER
Specification of 1.7.5 gives another proof of the isomorphism Mob(C)- Lor(M). 0

VIII.9 LIGHT CONE 269
Circles Let us consider a vector N E M of norm -1, quite explicitly
;bEC,a,dEl,ad-bb=-1
This determines a hyperbolic plane P = H3 fl N 1' . By the boundary 91? of P we

understand the set of isotropic lines contained in the linear hyperplane N 1
spanned by P. If we take advantage of the isomorphism t +PC(M) we get the
following equation for OP
=0
an equation which may be written

T
Y d -b z
9.4 =0
a w
which is the general equation of a circle in C.
Pair of ends An oriented geodesic k in H3 can be specified through its ends

which we may consider as an ordered pair (P,Q) of distinct points of C. Let us
write P = w and Q = V ] and show that the normal vector N of k is given by
9.5 L( ; D = det I zu
wv J
Let us transform this through the correspondence A 2Af - s12(C) to get
zv-}-wu -2zu
9.6 L ( P ,Q) = zv lwu
2wv -zv-wu
;P=[z1,Q=[°I
This matrix is easily seen to have determinant -1. Let us recall from 8.11 that
the matrix iL(P,Q) represents a Mobius transformation which represents a half-

turn with respect to the geodesic with ends P and Q. It follows from 9.2 and 9.5
that the symbol L(P,Q) obeys the transformation rule
9.7 aL(P,Q)a' = L(o(P),o(Q)) ; O 'E S12(C)
We shall evaluate the complex inner products of 2-vectors of the type 9.5 in terms
of the cross ratio on C. Recall from 1.9.8 that for four distinct points A,B,P,Q of
C the cross ratio [A,B,C,D] $ 0, 1, oo.
1
9.8 <L(A,B),L(P,Q)> =
[A,B,P,Q]+1
Proof According to 9.7 and 1.9.7 both sides of the formula are invariant under
S12(C). From the fact, 1.9.3, that the action of S12(C) on C is triply transitive it
follows that it suffices to verify the formula in the case where A = oo, B = 0,
P = 1, Q = q with q E C, q 0, 1. Recall from 1.9.9 that [oo,0,1,Q] = q. We find
that
+q -q
2
L(00,0) L(1,Q) = 1_' q
0 -1 2 -1-q
which gives tr L(oo,0)-L(1,Q) = q + 1 as required. 0

z
q-1
Corollary 9.9 Let A,B,P,Q be four distinct points of the Riemann sphere C.
The geodesics h and k in h3 with ends A,B and P,Q are coplanar if and only if
[A,B,P,Q] E R. The two geodesics h and k will intersect if and only if [A,B,P,Q] is
negative; h and k are perpendicular if and only if [A,B,P,Q] = - 1.
Proof The transformation q ,--, q + 1

q-1 is an involution which preserves k For
the remaining details see 111.10.8.
VIII.10 QUATERNIONS 271
VIII.10 QUATERNIONS
Poincare's half-space model of hyperbolic 3-space is the open subspace

C x ]0,+oo[ of C x R equipped with the global metric p given by the formula 11.7.2
2
10.1 cosh p(R,S) = 1 + IR2rs I R = (z,r), S = (w,s)
The group S12(C) acts as isometries on this model: if we agree to view the point R
above as a quaternion R = x + iy +j r, then the action is given by the formula
10.2 a(R) = (aR + b)(cR + d)-1 ;a=
see, for example, [Beardon] or [Fenchel].

We can give an S12(C)-equivariant isometry from the Poincare model to
the Hermitian model of 113 by assigning to the point R from Poincare's half-space
the complex 2 x 2 matrix
z + r2 z
10.3 F(R) = ; R=z+jr
z 1
2
trF(R)°F(S)=1+JS,R1
rs ;R=z+jr,S=w+js
which shows that F is an isometry as required, compare 11.4.1, 7.5. and 10.1. The
relevant formula for S12(C)-equivariance
10.4 F(o(R)) = a F(R) a" ; a E S12(C)
is left to the reader.

VIII.11 FIXED POINT THEOREMS
In this section we shall apply geometry to prove some fixed point theorems
for subgroups of PS12(C). Let us first recall that or E PS12(C), o # t, has one or
two fixed points on OH3. In the case of two fixed points, the geodesic through
these is stable under o and is called the axis of o. We shall analyse o in terms of
theorem 8.7 on half-turns. Details of some of the arguments can be found in the
exercises.
Elliptic transformation 1r2o E [0,4[ . The transformation o has two

fixed points on OH3 and all points of the axis are fixed points. The transformation
a can be realised as a product of half-turns in two geodesics through a given point
of the axis, both perpendicular to the axis. All planes perpendicular to the axis are
invariant under o. Elliptic transformations with a fixed axis h can be realised as
the product of two reflections in planes belonging to the pencil of planes through h.
Parabolic transformation tr2o = 4. The transformation o has preci-

sely one fixed point oo on OH3 and is the product of two half-turns in geodesics
through oo. The hyperbolic plane P spanned by these two geodesics is an invariant
plane. The transformation o can be realised as the product of two reflections in
planes belonging to the pencil of planes through oo perpendicular to P.
The parabolic transformations (including t) fixing a given point oo E OH3
make up the commutator group of the subgroup of PS12(C) fixing oo. This can be
seen by explicit computations, compare the proof of 1.9.13.
Hyperbolic transformation tr2o E ]4,oo[. The transformation has

two fixed points on OH3 but no fixed points on H3. The transformation leaves any
plane containing the axis invariant. The transformation can be realised as the
product of half-turns in two geodesics contained in a fixed invariant plane

perpendicular to the axis. A hyperbolic transformation with fixed axis h can be
realised as a product of two reflections in planes belonging to the pencil of planes
perpendicular to h.
Strictly loxodromic transformation tr2Q 0 [O,oo]. The trans-

formation has two fixed points on 8H3 but none on H3. The transformation can be
realised as a product of half-turns in geodesics perpendicular to the axis but not
contained in a plane. A transformation which is either strictly loxodromic or
hyperbolic is called loxodromic.
The group of transformations fixing two given points A and B of OH3 is
commutative, being the product of the group of rotations with axis AB and the
group of parallel translations along the geodesic AB. This group has index 2 in
the group of transformations stabilising the geodesic with ends A and B.
Elliptic fixed point theorem 11.1 A subgroup G of PS12(C)

