Elmer G.

Rees

Notes
on Geometry

Springer
U niversitext
Elmer G. Rees

Notes
on Geometry

With 99 Figures

4i Springer
Elmer G. Rees
School of Mathematics
The University of Edinburgh
James Clerk Maxwell Building
Edinburgh EH9 3JZ
Scotland
e-mail: [email protected]

Corrected Fourth Printing 2000, 2nd Printing 2004.

Library of Congress Cataloging-in Publication Data

Rees, Elmer G. Notes on geometry /Elmer G. Rees - Corrected 4th printing p. cm. -
(Universitext) Bibliography: p. Includes Index.
ISBN 3-540-12053-x
1. Geometry 1. Title
QA445.R43 1983b 516-- dc19 88-19082 CIP

Mathematics Subject Classification (2000): 51-01

ISBN 3-540-12053-X Springer Berlin Heidelberg New York

This work is subject to copyright. All rights are reserved, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data
banks. Duplication of this publication or parts thereof is permitted only under the provisions
of the German Copyright Law of September 9, 1965, in its current version, and permission
for use must always be obtained from Springer. Violations are liable for prosecution under
the German Copyright Law.
Springer is a part of Springer Science+Business Media
springeronline.com
0 Springer-Verlag Berlin Heidelberg 1983
Printed in Germany
The use of general descriptive names, registered names, trademarks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
Cover design: Erich Kirchner, Heidelberg
Production: LE-W Jelonek, Schmidt & Vdckler GbR, Leipzig
Printed on acid-free paper 41/3142YI. - 5 4 3 2 1
 V

Acknowledgements
I am very grateful to the very many people who have given me encouragement, help and criticism
both whilst giving the course and whilst writing these notes. I would like to single out
Christopher Zeeman who, in the early days at Warwick, inspired my interest in geometry of this
kind,
Michael Atiyah and Graeme Segal who encouraged me to give this course at Oxford,
Graeme Segal who made numerous suggestions and after my initial attempts gave a much better
course than mine. He was very willing for me to incorporate his improvements into these notes.
The approach to hyperbolic trigonometry is entirely his,
Wilson Sutherland who saved me from error many times, and finally
Pam Armstrong and David Stewart-Robinson who prepared the text and diagrams.
VI

Preface
In recent years, geometry has played a lesser role in undergraduate courses than it has ever
done. Nevertheless, it still plays a leading role in mathematics at a higher level. Its central role in
the history of mathematics has never been disputed. It is important, therefore, to introduce some
geometry into university syllabuses. There are several ways of doing this, it can be incorporated
into existing courses that are primarily devoted to other topics, it can be taught at a first year level
or it can be taught in higher level courses devoted to differential geometry or to more classical
topics.
These notes are intended to fill a rather obvious gap in the literature. It treats the classical topics
of Euclidean, projective and hyperbolic geometry but uses the material commonly taught to
undergraduates: linear algebra, group theory, metric spaces and complex analysis. The notes are
based on a course whose aim was two fold, firstly, to introduce the students to some geometry
and secondly to deepen their understanding of topics that they have already met. What is required
from the earlier material is a familiarity with the main Ideas, specific topics that are used are usually
redone.
The style of the course was informal and I hope some of the associated good aspects have
survived into this version. In line with this, I have taken a concrete viewpoint rather than an
axiomatic one. The view that I take is that mathematical objects exist and should be studied, they
are not arbitrarily defined as the axiomatic approach might suggest. This is the view of the vast
majority of mathematicians in their own work and it is a pity that this does not come across in more
undergraduate courses.
There are a large number of exercises throughout the notes, many of these are very
staightforward and are meant to test the reader's understanding. Problems, some of them of
interest in their own right are given at the end of the three parts. Some are straightforward and some
are more like small projects. The more difficult ones are marked with an asterisk.
 VII

Contents
Introduction 1

Part I: Euclidean Geometry 3

The Linear Groups 3
The Relationship Between O(n) and GL(n,R) 5
Affine Subspaces and Affine Independence 7
Isometries of R" 9
Isometries of R2 13
Isometries of R3 15
Some Subsets of R3 17
Finite Groups of Isometries 21

The Platonic Solids 23

Duality 27
The Symmetry Groups of the Platonic Solids 28
Finite Groups of Rotations of R3 35
Crystals 37
Rotations and Quaternions 42
Problems 45

Part II: Projective Geometry 51

Homogeneous Co-ordinates 52
The Topology of P' and P2 53
Duality 57
Projective Groups 59
The Cross-Ratio 61
Fixed Points of Projectivities 62
The Elliptic Plane 62
Conics 65
Diagonalization of Quadratic Forms 67
Polarity 69
Problems 75

Part III: Hyperbolic Geometry 79

The Parallel Axiom 79
The Beltrami (or projective) Model 79
Stereographic Projection 82
The Poincard Model 84
The Local Metric 88
Areas 89
Trigonometry 91
Hyperbolic Trigonometry 92
Lines and Polarity 94
Isometries 97
Elliptic Trigonometry 99
vnl
Problems 102
Further Reading 105
List of Symbols 106
Index 108
 Introduction
In Euclidean geometry, two triangles are congruent if one of them can be moved rigidly onto the
other. Definitions such as that of congruence, which tell us when two objects should be regarded
as being the same, are basic in geometry and are often used to characterize a particular geometry.
Two sets A, B are defined to be equivalent if there is an 'allowed transformation' f such that
fA = B. For Euclidean geometry the allowed transformations are the rigid motions. In his Erlanger
programme of 1872, Felix Klein formulated the principle that a geometry is defined by its allowed
transformations. The force of this principle is to make a close connection between geometry and
group theory.
If S is a set (an example to bear in mind is the Euclidean plane R2), consider the group Bij(S)
consisting of all bijections f: S S. (if S is a finite set with n elements this is the (familiar) symmetric
group Sn.) To impose a geometry on S is to consider a subgroup G of Bij(S); two subsets A, B being
equivalent for the geometry if there is an f e G such that fA = B. For Euclidean geometry, S is R2
and G is the group of all rigid motions. Klein's Erlanger programme not only says that the geometry
on S and the subgroup G determine each other but that they are, as a matter of definition, one
and the same thing. To obtain a worthwhile geometry, the subgroup G has to be chosen with some
care after considerable experience. Usually the set S has some structure and the group G
preserves this structure, examples are
i) S maybe a topological space and the elements of G are homeomorphisms of S.
ii) S may have certain subsets (for example, lines) that are mapped to each other by the elements
of G.
There are many other types of examples; in these notes we study the three 'classical' geometries,
Euclidean, projective and hyperbolic, but the approach is guided by Klein's Erlanger programme.
 Part I

Euclidean Geometry
We start by studying the linear groups. These are probably already familiar to the reader. They
play an important role in the study of geometry.

The Linear Groups

The ring M(n,R) of all n x n matrices over the field R of real numbers has the general linear group
GL(n,R) as its group of units, that is, GL(n,R) consists of all the invertible real n x n matrices. We
will often identify M(n,R) with the space of all linear transformations T: R" --* R". Note that M(n,R)
is a real vector space of dimension n2, and so can be regarded as the metric space R"'. The
determinant defines a continuous map
det: M(n,R) -.R
(continuous because it is given by a polynomial in the coefficients of a matrix), and GL(n,R) is
det-' (R \ (0)), so as R \ (0) is an open subset of R we see that
GL(n,R) is an open subset of M(n,R) = W.
The determinant is multiplicative and so defines a homomorphism of groups
det: GL(n,R) -. R \ (0).
Its kernel is the special linear group SL(n,R) consisting of matrices with determinant 1. The subset
SL(n,R) is closed in GL(n,R) and has dimension n2 - 1 (but is hard to visualize - try to do so for
n = 2).
Euclidean space R" will always be considered with an Inner product x.y defined on it, this
satisfies
i) (x+y).z = x.z + y.z for all x,y,z t R",
ii) (xx).y = X(x.y) for all x,y t R",1\ t R,
iii) x.y=y.xforallx,ytR", and
iv) x.x = 0 q x = 0.
The inner product defines a norm II II on R" by IIx0I2 = x.x and a metric d by d(x,y) =IIx-y;I. Note
that d(x+a,y+a) = d(x,y) so that distance is translation invariant. From the viewpoint of Euclidean
geometry the most important transformations are those that preserve distances. We will now study
such linear transformations.
A linear transformation T: R" -, R" is called orthogonal if Tx.Ty = x.y for all x, y t R".
A basis (e, , e2, ... , e") for R" is orthogonal if
e,.ei=0ifi#j
=1ifi=j.
If we write x =? x;e, and y =±y,e,, then
x.y = 1ty,
where on the right hand side x, y denote the column vectors with entries x,, y, respectively. If T is
an orthogonal transformation and A is the matrix of T with respect to an orthonormal basis, an easy
calculation shows that
LctA'Ay = k
and by choosing various suitable x, y one sees that A'A =1. Check this by using the following result.
Exercise Note that a;; = e,'Aejand henceshow that if x'Ay = xtBy for all x,y t R" then A = B.
4

The matrix of an orthogonal transformation with respect to an orthogonal basis is therefore

orthogonal in the usual sense for matrices. Note that a little care is needed in handling orthogonal
matrices because a matrix is orthogonal if and only if its columns (or rows) form an orthonormal
set of vectors.
If X is a metric space, a map f: X -. X is an Isometry if it is onto and distance preserving, that
is, d(fx,fy) = d(x,y) for all x,y F. X.
Exercise i) Show that a distance preserving map is one to one.
ii) Show that a distance preserving map f: X -. X is not necessarily onto by considering
the map f: R. -. R. defined by f(x) = x+ 1.
Exercise If X is a metric space, verify that the set of all isometries f: X -* X forms a group under
composition.
The properties of the isometries of a metric space X are intimately connected with the properties
of X itself. The importance of orthogonal transformations in Euclidean geometry arises because
they are isometries of R", moreover, apart from translations, they are in a sense all the isometries.
Lemma i) If T: R" -. R" is a linear isometry then T is orthogonal.
ii) if T: R" -. R" is linear and norm-preserving then T is orthogonal.
Proof Notice that any linear isometry is norm preserving because any such isometry satisfies
T(x-y).T(x-y) _ (Tx-Ty).(Tx-Ty) = d(Tx.Ty)2 =d(x.y)2 = (x-y).(x-y)
so by putting y = 0 one gets
I Tx112 = Mr.
Hence it suffices to prove ii). The map T preserves the norm of x-y so
T(x-y).T(x-y) = (x-y).(x-y).
Expanding these expressions using linearity gives
Tx.Tx-2Tx.Ty + Ty.Ty = x.x-2x.y +y.y.
But T also preserves the norms of x and y, so
Tx.Ty = x.y for all x, y e R",
and so T is orthogonal.
Of course there are isometries that are not linear, for example, the translations T0: R" - R"
defined by T,(x) = a + x are not linear unless a = 0. Later we will show that any isometry
1: R" -. R" such that 10 = 0 is linear.
The set of all orthogonal n x n matrices form the orthogonal group
ON _ (Ae GL(n,R)IA'A = I).
If A is orthogonal then det A = ± 1 because det A' = det A and so
1 = del I = det(A'A) = det A'.det A = (det A)2.
The group O(n) has
a normal subgroup, the special orthogonal group
SO(n) = O(n) n SL(n,R) consisting of the orthogonal matrices whose determinant is +1. This
subgroup of O(n) has index 2.
Examples O(1)= (±1), SO(1) _ (+1).
The group 0(2) consists of 2 x 2 matrices d, whose columns are orthonormal. An
a
elementary calculation shows that there is a 0 such that
r
LcJ-Lsins0JandLdj-±r
11

cost)
 5

(cost) -sin 8 This matrix represents a rotation through the angle

SO(2) consists of the matrices =1 cos
0 about the origin. -AsSO(2) has index 2 in 0(2), it has two cosets, one is SO(2) itself and the
other is SO(2). which consists of the matrices [sco 8 osej This matrix represents a
reflection Rein the line t.which makes the angle 0/2 with the x-axis.

(01)
\
[
RQ(a)

0/2

(0,0) -0,0)

R)(b)

The relationship between O(n) and GL(n,R)

A matrix in GL(n,R) has independent columns and a matrix in O(n) has orthonormal columns.
The Gram-Schmidt process transforms an independent set of vectors into an orthonormal set,
so it can be used to define a mapping GL(n,R) --+ O(n). To make this precise it is convenient to
introduce the group T. (n) consisting of the set of upper triangular n x n matrices whose diagonal
entries are positive IT for triangular, + for positive).
Proposition T. (n) is a subgroup of GL(n,R).
Proof If A F T. (n), then det A = a an ... an > 0. Hence T.(n) is a subset of GL(n,R). The
matrix A lies in T. (n) it and only if
aq = 0 for i >j
and Big >0.
If A. B e T. (n) then
(AB)+i = E a,kbki.
If i > j, then either i > k or k > i > j; in the first case alk = 0 and in the second bki = 0, so
(AB)q = 0 in both cases. It i = j, then (AB);, = akbk > 0. So AB f T, (n) if A, B e T,,. (n).
It remains to check that if A e T+(n) then so is A-'. Suppose A e T. (n) and AB = BA = I. We
show that b,i = 0 for i > j by downward induction on I. First consider i = n, then
-0.
Suppose that bki = 0 for all j < k if k > i, then
0=a,,bgforI>j,sob4 =0.
6

Given that b,, = 0 for i > j, one gets that

aubu = 1 for all i.
Hence as a,, > 0 one sees that bA > 0. Hence B e T+ (n).
(The reader may prefer to go through this proof explicitly in the case n = 2.)
Theorem 1 For a given A e GL(n,R), there are unique matrices B e O(n), C F T+(n) such that
A = BC.
Proof As suggested above, we use the Gram-Schmidt process to construct B from A and then
observe that they are related by A = BC with C F T+ (n). In detail: let a,, a2, ..., an be the columns
of A. The first stage of the Gram-Schmidt process is to find an orthogonal set f , , f2, ... f,,. This
is constructed by induction as follows.
fl =al,
fk = ak -'j { (a.f,)/(f,.f,) ) f,.
If F is the matrix with columns f , , f2, ... , in then F = AT,, where T, is in T+(n), and in fact T, has
ones along the diagonal. Note that the matrix F is obtained from A by a sequence of elementary
column operations, each new column involving only earlier columns. The second stage of the
Gram-Schmidt process is to normalise the set f,, f2, ... , f,,, that is, let b, = f, / fh,!I. If B is the matrix
whose columns are b,, b2.... , b then B - FT2 where T2 is a diagonal matrix with positive entries
(1 I iit,1l) on the diagonal, hence T2 a T+(n). If C =(T1T2)-' we have A = BC with B e O(n) and
C e T+(n). Moreover, it is clear from the formulae that the matrices B, C depend continuously on
the original matrix A.
It remains to check the uniqueness of this decomposition. Suppose A has two such
decompositions, B,C, and B2C2 say. Then D = B2 'B, = C2C1-' is In O(n) n T+(n). But we will
show that O(n) n T. (n) =(i) and so the decomposition is unique. Let D e O(n) n T+(n) then
D' = D"' and as T. (n) is a subgroup we have OF. T. (n). But D' is lower triangular so 0 must be
diagonal, and therefore D = D', so using orthogonality, D2 = I. So D has diagonal entries ± 1. As
D e T. (n) it has positive entries on the diagonal therefore D = I as required.
Corollary GL(n,R) is homeomorphic to O(n) x T. (n).
Proof The homeomorphisms are constructed as follows: A e GL(n,R) is mapped to (B,C) and
(B,C) a 0(n) x T. (n) is mapped to BC. These are clearly mutual inverses. The map (B,C) -+ BC
is continuous because matrix multiplication is continuous - the entries of BC are polynomials in
the entries of B and C. The map A- (B,C) is also continuous for a similar reason.
Note T. (n) is homeomorphic to R"1^`'ii2. A matrix in T. (n) has n(n-1)12 entries off the diagonal
and each of these can be an arbitrary element in R. There are n entries on the diagonal and each
of these is an arbitrary element of R+, so that T+ (n) x R^t"''L2 x (R.)". But R. is homeomorphic
to R (under log and exp as inverse homeomorphisms).
So GL(n,R) is homeomorphic to 0(n) x RM^+' u2.
Exercise The space SL(n,R) is homeomorphic to SO(n) x Rt"'*"'Zf'2.
Examples The space GL(1,R) is R \ (0), 0(1) is (±1) and so one can see directly that GL(1,R)
is homeomorphic to 0(1) x R.
The group 0(2) is the union of SO(2) and another coset of SO(2) but SO(2) is homeomorphic
to the circle S' = (z t C I izHH = 1), so 0(2) is homeomorphic to the union of two disjoint copies
of S' and GL(2,R) is homeomorphic to the union of two disjoint copies of S' x R3.
Exercise Show that the group GL(n,R) for n > 1 is not the direct product of its subgroups O(n)
and T. (n).
 7

Exercise Show that GL(2,R) has many subgroups of order three but that 0(2) x T, (2) has only
one such subgroup. Deduce that there is no isomorphism between GL(2,R) and 0(2) x T, (2). (It
is true that GL(n,R) is not isomorphic to O(n) x T. (n) for any n 3 2, but the proof is more difficult
for n -> 3. For n , 3, show that the centre Z of G = GL(n,R) consists of the scalar matrices and
that G/Z has no proper normal subgroups for n odd and only one such for n even. The group
O(n) x T. (n) modulo its centre has several proper normal subgroups.]

Atfine Subspaces and Affine Independence

It is often necessary to consider lines, planes, etc. that do not pass through the origin. Linear
subspaces always contain the origin but their cosets (in the additive group) do not and they are
called affine subspaces. However the affine subspaces have an intrinsic definition.
A subset A of R" is an affine subspace if as + µb e A for all a, b e A and all X, µ e R such that
x + µ = 1. A straightforward induction shows that the following is an equivalent condition:
a,a,eAforalla,eAand± X, = 1.
If V c R" is a linear subspace then it is easy to check that the set V + x = {v+x)v a V) is an affine
subspace of R" for any (fixed) x e R". Every affine subspace A is of this form, because if a e A and
V = A - a =(x-alx a A) then V is a linear subspace of R". Let K e R and
x - a e V then to check that X(x-a) e V we must check that a(x-a) + a e A but
X(x-a) + a = Xx + (1-,\)a
and x,a a A. If x-a. y-a e V then (x-a) + (y-a) e V because
(x-a) + (y-a) + a = x + y-a
which is a linear combination of elements of A, the sum of the coefficients being 1 + 1 - 1 = 1

Exercise If a, b e A and A is an affine subspace, show that A - a = A - b.

If A is an affine subspace, its dimension is the dimension of the linear subspace A -a of R".
If X c R" is any subset, its affine span Aff(X) is defined as
Affi(x)=(±Xx,Ix,ex,±X3=1}.
It is easy to check that Aff(X) is an affine subspace of R" and that it is the smallest affine subspace
containing X.
A set X = {x0, x,, ... , x,) is affinity Independent if r A,x, = 0 holds with f h, = 0
only if AO = x, A, = 0. it is easily checked that (x0, x,, ... , x,) is affinely independent
if and only it (x,-xo, x2-x0..... xe-xo) is a linearly independent set.
If X = (x0, x,, ... , x,) is affinely independent then Aff(X) has dimension r and X is called an
affine basis for Aff(X). Note that an affine basis for an r-dimensional affine subspace has r + 1
elements. If (e,, e2, .... e,) Is a basis of a linear subspace V then an affine basis for V is
(0,el,e2,.. ,e,).
An affine subspace H of R" whose dimension is n - 1 is called a hyperplane. If H is a linear
hyperplane of R", then there is a non-zero x e R" such that H = (x)l. This is because one can
choose an orthonormal basis for H and extend it (by a vector x) to an orthonormal basis for R":
it is then easy to check that H = (x)'. Hyperplanes arise as the perpendicular bisectors of line
segments.
Lemma If a, beR"witha*b,then B= {xI d(x,a)=d(x,b))isahyperplaneinR".
Proof It is clear that (a+b)/2 e B so we need to show that H = B - (a+b)/2 is an (n-1) dimensional
linear subspace. If c = (a-b)/2, it is easily checked using the translation invariance of distance that
H is the set {x I d(x,c) = d(x,-c)}. If c, e2, e3, ... , e" is an orthogonal basis for R", then
8

e2, e., ... , e" is a basis for H.

If H is any hyperplane in R" and x e R", then x can be written uniquely in the form y + z where
yeHand zIH.

More algebraically, let a e H, then H - a is a linear hyperplane, so H - a = (b)1 for some b. There
is a unique expression
x-a=Ab+cwhereceH-a.
Lety=c+a,z=Xb,thenyeHandzIH.
It remains to check the uniqueness. Suppose
Y, +z, =Y2+z2 with y1,Y2eH,z,,z2EHl
Then z2 -Z, = y, - Y2 e H -Y2 and z2 -z, e H -y2, hence z, = z2 and so y, = Y2.
If A c R", B c R'are both affinesubspaces, amap f: A-.B is an effine map if
f(aa+µb) = t,f(a) + Rf(b) for a,b e A and k + µ = 1.
An affine map is therefore one that takes straight lines to straight lines because the straight line
through the points a, b is the set (aa+p.bjx,p. R, X+p=1). If a e A, b e B then A-a and B-b are
linear spaces; if L: A-a -. B-bis a linear map then the map AL: A-. B defined by
A1(x) = L(x-a) + b
is an affine map. When checking this note carefully that L is only defined on A - a. All affine maps
arise in this manner as we now show.
Lemma if f: A -. B is an affine map then the map L,: A-a -. B-f(a) defined by
4(x) = f(x+a) -f(a)
is a linear map. The map f is obtained from Lo by the previous construction.
Proof We need to check that 4(x+y) = Lax) + 4(y) and that L.(Xx) = kL,(x). To check the first
we note that x + a, y + a, a e A and that x + y + a = (x+a) + (y+a) - a is a combination of them,
the sum of whose coefficients is 1. So
Lr(x+y) = f(x+y+a)-f(a)
= f(x+a) + f(y+a) -f(a) -f(a)
= Lr(x) + 4(Y)
 9

L,(xx) = f(xx+a)-f(a)
= f(x (x+a)+(1-x)a) - f(a)
_ Xf(x+a)+(14)f(a)-f(a)
_ XL,(x).
If L = L,, it is easy to check that if one takes b = f(a) then f = AL.
An important special case is the following.
Corollary If f: R" -. R" is an affine map then there exists a e R" such that the map L: R" -. R"
defined by L(x) = f(x) -a is linear, so f(x) = L(x) + a.

Isometries of R"
We have already seen that translations T,: R" -. R" defined by T,(x) = x + a and orthogonal
transformations T: R" R" which are linear and satisfy Tx.Ty = x.y are both examples of isometries
of R''. We will show that all isometries are combinations of these two basic kinds. In fact isometries
are affine maps. The first step is
Theorem 2 An isometry f: R" -. R" is uniquely determined by the images fao, fa,,... is, of a
set ao, a,, ... a" of (n+ 1) (affinely) independent points.
Proof Let f, g be isometries with fa, = ga, for O < I gi n. Then g-'f is an isometry with g''fa, =a,.
Let T be the translation defined by Tx = x - ao and let b, = T(a,) for 0 < i s n. Clearly, bo =0,
and the set {b,, b2, ... , b"} forms a basis for R". We will show that h = Tg 'fT"' is the identity,
and this shows that f = gas required.
Clearly ht: = b, for 0 s i n, so if y = hx we have that d(x,0) =d(y,0) and d(x,b,) = d(y,b) for
1 i n because h is an isometry. Hence x.x = y.y and (x-b,).(x-b,) = (y-b,).(y-b,) for
1,i n. By expanding these last n equations and manipulating one gets that x.b, = y.b, for
1 i n. As b,, b2, ... , b" is a basis, one has x.z = y.z for every z e R", hence x = y, proving
that his the identity.
This proof shows that a point in R" is uniquely determined by its distances from n + 1 independent
points. Note that, in general, a point x is not uniquely determined by its distances from n
independent points.
Theorem 3 If (ao, a,.... a") and (bo, b,, ... b") are two sets of (n+1) independent points in
R' with d(b,,b) for 0 < i, j < n then there is an isometry f: R" -. R" with fa4 = b, for
0'i<n.
Proof Using translations we can clearly assume that ao = bo = 0. Then (a,, a2, ... a") and
(b,, b2, ... b") are bases for R", and it is easy to see that the hypotheses imply that a,.a, = b,.b,
for all i, j. Let g be the unique (non-singular) linear transformation such that ga, = b, for
1 <_ i _< n. Let x -y =1 ;a,, then gx - gy = g(x-y) = E x,b; by the linearity of g. So
d(gx,gy)2 = E x,xib,.bi = E x,ala,.ai = d(x,y)2.
Hence g is a linear isometry. The required f is the composition of g with a translation, so it is affine.
We have already proved on page 4 that every linear isometry is orthogonal. Theorems 2 and
3 therefore combine to show that if f is an isometry of R" then f(x) = Ax + a where A e 0(n) and
a e R^ so that every isometry is the composition of an orthogonal transformation and a translation.
Hence every isometry is affine.
Exercise If X c R" is any subset and g: X -. IR" is an isometric map, show that there is an isometry
f: R" - R" such that fIX is g. If the affine subspace defined by X has dimension n - r, prove that
the set of such isometries forms a coset of 0(r).
10

We will now show how every isometry can be written as a product of reflections. This gives an
alternative approach to understanding isometries and yields independent proofs of some of our
previous results.
Definition If H is a hyperplane in R", reflection in H is the isometry R of R" defined by
RH(x) = Y-z
where x = y + z with y E H and z-L(H-y) using the decomposition given on page 8.
Note that RH2 is the identity and that RH leaves every point of H fixed. If H is the perpendicular
bisector of ab, RH interchanges a and b.
Example In R3, regard H as a two-sided mirror, then RH(x) is the mirror image of x.

Exercise If 0 e.H, that is if H is a linear hyperplane, show that RH is orthogonal. If a is a unit vector
perpendicular to H, show that RH(x) = x-2(x.a)a.
Theorem 4 Any isometry f: R" -+ R" that is the identity on an affine (n-r)-dimensional subspace
A (that is, fa = a for each a e A) can be expressed as the product of at most r reflections in
hyperplanes that contain A. Any isometry can be expressed as the product of at most (n+1)
reflections.
Note The last sentence can be regarded as a special case of the first if one makes the (usual)
convention that the empty set has dimension-1.
Proof Choose (n--r+1) independent points so, a,, ... , an., in A and extend them to a set
ao, a,, ... , an of (n+1) independent points In R". Let b, = is,, so b, = a, for 0 -i E n-r. As f is
an isometry, d(a,,ai) = d(b,,bl) so If H is the perpendicular bisector of
a, i , n-r. The idea now is to consider RHf, this is the identity on an (n-r+1) dimensional
affine subspace and so one can use induction on r to give the required result. In detail:
RHf = RH, ... RH, where H, ... H, are s hyperplanes (s<-r-1) containing A and a,,...,,, then f =
RHRH, ... RH. is a product of at most r reflections in hyperplanes containing A. To prove the last
sentence of the theorem, let H be the hyperplane bisecting af(a) for some a E R". Then RHf fixes
a 0-dimensional affine subspace, and so is the product of at most n reflections. So f is the product
of at most n + 1 reflections.
 II

Corollary1 If f: R" R" is an isometrywith f(0) = 0 then f is orthogonal.

