Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Hyperbolic Geometry
Birger Iversen
Aarhus University
AMBRIDGE
UNIVERSITY PRESS
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, Sao Paulo, Delhi
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521435086
A catalogue record for this publication is available from the British Library
INTRODUCTION ix
I QUADRATIC FORMS
I.1 ORTHOGONALITY 1
1. EXERCISES 50
II GEOMETRIES
II.1 GEODESICS 57
11.3 SPHERES 65
H. EXERCISES 84
IV FUCHSIAN GROUPS
IV.1 DISCRETE SUBGROUPS 125
V FUNDAMENTAL DOMAINS
V.1 THE MODULAR GROUP 150
VI COVERINGS
VI.1 HYPERBOLIC SURFACES 171
VI EXERCISES 198
CONTENTS vu
SOURCES 291
BIBLIOGRAPHY 292
SYMBOLS 295
INDEX 296
INTRODUCTION
PARALLEL AXIOM Through a given point outside a given line there passes a
unique line which does not intersect the given line.
z H az+b E S12(R)
cz + d
In fact the full group of isometries of the Poincare half-plane is the group PG12(R).
In particular the group G12(Z) acts as isometries on H2, a fact which lies at the
root of the applications of hyperbolic geometry to number theory.
INTRODUCTION xi
The Poincare disc The open unit disc D is the second most commonly
used model for the hyperbolic plane. We shall not write down the metric here, but
mention that the lines in this model are traced by Euclidean circles and lines
orthogonal to the unit circle OD. The Poincare disc has certain aesthetic qualities,
as shown below by a one of the many beautiful illustrations from [Klein, Fricke].
complete system of generators and relations for the group. In this way the topic
has had a big influence on what today is known as combinatorial group theory.
The theorem of Poincare is one of the main themes of this book. The proof of the
theorem is based on some geometric ideas which I shall outline next.
It follows that the boundary OHn is a sphere. In the end we find that the Lorentz
group acts on the sphere as the group of Mobius transformations.
INTRODUCTION xiii
;a,dER,bEC
tr2(S) = 4 <m,n>2
where m and n denote normal vectors for the geodesics a and b.
The realisation of Minkowski space through the space M of Hermitian
matrices has been used in physics for a long time. We shall introduce another tool
from physics namely the Dirac algebra, which is a concrete realisation of the
Clifford algebra of M. The use of Clifford algebras was suggested by the "strange
formulas" from the book of [Fenchel].
In the first two sections we lay down the general principles for quadratic
forms and the associated orthogonal groups. The third section will classify real
quadratic forms into Sylvester types, while the remaining sections will deal with
real forms of specific types.
Three types of real quadratic forms are of particular interest: positive
definite forms, parabolic forms and hyperbolic forms. The corresponding
orthogonal groups provide us with the isometry groups of the three basis
geometries: spherical geometry, Euclidean geometry and hyperbolic geometry.
For example, the orthogonal group of a hyperbolic form contains the
Lorentz group as a subgroup of index 2. The Lorentz group, on the one hand, is
the isometry group of hyperbolic geometry, while, on the other hand, it can be
identified with the somewhat older group of Mobius transformations. We shall see
that "the Mobius group is the boundary action of the Lorentz group".
1.1 ORTHOGONALITY
1 In this text we have applications only for the field R of real numbers,
the field C of complex numbers and occasionally the field Q of rational numbers.
2 QUADRATIC FORMS
is bilinear in x and y. The relation 1.2 between the bilinear form <x,y> and the
quadratic form Q(z) is called the formula for polarization. Observe that the
bilinear form <x,y> is symmetrical in x and y.
Proof A quadratic form Q(z) satisfies the formula Q(2z) = 4Q(z) as a special
case of 1.1. We can now substitute z for x and y in formula 1.2 to get that
which recovers the function Q(z) from the symbol <x,y>. Conversely, if we start
with a symmetrical bilinear form <x,y> we can use formula 1.4 to define a k-
valued function Q on E which is homogeneous of degree 2. Bilinearity and
symmetry give us the formula
<x + y, x + y> = <x,x> + <y,y> + 2<x,y> x,y E E
which shows that Q(z) = <z,z> is a quadratic form.
1.5 K1 = {eEEjVxEK<e,x>=0}
Proof Let us recall that a linear form on E is nothing but a k-linear map
:E--+k. Linear forms may be added and multiplied by a constant from k to form
a new vector space, the dual space E*. Let us prove that
To this end we simply pick a basis el,...,ell for E and introduce the coordinate
forms xl,...,xn E E* by the formula
ei x1(e)e1 ;eEE
Let us verify that any linear form on E satisfies the formula
= 1: i (e;) xi E E*
To this end we simply evaluate both sides of the formula on the basis el,...,en.
This shows that xl,...,xn generates the vector space E*. To see that xl,...,xn are
linearly independent, consider a relation of the form E )i x1 = 0 , A1,...,An E k.
Evaluate the relation on ej, j = 1,...,n, and conclude that Ai = 0.
Observe that the kernel for q is E 1 and use that Q is non-singular to conclude
that q is injective; formula 1.8 allows us to conclude that q is surjective as well.
Restriction of a form along the inclusion i:K- E defines a k-linear map
i*:E* K* ;fHoi,EE*
4 QUADRATIC FORMS
which is surjective: In the argument above pick the basis el,...,en for E such that
el,...,ek is a basis for K and observe that the coordinate forms for this basis are
xt o i,...,xk o i. We shall focus on the composite map
E q E* -, K*
Proof Let us pick a basis el,...,e. for E and let A denote the matrix for o,
thus o(ej) = E; A;ie1 for j = 1,...,n. Direct calculation gives us
<o(ei)e'Y(el)> = < E kAkiek, > hAhleh> = E h,k TAik<ek,eh>Ahi
which shows that the Gram matrix for Q o o is TAGA. In particular we find
that o is an orthogonal transformation for Q if and only if the matrix A obeys
1.12 T
AGA=G
Let us observe that the Gram matrix G is the matrix for the map q:E _ E* with
respect to the basis e1,...,en for E and x1,...,x11 for E*. Since Q is non-singular we
conclude that del G # 0. The formula 1.12 gives us
del TA del G del A = del G
Proof Let us observe that Q can't be identically zero since this implies that
the bilinear form <x,y> is identically zero. Thus we can find a vector v E E with
Q(v) # 0. Let us write Q(v) = A2, A E C, or Q(A-lv) = 1 to conclude that we can
find e E E with Q(e) = 1. Let K denote the line spanned by e and put F = K 1
and observe that E = K ® F (meaning that any vector x E E has a unique
decomposition x = f + k, where f E F and k E K). It is easily seen that the
restriction of Q to F is non-singular. It is left to the reader to complete the proof
by simple induction on dim(E).
6 QUADRATIC FORMS
Now, add these two formulas to see that r(e) = f. If e + f is isotropic, reflection p
along this vector interchanges e and - f. In conclusion the orthogonal trans-
formation rip maps e to f as required.
Proof Let us first treat the case where U is non-singular. This will be done
by induction on dim(U). In case dim(U) = 1 let us pick a non-zero u E U and put
A = Q(u). Since Q(o (u)) = A, we can apply 2.3 to extend v to an orthogonal
transformation of E.
To accomplish the induction step let us pick a non-isotropic vector e E U.
Since Q(e) = Q(v(e)) we can use 2.3 to pick r E O(E) such that o(e) = r(e). It
will suffice to extend r-1a: U,E to an orthogonal transformation of E. To recapi-
tulate it suffices to extend v:U-E under the assumption that a fixes a non-
isotropic vector e E U. Let us introduce the total orthogonal space W to L = k.e
in U and the total orthogonal space F to L in E. Observe that W and F are non-
singular and that W C F and o -(W) C F. Thus we can use the induction
hypothesis to extend the restriction a:W-F to an orthogonal transformation of F.
Let us now assume that U is singular. Let there be given data consisting
of a non-zero vector e E U fl U 1 and a linear complement R to L in U. We are
going to show that there exists an isotropic vector fin E which is orthogonal to R
and such that <e,f> = 1. Observe that we can find ,Q E E" with /3(e) = 1 and
/i(R) = 0 ; the vector b E E with q(b) _ /3 is orthogonal to R and satisfies
<b,e> = 1. Let us show that the vector f = b - 2<b,b>e is isotropic
<f,f> = <b,b> - <b,b><e,b> = 0
8 QUADRATIC FORMS
In this section we shall deal exclusively with quadratic forms over the field
R of real numbers. By a Euclidean vector space we understand a finite dimen-
sional vector space E with a quadratic form Q which is positive definite, i.e.
Q(e) > 0 for all non-zero e E E. The form Q is called negative definite if Q(e) < 0
for all non-zero e E E. Our first objective is to generalise the concept of an
orthonormal basis.
Proof Let us use induction with respect to dim(E) = n. If the form Q is identi-
cally 0, any basis for E is orthonormal. Thus we may assume the existence of an
e E E with Q(e) # 0. Let us write Q(e) = eat with e = ± 1 and A E R*. Put
el = A-le to get Q(e1)=± 1. Let us use the induction hypothesis to find an ortho-
normal basis e2,...,e. for (Re1) -1' . The basis e1,...,e1, meets our requirements. 0
Proof Let us fix the basis as in the statement of the proposition, and let E_
denote the subspace of E, generated by those elements e; from our basis for which
<e;,e;> = 0, - 1. Observe that Q(e) < 0 for all e E E_ and conclude that for any
Euclidean subspace F of E we have F fl E_ = 0. According to 3.4 we have that
10 QUADRATIC FORMS
This shows that q is independent of the basis chosen. Let us apply this to the
space (E,-Q) and conclude that p too is independent of the basis.
With the notation of the theorem we say that the quadratic form (E,Q)
has Sylvester tune (-p,q). Observe that two quadratic forms (E,Q) and (F,R)
with dim(E) = dim(F) of equal Sylvester type are isomorphic.
Proof Let us apply the Grassmann dimension formula to the linear map
f:U®V,E, f(u,v)=u-v ;uEU,vEV
observing that Ker f U fl V and Ini f = U + V.
This shows that sign detij<ei,ej> is independent of the basis considered. We leave
it to the reader to verify the formula in the case of an orthonormal basis.
Proof To begin let us just assume that the principal minors are different from
zero and let us show how to determine the signature of A. With the notation
above the subspace Ei of E generated by e1,...,ei is non-singular since del Ai # 0.
Let us pick a non-zero vector fi E Ei, orthogonal to Ei_1. The Gram matrices of the
two bases el,...,ei and el,.... ei_1,f1 have the same sign, thus
sign del Ai = sign <fi,fi> sign del Ai_1
In the case at hand we deduce that <fi,fi> > 0 for i = 1,...,n.
12 QUADRATIC FORMS
Proof The inequality is trivial if e and f are linearly dependent. If e and f are
linearly independent, they span a Euclidean plane and we get from 3.5 that
<e,e> <e,f>
det >0
<f,e> <f,f>
In the rest of this chapter we shall be concerned with the orthogonal group O(E).
The basic example of an orthogonal transformation is reflection r in a linear
hyperplane H. If n E E is a unit normal vector for H we have that
has index 2 in 0(E). We have seen that O(E) operates transitively on the sphere
We ask the reader to verify that SO(E) acts transitively on S(E), dim E > 2.
14 QUADRATIC FORMS
Evaluation of the determinants show that the first matrix defines a rotation, while
the second defines a reflection. It follows from the addition formulas for the
trigonometric functions that the first matrix in 4.8 defines an isomorphism
between the circle group SO(E) and the group R/2irZ.
<e,e> <e,f>
det > 0
<f,e> <f,f>
This is trivial when e and f are linearly dependent. When e and f are linearly
independent we get from 1.12 that the sign of the determinant is unchanged if we
replace e and f by an orthonorinal basis for the plane they span. 0
Let us say that a space (F,Q) is parabolic if Q is positive and the space
F 1 is of dimension 1. The isotropic line F 1 is invariant under orthogonal
transformations. It follows that we have a morphism of groups
5.4 µ:0(F)-;R*
where the multiplier p(o) E R* denotes the eigenvalue of o E 0(F) on the isotropic
line. This allows us to introduce a subgroup of 0(F) of index 2
Let us observe that the antipodal map 1, belongs to the centre of O(F)
and that µ(i) = -1. This defines a decomposition of the orthogonal group 0(F) of
a parabolic space F
This form is easily seen to be parabolic. The isotropic line is generated by the
vector (0,1) and is called the line at infinity. Let us use the inner product on F to
introduce the evaluation map ev: F x E . -> R
ev((x,l;),y) = <x,y> + ; x,y E E, l; E I$
The evaluation map identifies F with the space of affine functions on E.
Proof Let us first observe that a point e E E defines a linear form f ev(f,e)
on F with evaluation 1 against the isotropic vector (0,1). It is easy to see that any
linear form 0 on F with 4'(0,1) = 1 can be represented in this way and that such a
representation is unique.
Let us return to 0 E O(F) and observe that p(o)-1o is a linear automor-
phism of F which fixes the isotropic vector (0,1). It follows that for a given e E E
we can find a uniquely determined vector 0"(e) in E such that
ev(o(f),e) = µ(o) ev(f,o'(e)) ;fEF
Variation of e E E defines a transformation a :E-E. It follows at once that
(Qr) = T-Q , t- = 6 ; o,r E 0(F)
We conclude that o' is invertible and that the transformation i7 = 0'-1 satisfies
the required formula. In order to show that o is a similarity, we shall work our
way through three special cases.
10
A vector v E E gives rise to a transformation Ov E O(F) given by
Ov(z,O = (z,( - <v,z>) ; z E E, E 02
Coxeter matrix Let us consider an affine simplex D, i.e. the convex hull
of n+1 points eo,el,...,en, not contained in an affine hyperplane. Let
no, nt,..., nn
be the corresponding normal vectors: the evaluation of ns against ej is zero, i # s,
while the evaluation against es is positive. From this description it follows that
the normal vectors form a basis for F = E ® R. Let us write
and observe that the A is are strictly positive which follows from evaluation of
formula 5.9 on the vertices of the simplex. The Coxeter matrix for D is
<nr+ns>rs
Proof For s = 0,...,n let £s denote the linear form on E (D R with Qns) = as
and {$(nr) = 0 for s # r. The relation above ensures that s has evaluation 1 on
the isotropic vector (0,1). It follows from the remark made in the beginning of the
proof of 5.7 that we can find a point es E E with ev(f,es) = s(f) for all f E F. The
points eo,...,en define the simplex we are looking for.
Proof We have already seen that two similar simplices have the same Coxeter
matrices. With the notation above let m0,mi,...,mn be the normal vectors for the
simplex P enumerated such that
<nl.,n,> = <mr,ms> ; r, s = 0,1,...,n
Let v E O(F) be such that o (nl.) = o(mr), for r = 0,...,n. It remains to prove that
p(Q) > 0. To this end let us apply o- to the relation 5.9 to get that
)0m0 + Alm1 +... +.nrnn = (0,h(o,))
Let us evaluate this against an interior point p E P to get that
.Dev(m0,p) +... + µ(o,)
which shows that u(o) > 0 as required. 0
Proof Let us consider a real vector space F equipped with a basis eo,...,e. and
a quadratic form such that
<er,es> = a,,, ; r,s = 0,...,n
By assumption the form is positive but singular. We shall eventually prove that F
is parabolic. To this end we shall show that for a non-zero isotropic vector
n = E xsas we have as # 0 for all s = 0,1,...,n. From the inequality
Q(E xses) = E r,sarsxrxs >_ E r,s arslxr1Ixsl = Q(E Ix5Ies)
it follows that we can assume that all coefficients of n are positive. Let us for a
moment assume that the set J = { s I xs > 0) is different from 0,...,n. Let us recall
from 5.2 that the isotropic vector n is orthogonal to any vector:
0 = <n,es> = E r E J xr<el.,es> ; s = 0.... ,n
In particular we conclude that a1. = <er,es> = 0 for s 0 J, contradicting the
hypothesis that the matrix A is indecomposable.
Let us prove that F l has dimension 1. To this end let us introduce the
space N C R"+1 representing F 1
N = {(xo,...,x") E R"+1 I Q( Exses) = 0}
By the observations made above we conclude that the intersection between N and
the hyperplane x0 = 0 is zero; this implies dim(N) = 1. In conclusion if
Exses 0 is isotropic then all coefficients are non-zero of the same sign. Finally
we can use an isometry of F with E ® R to construct a basis for E ® R and apply
5.10.
1.6 LORENTZ GROUP 23
Lemma 6.2 Two points x and u of the hyperboloid satisfy the inequality
1 < I<x,u>I ; x,u E S(F)
Moreover, the inequality is sharp when x and u are linearly independent in F.
Proof Let us first prove that the plane R contains a vector of norm -1. Use
any orthonormal basis for F to exhibit a hyperplane E of Sylvester type (-n,0). It
follows from the dimension formula 3.4 that,
24 QUADRATIC FORMS
Proof Let us fix the point u E S(F) and let E denote the hyperplane orthogo-
nal to u. Since E has type (-n,0) we find that E separates S(F) into two parts
H={xES(F)I <x,u> > 0}, H-={xES(F)I<x,u><0}
These two parts are interchanged by reflection along u. We are going to prove
that stereo ranhic projection g:H-+E with centre -u maps H homeomorphically
onto the open unit disc
D={yEEI-<y,y><1}
The affine line through x E H and -u is parametrised (x + u)t - u, t E R. Its inter-
section g(x) with E is given by <(x + u)t - u,u> = 0 , which gives us
6.5 x - <x,u>u
g (x) = 1 + <x,u>
<g(x),g(x)> = 1 -+<x,u>
1
<x,u>
1.6 LORENTZ GROUP 25
From 6.2 we conclude that <x,u> > 1, so the formula above enables us to
conclude that g(x) E D as required. We ask the reader to work out that the inter-
section between H and the affine line through the points z E D and -u is given by
6.6 f(z) = 2
1+z <z,z>
+u
-u ; zED
The antipodal map x'-+-x is in the centre of O(F) but is not a Lorentz
transformation. It follows that Lor(F) has index 2 in O(F), more precisely
cosh s, sinh s)
coshs -sinks
sinks -coslcs
It follows that det(er) = -1 and lr(o-) = 0. This implies that a has eigenvalues 1
and -1. We ask the reader to verify that the basis e,f can be picked such that the
matrix for or corresponds to the case s = 0. This shows that a is a reflection along
a vector of norm -1. In the second case the matrix for o is
coshs sinks
L(s)_ ;sER
sinks coshs
From the addition formulas for the hyperbolic trigonometric functions it follows
that L(s) is a morphism of groups
This shows that the special Lorentz group is isomorphic to R when dim(F) = 2.