consisting of elliptic transformations has a fixed point on H3.
Proof Let us consider two distinct non-trivial members o and r of G and show
that their axes h and k intersect. To this end we shall rule out two alternatives:
common end and common perpendicular.
When h and k have a common end, the commutator oro 1r 1 is parabolic
and we conclude that o and r commute. The fixed point set h for c on H3 is
stable under r as follows from the "normaliser principle" IV.2.6. It follows that
h=k or ( when r is a half-turn) It is perpendicular to k.
When It and k have a common perpendicular, let A E It and B E k be
points such that the geodesic I through A and B is perpendicular to h and k. Pick
geodesics a through A and b through B such that a = a\ and r = A,3 where a,3,\
are half-turns in the geodesics a,b,l. Observe that or = a# and conclude from 8.7
that a and b intersect. This implies that the plane P spanned by a and b is perpen-
dicular to the axes h and k for a and r. We conclude from the proof of IV.2.8 that
A = B and therefore h = k.
Since any two axes of rotation are coplanar, we conclude that the axes of
rotation are members of some bundle X. Since any two axes intersect, we
conclude from 5.4 that all axes pass through the same point of H3. 0
Definition 11.2 A subgroup G of PS12(C) is called elementary if G leaves

a geodesic invariant or fixes a point of H3 or a point of the boundary 8H3.
Corollary 11.3 A subgroup G of PS12(C) with a finite orbit on H3 or 8H3

is elementary.
Proof Suppose that G has a finite orbit on H3 of order n > 1. Any non-
trivial y E G acts as a permutation on the set 0 on n elements and we find that
y"! has a fixed point on H3. From this it follows that y" is a elliptic (or t) and
we conclude that y is a elliptic: the class of loxodromic (resp. parabolic elements)
is stable under n!-powers. It follows from 11.1 that G has a fixed point on H3.
Suppose that G has an orbit on OOH3 consisting of n > 1 elements. In case
n = 1, the group has a fixed point on 8H3. When n = 2, the geodesic through the
two point orbit is stable. When n > 3, we find that all elements of G have finite
order and the elliptic fixed point theorem tells us that G has a fixed point on H3.
Corollary 11.4 A non-elementary subgroup G of PS12(C) contains a

loxodromic transformation.
Proof Let us assume that G does not contain a loxodromic element. According
to the elliptic fixed point theorem, the group G contains a parabolic element y,
fixing oo say. For any element a E G we have that
1r2(y"c)E[0,4] ;nEN
In order to exploit this we put
1 1 a b
; a,b,c,d E C, ad-bc = 1
0 1 c d
Direct calculation gives us
tr2(y°o) = (a + d + nc)2 E [0,4] ;nEN

But this is only possible if c = 0, which means that o(oo) = oo. This being true
for all o E G makes G elementary contrary to the hypothesis. 0
Invariant disc theorem 11.5 A non-elementary subgroup G of

PS12(C) fixes an oriented hyperbolic plane if and only if G doesn't contain strictly
loxodromic elements.
Proof Suppose first that G doesn't contain strictly loxodromic elements. Let
us show that any two hyperbolic transformations u,v E G have coplanar axes. To
this end we may assume that the axes m and it have a common perpendicular c.
Let us write p = ay and v = y,3 where y is a half-turn with respect to c, a is a
half-turn with respect to a geodesic a perpendicular to m and a is a half-turn with
respect to a geodesic b perpendicular to it. From the fact that none of the
transformations ay, 'y,(3 , a,6 are strictly loxodromic it follows that a,b,c are
pairwise coplanar, compare 8.7. It follows from 5.4 that we can find a plane P
such that either a,b,c are perpendicular to P or all lie in the same plane P. In
both cases we conclude that m and n lie in P.
Let us remark that G contains a hyperbolic transformation p with axis m
say, 11.4. For o E G we find that opo l is hyperbolic with axis o(m). It follows
that we can find another hyperbolic transformation v E G with axis n # m. At
this point we ask the reader to use 5.4 to find a plane R containing the axes of all
hyperbolic elements of G. Observe that R is stable under G.
It remains to prove that G preserves the orientation of R. To this end we
must prove that the group I' C Isom(R) induced by G consists of even transfor-
mations only. Assume for a moment that this is not the case. Taking into
account that r is non-elementary, it follows from IV.2.9 that we can find a glide
reflection y E r with axes g, say. Let us write y = rRKR where r is restriction to
R of a translation r in H3 along g and KR is restriction to R of reflection 1s in the
hyperbolic plane K perpendicular to R through g. Let p denote reflection in R and
conclude that rtzp E 1'. This is a contradiction since rKp is strictly loxodromic.
Conversely, let R be an oriented plane fixed by G. An element y E G
induces a transformation of the plane R which is even. It follows that y is the
product of two reflections in planes perpendicular to R. 0
Proposition 11.6 Any solvable subgroup G of PS12(C) is elementary.

More generally, a subgroup G of PS12(C) with a composition series
{t} =Go CG1
with G1 non-trivial abelian, is elementary.
Proof Let us start by assuming that G1 contains a parabolic element say,

with fixed point S E OH3. Since S is the only fixed point for o on OH3, we
conclude from the "normaliser principle", IV.2.6, that the fixed point set for G1 on
8H3 is {S}. Multiple applications of the normaliser principle show that G fixes S.
Let us assume that G1 contains an element K # t with two distinct fixed
points A and B on OH3 and let us prove that the geodesic k with ends A and B is
fixed by G1. When a2 = t (half-turn) we find that the set of fixed points for a on
H3 is k and we conclude from the "normaliser principle" that k is stabilised by G1.
When tie # t we find that k is the only geodesic fixed by n;: if is fixes the geodesic
h with ends C and D, we find that tit has fixed points A,B,C,D on 8H3 ; it follows
that h = k. We can apply the "normaliser principle" to the action of G1 on the
set of geodesics in H3 and conclude that G1 fixes k.
If k is the only geodesic fixed by G1, it follows from the "normaliser
principle" that G fixes k. Thus we may assume that G1 fixes two geodesics k and
1. It follows from the argument above that G1 consists of half-turns. By
inspection of a single half-turn, we conclude that k and l have a common
perpendicular h. It follows that G1 fixes the two perpendicular geodesics h and k.
VHI.11 FIXED POINT THEOREMS 277
The geodesic m perpendicular to h and k is fixed as well and we conclude that G1