Proof The theorem proves that f is the product of at most n reflections in hyperplanes through
0. But such reflections are orthogonal, and hence so is their product.
Corollary 2 Any isometry f: R" R" is the composition of an orthogonal transformation and a
translation. In fact T_,o.f and f.T,,o are orthogonal.
Proof T_,o.f and f.T,,o both fix 0 so the result follows from Corollary 1.
We will denote the group of all Isometries of R" by I(R").
Exercise If a E R", the translation T. is defined by T,(x) = x + a. Show that the set of all
translations forms a normal subgroup of I(R") which is isomorphic to the additive group R" and
whose quotient is isomorphic to O(n).
A metric d can be defined on I(R") as follows:
Choose a set of (n+ 1) independent points ao, a...... an in R" and let
d(f,g) = max d(fa,,ga,) for f,g e I(R").
0-1
By definition, d(f,g) = 0 if and only if fa, = gal for 0 -- i <- n and by Theorem 2, this holds precisely
when f = g. The triangle inequality is easy to check. This metric has the further property of left
invariance:
d(hf,hg)=d(f,g) forallf,g,haI(R").
Exercise On the subset of I(R") consisting of all the translations, show that d gives the usual
metric on R.
Different sets of independent points give different metrics on I(R"). These metrics give rise to
the same topology on I(R"):
If a = (a a,, .... a") and b = (b0, b,, ... , b") give rise to metrics da and db respectively,
then da and db both define the same topology. By symmetry between a and b it is enough to prove
that for every e > 0 there is a 8 > 0 such that
db(f,g) < S implies that da(f,g) < E.
First write the a, vectors in terms of the b, vectors:
o>,;ibiwith isxq=1.
12

Choose M so that Ixi,l, M for all i, j and now choose S > 0 so that (n+1)MS < F. If db(f,g) < S
then {11b1-gb;ll < S for each i. As I, g are both affine maps one has that
fa;-ga, = E A1(fb.-gb).
So
llfa,-ga,ll- lk,il llfbr-gbUU<(n+1)MS<e.
The result follows.
Corollary 2 above implies the following
Theorem 5 The space I(R") is homeomorphic to O(n) x R.
Proof We define the two maps which are mutual inverses and leave the verification that they are
both continuous to the reader.
Given f e I(R"), let f: R" -. R" be defined by i(x) = f(x) - f(0), so that f = T_ro.f. Then T(0) = 0,
so that f is orthogonal. The map I(R") -. O(n) x R" is now defined by f (1,1(0)), and its inverse
by (T, a) -. I, where f(x) = T(x) + a = TaT(x).
Exercise Prove that the group I(R") is not isomorphic to the direct product of its subgroups O(n)
and R" (the subgroup of translations). [Hint: find f e O(n) and g t R" so that fg * gf.)
The mapping f - f: I(R") -. O(n) is useful because it can be used to distinguish between two
different types of isometries.
Definition An isometry f is called direct or opposite according as det f = +1 or -1 where
T (x) = f(x) -f(0) if is orthogonal, so that one knows det f = ± 1 for any isometry f).
Care Sometimes in the literature (especially in physics textbooks) these are called proper and
improper motions.
Lemma The mapI(R")-. (±1} defined byf-det fis a group homomorphism.
Proof It is clearly enough to show that the map If -. If is a group homomorphism because det is
a multiplicative map. We show that fg = fg.
f(g(x)) = i(g(x) -g(0))
= i(g(x)) -i(g(0)) by linearity off
= f(g(x)) - f(O) - (f(g(0)) -f(0))
= f(g(x))-f(g(0)) = f (x)
Exercise It is worthwhile to compare this 'sign' of an isometry with the sign of a permutation (see
page 28). Define a monomorphism is S" -. O(n) by permuting the axes of R. Show that det(i(a))
equals the sign of the permutation a. (It is enough to check this for transpositions.)
Exercise (A refinement of Theorem 2).
Show that an isometry of R" is uniquely determined by the images of n independent points
together with its sign.
 13

Isometries of R2
We will now study the isometnes of R2 in some detail. First, we have the examples, we list their
properties and the reader should verify them.
Direct
1. Translations T,(x) = x + a.
The set of all translations forms a normal subgroup of I(R2) and is isomorphic with R2. The
translation T. is the product of two reflections in lines t, m both perpendicular to a and distance
hall / 2 apart. Ta is an element of infinite order in I(R2).
2. Rotations R(a,a) is the rotation through angle a about the point a. The rotation R(O,a) has
matrix
a Note that fR(a,a)f-' =R(fa,a) for a direct isometry f and that
coisa,.

fR(a,a)f-' =R(fa,-a) for an opposite isometry f.

For a fixed a, the set of all rotations R(a,a) forms a subgroup SO(2)a of I(R2) which is isomorphic
to SO(2) (= SO(2)0), and, for different a, each of these subgroups are conjugate. The rotation R(a,a)
is the product of two reflections in lines through a with angle a/2 between them. R(a,a) has infinite
order when 217/a is irrational, and in this case it and its powers form a dense subgroup of SO(2).
When 2;r/a is rational, R(a,a) has finite order equal to the numerator of 217/a, and it and its powers
form a discrete subgroup of SO(2)a.
Opposite
3. Reflections R` is reflection in the line 41. The identity and R4 forma subgroup of order 2.
For any pair of lines l ,m in the plane there is an isometry f such that ft = m; also fRef ' = R,e
for any isometry f. Hence any two subgroups of I(R2) generated by a single reflection are conjugate.
4. Glldes(or glide reflections) A glide is of the form G(C,a) = R1 Ta where a is a vector parallel
to 6. Note that in this case T,Rt = RiTa.

I
p r
t
i
a

fq

The square of a glide is a translation, (RuTa)2 = Tea and so a glide has infinite order.
Exercise i) Show that the translations Ta, Tb are conjugate in 1(R2) if and only if Ilal = Ibll
ii) Show that the glides Rg T RmTb are conjugate in 1(R2) if and only if IIaI I = IlblI.
Any isometry of R2 is either the identity or of one of the above four types. To prove this we will
need the following result about compositions.
Lemma T,R, is a glide if the lined is not perpendicular to a and a reflection ift is perpendicular
to a.
14

Proof

p+a r+a

q q+a
m

4
P r r/

q/

Choose pet, let q = p + a/2 and let r be the foot of the perpendicular from p +a tot. Let m denote
the line parallel to 1 through q. We prove that Ta Rt = RmT,..,, by checking that these
transformations are the same on the three independent points p. q, r (the case where these are
not independent is tautological) and then using Theorem 2. We check
TeRsP=Tap=p+a, RmT,-pp =Rmr=p+a,
T.R,eq = Teq' = q' + a, RmT,-vq =Rm(q'+a) = q' + a,
T,R, r = Tar = r + a, RmT,y,r =Rmr' =r + a,
from the diagram.
There are several ways of deducing the following theorem from the results that have been proved
already, the reader is urged to find some alternative proofs for himself.
Theorem 6 Any isometry of R2 is the identity, a translation, a rotation, a reflection or a glide.
Proof Suppose first that f has a fixed point a. (A fixed point of f is a point a such that fa = a.)
Corollary 1 to Theorem 4 says that, if a is taken to be 0, f is orthogonal. Hence f is represented
by one of the matrices
or
Lcs na cosa J `Sins -cosaJ
according as f is direct or opposite. The first matrix represents a rotation with angle a and the
second represents a reflection in the line y = x tan(a/2).
Secondly, assume f has no fixed point and that fa = b. Let .1 be the perpendicular bisector of
ab, then R,fa = a. By the above R1f is either a rotation about a or a reflection in a line through
a. If R,f is reflection in the line m, f = RgRm andf, m cannot meet, otherwise f has a fixed point,
sot, m are parallel and f is a translation. The remaining case is when Ref = R(a,a). Clearly
a £ 1 otherwise a would be a fixed point of f. The rotation R(a,a) can be written as R.R. and m
can be chosen to be any line through a. Choose m to be parallel tot, then f = R'eRmR = TaRn,
where a is the vector perpendicular to l and m of length twice the distance between Z and m. By
the lemma f = TBR is a glide or a refection, and as it has no fixed point is must be a glide.
We can summarize all this in the following table which gives a classification of the non-identity
elements of I(R2)
 15

Fixed Point?
Yes No
Direct Rotation Translation
Opposite Reflection Glide

Isometries of R3
The main examples are:
1. Translations
2. Rotations Let a be a directed line in R3, R(fm) denotes rotation about s through angle a.
If ,e is the z-axis, then R(s,a) has matrix
cosa -sina 0
sins cosa 0
1 0 0 1

3. Screws A screw is the composition of a translation and a rotation. Let a be a vector parallel
to the lines, then a typical screw is T,R(s,a) = R((,a)Ta.

nnnnn

Lemma For any a screw unless a is perpendicular toe.

Proof Write a = a, + a2 where a, is parallel to Z and a2 is perpendicular toe. Then
T,R(e,a) = T0,T6,R(s,a). The transformation T,,R(e,a) is a rotation about an axis m that is parallel
toe (essentially, this is a two dimensional situation). Hence, T,R(e,a) is the composition of a
translation Ta, and a rotation about an axis parallel to a,. Hence T3R(e,a) is a screw unless
a, = 0.
4. Reflections If H is a plane in R3, then RH is reflection in H.
5. Glides If H is a plane in R3 and a is a vector parallel to H, then a typical glide is
T. RH = RHTa.
Lemma For any a, H, the composite RHTa is a glide unless a is perpendicular to H, in which case
it is a reflection.
Proof The cases when a is perpendicular to H and when a is parallel to H give a reflection and
a glide respectively.
16

In general, let a= a, + a2 where a,1H and a2 is parallel to H, then

RHT.,T,? = RH'T.?
where H' is a plane parallel to H. Hence this is a glide.
6. Rotatory reflection If H is a plane in R3 and Z is a line perpendicular to H, then a typical
rotatory reflection is
RHR(C,a) = R(G,a)RH.
An important special case is when a = n, this composite is called an Inversion 1., or a reflection
in the point a. If t n H = {0} then inversion is linear and its matrix is diagonal with three minus
ones on the diagonal. Inversion at the point a is the composition of the reflections in any three
mutually perpendicular planes meeting in a.
7. A rotatory inversion is the composition of a rotation about a linee and inversion in a point
a contained in. Every rotatory inversion can be written as a rotatory reflection and vice versa.
Let R(e,a)1, be a rotatory inversion. Write R(Ca) as RH,RH where H,, H2 are two planes meeting
in. at angle a/2. Write 1, as RH?RH,RH, where H3, H4 are planes through a such that H2, H3 and
H4 are mutually perpendicular. Then R((,a)1, = RH,RH3RH,. But H,, H3 meet in a line m that is
perpendicular to H4, hence R(C,a)I, = R(m,(3)RH., a rotatory reflection.
Now consider a rotatory reflection RHR(.,a) wherei is perpendicular to H. Write this as RHRH.RH,,
then H, and H2 are both perpendicular at the point a = H n H, n H2. Then RHR(C,a) =
RHRH,RH,RH,RH? = I,.RH?RH?. But RH,RH? is a rotation about an axis that contains a, hence
RHR((,a) is a rotatory inversion.
Lemma If e is a line not contained in the plane H then RHR(C,a) is a rotatory reflection.
Proof Let I = RHR(t,a), then I = RHRH,RH, where H, H,, H2 are planes meeting in a = Z n H
and H, H, are perpendicular. Now 1. =RH,RH,RH if H3 is chosen such that H, H,, H3 are mutually
perpendicular at a. So I = I$RH3RH, and RH,RH, is a rotation about an axis through a. Hence f is
a rotatory inversion and so by the previous discussion it is a rotatory reflection.
Theorem 7 Every isometry of R3 is one of the above six types.
Proof We use Theorem 4 to show that an arbitrary isometry I of R3 is the product of at most four
reflections and then analyse each case.
The cases where I is the product of zero or one reflection are trivial.
When f is the product of two reflections, f = RH,RH,, there are two cases: H,, H2 meet or H,,
H2 are parallel. In the first case I is a rotation about the line where H,, H2 meet and in the second
case f is a translation perpendicular to the planes through twice the distance between them.
When I is the product of three reflections, by the previous case I is RHT. or RHR(C,a). By the
lemma on glides, RHT. is always a glide. By the lemma on rotatory reflections, RHR(C,a) is a rotatory
reflection.
When f is the product of four reflections there are two cases. First, if f has a fixed point, then,
by Theorem 4, t is a product of at most three reflections, hence I is a translation or a rotation.
Secondly, if f has no fixed point one can find a so that T.f has a fixed point (choose any x and let
a =-f(x)). Hence I is the composite of two translations, hence it is a translation or the composite
of a translation and a rotation T.R(e,a). Hence by the lemma on screws it is a screw.
Corollary A direct isometry of R3 that has a fixed point has a fixed line. This is often paraphrased
as "every rotation has an axis".
 17

We will now study the topology of t(R2) and the way that the four kinds of isometries define
subsets of it. The description will be in terms of certain subsets of R3, and we digress to define
them.

Some subsets of R3
We describe and name certain subsets of R3. Any two homeomorphic spaces will always be
given the same name.
First a subspace of R2, the circle

S' = {xER2lkxlj=1}.

1.
0
The torus = {x a R3. (r-2)2 + z2 = 1 where r2 = x2 +y2}.
This is the surface obtained by revolving the circle (x-2)2 + z2 = 1 about the z-axis.

Iz

Revolve

Exercise Show that S' x S' is homeomorphic to the torus. The map is
(e'°,e1')--* ((2+cosip)cose,(2+cosip)sinO.sincp).
The inverse map is
(x,Y,z) -+ ((x+iy)/Ix+iyl,lx+iyi-2+iz)
Check that these maps are in fact inverses and note that they are both continuous.
2. Thesolid torus = {x a R3 I (r-2)2 + z2 _ 1 }
This is the torus together with its "inside", and is homeomorphic to S' x D2, where
D2= {xeR2 1 IIIcI1}.
18

The open solid totes is the "inside" of the solid torus, that is
{xE R31 (r-2)2±Z2<1)
and is homeomorphic to S' x R2 because R2 is homeomorphic to the open disc
b2= (xER21l!xlk<1).
3. The MSbius band M is the subset of the solid torus defined by
x- r cosO, y= r sine,

As we consider planes rotating around the z-axis - the angle 0 increasing from 0 to 2-,7, the M6bius
band meets these planes in line segments of length 2 centred on the circle x2 + y2 = 2, z = 0
and the slope of these segments increases from 0 to IT. If we cut this MObius band along the line
segment where 0 = 0 we get its familiar representation:
C
A B

A It

B A
D

This notation means: take a rectangle and identify its opposite edges with a twist. More formally:
Consider the set (-1,1 ) x (-1,1) and the equivalence relation
(x,y) - (u,v) a x = u and y = v.
orx= 1,u=-1 andy=-v.
Then the space we want is [-1,1) x [-1,1)! - with the metric d([x),[y]) = min {d(x.y)lx E (x), y E [y)) .

The open MObius band is M n open solid torus.

Exercise Consider a MObius band in R3 and "thicken it". Show that the resulting solid is
homeomorphic to a solid torus.
Later we will need the following description of the open MObius band.
Lemma The space S' x R' /(x,y) - (-x,-y) is homeomorphic to the open MObius band.
 19

Proof The metric on the quotient space S' x R'/(x,y) (-x,-y) is given as above.
We cut S' x R' along 1 x R' to give [0,2n] x R'.
Note that the point -x a S' becomes x + it in [0,2n].

c d'
a
b
R1
b'
a'
d c
0 7r 27r

Under this equivalence relation, every point in the left hand half is identified with some point in the
right hand half (a with a' etc). Hence the quotient space is obtained from the left hand half by
identifying points on its vertical edges appropriately.
Before we go on to study the topology of the space I(R2), we prove an interesting result about
the space of all lines in R2. For ease of notation it is convenient to use complex numbers to
represent points in R2. Every line in R2 is given by an equation of the form
AZ+A2+B=0withAeC\ (0) and BE R.
The line C x + my + n = 0 corresponds to 2A = t - im and B = n.
The equation xAz + xAi + XB = 0 determines the same geometric line (for a e R \ (0)). By a
suitable choice of >, the equation can therefore be chosen to be Az + Al + B = 0 with JAI = 1
and B e R. The only remaining ambiguity is that (A,B) and (-A,-B) determine the same line. Hence
the set of all lines in R2 can be identified with
S' x R/(A,B) --(-A,-B).
This gives the set of lines a metric and we have
Theorem 8 The space of all lines in R2 is homeomorphic to the open M6bius band.
Exercise Show that the two lines represented by (A,,B,) and (A2,B2) are parallel if and only if
A, = ±A2. Show that the distance between the parallel lines represented by (A,B) and (A,C) is
18-Cl.
Exercise Show that the space of all directed lines in R2 is homeomorphic to S' x R.
Corollary The subspace of I(R2) consisting of reflections is homeomorphic to the open M6bius
band.
In Theorem 5 is was proved that 1(R2) is homeomorphic to 0(2) x R2. The group 0(2) is
homeomorphic to the union of two disjoint circles. Hence 1(R2) is homeomorphic to the union of
two disjoint open ton, one consisting of the direct isometries and the other of the opposite
isometries. These are the components of 1(R2).
To examine this space in detail we again use the complex numbers. A translation is
20

z --. z : b and a rotation about 0 is z - az with ;al = 1. The direct component consists of all
transformations of the form z -. az + b with dal = 1 and so is homeomorphic to S' x R2 (compare
Theorem 5). The subset 1 x R2 consists of translations and its complement consists of rotations.
The centre of z az + b is at b(1 -a)-' and it is a rotation through angle arg a. If the centre tends
to infinity in such a way that b is constant we see that a translation is the limit of rotations. Of course
this is clear in the diagram of S' x R2.

Now we consider the other component which is made up of all the opposite isometries. This
component is a coset of the subgroup of direct isometries. The transformation z -+: is a reflection.
and so this coset is made up of all the transformations of the form z -. a2 + b with jai = 1. Again
(a,b) E S' x R2 gives an explicit homeomorphism between the component and S' x R2. The
reflections can be characterized as the opposite transformations of order 2 and so correspond to
the set of (a,b) satisfying ab + b = 0. By Theorem 8 this subset is a homeomorphic to the MBbius
band, in fact it sits inside S' x R2 in the way described on page 18 because (e",b) satisfies
ef° b + b = 0 if and only if 2arg b =(8±r).
Exercise The transformation f: z -* of + b leaves the set F,,b = {z a Ciz = a2 + b} fixed. For
most values of a, b with Jai = 1 the set F,,b is empty, in this case f is a glide. By subsituting
z = az +b into z = aaz + b show that if F.,b is not empty then a5 + b = 0. Conversely, if
ab + b = 0 show that F.,b is non empty by checking that a + b/2 a Feb ,where a2 = a. In the case
F,,, is non-empty, F..b is the line Az + A2 + B = 0 where A = p, B = -jab and jig = -tea.
The complement of this Mbbius band is the set of glides. A glide is given by a reflection and a
translation. The translation is through the vector (a5+b)/2 and the reflection is
z-,a2+(b-ab)/2.
As the length of the glide tends to zero the glide tends to a reflection, and this can be seen in the
diagam of S' x R2. The space of all glides Is itself homeomorphic to an open torus. To see this
cut S' x R2 along 1 x R2 and remove the open MObius band that is inside S' x R2. We get the
following object.
 21

The 'twisted' line segment has been removed in each vertical section of the cylinder and the two
half R2's labelled A are to be identified one with the other as are the ones labelled B, as follows.

Identified

The resulting space is easily seen to be homeomorphic to the open solid torus.
Exercise Think of this S' x R2 as being homeomorphic to a MObius band made out of thick
cardboard. The process described above is then equivalent to cutting the MObius band down the
middle. This leaves a solid that is homeomorphic to a thick annulus and this is an open solid torus.

Finite Groups of Isometrles

If X c R" contains (n+ 1) independent points its symmetry group S(X) is the subgroup of I(R")
consisting of those isometries f such that fX = X. It is also the group of isometries of X (remember
the exercise after Theorem 3). One may need to consider subsets X of R" that do not contain (n+ 1)
independent points, there are two alternatives: the first is to consider isometrics of R" that send
X to itself but there may be many of these that are the identity on X, the second alternative is to
regard X as a subset of the affine space A spanned by it and considering the isometrics of A that
send X to itself.
Example 1 If C is a line in R2, it does not contain 3 independent points and the subgroup of 1(R2)
that preserves E contains a normal subgroup with two elements whose quotient is I(R' ).
22

Example 2 If X c R2 iS {(x,0)Ix a R} u ((0,y) j-1 < y < 1)

then S(X) = Z12 x Z/2, the three non-trivial elements being R,,, Ry, R(0,n). (Here R. denotes
reflection in the x-axis.)
Example 3 It P is a regular n-gon with centre 0 (say). is the dihedral group D. having
2n elements. Any f e keeps 0 fixed, hence c 0(2). It contains the elements
S = R(0,21r/n) and T = Rtwhere C is a line containing 0 and a vertex of P.

n= 6

The element S generates a normal cyclic subgroup {S} of order n and T has order 2. The
transformation S is direct and T is opposite, in fact {S} = n SO(2) and so has index 2 in
S(P ). So is made up of (S) and its coset T(S). All the elements of T(S)are reflections in
lines through 0 and either through a vertex or through the mid point of an edge of P. The group
S and T with the relations Sn = 1, r = 1, TST = S-'.
The dihedral groups in effect describe all the finite subgroups of I(R2):
Theorem g Every finite subgroup of I(R2) is either cyclic or dihedral.
Proof Let G c I(R2) be a subgroup of order n. Choose any point a e R2. Its orbit under the action
of G, Orb(a) = {gajg a G), is a finite subset of R2. Let c = (aFb ga)rnbe the centroid of the
set Orb(a). As the centroid c of a finite set X c R2 is a particular affine combination of the points
of X and an isometry is an affine map, the point fc is the same affine combination of the points
of fX, hence fc is the centroid of fX. If g e G, then gOrb(a) = Orb(a), hence gc = c for each
g e G. This proves that G is a subgroup of O(2)c the group of orthogonal transformations centred
at c. First consider the direct subgroup of G, Go = G r% SO(2). This subgroup consists of a finite
number of rotations. If S is the rotation with least angle amongst those of Gd, it is easy to see that
 23

Gd is the cyclic group generated by S. If G = Gd one is finished; otherwise Gd has index 2 in G,

because SO(2) has index 2 in 0(2). The set 0(2) \ SO(2) consists entirely of reflections, suppose
T = R, is a reflection in G: then G is generated by S and T, and it is easy to check that TST
S-' (this holds for any rotation S and reflection T). Hence thegroup G is dihedral.
This theorem shows that there is an intimate connection between the finite isometry groups of
the plane and regular figures in the plane. We will now proceed to prove the corresponding result
for R3; but first here is an alternative to the centroid argument in the above proof.
Exercise Let G be a subgroup of 1(R2). Show that G is infinite if it contains both a non-trivial
rotation about a point a and a reflection in a line e such that a f e (you should use the Lemma on
page 13). Hence show that G is infinite if it contains reflections in three non-concurrent lines.
Deduce that if G is finite then there is a point fixed under all the elements of G.

The Platonic Solids

We start with some definitions. A subset X of R" is convex if for every pair of points x, y e X
the line segment joining them lies entirely within X, that is, for each t with 0 _- t _- 1, the point
tx + (1-t)y is in X. That is, X is closed under taking non-negative affine combinations. A subset
of R" defined by a finite set of linear inequalities is clearly convex and is called a convex
polyhedron. (A finite union of convex polyhedra is called a polyhedron.) So a convex polyhedron
X in R3 is bounded by planes, a two-dimensional subset of X that is in such a plane is called a
face of X. the intersection of two faces an edge and the intersection of two edges a vertex. (We
will always assume that a convex polyhedron X has nonempty interior thus ensuring that X is really
'solid' and does not lie in any affine hyperplane of R3.) A face is homeomorphic to a closed disc,
an edge to a closed interval and a vertex is a point.

Faces: ABC, ACHI, ABDJI, BCFE, BDE,

CFGH, DEFGJ, GHIJ
Edges: AB, AC, Al, BC, BD, BE,CF, CH, DE,
DJ, EF, FG, GH, GJ, HI, IJ

A polyhedron is called regular if all its faces, edges and vertices are identical to each other. By
saying that two vertices are identical one means that there are the same number of edges at each
vertex and that the angles between them are the same. A convex regular polyhedron is also called
a Platonic solid because they were studied by the Platonic school and played a role in their
philosophy.
Exercise The definition of a Platonic solid given here uses "overkill". Find how many of the
assumptions can be dropped because they are consequences of others.
A Platonic solid is described by the number of its faces; 'hedron' means 'seat'in Greek and tetra
etc. are the Greek words for four, ... :
24

Tetrahedron

4 vertices each with 3 edges at rr/3

6 edges
4 faces each an equilateral triangle

Cube (or Hexahedron)

8 vertices each with 3 edges at 7d2

12edges
6 faces each a square

Octahedron

6 vertices each with 4 edges at n/3

12 edges
8 faces each an equilateral triangle
 25

Dodecahedron

20 vertices each with 3 edges at 3ni5

30 edges
12 faces each a regular pentagon

Icosahedron

12 vertices each with 5 edges at ir13

30 edges
20 faces each an equilateral triangle

The first three are simple enough to understand and models of them are easily constructed. I
will explain briefly how to construct the other two.
The icosahedron is obtained by first taking a pyramid on a regular pentagon A.
26

Take five more equilateral triangles B one on each edge of the pentagon A, interspace these with
five more such triangles C, forming a drum.

Finally complete the icosahedron by adding another pyramid D to the base of the drum. By
construction, all the faces will be equilateral triangles and one can deduce that the resulting object
is Platonic.

The dodecahedron is obtained by taking a regular pentagon A and adding to it 5 regular

pentagons B, as shown, to form a upside down bowl. If one takes another such bowl, the two fit
together to form a dodecahedron.
 27

Exercise Construct models of these solids using cardboard and sticky tape (or rods or ... ).

Duality
If one takes a Platonic solid, joins the mid points of adjacent faces by a new edge and fills in
the resulting solid, one obtains the dual solid which is again Platonic. If this process is done twice
it is easy to see that one recovers the original Platonic solid (or at least a smaller version of it).
The tetrahedron is self dual, the cube and octahedron are dual, as are the dodecahedron and
icosahedron. If one tabulates the number of vertices, edges and faces one sees the duality in the
table:
Vertices Edges Faces
Tetrahedron 4 6 4
Cube 8 12 6
Octahedron 6 12 8
Dodecahedron 20 30 12
Icosahedron 12 30 20

Note that Euler's formula V - E + F = 2 holds in all five cases. This formula holds for any polyhedron
that is homeomorphic to the 3-dimensional disc D3 = {x a R3 I IIxi < 1) so in particular it is true
for convex polyhedra. (Algebraic topology gives generalisations of this formula to polyhedra in R".)
As was known to the ancient Greeks these five are the only Platonic solids:
Theorem 10 There are precisely five Platonic solids.
Proof We have already constructed the five so we need only prove that there are no more.
Suppose that r faces meet at each vertex and that each face is a regular n-gon. It is clear that
each of r and n is at least 3. The sum of the angles at a vertex is less than 2a, and each angle
is (n-2)irln, being the angle of a regular n-gon. Hence we see that
(r(n-2)ir)/n < 21r
from which one deduces that
(r-2)(n-2) <4.
28

It is easy to see that any integral solution of this inequality with r, n -- 3 must have r or n = 3, so
the only solutions are
(r, n) = (3,3), (3,4), (3,5), (4,3) and (5,3).
If one knows the shape of a face (that is n) and how many faces meet at a vertex (that is r) then
there is only one possible solid with that particular (r,n).

The Symmetry Groups of the Platonic Solids.