Let us turn to the action of the Lorentz group on the isotropic cone C(F),
i.e. the set of vectors of norm 0. The multiplicative group IR* acts on C(F)-0; the
orbit space is called the proiectivised cone of F and is denoted PC(F). We may of
course think of PC(F) as the set of isotropic lines in F.
28 QUADRATIC FORMS
Proof Let us return to the proof of 6.3 and introduce the unit sphere Sn"1
= 8D
in the hyperplane E orthogonal to the base vector u. We shall make use of the
map 4:E-*F given by
Corollary 6.14 When dim(F) > 3, the Lorentz group Lor(F) acts
faithfully on PC(F).
The set of values of the function A is finite simply because it is a subset of the set
of eigenvalues of o. Since dim(F) = n+1 > 3, we can use the connectedness of S' 1
to conclude that A is constant.
Let us prove that F is generated by C(F). To this end we shall prove
that any vector f E F with non-zero norm is contained in a hyperbolic plane. If
<f,f> < 0, then f 1 is of type (-n+1,1), so we can select e orthogonal to f with
<e,e> > 0. When <f,f> > 0 the space f 1 has type (-n,0) so we will have no
trouble finding e orthogonal to f with <e,e> < 0. Now, observe that a hyperbolic
plane is generated by two isotropic vectors.
Let us observe that multiplication by a scalar A is not an orthogonal
transformation of F unless A = 1 or A = -1. We have already observed that
multiplication by -1 interchanges the two sheets of the hyperboloid S(F).
Proposition 6.15 The truncated isotropic cone C*(F) = C(F) - {0} has
two connected components provided that dim(F) > 3. The connected components
are preserved by the Lorentz group Lor(F).
Proof Let us fix a point u E S(F) and observe that C*(F) is the union of
C+(F)={ceFI <c,u> > 0 } , C (F)={ceFI <c,u><0)
It follows from the proof of the previous proposition that these two sets are the con-
nected components of C*(F).
Let us prove that the two connected components are unchanged if we
replace the point of reference u with an other point v of the same sheet of the
hyperboloid S(F). To this end let us focus on a point c E C*(F) and observe that
the function xH<x,c>, omits the value zero since the hyperplane
orthogonal to c has type (-n+1,0), compare the remark made after 6.4. It follows
that <u,c> and <v,c> have the same sign.
From the fact that o(u) and u belong to the same sheet of S(F) it follows
that <o(x),u> and <o(x),o(u)> have the same sign for x E C*(F). In particular
for x E C+(F) we have <o(x),o(u)> = <x,u> > 0, and we conclude that <o(x),u>
is positive; thus o(x) E C+(F).
30 QUADRATIC FORMS
It is easily seen that this space has Sylvester type (-n-1,1). Let us consider the
transformation t: E-> C(E + R2) given by
This induces a map from E into PC(E+922) which identifies E with the com-
plement of the point of PC(E+R2) representing the line through the point (0,1,0).
Let us extend this to a bijection t: E-+PC(E + R2), where L(oo) is the line through
the point (0,1,0).
Theorem 7.5 The action of the Lorentz group Lor(E (D922) on the
projectivised cone PC(E + R2) and the homeomorphism a E - + PC(E (D R2)
identifies the Mobius group M"ob(E) with the Lorentz group Lor(E+R2).
Proof Let us return to the sphere C from 7.1 and form the vector
C = (c,<c,c> -r2,1)
We ask the reader to verify the formulas
<C,C> = - r2 , )<C,tx> = Ix - c12 - r2 xEE
From this we can deduce a commutative square
E PC(E(D R2)
& [ 7C
E PC(E(D R2)
E PC(E (D R 2)
la 1 TN
E PC(E ® 082)
Proof Let us first check through three special cases. First observe that a A>0
gives rise to a Lorentz transformation
(x,a,b)'--' (x,Aa,A-ib)
which induces the dilatation of E. Next observe that a o E O(E) is induced
on E by the Lorentz transformation
(x,a,b) '-4 (o(x),a,b)
Let us finally observe that a vector v E E gives rise to the Lorentz transformation
(x,a,b) '- (x + by, a + b<v,v> + 2<x,v>, b)
which induces translation x F x + v on E. This shows that any similarity of E is
a Mobius transformation. In order to prove the converse, we must show that these
three types of transformation generate the group LorL(E ® 082) consisting of
Lorentz transformations which stabilise the line L generated by (0,1,0).
To this end we shall use the map (z,O -- (-z,2(,0) to identify E ®08 with
the subspace L -L of E ® R2. This allows us to express the evaluation map from
1.5 in terms of the inner product on E ® 082.
We ask the reader to complete the proof by means of 6.16 and 5.8. 0
DISCS Recall also that N = N(E ® 082) denotes the set of vectors in E ® 082 of
norm -1. A vector U E N defines a sphere f = {x E E I <tx,U> = 0} and decom-
poses the complement of Y into the union of two connected open subsets
36 QUADRATIC FORMS
Observe that the map t: M C factors through C+. It is useful to reinterpret the
projectivised cone as
PC(E ® R2) = C+(E e R2)/ R+ ; R+= ]0,+o0[
Proof We can identify the action of M'ob(E) on the set of discs in E with the
action of the Lorentz group Lor(E (DR2) on the space N(E (D R2) of vectors of norm
-1. Once this is done, the result follows from 6.16.
1.7 MOBIUS TRANSFORMATIONS 37
Proof Let K be a normal vector for one of the discs `D bounding 9G and let k
denote inversion in the sphere 9G. From o(9G) = 9G we conclude that o(`D) = `D or
o(S) = tc(s). Upon replacing o by tco we may assume that o(`D) = `D or otherwise
expressed o(K) = K. Let us observe that restriction of v to F = K 1 acts trivially
on the PC(F); it follows from 6.14 that the restriction of a to F is t, the identity.
This gives O = t, which implies that o = t.
Definition 7.13 We say that two spheres I and T are orthogonal if the
normal vectors for the bounding discs are orthogonal5.
Proof It is a consequence of 7.10 that the last three conditions are equivalent.
To proceed, introduce normal vectors H and K for two bounding spheres. The
correspondance 7.7 identifies the Mobius inversion r with the Lorentz reflection rK
along the vector K. From the formula
7K(H) = H + 2<H,K>K
we conclude that rKK(H) = ± H if and only if <H,K> = 0. This shows that
condition 1° is equivalent to condition 3°.
Proof Let N denote the normal vector for `D. A Mobius transformation
leaves `D invariant if and only if the corresponding Lorentz transformation fixes
N. It follows that we may identify M"ob(`D) with the Lorentz group of N 1 . This
group is generated by reflections rI< where <K,K> = -1 and <K,N> = 0.
Corollary 7.17 Let L denote a linear hyperplane in E and `ll one of the
half-spaces in E bounded by H. Restriction from k to L defines an isomorphism
Mob(9) M"ob(L)
From the investigations made in the previous section it follows that the inversive
product of I and T is preserved under Mobius transformations:
The relative positions of two distinct spheres S and T can be read off T*`J :
Proof Let H and K be normal vectors for discs bounding Y and T. These two
vectors span a plane R, and we can make the identification
InGT=PC(R1)
The geometry is ruled by the discriminant
0 = <H,H> <K,K> - <H,K>2 = 1 - (Y* r)2
From the discriminant lemma 6.3 we deduce that
t*T < 1 : 0 < 0 , the plane R has Sylvester type (-2,0) which implies that
40 QUADRATIC FORMS
R 1 has Sylvester type (-n+l,l). Thus PC(R 1) consists of at least two points.
Y*°J = 1 : A = 1 , the plane R has Sylvester type (-1,0) which implies that
R 1 has Sylvester type (-n,0). Thus #PC(R 1) = 1.
Y*°J > 1 : A < 0 , the plane R has Sylvester type (-1,0) which implies that
R 1 has Sylvester type (-n,0). Thus PC(R 1) = 0.
We shall prove more precisely that the inner product of the normal vectors
N('D),N(g) E N(E (D R2) is given by
A*% = 1 + 2 tan 2 n
Proof According to 8.5 we may assume that the circles A and `.B are concentric
with radii a and b, a < b. The condition for the existence of a chain of kissing
circles can easily be worked out by Euclidean trigonometry. The result is
0
sin 11= 1+B ' -b
42 QUADRATIC FORMS
a - 1-sin il
b 1+sin n
.A*`B=2(b+a)
Case (-1,0). Let us observe that the dual pencil R 1 has the same type
and that R fl R 1 = P is an isotropic line. Pick a circle N in the pencil R 1 and
observe that the circles in R are precisely the circles through P, orthogonal to N.
za
; aEC*
wa
1=
The group G12(C) acts on the Riemann sphere in virtue of the formula
a b z az+bw
9.1 ;ad-bc#0
c d w cz+dw
Let us remark that scalar matrices act trivially on C, thus we are dealing with an
action of PG12(C) = G12(C)/C* on C. In order to calculate the action of the
inverse matrix S-1 we can use the cofactor matrix S' given by
9.2 a b - d -b
c d -c a
u(00) 0.(L) =
9.4 0 - ( _A )
_
00
,
cz + b
zEC,z#-a
Proof Let us first observe that the group of upper triangular matrices acts on
C fixing oo and induces the group of even similarities of C. In general, the matrix
identity (c # 0)
a
C
c d 0 1
composite of inversion in the unit circle and reflection z- in the x-axis as follows
from the basic formula
Z1 _ZIZI-2 ;zEC*
This concludes the proof of the fact that G12(C) acts through even Mobius trans-
formations. In order to conclude the proof, observe that PG12(C) acts transitively
on t and that the group of even Mobius transformations fixing oo induces the
group of even similarities on C, compare 7.7. 0
where the symbol [p,q] denotes the 2 x 2 matrix with first column p and second
column q. For a Mobius transformation v we have
del [o(p),o(q)] = del(a) del [p,q]
From this it follows that the cross ratio is invariant under Mobius transformations
At this point let us remark that the cross ratio [P,Q,R,S] is well defined as long as
P,Q,R,S represent at least three distinct points of C. This allows us to fix three
distinct points P,Q,R of C and make a free variation of the fourth point S. The
result is a Mobius transformation:
Proposition 9.8 Let P,Q,R denote three distinct points of C. The map
S H [P,Q,R,S] ; SEC
is an even Mobius transformation, which transforms P,Q,R into 00,0,1.
1.9 RIEMANN SPHERE 47
-g2det[p,r] gldel[p,r] st
[P,Q,R,S] = I
-p2det[q,r] pl det[q,r] S2
I
Observe that the 2 x 2 matrix has determinant det[p,r] det[q,r] det[p,q] $ 0. Direct
calculation reveals that this transformation maps P,Q,R into oo,0,1.
Proof It follows from 9.7 that the condition is necessary. Conversely, if the
two cross ratios are identical, let o and r be the Mobius transformations given by
o(P,Q,R) = (oo,0,1) , r(X,Y,Z) = (c ,0,1)
According to formulas 9.7 and 9.9 we get that
[P,Q,R,S] = [oo,0,l,o(S)] = o(S)
and similarly, [X,Y,Z,W] = r(W). Thus the Mobius transformation 7-1a maps S
to W as required.
Proof In order to generate the full Mobius group we need PG12(C) and one
single odd Mobius transformation, for example complex conjugation
[wz , F'[ w]
Let us record two symmetry properties of the cross ratio. The verification
is immediate and is left to the reader.
1 1
; a E C*
0 1
I
Proof Let a be given by the matrix S E G12(C). Note that a non-zero vector
V E C X C is an eigenvector for S if and only if [v] E C is a fixed point for a. From
this we conclude that a $ e has one or two fixed points on C.
When a has two distinct fixed points A and B, let r be an even Mobius
transformation with r(oo) = A and r(0) = B. The transformation 7--1 ar will fix 0
and oo. It follows that a representing matrix must be diagonal. After multipli-
cation by a complex number we get a matrix of the first type. When a has only
one fixed point A, choose an auxiliary point C # A and let r be the even Mobius
transformation given by
r(0) = C , r(1) = o(C)
r(oo) = A ,
Observe that the transformation rlar fixes oo and maps 0 to 1. It follows that
this transformation can be represented by a matrix of the form
1.9 RIEMANN SPHERE 49
a 1
; aEC*
0 1
But the matrix can only have one eigenvalue so we must have a = 1. 0
Observe that lr2 S depends only on the conjugacy class of S in PG12(C). In parti-
cular, we may introduce the invariant 1r2o of an even Mobius transformation o.
Proof With the notation of 9.14 we have the following explicit formulas
a 0
trz 1 1 ]=4 , tr2 1=2+a+a-1 aEC-{0,1}
0 1 0 1
0 -1 a 0 0 1 J=[ 1 0 ] _ [ a- 0
1 0 0 1 -1 0 0 a 0 1
I EXERCISES
EXERCISE 1.1 Let E be a finite dimensional vector space equipped with a non-
singular quadratic form.
10 Let F be a non-singular subspace of E, show that E = F ® F 1 and that
(x,y)-(x,-y) ;xEF,yEF1
defines an isometry a E 0(E) with a2 = 1.
2° Conversely, let a E 0(E) be given with a2 = c. Show that E = E+ ®E_
where E+ is the eigenspace for 1 and E_ is the eigenspace for -1. Hint: Use that
x = z(x - a(x)) + z(x + a(x)) ;xEE
3° Given an isometry a E 0(E) with a2 = t. Show that the fixed point space
E. for a is a non-singular subspace of E and that a(x) = -x for x E E01 1.
EXERCISE 1.2 1° Let (E,R) and (F,S) be quadratic spaces. Show that the
function
Q(e,f) = R(e) + S(f) ; (e,f) E E ® F
defines a quadratic form on the direct sum E Fg F of E and F. The form Q on
E ®F is denoted (E,R) 1 (F,S)
2° Let (D,Q) be a quadratic form and E and F orthogonal subspaces of D
such that D = E + F and E fl F = 0. Show that (D,Q) is isomorphic to the form
(E,R) I (F,S) where R and S denote the restrictions of Q to E and F respectively.
EXERCISE 2.3 Let Q be a quadratic form on the vector space E, and let
OP(E) = {o E 0(E) I o(e) = e; e E E 11
1° Show that OP(E) is a normal subgroup of O(E) and that it contains all
reflections along non-isotropic vectors.
2° Given A E k*, show that OP(E) acts transitively on the set
{eEEI Q(e)=a}
Hint: Use the proof of proposition 2.3. 3° Show that any o E 0P(E) can be
written as a product of at most 2 dimE orthogonal reflections.
4° When dim E = 2 show that any or E OP(E) can be written as a product of
one or two orthogonal reflections.
EXERCISE 2.4 Consider the quadratic form on R 3 with Gram matrix G (below)
0 0 -1/s
r,s E R
G= A= R= -r 1 -r/s
s#0
-s 0 0
EXERCISE 3.1 1° Show that the determinant det is a quadratic form on M2(R)
and that its polarisation is given by
<A,B> = 2 it AB` ; A,B E M2(R)
where the symbol B' denotes the cofactor matrix of B given by
a b J=[ d -b
c d -c a
EXERCISE 4.2 Show that S03(UR) is a simple group, in the sense that any
normal subgroup N 54 {t} equals S03(R). Hint: Use the previous exercise to see
that it suffices to show that N contains a half-turn. Use the next exercise actually
to find a half-turn in N.
EXERCISE 6.1 1° Show that Lor+(F) acts transitively on each of the sheets Hn
of the hyperboloid S(F).
2° Show that the special Lorentz group Lor+(F) is connected.
EXERCISE 6.2 Show that the orthogonal group 0(F) has four connected
components.
Hint: Note that SOAP is similar to AOPB and deduce that d(O,A)d(O,B) = r2.
Alternatively, observe that the circle with diameter BP is orthogonal to the initial
circle, and passes through the point A; compare 1.7.14 and 1.8.7.
54 QUADRATIC FORMS
EXERCISE 7.2 1° Show that dilatation with centre 0 and ratio r>0 can be
realised as the composite of reflections in the Euclidean spheres with centre 0 and
radius 1 and q-r-.
2° Show that parallel translation in E can be written as a composite of
Euclidean reflections in two parallel affine planes.
EXERCISE 8.1 Let us consider Euclidean discs 1.t and 11 with centres u and v
and radii r and s. Show that
v12
niar - in - - r2 - s2
2rs
EXERCISE 8.2 Let us consider a square matrix with n+2 rows and n+2 columns
-1 1 1 1
diagonal entries: -1
1 -1 1 1
entries elsewhere: 1
1 1 -1 1
1 1 1 -1
EXERCISE 8.3 Let R denote a pencil of circles in the Euclidean plane E, and let
G denote the group generated by reflections in the circles of the pencil.
1° Show that the action of G on R identifies G with a subgroup of O(R), and
identify this subgroup depending on the type of the pencil.
2° Show that a product a/y of three reflections in circles of the pencil is itself
a reflection in a circle from the pencil.
EXERCISE 9.2 Let P,Q,R denote three distinct points of C. Show that the point
S E C belongs to the circle through P,Q,R if and only if [P,Q,R,S] E R.
EXERCISE 9.3 Let P,Q,R,S be four distinct points on a circle 9G. Show that
[ P,Q,R,S] < 0
if and only if the pair {P,Q} separates the pair {R,S}; meaning that P and Q
belongs to different connected components of 9G - {R,S}.
(R,D)(P,Q) ° (A,Q)(B,P)
When A = P and B = Q, use the transformation (A,B)(R,R) o (A,B)(C,R).
EXERCISE 9.5 1° Show that the cross ratio of four distinct points is a complex
number distinct from 0 and 1 which obeys the rules
[P,Q,R,S] = [Q,P,R,S]-1 = [P,Q,S,R]-1
EXERCISE 9.6 Fix three distinct points P,Q,R of C and consider the rational
function A on C given by A(S) = [P,Q,R,S]. Show that permutations of P,Q,R
give the six rational functions A, 1- A 1 1 A- 1 A
A-1
1,
EXERCISE 9.8 For o E PG12(C) let Fix(o) denote the set of fixed points for o on
C. Show that if r E PG12(C) commutes with or then o(Fix(r)) = r(Fix(o)).
2° Show the converse of 1°. In particular, show that Fix(o) = Fix(r) implies
that o and r commute.
3° Find the condition for two half-turns to commute.
11.1 GEODESICS
A map o,: X--+Y from one metric space to another is distance preserving if
'The geometries we are going to study all carry the structure of metric
space. The notion of a geodesic curve in a metric space, as introduced in section
11.1, is consistent with the notion of a geodesic curve in Riemannian geometry.
58 GEOMETRIES
The basic example of a metric space is the set R of real numbers equipped
with the standard metric d(x,y) y - x 1, x,y E R. It will be important for us to
know all isometries of R.