is a subgroup of the group (of order 4) generated by half-turns in h and k. If
#G1 = 4 we find that G1 has precisely one fixed point on H3 and it follows from
the "normalises principle" that G has a fixed point on H3.
We have reduced the problem to the case where #G1 = 2. In this case GI
is generated by a half-turn in a geodesic it which is stabilised by G2 by the
"normaliser principle". We can now repeat the arguments of the last paragraph
and conclude that G is elementary or #G2 = 2. We leave it to the reader to
conclude the proof by a simple induction on n.
Theorem 11.7 The group PS12(C) is simple, i.e. the group contains no
non-trivial normal subgroup.
Proof A non-trivial element o of PS12(C) can be written as a product o-=a,0

of half-turns in geodesics a and b respectively. Let us choose an end A for a and
an end B 54 A for b. We can write a13 = (ay)(-y(3) where y is a half-turn in the
geodesic AB. This provides a representation of v as a product of two parabolic
transformations. It is important to recall that any two parabolic transformations
are conjugated in PS12(C), compare 1.9.13.
Let us show that a normal subgroup N # {t} contains a parabolic transfor-
mation. If not, we can consider an element v with two fixed points A and B on
8H3. Pick a parabolic transformation a with fixed point A and observe that the
commutator a-Ivav-1 # 1 is parabolic and belongs to N.
At this point we refer the reader to [Beardon] for more information, in

particular Jorgensen's inequality.
VIII EXERCISES
EXERCISE 2.1 1° Suppose that dim E = 4. For a non-zero vector w E A 2E

show that dim EH, = 2 or dim E,,, = 4. Hint: Rule out the case dim EW = 3 by
means of proposition 2.5.
2° Let w be a non-pure 2-vector as above. Show that E,,, = E.W.
EXERCISE 2.2 Let Z:M-M be a skew adjoint operator on Minkowski space M.

Show that there exists a unique 2-vector z E A 2M such that zL = Z. Hint: A skew
adjoint operator Z defines an alternating 2-form <Z(x),y>, x,y E M. Show that
there is a unique z E A 2M with <z,x A y> = <Z(x),y>, x,y E M.
EXERCISE 4.1 Let P and Q be distinct hyperbolic planes in H3. Show that
there exists at most one geodesic h perpendicular to both P and Q. Hint: Use
proposition 4.6.
EXERCISE 6.1 1° Prove the general formulas

U+U- =t irU ;UEM2(C)
tr(UV) + tr(U"V) = trU trV ; U,V E M2(C)
2° Apply these formulas twice and deduce that
trUV + trU-V" = 2 trU trV - 2 tr(UV-) ; U,V E M2(C)
EXERCISE 6.2 Consider the complex multilinear form on M2(C)

vol(A,B,C,D) = -4 tr(AB"CD--A'BC'D) ; A,B,C,D E M2(C)
1° Show that the form changes its sign under the permutation
(A,B,C,D)-(D,A,B,C).
2° Show that vol(A,B,C,D) changes sign under the permutation of A and B.
3° Show that the form vol is alternating. Hint: Show that the group of per-
mutations of the letters A,B,C,D is generated by the permutations from 2° and 3°.
40 Show that vol takes the value 1 on the Pauli matrices o-0,q1,2,3
VIII EXERCISES 279
EXERCISE 6.3 For matrices A,B E M2(C) put A x B = BA'). With the
2(AB
notation of exercise 6.2. show that for all A,B,C,D E M2(C).
<A x B,C x D> = <A,C><B,D> - <A,D><B,C> - i vol(A,B,C,D)
Hint: First observe that the formula is valid for A,B,C,D E M. Second, observe
that the formula is C-linear in all variables. Third, observe that the complex
vector space M2(C) is spanned by M.
EXERCISE 7.1 Show that Dirac's algebra D is isomorphic to M2(H). Hint:

Identify the field of quaternions o-I with the ring of invariants under the involution
Xr-+X of M2(C). Observe that the scalar matrices 1,i form a basis for M2(C) as a
vector space over H and define a morphism of rings F:M2(C)- M2(H) by
F(X) = ; X=H+iK, H,KEH
Extend the definition of F to Dirac's algebra D by the convention F(AY) = r o]
EXERCISE 7.2 1° Given D E s12(C). View D as a 2-vector on M and show that

D[X = 2(DX - XD) ;XEM
2° Verify the identity VHI.3.2 D[iD[X = Im[<D,D>] X ;XEM
3° Verify the identity DLDLX + iDLiDLX = - DXD ;XEM
4° Consider R E s12(C) whose determinant is real and different from zero.
Show that
DXD-' = (del D)( D[D[X + D[iD[X) ;XEM
EXERCISE 7.3 Consider the following alternating C-valued 3-form on M

sEo sign(or) A1,A2,A3 E M
t
where the sum is over the group of permutations E3.

1° Show that this form identifies A 3M and N'Y where N is the space of
anti-Hermitian matrices, X* = -X.
2° Identify the lodge star *: A'M A 3M with multiplication by i:M-+N.
3° Identify the exterior product A 2M x A 1 M ---: A 3M with the pairing
2(WX-XW*) ; WEs12(C),XEM
EXERCISE 7.4 Let lb denote the set of matrices X E G12(C) with X- = - X.
; a,dER,bEC
10
Show that the determinant makes this into a Minkowski space. Hint: Use
the transformation T.A-M given by
b iI
-a I
2° Let us consider the following action of the group S12(C) on 13

o'.X=17 Xv' ;aESl2(C),XEA
Show that S12(C) acts on A through Lorentz transformations.
30 Verify the formula
T(crxa:-')=0- T(x)&-I ;XEM,cES12(C)
4° Define a multiplication on M2(C) (D M2(C) by the formula 7.1 replacing A
by A and show that the resulting algebra F is the Clifford algebra of (A,-del).
5° Show that the representation
D - M4(C) ; (A,B) i-->
can be decomposed into two simple representations. Hint: Show that vectors of
the form (z1,z2,z1,z2), z1,z2 E C make up a D-stable subspace of C4.
6° Conclude that the Clifford algebra of a form of type (-1,3) is M4(R).
EXERCISE 7.5 1° Define a multiplication on M2(C) ® M2(C) by the formula 7.1