If X is a Platonic solid whose centre is at 0 its symmetry group is
S(X) _ {f a O(3) I fX = X}
and its rotation group (or direct symmetry group) is
SCI(X) _ (f 4E SO(3) I fX = X).
The rotation group Sd(X) is a normal subgroup of index 2 in S(X). (Of course, for some subsets
X c R3 one may have S(X) = Sd(X) but this is notthe case if X is Platonic.)
We will use the following (standard) notation:
S is the symmetric group on n letters i.e. Bij({ 1,2, ..., n}). This group has n! elements called
permutations. An element of S. that merely interchanges two letters (say i and j) and leaves the
others fixed is called a transposition and is denoted by (ij) in the cycle notation. It is standard and
easy to show that every element of S can be written as a product of transpositions.
Exercise Show that any a E Sn can be written as a product involving only the (n-1) transpositions
(1,2),(2,3),. . ., ((n-1)n).
The group S, has an important (normal) subgroup of index 2 called the alternating group A.
It consists of those permutations that can be written as a product of an even number of
transpositions. It is not quite obvious that this is a well defined condition on a permutation (and
hence that the set of all such permutations forms a subgroup of Sr). Perhaps the easiest way to
see this is to introduce variables x,, x2, ..., xn and make a e Sn act on the polynomial ring
Z[x1, x2, ..., xn) by a(xi) = x,pi then
A. = { or e S1Iaf = f where f = n (x-x1)).
As an alternative one has:
Exercise Define sign: by
sign (a) =n
Verify that i)
sign (transposition) = -1
ii) sign is a homomorphism.
Deduce that An = Ker(sign).
Note We will now assume that the reader is quite familiar with the above facts about
permutations, he should therefore make sure that he understands and can prove them.

Let T denote the regular tetrahedron, we have S(T) a. S4 and Sd(T) a A4:
By labelling the vertices of T with the numbers 1, 2, 3, 4 and because any f E S(T) must send
a vertex to a vertex one obtains a homomorphism a: S(T) -. S4.
 29

First, we show that a is onto. Because S4 is generated by transpositions, it is enough to show

that any transposition is in the image of a. We show that (12) is in the image of a, the proof for
the others is identical. Consider the plane H through the vertices 3, 4 and through the mid point
m of the edge 12, H is the perpendicular bisector of 12, hence the reflection R is such that
aRH = (12).

The homomorphism a is also injective because, by Theorem 2. any isometry of R3 is uniquely

determined by the images of four affinely independent points (i.e. four non coplanar points). Hence
a is an isomorphism.
It is now apparent, because a reflection in a plane is mapped by a to a transposition, that every
direct isometry of T is mapped to an even permutation of (1,2,3,4) , that is, the image of Sd(X) under
a is contained in A4. As Sd(X) has index 2 in S(X) and A4 has index 2 in S4, a induces an
isomorphism: Sa(X) --. A4.
The other Platonic solids have a common feature - they are all centrally symmetric, that is the
map J: R3 -. R3 given by Jx = -x is a symmetry. This map J, often called central Inversion,
commutes with every linear transformation of R3. The group 0(3) is isomorphic (as a group) to
SO(3) x (t 1 ),the map being
A -(A,1) if det A = 1
A-. (JA,-l) rfdetA=-1 (detJA = 1 inthis caseJ,
As J commutes with any A, this map is a homomorphism and its inverse is given by
(A,1) -+ A. (A,-1) -. JA.
If the solid X is centrally symmetric, it is now easy to see that S(X) = Sd(X) x {± 1).
Another feature comes from duality. If X and Y are dual, then any symmetry of X is a symmetry
of Y and vice versa, so that we have an isomorphism S(X) 21 S(Y).
30

It is clear from these remarks on central symmetry and duality that if one determines the rotation
groups Sd(X) of the cube and the dodecahedron then all other symmetry groups are easy to find
from these.

Let C denote the cube, then its rotation group Sd(C) is isomorphic to S4. To show this one must
find four geometric objects in C that are permuted by the rotations of C. The four objects that we
will use are the four diagonals of the cube, labelling them 1, 2,3,4.
D

Diagonal 1: AG
2: CE
3: DF
4: BH
 31

Consider the map a: Sd(C) -. S4 obtained in this way. To show a is onto it is enough to find a
rotation f of C such that a(f) = (12), the other cases are identical. Consider the plane containing
the diagonals 1 and 2. It contains two edges of C, namely AE and CG; let . be the line joining the
mid points of these edges. If f denotes the rotation through angle A about t., one has that fA = E.
fC = G, fB = H and fD = F, and so f = identity. It is now clear that of = (12) as required.
To prove that a is injective we will use the "pigeon-hole principle" (any map between two finite
sets with the same number of elements is a bijection if it is onto). The group S4 has 4! = 24 elements
and we will now check that Sd(C) also has 24 elements. For any pair of vertices P, Q of the cube
C, there is a rotation of C that sends P to Q (one can write down such a rotation for each case)
and there are precisely three rotations that fix a given vertex. Hence there are 24 elements in Sd(C).
In the language of groups acting on sets, the group Sd(C) acts on the cube, the set of 8 vertices
is an orbit and the stabilizer of a vertex has 3 elements.
Exercise Prove that the points on the surface of a cube have orbits with either 24, 12, 8 or 6
elements under the action of Sd(C). Describe geometrically the points having these different orbit
sizes.

Let D denote the docahedron. We show that its rotation group Sd(D) is isomorphic to A5, a group
which has 60 elements. To do this one must find five geometric objects associated with the
dodecahedron such that they are permuted by the elements of Sd(D). Five such objects are the
cubes that are inscribed in the dodecahedron. The edges of these cubes are diagonals of the faces
of the dodecahedron. Here are two descriptions of these cubes.
First description
Remember that a diagonal of a regular pentagon, with edge length 1, has length
T = (1 +f)/2 = 2cos(>r/5).
The number ' is called the golden ratio, and is the positive root of x2 - x - 1 = 0. One way of
proving this is to consider the diagram.

Clearly AX = CX = AB = 1 and
AC=EC=x so EX=x -1. By
similar triangles AX/AC = EX/ED
which gives the result:
1/x = (x-1)/1 sox2 = 1 + X.

Now consider a cube whose edge length is r. On a face of the cube we construct a 'tent' as
follows: two faces are isosceles triangles with sides r, 1, 1 and the other faces are trapezoids with
32

sides T, 1, 1, 1 as shown. The trapezoid and triangle make angles a, p respectively with the base.
We will show that a + p = rr/2.
Let C be the height of the triangle and h the height of the tent. These are given by the equations
1 = e 2 + (7/2)2, e 2 = h2 + ((T-1 )/2)2.
Substituting gives us 1 = h2 + r2/2 - T/2 + 1/4. But T2 - T = 1, so h = 1/2. We now deduce that
tana = 2h/-r = 1 /T and tang = 2h/(T-1) = 1/(7-1). So tana tang = 1, and therefore a + 7r/2.
This shows that when we put two tents on adjacent faces of the cube as follows:

the pentagon ABCDE is planar.

Exercise Complete this argument to give another description of the dodecahedron.
In this way we have inscribed a cube of edge length r in a dodecahedron of edge length 1 in
such a way that each edge of the cube is a diagonal of a face of the dodecahedron. By starting
with each of the five diagonals of a face of the dodecahedron one can obtain five different cubes
inscribed in the dodecahedron. These are obtained from each other by applying 1, R, R2, R3, A4,
where R denotes rotation through angle 27r/5 about an axis perpendicular to a face of the
dodecahedron. Each cube contains two tetrahedra:
 33

If one considers one of these and its images under the group (R) one easily sees that the images
of the four vertices are all different and so are all the 20 vertices of the dodecahedron. This leads
to the
Second description
Consider the dodecahedron with one of its faces horizontal. Its vertices are then divided into fo
sets of five, each set at a different level. First label the vertices of the top face cyclically by 1, 2,
3, 4, 5. Secondly, label the vertices of the next level. Consider a particular vertex P at this second
level, P is joined by an edge to one vertex, say, 5 of the top face and by a chain of two edges to
two other vertices 1, 4 of the top face;

P is given one of the other labels, that is, 2 or 3 (there is a choice here but nowhere else, the two
choices essentially differ only by a reflection). The other vertices of this level are then labelled
cyclically in the same direction as the labelling of the top face. So far, the top face has been labelled
completely and each of the five faces touching it has four of its vertices labelled. It is easy to check
34

that for each of these five faces the four labels already assigned are different. The vertices at the
next level are now labelled so that their five faces have vertices all with different labels. Finally one
must label the vertices of the bottom face; each face touching it has three vertices already labelled:

The vertex marked X cannot be labelled with 1, 2,3 or 5 otherwise a face would have two vertices
with the same label, so it must be labelled 4. This procedure works to give a complete labelling.
Inspection shows that the four vertices with the same label form the vertices of a tetrahedron. Thus
one has a system of five tetrahedra inscribed in D. There is another system of five tetrahedra
inscribed in D obtained from the constructed system by a reflection. Alternatively, they would have
been constructed if one had made the other choice at the second stage of the labelling procedure.
Exercise Colour the faces of your model of the icosahedron with live colours corresponding to
the (dual of the) above procedure.
The twelve faces are labelled cyclically by (12345) and all the other 5-cycles obtained from
(12345) by even permutations. The cyclic labels of opposite faces are mutual inverses. These
twelve 5-cycles in S5 correspond under a to rotations through ±277/5 about the centre of the
corresponding face. The other twelve 5-cycles in S5 (obtained from (12345) by odd permutations)
correspond to rotations through ±477/5 about the centre of the faces.
Note It is a familiar fact that two permutations are conjugate in S if and only if their expressions
as the product of disjoint cycles involve cycles of the same length. This is not quite true for An: cycles
of length n (for n odd) are in two distinct conjugacy classes. This fact is clear in Sd(D) - rotations
through ±277/5 are not conjugate to rotations through ±477/5.
Exercise Check all these statements and use them to prove that 0(3) has no subgroup
isomorphic to S5. Find a subgroup of 0(4) that is isomorphic to S5.
To continue with the proof, one can check that the various rotations through ±277/3 about the
vertices give each of the (twenty) 3-cycles in As, and that the rotations through 77 about the mid
points of the edges give the (fifteen) elements of order 2 in As (such as (12)(34)).
We have proved that a: Sd(D) - As is onto. It is easy to check that Sd(D) has 60 elements and
so a is an isomorphism.
Exercise Show directly that there are elements of orders 3 and 5 in the image of a and that there
 35

is a subgroup of order 4 in the image of a. Deduce that a: Sd(D) - AS is onto.

Summarizing we have the following.
Table of Symmetry Groups
Direct Full
Tetrahedron A4 S4
Cube, Octahedron S4 S4 X W)
Dodecahedron, Icosahedron A3 A5 X (t')
Exercise Prove directly that the number of elements in the rotation group of a Platonic solid is
twice the number of its edges.

Finite Groups of Rotations of R3

Now we will prove the analogue of Theorem 9 in the case of R3. We need the following result.
(Remember that SO(n) is the group of orthogonal n x n matrices with determinant 1.)
Lemma If At SO(n) and n is odd, then 1 is an eigenvalue of A.
Proof We show that det(A-1) is singular.
det(A-I) = det(A-1).detA' = det(AA'-A')
= det(I-A') = det(I-A)' = det(I-A) = (-1)"det(A-I).
But n is odd so det(A-I) = 0.
Note This lemma is not true for all A E O(n), for example A= -I.
Corollary1 "Every rotation has an axis" - that is, if At SO(3) then there is an orthogonal matrix
P such that
Cost -Sins 0
P-'AP = sine coca 0
0 0 1

Proof By the lemma, there is an e3, with Ile3il = 1, such that Ae3 = e3. By using the Gram-Schmidt
process one obtains an orthogonal basis e1, e2, e3 for R3. Clearly if V is the subspace spanned
by e, and e2, then V i invariant under A, that is AV= V and ASV is in SO(2). The matrix of AjV
Cost -Sins
with respect to e1, e2 isLsina
cost
In the proof of the next theorem we will explicitly need
Corollary 2 If A E SO(3), A * I then {x: Ax = x) (the eigenspace with eigenvalue 1) is always
1-dimensional.
Theorem 11 Let G c SO(3) be a finite subgroup, then G is either cyclic, dihedral or the direct
symmetry group of a Platonic solid.
Proof The group SO(3) preserves distances and preserves 0, hence it acts on
S2=(xER3IlIxlj=1).
By Corollary 2 above, for each g * 1 in G, there is a unique set (x,-x} c S2 such that gx = x (and
so g(-x) = -x), we call such an x a pole of g (x and -x are the "north and south poles" of the rotation
g).
Suppose }GI = n, and for each x E S2, let m denote the number of elements in the subgroup
Stab(x) _ (g e G (gx = x),
that is, the number of elements (including the identity) of G that have x as a pole. Note that Stab(x)
is a finite subgroup of SO(2) and so by Theorem 9 (and its proof) it is cyclic. The point x E S2 is
a pole of some element g c G if and only if m > 1.
36

We will now study the orbit of a pole p under the action of G, that is
Orb(p)=
It is a standard fact about groups acting on sets that)Orb(p)l =n/mp. The set Orb(p) consists entirely
of poles because if hp = p then ghg-1(gp) = gp; and this also shows that mp = m9p , for each

As usual when studying orbits of actions of finite groups, one counts elements: for each pair of
poles ) p,-p) we have (mp 1) elements g e G \ (1) for which they are poles, that is gp = p. For
each orbit of poles we therefore have n(mP 1)/mp elements of G \ (11 ), but each element of G \ {1 }
determines two poles p and -p, so the number of elements in G \ {1 ) is
'/2 £ n(mp 1)/mp
where we have one summand for each orbit. Therefore one has the equation
2(n-1) = Z n(mp-1)/mp
or 2(1-1/n)= E1-1/mp) ')
To avoid the trivial group, one assumes that n 3 2. By definition of a pole, mp : 2. For each pole
p one has 1 > (1-1 /mp) , 1/2. Also 2(1-1 /n) -- 1. From this it is easy to deduce that there are either
2 or 3 orbits and we treat these cases separately.
2 orbits: Let the orbits have n/m, and n/m2 elements. Equation(*) gives 2/n = 1/m, + 1/m2 and
it follows that n = m, = m2. In this case, therefore there are two orbits with exactly one pole in
each. These must be fixed under every element of G and so one sees that G is cyclic, in fact this
case is essentially two dimensional.
3 orbits: This is the full 3-dimensional situation. Equation(*) becomes
1 + 2/n = 1/M1 + 1/M2 + 1/M3 ', )
and we assume m, > m2 3 m3. It is clearly impossible to have m, 3 3 for each i for then the right
hand side of (',) could not exceed 1. So one has m3 = 2.One therefore gets
1 /2 + 2/n = 1/m, + 1/m z *2).
Similarly, it is impossible to have m, , m2 -- 4, so m2 = 2 or 3. It is simpler to consider these two
cases separately.
m2 = 2 In this case equation CO becomes n = 2m,. Thus two of the orbits have mp = 2 and
have m, poles in them whilst the third orbit has mp = m, and has 2 poles in it. An easy analysis
shows that this is the dihedral case, the 2m, poles being perpendicular to the other two poles. The
2m, = n poles are arranged in orbits, illustrated here for n = 6 and 8.

O
n=6 n=8
 37

M2=3 In this case equation(* 2) becomes

1/6 + 2/n = 1/m,.
Clearly one has 3 < m, < 6 and (6-m, )n = 12m,. The only solutions are
(m,,n) = (3,12), (4,24) and (5,60).
These numbers correspond to the Platonic solids. A full analysis (which we omit) shows that each
of these cases can only arise from the symmetry group of a Platonic solid. The various orbits of
poles correspond to the centres of faces, to the mid points of edges and to the vertices of the
Platonic solid.
Note There is a difference between the dihedral group in R2 and the dihedral group of rotations
in R3 although the groups are abstractly isomorphic.
An interesting application of these kinds of ideas is to the study of crystals.

Crystals
In the mathematical theory of crystals it is assumed that the crystals are of infinite extent, an
important concept in their theory is therefore that of a lattice in R" (of course the main physical
interest is in the case n = 3).
A lattice L in R" is a subgroup of R" that is discrete as a topological subspace of R", and whose
elements span R". The discrete condition means that for each x E L there is an e > 0 such that
d(x,y) , e for ally F. L with y / x, or equivalently that L has no accummulation point.
The following exercise should help to make this concept familiar.
Exercise i) Show that any line through two points of a lattice L contains infinitely many points
of L.
ii) Show that L is isomorphic as a group to the subgroup of translations in its symmetry
group S(L).
iii) If L is a lattice in C (25 R) such that L (L is self conjugate), prove that there
is a basis for L whose vectors are two edges either of a rectangle whose edges are vertical and
horizontal or of a rhombus whose diagonals are vertical and horizontal.
An alternative definition of a lattice L c R" is that it is the set of integer combinations of a basis
of Rn.

L= (E rie,1r,EZ).
Therefore L is isomorphic, as a group, to Z. This definition is technically easier to use in proofs
but, in practice, it is often easier to check that something is a discrete subgroup.
Given a discrete subgroup L of R", one finds a basis for it as follows. Choose a non-zero vector
e, in L that is closest to the origin. Having chosen e,, e2, ..., ek choose ek+, to be a vector in L
not in Span(e,, e2, ..., ek) but closest to it. Some checking has to be done to see that L is the
set of integer combinations of the vectors e,, e2, ..., a". The other main thing to check is that there
is a closest vector at each stage.
An important restriction on the type of elements that can be in the symmetry group of a lattice
is given by the following result - often called the 'crystallographic restriction'.
Theorem 12 If L c R" is a lattice with n = 2 or 3 and R E S(L) is a rotation of order m, then
m= 2,3,4 or 6.
Proof We give three proofs. The first, for n = 2, is simple. The result for n = 3 can be deduced
from it by applying the fact (Corollary I on page 35) that any rotation of R3 is essentially a rotation
of R2. The other two proofs are only sketched and are of importance because their ideas can be
used to give corresponding results for dimensions n > 3.
38

1. Consider all the rotations in S(L). Their centres forma set M c R2. The set M contains L because
the rotation through n about any lattice point is in S(L). The set M is discrete: Let p be the centre
of a rotation R in S(L) and p k L and let q be a point of L closest to p, then p is the centroid of the
orbit of q under the cyclic group (R). Therefore p is not arbitrarily close to elements of L, hence
M is discrete.
Let R(p,21T/m) be the given rotation. Suppose p, is a point such that R(p,, 21T/m) a S(L) and such
that d(p,p,) is as small as possible. If m # 2, 3, 4, 6 we will show that there must be such a point
closer to p than p,. If S e S(L), then SR(p,2rr/m)S-' = R(Sp,2n/m) a S(L), so if
P2 = R(p,,27T/m)p we have that P2 is the centre of a rotation through 2n/m which is a symmetry
of L. But d(p2,p) = 2sin(n/m)d(p,p,) and so d(p,p2) < d(p,p,) if m > 6.

per- 2n
Pt

m>6

P2

A similar argument shows that p3 = R(p2,2ir/m)p, is such that R(p3,2n/m) a S(L). If m = 5 a check
shows that d(p,p3) < d(p,p,) and so one concludes that m ic 6 and m * 5.
 39

Each of these four remaining cases can occur, as seen by considering one of the following two
lattices.

Square lattice m = 2, 4 Equilateral triangle lattice m = 3,6

2. If f e S(L) and f is linear, let e,, e2, ..., e eL be a basis for L so that L is the set of integral
linear combinations of e,, e2, ..., e, then the matrix of f with respect to this basis has integer
entries.
When n = 2 or 3 and f is a rotation there is an orthonormal basis for R" with respect to which
f is represented by the matrix
Ae - rcose -sin91 or by r A, 0
IL sine cose J to 1

Under our hypotheses this matrix is similar to an integer matrix. As the trace of a matrix is invariant
under changes of bases it must therefore be an integer. In either case we see that 2cose e Z and
the only possibilities are cosO = 0, ± 1/2, ±1 giving that the rotation must have order 2, 3, 4 or
6.

As already mentioned this proof will give information in higher dimensions. For example when
n = 4 the matrix is integral for one basis and is of the form [As, A ,for another . So for
n = 4 (and also for in = 5) one has 2(cosel+cose2) a Z. This approach sums to lead to difficult
calculations in general and a slightly different approach is more tractable:
3. If f e S(L) is linear and of order m, then its minimal polynomial has integer coefficients and is
of degree -< n and it also divides x"'-1. These facts give information about m, for example:
Exercise i) If f is an irreducible transformation, show that n !%E ip(m) where p is Euler's function.
ii) If f e S(L) is not linear (for example, f may be a rotation whose centre is not even
a lattice point), explain carefully how to make proofs 2 and 3 complete.
It is clearly an interesting and important problem to try to describe all possible crystal groups,
that is subgroups G c I(R") which contain n independent translations but no arbitarily small
translation. Before this can be done a definition of equivalence of two crystal groups must be agreed
upon. This is a slightly thorny question and anyway the classification of the crystal groups even
for n = 2 is too long to be included here. (A proof that there are exactly 17 crystal groups in
40

dimension 2 is given in the article 'The seventeen plane symmetry groups' by R.L.E.
Schwarzenberger in the Mathematical Gazette, vol. 58 (1974) pages 123-131. Readers of these
notes should be able to read this article easily.) We will however give one classification result as
Theorem 13.
The group 1(R") consisting of all isometries of R" -contains the translations R" as a normal
subgroup, and the quotient group 1(R")/R" is isomorphic to O(n). In fact, as a group, I(R") can be
described as the semi-direct product O(n) z R", that is, its elements are pairs (A,x) with A e O(n)
and x e R" and the group operations are given by
(A,x).(B,y) = (AB, x+Ay)
and (A,x)-' = (A-',-A-'x).
The action on R" is given by (A,x)v = x + Av.
Exercise Let G be the group (1,a) of order 2. It acts on Z by a.n = -n. Show that the semi-direct
product G x Z is the infinite dihedral group. Find a subset of R2 whose symmetry group is
GxZ.
A crystal group G c I(R") has a normal subgroup of translations, G, = G r R", that forms a lattice
in R". The quotient group G = G/G, is called the point group of G. It is a subgroup of O(n) and
is finite. Notice however that the point group may not be isomorphic to the stabilizer of any point
in R":
Example If n = 2, let G be the crystal group generated by a glide g of length one along the x-axis
and a translation f of length one along the y-axis. in symbols
g(x,y) = g(x+ 1, y), f(x,y) = f(x,y+ 1).

The translation subgroup G, is generated by f and g2 and the quotient G has two elements. The
stabilizer of any point is however the identity.
Exercise In the above example, if G = (t,a), calculate how a acts on f and on g2. Show that
G x G, is generated by f, g2 and a where a(x,y) = (x,-y). Prove that a does not lie in G and that
g does not lie in G z G,. Both G and G x G, contain G1 as a subgroup of index 2.
Lemma If G is a crystal group, G acts on G,.
Proof Given A F G, there is an x e R" such that (A,x) a G. If v e G, then (I,v) a G. We show that
Ave G,, that is (I,Av) a G:
(A,x)(I,v)(A,x)-' _ (l,Av).
As this is independent of x, the action is well defined.
This lemma shows that the group G acts on the lattice G, by orthogonal transformations. Hence
the elements of G satisfy the crystallographic restriction even though the group G may not be a
subgroup of G. We will now proceed to say something about the classification of the various point
groups that arise in dimension 3. A point group is a finite subgroup of 0(3) and satisfies the
crystallographic restriction. We will classify such groups up to conjugation in 0(3), that is, we will
say that two subgroups G,, G2 are equivalent if they are conjugate. Two subgroups of 0(3) that
are isomorphic may not be conjugate, they are conjugate only if they act on R3 in the same way.
Exercise Show that any element of order 2 in 0(3) is conjugate to one of the three elements

and that these are not conjugate to each other.

 41

Now consider a finite subgroup G c 0(3) all of whose elements have order 2. 3. 4 or 6. If G lies
in SO(3) then by Theorem 11 it must be one of the following list.
1, Z12, Z13, Z/4, Z16, D2, D3, D4, D6, Tet, Oct (')
where Tet(= A4) and Oct(= S4) denote the rotation groups of the tetrahedron and octahedron
respectively. The classification given in Theorem 11 was also up to conjugacy.
Now suppose G is not entirely contained in SO(3) so that G has a normal subgroup Gd of index
2 consisting of the direct elements. The group Gd is in the list ('). Then G = GdngGd for any
gEG\Gd.If
1 0 0
J= 0 -1 0
0 0 -1
is in G then G = Gd x { 1,J} (compare the argument on page 29). This possibility gives a further
eleven groups. There remains the case that J k G, so G = Gd v JX for a subset X of SO(3). Because
G is a group and J commutes with every element in 0(3), the set Y = Gd u X is a subgroup of SO(3),
and 16[ = IYG. Every lattice has J assymmetry,
a so Y preserves the lattice that G preserves. Hence
Y is in the list ('), and has a normal subgroup of index 2. An examination of the various possibilities
shows that a further ten groups G arise in this way, so we obtain
Theorem 13 Up to conjugacy there are exactly 32 finite subgroups of 0(3) that satisfy the
crystallographic restriction.
Each of these groups arises as the point group of some crystal group, and with one exception
they all occur as a point group of some known, physically occurring crystal. In dimension three the
total number of crystal groups is 230. It was proved in 1910 by Bieberbach that there are only a
finite number of crystal groups in each dimension.
For more information the reader is urged to consult books on crystallography; a good first
exercise would be to search in the book for the equivalent of Theorem 13. A suitable introduction
is the book'Elementary Crystallography' by M.J. Buerger (published by J. Wiley in 1956).
It is interesting to note that this discussion is quite close to an unsolved mathematical problem:
How densely can one pack balls of equal radius in R3?
Let us assume that the balls have radius 1 and let li be the cube with edges of length 2E and
centre at 0. The density of a packing of R3 by an arrangement of non overlapping balls is defined
to be
lim (vol I of Its in I
e 8C
For any packing this limit is known to be greater than a number that is approximately 0.7796. A
lattice gives rise to a packing by putting the centres of the balls at the lattice points and then
changing the radius of the balls until they just touch. The densest lattice packing comes from the
'face centred cubic' lattice, its density is it/H8 = 0.7405, and it is conjectured that this is the densest
possible packing of R3.
42

face centred cubic lattice

The face centred cubic lattice has points at the vertices of a family of cubes and at the centre of
each face. It occurs in the crystal structure of gold.
Exercise i) Prove that the face centred cubic lattice is spanned by the ('/2,1/2,0), (1/2,0,1/2) and
(0,'/2,'/2).
ii) Find the densest packing of R2.