Proof Let us consider a point a E R and pick an open interval J around a such
that the restriction of y to J is distance preserving. It follows from the sharp
triangle inequality 1.4.2 that for any three distinct parameters r,s,t E J the points
y(r), y(s),y(t) lie on an affine line. If we fix two distinct points c,d E J we find
that y(J) is contained in the affine line through y(c) and y(d). It follows from 1.5
that we can find a vector e of norm 1 such that
y(t) = e(t - a) + ry(a) ;tEJ
This implies that y is a differentiable curve whose derivative y'(t) is locally
constant. Since IR is connected, we conclude that ye(t) is independent of t E R. It
follows from elementary calculus that y has the required form. 0
The isometry r fixes H while the image r(x) of a point x 0 H can be described by
observing that the affine line through x and r(x) is perpendicular to H and that H
intersects this line in the midpoint of x and r(x).
Lemma 2.4 Let there be given two sequences A1,...,Ap and B1,...,Bp of
points of E such that
d(Ai,Ai) = d(Bi,B3) ; i,j = 1,...,p
Then there exists an isometry a of E composed of at most p reflections such that
cT(Ai) = Bi ; i = 1,...,p
Proof Let us recall that the bisecting hvnernlane for two distinct points A
and B of E is given by
{ X E E I d(X,A) = d(X,B) }
It is easily seen that reflection in the bisecting hyperplane interchanges A and B.
We shall prove the lemma by induction on p. Observe that we have just
taken care of the case p = 1. To accomplish the induction step, suppose that an
isometry p composed of at most p-1 reflections has been found to transform
AI,...,Ap-1 into BI,...,Bp-1. When p(Ap) = Bp we can take a = p. When
p(Ap) $ Bp, observe that
d(p(Ap),Bi) = d(p(Ap),p(Ai)) = d(Ap,Ai) = d(Bp,Bi) i = 1,...,p-1
;
which means that the points BI,...,Bp_1 all lie on the hyperplane bisecting p(Ap)
and B. If r denotes reflection in this hyperplane then the isometry o = rp serves
our purpose. 0
Proof Put o = /34a and assume that P is a point of E where o(P) P. From
d(o(P),Ai) = d(o(P),o(A1)) = d(P,Ai) ; i = o,...,n
it follows that the points A0,...,A. all belong to the affine hyperplane bisecting P
and o(P). This contradicts that the points A0,...,An form a simplex.
Proof Let us assume that o $ t and pick P E E with o(P) # P. It follows from
the proof of 2.5 that K equals the perpendicular bisector for o(P) and P. Let us
form the composite of r, reflection in K, and o to get an isometry ro which fixes
P and the points of K. We conclude from 2.5 that ro = t.
Proof Let us pick a Euclidean simplex A0,...,A. in E. We can use lemma 2.4
to pick an isometry o which is the composite of at most n+1 reflections with
o(Ai) = /3(Ai) ; i = 0,...,n
It follows from lemma 2.5 that o = 0.
The translations form a subgroup 7(E) of the group Isom(E) of all isometrics.
Proof Let Isom'(E) denote the set of isometrics of E which can be written in
the form TeU as above. We ask the reader to verify the formula
are °1=To.(e) ;e EE, 0EO(E)
We can now verify that Isom'(E) is closed under composition
TV a Te P = TV TQ(e) P = TV+O-(e) vP
Observe also that Isom'(E) contains all reflections in hyperplanes in E as follows
by inspection of the formula 2.3. Thus we get Isonz(E) = Isom'(E) from 2.7. 0
As a final general remark on the structure of Isom(E), we ask the reader to verify
2.12 . TV O 1 =
T.i(V)
;oEIsom(E),vEE
Proof Let us first show that the two subspaces Ker(a-t) and Im(a-t) of E
are orthogonal to each other
<x,a(y)-y> = <x,a(y)> - <x,y> = <a(x),a(y)> - <x,y> = 0
This gives us Ker(a - t) n Im (a - t) = 0 . From Grassmann's dimension formula
applied to the linear map a-t we get that
dim Ker(o - t) + dim Im(a - t) = n
The result follows from the dimension formula 1.3.4.
Let us return to the classification theorem 2.13. The fact that rv and 0
commute can be written in the form
64 GEOMETRIES
O(x+v)=O(x)+v ; xEE
This shows that the fixed point set for 0 is stabilised by the translation rv.
Let us apply the classification theorem to dimensions 2 and 3. As a
consequence of our investigations of O(E) made in 1.4.8 and 1.4.9 we get
11.3 SPHERES
Lemma 3.3 Let A and B be points of the sphere Sn. A unit tangent vector
U to S" at A can be chosen such that
B = A cos d(A,B) + U sin d(A,B)
Proof Let us at first suppose that A and B are linearly independent. Let U
be a unit tangent vector at A in the plane spanned by A and B, and let us write
66 GEOMETRIES
B =xA+yU ;x,yER
Using <A,U> = 0 we get x2 + y2 = 1. Thus we can determine s such that
B = A cos s + U sin s ; sE[-7r,7r]
Notice that the formula is unchanged under the substitution (s,U)H(-s,-U), and
conclude that we can write B as above with s E [0,7r]. Observe that
cos d(A,B) = <A,B> = cos s
and conclude that s = d(A,B). If A and B are linearly dependent we have B = A
or A = -B. It follows that sin d(A,B) = 0 and that any unit tangent vector U to
Sn at A will serve our purpose.
a = d(B,C)
b = d(A,C)
c = d(A,B)
Let us use lemma 3.3 to pick tangent vectors U and V at A such that
B=Acosc+Usinc , C=Acosb+Vsinb
We can use this to introduce LA = a as the angle between U and V
1
cos a = cos b cos c + sin b sin c cos a
Proof When A,B,C are linearly independent we can use the cosine formula
with cosa < 1 to get the inequality
cosa < cos(c-b)
When c-b is positive we deduce that a > c-b which gives c < a+b. Observe
that this inequality is trivial when c-b is negative. This concludes the proof of
the sharp triangle inequality. When A,B,C lie in a plane we can modify the
trigonometry to maintain the cosine formula 3.5 allowing for a = O,ir. We can
deduce the triangle inequality from 3.5 using the inequality cosa < 1.
Let us observe that the metric we have introduced on the sphere Sn defines
the topology on Sn induced from the ambient Euclidean space F as follows from
Let us show that two geodesic curves o,,-f: R- Sn which coincide in a neighbour-
hood of 0 are identical. To this end we introduce the set
{ t E R I o(t) = -y(t) , o'(t) = -y(t) }
Observe that the set is closed since 7,0,7',o' are continuous. The set is open as
well, as follows from formula 3.9. Since the set is non-empty, by assumption, we
get from the connectedness of R that y = o. Apply this to the given geodesic curve
7 and the geodesic curve o given by the right hand side of 3.9 with u = 0, and the
result follows.
11.3 SPHERES 69
Lemma 3.10 Let there be given two sequences A1,...,Ap and B1,...,BP of
points of S" such that
d(A1,Aj) = d(Bj,Bj) ; i,j = 1,...,p
Then there exists an isometry v composed of at most p hyperplane reflections with
o,(Ai) = B. ; i = 1,...,p
Proof Consider two distinct points A and B of S", let us analyse the set
{ P E S" I d(A,P) = d(B,P) }
From the definition of the spherical distance it follows that this is the intersection
between S" and the linear hyperplane K orthogonal to the vector A - B.
Orthogonal reflection in K will interchange A and B. We can now conclude the
proof by the method used in the Euclidean case, see the proof of 2.4.
Proof Let us pick a spherical simplex in S", i.e. a sequence A0,...,An of points
of S" linearly independent in F. We can use lemma 3.10 to pick an isometry o
which is the composite of at most n+1 reflections in hyperplanes with
o(A;) = /3(Ai) ; i = 0,...,n
We can conclude the proof by the method used in the Euclidean case, see 2.5.
70 GEOMETRIES
Proof Consider points A,B,C on H" and put a = d(B,C), b = d(C,A) and
c = d(A,B). We shall calculate the determinant of the Gram-matrix 1.1.10 of
A,B,C using the standard abbreviations cha = cosha and sha = sinha.
1 cha chb
4.3 A = del cha 1 chc
chb chc 1
H.4 HYPERBOLIC SPACE 71
Let us assume that the vectors A,B,C are linearly independent in the ambient
space F. It follows that they generate a subspace of F of type (-2,1) and we
conclude from 1.3.5 that A > 0. If c > a and c > b we have that
p-a=2(c-a)+2b>0, p-b=2(c-b)+2a>0
Using A > 0 and p > 0 we conclude from the factorisation 4.4 that p - c > 0. It
follows that a + b - c = 2(p - c) = 0.
If the vectors A,B,C are linearly dependent in the ambient space F, then
A = 0. If A,B,C do not represent the same point then p $ 0 and we conclude from
the factorisation 4.4 that p - a = 0 , p - b = 0 or p - c = 0. The remaining details
are left to the reader.
The tangent space TA(I-I") inherits a quadratic form from the ambient
space F. It is easily seen that the tangent space has Sylvester type (-n,0)
72 GEOMETRIES
Geodesics Starting from a point A E H" and a unit tangent vector T to H"
at A we can construct a curve 0: R - II" by the explicit formula
Let us calculate the hyperbolic distance between y(s) and y(t), s,t E R
cosh d(y(t),y(s)) = <y(t),y(s)> = cosh(t-s) = cosh It-si
From this we conclude that
d(y(t),y(s)) = I t - s ;s,tER
This shows that 0: R - H" is a geodesic curve. The image of the curve is contained
in the plane R spanned by A and T; in fact it follows from lemma 4.6 that the
image of y is H" fl R.
11.4 HYPERBOLIC SPACE 73
Proposition 4.7 Any geodesic curve 7:R--+H" can be written in the form
y(t) = A cosh t + T sinh t ; tER
where A E H" and T is a unit tangent vector to H" at A.
Proof Let us fix a point u E R and pick an open interval J around u such that
the restriction of y to J is distance preserving. It follows from the sharp triangle
inequality 4.2 that for any three distinct parameters r,s,t E J, the points
y(r),y(s),y(t) are linearly dependent. In particular if we fix two distinct points
c,d E J such that 7(c) and y(d) are linearly independent we find that O(J) is
contained in the plane R spanned by y(c) and y(d). Put A = y(u) and pick a unit
tangent vector T E TA(H') contained in R. It follows from the discussion
preceding 4.7 that we can use lemma 1.5 to find e = ± 1 such that
y(t) = A cosh c(t-u) + Tsinh E(t - u) ;tEJ
This implies that y is continuously differentiable on J with y'(u) = eT. Thus
y(t) = y(u) cosh(t-u) + yo(u) sinh(t-u) ;tEJ
The merits of this formula is to reconstruct y in a neighbourhood of u E R from
the values of 7(u) and yo(u). We ask the reader to conclude the proof by the
method used in the proof of 3.8.
Lemma 4.8 Let there be given two sequences A1,...,Ap and B1,...,Bp of
points of Hn such that
d(A;,Ai) = d(Bi,Bj) ; i,j = 1,...,p
Then there exists an isometry a composed of at most p Lorentz reflections with
o-(Ai) = B; ; i = 1,...,p
Proof Consider two distinct points A and B of Hn and let us analyse the
perpendicular bisector for A and B
{ P E HIn I d(A,P) = d(B,P) }
From the definition of the hyperbolic distance it follows that this is the
intersection between H' and the linear hyperplane orthogonal to the vector
N = A - B. We can use 1.6.2 and 1.6.4 to conclude that
<N,N> =2-2<A,B> <0
Reflection along N will interchange A and B. We can now conclude the proof by
the method used in the proof of 2.4. 0
For hyperbolic space H" we take the sheet of the unit hyperbola in E ® R
consisting of points (A,a) with -<A,A> +a 2 = 1 and a > 0. We shall make use of
the parametrisation p:D"-+H" given by the formula
(A,1)
5.2 ;AED"
P(A)
- 1 - <A,A>
Geometrically, the point p(A) is the intersection of H" and the line through (A,1)
and 0. The Klein model of hyperbolic n-space is D" equipped with the metric
obtained by transporting the hyperbolic metric along p: D" - H". It follows from
5.2 and 4.1 that the metric in the Klein model is given by
I - <A,B>
5.3 coshd(A,B) = ;A,BED"
4 (1 - <A,A>)(l - <B,B>)
The geodesic lines in the Klein model are precisely traces on D of ordinary affine
lines in E. This can be seen from the description of the geodesic lines H" given in
4.7 and the geometric origin of the parametrisation p:D" H".
The boundary 8H" (in the sense of 6.18) can be parametrised by the
Euclidean boundary of OD" by assigning to a point A E 8D" the isotropic line in
E ®R through the point (A,1).
76 GEOMETRIES
Let us once again start with an n-dimensional Euclidean space E and its
open unit disc D. We shall make use of the auxiliary space E ® R equipped with
the hyperbolic form given in 5.1 and the sheet Hn of the unit hyperboloid in E ® R
passing through (0,1). This time we shall consider the parametrisation f: Dn - Hn
given by the formula
From the geometric point of view, this is the stereographic projection with centre
(0,-1) already considered in 1.6.6. The result of transporting the hyperbolic
metric along f is given by
Proof Let us first record from 1.6.5 that f-1:Hn-*Dn is given by the formula
f-1(A,a) = A(l+a)-1 ; (A,a) E Hn
Let r denote reflection along a vector (N,n) in E 9 R of norm -1. Let us trans-
form r to Dn. We ask the reader to verify the formula
P+(InP-N12-1)Nii1
' rf(P) _
f1 nP-N 12
= Nri +ii 2 P-Nn1
1
; P E Dn
1 P-Nn-1 12
This shows that f-1rf is induced by inversion in the Euclidean sphere N with
centre Nn-1 and radius n-1. Let us observe that
(n-1 )2+ 1 = <Nn-1, Nii 1>
which shows that X is orthogonal to Sr-1 = DDn.
Proof Let us focus on the point P = 0(0) E Dn. If P = 0, the result follows
from 6.4. If P $ 0, put R = o(P) where a is inversion in the sphere Sn"1. Let
be a circle with centre R orthogonal to Sn"1. We ask the reader to verify that
* Sn 1 = 2i. I p(0) - o(R) I
Observe that o(R) = P and conclude that p(O) = P or p(P) = 0. It follows that
po fixes 0 and we conclude from the previous lemma that pO E 0(E).
78 GEOMETRIES
Let us start from a Euclidean vector space L of dimension n-I and form
the Euclidean space E = L ® R. We shall be concerned with the inner half-space
E+ consisting of points (p,h) of E with h > 0. Inversion o in a sphere with radius
and centre in the south pole S = (0,-1) will map E+ onto the unit disc Dn
7.1 (p , I+h)
o(p,h) = (0,-1) + 2 IP12 ;pEL,heR
+ (I +h)2
Let us use o to transport the metric of the Poincare disc onto the upper half-space
E+. Explicit calculation based on 6.2 and 7.1 gives us
2
7.2 coshd(P,Q) = 1+ 1P -Q P = (p,h), Q = (q,k)
ph
The open half-space E+ equipped with this metric is called the Poincare half-space.
_ 12
8.1 coshd(z,w) = 1 + 2 Im[w] II [z] ; z,wEH2
For the convenience of the reader we shall give two different formulas
Iw-zl I w -Y I
8.2 sinhd(z,w)
z _ Im[w] Im[z] '
cosh d(z , w) _
2 2 Im[w] Im[z]
In order to describe the isometries of H2 recall from 1.9 that G12(R) acts on C-R
through the formula
a b az+b
c d cz + d
We ask the reader to work out the following useful expression for the imaginary
part of this fraction.
IM[ az + b _ Im[z] bl
ra d]
8.3
cz+d] -Icz+d12 det Lc c a] E G12(R)
az + b
cz + d
deter>0
8.4 a+b 0- = Lc db]
cz + d deto, <0
t SZ
e)
cosh (s - t) =2t(e8-t + et-sl =
e2t,+e2s
2etes
_1+ (e
-
2etes
=cosh d(ies,iet)
It follows that dJ, the positive y-axis, is a geodesic. Let us observe that PG12(R)
acts transitively on the set of circles in C orthogonal to A as follows from the fact
that PG12(R) acts transitively on R. The action of PG12(R) on the set of geodesics
in H2 is transitive as well as follows from 8.5 and 4.9.
Let us amplify the fact that the isometries of H2 are induced by Mobius
transformations by asking the reader to verify that the distance can be written
z w
z* w*
II EXERCISES
EXERCISE 1.1 Let X and Y be metric spaces. For points P = (x,y) and Q = ({,rl)
on X x Y let us define the distance by
d(P,Q) = d(yrl)2
Show that this defines a metric on X x Y.
EXERCISE 2.1 Give a direct proof of the fact that an isometry f of Euclidean
space E with f(O) = 0 is linear. Hint: Use polarisation to show that
<f(x),f(y)> = <x,y> ; x,y E E
Conclude from this that vectors of the form below have norm zero
f(x+y) - f(x) - f(y) , f(rx) - rf(x) ; x,y E E, r E R
EXERCISE 2.2 Let o and p be rotations of the Euclidean plane with distinct
centres A and B.
10
Justify the following procedure for the calculation of pa-: Write p = 07
and o = ya where y is reflection in the line through A and B and Q is reflection in
a line b through A and a is reflection in a line a through B. If a and b are parallel
we find that pa. = 6a is a translation. If a and b intersect in C we find that
po = ,3a is a rotation with centre C.
2° Suppose that p and a are rotations with the same angle -27r/3 but
distinct centres A and B. Show that pa is a rotation with angle 27r/3 and that its
centre C forms an equilateral triangle with A and B.
30 Let DABC be a triangle in the Euclidean plane. Let A* denote the point
in the plane such that ABC is equilateral and such A and A* lie on opposite
sides of the line through B and C. We let BA denote the rotation about the
circumcentre OA of DA*BC which transforms B to C. Define similar transfor-
mations BC and BB and show that 8C BB BA = c. Hint: Show that 0C BB BA is
rotation with centre B and angle 0.
4° With the notation above, show that the circumcentres OA, OB and Oc
form an equilateral triangle. Hint: Use the method from 1° to calculate 0B BA.
II EXERCISES 85
EXERCISE 3.1 Given two lines k and I in F through the origin. Let us define
the acute angle L(k,l) E [0,2] between k and I by the formula
cos L(k,l) = I<k,l>I
where k and I are unit vectors generating k and I.
10
Show that the acute angle defines a metric on projective space P(F), the
set of lines in F. Hint: Interpret P(F) as the orbit space for the action of the
EXERCISE 3.2 Let S2 denote the unit sphere in R3 and let the standard
spherical coordinates o': l 2 S2 be given by
o(9,0) = (silt 0 Cos 0, Sill 0 sill 0, Cos 0)
1° Show that a continuously differentiable curve y: [a,b] S2 which does not
pass through the north and the south pole can be written y = a(0,0) , where
0,0:[a,b] R are continuously differentiable functions.