replacing A by A in the definition 7.1 and show that the resulting C-algebra F is
the Clifford algebra of (s12(C),-del).
2° Repeat 1° with C replaced by R.
EXERCISE 8.1 1° Show that the only involutions in Lor+(M) are the half-
turns. Hint: See exercise I.1.1.
2° Find all involutions in the group Lor(M).
VIII EXERCISES 281
EXERCISE 8.2 Show that an even Lorentz transformation o of M is a product of

two plane reflections if and only if tr-2v E [0,+00[.
EXERCISE 8.3 For U E r show that UU' E ode and UU#' E Ei. Verify that the
spinor norms N:r-R and S:r-*R defined by
N(U)i = UU' , S(U)t = UU#'c ;UEF
are morphisms of groups.
20 Explain how the spinor norm N was used in the proof of Theorem 8.5.
3° Show that U E r acts on M as a Lorentz transformation if and only if the
spinor norm is S(U) is positive.
EXERCISE 8.4 Let r denote an oriented geodesic with normal vector R. Show
that a half-turn p with respect to r acts on M through the formula
p = RL RL + *RL *RL
Hint: Consider a positively oriented orthonormal basis A,T,U,V for M where A E r
and T is a positively oriented unit tangent vector to r at A.
EXERCISE 8.5 Verify the following table of involutions.

Lorentz transformation r S E G12(C)
point reflection iSW SE M, detS=1
half-turn is S E s12(C), det S = -1
plane reflection S4 SEM, detS=-1
EXERCISE 8.6 1° Show that the Clifford group r+ acts on M as orientation

preserving transformations and the induced map r+- SO(M) is surjective.
Hint: Show that r+ is generated by S12(C), i and R*.
2° Show that the action of r on M defines a surjective map 17->0(M).
Hint: Observe that r = r+ U r+1P
EXERCISE 9.1 Show that <L(A,B),L(B,C)> = 1 ; A o B i6C

EXERCISE 9.2 An even isometry a of H3 interchanges two distinct points A and

B of OH3. Show that a is a half-turn with respect a geodesic perpendicular to the
geodesic with ends A and B.
EXERCISE 9.3 Consider two non-intersecting discs in the complex plane and let
the Euclidean line through their centres intersect the bounding circles in the points
A,B,C,D in this order.
10 Show that the hyperbolic distance distance d between the corresponding
hyperbolic planes is given by
tanh.2 2cl =AAD BC
C BD
20 In the case where the two circles are concentric, show that d is the
logarithm of the ratio of their radii.
EXERCISE 11.1 1° Let z = x + iy be a complex number. Verify the formula
eZ 0
1
2
tr 0 ez = cosh x cos y+ i sinh x sin. y
Discuss the action of the matrix on H3 in the cases x = 0 and y = 0 separately. In

general observe that eZ = ece'y.
2° Use this to discuss the general loxodromic transformation.
EXERCISE 11.2 1° Let a 54 1 be a parabolic transformation of H3. Show that

the set of hyperbolic planes in H3 invariant under a form a pencil. Hint:
Investigate the special case a = Lo 1], compare 1.9.15.
2° Let P be an invariant plane for a. Show that or preserves the connected
components of H3-P.
EXERCISE 11.3 Let 9 denote a pencil of hyperbolic planes. Show that the
product of three reflections in planes from 9 is itself a reflection in a plane from 9.
Hint: Show that there exists a hyperbolic plane Q whose normal vector is
orthogonal to the normal vector of any plane from 9.
VIII EXERCISES 283
EXERCISE 11.4 1° Show that an odd isometry ? of H3 can be decomposed as

v = ap where ti is a half-turn and p is a reflection in a plane. Hint: When c.2 # c,
pick a point A E OH3 (distinct from the fixed points of o2) and let p denote
reflection in the plane through B = o(A) perpendicular to the geodesic with ends A
and C = a(B). Observe that ap interchanges the points B and C and conclude
that te = up is a half-turn.
2° Show that an odd isometry o of H3 can be decomposed into a product of
three plane reflections u = a/3y where the mirror plane of y is perpendicular to the
mirror planes of a and ,13. Hint: Consider a decomposition u = icp a in 1° and use
that the pencil of planes containing the axis k of K. has type (0,2).
APPENDIX: AXIOMS FOR PLANE GEOMETRY
The aim of this appendix is to prove that a metric space which satisfies
the incidence axiom and the reflection axiom is isometric to the Euclidean plane or
(after rescaling) the hyperbolic plane. For a precise statement see the first page of
the introduction. The need for a proof comes from the fact that our axioms are
somewhat untraditional. The proof is presented as a systematic series of
deductions from the axioms. This will be carried out to the point where we have
proved the "Pasch's axiom", see the discussion at the end of the appendix.
Let us agree that a straight geodesic curve in a metric space X is a
distance preserving map y: R-X and recall that a line in X is the image of a
straight geodesic curve, compare 11.1. In the following we let X be a plane i.e. a
metric space X which satisfies the following two axioms
INCIDENCE AXIOM Through two distinct points of X there passes a unique

line. The space X has at least one point.
REFLECTION AXIOM The complement of a given line in X has two connected

components. There exists an isometry or of X which fixes the points of the line,
but interchanges the two connected components of its complement.
Sharp triangle inequality A.1 For a point B of the plane not on the
line through two distinct points A and C we have that
d(A,C) < d(A,B) + d(B,C)
Proof We shall turn the problem upside down and assume that we are given
three distinct points A,B,C with d(A,C) = d(A,B) + d(B,C) and proceed to
construct a line through all three points. Pick a,b,c E R with b - a = d(A,B) and
c - b = d(B,C) and choose three straight geodesic curves 0, A, It in X with
0(a) = A , q(c) = C, A(a) = A, .\(b) = B, µ(b) = B, p(c) = C
Let o:R-X denote the curve with restriction 0 to ]-oo,a], restriction A to [a,b],
restriction it to [b,c] and restriction 0 to [c,+oo[. We shall prove that o:IR-+X is a
straight geodesic curve. The triangle inequality gives us (compare VI.2.6)

d(c (x)),a (y)) < y - x ; x,y E R , x<y
In order to prove the opposite inequality, choose r E R smaller than a and x and
choose w E R bigger than c and y to get that
d(o(r),o(w)) d(a(r),o-(x))+d(o-(x),or(y))+d(o,(y)),or(w)) < x-r+d(o(x),o(y)+w-y
Observe that d(v(r),o (w)) = w - r and deduce that y - x < d(o (x),o (y)).
For two points A and B of the plane we define the segment