Rotations and Ouatemions

There is an important connection between the group of rotations SO(3) and the quatemions.
Before describing this connection we will study SO(3) as a topological space. The 3-dimensional
ball (ordisc) of radius r is denoted by
Dr3= (xeR31IIxII-r).
Theorem 14 The space SO(3) is homeomorphic to the quotient space D3/-- where -- is the
equivalence relation
x - yifandonlyif x=y or x= -y and {IxJI=1.
(If you need reminding about how to make a quotient space into a topological space, look again
at the discussion about the MObius band on page 18.)
Proof By Corollary 1 on page 35 each element of SO(3) is determined by its axis and the angle
of rotation. Consider the disc D3, of radius if. Define a map f: D3, --. SO(3) as follows:
f(x) is the rotation through angle IIx1I about the axis Ox.
Note that f(x) is well defined, for f(0) is the identity, so it does not matter that for x = 0 one does
not have an axis. By Corollary 1 on page 35, f maps D3, onto SO(3). Moreover f is injective except
that when IjxII = r, one has f(x) = f(-x). Hence it follows that f defines a continuous bijection between
D3, i-- and SO(3); as both of these spaces are compact f induces a homeomorphism between
them.
 43

The quatemions are defined in a somewhat similar way to the complex numbers. The complex
numbers C are defined as a vector space over R with basis (1,i) and multiplication is defined to
be bilinear, i2 = -1 and elements of R commute with everything. Similarly the quaternions H (H
stands for Hamilton their discoverer; one cannot use 0 as that is reserved for the rational numbers)
form a four dimensional vector space over R with basis (1,i,j,k), the multiplication is again bilinear
and R is central, also
i2=j2=k2=-1
ij=k,jk=i,ki=j,ji=-k,kj=-i,ik=-j
(for these last relations use + sign if the symbols are in the cyclic order i, j, k otherwise use - sign).
Exercise Verify that (x+iy+jz+ku)(x'+iy'+jz'+ku') equals
xx' - yy' - zz' - uu' +i(xy' +yx' +zu'-uz') + j(xz'-yu' +zx' +uy') + k(xu' +yz'-zy' +ux').
An ralternative approach is to consider the subset of M(2,C) consisting of all the matrices of the
form I P1. The quatemion x + iy + jz + ku is represented by the matrix
j J x+iy z+iu
-z+iu x-iy
Matrix multiplication corresponds to quaternion multiplication. The uaternion q can be regarded
as a point in R4, and so has a norm Ijgjl. If q is represented by A = a] then detA and
so Iq,g211= Ilq,ihlg211. L
A quaternion q has a conjugate q = x - iy - jz - ku. Then qq = Iq;j2, and hence a non-zero
quatemion q has an inverse q-' = q / Ilgjj2. Thus H is a non-commutative division ring, and in many
respects behaves like a field. It is now easy to verify that the 3-dimensional sphere
S3={geHllbll=l)
is a group - the group of unit quatemions.
Let U(n) denote the unitary group
U(n) = (At M(n,C)IAA` = I)
and SU(n) denotes the special unitary group, those unitary matrices with determinant 1. One
easily checks that the above correspondence between 2 x 2 matrices and quaternions gives rise
to an isomorphism between the groups SU(2) and S3. The rows of a unitary matrix are orthonormal
with respect to the Hermitian inner product
<x,y> = F x,9, on C".
Hence an element of U(2) is of the form A = t x; X- I with as + {3{3 = 1 and al = 1. The
matrix A lies in SU(2) if and only if A = 1, and the inverse of such an A is B

To make the connection between the unit quatemio ns and roto tions one needs a three
dimensional real vector space on which S3 acts. This is the space R3 = {iy+jz+ku} of purely
imaginary quaternions.
Proposition The group S3 acts orthogonally on R3 by conjugation
Proof a We make use of the representation of quatemions as 2 x 2 complex matrices of the form
a
a . Such a matrix represents a purely imaginary quaternion if and only if its trace vanishes.
As the trace of a matrix is preserved by conjugation, that is,
tr(PAP-) = tr(A),
one sees that if q e S3 and q, a R3 then qq,q ' is also in R3. So for each q e S3 one has a map
p(q): R3 .- 83 defined by p(q)q, = qq,q-1. It is clear that the map p(q) is linear; it is also norm-
preserving because
Ilgq,q `II=Iiglj.IIq,II.Igl ' = Ik1,ll.
44

So by the lemma on page 4, p(q) is orthogonal.

Theorem 15 Conjugation of quaternions induces an isomorphism p: S3/{±1) -+ SO(3).
Proof The proposition shows that p: S3 0(3) is a homomorphism. The space S3 is connected
and p is continuous so the image of p is connected. The group 0(3) has two components, the
component of the identity is SO(3) and so the image of p is in SO(3). If p(q) = I then
qq,q-' = q, for all q, a R3. A straightforward check shows that q is real. The only reals in S3 are
±1, so the kernel of p is {±1 ). It remains to prove that p is onto. Consider q = X + aµ with
a e R3, Ilall = 1, x, p. e R and x2 + µ2 = 1. Clearly qaq'"' = a so that p(q) is a rotation with axis
a. The trace of a rotation with angle 9 is 1 + 2cos9 so to determine the angle of rotation we calculate
trp(q). The diagonal entries of the matrix of p(q) with respect to the basis i, j, k are Re(-iqiq),
Re(-jqjq) and Re(-kqkq). If a = a,i + a2j + a3 k and q = X + aµ an easy calculation shows that
Re(-igiq) = x2 + µ2(a,2-a22-a32). Similarly forj and k. So
tr(p(q)) = 3\2-tL21jalj = 312-µ2 = 1 + 2(X2-µ2) =.1 + 2cos9
if x = cos9/2, p. = sing/2.
Hence, by choosing x, µ and a suitably, one can write any rotation in the form p(q).
From this theorem it follows that SO(3) is homeomorphic to the space S3/x - -x. This fact ties
in well with Theorem 14: the lower hemisphere of S3 given by u -- 0 is clearly homeomorphic to
D3 by the projection (x,y,z,u) -+ (x,y,z). Every point of S3/x -- -x has a representative in the lower
hemisphere, and this representative is unique unless u = 0. A point of S3/x -- -x with u = 0 has
two representatives in the lower hemisphere and they are antipodal points of the boundary of D3,
that is, points x, -x with If x11 = 1. This discussion gives an alternative proof of Theorem 14.
The description of the group SO(3) as S3/{±1}, or equivalently as SU(2)/(±I), is useful in
quantum mechanics where it is used for the study of spin.
 45

Problems
1. If f is a distance preserving map of R" to itself, show that f is onto. Try to find conditions on
a metric space X that ensure that any distance preserving map f: X - X is onto.
2. Show that GL(n,C) is homeomorphic to U(n) x R^' where U(n) is the space of unitary matrices.
3. Describe the space of all real 2 x 2 matrices of rank 1.
4. Show that M(n,R) has a linear subspace V of dimension n(n-1) which consists entirely of
singular matrices. If n > 1, show that there are infinitely many such spaces. It is true that there
is no linear space V of singular matrices in M(n,R) with dim V > n(n-1), but this is much harder
to prove.
5. Show that Id(R2) is isomorphic to the group of all complex 2 x 2 matrices of the form [o f ]
with ;wl = 1. J
6. Find a group of (n+ 1) x (n+ 1) real matrices that is isomorphic tol(R").
7. Let a, b be unit vectors in R2 and R be the (unique) rotation about the origin such that
Ra = b.
i) Express Rb in terms of a, b and a.b.
ii) Express Rx (for any x e R2) in terms of a, b, x, a.b, a.x and b.x.
8. Work out the compositions of the various elements in 1(R2); for example, what is
R(a,a)R(b,-a) for a* b?
9. Suppose that f: R" -. R" is an isometry such that there is a K > 0 with d(fx,x) < K for all
x E R", show that f is a translation.
10. If A F O(n), show that there is an orthonormal basis for R" with respect to which the matrix
of A has elements of 0(1) and 0(2) along the diagonal. Deduce that any f e I(R") can be written
in the form
T°HbR`with a,b=0or 1,c.0and a+b+2c,n,
where T is a translation, H is a reflection and R` is the composition of c rotations about (n-2)
dimensional subspaces. Show further that different triples (a,b,c) always give different types
of isometries. Hence show that, apart from the identity, there are 2n types of isometries of
R"
11. The half turn Ha is the rotation through ;r about the point a in R2. Show that
i) the product of two half turns is a translation,
ii) every translation can be written as HaHb and that either a orb can be chosen arbitrarily,
iii) every opposite isometry is the product of a half turn and a reflection,
iv) HaHbHc = HcHbHa and is a half turn.
12. Show that (RCRmR")2 is a translation and give necessary and sufficient conditions for
R{ RmR" to be a reflection. If t, m, n are the sides of a triangle, find the compositions of Re,
Rm, R" in the various orders.
46

13. Prove that

i) RtR",=R_Rtamisperpendicular tot,
ii) HaRe = Re Ha a e e,
iii) Re RmR" = R"RrRt q e, m, n are concurrent or parallel,
iv) H,Hb = H0H, G:- b is the mid-point of ac,
v) HaRt = Rt Hb er a is the perpendicular bisector of ab.
Translate other conditions about points and lines into ones about reflections and half turns.
14. i) If f and g are direct isometries of R2, prove that fgf-'g ' is a translation. Show that every
translation can be written in this way.
ii) Show that every direct isometry of R2 can be written in the form tgf-'g`', where f, g are
isometries of R2.
15. Let a, b, c be a clockwise oriented triangle in R2 whose angles area, a, y at a, b, c respectively.
Let R(p,8) denote rotation about p through 0 anticlockwise. Prove that
R(c,2y)R(b,20)R(a,2a) is the identity.
16. If a, b, c are affinely independent points in R2 then x e R2 can be written uniquely as
x = xa + µb + vc with x + µ + v = 1. Prove that the ratios X:µ:v are equal to the ratios
A (xbc): G (axc): A (abx) where A (xbc) denotes the area of the triangle xbc; this area is to be
interpreted with a certain sign which you should explain.
17. If a, b, c and a', b', c' are two sets of affinely independent points in R2, show that there is
a unique map f: R2 -. R2 of the form fx = Ax + d where A e GL(2,R) and d E R2 such that
fa=a',fb=b',fc=c'.Show that
d(a'b'c') = detA.A(abc)
where G denotes the area of a triangle.
Let a,x + b,y + c, = 0 (i = 1, 2, 3) be three non-concurrent lines in R2 such that no pair
is parallel. By using a suitable affine transformation, show that the area of the triangle
bounded by the three lines is (detA)2/2IdetC,detC2detC31, where A is the matrix
ra, b, c,l
C2
a3
Lag b3
b2 C3
and C; is the 2 x 2 cofactor matrix of the entry c,.
18. Prove that any two reflections of R" are conjugate in 1(R"). If G is any abelian group and
a: I(R") -+ G
is a homomorphism, show that a(RI) = a(R2) for any two reflections R,, R2. Deduce that there
is essentially only one non-trivial homomorphism from 11(R") to any abelian group, namely the
sign homomorphism defined on page 28.
19.' If H is a linear hyperplane in R", let
t'H = (f a GL(n,R)Ifx = x forx a H).
If a is a point not in M, prove that there is a real number a independent of a such that
fa = ka + h for some h e H. If x = 1,f is called a transvection. Show that the set of transvections
is normal in rH. If x * 1, f is called a dilation. Prove that every element of GL(n,R) can be
written as a product of transvections and dilations, and investigate the extent to which this
can be done uniquely.
 47

20. Let Aff(R",R") denote the set of all affine maps 1: R^ -. R. Let L, denote the linear map that
corresponds to f, that is L,(x) = f(x) -f(0). Show that
L: Aff(R",R") -- M(n,R)
is a map of rings.
By identifying R" with the set (x a R"*'`x",, = 1) show that Aff(R",R") can be identified
with a subset of M(n+ 1,R). Multiplication corresponds under this identification but addition
does not.
21.' An affinity of R" is an affine isomorphism of R".
i) Show that the set of all affinities of R" form a group A(Rn) and that the set of translations
form a normal subgroup.
ii) When AIR n) is given its obvious topology, show that it is homeomorphic to
GL(n,R) x R".
iii) If (a0, a , , ... , a") and {bo, b,, ... , b"} are two sets of independent points in R", prove
that there is a unique affinity f such that fa, = b, for each i. Use this fact to find the
dimension of the space A(R") and check that it agrees with that obtained from ii).
iv) If X is a subset of R" whose centroid is at c, show that the centroid of fX is at fc for f
A(R").
v) Which of the following concepts are preserved by an affinity of R1" Circle, ellipse,
hyperbola, parabola, parallelogram, mid-point of a line segment.
22.' If f is a line in R2, let Fix(C) = (g a GL(2,R)Egx=x, Vx e e). Prove that Fix(s) is homeomorphic
to two copies of R2 and is isomorphic to the group A(R'). Generalise these facts to higher
dimensions.
Describe Fix(() as a subspace of GL(2,R), remembering that GL(2,R) is 0(2) x R3. (To
visualize this, it may help to consider Fix(C) n SL(2,R); SL(2,R) is homeomorphic to an open
torus.)
Let Inv(e) = (g a GL(2,R)Igx e I!, Vx e t). Show that Inv(t) is homeomorphic to the union
of four copies of R3, and investigate how it sits inside GL(2,R).
23. If we represent the circle C in R2 with centre (p,q) and radius R by the point
Pc = (p,q,r) a R3 where r = p2 + q2 - R2, describe the subset X of R3 that corresponds to
the set of all circles in R2.
Verify that the family of circles that corresponds to the points of a line in R3 is a coaxal family,
and show that there are three types of coaxal families corresponding to the cases where the
line meets, misses or touches the boundary of X.
If t, m are skew lines in R3 and a 4 C u m, prove that there is a line in such that a e n,
t n n m and m n n * 4,. If x,x',y,y' are four non-concyclic points in R2 one of them lying
on the circle C, then there are points z, z' E C such that {x,x',z,z'} and (y,y',z,z') are
concyclic.
24.' If r2 = x2 + y2, then the subset of R3 defined by (r-a)2 + z2 = b2 (a > b) is a torus. The torus
is the union of a family of circles, the meridians, all of radius b. They are the various images
of the circle C: (x-a)2 + z2 = b2, y = 0 under the various rotations about the z-axis. It is also
the union of another (orthogonal) family of circles, the parallels, whose radii vary from
a - b to a + b. They are the circles of revolution obtained as the images of single points of
the circle C as C revolves round the z-axis.
48

There are other interesting families of circles on the torus:

Show that each of the planes
Z : b(x cosa+y sins) z a -b
meets the sphere (whose radius is a)
S.: x2 + y2 + z2 + 2b(x sins-y cosa) = a2 - b2
in a great circle of radius a that lies on the torus. As a varies one gets two families of circles
on the torus (corresponding to the ± in the equation of the plane). Verify that the circles of
one system do not meet each other, and that they cover the torus, but that two circles
belonging to different systems meet each other in exactly two points. (This example is treated
in Coxeter,'Introduction to Geometry' pages 132-133.)
25. Let G be a group generated by two elements a, b both of order two. If the order of ab is n
show that G is isomorphic to D,,, the dihedral group of order 2n. (Dihedral groups are
characterized by the property that they are generated by two elements of order two.)
26. Show that there are exactly seven distinct infinite subgroups G c I(R2) such that Z is an orbit
of G, where Z = ((n,0): n e Z}. For each such G, draw a subset X(G)c R2 such that
S(X(G)) G. Sometimes these subsets in R2 are called the seven wallpaper patterns.
27. Draw a'plan' projection of the dodecahedron in a plane perpendicular to a face. Do the same
for an icosahedron in a plane perpendicular to a main diagonal.
28. Show that any two diagonals of a regular pentagon divide each other in the ratio -,:l - the
'golden section' ,r equals (1 +f5)/2.
The twelve vertices of a regular icosahedron of edge length 2 are the vertices of three
mutually perpendicular (golden) rectangles whose sides are of length 2 and 27. Draw them.
If these rectangles are in the three co-ordinate planes, verify that the twelve vertices have
co-ordinates (0,±T,±1), (±1,O,±T) and (±T,±1,0). Show that the octahedron whose six
vertices are at the points (±72,0,0), (0,±72,0) and (0,0,±T2) is regular. The twelve vertices
of the icosahedron lie on the twelve edges of the octahdron and divide them in the ratio 7: 1.
(For more on 7 see Coxeter'Introduction to Geometry' chapter 11.)
29. Which of the Platonic solids have the property that their faces can be coloured black and white
and so that every pair of adjacent faces are differently coloured?
30. Given one edge of the dodecahedron, show that there are exactly five others that are either
parallel or perpendicular to it. In this way the thirty edges can be divided into five sets of six
edges. Describe the precise relationship of these sets to the five inscribed cubes and show
that this gives another way of understanding the symmetries of the dodecahedron. This
method also works for the icosahedron.
31. If r is a connected graph in R2 with V vertices, E edges and F faces, use induction on E to
show that V - E + F = 1. (There will be two cases - the new edge may join two old vertices
or it may have a new vertex. If this is not enough of a hint consult Coxeter 'Introduction to
Geometry' page 152.)
Consider a graph r on S2. By projecting r stereographically to R2 (see page 82) from a
point not on r to obtain a graph r, c R2 and using the above result for r,, prove that
V-E+F=2forr.
What is V - E + F for a graph on a torus? Deduce the formula from the above result for
a graph in R2.
 49

32. A convex polyhedron has f" faces with n edges and v" vertices at which n edges meet. Show
that
i) IN, = Ynv",
ii) $f2"+t iseven,
Iii) V3+f3>O.
33. If L is a lattice in Rn, show that there is a compact subset F c R" such that the sets x + F
(x a L) cover R" and the sets x + P are all disjoint.
34. A basis (e, , e2, ... , e") for R" determines a parallelepiped
P = {xFR"Ix= i; x,e;with 0-- x,_< 1)

and a lattice L = (x a R" I x = E m;e, with m; a Z).

Let S = max{j)e,fl, IIeAJ, ... , 00"l1) be the length of the longest edge of P. Prove that, for
n c 4, if x e P then there is an e e L such that Iix-ell _< S. Show that this is false for n 5.
e 0 1
35.' i) Show that anYelement A F. SU()2 is conlug ate to a diagonal
9 matrix U v = 0 e-'VJ
That is, there is a P e SU(2) such that P-AP = Uy. The angle can be cho en unique y
in (0,n and i$cos ' (trA/2)
a
ii) 1113= -] ,show that tr(BU,B-'Up-') = 2(aa+p(ocos2qp).
iii) Let H a normal subgroup of SU(2) that contains an element A such that
trA = 2cos(po ±2. Use i) and ii) to show that H contains matrices with all possible traces
between 2 and 2cos2tpo. Hence show that H = SU(2).
iv) Deduce that if H, c SO(3) is a normal subgroup then H, _ (1 ) or SO(3). This means
that SO(3) is a simple group.
36. Prove that two non-zero quatemions x, y are orthogonal if and only if x ' y is purely imaginary.
37. Regard a purely imaginary quatemion as a vector in R3 and vice versa. Show that the
quatemion product xy of x, y e R3 is -x.y + xxy (xxy is the vector product of x and y).
If 1, J, K is a right-handed set of orthonormal vectors in R3, show that 12 = JZ = K2 = -1
and
IJ=K,JK=I,KI=J,JI=-K,KJ=-L1K=-J.
38. If q E S3, write q = cosO + Isine with I e S2. Prove that p(q) a SO(3) is the rotation through
20 about the axis 01.
39. Show that the elements ±1, ±i, ±j, ±k and (±1 ±i±j±k)/2 form a subgroup of S3. What is
its image in SO(3)?
40. (For those with a knowledge of the calculus of several variables.)
The spaces S3 and SO(3) are both connected, compact, closed, smootn manifolds of
dimension three. The map p: S3 -. SO(3) is smooth, everywhere regular and Jp 'xI - 2 for
each x E SO(3). Deduce that p is onto.
 Part II

Projective Geometry
Projective geometry was invented in the 17th century, the first important contributions to the
subject being made by a French architect Gerard Desargues. These studies arose from attempts
to understand the geometrical properties of perspective drawing. When one draws a pair of parallel
lines, such as the sides of a long, straight road, it is usual to draw them as meeting at infinity'.
In the drawing one has'points at infinity', these lie on the 'vanishing line'. The vanishing line consists
of the points where pairs of parallel lines on the (flat) surface of the earth are depicted to meet.
The need to introduce and study points at infinity led to projective geometry.
We start by considering the simplest kind of projections. If C and m are two lines in the Euclidean
plane R2, consider the problem of projecting one of them onto the other from a point p I!f a m.
This is done as follows:
If x e e and the line px meets mat y. The projection from e to m is obtained by sending x toy.

Unless e and m happen to be parallel this map is not everywhere defined. If x e C is such that px
is parallel to m (x = a in the diagram), then there is no y. Similarly the point b is not in the image
of the projection. As x tends to a, y tends to infinity and as x tends to infinity, y tends to b. If a point
at infinity is added to each of e and m then projection becomes well defined and bijective on the
extended lines. These extended lines are called projective lines.
Similarly if >r,, 772 are two planes in R3 and one wants to consider the projection of the plane rr,
from a point p to the plane a2. Lines at infinity have to be introduced in order to make the projection
well defined. The Euclidean plane together with the (projective) line at infinity forms the projective
plane. The mathematician's 'line at infinity' corresponds to the artist's 'vanishing line'. This
vanishing line is the image of the line at infinity of the plane of the earth under projection onto the
artist's canvas.
52

Homogeneous Co-ordinates
The points of the projective line are those of R' together with the point at infinity. For some
purposes this description is adequate. (Compare with the treatment of the Riemann sphere in
complex analysis. As we will see later, the Riemann sphere does fit in well with projective
geometry.) A major drawback of this direct approach is that the point at infinity seems very different
from the finite points. From another viewpoint this is not so. The points of P' all have the same
standing, in other words P' is homogeneous. The co-ordinates that show this are the
homogeneous co-ordinates The points of P' are represented by ratios xly of real numbers. If
y 0 one has x/y a R', and if y = 0, xly is the point at infinity. More formally, a point in P' is an
equivalence class of non-zero pairs (x,y) of real numbers under the equivalence relation: (x,,y,)
(x2,y2) if and only if there is a non-zero x e R with (x,,y,) = A(x2,y2). So P' = R2 \ (0)/x _ xx.
Usually the equivalence class containing (x,y) is denoted by [x:yJ. This notation is meant to suggest
that one is considering the ratio of x and y.
Similarly there are homogeneous co-ordinates for the projective plane p2. A point is [x:y:zJ and
one has (x:y:z] = [xx:xy:xz) it .\ # 0. As before p2 = R3 \ (0)/x - )Lx. The points with z * 0 have
unique representatives of the form (x:y:1) (divide by z); such points correspond to the points
(x,y) a R2. The points with z = 0 are [x:y:0] and correspond to the points (x:yJ of the (projective)
line at infinity. From this description it is clear that there is no real distinction between the points
at infinity in p2 and the finite points, so that P2 is homogeneous. The three points (1:0:0), [0:1:0]
and (0:01] are regarded as base points. They are usually referred to as the vertices of the triangle
of reference.
With this step from R2 to P2 one has eliminated a nuisance of Euclidean geometry. This arises
because in R2 some pairs of lines meet and other pairs (parallels) do not meet. In P2 all pairs of
lines meet and this eliminates the need for special cases in the statements of some theorems and
proofs. To illustrate the use of homogeneous co-ordinates we give a proof of this fact.

Proposition Any pair of distinct lines in P2 meet in a point.

Proof A line in P2 is given by an equation
ex+my+nz=0.
Note that if one representative of the point [x:y:zJ satisfies this equation so does any other,
(xx:xy:xz).Two lines f,x+m,y+n,z=0 and e2x+m2y+n2z=0
are distinct if one equation is not a multiple of the other, in other words if the matrix
m, n+
A
12 m2 n2
has rank 2. A point [x:y:zJ lies on both lines if the vector

x
=LzJ
is a solution of Ax = 0. But as A has rank 2, the system Ax = 0 has a one dimensional solution,
and all the non-zero solutions represent the same point in p2.
This proof illustrates why some aspects of projective geometry are easier to study than the
corresponding aspects of Euclidean geometry. One deals with homogeneous rather than
inhomogeneous equations, and this makes the study of linear equations much simpler.
Exercise Show that an equation f(x) = 0, where x = (x,y,z), defines a subset of p2 if and only
if f(x) = 0 implies f(xx) = 0. Deduce that f must be homogeneous. Can the subset defined by f
be empty?
 53

The Topology of P' and P2

We will first consider the projective fine P' = R' u {ao}. The real line is homeomorphic to any
open interval (for example, the map x --* x/(1-jxj) defines a homeomorphism between (-1,1) and
R). So P' is homeomorphic to an open interval with a point at infinity, in other words P' is
homeomorphic to a closed interval with its two end points identified. This shows that P' is
homeomorphic to the circle S' = {(x,y) jx2+y2=1). By using stereographic projection one can, in
a more direct fashion, set up a homeomorphism between S' and R' u { x } . If n e S' denotes the
north pole - the point (0,1), any line through n meets S' again in p and meets R in q. The map
p - q: S' \ (n) -. R is a homeomorphism and one can also send n to -. This defines a
homeomorphism S' -. R' u ( -) = P'.

More formally, a homeomorphism can be set up by using homogeneous co-ordinates, every

point of P' has a representative [xy] such that x2 + y2 = 1 (divide x and y by (x2+y2)12). In fact,
every point of P' has exactly two representatives of this form, namely [x:y] and [-x: y]. So P' is
the space S'/x--x and this defines its topology. There are two simple ways of seeing that the space
S'/x---x is itself homeomophicto S'.
a) Every point of S'/x--x has a representative in the lower semicircle, unique except for the two
end points. When these two points are identified one gets a circle.

b) The map z - z2: S' -. S' (z a complex number with jzj = 1) identifies precisely the points
z and -z. So the second circle can be regarded as the first circle modulo the identification x - -x.

To 'see'the line at infinity in

p2 = R2 u (line at infinity)
choose a homeomorphism between R2 and the open disc b2 = (x a R2 j jlxj < 1) ; such a
54

homeomorphism is given, in polar co-ordinates, by

(r,e)-.(r/(1-IrI),e): 62 R2.
Two parallel lines It and m in R2 when mapped into D2 are as shown in the diagram.

They 'want' to meet in p and q where p = -q, so to make a space in which they meet in only one
point, one takes the disc D2 = ((x,y): x2+y2_-1) and identifies antipodal pairs of points on the
boundary. The space is P2 = D2/_- where x - y if and only if either x = y or x = -y and
ILII = 1. The new points added to R2 are S'/x -- -x which is a P'. this is the line at infinity in P2.
What we have done here is very reminiscent of our discussion of the topology of SO(3). In exactly
the same way as was done there we see that P2 is homeomorphic to S2/x - -x. The lines in P2
are the images of the great circles on the sphere S2 = ((x,y,z) e R31x2 + y2 + z2 = 1). The great
circles being the intersections of S2 with planes in R3 containing 0. We make all this formal by setting
up the homeomorphism in homogeneous co-ordinates. From its description
p2 = R3 \ (0) /x - -x-
each point of P2 is represented by precisely two (antipodal) points in S2. So that P2 = S2/x - -x.
One can define P" for all n -- 1 as S"/x - -x, and then the conclusion of our previous discussion
about SO(3) is that SO(3) is homeomorphic to P3.
The above description of P2 as D2 with antipodal points on the boundary identified coincides
almost exactly with the description given in popular accounts of elementary topology. In such
accounts the projective plane is obtained from a square by identifying edges according to the
following diagam:
 55

Under a suitable homeomorphism of the square with the disc, these identifications correspond
exactly to the identification of antipodal points.
Proposition The space P2 \ (x) is homeomorphic to an open Mobius band for any x e P2. The
space p2 \ B(x) is homeomorphic to a closed Mbbius band for any open disc neighbourhood D(x)
of X.
Proof 1 This is an elementary 'cutting and pasting' argument. One is allowed to cut the space
as long as it is eventually pasted back together in the same way as it was cut.
It is clear that the two sentences of the proposition are essentially equivalent statements. For
this first proof it is more convenient to prove the second sentence.
Consider P2 \ D(x), diagramatically we have:

where the shaded region has been removed. If one cuts along AU and YC and sticks AD to BC
there results the following.
56

After sticking AID to CID this is clearly homeomorphic to

Y2 U,

x 11D- B

U2 A,. C, Y,

This is the MSbius band.

Proof 2 For this second proof it is more convenient to prove the first sentence of the proposition.
Consider P2 \ (x) , this is homeomorphic to S2 \ (n,s)/a - -a where n and s are the north and south
poles of the sphere S2. But S2 \ (n,s) is homeomorphic to S' x R' and - corresponds to the
equivalence relation (x,y) -- (-x,-y).