2° With the notation above show that the length of the curve y is given by
where y runs through the set of piecewise continuously differentiable curves on the
cylinder Cyl.
10 Show that the distance defined above is a metric on Cyl.
2° Recall that the standard cylinder coordinates tc:R2-> Cyl are given by
rc(O,r) = (cos 0, sin 0, r)
Show that a continuously differentiable curve y:[a,b]-+Cyl can be factored y=4cp
where µ:0t-*R2 is a continuously differentiable curve.
3° With the notation above show that the length of the curve y on Cyl is
1(y) = 1(µ)
4° Show that the distance between points p and q on Cyl is given by
d(p,q) = inf d(P,Q)
P,Q; p=ti(P),q= c(Q)
exists a constant k > 0 such that for all P E R 2 the disc with centre P and radius k
is mapped bijectively onto the metric disc with centre p = v(P) and radius k.
Show that an affine line in R2 is mapped by v onto a geodesic in S1 x S1,
3°
and that all geodesics on S1 x S1 have this form.
4° Let there be given points p and q on S1 x S' with d(p,q) = d. Show that
we can find a geodesic curve y:R->S' x S' with y(0) = p and y(d) = q.
EXERCISE 3.5 For real numbers a,b,c consider the determinant g given by
1 cosa cosb
cosb cosc 1
1° Show that
EXERCISE 5.1 For d E D', d # 0, let d* denote the mirror image of d in the
unit sphere Si-1 and let 'cd denote reflection in the circle with centre d*
orthogonal to Sr"1. Let Pd denote Euclidean reflection in the hyperplane orthogonal
to d and put rd = t'd Pd-
1° Extend the definition of the symbol rd to all of Dn by the convention
ro = t , and show that any µ E Mob(D") can in a unique way be written
p=rdo ;dEDn,aEO(E)
20 Use the previous result to establish a homeomorphism
Mob(Dn) -Z D' x O(E)
The topology on Mob(Dn) is uniform convergence on Sn-1. (See [Beardon] §3.7.)
In this chapter we shall make a detailed study of the hyperbolic plane H2:
classification of isometries and trigonometry. The even isometries of H2 can be
divided into three classes: rotations, translations and horolations. This division
follows the natural occurrence of geodesics in three types of pencils. We shall
keep track of the pencils by assigning a normal vector to an oriented geodesic.
This will be dealt with in a specific model for H2 referred to as the s12(R)-model.
The big advantage of the s12(R)-model over say the Poincare models is a very rich
vector calculus 1.
s12(R) = { R E M2(Id) I it R = 0 }
a b a b 0
_ (a2+ be)
C -a c -a 0 1
11
It is time to show that the Sylvester type of our inner product is (-2,1). To this
end we ask the reader to verify that the following three matrices form an
orthonormal basis for s12(R)
I
The determinants of these matrices are -1, -1, +1.
It is now time to introduce an important tri-linear form vol on s12(R).
The evaluation of the form vol on three vectors K,L,M is given by
This is actually an alternating form, in the sense that the form annihilates any
triple where two of the vectors are equal: for example if K = M we get from
K2 = - t del K
The alternating form 1.5 is a volume form for s12(R) in the sense that it
takes the value 1 or -1 whenever K,L,M form an orthonormal basis. To see this,
we ask the reader to verify that vol has evaluation 1 against the basis above.
Let us introduce a third opeation on s12(R), the wedge product of two
vectors K and L from s12(IR), given by the formula
We shall need one more formula relating the inner product and the wedge product
This formula can be deduced directly from the definitions using formula 1.4:
4(<A,C>B - <A,B>C) =
- (AC + CA) B - B (AC + CA) + (AB + BA) C + C (AB + BA) _
ABC - ACB + CBA - BCA = 4 A A (B A C)
The inner product of two wedge products can be evaluated by the formula
Let us present a very useful volume formula. For two triples of vectors
A1,A2,A3 and B1,B2,B3 in s12(R) we have
Proof Let us first fix A1,A2,A3. Both sides of the formula are alternating in
B11B21B3 so we may assume that B1,B2,B3 is a fixed positively oriented or-
thonormal basis. We can make a variation of A1,A2,A3 and observe, that both
sides are alternating in the A1,A2,A3. Thus it suffices to treat the case where the
A1,A2,A3 and B1,B2,B3 form the same orthonormal basis. In this case the result
follows from the fact that s12(R) has Sylvester type (-2,1).
Clifford algebra 1.11 Let us add some remarks for readers familiar
with the formalism of Clifford algebra. For reference see [Deheuvels] or the
opening chapter of [Gilbert,Murray], but it is our personal opinion that the
arguments of Clifford algebra are best learned through concrete examples.
The real algebra M2(C) is easily seen to be generated by the linear
subspace isl2(R). The square of an element X of this subspace is a real scalar. In
fact X2 is a real quadratic form of X E isl2(OR) as follows from 1.1:
(iS)2 = det S ; S E s12(R)
Observe that dimR M2(C) = 23 and conclude from general principles that M2(C) is
a concrete realisation of the Clifford algebra of the quadratic space s12(l) of
Sylvester type (-2,1). A good many of the definitions and arguments of this
section conform with the general notions of Clifford algebra. This is particularly
true for the definition of the wedge product 1.6 and the arguments presented in
section 111.4.
92 IIYPERBOLIC PLANE
A= ; a,b,c E R, detA = 1
Observe that the entry c $ 0 which follows from - a2 - be = 1. Thus the sign of
entry c distinguishes between the sheets of the hyperboloid. We let H2 denote the
sheet consisting of matrices A as above with c>0
Normal vectors Let us recall from 11.4.7 that a geodesic curve y:R-H2
has the explicit parametrisation
7(t) = A cosh t + T sinh t ;tER
where A = 7(0) and T = y'(0) is a vector orthogonal to A with <T,T> 1.
This follows from 2.1 using the general formula 1.8. Let us remark that the
normal vector N depends on the orientation of the geodesic curve y. Thus the
normal vector of an unoriented geodesic is determined up to a sign only.
111.2 PENCILS OF GEODESICS 93
The two connected components of the complement of h are called the sides
of h. Once the geodesic h is oriented by a normal vector N, we can distinguish
between the two sides of h: a geodesic curve through a point A E h with tangent
vector N at A passes from the negative side of h to the positive side of h (warning:
the function X F-+ <X,N> takes negative values on the positive side of h).
94 HYPERBOLIC PLANE
or = Lor(V,-U)
-U=Vcosa+VAAsina
Using 1.7 and 1.8 we deduce that <U,V> = cosa and U A V = A sina. For the
corresponding normal vectors H = U A A and K = V A A we deduce from 1.8 that
Proof Suppose that the normal vectors H and K satisfy I<H,K>l < 1.
Consider the discriminant A for the plane E spanned by H and K
A = <H,H><K,K> - <H,K>2 = 1- <H,K>2
Observe that A>O and conclude from the discriminant lemma 1.6.3 that E has
Sylvester type (-2,0). It follows that the line E L has type (0,1) and that the
point A of intersection between H2 and E L lies on h and k. The converse
implication follows from 2.5.
Proof Two geodesics which meet at a right angle will have <H,K> = 0 as it
follows from 2.5. Conversely, if <H,K> = 0 the two geodesics will meet according
to 2.6. The angle at which they meet must be a right angle, 2.5.
H=AAL
K= -BAL
Let us evaluate the inner product between the normal vectors H and K
<H,K> = <A,L><L,B> - <A,B><L,L> = <A,B>
This gives us the first of the following two formulas
In order to prove the second formula, observe first that L A K = B as follows from
1.8 using K = L A B. Next, observe that vol(H,L,K) = <H,B> to get that
B = A cosh d(A,B) + H sinh d(A,B)
This gives <H,B> = <H,H> sinh d(A,B) = - sinh d(A,B) as required.
Theorem 2.9 Two distinct geodesics h and k are perpendicular to the same
geodesic if and only if their normal vectors H and K satisfy I<H,K>l > 1. The
common perpendicular geodesic is unique whenever it exists.
Proof Observe that H and K span a plane E, and form the discriminant
A = <H,H><K,K> - <H,K>2 = 1 - <H,K>2
It follows that A<O if and only if I<H,K>l > 1. If A < 0 we conclude from the
discriminant lemma I.6.3 that E has Sylvester type (-1,1). It follows that E -1-
contains a vector L of norm -1. The geodesic 1 with normal vector L is perpen-
dicular to h and k.
Conversely, consider a geodesic I perpendicular to both h and k. It follows
IH.2 PENCILS OF GEODESICS 97
from 3.3 that a normal vector L for 1 is orthogonal to both H and K, i.e. L E E 1
This makes L unique up to a sign.
Proof Observe that H and K span a plane E, and form the discriminant
A = <H,H><K,K> - <H,K>2 = 1 - <H,K>2
If h and k have a common end S we must have S = E 1 . In particular, we
conclude that the line E 1- is isotropic. It follows from the discriminant lemma
1.6.3 that the plane E has type (-1,0) and that A = 0, i.e. I<H,K>l = 1.
Conversely, if I<H,K>l = 1 we get A = 0 and we get from 1.6.3 that E 1 is an
isotropic line, in fact a common end for h and k.
Corollary 3.2 The perpendicular bisectors ma, mb, m, for triangle DABC
lie in a pencil.
Proof Let 9 denote the pencil containing mb and ma and let a E 9 be the
geodesic through A. Reflections in the geodesics ma, mb, a are denoted µa, Pb and
a. It follows from 3.1 that µaµba is a reflection in a geodesic h E 9. Observe that
µaµba transforms A into B and conclude that h = mc. Thus me E 9.
Proof For three reflections p,a,r in geodesics with end S we have par = rap,
as follows from par = (par)-1 = 7_ 1
0`
1P- 1 . For four such reflections, we get that
(a1)(7) = (aQy)b = (7)3a)b = y(3ab) = y(ba,Q) = (yb)(a/3)
and the result follows.
A closer look at the proof of 3.1 reveals that the group of horolations with
centre S is isomorphic to the additive group of R, see also IV.2 The orbits on H2
for the group of horolations with centre S are called horocycles with centre S.
111.3 CLASSIFICATION OF ISOMETRIES 101
Corollary 3.4 Two distinct points A and B belong to the same horocycle
with centre S if and only if the perpendicular bisector m for A and B has end S.
Proof Suppose that the perpendicular bisector in for A and B has end S. Let
It be a reflection in m and let a be a reflection in the geodesic through A with end
S. With this notation, the horolation µa transforms A into B. Conversely, let a
be a horolation with B. It follows from the theorem on three reflections
that as is a reflection in a geodesic it with end S. Note that o-a(A) = B and
conclude that n is the perpendicular bisector for A and B.
Proposition 3.5 The group of horolations with centre S acts on the set of
geodesics with end S in a simply transitive manner.
Proof Let a and b be geodesics with end S and let M and N be normal
vectors for a and 6 respectively. The linear automorphism a of the plane P = S 1
with o,(S) = S and a(M) = N is orthogonal. It follows from 1.5.3 that o can be
written as a product of one or two reflections. Upon replacing N by -N we can
assume that a is the product of two reflections. It is immediate to extend a to a
horolation with centre S.
Let us assume that the horolation cr fixes the geodesic a. For A E a we
must have a(A) = A: otherwise, the perpendicular bisector for A and has end
S, contradicting the fact that two perpendicular lines can't have the same end.
We shall present the remaining two pencils leaving the detailed treatment
to the reader. This can be done by algebra as above, but elementary geometry may
be applied as well. See also 11.8.
Proof Let us fix A E H2 and focus on the point O(A). If A = O(A), we find
that 0 is a reflection in a geodesic through A. If A # O(A), let in denote the
geodesic through A and q5(A) and let l denote the perpendicular bisector for A and
O(A). The point of intersection between in and I is denoted M while µ and A
denote reflections in these geodesics. Observe that \q is even and fixes the point
A. Thus a4 is a rotation around A, and we can use the theorem on three
reflections to write a¢ = ACV where v is a reflection in a geodesic n through A. Let
us introduce reflection ti in the geodesic k through M perpendicular to n and write
Note that \p and inc are half-turns around points of k. It follows that the isome-
try r = () p)(vic) is a translation along k.
104 HYPERBOLIC PLANE
Let us first bring the whole group G12(R) to act on sl2(R). To this end we shall
make use of the sign of the determinant
With this notation, the group G12(R) acts on s12(R) through the formula
<of 10 01
],y-', [ ° 1]> = 2(deto)-1(a2+b2+c2+d2)
where a,b,c,d are the entries of o. It follows from this and 1.6.4 that conjugation
with o interchanges the two sheets of the hyperbola if and only if sign(o) < 0. It
follows that the formula 4.3 defines a Lorentz transformation of s12(UR).
From the fact 1.6.11, that Lor(sl2(R)) is generated by reflections, it follows that
the map G12(R) -> Lor(s12(R)) is surjective.
Let us show that the kernel of the map G12(R) -; Lor(sl2(R)) consists of
scalar matrices. Consider a o E G12(R) with dec(o) > 0 which acts trivially on
s12(R). Observe that the vector space M2(R) is generated by t and s12(R) and
conclude that o commutes with all matrices X E M2(R). From this it follows that
o is a scalar matrix. Next, consider a a E G12(R) with dec(o) < 0 which acts
trivially on s12(R). This means that o anti-commutes with s12(R). We ask the
reader to verify the general fact that a matrix o E M2(R) which anti-commutes
with s12(R) is the zero matrix.
Corollary 4.6 The action of S12(R) on s12(R) identifies the group PS12(R)
with the special Lorentz group Lor+(s12(R)).
Proof We shall show that the action of a o E G12(R) defined by 4.3 is an even
Lorentz transformation on s12(R) if and only if dec(o) > 0. To this end, we pick a
decomposition of the action of o on s12(R) into a product r1...rs of reflections
along vectors K1,...,Ks of norm -1. It follows from 4.5 that the matrices o and
K1...Kg E G12(R) have the same action on s12(R). Thus we can use 4.4 to pick a
scalar r E R* such that
o = rK1..K5
Calculate the determinant of both sides to get that dec(o) = r2(-1)$. This shows
that dec(o) > 0 if and only if s is even.
106 HYPERBOLIC PLANE
Proof From the end of the proof of the previous proposition follows that we
can write o = eHK, where e = ± 1. This gives us
ir o = e lr(HK) = - 2e<H,K>
The result follows by taking the square of this relation. 0
tr o)2
(det
For a general a E G12(R) we shall make use of the invariant tr2(o) =
)
introduced in 1.9.14.
Proof It is known from linear algebra that any conjugacy class in G12(R)
contains a matrix from the following list
A 0 a -,3 A 1
0 µ (3 a 0 A
111.5 TRIGONOMETRY
In the next four sections we shall study the trigonometry of the hyperbolic
plane and apply this to a number of existence and classification problems.
Our treatment is based on the vector calculus from section III.1 and the
concept of normal vectors from 111.2. Let us use the convention that the natural
orientation of a closed polygon is such that the polygon lies on the positive side of
its edges. In other words, the corresponding normal vectors are inward directed, see
the remarks made after 2.3. Let us start with the hyperbolic triangle AABC.
sinh a - sinh b
sin a - sin Q
sinh c
sin y
c
Proof Let us first prove the cosine relation. Recall from 2.5 that
B = A cosh c + V sinh c , C=Acoshb - Usinkb
Let us use this and <U,V> = cosa to calculate the inner product of B and C
<B,C> = cosh b cosh c - sink b sink c cos a
Observe that <B,C> = cosh a and the cosine relation follows.
In order to prove the sine realations we observe that 1vol(A,B,C)I is
symmetrical in A,B,C. Let us evaluate this in an asymmetrical manner: with the
notation from 2.5 we have
vol(A,B,C) = vol(A,U,V) sink b sinh c
A few lines before 2.5 we found that U A V = A sin a, thus
vol(A,B,C) = sin a sinh b sink c = sin a sinh a sinh b sinh c
sinh a
and the result follows since I vol(A,B,C) I is symmetrical in A,B,C. 0
108 HYPERBOLIC PLANE
Proof Let H,K,L be the inward directed normal vectors of the edges of the
triangle
Proof Let us assume that a > ,3 and > y. We ask the reader to use
formula 5.2 to rewrite the inequality cosh a > 1 in the form
cos(7r - a) < cos(,3 +'Y)
When ,8 + y < it we conclude that 7r - a > 13 + y or a + a + y < 7r as required.
When 7r <,3+y we can rewrite the inequality above as cos(a+7r) < cos(,13+y);
this gives a + 7r < 0 + y, which contradicts a > / and a > y.
111.5 TRIGONOMETRY 109
Theorem 5.4 Let a,Q,y E ]0,lr[ be real numbers with a +,3 + y < a, then
there exists a triangle DABC in H2 with LA = a, LB =,3, LC = y.
Proof Let us rewrite the basic assumption 0 + y < Tr - a and conclude that
cos (f3+y)>cos(ir-a)
Using the trigonometric addition formulas we find that
sin 0 sin y < cos,3 cos y + cos a
It follows that we can determine a E ]0,+Oo[ such that
sin / sin y cosh a = cos 0 cos y + cos a
Let us now pick two points B and C with d(B,C) = a , and draw a geodesic I
through B forming an angle ,3 with BC and a geodesic k through C forming an
angle y with the geodesic h through BC. We let H,K,L denote inward directed
normal vectors for these geodesics
This gives us cos a = <K,L>, which shows that I<K,L>I < 1. We can use 2.6 to
conclude that the geodesics I and k intersect. It follows from 5.2 that the angle of
intersection is a. 0
110 HYPERBOLIC PLANE
0 <K,L> <K,H>
5.7 vol(K,M,L) vol(M,L,H) = det -1 0 <M,H> = <K,H>
0 -1 0
Proof Let us first observe that formula 5.5 remains valid in this context. It
follows from 2.10 that <K,L> = ± 1. In order to decide between +1 and -1,
observe that the left hand side of formula 5.5 is positive. . 0
cosh a sin /3 = 1
6.2 sinh a tan,6=I
lanha sec(3=1
The last two formulas are elementary consequences of the first. In the third
formula, we are using the old convention sec,3 = 1/cos /3.
112 HYPERBOLIC PLANE
7.1
cosh as = sinha2 sinha3
cosha3 = cotha2 cotha4
Proof We shall assume that the enumeration of the vertices is consistent with
the natural orientation of the sides of the polygon, 11.5. For i = 1,...,5 we let Ni be
the inward directed normal vector for the edge P;Pi+1. Let us at once record that
<Ni,N;+1> = 0 ; i = 1,...,5 , N6 = Ni
From the general formula 1.10 (or the more specific formula 5.7), we get that
Evaluate this by means of 2.8 to get the first of the two formulas in 7.1.