[A,B] = { P I cl(A,P) + d(P,B) = d(A,B) }
We proceed using the reflection axiom. For convenience we shall use the word
sides for the connected components of the complement of a line k.
Proposition A.2 Given a point P in the plane outside the line k. There
exists a unique point Pk E k such that
d(P,Pk) < d(P,K) ; K E k, K o Pk
Proof Let tc be an isometry of the plane which fixes k pointwise but inter-
changes the sides of k. For a P 0 k, the points K(P) and P lie on different sides of
k. It follows that the line n through P and h(P) will meet k in a point Pk say.
11
K Pk k
1c(P)
Notice, that the point Pk is fixed by K. It follows that the line n is stable under k
since the points P and Pk are mapped into n. By examination of the restriction of
1c to n we conclude that t4tc(P)) = P and that Pk is the midpoint of P and tc(P).
Let us now investigate the distance from P to an arbitrary point K E k,
286 APPENDIX: AXIOMS FOR PLANE GEOMETRY
K :A Pk. From the sharp triangle inequality A.1 it follows that

d(P,a(P)) < d(P,K) + d(K,x(P)) = 2d(P,K)
The left hand side of this inequality can be rewritten as
d(P,tc(P)) = d(P,Pk) + d(Pk,tc(P)) = 2d(P,Pk)
and we conclude that d(P,Pk) < d(P,K) as required. 0
The point Pk E k nearest P is called the perpendicular projection of P onto

the line k. Let us make the general remark that for any isometryp of the plane
A.3 µ(pn) = µ(p),,(n)
The isometry c whose existence is postulated in the reflection axiom is unique and
is called reflection in the line k.
Proposition A.4 An isometry a of the plane which fixes the points of

the line k is either reflection K in the line k or the identity t.
Proof Let K denote reflection in k. If a interchanges the sides of k we have

a = tc according to the remarks above. If or stabilises the sides of n,, then ica will
interchange the sides of k and therefore Ka = a, and we find that a = t. 0
We say that a line it is perpendicular to a line k, n 1 k, if to is stable under

reflection in k but different from k. If n I k , then for any point N E n we have
that Nk E n as follows from the proof of proposition A.2.
Proposition A.5 Given lines a and k in the plane. If n. 1 k then k 1 n.
Proof Assume that it I k and pick a point P of k but not on n. Let K be

reflection in k and conclude from A.3 that tc(Pn) = n;(P)a`(n) = Pn This shows
that Pn E k and it follows that k 1 n. 0
Saccheri's inequality A.6 Let A and B be points outside a given line

k. Their perpendicular projections Ak and Bk satisfy d(A,B) > d(Ak,Bk).
Proof Let us put A0 = A, P0 = Ak, Al = B, P1 = Bk and reflect our

quadrangle in the line A1P1 to get a new quadrangle P1A1A2P2. Reflect this in
the line A2P2 to get a new quadrangle P2A2A3P3...
A0 Al A2 A3 A2n-1 A2n
P2 P2n-1 P2n
From the triangle inequality we find that

d(Po,P2n) < d(P0,A0) + d(A0,A1) + d(A1,A2) + ... + d(A2n,P2n)
Using that reflections are distance preserving we find that
2n d(Po1P1) < 2n d(Ao,A1) + 2 d(P0,A0)
This may be rewritten as
d(Ak,Bk) - d(A,B) < 1/11 d(A,Ak)
Saccheri's inequality follows from this by letting n tend to infinity.
Proposition A.7 For any point K on the line k there exists a point P off k
with Pk = K.
Proof Let us fix one of the sides S of k and observe that perpendicular
projection p:S-*k is continuous as a consequence of Saccheri's inequality. Using
that S is connected, we conclude the image p(S) is a connected subset of k. It
remains to prove that p(S) is dense in k, i.e. for K E k and e > 0 given, there exists
Q E S with d(K,Qk) < c. Observe that for K E k and Q E S we get from A.5 that
k is perpendicular to the line n. through Q, Qk. It follows that d(K,Qk) < d(K,Q).
Finally, observe that for K E k and e > 0 given there exists a point Q E S with
d(Q,K) < e : such a point can be found on any line through K which meets S.
Let us fix a point S of the plane and define a half-turn o around S as

follows: we put o(S) = S and for P $ S we draw the line through P and S and
determine o(P) such that S is the midpoint between P and o(P). The construction
of the map o requires only the incidence axiom. Taking the reflection axiom into
account we shall show that o is an isometry.
Proposition A.8 A half-turn with respect to a point S of the plane is an

isometry of the plane.
Proof Let us use A.7 to pick two perpendicular lines in and it through S and
let p and v be reflections in in and it respectively. We intend to show that pv is a
half-turn with respect to S. It is easy to check that the isometry pvpv is the
identity on in and on it. We conclude from A.4 that pvpv = t, the identity. It
follows that o = pv is an involution with fixed point S. We proceed to show that
S is the only fixed point for o. It is easily checked that S is the only fixed point
for o on m U n. Let us turn to the complement of in U it. Reflection v in n leaves
the two sides M1 and M2 of in invariant: from it 1 in it follows that M1 and M2
both meet it, and we conclude that v has a fixed point on both of them. In the
same way we conclude that p leaves the two sides N1 and N2 of n invariant. The
complement of in U it is decomposed into the union of four disjoint "quadrants"
M1 n N1, Min N2, M2 n N1, M2 n N2
Observe that o interchanges the first quadrant with the third and the second
quadrant with the fourth. It follows that S is the only fixed point for o.
From the fact that o is an involution with S as the only fixed point it
follows that o is a half-turn with respect to S: For a given point P # S the two
points P and o(P) are interchanged by or. It follows that the line h through P and
a(P) is stable under a. Second, we deduce that the midpoint of P and o(P) is a
fixed point for o. In conclusion, the point S is the midpoint for P and o(P). D
Theorem A.9 A line h in the plane which meets a second line k must meet
both sides of k.
Proof Let us show that a half-turn a around a point S E k interchanges the