S2 \ (n,s) S' x R'

a -+ a'

By the lemma on page 18, the space S' x RI/-- is homeomorphic to the open Mbbius band.
 57

Duality
With no extra difficulty, one can define the projective space corresponding to a vector space V
over any field F. It is
P(V) = V \ (0)1-
where u -- v if and only if there is a (non-zero) a e F with u = av. In our previous notation
P' = P(R2) and P2 = P(R3). Projective spaces of this type are much studied, but we will concentrate
almost exclusively on lines and planes over the real and complex fields.
Exercise Define an embedding Vc P(V ® F). What is the relationship between the projective
subspaces of P(V®F) and the affine subspaces of V?
With this elegant notation we can now explain one of the more beautiful aspects of projective
geometry -duality.
If W is a subspace of V then one has an inclusion of the corresponding projective spaces
P(V) c P(W). A vector space V has a dual space V' = Hom(V,F). This is the space of linear
transformations from V to F (also called linear functionals on V). The projective spaces P(V) and
P(V') are said to be dual. A point p e P(V') determines a codimension one subspace W of V (that
is W has dimension one less than V): Suppose p is represented by 'p a V', then W = Ker p. and
asp is non-zero, it follows that dim W = dim V - 1. It is easily verified that W does not depend
on which representative p of p was chosen. Hence a point of P(V') determines a codimension one
projective subspace of P(V) and also vice versa, at least if V is finite dimensional, because then
V = V". For example a point in P(R3') corresponds to a (projective) line in P(R), and a line in
P(R3') corresponds to a point in P(R).
Exercise i) Show that three points in P(R3') are collinear if and only if the corresponding three
lines in P(R) are concurrent.
ii) Make corresponding statements in higher dimensions and for other fields.
An inner product x.y on R" induces a linear isomorphism «: R" -. R by a(u)v = v.u and so
P(R') is then naturally isomorphic to P(R"'). An easy consequence is that the space of codimension
one subspaces in P(R") is homeomorphic to P(R") itself. In particular, the space of lines in P2 is
homeomorphic to p2 itself. The lines in P2 correspond to the lines in R2 together with the line at
infinity, so that the space of lines in R2 is homeomorphic to P2 \ (pt). We have already come across
this fact, but the present proof is more conceptual.
One of the main reasons that duality is useful is that every theorem has a dual. The dual of a
true theorem is also true, so that only one of the two needs to be proved. As an example consider
the theorem of Desargues:
Theorem 16 If PIP2P3 and g1Q2Q3 are two triangles in p2 with vertices p,, q, such that the three
lines p,q, are concurrent then the three points r,i = p,pin q,qf (i 4 j) are collinear.
The dual is
Theorem 16d If t 1e2e3 and m1m2m3 are two triangles in P2 with sidesC,, m, such that the three
points e, nm, are collinear then the three lines (e;nej).(m)mf) are concurrent.
In this case it happens that the dual theorem is also the converse of the original. Desargues
theorem can be proved just using properties of p2 but the simplest proof uses P3.
Proof Assume P2 C P3, choose a projection point 0 ti P2 and suppose that the three lines pq,
meet at v.
58

Choose some point q3' q3 on Oq3, the line vq3' meets Op3 in p3' (as vq3' and Op3 are coplanar).
The triangles P1P2P3' and g1g2q3' lie in different planes in P3, these planes meet in a line e.

r12

The points r,, are the projections of the points

P1P2nglg2, P,P3'nglg3 and P2P3'n g2g3'
to p2, hence the three points rq lie on the projection of t from 0 to P2. This shows that the points
r,i are collinear.
By applying the theorem to the dual P2 one has simultaneously proved the dual and hence (in
this case) the converse.
Exercise By choosing a suitable line at infinity deduce the following: If PIP2P3, glg2q3 are two
triangles in R2 with the three lines p,q, parallel, show that either the three points pip, n q;qi are
collinear, or the three pairs of lines p,p,, q,qi are parallel.
 59

Projective Groups
In the introduction to these notes the importance of a group acting on a geometry was stressed,
and the reader should now be familiar enough with projective geometry for the group to be
introduced.
If T: V -. V is a linear isomorphism, it clearly induces an isomorphism P(V) -. P(V). Some linear
isomorphisms of V induce the identity on P(V). If T(v) = Xv for a scalar k then one certainly gets
the identity on P(V), because one has divided out by the scalars in forming P(V). Scalar
multiplications are the only linear isomorphisms of V that induce the identity on P(V). When
V = F, its group of linear isomorphisms is GL(n,F).
The quotient group PGL(n,F) = GL(n,F)/{xI} is called the projective group and its elements
are called projectivities.
We will illustrate this definition by considering the case n = 2 and F = C in some detail. You
might have come across this already in complex analysis. To conform with a standard notation we
will write P"-'(F) for P(F").
Example P'(C) is the Riemann sphere because P'(C) is C together with a point at infinity. This
latter description is the usual description of the Riemann sphere. The points of P'(C) are, in
homogeneous co-ordinates, pairs [z:wj where (z,w) * (0,0). If w * 0, [z:wJ corresponds to the point
= z,,w E C, and if w = 0, it is'the point at infinity,
Stereographic projection shows that P' (C) is homeomorphic to the sphere S2 = (x E R3i IIx'II= t } .

Stereographic projection is the map p -+ q where npq are collinear. It can also be described as
the map obtained by rotating the map S' -+ P1 (R), defined on page 53, about the vertical axis.
Its properties are discussed in more detail in PartI l 1.

To understand the action of PGL(2,C) on P' (C) one must take representatives,
d]EGL(2,C)and LWJEC2\ (O),
c
of the elements of PGL(2,C) an P' (C) under consideration. One gets

[c d] [w] - [cz+dw]
that is, the point [az+bw:cz+dw] f P'(C). If w * 0, let = z/w, then this transformation is the
fractional linear (or MOblus) transformation
4-+ for /c.
The point at infinity is mapped to a/c and t e point -d/c is mapped to the point at infinity. (Remember
that we are assuming ad - be * 0.) This transformation is clearly a self homeomorphism of P' (C).
60

Unless c 4: 0 it is not a well defined map from C to itself. Projective space here plays the same
role as it did in our discussion of projections (on page 51) - it allows certain interesting maps to
be well defined.
When handling projective lines, it is often convenient not to have two variables z, w but merely
to have one variable l; whose values are in C u (- ). One uses some 'rules of thumb' when handling
the case ?; = -; these can be proved using the more rigorous approach using the two variables
z,w: for example (a-+b)/(cc+d) is a/c.
If f: P'(C) - P'(C), it makes sense to say that f is analytic at any point of P'(C). In particular,
one can consider analytic maps f: P' (C) -. P' (C) which are bijections and whose inverses are also
analytic, such maps are called blanalytic Isomorphisms. A Mobius transformation is a bianalytic
isomorphism and there are no others:
Theorem 17 The group of bianalytic isomorphims f: P'(C)-.P'(C) is PGL(2,C).
Proof Suppose that f is a bianalytic isomorphism. Then there is a single point a e P'C such that
f(a) _ -. By composing with a Mobius transformation, we can assume a = 0. Let f have residue
a at 0, then
f(z) = g(z) + a/z,

where g(z) is analytic and finite valued everywhere on P'C. By Liouville's theorem, g is a constant
c. Hence
f(x) = c + a/z

and so f is a Mobius transformation.

Two important properties of Mobius transformations are:
Theorem 18 If f is a Mobius transformation, then f is conformal, that is, it preserves angles, and
f sends circles to circles.
Proof The map f is conformal because any bianalytic map is conformal.
The equation of a circle is
Az2+Bz+B2+C=0
where A. C are real. If A = 0 the circle degenerates to a straight line, that is, a circle through =.
Let z = (aw +b)/(cw+ d) then w satisfies
A aw+b aw+b + B aw+b
cw+ t + cw+ ( + ; +C=0
which reduces to an expression of the required form.
Projections are projectivities as we will now show forthe case n = 1.
Example A projection f: P' (F) -. P' (F) is an element of PGL(2,F).
We consider two lines C,, e2 in P2(F) and a point p not in either. By choosing a suitable line at
infinity, we may assume that ti, C 2 meet in F2 and that p e F2. Further by choosing a suitable basis
for F2 one can assume that C, is x = 0, C2 is y = 0 and that p is the point (1,1). The map
e, -- e 2 is then defined by f(x) = y if (x,0), (0,y) and (1,1) are collinear; that is, if y = (x-1)/x. So
f(x) = (x-1)/x and f is in PGL(2, F).
 61

We now study how the group PGL(2,F) acts on P' (F):

Theorem 19 Given two sets {p,q,r}; {p',q',r') each consisting of three distinct points in P'(F).
There is a unique f PGL(2.F) such that fp = p', fq = q' and fr = r' (such an action is called triply
transitive).
Proof We will use the following lemma to choose good representatives in F2 for the points in
P'(F).
Lemma Given three distinct points p. q, r e P'(F), they have representatives'p, q,'r a F2 such that
p=q +`rr.
Proof of lemma Choose any three representatives p`,,'q,, ,. As q, r are distinct, 'q, and , are
independent and therefore span F2. Hence there are constants X, µ such that p, µr,. The
representatives we need are then p = `p, , q = aq, and (= µ'r, .

Using this lemma choose representatives p, q, and 0', q', ?' such that `p = q + F and
' = q'' + 7'. There is a non-singular linear transformation f: F2 -, F2 defined by'fgq = q'' and f
= r (remember that {q;r) and {q',r } are bases for F2). Clearly fp = ', and so the image of f'
in PGL(2,F) has the required property.
To prove uniqueness, it is easy to see that one needs only to verify that, if p. q, r e P'(F) are
distinct points such that fp = p, fq = q and fr = r, then f is the identity. Choose representatives
f e GL(2,F), `p, 'q, r` a F2 such that 'p = `q + i. As fp = p etc., there are >`, µ, v e F such that
fp=x`p,fq=p. andfi=vr.So
.\(q+r) = xP = to = f(q+ =7q+fr'= µq + vf'.
But q', i are independent so x = µ = v. So f'= lit and hence f is the identity in PGL(2,F).
This theorem can be used to introduce an important invariant of projective geometry.

The Cross-Ratio
The three points 0, 1, - e P'(F) are called the standard reference points. In homogeneous co-
ordinates they are [0:1 ], (1:1 J and [1:0].
Given any four distinct points p, q, r, s e P'(F), Theorem 19 provides us with a unique
f e PGL(2,F) such that fp = w, fq = 0, fr = 1, the image of the fourth point s is then determined,
say X = fs, and is called the cross-ratio of p,q,r,s. It is usually written as a = (pq,rs) or (p,q;r,s).
This cross-ratio can be defined directly in terms of the points p, q, r and s as follows: Define the
arithmetic rules on P'(F) by using the usual ones on F and treating - using the 'rule of thumb'
mentioned on page 60; that is, write oo as 1/0 and then multiply throughout by 0. This could be
formalised but it is easier to do it informally. With this convention, the cross-ratio can be evaluated
numerically as
(pq,rs) =
lP s)(q r)
Proposition The two definitions of the cross-ratio agree.
Proof If p, q, r, s e P'(F) and f e PGL(2,F) we need to show that (fp,fq;fr,fs) = (p,q;r,s). Let p =
[po:p, ] etc., then an easy calculation shows that the cross-ratio (p,q;r,s) is
deaf deaf s where (p,r) is the matrix
ps r°
q lp
Note that det(p,r) is not well-defined but that rthe, cross-ratio` is we I defined: If 'f a GL(2,F)
represents f then fp is the point [po':p,'J where Lp°,1 = ff p°l so det(fp,fr) = det f.det(p,r). The
result now follows. JJ
LJ
62

Fixed Points of Projectivitles

Finding the fixed points of a map f: P'(F) --. P"(F) for f ePGL(n+1,F) is the same as finding the
eigenvectors of a representative f e GL(n+1,F) because fp = p if and only if there is a x with
fp = \`p for a representative p of p. Projectivities are sometimes named according to their fixed
point sets. The reader may verify that the following cover all the possibilities in each of the stated
cases:
n = 1, F = C. There are two types of element apart from the identity.
Parabolic: Exactly one fixed point e.g. p 11] (not diagonalizable)
Hyperbolic: Exactly two fixed points e.g. LO 2] (diagonalizable)
n = 1, F = R There are three types of elemen apart from the identity; these are distinguished
using z = del [a d] and t = trace [c d]
Elliptic: No fixed point t2 <4A
Parabolic: Exactly one fixed point t2=4A
Hyperbolic: Exactly two fixed pants t2>4i .

The Elliptic Plane

Thus far we have considered all the projective transformations of a projective space P"(F). When
F = R one can define a good metric on P"(R) = P", and then consider the group of isometries.
We will do this in the case n = 2. It may be helpful to have the following general remarks about
lines in metric spaces:
A curve (or path) in a metric space X is the image of the unit interval [0,1 ] under a continuous
map [0,11 - X. A geodesic in X is a curve of shortest length between two points of X. This is a
naive form of a more complicated definition, and is not quite rigorous but it is good enough to convey
the general ideas. One has to define geodesics locally and then extend these globally. For example
consider the sphere S2, on a great circle both the long curve from a to b and the short curve are
local geodesics, but according to our definition only the short curve is a geodesic. One can think
of the geodesics as the straight lines in the space.
 63

The geodesics on the sphere S2 are the great circles - the intersections of linear planes in R3
with S2 (as is well known to navigators!). One can therefore consider a geometry on S' on which
the lines are these great circles. The main defect with this as a geometry is that any two lines meet
in two (antipodal) points. However the geometry is quite tractable, calculations can be performed
by using 'spherical trigonometry'. One can get rid of the defect by identifying all pairs of antipodal
points with each other, x - -x. Thus one obtains the projective plane with a metric - this is called
the elliptic plane and the geodesics are precisely the projective lines. As the sphere S2 has unit
radius, one can easily calculate that the total area of the elliptic plane is 21r (half the area of S2,
because of the identifications) and that the length of a projective line is rr.
The group of isometries of the elliptic plane consists of those elements of PGL(3,Rl that preserve
the metric. Any isometry of P2 arises from an isometry of S2 and the isometries of S2 make up the
group 0(3). Hence the group of isometries of the elliptic plane is the image of 0(3) c GL(3,R) in
PGL(3.R); call this PO(3). The only scalar matrices in 0(3) are ±1 so PO(3) = O(3)/{±1). As
-11 SO(3) and 0(3) = SO(3) x (±I), we see that PO(3) is isomorphic to SO(3). Note however
that the groups PO(n) and SO(n) are not always isomorphic; PO(n) has two components for even
n. In particular, the isometries of P1 form the group O(2)/(±I). The elements in the component
SO(2)!(±l) correspond to rotations of S' and have no fixed point (except that the identity has fixed
points), so this component consists entirely of elliptic elements. The other component contains
reflections of S' and each element has exactly two fixed points in Wand so is hyperbolic:

Q 9

Fp
Reflection in E.

The two fixed points are the images of p and q. Any parabolic element of PGL(2,R) is not an
isometry.
As an example of a calculation in the elliptic plane we have
Example Let A be a triangle in the elliptic plane whose angles are a, 0 and y. Its area is
(a+P+y) -;r.
Ais bounded by (segments of) three projective lines. Consider the corresponding three great
circles t, m, n in S2.
64

The area of the elliptic triangle A dearly equals the area of the spherical triangle abc. The area
of the'lune' between the lines m and n is 2a.

Calculating, one gets

A+ = 2a
A+p2=2p
t +1 3=2y
0+01+A2+L3=2R
So2i+2A=2a+20 +2y
giving the result.

Exercise Calculate the area of a'convex' spherical polygon in terms of its angles.
 65

Conics
Conics have been studied since the time of the ancient Greeks. They seem to have been
invented to solve the Delian problem: to find a geometrical method of constructing a number x such
that x3 - 2. It can be proved, using a little of the theory of fields, that such an x cannot be constructed
by ruler and compass alone. Consider the two parabolae x2 = y and y2 = 2x. The
x-coordinate of their (non zero) intersection satisfies x3 = 2. This is one of the solutions of the Delian
problem given by Menaechmus in the 4th century BC. The Greeks gave numerous other solutions
of this problem that used conics in various ways. For these and other interesting uses of conics
see the book' 100 Great Problems of Elementary Mathematics' by Dorrie (Dover publications).
Conics. as their full name conic sections suggests, are obtained by considering the planar
sections of a cone. For example, if C is the cone x2 + y2 = z2 in R3 whose vertex V is 0, one can
consider C o a for various planes it in R3:
66

We list the various possibilities for r:

1. If V e r, we get the singular tonics:
a. , meets C only at V, the conic is one point.
b. r is tangent to C along a generator, the conic is a straight line.
c. r meets C in two straight lines.
11. If V J ,-r, there are three cases corresponding to which of the three possibilities of 1 above holds
for the plane;-,' parallel to r and passing through V. These are the non-singular conics.
a. r meets C in an ellipse (C 1 in diagram).
b. r meets C in a parabola (C2 in diagram).
c IT meets C in a hyperbola (C3 in diagram).
Exercise Let D3 be a solid ball in R3, p e R3 \ D3 and r a plane in R3 not containing p and not
meeting D3. What shape is the projection of D3 from p onto the plane n?
From the projective viewpoint all the non-singular conics are the same, because they can clearly
be projected to each other from the vertex V - see the above picture. It is therefore easier to study
conics from the projective viewpoint. There is only one type, the difference being only in the choice
of the line at infinity. There are three possible choices relative to the conic C:
i) ellipse, the line at infinity does not meet C
ii) parabola, the line at infinity touches C
iii) hyperbola, the line at infinity meets C in two points.
Theorems about conics are therefore best proved in P2 and to obtain special theorems about one
of the three types in R2 the line at infinity is chosen carefully in p2.
By suitably choosing the line at infinity, a non-singular conic in P2 can be regarded as an ellipse
in R2. Hence any non-singular conic in P2 is homeomorphic to a circle. We have seen that any
projective line in p2 is also homeomorphic to a circle. However a line can be distinguished from
a conic topologically, because their complements in P2 are not homeomorphic - they lie inside P2
in different ways. This is not easy to visualise directly because P2 itself is not easily visualised, but
it is worth making the effort. A similar phenomenon occurs with the Mbbius band M: consider the
'middle circle' Q of M which goes round M once and a circle m that is very close to the edge of
M and so goes round M twice.

b C
a a
c b
 67

The complement of e is connected and is homeomorphic to an annulus but the complement of m

has two components. One is homeomorphic to an annulus and the other to a Mobius band. (If you
find this hard to visualise you can cut two paper Mobius bands, one along e and the other along
m.) Remember the proposition on page 55 that P2 \ {open disc) is homeomorphic to the Mobius
band. A line in P2 is like the line C of the Mobius band and a conic in P2 is like the line m of the
Mobius band. In fact, after removing a point not on the line or conic the situation is exactly like this,
to see this we use a slightly different viewpoint:
The space R2 \ (line) is disconnected but P2 \ P' is connected. Consider the sphere S2 with a
great circle removed. Under the map n: S2 P2 this space is mapped to P2 \ P' and consists of
two hemispheres which are identified under rr so P2 \ P' is homeomorphic to an open disc and
therefore to A2. Alternatively, think of Pi c P2 as the line at infinity, removing it leaves R2. The space
p2 \ (conic) is disconnected, one component (the 'inside') is homeomorphic to R2 the other
component (the 'outside') is homeomorphic to an open Mobius band. To see this, choose the line
at infinity to miss the conic so that the conic is an ellipse in R2, its inside is homeomorphic to R2
and its outside is homeomorphic to an annulus. But its outside in P2 is homeomorphic to p2 \ (disc)
which is a Mobius band.
The equation defining a conic in R2 is a polynomial of degree two
axe+by2+2dy+2ex+2fxy+c=0.
An equation defining a subset of a projective space must be homogeneous otherwise some co-
ordinates representing a point may satisfy the equation and others may not. To obtain a
homogeneous equation from the above equation, one multiplies each term with various powers
of z so that each term has degree two giving the equation
q(x,y,z) = ax2 + bye + cz2 + 2dyz + 2ezx + 2fxy = 0
(More properly we should have taken new co-ordinates g, ,t, g with x = E);, y = 'r I giving the co-
ordinates {g:ii:;j in p2 and (x,y) in R2 with = 0 being the line at infinity.)
In matrix notation
q(x,y,z) s xtAx
x a f e
where x is the vector ryl and A is the matrix f b dl
Lz ed c
An easy check shows that the conic is non-singular if and only if the matrix A is non-singular. With
this algebraic definition one can discuss conics in P2(F) for any field F. They are non-singular if
the matrix A is non-singular.
Exercise (For those with a knowledge of the calculus of several variables.) The projective plane
P2 is a manifold (indeed so is P" for any n). Show that the conic C is non-singular if and only if
the function q(x,y,z) has non-vanishing derivative at each point of C and that then C is a manifold.

Algebraically a conic is given by a quadratic form in three variables, so that conics can be studied
using the theory of quadratic forms.

Diagonalization of Quadratic Forms

The algebraic invariants of quadratic forms arise from the attempt to diagonalise them. We will
do this for the fields R and C but it can be done over other fields. The interested reader can do
this himself.
68

Over R The invariants of a real quadratic form are its rank and signature. Given the rank and
signature, there is a linear change of variables which transforms the quadratic form to a canonical
form. In our case this linear change of co-ordinates gives a projective transformation of p2 that
sends q to a canonical form. This canonical form is easy to analyse and we summarize the results
in the following table. As we are considering the points where q vanishes, the sign of q is immaterial
so we need only consider quadratic forms q with positive signatures.
Rank Signature Canonical Form Point Set of Conic
3 3 x2+y2+z2=0 Empty
3 1 x2 + y2 = z2 Non-singular conic
2 2 x2 + y2 = 0 One point: [0:011
2 0 x2=y2 Two lines, x = ± y
They meet at [0:0:1 ]
1 1 x2 = 0 One line, x = 0, of multiplicity 2,
"double line" [0:a:b]
It is worth noting that the three singular cases remain distinct in P2 and that the empty conic does
not arise as a conic section unless the cone itself degenerates to a line.
Over C In this case the only invariant is the rank.
Rank Canonical Form Point Set of Conic
3 x2 + y2 + z2 = 0 Non-singular conic
2 x2 + y2 = 0 Two lines, x = ±ry
1 x2 = 0 One line, x = 0, of multiplicity 2
"double line"
The classification becomes much simpler on passing from the Euclidean to the projective case
and again on passing from the real to the complex case. A good way of understanding the
classification is to go backwards through these stages. That is why complex projective geometry
was so extensively studied in the last century -theorems in the other situations being proved by
specialising theorems of P2(C). Of course the amount gained by doing this for conics is not great,
but in the study of higher degree curves (the zero sets of polynomials of degree 3 3) the advantages
are much more substantial. Newton was the first to attempt to classify cubics (curves of degree
3) in R2 and he obtained 72 of the possible 78 forms. In P2(C) there are just 3 forms.
Now we investigate the space of all conics in P2(C) (this works over any field but there is no
significant loss in just thinking about the case F =C).
Theorem 20 The space of all conics in P(C) is P'(C).
Proof This is immediate because a conic is determined by the six numbers a, b, c, d, e, f and
multiplying throughout by a non-zero scalar gives the same conic.
Note that we are really saying that the space of quadratic forms in three variables, up to scalar
multiplication, is P5(F). These forms do not always correspond to different point sets in P2(F). For
example, over R there are many different quadratic forms that give the empty point set. However
for an 'algebraically closed' field such as C the correspondence of the theorem is also an exact
one for point sets.
Exercise let v: P2(F) -. Ps(F) be the Veronese map, given by
v(x:y:z] = (x2:y2:z2:2yz:2zx:2xy]
tier the images of We cores in P2,F) are the intersections of the image of v with the various
hyperplanes of Ps(F) (note that v is always injective). This gives a correspondence between the
set of conics in P2(F) and the dual of the P5(F) in which the image of v lies. The image of the map
v is called the Veronese surface.
The various types of conics will form certain subsets of P5(F), which we now investigate when
F=C.
1. The rank one conics correspond to the matrices of rank one and so are given by the subspace
of P5(C) defined by the equations:
ab = f2 ac = e2 ad = of be = d2.
This subset R, is given by four equations, nevertheless it is a two dimensional subset (because
the equations are not 'independent'). In fact R, is isomorphic to P2(C), because a rank one conic
is a line of multiplicity two and the space of lines in P2(C) is isomorphic to
P2(C) itself by duality.
2. The rank two conics correspond to the matrices A given by the equation det A = 0 (more
correctly this is the subset consisting of rank one and rank two conics) and so this is the set defined
by the equation
abc + 2def-ad2-bee-cf2 = 0.
Such a set is called a hypersurface of degree 3 in P5(C). The rank two conics are pairs of lines.
The set of unordered pairs of lines is P2(C) x P2(C)/(x,y) -- (y,x). The rank one conics sit in this
set as the diagonal elements (x,x). This degree three hypersurface is homeomorphic to P2 (C) x
p2(C)/_
3. The rank three conics correspond to the complement of this degree three hypersurface.

Polarity
This is a concept that applies in any projective space P"(F) but we will concentrate on the case
n - 2. As a small project, the interested reader could carry out the generalization himself.
The concept of polarity involves a fixed, non-singular conic C whose equation will be taken to
be x'Ax = 0 where A is a 3 x3 symmetric matrix. Two points a, b e P2(F) will be called polar with
respect to C if a'Ab = 0. This is a symmetric relation (a`Ab = 0 4 b`Aa = 0) because A is a symmetric
matrix. Note also that a point a e C is polar to itself. For a given point a, the set (xla'Ax = 0) of
points polar to a forms a line called the polar line of a.
Exercise Let (be a line in P2(F). Show that the polar lines of the points off are all concurrent.
Sometimes, there is a simple geometrical description of the polar line of a point:
Proposition Suppose that there are two tangents C,, C 2 from the point a to the conic C and that
they are tangent to Cat b,, b2 respectively. Then the polar of the point a is the line b1b2.
70

Before giving the proof, it is appropriate to explain what is meant by saying that the line t is
tangent to C.
The points of the line a joining a and b are of the form a + l,b (X= - gives b). Sob meets C
at the points given by
0 = (a+Xb)'A(a+xb) = a'Aa + 2]Xb'Aa + a2b'Ab.
This quadratic equation has either 0, 1 or 2 roots. If it has only one root, X0 say, then 8 is said to
be tangent to Cat the point a + fib. The condition for a single root is
a'Aa.b'Ab = (b'Aa)2 ')
If b t C, this condition becomes b'Aa =0 and so the equation of the tangent at b is b'Ax = 0.
Exercise If b k C, show that a satisfies (') if and only if a lies on one of the tangents from b to
C.
Proof of Proposition As a lies on both tangents, we have
b,'Aa=b2Aa=0.
Hence, a'Ab, - a'Ab2 = 0, so the polar line of a passes through b, and b2, proving the
Proposition.
When F = R and the conic C is non-empty this Proposition gives a method of constructing the
polar of a point a. There are three cases depending on whether a is outside, inside or on the conic.
If a is outside the conic, draw the two tangents from a to the conic and the polar line is the line
joining the two points of contact.
If a is inside the conic, draw two lines t,,t2 through a meeting the conic at b,, b,' and b2,b2'
say.