Let us now turn to the second formula, which we, by the way, intend to
generalise beyond the case where LP1 is a right angle. We shall first establish the
following formula
Observe that this formula is linear in R and check the formula directly in the case
where R=N1,N2,N3. We ask the reader to use 1.10 to get that
vol(N1,N3,N4) vol(N3,N4,N5) = <N1,N5> + <N1,N3><N3,N5>
Combine this formula with 7.3 with the specification R = N4 to get that
111.7 RIGHT ANGLED PENTAGONS 113
Proposition 7.6 Let a1 > 0 and a2 > 0 be real numbers. There exists a
right angled pentagon with two consecutive sides of length a1 and a2 if and only if
sinhat sinha2 > 1
Proof The condition is necessary as follows from the first formula in 7.1. To
see that the condition is sufficient, start from a right angle LP1P2P3 with
d(PI,P2) = a1 and d(P2,P3) = a2. Draw the geodesic k3 through P3 perpendicular
to P2P3 and the geodesic k5 through P1 perpendicular to P1P2
Formula 5.7, used in the analysis of Lambert quadrangles, at once gives that
sinhat sinha2 = <N3,N5>
It follows that <N3,N5> > 1 which implies that k3 and k5 have a common
perpendicular geodesic, compare 2.9.
114 HYPERBOLIC PLANE
Let us now turn to the right angled hexagon. The method we have used
in the analysis of the right angled pentagon applies to the hexagon as well. We
shall show that right angled hexagons are classified by the length of three
alternating sides a2,a4,a6.
For geometric applications see [Fathi, Laudenbach, Poenaru], in parti-
cular, "expose 3 et appendice aux expose 8". The point is that the basic building
blocks of hyperbolic geometry, hyperbolic pants, can be sewed out of two isometric
right angled hexagons, see VII.2.6.
P4
P1
Proof With a notation similar to the one applied in the previous section
we get from formula 7.2 that
vol(N4,N5,N6) vol(N5,N6,N1) = <N4,N1>
vol(N1,N2,N3) vol(N2,N3,N4) _ <N1,N4>
Using 2.8 we deduce equality between the two first ratios. The remaining formulas
follow by symmetry.
HI.8 RIGHT ANGLED HEXAGONS 115
Proof Let us observe that formula 7.5 remains valid in this context. Thus
the result follows after the specification <N1,N5> = cosh a6 , compare 2.8. 0
Proof Let the number a3 > 0 be determined by the cosine relation 8.2, and
start the construction of the hexagon on the basis of the numbers a2,a3,a4
Let N1 and N5 denote the inward directed normal vectors for h1 and h5 and deduce
from the relations 8.2 and 7.5 that
<N1,N5> = cosh a6
We conclude from this that the geodesics h1 and h5 have a common perpendicular
h6 say. We leave the remaining details to the reader.
z-wI2
9.1
I
coshd(z,w) = 1 + z,wEH2
2Im[z]Im[w]
x _IzI2
9.2 F(z) = 1
Y ;z=x+iyEH2
1 -x
Proof Let z vary along a smooth curve in H2 and differentiate F(z) to get
x -I z2 I
x/ -2(xx'+yy')
F(z)I 32
1 -x 0 -x/
1 -2x I x -x2+y2
_y1
TZF(u) = u v ; u,v E R
0 -1 y2 1 -x
The norm of the tangent vector TZF(u) at the point F(z) is given by
111.9 FROM POINCARE TO KLEIN
Let us end this section with some remarks for readers familiar with differential
forms. The formula 9.6 may be written
9.7 F*vol = 2 dx A dy
y
where the sum is over all vertices P of A. We ask the reader to check that the
area is additive under simple subdivision of polygons. Once this is done, it follows
from 5.3 that Area A> 0. In the Poincare half-plane model the area can be
calculated from the area-form 9.7
There are hints on how to verify this in the exercises. In general, formula 9.8 may
be taken as the definition of area.
118 HYPERBOLIC PLANE
zw -Z
10.1 L(e)
w2 - zw
e- z
w
E R2
'
The matrix L(e) has determinant 0, more precisely the vector e belongs to the
kernel of the linear map with matrix L(e). From this it easily follows that our
construction defines a bijection L: R- + PC(s12(K)). This map obeys the trans-
formation rule (see 1.9.2 for notation)
zv+wu - 2zu
10.3 L(Z,U) = wu 1
zv Z=[z],
w U=[v]
2wv - zv - w u
It follows from 1.6 that the geodesic in H2 with normal vector L(Z,U) has ends
(Z,U). Moreover, we conclude from 10.2 and 10.4 that
We shall now evaluate the inner product of these vectors in terms of the cross ratio
on A. For four distinct points A,B,P,Q we shall prove that
Proof According to 10.5 and 1.9.7 both sides of the formula are invariant
under S12(R). From the fact, 1.9.3, that the action of S12(R) on U2 is triply
transitive it follows that it suffices to verify the formula in the case A = oo, B = 0,
P = 1, Q = q, q E R. Let us recall from 1.9.9 that [oo,0,1,q] = q. Thus it suffices
to prove that
<L(oo,0),L(l,q)> = 9 + l qER
10.8
The column entitled Sylvester type refers to the pencil generated by h and k, see
the end of 111.2.
III EXERCISES 121
III EXERCISES
EXERCISE 2.1 1° Show that the vertices of an equilateral triangle lie on a circle.
2° Show that this is not true for a general triangle in H2.
EXERCISE 2.2 Let h be a geodesic in H2. For a point P E H2 let Ph denote the
perpendicular projection of P onto h. Show that the map P --, Ph decreases
distance, in the sense that d(P,Q) > d(Ph,Qh) for all points P and Q.
EXERCISE 3.2 Let S be an isotropic line in s12(R). Show that the horocycles
with centre S are cut by the affine planes in s12(R) parallel to TS(H2) = S 1.
EXERCISE 3.3 1°Given a triangle DABC. Let PA denote the rotation with
centre A which transforms the ray AB into the ray AC. Show that PA2 = -YO
where y is reflection in the geodesic AB and ,Q is reflection in the line AC.
122 IIYPERBOLIC PLANE
2° Define rotations PB and pC as above and prove that PA2 ° PB2 o pC2 =t.
3° If DABC is equilateral show that PA,PB,pC are conjugated in Isom+.
EXERCISE 4.2 Let A E s12(R) represent a point of H2. Show that the
transformation X r» AXA"1 of s12(R) is half-turn with respect to the point A.
a b
A= _ a,bEC, Ja12-lb12>0
b a
EXERCISE 5.1 Consider a triangle DABC with LC a right angle. Show that
sin A = lim cos A= lim b
a-40 c b-->o c
III EXERCISES 123
EXERCISE 5.2 Show that a triangle LABC has an inscribed circle with radius
EXERCISE 6.1 Let S be an isotropic line in s12(R). Show that the set of vectors
of norm -1 in S 1 has two connected components. Use this to replace the second
half of the proof of 6.1 by a "deformation argument".
EXERCISE 8.1 1° Show that the normal vectors of a right angled hexagon
satisfy the formula
vol(N6,N1,N2) <N1,N3> vol(N3,N4,N5) <N2,N4> =
<N21N5><N2,N6> + <N31N5><N3,N6>
2° Prove the following cotangent relation for a right angled hexagon
cosh a2 cosh a3 = coth a1 sinh a3 + coth a4 sinh a2
3° Investigate an improper hexagon with five right angles in which the sixth
vertex P6 lies at infinity. Hint: Derive a formula for the normal vectors similar to
the formula above (the term <N5,N6> must be added to the right hand side).
EXERCISE 9.1 Let DABC be a triangle in the hyperbolic plane H2, and let
a,Q,y be translation from B to C, C to A resp. A to B.
1° Show that /3ay is a rotation around A of angle -Area A relative to the
orientation defined by the triangle. Hint: For two distinct points P and Q we let
tpQ denote the unit tangent vector at P to the geodesic from P to Q. Justify the
following "mod 27r" calculation
Lor(tCA, aytAB) = Lor(tCA, tCB) + Lor(tCB, aytAB) = LintA - Lor(tCB, atBA)
LintA - Lor(a 1tCB, tBA) = LintA + Lor(tBC, tBA) = LintA + LintB
Let 9 denote the rotation angle in accordance with 3.7. Justify the calculation
Lor(QtCA, QaytAB) = Lor(QtCA, tAB) + Lor(tAB, /3aytAB) = - LintA + 9
Finally, combine the two results.
2° Generalise the result to an arbitrary polygon in H2.
30 Prove a similar result for the sphere S2.
124 HYPERBOLIC PLANE
EXERCISE 9.3 Let the hyperbolic area of say a compact subset . of the
Poincare half-plane H2 be defined by formula 9.8. Show that
Area(o(.)) = Area(.) a E Isom(H2)
2° Given r>O and v E [0,2]. For s>r let E(s) denote the subset of H2 bounded by
the y-axis, x = rcosv, y = s and the circle with centre 0 and radius r. Show that
lims_;00 Area E(s) = 2 - v
3° Prove that the area of .ABC is 7r - LA - LB - LC. Hint: Move the triangle
such that two vertices are located on the y-axis and apply 2°.
IV FUCHSIAN GROUPS
The group PG12(R), being the factor group of the topological group G12(R)
with respect to a its centre R*, carries the structure of a topological group: the
open subsets of PG12(IR) correspond to open subsets of G12(R) stable under R*. We
shall use the action of PG12(R) on H2 to study discrete' subgroups of PG12(R).
But first, we shall exhibit some convenient compact neighbourhoods of the identity
t in PG12(R).
a2+b2+c2+d2 a b
1.2 hall = I det or I
a
c d
Observe the formula II as II = II a II , a E 13*, and conclude that the norm induces a
continuous function on PG12(IR). It follows that a set of the form
i=
1 a b 0 -1 d -b
a( : ) lad - bcl c d 1 0 -c a
The result follows from the fact that sets of the form 1.3 are compact
neighbourhoods of the identity c in PG12(R).
IV.1 DISCRETE SUBGROUPS 127
Proof Consider an orbit A for r and pick a. base point z E A. A disc D with
centre z E A will contain finitely many points only from A as follows from 1.5.
Thus we can find e > 0 such that D(z;c) fl .A={z}. Let us show that in fact
To see this choose o E F with o(z) = u. For a second point v # u of the orbit A
we have o(v) 0 D(z;c) and consequently d(u,v) = d(o(u),o(v)) > c. In order to
see that A is closed, consider a point w 0 A. The open disc D(w;c/2) contains at
128 FUCIISIAN GROUPS
most two points of A as follows from 1.8. It is now easy to find a disc with centre
w not meeting A.
Proof Given a compact set K, a point z E H2 and a number r > 0, let us show
that D(z;r) and a(K) are disjoint for all but finitely many a E F. Choose s > 0
with K C D(z;r+s) and conclude from 1.5 that S = {y E F Iy(z) E D(z;r+s)} is
finite. Observe that yD(z;r) = D(y(z);r) and conclude that K fl yD(z;r) = 0 for all
y E F - S. Given a second compact set L, choose z E L and r > 0 such that
L C D(z;r). It follows that a(K) and L meet for at most finitely many a E F.
Proof For all but finitely many a E F the sets K and a(K) are disjoint. It
follows that only a finite number of elements from r have fixed points on K. 0
Corollary 1.14 Let the locally compact group G act continuously on the
topological space X, and let r be subgroup of G. If there exists a point x E X,
isolated in its orbit rx, and such that the stabiliser rx is discrete in G, then IF is a
discrete subgroup of G.
Let us start the study of subgroups of PG12(R) with a list of its so called
elementary subgroups. By this we mean a subgroup G which fixes a point of H2,
fixes a point of 8H2 or stabilises a geodesic in H2. We shall see that a subgroup
G of PG12(R) is elementary if and only if the group G is solvable.
a b a b
;a2+b2#0
-b a b -a
Proof It is easily seen that the matrices from the list above stabilises the point
i E H2 considered in the proof of 1.1. From formula 1.4 follows that matrix [a d]
stabilises i if and only if
a2+b2+c2+d2 = 2Iad-bcl
When ad - be > 0, this can be untangled via the identity
(a - d)2 + (b + c)2 =a2+b2+c2+d2-2(ad-bc)
The case ad - be < 0 is handled in a similar manner. It follows that the stabiliser
of i in G12*(R) = {o E G12(R) ( dela = ± 1} is 02(R). This shows that the
stabiliser of i in PG12(R) is 02(R)/{-t}.
Ia 0 0 b
;ad$0,be960
0 d c 0
132 FUCHSIAN GROUPS
Proof Let y be the geodesic with normal vector N =[o 01. Let us show that
the matrices we have listed above make up the set of transformations which acts
on s12(R) with N as eigenvector. Direct calculation gives us
0 11 d -b ad+bc -2ab
-1 -c a 2cd -ad-be
Proof it is cleare that a matrix of the above form stabilises the special end
00 = Eo of Conversely, let the matrix a, E G 12( R) stabilise the en d oo. Direct
calculation give s us
a b 0 1 d -b -ac a2
c d 0 0 -c a -c2 ac
I
From this we conclude that c = 0, compare III.4.3.
Discrete groups Let us find the discrete elementary groups. From 2.1 we
find that a group I' which fixes a point is discrete if and only if I' is finite. From
2.2 we find that a discrete group I' which stabilises a geodesic k is discrete if and
only if the translation subgroup is cyclic. In the case of a fixed point on 8H2 we
have the following result.
We ask the reader to verify that after conjugation by a matrix of the form [1 i]
we may assume that r = 0. Let us study the group generated by o and the
commutator y = orO 1T 1
1 c
;u>1
0
I
1
1 c
; nE?L
0
Proof Let us first show that r contains a. translation p which doesn't fix oo:
pick p E I' which doesn't fix oo. The geodesics p(h) and p(k) have p(oo) as an
end. The second end of one of these geodesics must be different from oo. It
follows that one of the translations pap-1 and µrµ 1 doesn't fix oo.
The group r of horolations in r with centre oo is non-discrete as we saw
during the proof of 2.4. Let us investigate the commutator 71Pil 1p 1 as r/ varies
through I'... To this end we put
134 FUCIISIAN GROUPS
a b 1 e
P=
c d 11 0 1
Proof The elementary groups we have displayed in the three first propositions
of this section are easily seen to be solvable. Conversely, let us consider a solvable
subgroup G # 0 and form the derived series of commutator groups
G=D°GDD1GJ ...Di-1GJDT'GD D°+1G={t} DG {t}
Let us first assume that G consists of even transformations. We shall make use of
the classification of even isometries and proceed accordingly.
10 D"G contains a rotation o with centre A. This means that the fixed
point set for o on H2 consists of A only. Let us prove by decreasing induction that
IV.2 ELEMENTARY SUBGROUPS 135
the fixed point set for D'G on H2 is {A}, i = 0,...,n. For i = n observe that the
group D"G is abelian and use the "normaliser principle" 2.6. The induction step
is likewise conducted by means of 2.6.
2° DG contains a horolation o with centre P E (9H2. It follows that the
fixed point set for o on OH2 is {P}. We can proceed as above and conclude from
2.6 that the fixed point set for G on OH2 is {P}.
3° DG contains a translation o $ t along the geodesic k. We shall utilise
the action of G on the set g of geodesics in H2; let us show that gjo = {k}. Let 0
stabilise the geodesic h which means that the set {A,B} of ends for h is stabilised
by o,. It follows that o2 fixes A and B. Observe that v2 $ t has precisely two
fixed points on OH2 and conclude that A and B must be the ends for k, thus h = k.
We conclude as above that the fixed point set for G on fj is {k}.
Let us now return to the general solvable group G and observe that G+ is
solvable as well. When G+ # {t} we can assume that G+ has precisely one fixed
point on either one of the following three sets H2, OH2 or g and we can use the
"normaliser principle" 2.6 to conclude that G has a fixed point as well. 0
Proof Let us make the general remark that the commutator of two rotations o
and r is always a translation. To see this, let their centres be A and B and let /3
denote reflection in the geodesic through A and B. Let us use the theorem on
three reflections 111.3.1 to write o = aO and r = /37 where a and y are reflections
in geodesics through A and B. Direct calculation gives us
orU 1T 1 = a,(3,3"Y(3a"y0 = (a7)33)2
Using that ay,(i that is a glide reflection, compare 111.3.8, we conclude that its
136 FUCHSIAN GROUPS
Proof Let us assume that all odd transformations from G are reflections.
Pick a non-trivial transformation o E G+. If o is a rotation around the point A,
then we conclude from 2.10 that all reflection axes of elements of G pass through
A, and we conclude that A is fixed by G. If o is a horolation with centre S, then
we conclude from 2.10 that all reflection axes from G pass through S, which
implies that G fixes S. If o is a translation then we conclude from 2.10 that all
reflection axes from G are perpendicular to the translation axis s of o. This
implies that s is stabilised by the group G.
Proof Observe that tr2(n;o) = 0, exercise 11.8.1, and use explicit calculation in
the Poincare model taking for k the y-axis. See also exercise IV.3.1.
IV.3 GEOMETRY OF COMMUTATORS 137
In this section we shall work out a number of trace formulas for the
composition of two even isometries and for their conmmutator.
Proof Let us represent a by the matrix A E S12(R) with it A > 0. This allows
us to assign to a the transformation vector U= 2(A' - A). Observe that
it U = 0, otherwise expressed, U E s12(R). The axis a of the translation a has a
natural orientation and M is defined as the corresponding normal vector. We shall
prove that U and M and the translation length T are related through
3.2 U = sinh(1T)
2
M 12 it A = cosh(21T)
To see this, let us pick points P and Q on a such that Q is the midpoint of P and
a(P). Let H and K be positively directed unit tangent vectors to a at P and Q
respectively
138 FUCHSIAN GROUPS
Proof The elementary relation A + A' = t trA gives us U = 2(t irA - 2A) and
similarly V = 2(t trB - 2B). Straightforward calculation yields
-2<U,V> = 1r(UV) = a tr[(t trA - 2A)(t trB - 2B)] _
Irt trA trB + trAB - 4trA 1r2B - 4Ir2A trB
from which formula 3.3 follows. 0
Proof From the formulas 3.2 and lemma 3.5 we deduce that
2 tr(,3a/3"la 1) = cosh 2(IT)
2
+ sinh2(IT)
2
where M is a normal vector for a. Observe that /3(M) is a normal vector for /3(a)
Corollary 3.7 Let a and /3 be two translations with the same translation
length T, whose axes a and b intersect at an angle 0 E ]O,Iir]. The commutator
/3a/3-la' is not a rotation if and only if
sinh2(2T) sin 0 > 1
IV.4 JACOB NIELSEN'S THEOREM 141
When s and r(s) intersect at an angle 0 E ]0,2] say, we can apply corollary 3.7 to
142 FUCIISIAN GROUPS
Let us fix r for a moment. We can then use 4.2 and 4.3 to conclude that the
numbers T. are bounded away from 0 as o runs through r,-{t). It follows that
IF, is discrete, in fact, r, is cyclic generated by an element o of translation length
T, say. For any r E r such that s and r(s) have a common perpendicular we get
from 4.2 that
For any r E I' such that s and r(s) intersect we get from 4.3 that
We shall now study the action of r on N = N(s12(68)), the space of vectors in sl2(lt)
of norm -1. Let us pick a normal vector N for s and observe that the stabiliser
for N is r,. Let us recall from 111.2.5 and 111.2.8 that for the inner product
between N and rN we have respectively
(<N,rN>l = cosh d , I<N,rN>l = cos 0
We conclude from 4.4 that d is bounded away from 0 and from 4.5 that 0 is
bounded away from 0. It follows that we can find a constant c>0 such that
I<N,rN>I 0 ]1 -c,l +c[ ; r o rN
From this we conclude that N is isolated in its orbit rN in N. Recall once again
that rN is discrete and conclude from 1.14 that r is discrete.