two sides of k: pick a line n through S perpendicular to k and observe that a and
reflection is in k have the same restriction to it. Let us now assume that h and k
meet at the point S E h. Pick a point H E it belonging to one side of k and
conclude that a(H) E it belongs to the other side of k.
Corollary A.10 Let A and B be two points of the plane outside a line k.
Then A and B lie on different sides of k if and only if k meets [A,B].1
Proof If k does not meet [A,B], then the connected set [A,B] is contained in
the complement of k and must be contained in one of its connected components. If
k meets [A,B] in S, then S divides the line it through A and B into two open half-
lines a and b with A E a and B E b. Since a and b are connected, they are entirely
contained in the connected components of the complement of k. It follows from
A.9 that they are contained in different components.
Corollary A.11 Through a point P of the plane there passes a unique line
perpendicular to a given line k.
Proof If P is outside k, a line it through P perpendicular to k must pass

through Pk, which makes it unique. When P E k, let us analyse a fixed line n
through P perpendicular to k. Reflection tv in k must stabilize the two sides of n
since k meets both of them, A.5. For a point S 0 it we conclude that S and tc(S)
lie on the same side of n and we conclude from A.10 that n is disjoint from
[S,n(S)]. In particular Sk 0 n. and therefore Sk # P.
1 As a consequence of this we find that the relation " [A,B] n k = 0 "

is an equivalence relation on the complement of k. This is the exact content of
Pasch's axiom (M.Pasch 1843 - 1930) which played a key role in Hilbert's axioms.
Let A and B be two distinct points of the plane. The line n through the
midpoint of A and B and perpendicular to the line through A and B is called the
uernendicular bisector for A and B.
Proposition A.12 Given two distinct points A and B of the plane. A

point P of the plane belongs to the perpendicular bisector n of A and B if and only
if d(A,P) = d(B,P).
Proof Reflection v in n interchanges A and B as follows from A.S. Thus

d(P,A) = d(v(P),v(A)) = d(P,B) ;PEn
In order to prove the converse, we shall show that a point P 0 n on the same side
of n as B satisfies d(A,P) > d(B,P). I
n.
To this end, let N be the point of intersection for [A,P] and n. Since P and B are
on the same side of n we have N 0 [P,B] and the sharp triangle inequality gives
d(P,B) < d(P,N) + d(N,B) = d(P,N) + d(N,A) = d(P,A)
The last equality follows from N E [P,A]. 0
This is a good place to stop. Readers with a sense of humour could once
again consult the opening chapter of [Coxeter] on Pons Asinorum. Now we have
proved "Pasch's axiom", a proof of the classification theorem, presented in the
introduction to the book, can be found in [Borsuk, Szmielew] or [Martin].
SOURCES
Let me try to make a list of the sources which have been most useful to
me in preparing the book. At the same time I hope to give an introduction to the
bibliography and some suggestions to further reading.
Chapter I The two volumes of [Berger], the book of [Coxeter2] and

the text of [Wilker]. For more information on orthogonal groups: [Deheuvels].
Chapter 11 The notes of [Rees] and the book of [Nikulin, Shafarevich].
Chapter III Inspired by the book of [Fenchel].
Chapter IV Nielsen and Fenchel, On discontinuous groups of isometric

transformations.... Courant Anniversary Volume. Reprinted in [Nielsen].
Chapter V Most of the material is taken from [Beardon]. The section

on cusps is inspired from [Shimura].
Chapter VI The book of [Nikulin, Shafarevich] contains a lot of ideas

fundamental to the geometric concepts in this book. For further information see
the collected works of [Nielsen] and [Casson, Bleiler].
Chapter VII The proof of Poincare's theorem leans heavily on its prede-
cessors [de Rham], [Maskitl], [Beardon]. For the organization of the proof in case
of non orientation preserving transformations, it has been beneficial to conform to
the principles of combinatorial topology. Here the book of [Zieschang, Vogt,
Coldewey] have been very useful.
Chapter VIII The book [Fenchel] and a paper by [Coxeter']. For the
section on fixed points [Lyndon] and [Beardon].
292 BIBLIOGRAPHY
BIBLIOGRAPHY
A.F. Beardon, The geometry of discrete groups.

Springer Verlag, Heidelberg 1983.
M.Berger, Geometry I and II.
Springer Verlag, Berlin 1987.
K.Borsuk, W.Szmielew, Foundation of geometry.
North-Holland Publishing Co. Amsterdam 1960.
N.Bourbaki, Algeebre III, Algebre m.ultilineaire. Elements de Mathematiques,
Hermann, Paris 1958.
N.Bourbaki, Groupes el Algebres de Lie IV,V,VI. Elements de Mathematiques,
Hermann, Paris 1968.
A.J.Casson, S.A.Bleiler, Automorphisms of surfaces after Nielsen and Thurston.
London Math. Soc. Student Texts 9. Cambridge.Univ.Press 1988.
D.J.Collins, H.Zieschang, Combinatorial group theory and fundamental groups.
Preprint. I.H.E.S., Bures sur Yvette 1989.
H.S.M. Coxeter, The inversive plane and hyperbolic space.
Abh.Math.Sem.Univ.Hainburg 29(1966),217-142.
H.S.M. Coxeter, Regular polytopes,
Dover, New York 3rd edition , 1973
H.S.M. Coxeter, W.O.J Moser, Generators and relations for discrete groups.
R. Deheuvels, Formes quadratiques et groupes classiques.
Presses Universitaires de France, Paris 1981.
H.M.Farkas, I.Kra, Riemann surfaces.
A.Fathi, F.Laudenbach, V.Poenaru, Travaux de Thursten sur les surfaces.
Asterisque 66-67 (1979).
W. Fenchel, Elementary geometry in hyperbolic space.
Walter de Gruyter, Berlin 1989.
BIBLIOGRAPHY 293
J.E.Gilbert, M.A.M.Murray, Clifford algebras and Dirac operators in harmoic