Suppose the tangents at b;, b,' meet at c; (i = 1,2). The line c,c2 is then the polar of a.
If a is on the conic then the tangent to Cat a is the polar line of a.
Exercise When F = C and a i C prove that there is a line through a and tangent to C. Hence
in this case C has no inside or outside.
Exercise Polarity is a generalization of orthogonality: If F = R and A = 1, then a is polar to b
it and only if corresponding vectors a, b t R3 are orthogonal. In this case, the conic is empty.
An important role is played by self-polar triangles: A triangle abc is self-polar with respect to
C if each side is the polar line of the opposite vertex.
Proposition Self polar triangles always exist.
Proof Take any point a , C and let? be its polar line. Take any b t 8 , b k C and let m be its polar
line; m passes through a and meets 4 inc. The triangle abc is self polar.
 71

If the triangle abc is taken as the triangle of reference, that is, one chooses a co-ordinate system
(or basis) so that a is e, = [1:0:0], b is e2 = [0:1:0] and c is e3 = [0:0:1 ] then the condition that
abc is self polar with respect to x'Ax = 0 is the same as the condition that A is diagonal: Let
A = (a) then e,'Ael = ay and if abc is self polar e; Ae, = 0 if i j so A is diagonal. The converse
is similar.
There are very many self-polar triangles, the space of all (ordered) triangles in P2(F) is a six-
dimensional space (in fact (P2 X P2 x P2) and from the above proof one sees that the space of
self polar triangles is three-dimensional. If two three-dimensional spaces lie in a six dimensional
space then one expects them to meet (just as two general three dimensional affine spaces in R6
meet). On these rather general grounds one would expect two conics to have a common self polar
triangle and this is in fact the case:
Proposition Suppose C,, C2 are two conics in P2(F) that meet in four distinct points. Then, there
is a (unique) triangle which is self-polar with respect to both C, and C2.
Exercise If C1, C2 are distinct, non-singular conics, show that C, n C2 has at most four points.
As a preliminary we will study how the conic, a point and its polar meet a line. But first we need
a new notion.
Definition Two pairs of collinear points, a,b and a,p are harmonic if (a,b;a,p) = -1.
Lemma The pairs a,b; a,p are harmonic if and only if (a,b;ot,p) = (b,a;a,p).
Proof If x = (a,b;a,p) then X-' = (b,a;a,p). If the four points a, b, a, p are distinct then X/ 1.
So (a,b;(x,p) = (b,a;a,p) q x2 = 1 <-z- X _ -1.
Harmonic pairs arise naturally from complete quadrilaterals:
Proposition Let a, b, c, d be four points in P2(F). Let p = (ad) n (bc), q = (ac) n (bd) and
r = (ab) n (cd). Let s = (qr) n (ad) and t = (qr) n (bc). Then (q,r;s,t) is harmonic.
72

Proof Let u = (pq) n (ab) and v = (pq) n (cd). Projecting from p one gets
(q,r;s,t) = (v,r;d,c)
and by projecting from q one gets
(v,r;d,c) _ (u,r;b,a).
Projecting once again from p gives
(u,r;b,a) _ (q,r;t,s).
The result now follows from the lemma.
Exercise If a, b, c E R then the pairs a, b and c, ar are harmonic if and only if c is the mid point
of ab.
We now return to conics:
Lemma Let C be a non-degenerate conic and let t be the polar of the point a. If m is a line through
a meeting C at b and meeting C in a, p then the pairs a,b; a,p are harmonic. Conversely if
(a,b;a, p) = -1 then b is the point of intersection of{ and m.
Proof

Choose co-ordinates so that a is [1:0:0] and p is [0:1:0] so m is the line z = 0. The equation
of C is therefore xy + zX = 0 where X is some linear expression in x, y. z. Let a be [1:s:0] and
b be [1:t:Oj then a and b are polar if and only ifs + t = 0. But (a,b;a,p) = sttand so the result follows.
Now we can return to the proof of the existence of a self polar triangle.
 73

Proof Let a, b, c, d be the four points where the conics meet, construct the complete quadrilateral
from them as on page 72. The triangle p, q, r is self-polar: It is enough to prove that pq is the polar
of r. The pairs a,b:r,u are harmonic so, as a, b e C, u must lie on the polar of r. Similarly v lies on
the polar of r, proving the result.
Combining the relationship between diagonalization and self polarity with the existence of a
common self polar triangle one sees that it is almost always possible to diagonalise two quadratic
forms simultaneously. A suff icient condition being that the two conics must meet in four points.
Exercise Prove that two quadratic forms cannot be diagonalized simultaneously if the
corresponding conics touch.
Given a triangle, there are many conics with respect to which it is self polar:
Proposition Let abc be a triangle and p E (ab), q e (ac) then there is a unique conic C such that
abc is self polar and such that p, q both lie on C.
Proof We can take abc to be the triangle of reference. Then abc is self polar with respect to a
conic C if and only if the equation of C is
Xx2+µy2+vz2=0.
The conditions that p. q lie on C uniquely determine X, v up to a constant multiple.

Now, we discuss the famous theorem of Pascal. To prove it we need to consider the cross-ratios
of points on a conic.
Lemma Let p, q be two distinct points on a non-singular conic C and let a, b, c, d be four other
distinct points of C, then (pa,pb;pc,pd) = (qa,qb;qc,qd).
Proof Let pa, qa, pb, qb have equations tax = 0, max = 0, 0, mbx = 0. Any line through
p has equation (Ata+µtb)x = 0 and similarly for the lines through q. Without loss of generality,
we can assume that the equations of pc, qc are (t a tb)x = 0 and (ma-mb)x = 0.
The conic C passes through the four points of intersection of the two fine pairs (pa) L (qb) and
(pb) u (qa). The equations of these line pairs are
(eax)(mbx) = 0 and ((bx)(max) = 0
The matrix of the conic C must be a combination of the matrices of these two line pairs as they
all have four points in common. So the equation of C is of the form
s(.ax)(mbx) + t(.t,x)(max) = 0.
But the equation of C is also a combination of those of (pa) u (qc) and (pc) u (qa) so s = -t = 1.
Hence the equation of C is
( ax)(mbx) _ (ebx)(max) (')
A calculation now shows that
(pa,pb;pc,pd) =tad/tbd
and (ga,gb;qc,qd) = madlmbd.
But d c C so by (') the two cross-ratios are equal.
Corollary Given four points a, b, c, d on a conic C one can define their cross-ratio (a,b;c,d) to
be (pa,pb;pc,pd) for any p e C.
Pascal's Theorem Let a.b,c;a',b',c' be six distinct points on a non-singular conic. Then the three
points
p = (bc') n (b'c), q = (ac') n (a'c) and r = (ab')n(a' b)
are collinear.
74

Proof

Lets = (ab') n (a'c) and t = (ac') n (b 'c).

By projecting from a' and using the above corollary one gets (a,s;r,b') _ (a,c;b,b').
Similarly, using c' one gets (a,c;b,b') = (t,c;p,b').
So (ga,gs;qr,qb') = (gt,gc;gp,qb').
But qa = qt and qs = qc so qr = qp, hence p, q, r are collinear.
Exercise Use the same method to prove Pappus's theorem: If a, b, c are collinear and
a', b', c' are collinear, prove that p, q, r are collinear where p = (bc') n (b'c), q = (ac') n (a'c) and
r = (ab')n (a'b).
Note that Pappus's theorem can be regarded as a degenerate case of Pascal's theorem - when
the conic is singular.
 75

Problems
1. Show that the equation of the line joining the two distinct points (a1:b,:c,] and (a2:b2:c2J in
P2 is
x y z
del a, b, c, = 0
a2 b2 c2

2. Let V be an m-dimensional subspace of R" and let L,, L2 be two (n-m)-dimensional subspaces
of R" such that V n L, = V n L2 = (0). Define a mapping f: L, L2 as follows: for x e L,,
consider the affine space x + V, it meets L2 in a single point y, define f(x) = y. Prove that
f is a linear map.
3. Show that the maps f: S2 -+ R4, g: D2 -. C2, defined by
f(x,y,z) = (x2-y2,xy,yz,zx),
and g(re'e) = (re21,r(1-r)e" ),
induce continuous, injective maps of P2 into R4.
4. The faces of a regular octahedron can be coloured black and white so that no two adjacent
faces are of the same colour. Consider the object formed by the union of the black faces and
the three diagonal squares of the octahedron. This is called the heptahedron. Show (by
cutting and pasting) that there is a continuous map from P2 to the heptahedron. So the
heptahedron is a model for P2 in R3.
5. Formulate the dual of Pappus's theorem. You should dualise some other theorems of
projective geometry.
6. Four points a, b, c, d e P2, no three of which are collinear, determine a complete quadrangle.
It has six lines, which have three other points in common. Draw this configuration. What is
its dual?
7. Let n, 7r2 be two copies of P" '(F) in P"(F) and let p k 1 v n2. Prove that the projection of
n, onto n2 from p induces a projectivity of P' '(F). (Sometimes such a map is called a
perspectivity with centre p.)
Show that any projectivity: P'(F) -. P'(F) can be expressed as the composite of two
perspectivities.
8. By considering the Jordan normal form, show that there are six essentially different types of
elements in PGL(3,C). Can they be distinguished by their sets of fixed points?
9. Check that the obvious homomorphism SL(2.R) -. PGL(2,R) has kernel of order two, but that
the two spaces are homeomorphic.
10. Let a0, a,, ..., a" be (n+1) points in P"(F) that do not lie in any hyperplane. Prove that the
subgroup of PGL(n+ 1,F) that consists of those f such that fa, = a, for 0 ci <_ n is isomorphic
to (F')". [F' = F \ (0).]
11. Prove that the map f: C -+ C' defined by f(z) = exp(2niz) is a homomorphism whose kernel
is Z. Deduce that C' a C2.
12. Given (n+2) points po, p,, ..., p"+, in P"(F), no (n+1) of which he in a hyperplane. and
another such set q0, q,, ..., q,,,,, prove that there is a unique f E PGL(n+1,F) such that
fp,=q,forall0-- i_n+1.
76

If po, p,... , p, are distinct points in P2(F) and po, p,. P2 are not collinear but both the sets
po. P1. P3 and po, P2, P4 are collinear, prove that there is a unique f e PGL(3,F) sending the
p's onto any set qo, q1...., q4 that has the same properties.
13. Find a monomorphism 5,,,, - PGL(n,C) where S,,,1 is the symmetric group on n + 1 letters.
Describe the image of this homomorphism geometrically.
14. Show that the image of SU(2) ---. GL(2,C) -PGL(2,C) is isomorphic to SO(3) and that the
action of f e PGL(2,C) on P'(C) is the same as that of the corresponding element of SO(3)
on S2. (Corresponding under stereographic projection.) In this way the rotation groups of the
Platonic solids can be identified with groups of Mobius transformations. Verify the following.
i) The tetrahedral group is generated by the Mabius transformations T1(z) = (z+i)/(z-i),
T2(z)= -11z.
ii) The octahedral group is generated by T,, T2 and T3(z) =iz.
iii) The icosahedrat group is generated by
wZ, I(w4- )z + (w2-3))/[(w2-W3)Z + (w- 4)),
where w is a primitive fifth root of unity.
Find M6bius transformations that generate the dihedral groups.
15.' Show that, for each k > 0, the group GL(n,C) acts on the space of homogeneous polynomials
of degree kin n variables. Verity that the three polynomials
x4 ± 20 x2y2 + y4, xy(x4-y4)
are each invariant (up to a constant factor) under the action of the tetrahedral group. Is their
stabilizer equal to the tetrahedral group? How are the zeros of these polynomials related to
the tetrahedron?
16. For f, g e PGL(2,C), prove that fg = gf it and only if either
i) f or g is the identity
ii) f and g have the same fixed point sets, or
iii) f2 = g2 = 1 with fixed points x1, x2 ; y1, y2 that satisfy (x1,x2,y1,y2) _ -1.
17. If (ab,cd) = X, calculate the cross-ratios of the various permutations of a, b, c, d in terms of
K (for example, (ad,bc) Hence find a non-trivial homomorphism S4 -, S3 between
the symmetric groups.
(Hint: consider the action of S4 on the set (2,2-',-1) obtained using k = -1.1
Use a similar method to construct a non-trivial homomorphism S4 -+ {± 1).
Note that for n > 4, the only non-trivial homomorphism defined on S. is the sign
homomorphism.
The non-trivial homomorphism S4 --. S3 can be constructed more geometrically as follows.
Consider the complete quadrilateral drawn on page 72. Regard S4 as the group of bijections
of the set (a,b,c,d) and S3 as the group of bijections of {p,q,r}. A permutation of (a,b,c,d)
induces one of {p,q,r}.
18. Show that, for any three distinct points a, b. c e P' (F), the map x -, (ab,cx) defines a bijection
P1 (F) \ (a) -+ F.
19. Let f: P'(R) be a map such that (ab,cd) = (fafb,fcfd) whenever a, b, c, d are distinct.
Show that f e PGL(2,R). Is the corresponding result true for P' (C)?
 77

20. The set of lines through a fixed point p e R2 can be identified with P'(R). If (, m are two lines
meeting at p. and b,, b2 are the bisectors of the angles between them, show that
(t b,,mb2) _ -1.
21. Set up projective spaces over the quatemions.
Find out about the Cayley numbers (an eight dimensional algebra over R). In this case one
can define the projective plane but not a higher dimensional projective space. Investigate.
22. On a sphere of unit radius consider a
spherical triangle with angles and sides as
shown, each angle being less than n. Prove
that
i) cosa=cosbcosc+sinbsinccosa
ii) sin a sino = sin b sins.

23. Show that there is a spherical triangle with angles O < ac 0 c -y if and only if
p+y<rr+a.
For which integers p, q, r is there a spherical triangle whose angles are zr/p, zr1q and rlr?
24. Given the area and base of a spherical triangle, find the locus of the third vertex.
25. A circle in the elliptic plane is a set (yJd(y,x) = r) for a fixed x e P2 and r > 0. By drawing
the corresponding sets in S2, find two circles in the elliptic plane that meet in four points.
26. Sketch the curve y2 = x3 + x2 in R 2. What is the equation of the corresponding curve in
terms of homogeneous coordinates for P2(R)? Sketch that part of the curve that lies in the
part of P2(R) given by x = 1, indicate in this sketch those points of the curve that lie in the
original R2.
27. Let C be a non-singular conic in P2(C). Prove that there are linearly independent
homogeneous quadratics (i = 1, 2,3) such that each point of C is of the form
[f , (X,1+):f 2(X,µ):f 3(), µ))

28. Given five points in P2(F) no three of which are collinear, prove that there is a unique non-
singular conic passing through them. What happens when three of the points are collinear?
Are there any special cases?
29 Let V be a vector space over the field F and W: V F2 be a linear map of rank 2 with kernel
K. Show that there is an induced map
,p: P(V) \ P(K) -. P'(F).
If a, b, c, d are collinear points in P(V) \ P(K) show that (ab,cd) - (,pa,pb,gccd). By constructing
suitable maps pp, show that if a, b, c, d are points on a non-singular conic C in P2(F), then
the cross-ratio (ab,cd) can be defined as the cross-ratio of the lines joining any point p e C
to a, b, c, d.
30.' Consider the space X of all lines in P3(F). If I is the line joining [ao:a,:a2:a3) and [bo:b,:b2:b3),
let Pi be the 4 x 4 skew-symmetric matrix with entries
a ai
del b' p,]
{b,
Show that these entries satisfy
Po,P23 + Po2P31 + Po3P12 = 0 V)
78

Show that t -. PP defines a map X Q where 0 is the quadric in P5(F) whose equation
is ('). Conversely show that each point 0 corresponds to a unique line in P3(F), so that X is
isomorphic to Q.
31. Let Q be a non-singular quadric in P3(R), that is Q = {xlx'Ax = 0) where A is a non-singular,
symmetric real 4 x 4 matrix. By putting A into diagonal form, show that if 0 is not empty then
it is homeomorphic either to a torus T2 or to a sphere S2.
32.' Consider a surface X in P3(C) defined by a non-singular quadratic form. Show that by using
a projectivity its equation can always be put in the form xox3 = x,x2. Deduce that through every
point x of X there are two lines gg,, e2 c X. If m,, m2 are the lines through a different point
y of X, show that one of m,, m2 meets e , in one point and the other does not meet e,. This
gives two families of lines on X.
Find a surface in R3 with a similar property.
33. Investigate the subsets of P5(R) corresponding to the five types of real conics.
34. (For those with a knowledge of the calculus of several variables.) Consider the map
h: S5 R defined by
S(a,b,c,d,e,f) = abc + 2def -ad 2-bee-cf2.
At which points of S-'(0) is S regular? Show that the image of 8-'(0) in P5 is P2 x P21- where
(x,y) _ (y,x). Is P2 X P2/-- a manifold?
35. Consider the family of conics x2 + y2 = tz2 in p2(R) and P2(C), as t varies. If t k 0, a the
conic is non-singular. Describe how it degenerates to a singular conic as t -+ 0 and t -. .

Repeat with the family x2 + ty2 = z2.

36. What is the intersection of the Veronese surface (page 69) with a linear P3(F) C P5(F)?
37. Consider four lines in P2, that meet in six points. Label the six points as a, a', b, b', c, c' in
such a way that a and a' do not lie on any one line and similarly for b, b' and c, c'. Let C
be a non-singular conic such that a, a' and b, b' are polar pairs with respect to C. Prove that
c, c' are also polar pairs.
Hence or otherwise show that a triangle and its polar are in perspective, that is the lines
aa', Wand cc' are concurrent.
38. Define a map I: R2 \ {0) -+ R2 \ (0) by lOIxH.HOxl = 1 and Ix is on the half line Ox. Show that
1 maps circles to circles except that circles through 0 are mapped to lines and vice versa.
Is I conformal?
The point 0 is 'sent' to infinity under I. Find a space X containing R2 and a map T: X -. X
such that 1I R2 \ (0) is 1. Can you write down a formula for I?
 Part III

Hyperbolic Geometry
The Parallel Axiom
Euclid gave certain axioms that were meant to characterize plane geometry. For twenty centuries
much controversy surrounded one of these axioms - the parallel axiom. It was felt that this axiom
was not as basic as the others and considerable efforts were put into attempts to deduce it from
the other axioms. Before the nineteenth century the most successful attempt seems to have been
made by Saccheri in 1773. He produced a thorough investigation of the parallel axiom, in fact he
claimed to have deduced the parallel axiom from the others, but his otherwise excellent piece of
work had one flaw in it. At the beginning of the nineteenth century Gauss, in work that he did not
publicise, showed how a large body of theorems could be deduced from a variant of the parallel
axiom, thus discovering hyperbolic geometry. Because Gauss did not publish his work, this
discovery is usually associated with the names of Bolyai and Lobachevski who published the same
discoveries in 1832 and 1836 respectively. Despite the fact that Euclid's parallel axiom was shown
to be necessary several defects were found, during the nineteenth century, both in the axioms and
logic used by Euclid. For a discussion of Euclid's axioms and their defects, the reader may consult
M.J. Greenberg's book "Euclidean and Non-euclidean geometry".
t
The parallel axiom of the Euclidean plane may be stated as: given a line and a point p a( Mere
isauniquelinemwithpemand
The corresponding fact for the projective plane is: given a line t and a point p t there is no
line mwith pemand to m=$.
The only alternative (assuming homogeneity) is: given a line t and a point p t t there is mo
than one line m such that p e m and Y n m= 4,.
It is this third possibility that holds in hyperbolic geometry. At first sight it seems hard to imagine
such a geometry. The first simple 'model' of such a geometry was discovered by Bettrami in 1868.

The Beltraml (or projective) Model

Let C be a non-singular conic in P2; as we have seen C divides P2 into two parts - an inside
(homeomorphic to R2) and an outside (homeomorphic to the Mdbius band M). The inside will be
denoted by H and is the space of the geometry, it remains to describe the lines of the geometry.
A line in H is a chord of the conic C, that is H n e where t is a line in P2. The conic C is called
the absolute conic of H. To obtain a picture let the conic C bean ellipse in R2.

C
80

For a point p e H with p k e , there are many lines of the geometry not meeting t. So, the third
possibility for the parallel axiom holds.
Two lines are parallel if they meet on C. they are ultraparallel if they meet outside C.
An obvious group that sends lines of H to lines of H is the group
Gc = (f a C).
That is, it consists of the projectivities of P2 that send the absolute conic C to itself. Such a
projectivity must send the complement of C to itself. This complement has two components, and
they are not homeomorphic to each other. So, a projectivity t that sends C to itself must send H
to itself.
It turns out that there is a metric on H, such that the group Gc is the group of direct isometries.
It is not at all clear that there should be such a metric. The next section attempts to give general
reasons why there might be such a metric, and later the metric is constructed.
The reader who only wants to see how the metric is constructed may skip the next section.

Groups of plane isometries

We analyse the 'size' of the groups of isometries of the Euclidean and elliptic planes. Then we
show that Gc has the same size. We measure the size by answering the following question. Given
two points p, q of the geometry and two lines(, m such that p e t and q e m, how many isometries
f are there such that fp = q and f f = m?
a) The Euclidean plane
There is a translation that sends p to q and composing with a rotation about q shows that there
is at least one isometry f with the required property. There are exactly four such, all obtained from
each other by first composing with reflection in (, reflection in a line perpendicular tot, and the
halt turn about p which is the composite of the two reflections.
b) The elliptic plane
The isometres of the elliptic plane form the group PO(3), and this is isomorphic to SO(3) the
group of rotations of S2. So we consider SO(3) acting on S2. Given a point p E S2 and a great circle
f containing p, it is clear that there is a rotation sending the pair p, t to any other such pair q, m.
It remains to show that there are exactly four rotations that take p, t to itself or to -p, e, because
these are the only points that are identified under the projection from S2 to the elliptic plane P2.
Consider the diagram

p,ree
Op lOr
Or lOs
OP IIDS
 81

The only rotations that send p top and t to t are the identity and the halt-turn about Op. The only
rotations that send p to -p and C to t are the half-turns about Or and Os. These four rotations give
four distinct elements of PO(3) because SO(3) -+ PO(3) is an isomorphism.
c) The Bettraml model
To prove the result in this case we need two results of projective geometry.
1) Given two triangles abc and a'b'c' in P2(R) and points x e ab, y e ac, x' a a'b', y' a a'c',
there is a unique f e PGL(2,R) sending x, y, a, b, c to x', y', a', b', c' respectively. (Problem 12
page 76.)
2) Given a triangle abc and points x e ab, y e ac, there is a unique conic C such that the triangle
abc is self-polar with respect to C and such that x e C, y e C. (see page 73).

Let C be the absolute conic of the model. Given p e t, there is a unique self-polar triangle pp1 P2
such that p, e t : P2 is the polar of t and p, is the point where the polar of p meets 4. Similarly let
qq1 q2 be self-polar with q, a m.
Choose x, y such that x e t nC, y e pp2 n C. Let (XI A2} = m n C and (y1,y2) = gg2n C.
By 1) above there is a unique projectivity fq sending p, p1, p2, x, y to q, q1, q2, x,, yi. By construction,
x, e m n C, y, a qq2 n C and gq,g2 is sect-polar with respect to C. The conic fC has the same
properties, but by 2) above, there is a unique such conic.so fC = C. Hence there are four elements,
f, 1, f1 2, f21 and f22 of Gc that send p,[ to q,m.
We will construct the metric on the Beltrami model but first we will construct another model, the
Poincar6 model. To motivate this we need to study stereographic projection.
82

Stereographic Projection
This is the map that identifies the two dimensional sphere S2 c R3 with P' (C). Here we will regard
it as a map S: S2 \ (n) --p C, where n is the north pole of the sphere. There are several variants
of the definition. We choose the one where C is the plane through the equator of S2 (any parallel
plane would do almost as well). Define Sp = p' if n, p and p' are collinear.

This map can be described in co-ordinates by

S(x,y,z) = (x+iy)/(1-z) forz k 1.
Exercise Show that two points p, q e S2 are antipodal, that is p = -q if and only if

S(p)S(q) = -1.

Stereographic projection S has two important properties that we will need:

1. Conformality, in other words it preserves angles between curves.
2. Circle preserving, a circle on S2 gets mapped to a circle in C except that a circle through n
gets mapped to a straight line in C.
The main features of the proofs of these properties will now be given.

1. Let a be the tangent plane to S2 at p and pr be a line in n. Let p'r' be the image of pr under
S. The plane n makes the same angle with C as it does with the tangent plane at n.
 83

The line np' is perpendicular to the line it n C, because it is perpendicular to the intersection
of the tangent planes v and rr and v is parallel to C. Hence the angles rps and r'p's' are equal
- as was required to be proved.
2. Consider a circle C on S2 not through n, the cone that is tangential to the sphere along C has
its vertex at x say, we will show that the stereographic projection of C is a circle whose centre
is the projection y of x onto the plane C.

This diagram is in the plane of n, p and x; xq is drawn parallel to p'y. We will show that the
length p'y is independent of where p is on the circle C. Using the symmetry of pp' with respect
to the planes rr and C, it is clearthat the angles qpx and qp'y are equal. So [pxl = igxi. Therefore
jp'yi = Inyl.IIgxJ/Inxi = inyi.ipxilinxi. As ipxi is constant for p e C we see that ip'yJ is constant
as required.
84

The case where the circle C goes through n is much simpler.

Clearly the projection of the circle S2 n iris the line C n rr.

The PoincarA Model

We will define this model as the image of the Beltrami model under certain projections. Let D
be the open disc of unit radius in C,
D= (z F. CI!zI<1)
and L be the lower hemisphere of S2
L= {(x,y,z)eS2IZ<0}.
Clearly stereographic projection S defines a map S: L-. D.
There is another projection V: L--. D given by vertical projection, in co-ordinates,
V(x,y,z) = x + iy.
The Beltrami model can be taken to have D for its space of points and the chords of the boundary
circle to be its lines.
The Poincard model has D for its space but its 'lines' are nothing like ordinary straight lines.
Its lines are defined to be the images of the lines of the Beltrami model under the composite
SV-':D-.L--D,
where we take the Beltrami model with its boundary conic C to be the unit circle. Notice that V-'
takes lines in D to arcs of circles in L that are orthogonal to the boundary. Using the two properties
of the stereographic projection S given above, it is clear that these arcs in L are mapped by S to
arcs of circles in D that are orthogonal to the boundary.
 85

Examples of lines in the Poincard model are:

One could define the Poincar8 model directly in this way: its space is D and the lines are arcs of
circles perpendicular to the boundary. For the Poincart model it is not quite obvious even that there
is a unique line through any two points, although of course one knows it is true because it is true
in the Beltrami model. The information can be transferred from one to the other using the map SV
and its inverse VS-1.
The reason for introducing the Poincare model is that it is very important in mathematics. It was
Poincare's discovery of this model that brought hyperbolic geometry from a mere curiosity to an
object playing a central role in many parts of mathematics. Of particular importance is the
relationship with complex analysis.
By Theorem 17 the fractional linear transformations form the group of all bianalytic
transformations of P' (C). It is interesting to consider the group of bianalytic transformations that
send the unit disc D (which we have also called L in P' (C)) to itself. It can be calculated explicitly:
Exercise Prove that a matrix La d] satisfies the condition ;zI < 1 implies jaz+bl < 1, that is,
it sends D to itself if and only if
b
[c d] S,
with a=8,(3=yand X=ad -bc,so aa-PO =1.
We will use a different model for D in which it is easier to draw pictures and for which the
calculations are slightly easier.
As we saw, the unit disc D can be regarded as the lower hemisphere L of the Riemann sphere.
This set L can be rotated round to be a hemisphere that is bounded by the great circle that projects
to the real axis. There are two such hemispheres, we will take the one that projects to the upper
half plane
H= Imz>0}.
This rotation that sends L to H is a fractional linear transformation (by Problem 14, p. 77). In fact
it is given by f: D H where f(z) = (z+ 1)/i(z-1), its inverse is r' (z) = (z-i)/(z+ i).
Note that f(O) = i, f(1) = -, f(-1) = 0, f(ef°) = -cot(9/2).
86

j o--w-- -- -

H
0
f

The hyperbolic lines in D become semicircles in H whose centres are on R or they become half
lines perpendicular to R (these are the hyperbolic lines in D meeting the boundary of D at z = 1.
in H they can be regarded as semicircles with centres at x). It is straightforward to calculate the
group of bianalytic transformations sending H to itself:
Theorem 21 The fractional linear transformation f defined by [a b sends H to itself if and only
if a, b, c, d are all real multiples of a complex number µ and (ad-bc)0. In other words PSL(2,R)
is the subgroup of PGL(2,C) that sends H to itself.
Proof If f maps H to itself, it must map R to itself as R is the boundary of H, in other words
(az+b)I(cz+d) is real for each real z. It is now straightforward to calculate that a, b, c, d must all
be multiples of one complex number µ, in particular they can all be chosen to be real, so
f c PGL(2,R). If ad-bc < 0, H would be mapped to the lower half plane, so (ad-bc) must be positive;
dividing by a suitable constant ensures that ad-bc= 1 and sotePSL(2,R).