Proof Let us assume that r is non-elementary and that the elliptic elements do
not accumulate at t. Let us use 2.8 to pick a proper translation o E I' with axis s,
say, and introduce the group rs made up of the translations in I' with axis s. Let
us now pick a neighbourhood V of t in PG12(R) such that
<N,7-N> < 0 and TUT 1u 1 is not a proper rotation ; o,T E V
Let us analyse a pair of elements T E V-rs. o E rs fl V, o # t. Observe that T is
not a half-turn around a point P of a half-turn 1r around P satisfies
s:
<N,irN> = 1. We conclude that T(s) # s. Let us remark that s and T(s) cannot
have a common end as follows from 2.5 applied to o and 7017 -1. If T(s) intersects
s, then we can observe that
U(TU lr1)U 1(TQ 1r1)-1 = (TUT 1o 1)-2
is not a rotation and we conclude from 3.4 that inequality 4.3 holds. If T(s) and s
have a common perpendicular then we conclude from <N,7-N> < 0 that c 1 and
Tor 1 translate in opposite directions, compare the proof of 3.2. We conclude from
3.1 that inequality 4.2 holds. This is sufficient to conclude that rs is discrete, so
let us pick a definite generator o E rs and change the neighbourhood V of t such
that
<N,TN> < 0 and To7'1o 1 is not a proper rotation ;TEV
At this point we leave the remaining details to the reader.
144 FUCIISIAN GROUPS
IV.5 CUSPS
Proposition 5.1 Given a cusp S E 8II2 for the discrete group I' of
isometries of H2. There exists a horodisc D with centre S E H2 such that
y(D) n D= O ; y E t -Fs
Proof We shall work with the point oo of the Poincare upper half-plane.
Moreover, we shall assume that the group 1 s+ is generated by A = [o i]. Let
B = [ a] E G12(R) with det B = f 1 represent 13 E FF. According to lemma 5.3
we have jcj > 1. An elementary estimate based on 11.8.3 and 11.8.4 gives us
Im[/3(x + iy)] = y Ic(x ± iy) + d 1-2 < y c 2y 2 < y-t
Otherwise expressed
It follows that the horodisc Im[z] > 1 will serve our purpose.
B= c#0, detB= ± 1
Proof Let us proceed under the assumption that Icl < 0. In order to derive a
contradiction we consider the sequence of matrices (Bn)n E N given by BO = B and
1
Bn+1 = BnABn ; nEN
We ask the reader to establish the recursion formula
Using co = c and Icl < 1 we conclude from this that c11-+0 and ancn-+0 for n-+oo.
Put this information into the right hand side of the recursion formula and
conclude that Bn-+A for n-+oo. Let us assume that the corresponding classes a
and (3 generate a discrete subgroup of PG12(R). Then we get that fl. = a for n
large and deduce, by descending induction on n, that oo is fixed by
Since 0 = Qo this contradicts the assumption c t- 0.
With the notation of 5.1, let us observe that the orbit of a point A E 8D
does not meet D: for y E rs we have y(OD) = OD and for y E F-I s we find that
y(8D) and D are disjoint.
Proposition 5.4 Let S E 8H2 be a cusp for the discrete group r of iso-
metries of H2. For any compact subset K of H2 there exists a horodisc D with
centre S E H2 such that y(D) fl K = 0 for ally E F.
Proof Let us return to the proof of 5.1 and choose r>1 such that K lies in the
vertical strip of the Poincare half-plane bounded by Im[z] = r and lm[z] = r"1. For
y E FS we have y(D) = D and for y E f -I s we have y(D) fl K = 0 as a
consequence of the inequality 5.2.
146 FUCHSIAN GROUPS
Corollary 5.5 Given cusps S,T E 49H2 in different 17-orbits, for any horo-
disc E with centre T there exists a horodisc D with centre S such that
y(D)f1E=0for all yEr.
Proof Let us pick a compact subset K of dE such that FTK = E. Use 5.4 to
choose a horodisc D with centre S such that D fl rK = 0. This gives rD f OE = 0.
For y E r we conclude from the connectedness of yD that yD C E or yD fl E = 0.
The first option can be ruled out on the grounds that 7(S) # T.
Proof Let us first observe that the projection H2-H2/r maps open subsets to
open subset. Next, cover H2 with open discs and apply the Borel-Heine theorem
to the images of the discs in the compact space H2/r. The upshot is that we can
find a finite set of discs in H2 which meet all r-orbits. It follows that we can find
a compact set K in H2 which meets all r-orbits. We conclude from 5.4 that F
contains no horolations.
Proposition 5.7 With the notation above, the topological space X = Y/r
is a Hausdorff space in which every point has a neighbourhood homeomorphic to a
neighbourhood of 0 in R2 or a neighbourhood of 0 in R x [0,oo[.
Proof The space x = Y/r is Hausdorff as follows from 5.4 and 5.5. In order
to construct a neighbourhood of a cusp point, it suffices to treat the point oo of the
Poincare half-plane. It follows from 5.1 that it suffices to treat the case where
r = rte. Let us assume that r is generated by zi--+z+k, k>O. The space Y/r is
homeomorphic to the open unit disc D in the complex plane
Y/r - + D , z i-+ exp(21rik-1z) , z E H2
In the general case we can arrange for r to be generated by zF + - i and zF--*z + k
as above. In this case we find that the orbit space X is homeomorphic to the orbit
space for the action of complex conjugation on the open unit disc D in C.
In order to investigate a finite point, it suffices to treat the origin 0 in the
Poincare disc D. According to 1.5 (for more details see VI.5.8) we may assume
that the full group r fixes 0. Let us assume that r+ is generated by zHBz, where
0 is a primitive nth root of unity. Using the well known fact that z-*z" maps open
set to open set, we can make the identification
H2/r+=, D
A non-trivial element of r/r+ corresponds to reflection of D in a line through 0.
Corollary 5.8 The space X = Y/r is compact if and only if there exists a
compact subset K of H2 and a finite number of horocycles D1,...,Dr whose centres
S1.... ,Sr are cusps for r such that any r-orbit in H2 meets K U D1 U ... U Dr.
IV EXERCISES
EXERCISE 1.1 Let 1' be a subgroup of G12(R) and put A = 1' fl S12(R). Let P1'
and PO denote images of I' and A in PGI2(R) and assume that [PI':PD] is finite.
Show that PI' is discrete in PG12(R) if and only if PA is discrete in PG12(R).
EXERCISE 1.2 The additive group U8 of real numbers acts on R/27rZ. Show that
the i-orbit of 1 mod 2a is not discrete.
EXERCISE 2.2 Let a and ,3 be translations whose axes have a common perpen-
dicular. Show that the commutator a/3a-1/3-1 is a translation.
EXERCISE 2.4 Let n. E PG12(R) be a proper translation with fixed points A,B.
1° Show that for any point Q 0 {A,B} the set {t:n(Q) I n E Z } accumulates
IV EXERCISES 149
1rof
at A and B. Hint: Take A=O and B=oo, in which case K =
110, 1
EXERCISE 3.1 Let S E G12(R) and put U = 2(S' - S). Show that U E s12(R) is
an eigenvector for or and use detS = (2 trS)2 - trS2 (exercise 1.3.1) to prove that
<U,U> = a det S (4 - tr2 S)
2° For a second matrix K E s12(R) prove that
<U,K> = IrSK -1 IrS trK
3° Use this to give an alternative proof of lemma 2.10.
EXERCISE 4.1 1° For A E H2 and r > 0 fixed, prove that the set below is
compact
{ N E N(s12(R)) (I<N,A>l < r }
Hint: Pick a basis S,T for the tangent space of H2 at A and use that any point of
F can be written in the form N = xA + yS + zT , x,y,z E R.
2° Let `ll be a discrete subset of N(42(R)). Given a point A E H2 and r > 0,
show that only finitely many geodesics in H2 with normal vectors in 9 will meet
the hyperbolic disc with centre A and radius r.
V FUNDAMENTAL DOMAINS
= erp2ai/3
V.1 TIIE MODULAR GROUP 151
The modular figure 1.1 The three geodesics in the upper half-plane H2
Re[z]=2, Re[z]=-2, zj = 1
bound an open set M which is a fundamental domain for the group PSl2(7).
Proof Let us fix a point w = yi, y > 1 and introduce the three transformations
o(z) = -z 1, T(Z) = z + 1 , W(Z) = z - 1
We ask the reader to verify that the three geodesics bounding M are the per-
pendicular bisectors for w,w(w), w,o(w) and w,T(w). In order to find a point of M
in the orbit through a given point z E H2 we shall minimise the hyperbolic distance
d(w,p(z)), p E PS12(7L). Precisely if v E S12(7L) is chosen such that
d(w,v(z)) < d(w,p(z)) E S12(7)
Then v(z) E M as follows from the inequality
d(w,v(z)) < d(w,c,i 1(z)) = d(w(w),v(z))
and similar inequalities with w replaced by o and r.
Let us consider an orbit 0 of S12(Z) in H2 and let us show that Im[z] is a
bounded function of z E 0 which assumes its upper bound whenever z E M. In
concrete terms, let us consider a z E M and p E r represented by
1.2
We shall see in the next, section that the modular group is generated by o and T.
The transformation p = ro is a rotation of order 3. Thus we have the relations
1.3 2=t , p3
I
0" i p=To-
It follows from Poincare's theorem VII.3.7 that this set of relations is complete.
The hyperbolic triangle A with vertices i, C, oo is a fundamental domain
for the extended modular group PG12(7L) as follows from the fact that reflection 6
in the y-axis divides the fundamental domain for the modular group into two
halves. Let us introduce reflections a,/3,ry in the sides of A as given by the figure
The group G12(F2) is isomorphic to the permutation group S3 through the action
of G12(F2) on the three lines in F2 ®12. This group is lifted back to PG12(7L) by
the dihedral group D3 generated by a and y whose matrices are
1 -1 1 , [ 0 1
0 1 1 0
The level 2 modular group I'(2) = G(2)+ has fundamental domain the
quadrangle 1,0,-l,oo as follows from 1.4 applied to the reflection 0. We shall see
in the next section that r(2) is generated by
z
v(z) =
µ(z) = z + 2 ,
-2z+1
In fact we shall see as a consequence of Poincare's theorem VII.3.7 that the level 2
modular group I'(2) is free on these two generators.
I'(2) = G(2)+
-1 0 1
V.2 LOCALLY FINITE DOMAINS 155
Let I' denote a discrete group of isometries of the hyperbolic plane H2 and
let us consider a fundamental domain P C HZ which is locally finite in the sense
that any point z of H2 has a neighbourhood which meets o(P) for at most finitely
many a E F. We shall relate the group theory of I' and the topology of H2/F to
the geometry of P.
Proof Since our map p is continuous, we must prove that p maps open subsets
of P/r to open subsets of H2/r. To this end it suffices to show that for any given
(relative) open subset B of P the set
V=UyEry(B)
is an open subset of H2. It suffices to consider a point b E B and construct an
open disc with centre b entirely contained in V. Let us use the fact that P is
locally finite to see that the following subset S of r is finite
S={aEIF Ivl(b)EP}
Since P is locally finite we can choose a>O so small that
D(b;c)na(P)=0 ; o¢S
The disc D(b;c) is contained in the union of all IF-translates of P, thus
D(b;) C Uo E S o(P)
Let us take advantage of the fact that S is finite to decrease e > 0 further to get
D(Q 1(b);e) (l PC B ; or E S
Let us combine these two pieces of information to get
D(b;e) = Uo E S o(P) n D(b;e) = Uo E
So(P n D(v 1(b);e)) C Ua E So(B)
This concludes the construction of the required disc. 0
156 FUNDAMENTAL DOMAINS
Proof A closer look at the proof of 2.1 reveals that the projection
P onto open subsets of H2/I'. If H2/F is compact
then we can use Borel-Heine to find a finite set of open discs D1,...,Dn with
centres z1,...,zn in P such that each r-orbit in H2 meets one of the sets
P fl D1,...,P fl D. It follows that P is contained in the union of these discs.
Proposition 2.4 Let P be a locally finite fundamental domain for 1' and
A E OH2 for which there exists a point L E P such that the geodesic segment [L,A[
is contained in P. A transformation y E r+ with y(A) = A is parabolic.
Example 2.5 Let I' denote the group of isometries of the Poincare half-
plane generated by 7(z) = 2z, z E H2. The region P bounded by the two curves
x+i(21xl +1) , x+i(21xl+2) xER
is a locally finite fundamental domain. Observe that y(oo) = oo and that any horo-
disc with centre oo meets P. Thus oo E al,(P) but oo is not a cusp point for I.
158 FUNDAMENTAL DOMAINS
Proposition 2.6 Let A E 0112 be a cusp for the discrete group r and D a
horodisc with centre A. A locally finite fundamental domain P for IF meets (at
least one but) at most finitely many F-translates of D.
Proof Let us assume that the horodisc D is closed and use the remarks fol-
lowing IV.5.3 to pick an open horodisc U contained in D such that there exists a
full F-orbit outside U. Let y E F be a parabolic element with fixed point A.
Moreover, we pick a geodesic It emanating from A and let C be the compact subset
of D - U bounded by h and y(h). Observe that D-U is union of translates of C
by powers of y. Let us enumerate the F-translates of P which meet C
ol(P),...,on(P) ; ol...... n E F
Let us now assume that o-1(D) meets P or what amounts to the same thing that
o(P) meets D. Assume for a moment that o(P) C U; this implies o(P) C U which
contradicts the existence of a full F-orbit outside U. In conclusion o(P) meets
D - U. It follows that we can find s E Z such that yso(P) meets C. This proves
that yso = of for some i = 1,...,n, and we conclude that o(D) = o1-y--s(D) = o1(D).
It follows that P meets at most finitely F-translates of D.
Corollary 2.7 Let F denote a discrete group and P a locally finite funda-
mental domain for F. The f-orbit of a cusp meets the horocyclic boundary 01iP.
Proof Let us pick a horodisc K with its centre at a cusp A and let
K, a1(K)...... on(K) ; 0"i,---,0'n E IF
be a complete enumeration of the F-translates of K which meet P. For any
horodisc L contained in K we find that at least one of the horodiscs
L, ol(L)....... n(L) ; 01,...,on E F
will meet P since P meets any F-orbit. From this it follows that at least one of
the points A,o1(A),...,on(A) belongs to 01iP.
V.3 CONVEX DOMAINS 159
Proof Let S' denote the unit circle in the tangent space TA(H2). Let us define
a map p:U-S' as follows: for Q E U draw the oriented geodesic from Q to A and
let p(Q) E S' denote its unit tangent vector. The map p is continuous and the
image V = p(U) is an open subset of S' which does not contain antipodal points.
Let .A:S'-+S' denote the antipodal map, and observe that V and A(V) are non-
empty open disjoint subsets of S'. Since S' is connected we can find a point s E S'
not in V U A(V). It follows that neither s nor A(s) belongs to V. It follows that
the geodesic through A with tangent vector s does not meet U.
'In terms of Klein's disc model 11.5, the set R is convex in the hyperbolic
sense if and only if R is a convex subset of the ambient Euclidean plane.
160 FUNDAMENTAL DOMAINS
Corollary 3.4 Let U be a convex open subset of H2. The interior of the
closure U equals U.
Lemma 3.5 Let P be a locally finite convex fundamental domain for the
discrete group y. A set of the form
Pflcr(P) 0- 0t,0' Er
is either empty or a closed geodesic segment. A set of the form
Pflo(P)flr(P) ; t#o#r#t, o,rEI
is empty or reduced to a point.
Definition 3.6 Let P be a locally finite convex fundamental domain for the
discrete group I'. By a side of P we understand a closed geodesic segment, not
reduced to a point, of the form P fl o(P) with c E r-{t}. By a vertex of P we
understand a point which can be presented in the form P n o(P) fl r(P),
o,r E r-{t}, o 54 r.
162 FUNDAMENTAL DOMAINS
Returning to the situation at hand, we cannot have P n a (P) = {z} for all o E S
since D-{z} is connected. Thus we have found a side of P through z E 8P. 0
Proof We have already shown during the proof of 3.5 that an interior point
of s is not a vertex of P. Let us investigate a point z of s = P n a(P), which is not
a vertex of P. From the previous proof we find an open disc D with centre z such
that D C P U o(P). This gives us a partition
D = DnP U Dno(P) U Dns
Since D - D n s is disconnected and D n s is a geodesic arc through the centre z of
D, we conclude that D n s is a diameter of D. 0
V.3 CONVEX DOMAINS 163
Proof Consider a open disc D with centre A which meets only those sides of
P which have vertex Az. Let these sides be enumerated cyclically $1,s2,.... sr. Note
also that D fl P is connected since it is a convex set. Let us change the
enumeration above such that D fl P C L(sl,s2). Observe that s fl D C P for
i = 1,...,r and conclude that r = 2. 0
Key lemma 3.12 Let s1 be an edge of P with initial vertex A and let
st,...,Sri... ol,...,Orr,...
be the sequence of *-iterates of st and corresponding side transformations. There
exists n > 2 with t and such that the first of the two unions
n n
U 0"1o2...o1(P) U 0-10-2...oi(P)
i=1 i=1
is neighbourhood of A while the second union is a disjoint union.
Proof Let us make use of the orientation of H2 which has P on the positive
side of is1. Let us define the sign of an edge s of P by sign(s) = +1 when P is on
the positive side of s and sign(s) = -1 otherwise. Observe that
sign(*s) = - det(os) sign(s) ;sEg
Using sign( is) sign(s) we get for s = s1_1 that
sign([si) = det(o1-1) sign(Tsi_1) ; i = 2,...