analysis. Cambridge University Press. Cambridge 1991.
M.J.Greenberg, Euclidean and non-Euclidean geometries.
Freeman. San Francisco 1973.
H.B.Griffiths, Surfaces.
Cambridge University Press, Cambridge 1976.
Klein,Fricke , Vorlesungen uber die Theorie der elliptischen Modulfunctionen.
Vol 1,2. Teubner, Leibzig 1890.
S.Lang, Introduction to modular forms. Grundlehren 222.
J.Lehner, A short course in automorphic functions.
Holt, Rinehart & Winston. New York 1966.
R.C.Lyndon, Groups and geometry. London Math.Soc.Lecture Note Series 101.
Cambridge University Press 1985.
R.C.Lyndon, P.E.Schupp, Combinatorial group theory. Ergebn.Math.Grenzgeb. 89
A.M.Macbeath, The classification of non-Euclidean plane crystallographic groups.
Canadian Journal Mathematics, 19(1967) 1192-1205.
A.M.Macbeath, A.H.M.Hoare, Groups of hyperbolic crystallography.
Math.Proc.Cambridge Phil.Soc.79(1976), 235-249.
W.Magnus, Noneuclidean tesselations and their groups.
Academic Press. New York 1974.
Y.I.Manin, Three dimensional hyperbolic geometry as oo-adic Arakelov geometry.
Preprint. I.H.E.S. Bures sur Yvette 1990.
G.E.Martin, The foundations of geometry and the non-Euclidean plane.
B.Maskit, On Poincare's theorem for fundamental polygones.
Advances in Math. 7(1971)219-230.
B.Maskit, Kleinian groups.
W.S.Massey, Algebraic topology: an introduction.
Harcourt, Brace & World Inc. New York 1967.
294 BIBLIOGRAPHY
J.Milnor, Hyperbolic geometry: The first 150 years.

Bull.Amer.Math.Soc.6 (1982)9-24.
M.Newman, Integral matrices.
Academic Press 1972.
V.V.Nikulin, I.R.Shafarevich, Geometries and groups.
J.Nielsen, Collected Mathematical Papers. 2 Vols.
Birkhauser, Basel 1986.
H.Poincare, Theorie des groupes fuchsienne.
Acta.Math.1(1882) 1-62.
G. de Rham, Sur les polygones generateurs des groupes Fuchsians.
I'Ensignement de Math. 17 (1971) 49-62.
E.G.Rees, Notes on geometry. Universitext,
I.R.Shafarevich, Algebra I. Encyclopaedia of Mathematical Sciences vol 11.
G.Shimura, Introduction to the arithmetic theory of modular forms.
Iwanami Shoten and Princeton University Press, Princeton 1971.
J.H.Silverman, The arithmetic of elliptic curves.
W.P.Thuston, Three dimensional manifolds, Kleinian groups and hyperbolic
geometry. Bull.Amer. Math. Soc.6 (1982) 357-381.
J.B.Wilker, Inversive geometry. The geometric vein, edited by C.Davies,
B.Grunbaum, F.A.Scher. Springer Verlag, Heidelberg 1981.
M.C.Wilkie, On noneuclidean cristallographic groups.
Math. Z. 91, 87-102, 1966.
Zieschang, E.Vogt, H.D.Coldewey, Surfaces and planar groups.
Lecture Notes in Math. 835, Springer Verlag, Berlin 1980.
SYMBOLS 295
SYMBOLS
K1 2 vol(K,L,M) 89
O(E), O(Q) 4 HAK 90
0n(k), SO0(k) 5 Lor(S,T) 94
L(e,f) 12 LintP 94, 117, 166, 201

SO(E) 13 Area A 117, 230, 231
S(E) 13, 23 Gl:(7L) 127, 150
GOO(F), µ 17, 30, 35 PSI2(7L) 151, 221
Siml(E) 18 PGl2(7L) 152, 221
Lor(F), Lor+(F) 25 G(2) 153
C(F), PC(F) 27 F(2), r(n) 154, 225

C"(F) 29 r2 r3 224, 225
N(F) 31 156, 158, 218
P(F), PS(F) 31 163, 201, 214
Hn, OHn 31, 70 163, 201, 214
E, 00 32 171
'bb(E) 32 G,t 187

S1
39 201
C ,0 44 goo 214
PG12(C) 44 G, [G, G] 221
S" 44, 91, 106, 119, 136, 260 r, r2, r3 223
[P,Q,R,S] 46, 82, 119, 270 X(X) 224, 230

tr2(S) 49, 82, 265 A IE 240
0 62 ej v 240
TV 62 *w 242
Sn, TA(Sn) 65 or 242
TA(Hn) 70 SO3(C) 250
sl2(R) 88 A 262
<A,B>, A,B E s12(R) 89 262

296 SYMBOLS
INDEX
accidental cycle 204 elementary subgroup 131, 274
alternative cosine relation 108 elliptic fixed point theorem 273
area, hyperbolic 117, 230, 231 elliptic transformation 82, 106, 272
bisecting hyperplane 60, 74 ends of geodesic curve 97, 250
boundary cycle 210 Euclidean vector space 9, 12
branch scheme 225 Euler characteristic 224, 230
bundle of geodesics 255 Euler's proposition 14
Cauchy-Schwarz 12, 16 even Lorentz transformation 25, 34
Clifford group 265 free cycles 214
closed geodesic 197 free edges 213
commutators 137 Fuchsian group 125
complexification 15 fundamental domain 150
conjugated with respect to a sphere 38 genus of surface 224, 229, 230
convex set, hyperbolic 159 geodesic curve 58
coplanar geodesics 255, 259 geodesic lifting property 182
cosine formula, hyperbolic geometry107 glide reflection, Euclidean 64
cosine formula, spherical geometry 67 glide reflection 103, 136
Coxeter matrix 20 Gram matrix 4
cross ratio 46 Grassmann dimension formula 3
cusp 144, 156 Hahn-Banach principle 159
cycle condition 203 "half cycle maps" 214, 219
cycle map 166, 201 half-turn 55, 102, 267
dimension formula of Grassmann 3 Hermitian matrix 260
discrete subset, subgroup 129 Hodge star operator 242
Dirac's algebra 262 homotopy, free 197
Dirichlet domain 167 homotopy lifting lemma 196
discontinuous action 186 Hopf-Rinow theorem 175
edge 163, 200 horocycle, horolation 83, 100
edge cycle 165, 201 horocyclic boundary point 156