In this proof we have considered the boundary of H. In P'(C) the boundary is a circle and is
naturally P'(R) c P'(C). Every hyperbolic line in H meets P'(R)in two points. If one of them is at
infinity then in the picture it is a straight line perpendicular to R.

Hyperbolic lines in H
 87

When one regards H in P'(C) all hyperbolic lines are the same. The group PSL(2,R), being a
subgroup of PGL(2,C), preserves angles and circles and hence sends hyperbolic lines to
hyperbolic lines.
A useful result for handling hyperbolic lines is the following.
Theorem 22 The four points z1, z2, z3, z4 a P'(C) lie on a circle if and only if the cross-ratio
(z 1 z2,z3z4) is real.
Proof By Theorem 19, there is a unique f e PGL(2,C) such that f(z1) f(z2) = 0, f(z3) = 1.
The map f sends circles to circles (Theorem 18) so f(z4) lies on the circle through 0, 1 but this
circle is P' (R) so (z1z2,z3z4) = f(z4) is real.
Conversely, if i' a P'(R) then cc, 0, 1, K lie on a circle, hence so do z1, z2, z3, z, because f-'
preserves circles.
This Theorem could have been proved by a direct calculation.
Note We have tacitly assumed in the proof that the four points are distinct. If they are not then
they are always concylcic, also the cross-ratio is always 0 or - and hence is real.
Corollary If q, r E P'(R) then p e H lies on the hyperbolic line through q, r if and only if (gr,pp)
is real.
Proof The hyperbolic line is half of a circle C whose centre is on P(R), and if p e C so is p. The
converse is also clear.
Example We check directly that plies on the line through 0 and cc if and only if (0=,pp) is real.
(0c,pp) = (0-p)(c-P)/(O--P)(cc_p) = p/p = p2/pp

so (0cc,pp) is real if and only if p2 is real. Let p = x + iy, p2 is real if and only if xy = 0 but
y k 0 as p e H so p2 is real if and only if x = 0 - as required.
Now we are preparing to prove that the group PSL(2,R) is a group of direct isometries of H.
Theorem 23 The group PSL(2,R) is transitive on H, it is even transitive on pairs (p,f) such that
p e! . The stabilizer of a pair (p,4) has two elements.
Proof Given p f t and q e m, let t meet P' (R) in a, b and m meet it in c,d.

By Theorem 19 there is a unique f e- PGL(2,C) such that fp = q, fa = C, fb = d. By Theorem

18, 1 preserves circles; hence f sends I to m. Also f is conformal so it sends a circle orthogonal
to t to another circle orthogonal to m; hence it sends PI(R) to itself; so by the proof of Theorem
21, f e PSL(2,R). This shows that PSL(2,R) is transitive on pairs (p, i.) such that p e L.
The only elements of PGL(2,C) that send p to p and t to 4 are the identity and a transformation
88

g such that gp - p. ga = b, gb a (again by Theorem 19). Clearly g2p = p, g2a = a and

g2b = b, sog2 = identity.
Our conclusion is that PSL(2,R) is analogous to a group of direct isometrics of H. To get the
(larger) group of all isometries one adds the isomorphism z --. 1/2 and considers the group of
transformations generated by PSL(2,R) and this extra element. Notice that z -. 1/2 also preserves
circles and angles (up to sign). It sends P'(R) to itself and i to itself. When we need to use this
larger group we denote it by PSL'(2,R).
We could now proceed to define a metric on H by using our previous formula. If p, q e H and
the hyperbolic line through p, q meets P'(R) in a, b then define d(p,q) = Iog(pq,ab)I/2. This distance
is preserved by the whole group PSL'(2,R) and it can be verified that this is the group of all
isometries of the Poincard model H. For ease of calculation and for novelty we take a different
approach.

The Local Metric

Given a point p e H. what is the element of distance dsp at p? Such an element of distance is
called a local metric and is exactly what is required to define geodesics locally. Clearly ds, must
be given by an expression of the following form
dsP2 = a(p)dx2 + 2b(p)dxdy + c(p)dy2,
for some real valued functions a, b, c defined on H. It is somewhat more convenient to rewrite this
as
ds2 = A(z)dz2 + B(z)dzd2 + C(z)d22 for z e H (')
Exercise By comparing these two expressions; prove that B(z) is real and that A(z) = E(-z-). Use
these facts to simplify the following proof.
Theorem 24 The only metrics of the form (') that are invariant under the group PSL(2,R) are of
the form dsz2 = k(dx2+dy2)/y2 where z = x + iy and k is a positive constant.
Proof The metric must be invariant under the transformation z -. z + X = w for X F. R. In this
case dz = dw and it is easy to see that then the functions A. B, C of (') are independent of x.
The metric is also invariant under z - xz = w for x > 0. In this case dw = adz so one gets that
A(Az)X2 dz2 = A(z)dz2
and hence A is homogeneous of degree -2 in z. Similarly B, C are also homogeneous of degree
-2 in z, but all three are independent of x, so there are constants A, B, C so that
A(z) = A/y2, B(z) = B/y2, C(z) = C/y2.
Finally, the metric is invariant under z -1 /z = w. In this case dw = dziz2 -nd
w = u + iv = -(x-iy)lizl2,
soy = zj2v. We have A(w) = A/v2 and A(z) = A/y2, so
Adze/y2 = Adze/v2z4 = AIz14dz2/y2z4
Hence (z4-Iz14)A = 0 for all z e H, so A = 0. Similarly C = 0. So dsz2 = B(dx2+dy2)/y2. Clearly
B>0.
When one has a metric on a locally Euclidean space (more precisely, on a smooth manifold)
one can define a local metric ds by considering the lengths of small line segments near a point.
This can be viewed as differentiation. Conversely if a local metric ds is given on such a space, one
can define a global metric by integration. The distance between two points p, q is defined to be
if Ids,
h
where the infimum is taken over all paths a from p to q. This is often called the geodesic distance
 89

- the length of the shortest path between two points. This all works well when the space is
compact but when the space is not compact the distance may not be realised as the length
of any curve. For example, consider R2 \ (0) and the points to be (1,0) and (-1,0).

Areas
Similarly one can consider area elements, in our case the area element at the point
z = x + iy e H is clearly dxdy/y2. We will now use this formula to calculate the area of a
hyperbolic polygon in H, that is, an area bounded by a finite number of hyperbolic lines of H.
An example of such a polygon is the following:

By our definition, the area is

I7
F

To evaluate this we use Green's theorem

J ax I
by letting u = V y and v = 0, so au/ay = -1 /y2. This gives
udx + vdy

dxdy
I
To evaluate this integral one calculates the value of the right hand side as one traverses a
segment of a hyperbolic line - which is an edge of F. Let L be a segment of a hyperbolic line
(whose euclidean radius is r: L is the portion between angles 0 and a as shown below)
90

Then one has that

[dx
y
hence this integral is independent of the radius r. Note that the integral vanishes if L is a vertical
line. This can be seen from this calculation or directly because dx vanishes.
With this preliminary we can now calculate the area of F. Suppose its edges are L,, Lz, ..., I.,,,
with Lk defined by the angles (3k, yk - the radius of the corresponding hyperbolic line is not needed.
If Lk is a segment of a vertical line we take (3k = Yk (= 0). The conclusion is that the hyperbolic
area of F is r. (Pk-Yk). This sum can be simplified by considering the change in the direction of
the outward normal to F as we traverse the boundary aF in an anticlockwise direction. Let the vertex
at which Lk meets 44.1 be Ak and let the internal angle there be ak.

As we traverse Lk the change in the direction of the outward normal is Yk - Yk. At the vertex Ak
the change is >r-0k. So the total change as we traverse around F is
Mr + (Yk Yk-k)
 91

But the total change is clearly 2;r, so one has the equation
21r = mrr +1 (Yk-00 - £ ak
But Area(F) = E ()3k-yk)
so one deduces that Area(F) = r(m-2) - £, ak
So the area of a hyperbolic polygon depends only on its internal angles.
In particular a hyperbolic triangle whose angles are a, p, yhas its area given by
Area(L) = a-(a+p+y).
It is interesting to compare this with the area of an elliptic triangle (given on page 64) which is
(a-I3+Y)-rr.
This gives a way in which the three plane geometries can be distinguished and is closely
connected with the'curvature' of the space:
Given any triangle of a plane geometry, let its angles be a, p, y. One can decide which plane
geometry it belongs to by considering the sign of (a+p+y)- it. Similar measurements can be
carried out in three dimensional space - this has been done to try to discover the curvature of the
space that we live in. Non zero results that are outside experimental error have not been found
- so that the curvature may be zero. The study of invariants such as curvature is called differential
geometry and plays an important role in general relativity.
We have the summarising table:
(a+p+y)-a Geometry Curvature
<0 Hyperbolic negative
>0 Elliptic positive
0 Euclidean flat
Unfortunately, it is cumbersome to calculate the trigonometry of hyperbolic figures using the
Poincare model. It is easier to use a slight variant of the Beltrami model, this we now do.

Trigonometry
We are going to study hyperbolic and elliptic (or spherical) trigonometry (I presume that the
reader knows about euclidean trigonometry but it will not be used here). Some preliminary work
is common to both cases and anyway the two are very similar.
From now onwe will have a fixed non-degenerate symmetric bilinear form on R3 which will be
denoted x.y (rather than the more cumbersome x'Ay used in the discussion of conics). Using the
diagonalization of quadratic and bilinear forms over R, choose a basis e, , e2, e3 forR3 so that
e;.e) = E,ti,i with E, _ ±1.
Three points x, , x2, x3 a R3 determine a 3 x 3 matrix (f;,i) by x, = I E,ei. Define
voI(x1,x2,x3) to be det(f;;).
Note vol(x1,x2,x3)2 is independent of the choice of basis because it equals det(x,'x1).
(In the case of the standard inner product, vol(x,x2x3) is the volume of the parallelepiped with edges
Ox 1, Ox2, Ox3.) To define the vector product we use the following general result:
Lemma Suppose that x.y is a non-degenerate bilinear form on R", then for every linear mapping
f: R" - R there is a unique x, a R" such that f(y) = x,.y for all y e R.
Proof Choose a basis e, , e2, ..., an for R" such that
e,.e1= e,S;i for all i, j.

Define x, to be ± e,X,e, where > =1(e,) for each i.

92

Then x,.e, = j e,X,e,.e, e,x,e;&,i =X, = f(e).

By linearity x,.y = f(y) for ally.
The uniqueness of x, follows from the non-degeneracy of x.y.
Corollary For every x,y a R3, there is a unique vector x x y e R3 such that (xxy).z = vol(x,y,z)
for all z e R3.
Proof The map z -. vol(x,y,z) is linear and so the result follows.
Proposition i) (xx) x y = X(xxy)
ii) yxx=-xxy
iii) (xxy).z = (yxz).x = (zxx).y = x.(yxz) = y.(zxx) = z.(xxy).
These all follow from standard properties of the determinant.
Now definer to be e,e2r3 and note that a is independent of the choice of the basis e,, e2, e3
because it is the sign of the determinant of the matrix of the bilinear form.
Proposition x x(yxz) = e[(x.z)y-(x.y)zJ.
This can be proved by following a proof of the usual case. The proof that follows is just one of
many such proofs.
Proof It is clear that both sides are linear in each of the three variables. It is therefore enough
to check that the formula holds when the three variables are taken to be the basis vectors. By the
symmetry of e,, e2. e3 we may take x to be e,. If either a) y = z or b) neither y nor z equals
e, then the left hand side vanishes; it is also clear that the right hand side vanishes in both these
cases. When y and z are interchanged, both sides change sign. Hence it suffices to consider
y = e, and z = e2, e3. The formula is easily checked in both these cases.
Corollary (xxy).(yxz) =
Proof (xxy).(y)<z)= [y x (yXz)].x= r.[(yz)(yx)-(yy)(z.x))
In both the applications given in these notes a will be +1 so the reader need not worry unduly
about it.

Hyperbolic Trigonometry
We will work with the Beltrami model whose absolute conic is
x,2-x22-x32=0.
The inside of the conic is the set of points [x,:x2:x3] in P2(R) for which x12 - x22 - x32 > 0 and this
is the model H c P2(R) for the hyperbolic plane, the lines are the intersections of the projective lines
with H.
Representatives for the points of H can be chosen on the hyperboloid x,2 - x22 - x32 = 1 in R3,
this hyperboloid has two sheets, each point of H has two such representatives. We will always
consider the representative on the sheet with x, > 1 (the other sheet has x, -- -1). The points on
the outside of the absolute conic each have two representatives on the hyperboloid of one sheet
X,2-X22-X32 =-1.
There is no simple way of singling out one of these representatives.
 93

We will always have in mind the fixed bilinear form

x.Y = x1y1-X2Y2-X3Y3
Two distinct points x, y e P2(R) determine a plane (x,y) in R3, (x,y) is spanned by the
representatives of x and y. The bilinear form restricted to the plane (x,y) can be of one of three
types:
1. Negative definite (type (0,2), that is, no positive terms and two negative terms in its
diagonatized form). In this case the line xy does not meet the absolute conic.

Case 1

2. Negative semidefinite (type (0,1)). In this case the line xy is tangent to the absolute conic (at
z, say). The representatives of z are the non-zero vectors in the plane (x,y) that satisfy
iz=0.
3. Indefinite (type (1,1)). In this case the line xy crosses the absolute conic.
Using the discriminant of a quadratic form q on R2, it is easy to prove the following algebraic
criteria
type of q criteria
(2,0) x.x > 0 and (x.x)(y.y) > (X. y)2

(1,1) (x.x)((x.y)2 -(x.x)(yy)) > 0

(0,2) (x.x) < 0 and (x.x)(y.y) > (x.y)2

degenerate (x.x)(y.y) = (x.y)2.
94

Lines and Polarity

Lemma If t is a line in H, its pole a lies outside C.
Proof Suppose x e 2 and x.x > 0, then there is an a such that x + as e C so
0 = (x+aa). (x+aa) = x.x + 2aa.x + a2a.a; a.x = 0 and x.x > 0
hence a.a < 0.
Conversely, if a is outside C, its polar line t must cross C: otherwise every x e t satisfies
x.x < 0. Every point in P2(R) can be written as as + (3x for some x e E and then
(aa+px).(aa+3x) < 0 because a.x = 0, hence the form would have to be negative definite on the
whole of R3, which it is not. This argument also proves
Proposition Consider the point x e P2(R), then
x lies inside C if and only if its polar does not meet C
x lies on C if and only if its polar is tangent to C
x lies outside C if and only if its polar crosses C.

As in the Euclidean case, the space of all lines in H is a Mbbius band because there is a
homeomorphism (obtained by taking the pole) between the space of lines and the space of points
lying outside C; the outside of a conic is homeomorphic to a Mbbius band.
Notation A line in H will be denoted by al where a is its pole.
The two choices of the representatives of the pole of the line t on the hyperboloid x.x -1
correspond to the two directions of £. If we wish to consider the direction of a line t then we will
write it as `a= for a particular choice a for the representative of the pole a oft.
Proposition The lines al, b- meet, are parallel or utlraparallel according as the plane (a,b) is
of type (0,2), (0, 1) or (1, 1 ),
Proof The plane {a,b} represents the polar of al n b'. The lines al, bl meet if and only if
al n bl is inside C, they are parallel if and only if al n bl lies on C and they are ultraparallel if
and only if aln bl lies outside C.
It al, bl meet then (a,b} is of type (0,2) and a.a = b.b = -1, one has that (a.b)2 < 1. Hence
the following is well defined:
Definition If a^, b- meet then the angle between them is defined to be 9 where cost) = a.b.
We now define the distance between two points of H:
Proposition If x, y e H then x.y -,1.
(Note that His taken ambiguously to be the inside of the absolute conic
or{xcR3[X.x=1andx,-- 1).)
Proof Let x, y e H and m = (x+y)/2 be the mid point of the line segment xy in R3. By the convexity
of the hyperboloid, m.m -- 1. Hence (x+y).(x+y) , 4, so x.y -> 1. x
Now we can define the metric don H.
(Ir
Definition It x,y a H, let x.y = cosh d(x,y) with d(x,y) --0.
Proposition If xyz is a hyperbolic triangle then c b

cosh b = cosh a cosh c - sinh a sinh c cos t; (1 )b

V
c os(3 = -co s y cosy + cos h b s in s sin a 1) ". Y a z
There are identical formulae (1)e, (1)c, (1). , (1)y for cosh a, cosh c, cosa and cosy respectively.
In the proof we will need two preliminary results.
 95

Lemma Let x. y e H be distanced apart, so x.y = cosh d, then x x y has magnitude sinh d and
it is the pole of the directed line xy.
Proof Clearly (x x y).x = (x x y).y = 0, so the line xy is the polar of x x y.
Also (x x y). (x x y) = -(x x y). (y x x) = 1- (x.y)2 by the Corollaryon page 95.
Hence (x x y).(x x y) = -sinh2d as required.
If x. y, z are three non-collinear points in H then the directed angle e enclosed by the line
segments xy and yz has the same sign as (x x y). (y x z).
So cos e = (x x y).(yx z)/sinh d(x,y)sinh d(x,z).
Proof of the proposition Consider the identity
(x )<Y) (y x z) =
given by the Corollary on page 92 and interpret the terms. The left hand side is
cos(3 sinh c sinh a
and the right hand side is cosh c cosh a-cosh b. The formula (1)t, follows.
Corollary The triangle inequality is satisfied by d.
Proof By the proposition
cosh b < cosh a cosh c + sinh a sinh c = cosh(a+c)
But cosh x is an increasing function for positive x, hence b < a + c as required.
To prove formula (1)a, we consider the poles of the sides of the triangle. The pole of the line
xy is (x x y)/sinh c. Let (x x y) = t sinh c, (z x x) _ 1sinh b and (y x z) = +:sinh a. Then
Tl)
But(txj).(Ex71) 1= sin2-Y

So z siny.x sina = cos- cosy + cost, and as x.z =cosh b, formula (1)A follows.
Proposition If xyz is a hyperbolic triangle right angled at y, then
i) coshb = cosh a cosh c,
ii) cosy = cosh a sin-f,
cosy = cosh c sins,
iii) sinh a = sinh b sins,
sinh c = sinh b sin-y.
Proof The formulae i) and ii) follow immediately from the above formulae (16 (1). and (1)),
To prove formula iii), consider (1),
cosh a = cosh b cosh c -sinh b sinh c cosa,
substitute cosh a cosh c for cosh b from i), use cosh2c = 1 + sinh2c and divide by sinh c to get
cosh a sinh c = sinh b cosa.
Now substitute cosh a sin-y for cosy from ii) and divide by cosh a to give the result.
Proposition (The analogue of the law of sines).
If xyz is a hyperbolic triangle, then
sinh a/sina = sinh b/sins = sinh clsiny
To prove this we need to drop a perpendicular from x to the line yz. We will prove that this can
be done after giving the rest of the proof. x
Proof Let p be the foot of the perpendicular and
h = d(x,p). By iii) above applied to the two right
angled triangles, one gets
sinh h = sinh c sin(3
and sinh h = sinh b siny
96

So sinh b sin-y = sinh c sinO and the result follows.

Lemma If al is a line in H and b is a point not on al, then there is a perpendicular from b to a'
and the distance from b to al is d where sinh d = a.b.
Proof

The line ab is perpendicular to al, because (b x a).a vanishes.

Let p be the point where al meets the line ab. Then p = Ab + µa with p.a = 0, p.p = 1 and
cosh d = b.p. Expanding everything out yields
=1 a(a.b) -µ. = 0 X + µ(a. b) = coshd.
Sox 1 /cosh d and (a.b) = sinhd.
This lemma also gives us a very famous formula of hyperbolic geometry.
Let a, b be as above, then there are two lines through b that are parallel to al, they meet a'
on the absolute conic.

These two lines make equal angles 0 with the perpendicular by from b to al. This angle is called
the angle of parallelism, it is given by the following equivalent formulae:
 97

sing = sech d, tan6/2 =

Proof Consider the right angled triangle bpx. By the formulae for such a triangle,
cos fi = cosh d sina.
In the limit, as x tends toy, the angle (i tends to zero, hence
1 = cosh d sing.
This gives the first formula and the fact that the two angles of parallelism are equal.
Let t = tanei 2, then sinO = 2t/(1 +t2).
But sech d = 2/(ed+e-d).
Hence (e('+e'°) = t + t-'sot = e' or e'.
But 0 < 0 < 7/2 so 0 < tane/2 < 1, hence t = e-'d.

Isometries
Definition The group 0(3') consists of the linear transformations of R3 that preserve the form
x.y, that is,
O(3')= (TeGL(3,R)ITx.Ty=x.yforallx,ycR3).
The elements of 0(3') define mappings H -+ H that are isometries of H with the metric defined
by cosh d(x,y) = x.y. The elements of O(3') also preserve angles.
Proposition The group 0(3') acts transitively on pairs (x,e) where x is a point of the line 1.
Proof If x is any point in Handy a x- L, let z = y.Ln xl. Consider the linear transformation defined
by Tx = (1,0,0), Ty = (0,1,0) and Tz = (0,0,1). As y, z e xl and xl does not meet the absolute
conic we have y.y = z.z = -1. Hence T E 0(3'). The line T2 passes through (1,0,0). It is clear that
0(3') contains all rotations about the x, axis, so there is an S e 0(3') such that S(1,0,0) _ (1.0,0)
and STt is any given line through (1,0,0).
The image PO(3') of 0(3') in PGL(3,R) preserves the conic and preserves cross-ratios. Hence
PO(3') is the group of isometries of H.
In hyperbolic geometry there are many other interesting constructions and we end by considering
some of them.
Proposition For every n , 3, there is a regular n-gon with angle a for any a satisfying
0 < a < (n-2)a/n.
Proof Consider the Poincare disc model - in which the lines are arcs of circles perpendicular to
the edge of the disc. Taken rays emanating from the centre, with equal angles between them. Take
points P,, P2, ..., P on each ray at equal distances d from the centre. Then P, P2 ... P. is a regular
n-gon. When d is small, the angle a is close to (n-2)a/n and when d is large, the angle a becomes
close to 0. By continuity any value of a between 0 and (n-2)aln is attained for some value of d.
98

Corollary For every n 3 5, there is a regular n-gon all of whose angles are right angles.
Of course, these particular n-gons we have constructed can be moved around using elements
of 0(3') to give a large number of others.
As a further example we consider the trigonometry of a right angled hexagon.

Proposition In a right angled hexagon with side lengths as shown one has
cosh 0 = -cosh a cushy + sinh a sinh y cosh b.
Proof As the angles are right angles, we have that al.0l al.-y1= 0 where al denotes the pole
of the side labelled a, etc. So ( y both lie on the line a. So we have:
al

As usual we use the identity a

but we need a preliminary result
Lemma Two lines&, min the hyperbolic plane are ultraparallel if and only if they have a common
perpendicular. Their distanced apart along the common perpendicular is given by
coshd =C1.m1 .
 99

Proof Remembers, m are ultraparallel if and only if they meet outside C. Let p = e n m, then
p . a 1= 0 so p1 is orthogonal to t, similarly it is orthogonal tom hence p' is a common perpendicular
Conversely suppose n is a common perpendicular then n'.t'= 0 and n'.m' = 0 so
n 1= t n m, but n' is outside C, sot, m are ultraparallel.
Nowletx,ybethepoints snnandmn n. We must show that x.y=tl.m1.
Clearlysl = \[(x.y)x-(x.x)y] and m' = µ[(y.y)x-(x.y)y).
The constants x, µ are determined by ml.m1= -1 and are x2 = µ2 = -1 /(1-(x.y)2)
But 1'.m' = x .I1-(x.y)2)(x.y).The result follows.
So al.b1 = cosh y, a1.cl = cosh p, bl.c' = cosh a and a1.a1= bl.b1= -1.
(a1 x b1) = (al.a1)(bl.b1)
= 1 -cosh2y =-sinh2y
and(blxcl).(blxcl) = -sinh2a.
Therefore (al x b1).(b' x c1) = sinh a sinhy cosh band the formula follows.
Spherical and elliptic trigonometry can be treated in a similar manner, for completeness we now
discuss this.