Let us investigate the angle Lor(jsi,sl) with vertex A1. We have that
Lor(lsi,si) = sign (Jsi)L1ncA1 mod 29r
Observe that o1...oi_1P has vertex A = o1o2...o1_1(A1) and angle
Lor(0*1 ...a1_1 f 1,01...oi_lsi) = det(o1...0i-1)Lor(jsi,si)
From the above formulas and the normalisation sign(js1) = 1 we get that
Lor(ot...oi_1 jsi,ot...oi_tsi) = Lint Al mod 27r
Let us also observe that
ot...oi_t Jsi = ol...oi_1*si_1 = Q1...oi_25i_1
It is now clear that the integer n for which 21r does the job.
V.3 CONVEX DOMAINS 165
Let sl,...,s,, be a full edge cycle, i.e. a full orbit for $ on C. The
corresponding sequence Al,...,Ar of initial points is called a vertex cycle. The
second vertex cycle which starts out with Al is A1,Ar.,Ar_1,...,A2, as follows by
observing that is1,1sl.,Jsi._1,...,Js2 form a full edge cycle.
This gives -y- I = 0102...0j Evaluation of this formula at the vertex Al+1 gives
1(A;+1) = A or Ai = y(A).
y
Corollary 3.15 Let sl,...,sr denote a full edge cycle for the operator 'I,
and let ol,...,or denote the corresponding side transformations. The cycle man
o = ol...or
is a rotation around the initial vertex Al of s1. The order of o is given by
2a = ord(o) (LintA1 + ... + LintAr)
where Al,...,Ar are the initial vertices of the edge cycle.
Proof Let us return to the proof of 3.12. Evaluation of the formula al.... on =
on the edge sn+1 gives us sl = sn+1 Taking into account that r is the length of
the cycle, we find that q = n/r is an integer. From the proof of 3.12 we get that
sign1(s,.+1) = del(o1...o,.) sign(sl)
Since sr+1 = s1, we get that del or = 1 which makes or a rotation. Observe that
27r = E i "1 LintAl = cl E i r l LintAl
and the result follows.
Proof Let us first observe that for any point Q E P and any point A E 8hP the
geodesic segment [Q,A[ is contained in P as follows from 3.1.
Let us assume that A E 8hP is a cusp and let us pick a horocycle D with
center D. Recall from 2.6 that D meets at most finitely many translates of P and
conclude that D meets at most finitely many sides of P. This allows us to shrink
D such that all sides of P which meets D have ends at A. Let us observe that
D fl P # 0 and D T= P and conclude from the connectedness of D that D f OP # 0.
It follows from 3.7 that D f OP is traced on D by a number of geodesics hl,...,hr
through A. It follows from this that D fl P is fibred by traces of geodesics.
Observe also P fl yP = 0 for a generator y of r A. The remaining details are left of
the reader.
V.4 DIRICIILET DOMAINS 167
Lo(w)
4.1
4.2 P(w) =
CEr,o#t
fl Ho(w)
Proof Let us verify that the intersection P(w) fl D(w;r) is open for all r > 0.
Observe that the set S = {o E r-{t} I o(w) E D(w;2r)} is finite and that
P(w) fl D(w;r) = n a E S Ho(w)
168 FUNDAMENTAL DOMAINS
We find that P(w) is open, since P(w) fl D(w;r) is open for all r>O. Next, we ask
the reader to verify the formula
7(P(w)) = P(7(w)) Er
From this it follows immediately that P(w) and y(P(w)) are disjoint for 7 54 t.
It is now time to verify the formula for P(w). Observe that the left hand
side is included in the right hand side since the intersection of a family of closed
sets is closed. To verify the opposite inclusion, consider a point z E H2 which
belongs to Ho(w) for all o E r-{t). Notice that [w,z] belongs to Ho(w) for all o
and conclude that [w,z[ belongs to P(w); it follows that z E P(w).
Let us show that any orbit A meets P(w). To this end pick a z E A with
the shortest possible distance from w. This gives
d(z,w) < d(o 1(z),w) ; o E IF, o 0 t
or d(z,w) < d(z,o(w)) for all o t, which shows that z E H. for all o E F -1t). It
follows that z E P(w), since P(w) is the intersection of such sets.
Proof Let us fix a number r > 0 and show that the open disc D(w;r) with
centre w and radius r > 0 meets o(P) for at most finitely many o E 17. If o(P)
meets P, we can find z E P with o(z) E D(w;r); this gives us
d(w,o(w)) <d(w,o(z))+d(o(z),o(w)) < r + d(z,w)
On the other hand, from the fact that z E P it follows that
d(z,w) < d(z,a 1(w)) = d(o(z),w) < r
But indeed, only finitely many o E I satisfy d(o(w),w) < 2r.
The modular figure from 1.1 is in fact a Dirichlet domain for the modular
group as follows from the following lemma.
V.4 DIRICIILE'F DOMAINS 169
Lemma 4.5 With the notation of 4.2 let S be a subset of r-{r} such that
Ps(w) - fl Hb(w)
b E S
The reader will find much more information about Dirichlet domains in
[Beardon]. In particular, the change of base point for the Dirichlet domain is
discussed and many applications are given. See also [Macbeath, Hoare].
170 FUNDAMENTAL DOMAINS
V EXERCISES
EXERCISE 1.1 The group PG12(Z) contains the congruence group G(2) as a
normal subgroup.
10 Show that PG12(Z) has a unique subgroup N of index 2 containing G(2).
2° Show that the hyperbolic quadrangle i, 1 + i, exp(7ri/3), 00 is a
fundamental domain for N. Show that the sides of the quadrangle can be paired
by a rotation about exp(3ai) and a glide reflection fixing oo, both belonging to N.
EXERCISE 1.2 Show that PS12(Z) contains three subgroups of index 3 which
contain I'(2). Find fundamental domains for these groups.
VI COVERINGS
For the convenience of the reader we shall recall the definition of the arc
length of a curve y:[a,b]--X in a metric space X. To this end we consider a
subdivision a = a0 < a1... < an = b of our interval and form the sum
n
E d(y(ai-1),-y(ai)) ; a=a0<al... <an=b
i=1
We say that the curve y:[a,b]-%X is rectifiable if y is continuous and sums of the
above form are bounded. The length of a rectifiable curve y:[a,b]-:X is defined to
be the supremum over sums of this form taken over all subdivisions of [a,b]
It follows from the triangle inequality that d(x,y) < l(y). We shall mostly be
concerned with metric spaces which have the following property.
172 COVERINGS
Definition 1.2 We say that a metric space X has the shortest length
property if, first of all, any two points x and y of X can be joined by a rectifiable
curve y. Moreover, it is required that
d(x,y) = inf7 1(y)
where the infimum is taken over all rectifiable curves in X joining x to y.
The hyperbolic plane Hz has the shortest length property: the distance
d(x,y) between x,y E H2 is realized as the arc length of the geodesic arc [x,y]. We
shall be concerned with metric spaces X with the shortest length property which
locally look like H2.
Lemma 1.3 Let X be a metric space with the shortest length property and
D an open disc in H2 with centre a and radius r. For any isometry o:D-+X of D
onto an open neighbourhood of o(a) in X we have d(x,o(a)) > r for all points
x E X - o(D).
Proof Let there be given a rectifiable curve y in X which joins o(a) to the
point x outside o(D). For a number p E ]O,r[ let K denote the compact disc in H2
with radius p and centre a and let us prove that y meets o(OK). Observe that
o(K) is a compact subset of X and conclude that V = X - o(K) is an open subset
since X is Ilausdorff. This gives us a partition of X into the closed subset o(8K)
and the two open subsets V and U = o(K - OK). Observe that y meets V and U
and conclude from the connectedness of the image of y that y meets o(8K) in y
say. This gives us 1(y) > d(o(a),y) = p. Take supremuwn over p E ]O,r[ to get
that 1(y) > r. Finally, take infimum over all y and conclude that d(o(a),x) > r.
where the infimum is taken over all continuous curves -y in X joining x to y such
that fy is rectifiable in Y. We shall establish, during the proof of 1.6, that this
defines a metric on X.
Proof Let us assume that 0 E J and let us introduce the following set
E = Is E ]0,+00[ l a and 0 agree on [0,s]}
From the very definition of E it follows that E is a non-empty interval with 0 as
its left end point. Let us show that E is an open subset of R. To this end we
consider a point s E E and a small open disc D in X with centre a(s) =,3(s). We
conclude from our study (made during the proof of 11.4.8) of geodesics in H2 that
a and ,6 agree in a neighbourhood of s E It
Let us show that E is unbounded. To this end we must rule out the
alternative E=]O,p[ for some p E It by the definition of E we find that a and Q
agree on [O,p[. By continuity of or and 3, this implies that a and 3 agree on [O,p],
and we get p E E, a contradiction. This argument shows that a and ,Q agree on
[0,+oo[. The interval ]-00,0] is handled in a similar manner. 0
176 COVERINGS
Proof Let us first remark that the result is true for X = H2 as follows from the
proof of 11.4.8. In the general case let us assume that 0 E J and try to extend y to
the right of J. To this end put
j = { x E ]0,+co[ 1 y extends to ]0,x[ }
The set is a subinterval of ]0,+oo[. Let us prove that a E I implies a+e E j for
some a>O. From the inequality 2.1 we conclude that y:[O,a[-+ X is uniformly conti-
nuous. From the completeness of X it follows1 that y can be extended to a conti-
nuous curve on [O,a]. In order to extend y to the right of the point a consider a
hyperbolic disc in X with centre y(a) and use the result for H2. It follows that
I = ]0,+oo[. Extension of y to the left of J can be done in a similar way.
Proof Let us introduce an abbreviation which will facilitate the exposition and
underline the strategy of the proof
Let us first show that for any two distinct points x and y of X, we can find z # x
such that %%(x,z,y). To this end choose an r E ]O,d[ such that the disc D(x;p) is
hyperbolic for some p > r. The circle S(x;r) with centre x and radius r is compact
as follows from 1.3. Let the continuous function
z ,--, d(z,y) ; z E S(x;r)
take its minimum at the point z E S(x;r). Consider a rectifiable curve y:[a,c]--, X
with y(a) = x and y(c) = y. Choose a point b E ]a,c[ with d(x,y(b)) = r. We have
1(y) _ k(-Y(a,h)) + 1(y[b,e]) >_ d(x,7(b)) +d(7(b),Y) > r+d(z,Y)
Take infimum over all rectifiable curves y from x to y to get
d(x,y) > d(x,z) + d(z,y)
The opposite inequality is the triangle inequality and we have verified ]b,(x,z,y).
Let us establish the basic relation
Simple manipulations of the two identities and the triangle inequality yield
d(x,w) + d(w,y) < d(x,z) + d(z,w) + d(w,y) = d(x,z) + d(z,y) = d(x,y)
Another application of the triangle inequality gives us
d(x,w) + d(w,y) = d(x,y)
Put this information back into the first inequality and deduce that in fact
d(x,z) + d(z,w) = d(x,w)
It follows from this and lemma 2.6 that x and w satisfy the Hopf-Rinow theorem.
To conclude the proof, let x and y be points with d = d(x,y). Pick a point
v $ x satisfying N (x,v,y), put e = d(x,v) and use 2.3 to construct a geodesic
curve o:R- X with o(0) = x and o(e) = v. We shall study the set
J = { s E [O,d] I d(x,o(s)) = s & d(o(s),y) = d - s }
It is easily seen that the set J is a closed subinterval of [O,d] containing 0. Let us
178 COVERINGS
show that s E ]O,d[ fl J implies t E J for some t > s. Observe that z = a(s) satisfies
3G%(x,z,y). Pick a point w $ z such that 3G%(z,w,y) and conclude from 2.8 that
36,(z,w,y). It follows from 2.6 and 2.3 that w belongs to the image of o. The
parameter t E R with o(t) = w satisfies our requirements. It follows from this that
J=[O,d]. In particular d(o(d),y) = 0, i.e. o(d) = y. 0
Lemma 2.6 Let -r :[a,c]-+X be a curve in the metric space X and let there be
given b E [a,c] such that
d(7(a),7(b)) + d(7(b),7(c)) = d(7(a),7(c))
If the restrictions y:[a,b]-+X and y:[b,c]-X are distance preserving, then the curve
y:[a,c]--->X is distance preserving.
VI.3 UNIFORMIZATION
Proof Let us pick three points A,B,C of D not on a geodesic. We can use
lemma 11.4.8 to pick an isometry r:H2_H2 which agrees with o at the three points
A,B,C. It follows from the proof of 11.4.9 that the restriction of r to D is actually
equal to a. 0
Proof Let 0 denote the centre of D and let us use 2.3 to extend the
isometry D-+X to a map 0:H2-.X whose restriction to each hyperbolic line
through 0 is a geodesic. Let us fix a point A of H2 and proceed to find an open
disc with centre A which is mapped isometrically into X by 0. To this end use the
Borel-Heine theorem to find an r > 0 such that for all C E [0,A] the metric disc
D(O(C);2r) is a hyperbolic disc. Let us first prove that
d(,A(A),O(B)) = d(A,B) ; B E D(A;r)
To this end consider a subdivision of the segment [0,A]
0 = A0, A1,.... Ai_1, A = A d(Ai_1)A;) < r , i = 1,...,n
and let B0,B1,...,Bn be the perpendicular projections of the sequence of As onto
180 COVERINGS
and the image configuration in `.bi. From the induction hypothesis we conclude that
LAi-1A;B1-1 = L0(Ai-1)0(Ai)0(Bi-1)
LBi-iA;Bi = L0(Bi-1)c5(Ai)c5(B;)
with the conclusion that
LA;_1A;B; L0(Ai-1)0(Ai)0(Bi)
From this we draw the same conclusion for the complementary angle
LAi+1A;Bi =L0(Ai+1)0(Ai)0(Bi)
We can now use elementary geometry to conclude that
AAi+IAiBi = A0(Ai+1)0(Ai)0(Bi)
In a similar way we can first deduce that
LBi-1BiAi = Ld'(Bi-1)0(Bi)0(Ai)
LA,B;AJ+1 =
with the conclusion that
LAi+1BiBi+1 =
Another reference to elementary geometry gives us
AAi+1B;Bi+1 A0(Ai+1)0(Bi)0(Bi+1)
which concludes the induction. Let us now prove that
AAnB11B OO(An)O(Bn)O(B)
Observe that LBn=LO(Bn) is a right angle, and conclude as above that the
corresponding sides have equal length. In conclusion d(A,B) = d(O(A),0(B)). Let
us remark that at the same time we have proved that the following two angles are
equal LAn_1AB = L0(Ai-1)0(A)0(B).
It remains to prove that the distance between any two points B and C of
D(A;r) is preserved. To see this, observe that the proof above shows that the
angle LOAB is preserved by 0. A similar remark applies to the point C and the
result follows by hyperbolic trigonometry, 111.5.1.
VI.4 MONODROMY
Proof Consider a point x E X and an r > 0 such that the metric disc D(x;r) is a
hyperbolic disc and such that the restriction f:D(x;r)-Y is an isometry. From the
fact that f is open it follows that D(x;r) is mapped homeomorphically onto an
open neighbourhood of f(x) in Y. We conclude from lemma 1.3 that
f(D(x;r)) = D(f(x);r)
Let there be given points x,y E X and a continuous curve y:[a,b]-X with y(a) = x
and y(b) = y. We ask the reader to use the Borel-Heine theorem to find an r > 0
such that for all s E [a,b] the point y(s) is the centre of a hyperbolic disc with
radius r such that f maps D(y(s);r) isometrically onto D(fy(s);r). A simple
subdivision argument shows that y is rectifiable in X if and only if fy is rectifiable
in Y. For a rectifiable curve -y in X we find that
1(fy) = 1(y)
The formula 4.1 follows from variation of y. 0
Proof Choose r>0 such that f induces an isometry D(x;r)- D(y;r). Next,
choose an open interval J around 0 in R such that 71(J) C D(y;r). Let f:J-D(x;r)
VI.4 MONODROMY 183
be given by ff =,9. Let us use 2.3 to extend to all of O and use 2.2 to conclude
that the identity ff _ 17 is generally valid. 0
Proof We have already observed that f(X) is open. Let us show that f(X) is
closed. To this end consider a point w in the closure of f(X). Pick a hyperbolic
disc D with centre w and a point y E D fl f(X). Let us join the centre w of D with
y along the radius of D and extend this to a geodesic ij:l ->Y. According to 4.2 we
can find a geodesic x:R,X with p = From this it follows that all points of y
belong to f(X). Since f(X) is open and closed we can use the fact that Y is
connected to conclude that f(X) = Y.
Let us now investigate the point x. For small values of r > 0, the map f
induces an isometry D(x;r)-+D(y;r). The inverse of this induces an isometry
s:D(y;r)-+X with fs = t and s(y) = x. Let us use the uniformization theorem 3.2 to
extend s to a local isometry s:D(y;p)--X. The extension will still satisfy s(y) = x
and fs = c as follows from 2.2 applied to the various radii in the disc D(y;p). In
fact s is an isometry: for u,v E D(y;p) we get from the inequality 4.1 applied twice
d(u,v) = d(f(s(u)),f(s(v)) < d(s(u),s(v)) < d(u,v)
which shows that d(u,v) = d(s(u),s(v)). We conclude from the proof of 4.1 that
s(D(y;p)) = D(x;p). Let us recall the relation fs = t and it follows that D(x;p) is
mapped by f isometrically onto D(y,p).
Proof Let us put D = D(y;p) and make a general remark on two continuous
sections r,s of f:X-}Y over D C Y. Let us show that
r(D)fls(D)54 0 r=s
We may write a common point u of r(D) and s(D) in the form u=r(vl)=s(v1),
where v1,v2 E D. This, however, gives f(u) = v1 = v2. Thus we have found a
v E D with r(v) = s(v). Observe that the connected set D is partitioned into the
two sets
{wEDIr(w)0s(w)}, {vEDIr(v)=s(v))
The first set is open since r and s are continuous and X is Hausdorff. To see that
the second set is open consider a point v E D with r(v) = s(v); choose a neigh-
bourhood U of r(v) = s(v) such that the restriction of f:X-'Y to U is injective and
choose a neighbourhood V of v in D with r(V) C U and s(V) C U. We have that
w= f(r(w)) = f(s(w)) ;wEV
which gives r(w) = s(w) for all w E V. In conclusion we have partitioned the
connected set D into two open subsets of which the second is non-empty. It
follows that r = s.
To conclude the proof of the first statement, observe that at a point of the
fibre x E f-1{y} we can use the restriction of f to the disc D(x;p) to get an
isometry ,: D(x;p) D(y;p). The inverse s,, = f,-' defines a section s, of
f:X--*Y over D with s,(D) = D(x;p).
Consider a point w E X with d(f(w),y) = r < p. Let q:R-4Y denote a
geodesic with q(O) = f(w) and il(r) = y. Let us use 4.2 to find a geodesic e:R-+X
with f. = p and (O) = w. This gives f(i(r)) = zl(r) = y. In conclusion we have
found a geodesic {: R-X with (0) = w and fi(r) = x E f-1 {y} where r E ]0,p[. It
follows that w E D(x;p) as required.