INDEX 297
horocyclic compactification 147 modular figure 151
horodisc 144, 158 modular group 127, 150, 221
hyperbolic distance 70 modular group, level 2 153
hyperbolic pants 114, 211 monodromy theorem 171, 185
hyperbolic space 70 non-isotropic vector 6
hyperbolic surface 173 non-singular form 3
hyperbolic transformation 83, 106, 272 normal vector 92, 248, 251
hypercycle 83, 102 normaliser of subgroup 134, 190
indecomposable matrix 21 odd Lorentz transformation 25
infinity, line at 18 orientable surface 229
interior angle of polygon 94 orientation of vector space 241
inversion in sphere 32 orientation vector 242
involution 6, 262 oriented angles 94
isometry of metric spaces 57 orthogonal group 4
isotropic cone 27 orthogonal spheres 37
isotropic vector 6, 7, 17 orthogonal vectors 2
inversion in sphere 32 orthonormal basis 9
inversive product of spheres 39 parabolic cycle map 214

Klein bottle 229 parabolic form 16
Klein disc, model 75 parabolic transformation 83, 106, 272

Lambert quadrangle 110 parallelogram group 208
linearisation 62 pencil of geodesics 97, 255

local homeomorphism 173, 204 pencil of planes 251
Lorentz group 25 perpendicular geodesics 95, 249

Lorentz transformation 25 perpendicular, plane to line 252
loxodromic transformation 273 Poincare disc 76
metric space 57 Poincare extension 38
Minkowski space 243 Poincare half-plane 80
Mobius band 203, 229 Poincare half-space 78
Mobius group 32 polarisation 2, 89, 260

Mobius transformation, even, odd 34 positive definite form 9
298 SYMBOLS
positive form 16 simplex, spherical 69

positively oriented 241 side of convex fund. domain 161
principal minor 11 sine relation 107, 114

projective plane 229 skew adjoint 247
projective space 31 solvable group 134, 276
projectivized cone 27 special Lorentz group 25
pure exterior vector 245 special orthogonal group 13
quadratic form 1 spherical distance 65
rectifiable curve 171 Steiner's alternatives 41
reflection along a vector 6 Sylvester type 10
reflection, Euclidean 13, 60 symmetry with respect to a point 102
reflection, hyperbolic 27 tangent vector to hyperbolic space 70

reflection relations 200 tangent vector to sphere 65
Rieinann surface 147, 224 topological manifold with boundary 146

rigid, rigidity 179 total orthogonal subspace 2
rotary reflection 64 transformation vectors 137

rotation angle 102 transitive action 6
rotation, Euclidean 14 translation axis, length 102, 137

rotation, hyperbolic 102, 106, 272 translation, hyperbolic 83, 106
screw, Euclidean 64 triangle group 209
"shortest length property" 172 uniformization theorem 179
side pairing 163, 200, 213 unorientable surface 229

side transformation 163, 200 vertex of convex fund. domain 161
signature, Fricke-Klein 230 vertex cycle 165, 201

similarity, Euclidean 18, 35 volume form 89, 244
simplex, Euclidean 21, 60 wedge product 90, 240, 263
simplex hyperbolic 74 Witt's theorem 7, 30, 31, 40, 259

(Birger Iversen) Hyperbolic Geometry (B-Ok - Xyz) PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

(Birger Iversen) Hyperbolic Geometry (B-Ok - Xyz) PDF

Caricato da

Copyright:

Formati disponibili

LONDON MATHEMATICAL SOCIETY STUDENT TEXTS

Managing editor: Dr C.M. Series, Mathematics Institute

1 Introduction to combinators and A.-calculus, J.R. HINDLEY & J.P. SELDIN

Cambridge University Press

© Cambridge University Press 1992

This publication is in copyright. Subject to statutory exception

First published 1992

ISBN 978-0-521-43508-6 hardback

1.2 WITT'S THEOREM 6

1.3 SYLVESTER TYPES 9

1.4 EUCLIDEAN VECTOR SPACES 12

1.5 PARABOLIC FORMS 16

1.6 LORENTZ GROUP 23

1.7 MOBIUS TRANSFORMATIONS 32

1.8 INVERSIVE PRODUCTS OF SPHERES 39

1.9 RIEMANN SPHERE 44

11.2 EUCLIDEAN SPACE 59

II.4 HYPERBOLIC SPACE 70

11.5 KLEIN DISC 75

II.6 POINCARE DISC 76

11.7 POINCARE HALF-SPACE 78

11.8 POINCARE HALF-PLANE 80

III HYPERBOLIC PLANE

111.2 PENCILS OF GEODESICS 92

111.3 CLASSIFICATION OF ISOMETRIES 99

111.4 THE SPECIAL LINEAR GROUP 104

111.5 TRIGONOMETRY 107

111.6 ANGLE OF PARALLELISM 111

111.7 RIGHT ANGLED PENTAGONS 112

111.8 RIGHT ANGLED HEXAGONS 114

III EXERCISES 121

APPENDIX: DISCRETE SUBSETS 129

IV.2 ELEMENTARY SUBGROUPS 131

IV.3 GEOMETRY OF COMMUTATORS 137

IV.5 CUSPS 144

V.2 LOCALLY FINITE DOMAINS 155

V.3 CONVEX DOMAINS 159

VI.2 HOPF-RINOW THEOREM 175

VI.3 UNIFORMIZATION 179

VI.4 MONODROMY 182

VI.5 ORBIT SPACES 186

VI.6 CLASSIFICATION 189

VI.7 INSTANT JET SERVICE 191

VI.8 HOMOTOPY 196

VII POINCARE'S THEOREM

VII.1 COMPACT POLYGONS 200

VII.2 TRIANGLE GROUPS 209

VII.3 PARABOLIC CYCLE CONDITIONS 213

VII.4 CONJUGACY CLASSES 218

VII.5 SUBGROUPS OF THE MODULAR GROUP 221

VII.6 COMBINATORIAL TOPOLOGY 226

VII.7 FRICKE-KLEIN SIGNATURES 230

VII EXERCISES 235

VIII HYPERBOLIC 3-SPACE

VIII.1 EXTERIOR ALGEBRA 240

VIII.2 GRASSMANN RELATIONS 245

VIII.3 NORMAL VECTORS 248

VIII.4 PENCILS OF PLANES 251

VIII.5 BUNDLES OF GEODESICS 255

VIII.6 HERMITIAN MATRICES 260

VIII.7 DIRAC'S ALGEBRA 262

VIII.8 CLIFFORD GROUP 265

VIII.9 LIGHT CONE 268