Elliptic Trigonometry
The elliptic plane is denoted by P2 and representatives of its points are vectors
x e S2 = {x a R3j jjxjj = 1 }. Both x and - x represent the same point of P2. The distance between
two points x, y e P2 is the (acute) angle between the lines Ox, Oy in R3. In symbols, cos d(x,y) x.yI
X
where x.y is the (usual) Euclidean inner product of R 3 .
We now study the trigonometry of an elliptic triangle xyz.
Let a = d(y,z), etc. and a = the angle at x. etc. as shown.
Theorem In an elliptic triangle cos b = cos a cos c + sin a sin c cosp -0b.
There are similar formulae (1). and (Dc for cos a and cos c.
Proof We use the identity (xxy).(yxz) = (x.y)(y.z) - (x.z)(y.y) = cos c cos a-cos b
Now -(xxy).(yxx) _ -(x.y)2 + (x.x)(y.y) = - cosec + 1 =sin 2C
and iiy x zir = sin2a.
The angle between the lines polar to (y x x) and (y x z) is p.
So (x x y). (y x z) _ -sin a sin c cosp, giving the result.
Corollary The triangle inequality: d(x,z) , d(x,y) + d(y,z).
Proof cosb = cos a cos c + sin a sin c cosp >- cos acos c - sin a sin c = cos(a+c).
Hence b , a + c because cos is a decreasing function of x for 0 E x < a.
One can obtain further formulae by considering the polar triangle. (Remember that if the line is
the equator of the sphere the pole is the north (or south) pole, this probably explains the use of
the words pole and polar.) The edges of the polar triangle of xyz are the polars of x, y, z. Let the
polar triangle be x'y' z' where x' is the point where the polars of y and z meet etc. Let its edge lengths
be a', b', c' and its angles be a', p', y'.
Proposition a + a' = rr, p + b' = n and y + c' = n.
Proof Suppose the lines xy and xz meet y'z' in p, q respectively. Then a equals d(p,q).
100

The polars of x and z meet at y', the point q is collinear with x and z so its polar passes through
y', hence d(q,y') is 7r/2. Similarly d(p,z') = rr/2 so d(p,q) + d(y',z') = rr, proving the result.
Corollary cos p = -cos a cos y + sin a sin y cos b 1Qp

and also ,
Q},
Proof Apply the trigonometric identity that we have already proved to the polar triangle and use
the proposition.
When we apply these formulae to a right angled triangle we get:
Proposition if xyz is a triangle right angled at y, then
i) cosb=cosacosc
ii) cosa=cosasin y
cosy = cos C Sin a
sin a = sin b sin a
iii)
sin c = sin b sin y.
Proof i) Put cos p = 0 in (D b.
ii) Put cos = 0, sin p = 1 in t
iii) Use 1)6: cos a = cos b cos c + sin b sin c cos a
Substitute cos a cos c for cos b (from i)) and use coszc = 1 - sin2c. Divide by sin c to get
cos a sin c = sin b cos a. Substitute cos a sin y for cos a (from ii)) and divide by cos a to give
the result.
Proposition (Law ofsines)
Let xyz be any triangle, then
sin a/sina = sin b/sina = sin c/siny.
Proof Drop the perpendicular from x to yz
By iii) of the previous proposition
sin h = sin csinp
and also sin h = sin b sing.
So sin b siny = sin c sing and the result now follows easily.
 101

Note Euclidean triangles are very close to elliptic triangles with very short sides, so one can
recover the usual Euclidean trigonometric formulae by suitable approximations (remember the
angles are not small!). For example by putting cost = 1 -1212 and sinL = Z with a, b, c in
the formula a one gets b2 =a 2+ c2-2ac cos(3.
Exercise Show how the Euclidean cosine rule can be derived from the formulae of hyperbolic
trigonometry by a similar approximation.
102

Problems
1. Consider a disc D2 made of a material such that the speed of light at a point p is proportional
to the Euclidean distance of p from the boundary. Prove that the light rays are the hyperbolic
lines of the Poincar6 model.
2.' Certain map projections from parts of S2 to the plane are conformal. They can be described
as fS where f is a conformal map from R2 to itself and S is stereographic projection. When
f(z) = -i log(z), show that one gets Mercator's projection.
A theorem on map projections states that, in a neighbourhood of a point, any two map
projections differ, up to first order, by a linear map: R2-. R2. Prove this.
3.' Show that the co-ordinates in P2(R) can be chosen so that the equation of a non-singular conic
C is xz = y2. Prove that any elment of PGL(3, R) that maps this conic to itself has the form
1a2 2ab b21
T = ac ad+bc bd J
c2 2cd d2
The map a; + i,l -. [a;2+,12:k:1 j defines a homeomorphism h: H, y H2 where H, is the upper
half plane and H2 is the interior of C. If f e PSL(2,R) is z -. (az+b)/(cz+d), show that
h-'Th = f. This gives an explicit isomorphism between PSL(2,R) and Gc. Use this to prove
that Gc is the group of all direct isometries of the Beltrami model.
4. The circle of radius r and centre a in a metric space X is the set {x I d(x,a) = r). If the absolute
conic C is a unit circle in R2, what is the shape of a circle under the Beltrami metric? [First,
consider the case where a is the centre of C.] Show that a circle in the upper half plane model
is an Euclidean circle but with a different centre.
5. Given p e t, let pq be perpendicular tot. Consider the hyperbolic circle whose centre is at
q and touches 4 at p. As q tends to infinity along the perpendicular one gets a limiting curve
called a horocycle (in Euclidean geometry the horocycle equals t but this is not the case
in hyperbolic geometry). If the absolute conic is a circle C in R2 take t to be a diameter of
C and p to be the centre of C and describe the corresponding horocycle. In the upper half
plane model, describe the horocycle when p = i andt is the line x = 0.
6. It p e 8, define d(p,t) to be d(p,q) where pq is perpendicular tot. A hypercycle is the locus
of points equidistant from a line, that is (xjd(x,e) is constant). Find the equations of some
hypercycles for the Beltrami at d upper half plane models.
7.' Let V be a finite dimensional vector space over R and (p be a symmetric, non-singular, bilinear
form on V. Let O(y) be the subgroup of GL(V) that preserves cp, show that det T = ±1 if
T E O(cp). (We have previously considered the Euclidean case, when the signature of ip is
± dim V.) Prove that O(,p) is compact if and only if ip is definite (the.Euclidean case). When
Sign(4p) * ±dim V, v e V is isotropic if ip(v,v) = 0, describe the subset of all the isotropic
vectors. O(,p) acts on the projective space P(V), how many orbits does it have? If
Sign(g) = dim V - 2, show that O(p) has four components and that if V is two dimensional
each component is homeomorphic to R. When dim V = 4, O((p) is called the Lorentz group
and the set of isotropic vectors form the light cone (c.f. relativity).
 103

8.' Take dim V = 2 and use the notation of the previous question. Find a basis such that
pP(x,y) x,y2 + x2y,. The isotropic vectors form two linese,,e2; prove that m, n are orthogonal
lines with respect tow if and only if the cross-ratio of the four lines is -1.
Prove that O(c) acts on the set of half lines through the origin and that it has
four orbits on each of which the action is simply transitive. Show that the matrix
0_,, that sends m' to n' satisfies x-2. Hence one can define
angles by the formula angle (m' ,n) = '/Zlog(e, 42,m' n' ). Continue with this approach to obtain
a model for the hyperbolic plane.
9.' U(p,q) is the set of non-singular transformations T: CP*Q --* CP+Q that preserve the form
Z,Z, +. . . + zPZP-zP+,;P+1 -... -44Q1PjQ
Show that U(p,q) is a group; for which p,q is it compact?
Describe the matrices in U(1,1). If T is the matrix show that
Cj 101
T-' AT e U(1,1) if At SL(2,R). What is the exact relationship between U(1,1) and SL(2,R)?
10. If p, q, r, a, b are points in P' (F), show that
(pq,ab)(qr,ab) = (pr,ab)
Let H be the inside of the unit circle C in R2. If p # q are two points in H, let a, b the points
where the line pq meets C, define
2d(p,q) = Ilog(pq,ab)I.
If pqr is a triangle, right angled at q, verify that d(p,r) > d(p,q). Deduce that d is a metric on
H. Verify that d is invariant under the group Gc.
If H is the unit disc in the plane x, = 1 in F113, let n denote the projection of H from 0 onto
the sheet x, -_ 1 of the hyperboloid x,2-x22-x32 = 1. Verity that n is an isometry.

11. Draw pictures of surfaces in R3 that have negative curvature somewhere. Can they have
negative curvature everywhere?
Consider the tractrix
x=sechu, z=u-tanhu, y=0
The tracloid is obtained by revolving the tractrx about the z-axis. What is its curvature? Draw
pictures of geodesics on the tractoid.
104

12. Let xyz be a hyperbolic triangle, right angled at y. It has five quantities associated with it.
namely, a. y, a, b and c. Show that given any two of these quantities, the other three are
determined. (There are six essentially different cases to consider.)
13. A hyperbolic triangle has six quantities associated with it, namely, a, a, b, and c. The
triangle 'can be solved' from given data if all six quantities can be calculated from the given
data. Show that one can solve the triangle if one is given a, 0, y as data. From which other
triads of the six quantities can the triangle be solved?
14. A right angled pentagon (all angles are irl2) has edges of length a, b, c, d and e, labelled
cyclically. Prove that
sinh a sinh b = cosh d.
Show that a right angled hexagon can be divided into two right angled pentagons. Deduce
a law of sinhs for the hexagon.
15. Consider the points (1,0,0), (cosh u, sinh u, 0) and (cosh v, sinh vcosa, sinh v sina) on the
hyperboloid x12 - x22 - x32 = 1 for suitable u, v to prove that there is a hyperbolic triangle
whose angles are any given positive numbers a, p, y such that a+ p +,y < ir.
16. An ideal triangle is one such that at least one of its vertices is on the ideal conic C.
i) If x E C and y, z t H, find a relation between a, p and y. What is the area of this triangle?
ii) If x, y E C and z f. H, can one say anything about y?
iii) If x, y, z E C and x'. y, z' E C, prove that there is an isometry in Gc that sends one triangle
to the other.
17. It p. q, r are positive integers, prove that there is a triangle in one of the three plane geometries
with angles Trip, rlq, rr/r. Verify that the space of the relevant geometry can be covered by
such triangles, any two triangles only meeting along edges or at vertices. Such a covering
is called a tesselation of the plane. Draw pictures for the cases (p,q,r) = (2,3,5), (2,3,6) and
(2,3,7).
18. In the upper half plane model, if p = i and q = led with d > 0, show that d(p,q) = d. If pqr
is a triangle, find the locus of r if
i) a=n/2
ii) p - n/2
iii)' -y = n/2 (do the calculation in the hyperboloid model)
where a, p, -y are the angles at p, q. r respectively.

19. Consider the curve x2 = 0 on the sheet of the hyperboloid x,2 - x22 - x32 = 1 given by
x, -- 1, this is a line a in H. Verify that the map f: R -, tdefined by
f(u) = (cosh u, 0, sinh u)
is an isometry. Deduce that every hyperbolic line is isometric with R.
 105

Further Reading
The best introduction, particularly for the group theoretic approach to Euclidean Geometry, is
H. Weyl, 'Symmetry', (Princeton University Press). It is an easy book to read and it deals with
symmetry in many contexts.
A classic book containing a wealth of material at a reasonably low level is
D. Hitbert and S. Cohn-Vossen, 'Geometry and the Imagination', (Chelsea Publishing Company).
It is very readable and its style is discursive.
A text book that covers similar ground to these notes but without assuming as much from other
parts of mathematics is
H.S.M. Coxeter,'Introduction to Geometry', (J. Wiley and Sons, Inc.).
A much more comprehensive account written from a similar viewpoint to that taken here is
M. Berger,'Geometrie' (5vols.), Nathan).
For an account of the axiomatic approach to geometry at about the level of these notes the reader
should read
M.J. Greenberg,'Euclidean and Non-Euclidean Geometries', (W.H. Freeman and Co.).
Other books that contain much relevant material are
L. Fejes Toth,'Regular Figures', (Pergamon Press).
N.H. Kuiper,'Linear Algebra and Geometry', (North Holland Publishing Co,).
1.R. Porteous,'Topological Geometry', (van Nostrand and Cambridge University Press).
H. Eves, 'A Survey of Geometry' (2 vols.), (Allyn and Bacon, Inc.).
Most books on the history of mathematics contain many geometrical topics. A good example is
H. Eves, 'An Introduction to the History of Mathematics', (Holt, Reinhart and Winston).
A great deal of mathematical insight can be gained by learning about the history of the subject.
There are many other excellent books on geometry. The above are a sample that the reader
may wish to look at first.
106

List of Symbols
Bij(S) the group of bijections of the set S.
S" Bij(S) where S = {1,2,...,n }, the symmetric group.
F" the vector space of all n-tuples of elements of the field F.
R" n-dimensional Euclidean space.
M(n,F) the ring of all n x n matrices over F.
GL(n,F) the group of all invertible n x n matrices over F.
SL(n, F) the group of n x n matrices over F whose determinant is 1.
x.y an inner product on R" or C".
I
li the corresponding norm on R" or C".
d a metric.
O(n) the group of all orthogonal n x n matrices.
SO(n) the group of orthogonal n x n matrices with determinant 1.
Aff (X) the affine span of the set X.
T. (n) the group of upper triangular real n x n matrices with positive diagonal entries.
S"-' the space of vectors of norm 1 in R" (the unit sphere).
R reflection in the hyperplane H.
I(R") the group of all isometries of R".
T. translation by the vector a.
R(a,a) rotation through angle a about the point a.
G(R,a) glide by the vector a along the line it.
is inversion in the point a.
M MObius band.
S(X) the symmetry group of the set X.
D" the dihedral group with 2n elements.
Sd(X) the rotation group of the set X.
A" the alternating group on n letters.
sign the sign of a permutation.
J central inversion.
T tetrahedron.
C cube.
D dodecahedron.
the golden ratio.
Stab(x) the stabilizer of x (under a group action).
Orb(x) the orbit of x (under a group action).
L lattice in R".
D" the space of vectors of norm < 1 in R", the unit disc.
H the space of quatemions.
i,j,k unit quatemions.
U(n) the unitary group.
SU(n) the special unitary group.
H. half turn about the point a.
Aff(R",R") the set of all affine maps f: R"-+ R".
 107

P^ real projective n-dimensional space.

P(V) the projective space associated to the vector space V.
P"(F) P(F"-' ), so P" = P"(R"'') = P"(R).
[x:y.zj homogeneous co-ordinates on p2
V. Hom(V, F), the space of linear functionals on V.
PGL(n,F) GL(n.F) (x1), (the subgroup (h 1) is the centre of GL(n,F)).
=(pq,i the cross-ratio of the points p, q, r, s EP' (F), it equals (p-r)(q-s)i(p-s)(q-r).
PSL(2,R) the image of SL(2,R) -. GL(2,R) -. PGL(2,R)
G the group of the Beltrami model whose absolute conic is C.
H the hyperbolic plane.
a the polar line of the point a.
the pole of the line e .
0(3') the subgroup of GL(3,R) that preserves the form x.y = x, y, - x2y2 - x3Y3
108

Index
absolute conic 79 half turn 45
affine basis 7 harmonic 71
affine map 8 hexahedron 24
affine span 7 homogeneous co-ordinates 52
affine subspace horocycle 102
Z
hypercycle 102
affinely independent Z
hyperplane 7
affinity 47
alternating group 28 icosahedron 25.26
angle in Beltrami model 94 inner product 3
inverse quaternion 42
angle of parallelism 96
inversion 18
Beltrami model 79.94 isometry 4
bianalytic isomorphism 64 isometry group of R" 11
central inversion 29
isotropic 102
centrally symmetric solid 22 lattice 37
circle 6,17 local (Poincare) metric 88
conic 65,67 meridian of torus 47
conjugate quaternion 43 metric in Beltrami model 94
convex polyhedron 23 MObius band la
cross ratio 61 MObius transformation 59
crystal group 39 norm 3
crystallographic restriction 32 octahedron 24
cube 24.30 open MObius band la
curve 62 open solid torus 18
Desargues' theorem 57 opposite isometry 12
dihedral group 22 orbit 36
direct isometry 12
orthogonal 3
orthogonal group 4
direct symmetry group 28
orthonormal 4
dodecahedron 25,26,27,31
dual solid 97 Pappus's theorem 74
parallel hyperbolic lines 8n
edge 23 parallel on torus 4Z
elliptic plane 63 Pascal's theorem 74
Euclidean metric a Platonic solid 23
even permutation 28 Poincar6 model 84
face 23 point group 412

fractional linear transformation 59

polar 69
polar line 69
general linear group 3 pole 35,69
geodesic 62 polyhedron 23
glide 1315 projective group 59
glide reflection 13 15 projectivity 59
golden ratio quaternion 42
Gram-Schmidt process fi
 109

reflection 10,13,15
regular polyhedron 23
rotation 13,15
rotation group 2$
rotatory inversion 16
rotatory relfection 16
screw 15
self polar triangle ZQ
sign of permutation 26
simple group 42
solid torus 1Z
special linear group 3
special orthogonal group 4
special unitary group 4a
stabilizer 35
stereographic projection 53,59.82
symmetric group 26
symmetry group 2]
tangent to conic 70
tesselation 104
tetrahedron 24.28
torus 1Z
translation 13.15
transposition 26
triangle of reference Z]
ultraparallel lines $Q
unit quaternion group 43
unitary group 43
upper half plane model 85.86
Veronese surface 66
vertex 23
Universitext
Aksoy, A.; Khamsi, M. A.: Methods in Fixed Boltyanskii, V. G.; Efremovich, V.A.: Intui-
Point Theory tive Combinatorial Topology
Alevras, D.; Padberg M. W.: Linear Optimi- Booss, B.; Bleecker, D. D.: Topology and Ana-
zation and Extensions lysis
Andersson, M.: Topics in Complex Analysis Borkar, V. S.: Probability Theory
Aoki, M.: State Space Modeling of Time Se- Carleson, L; Gamelin, T. W.: Complex Dy-
ries namics
Arnold, V.1.: Lectures on Partial Differential Cecil, T. FL: lie Sphere Geometry: With Ap-
Equations plications of Submanifolds
Audin, M.: Geometry Chae, S. B.: Lebesgue Integration
Aupetit, B.: A Primer on Spectral Theory Chandrasekharan, K.: Classical Fourier
Transform
Bachem, A.; Kern, W: Linear Programming
Duality Charlap, L. S.: Bieberbach Groups and Flat
Manifolds
Bachmann, G.; Narici, L; Beckenstein, E.:
Chern, S.: Complex Manifolds without Po-
Fourier and Wavelet Analysis
tential Theory
Badescu, L.: Algebraic Surfaces
Chorin, A. I.; Marsden, J. E.: Mathematical
Balakrishnan, R.; Ranganathan, K.: A Text- Introduction to Fluid Mechanics
book of Graph Theory
Cohn, H.: A Classical Invitation to Algebraic
Balser, W.: Formal Power Series and Linear Numbers and Class Fields
Systems of Meromorphic Ordinary Differen- Curtis, M. L: Abstract Linear Algebra
tial Equations
Curtis, M. L: Matrix Groups
Bapat, R.B.: linear Algebra and Linear Mo-
dels
Cyganowski, S.; Kloeden, P.; Ombach, J.:
From Elementary Probability to Stochastic
Benedetti, R.; Petronio, C.: Lectures on Hy- Differential Equations with MAPLE
perbolic Geometry
Dalen, D. van: Logic and Structure
Benth, F. F.: Option Theory with Stochastic
Das, A.: The Special Theory of Relativity: A
Analysis
Mathematical Exposition
Berberian, S. K.: Fundamentals of Real Ana- Debarre, 0.: Higher-Dimensional Algebraic
lysis Geometry
Berger, M.: Geometry I, and II Deitmar, A.: A First Course in Harmonic
Bliedtner, I.; Hansen, W.: Potential Theory Analysis
Blowey, J. F.; Coleman, J. P.; Craig, A. W. Demazure, M.: Bifurcations and Catas-
(Eds.): Theory and Numerics of Differential trophes
Equations Devlin, K. J.: Fundamentals of Contempo-
Burger, E.; Gradel, E.; Gurevich, Y.: The Clas- rary Set Theory
sical Decision Problem DiBenedetto, E.: Degenerate Parabolic Equa-
Bottcher, A; Silbermann, B.: Introduction to tions
Large Truncated Toeplitz Matrices Diener, F.; Diener, M.(Eds.): Nonstandard
Boltyanski, V.; Martini, H.; Soltan, P. S.: Ex- Analysis in Practice
cursions into Combinatorial Geometry Dimca, A.: Sheaves in Topology
Dimca, A.: Singularities and Topology of Gustafson, S. J.; Sigal, I. M.: Mathematical
Hypersurfaces Concepts of Quantum Mechanics
DoCarmo, M. P.: Differential Forms and Ap- Hahn, A. J.: Quadratic Algebras, Clifford Al-
plications gebras, and Arithmetic Witt Groups
Duistermaat, 1. 1.; Kolk, I. A. C.: Lie Groups Hcjek, P.; Havr6nek, T.: Mechanizing Hypo-
thesis Formation
Edwards, R. E.: A Formal Background to Hi-
gher Mathematics Ia, and lb Heinonen, J.: Lectures on Analysis on Metric
Spaces
Edwards, R. E.: A Formal Background to Hi-
gher Mathematics Ila, and Ilb Hlawka, E.; Schoi&ngeier, J.; Taschner, R.:
Geometric and Analytic Number Theory
Emery, M.: Stochastic Calculus in Manifolds
Ho!mgren, R. A.: A First Course in Discrete
Endler, 0.: Valuation Theory Dynamical Systems
Erez, B.: Galois Modules in Arithmetic Howe, R., Tan, E. Ch.: Non-Abelian Harmo-
Everest, G.; Ward, T: Heights of Polynomials nic Analysis
and Entropy in Algebraic Dynamics Howes, N. R.: Modern Analysis and Topo-
Farenick, D. R.: Algebras of Linear Transfor- logy
mations Hsieh, P-F; Sibuya, Y. (Eds.): Basic Theory
Foulds, L. R.: Graph Theory Applications of Ordinary Differential Equations
Franke, J.; Hdrdle, W.; Hafner, C. M.: Statis-
Humi, M., Miller, W.: Second Course in Or-
dinary Differential Equations for Scientists
tics of Financial Markets: An Introduction
and Engineers
Frauenthal, I. G: Mathematical Modeling in
Hurwitz, A.; Kritikos, N.: Lectures on Num-
Epidemiology
ber Theory
Friedman, R.: Algebraic Surfaces and Holo-
Huybrechts, D.: Complex Geometry: An In-
morphic Vector Bundles
troduction
Fuks, D. B.; Rokhlin, V. A.: Beginner's Course Isaev, A.: Introduction to Mathematical Me-
in Topology thods in Bioinformatics
Fuhrmann, P A.: A Polynomial Approach to Iversen, B.: Cohomology of Sheaves
Linear Algebra
Jacod, J.; Protter, P.: Probability Essentials
Gallot, S.; Hulin, D.; Lafontaine, J.: Rieman- Jennings, G. A.: Modern Geometry with Ap-
nian Geometry plications
Gardiner, C. F.: A First Course in Group Jones, A.; Morris, S. A.; Pearson, K. R.: Abs-
Theory tract Algebra and Famous Inpossibilities
Garding, L.; Tambour, T.: Algebra for Com- Jost, J.: Compact Riemann Surfaces
puter Science
lost, J: Postmodern Analysis
Godbillon, C: Dynamical Systems on Sur- Jost, J.: Riemannian Geometry and Geome-
faces tric Analysis
Godement, R.: Analysis I Kac, V; Cheung, P.: Quantum Calculus
Goldblatt, R.: Orthogonality and Spacetime Kannan, R.; Krueger, C. K.: Advanced Ana-
Geometry lysis on the Real Line
Gouvea, F. Q.: p-Adic Numbers Kelly, P.; Matthews, G.: The Non-Euclidean
Gustafson, K. E.; Rao, D. K. M.: Numerical Hyperbolic Plane
Range. The Field of Values of Linear Opera- Kempf, G.: Complex Abelian Varieties and
tors and Matrices Theta Functions
Kitchens, B. P.: Symbolic Dynamics Nikulin, V. V.; Shafarevich, I. R.: Geometries
Kloeden, P.; Ombach, J.; Cyganowski, S.: and Groups
From Elementary Probability to Stochastic Oden, 1. J.; Reddy, J. N.: Variational Methods
Differential Equations with MAPLE in Theoretical Mechanics
Kloeden, P. E.; Platen; E.; Schurz, H.: Nu- Oksendal, B.: Stochastic Differential Equa-
merical Solution of SDE Through Computer tions
Experiments
0ksendal, B.; Sulem, A.: Applied Stochastic
Kostrikin, A. 1.: Introduction to Algebra Control of Jump Diffusions
Krasnoselskii, M. A.; Pokrovskii, A. V.: Sys- Poizat, B.: A Course in Model Theory
tems with Hysteresis
Polster, B.: A Geometrical Picture Book
Luecking, D. H., Rube!, L. A.: Complex Ana-
lysis. A Functional Analysis Approach Porter, J. R.; Woods, R. G.: Extensions and
Absolutes of Hausdorff Spaces
Ma, Zhi-Ming; Roeckner, M.: Introduction
to the Theory of (non-symmetric) Dirichlet Radjavi, H.; Rosenthal, P.: Simultaneous
Forms Triangularization

Mac Lane, S.; Moerdijk, 1.: Sheaves in Geo- Ramsay, A.; Richtmeyer, R. D.: Introduction
metry and Logic to Hyperbolic Geometry
Marcus, D. A.: Number Fields Rees, E. G.: Notes on Geometry
Martinez, A.: An Introduction to Semiclas- Reisel, R. B.: Elementary Theory of Metric
sical and Microlocal Analysis Spaces
Matousek, J.: Using the Borsuk-Ulam Theo- Rey, WI. I.: Introduction to Robust and
rem Quasi-Robust Statistical Methods
Matsuki, K.: Introduction to the Mori Pro- Ribenboim, P.: Classical Theory of Algebraic
gram Numbers
Mazzola, G.; Milmeister G.; Weissman J.: Rickart, C. E.: Natural Function Algebras
Comprehensive Mathematics for Computer Rotman, J. J.: Galois Theory
Scientists i
Rubel, L. A.: Entire and Meromorphic Func-
Mazzola, G.; Milmeister G.; Weissman J.: tions
Comprehensive Mathematics for Computer
Ruiz-Tolosa, J. R.; Castillo E.: From Vectors
Scientists 2
to Tensors
Mc Carthy, P. J.: Introduction to Arithmeti-
Rybakowski, K. P.: The Homotopy Index and
cal Functions
Partial Differential Equations
Meyer, R. M.: Essential Mathematics for Ap-
Sagan, H.: Space-Filling Curves
plied Field
Samelson, H.: Notes on Lie Algebras
Meyer-Nieberg, P.: Banach Lattices
Mikosch, T: Non-Life Insurance Mathema- Schif , J. L.: Normal Families
tics Sengupta, J. K.: Optimal Decisions under
Mines, R.; Richman, F.; Ruitenburg, W.: A Uncertainty
Course in Constructive Algebra Seroul, R.: Programming for Mathemati-
Moise, E. E.: Introductory Problem Courses cians
in Analysis and Topology Seydel, R.: Tools for Computational Finance
Montesinos-Amilibia, J. M.: Classical Tessel- Shafarevich, L R.: Discourses on Algebra
lations and Three Manifolds Shapiro, J. H.: Composition Operators and
Morris, P.: Introduction to Game Theory Classical Function Theory
Simonnet, M.: Measures and Probabilities Tondeur, P.: Foliations on Riemannian Ma-
Smith, K. E.; Kahanpda, L.; Kekalainen, P.; nifolds
Traves, W: An invitation to Algebraic Geo- Verhulst, F.: Nonlinear Differential Equa-
metry tions and Dynamical Systems
Smith, K. T.: Power Series from a Computa- Wong, M. W.: Weyl Transforms
tional Point of View
Smorynski, C.: Logical Number Theory I. An
Xambo-Descamps, S.: Block Error-Cor-
Introduction recting Codes
Stichtenoth, H.: Algebraic Function Fields Zaanen, A.C.: Continuity, Integration and
and Codes Fourier Theory
Stillwell, /.: Geometry of Surfaces Zhang F.: Matrix Theory
Stroock, 1). W.: An Introduction to the Zong, C.: Sphere Packings
Theory of Large Deviations
Zong, C.: Strange Phenomena in Convex and
Sunder, V. S.: An Invitation to von Neumann
Discrete Geometry
Algebras
Tamme, G.: Introduction to ttale Cohomo- Zorich, V. A» Mathematical Analysis I
logy Zorich, V. A.: Mathematical Analysis II
Universitext

ELMER REES was born and brought up in Wales in a Welsh

speaking community. He studied at Cambridge University
and obtained his PhD in Algebraic Topology at Warwick
University in 1967. He has held teaching positions at several
UK Universities, Hull, Swansea, Oxford and since 1979 he has
been Professor of Mathematics at Edinburgh University.
He has been a member of the Institute for Advanced Study,
Princeton and held visiting positions in Berkeley, Sydney,
IHES, Paris and MPI, Bonn. He is a Fellow of the Royal Society
of Edinburgh, has been President of the Edinburgh Mathemat-
ical Society, Vice President of the London Mathematical Soci-
ety and the Forder lecturer in New Zealand. His main re-
search interests are in the applications of algebraic topology
to geometric problems. He enjoys travelling particularly to
visit collaborators and to participate in mathematical events.

Notes on Geometry
This book offers a concrete and accessible treatment of Euclide-
an, projective, and hyperbolic geometry, with more stress on
topological aspects than is found in most textbooks. The au-
thor's purpose is to introduce students to geometry on the ba-
sis of elementary concepts in linear algebra, group theory, and
metric spaces, ant to deepen their understanding of these top-
ics in the process. A large number of exercises and problems is
included, some of which introduce new topics.

ISBN 3-540-12053-x

11111111111
9"783540"120537

springeronline.com