VI.4 MONODROMY 185
Proof Use the Hopf-Rhiniw theorem 2.4 to pick a geodesic curve il:[O,d]->Y
with 9(0) = y and q(d) = w, where d = d(y,w). Let :[O,d]-*X be a lifting of rl to
X, compare 4.2. This gives points x = (O) and v = (d) with f(x) = y and
f(v) = w. It follows from the inequality 2.1 that d(x,v) < d. The required identity
is now a consequence of the general inequality 4.1.
Proof Let us first show that any local isometry o:H2-+H2 is an isometry. To
this end choose an open disc D C H2 such that the restriction o': D-+H2 is an
isometry. Next use 3.1 to extend o' to a global isometry o": H2-*H2. Observe
that or = o" as follows by restriction to hyperbolic lines through the centre of D.
In the general case, let us use the uniformization theorem 3.2 to construct
a local isometry ir:H2-+X. The composite Oir:H2-H2 is a local isometry and
consequently a bijection. We conclude from this that it is injective and from 4.3
that rr is surjective. This shows that 0 is a bijection as well.
186 COVERINGS
For the G-orbit A through x we conclude the existence of r > 0 such that
D(x;r) fl A = {x}
For two distinct points p,q of A we can choose o E G with o(p) = x and conclude
that d(o(p),o(q)) > r. Since o is an isometry we can conclude that
We intend to put a metric on the space X/G of orbits. To get started let
us introduce the distance between two G-orbits A and `.9 by the formula
Proof When X is complete, we get from 5.7 that any closed metric disc in X is
compact. It follows from 5.4 that any closed metric disc in X/G is compact. To
prove that a Cauchy sequence {z11) in X/G is convergent we can refer to the
elementary fact that a Cauchy sequence is bounded, i.e. contained in some closed
metric disc in X/G. Let us finally recall the fact that a Cauchy sequence in a
compact metric space is convergent.
Suppose conversely that X/G is complete and let us investigate a Cauchy
sequence {x17} in X. The sequence {p(x11)} is a Cauchy sequence in X/G and
converges towards a point y E X/G, say. Let 9J be the corresponding G-orbit in X
and choose r>0 such that the discs D(x;3r)x E c are disjoint. Let us choose an
n E N such that d(p(xn),y) < r and such that d(xn,xp) < r for all p > n. Let the
point x E qJ be given by the condition d(x1l,x) < r. It follows that
xp E D(x;2r) ; p>n
Upon shrinking r we can obtain that the closure of D(x;2r) is compact, compare
1.3. It follows that the sequence {x11} is convergent in X. 0
VI-6 CLASSIFICATION 189
VI.6 CLASSIFICATION
Consider points z,w E 112 with x = a(z) = a(w). Let us show that we can
find a transformation y E r with y(z) = w. The restrictions it and
?r,,:D(w;r)->D(x;r) are isornetries for small values of r > 0. Introduce the isometry
0 = aw 11r : D(z;r) -' D(w;r)
and use rigidity 3.1 to extend 0 to a global isometry o of H2 into itself. The
identity ao = n holds in D(z;r), but it extends to all of H2 by rigidity 3.1.
We can now use 7r:H2->X to produce a bijection 0: H2/r X. But 4.5
allows us to conclude that 0 is an isometry.
Let us now study local isometries from H2/r to H2/E. Let us first show
that a o E PG12(U8) with arc r-1 C E induces a map
In other words, we must verify that " x - y mod r o(x) u(y) mod E ".
To this end we write y = 7(x) with y E r to get o(y) = oy(x)) = (uyo1)o(x) and
observe that oyo1 E E to get o(x) = o(y) mod E as required. It follows from
the remarks from beginning of this section that the map 6.2 is a local isometry.
Let us finally verify that the transformation 6.2 is an isometry if and only
if oro' = E. To see this make the factorization of 6.2
H2/r H2/oro 1 , H2/E
where the first map is the isomorphism induced by o and the second map is simple
projection. This second map is bijective if and only if oro 1 = E. 0
Example 7.2 Consider the closed interval Z=[0,1] with the identifications
t = 1 -t t E ]0,1[. The "instant jet service" distance between {0} and {1} is 0.
192 COVERINGS
Proof Let us first observe that a sum of the type which occurs in 7.1 gives rise
to the following inequality, where we have put z11 = wn
n n n
E d(zi-1,wi) ? E d(f(zi-1),f(wi)) ? E d(f(zi-1),f(zi)) ? d(f(zo),f(wn))
i=1 i=1 i=1
Take infimum over all sequences of the type considered and conclude that the map
f:X/g->W is distance decreasing.
Proof A simple application of lemma 7.3 shows that f:Z/gj X/I decreases
distances. Let us prove that the inverse map f-1 decreases distances. To this end
we consider points z,w E Z with equivalence classes Z,W E Z/gj and show that
d(z,u(w)) > d(%,°W) ; o E 1'
Let us fix c E I' and consider a rectifiable curve y:[a,b]-X with y(a) = z and
y(b) = c(w). We shall show below that there exists a subdivision of [a,b]
a = to < t1... < to = b and elements i = c1,...,cn = c of r such that
n n
1(7) ? . d(y(ti-1),7(ti)) _ d(o;(zi_1),oi(wi)) _ . d(z;_1,wi) ? d(`%,`W)
Finally, make a variation of y and deduce that d(x,o(w)) > d(Z,` V) as required.
Let us start the construction of a subdivision of the interval [a,b] subject
to 7.5 by observing that y([a,b]) meets only finitely many I'-translates of Z as
follows from Borel-Heine. When y(b) E Z we can use n = 2 and ti = b. Let us
assume that o(w) 0 Z and let tl E [a,b] be the last time -y leaves Z:
t1 = sup{t E [a,b] I y(t) E Z)
Let us observe that any union of I'-translates of Z is a closed set as follows from
the fact that the covering of X with F-translates of Z is locally finite. We can
now conclude that
1(t1) E U , 54 , o(Z)
since the alternative would make y(t1) an interior point of Z, contradicting its
definition. Choose 0-2 of = c such that y(tl) E o2(Z). When y(b) E o2(Z) we
can use n = 3 with t2 = b. If y(b) ¢ o2(Z), let t2 E [a,b] be the last time y leaves
o2(Z) and observe that
y(t2) E U o 0 al p2 o(Z)
since the alternative makes y(t2) an interior point of ol(Z) U o2(Z) contradicting
the definition of t2 and y(b) 0 o2(Z), that is t2 < b. This procedure terminates
since y meets only finitely many F-translates of Z.
We are going to stydy some examples which are usefull for the
understanding of Poincares theorem in VII.3 in case of unbounded polygones.
h k
zi=awi
Consider the sequence {zi}i E N in Z given by the recursion formulas above starting
from zo=a+ib, b > 0. It will be useful to observe that Im[zi] = a'b, i E N. Let us
recall from 11.8.2
d(z,w) < 2 sinh 2 (z, w) = 1w - zi (Ir[w]Im[z])-1/2
Recall that a > 1 to see that the infinite series Ed([zi_1],[zi]) is convergent and
conclude that [zi]i E N is a Cauchy sequence in Z/R. In order to show that our
sequence is not convergent, consider the function f:Z-HIS given by f(z) = y/x. The
function f induces a continuous function on Z/% (see exercise 5.2). If our sequence
was convergent, then f would be bounded on the sequence, which is not the case.
Proof Suppose that the ends of h and k are distinct. Consider a sequence
(wn)n E N of points of Z representing a Cauchy sequence in Z/`3G. We shall prove
that the sequence is bounded in H2. Once this is done, we conclude that
{wn}n
EN
is contained in a compact subset of Z. It follows that our Cauchy
sequence is contained in a compact subset of Z/%, which makes it convergent.
Let us introduce the geodesic s through A and D and consider the function
f,:H2-+R which is zero on the half-plane containing B and C, while its value at a
point z on the other half-plane is the hyperbolic distance from z to s. We ask the
reader to show that f,:X;R is distance decreasing and we conclude from lemma
7.3 that the sequence f,(zn)n E N is a Cauchy sequence. This implies that we can
find a constant c such that
fs(z1l) < c ;nEN
The same conclusion is available for the geodesic through B and C.
In order to show that our initial sequence is bounded away from the points
A and D, pick a horocycle A with centre A and define the polar function
pA:H2--+R which is zero outside the horodisc bounded by A while the value at a
point z of the horodisc is the distance from z to A measured along the geodesic
through z and A. Let us pick a second horocycle 9 with centre D disjoint from A
and such that c(A fl h) = 9 fl k. The function p = 2(pA + pct) is distance
decreasing and constant on the equivalence classes modulo `J,. It follows from
lemma 7.3 applied to p:Z-*R that p is bounded on our sequence. The same
conclusion is available for the points B and C. Thus we have proved that our
sequence is bounded in H2.
Let us now assume that A = D and that a is a horolation with centre A.
Pick a horocycle A with centre A and observe that the function PA is constant on
the equivalence classes of !. As above we conclude that Z/`R is complete.
When A = D and a is a translation we can use example 7.6 to conclude
that Z/% is not complete. 0
196 COVERINGS
V1.8 HOMOTOPY
I I f
T x [0,1] -, Y
can be completed with a dashed arrow to a commutative diagram in the category
of continuous maps.
Proof Let us use the uniformization theorem 3.2 to find a local isometry
f:H2-+X and let us use 4.3 to pick a base point z E H2 with f(z) = x. For a curve
7:[a,b]-+X with y(a) = x and y(b) = y there is a unique curve ,6:[a,b]-+H2 with
,6(a) = z and f,6 = y. It follows from 8.2 and 8.3 that the homotopy type of ,6 is
determined by the point ,3(b) E f-1 {y}. The image in Y of the geodesic line
segment [,6(a) 6(b)] represents the homotopy class of ,6. The uniqueness in the
statement follows from the geodesic lifting property 4.2.
Let us recall that two continuous curves a,/3:S1-X are freely homotopic if there
exists a continuous map W:S1 x [0,1]-X such that
'(s,0) = a(s) , 'Y(s,1) = Q(s) s E S'
Proof With the notation from the proof of 8.1, let us consider a curve
g:[0,1]-X with g(0) = g(1) = x. Let h:[0,1]-i[2 be a lifting of g with h(0) = z and
let y E IF be determined by h(1) = yz. Since g is not homotopic to zero we have
y # t. We conclude from IV.5.6 that y is a translation with translation axis say k.
We intend to show that the restriction of the projection f:H2-X to k defines a
closed geodesic L---+X which is freely homotopic to g. A free homotopy is formed by
the restriction of f:II2--*X to the hyperbolic quadrangle given below
7z
VI EXERCISES
EXERCISE 1.2 Let us fix a point u+iv of the Poincare half-plane and consider
the real function V)(x,y) in two variables given by
1G(x,Y) = d(x+iy,u+iv)2
10 Show that the Hessian of this function at the point u+iv is
= v(dx2 + dy2)
d2.0(x,Y)
EXERCISE 1.3 Show that a circle in H2 with radius r has arc length 27r sink r,
compare VI.1. Show for example first that the length k11 of the edge of an inscribed
n-gon is given by the formula
sinh(k,i) = sinh(r) 2 sin(7f/n)
and use this to calculate the length of the circle as lien nkn
n--;00
EXERCISE 1.4 Consider a fixed real number y > 0 and let Clf denote the
Euclidean line in the upper half-plane through iy and parallel to the x-axis.
VI EXERCISES 199
10 Show that 'J is a horocycle, and that the are length is given by
length (iy,x+iy) = ; x>0
Y
2° Verify the formula
length (iy,x+iy) = 2 sinh 1-2d(iy,x+iy) ; x>0
EXERCISE 2.1 Let X be a metric space and 7:[a,b]-#X a geodesic curve. Show
that the curve y is rectifiable with arc length 1(y) = b - a.
EXERCISE 5.2 Let a be the isometry of the Poincare half-plane given by zp- az,
where a > 1 is a fixed real number.
10 Show that the group IF generated by a acts discontinuously on
X = {zEH2IRe[z]>0).
2° Let f:X-Q be the function given by f(x+iy) = y/x, x+iy E X. Show that
f induces a continuous function on the surface X/I'.
3° Show that the surface X/I' is not complete. Hint: Use 5.9.
EXERCISE 7.1 With the notations and assumptions of proposition 7.7. Show
that the set X = U EZ
onZ is open and connected. Prove that Z/' is
complete if and only if X = H2. Hint: Apply propositions 5.9 and 7.4.
1.1 -1
;sEg
Recall that an edge s is an oriented side and that s-1 denotes the result of changing
the orientation. For s E 6 we let ors denote the isometry of H2 which maps *s to s
in such a way, that locally, the half-plane bounded by *s containing A is mapped
to the half-plane bounded by s but opposite A. The isometry as is called the it
transformation determined by s. Let us at once observe that we have the
Given an edge s of A, we let Is denote the edge with the same initial vertex as s
but different final vertex. This defines for us a second involution on 6, su- s. The
combination of these two operators defines the edge operator T:9-9 given by
Lemma 1.4 Let sl,...,s, be an edge cycle with vertex cycle P1,...,Pr and side
transformations (oi = o-S , i = 1,..,r). The cycle ma o = 03 r is a
rotation around P1 of angle (measured in the direction determined by A and sl)
congruent mod 27r to the sum of the interior angles
LintPi + LintP2 + ... LintPr
Proof Let us choose the orientation of 112 which places A on the positive side
of s1 near P1. With the notation of 111.2.4 we have
Lor(si,1si) = sign(si) LintP1 mod 27r ; i = 1,..,r
where sign(si) = ei equals +1 (resp.-1) in case z lies on the positive(resp.
negative) side of si. From the very construction of the side transformations it
follows that
sign(sm) = sign(s,u+1) m = 1,...,r
Let us multiply these equations together for m = 1,...,r to get that
sign(s1) = det(ol...or) sign(sr+1)
Since sr+1 = s1 we conclude that the cycle map is even. Let us proceed to
calculate the rotation angle of the cycle rrta.p. Let us prove a general formula
Ei=m...r.int Pi = emLor (sm, om...or(sl)) m = 1,...,r
To this end we recall that
si = oi(*si) , °i+1 = *si , (Si = 0i(si+1)
202 POINCARE'S THEOREM
This will be done by decreasing induction on in. The case m = 1 runs as follows
LintPr = Er' or(Sr'Tsr) = ErLor(Sr,0r(Sl))
The induction step m+1 follows from the formulas we have just derived
Ei=m.... rLint Pi = emLor(sm'lsm) + em+1Lor(Sm+1, 47m+1"'0r(Sl))
emLor(Sm,lsm)+Em+l det(om)Lor(lsnl, am"'Qr(Si)) _
emLor(Sm,lsm) +EmLor(1Sm,Om...Or(S1)) = EmLor(sm, om...or(sl))
This concludes the proof of our formula. Finally, put m = 1 in our formula and
recall that el = 1 to get the announced rotation angle. 0
P3
a=0a, 0 =0-b,'Y=0-c1 b=od.
Our next example is an isoceles triangle with the side pairing indicated on the
drawing, the base edge c is paired with itself *c=c. The side transformation a=oa
is a rotation with centre C, while -y=ac is a reflection in the geodesic through c.
C
Edge cycles a *a 1 , a c *a c
Vertex cycles C C , BAAB
Cycle maps a, a 1
, ayay
Let us consider a different side pairing on the same triangle given below: again
*c = c. The transformation y = o, is a reflection as before, while the
transformation a = oa is a glide reflection.
Remark 1.5 Let us consider an edge cycle SI S2,...Sr. From sj+1 = 1*sj,
i = 1,...,r, we conclude that Jsj+1 = *si. This gives us
I'(lsi+1) = ls, ; i = 1,...
It follows that lsl,Jsr,lsr_1,...,1s2 form an edge cycle. Notice that the corresponding
vertex cycle is P1,P,,Pi_1,...,P1, which is nothing but the reverse cycle of the
original vertex cycle. It follows that a vertex cycle makes up a full equivalence
class of A under the equivalence relation generated by the side pairing. Let us
observe that
1S111Sr,1sr_11 ...,152 = *Sr,*Sr_1,...,*S1
and conclude that the cycle map is o,. 10,, 2 1...01 1, which is the inverse of the
cycle map of the original cycle.
Cycle condition 1.6 We say that an edge cycle s11 ...,sr satisfies the
cycle condition if the angle sum along the corresponding vertex cycle P1,..-,Pr has
the form
The cycle map o = al-or of an edge cycle which satisfies the cycle condition is a
rotation about P1 of order no.1
Proof Let I' denote the universal group on the generators as , s E g, subject to
the reflection relations 1.2 and the cycle relations. The canonical map F-+G is
denoted On the basis of these data we shall construct a topological space X
and a local homeornorphism f: X- H2, compare the introduction to chapter VI.
The set underlying X is the quotient space for the equivalence relation on r x A
generated by relations of the form
The equivalence class of the point (y,k) is denoted y. k E X. Let us equip I' x A
with the product topology: an open subset has the form U = U-7 E r {0.) x U0.
where U. is an open subset of A, u E r. The topology on X is the quotient
topology: open subsets of X correspond to open, - stable subsets of IF x A.
The evaluation map (y,h)i-- (h), I' x A-+H2 is continuous and compatible
with our equivalence relation, so it will induce a continuous map f.X-*H2
y(h) ; 7EI, heA
The natural action of r on r x A induces an action of r on X through the formula
a(13. s) = (a13). s I',sEA
Let us make the observation that f:X- H2 is F-equivariant in the sense that
f(rx)=rf(x) ;rEI ,xEX
Let us prove that X is connected. To this end let us note that sr-+t.s induces a
continuous section of f:X-H2 over A and conclude that t.A is a connected subset
of X. For x E X given, let us construct a connected subset of X containing x and
meeting t.A. It is clear from the construction that X is the union of all 17-translates
VII.1 COMPACT POLYGONS 205
of t.0. So let us pick r E r with x E r.4 and let us write r = al...a. where the cis
are side transformations with respect to the edges s1,...,sn of A. For i = 1,...,n we
get from 1.8 that t.0 meets oi.A, which implies that Oi...Oi_i.0 meets
The chain
t.0 U o 1.0 U QiQ2.0 U ... 0, 10'2...On_1.0 U T.0
is a connected set of the required sort.
We proceed to construct some special neighbourhoods of a given point
x E X. For convenience we introduce the following terminology concerning the
equivalence class of r x A represented by x. But first let us agree that for z E A
and c > 0 we shall put A(z;c) = {x E i d(x,z) < (}.
We are going to show that all equivalence classes of r x A have the local
tesselation property. Observe that if the equivalence class C has the local
tesselation property, then the same is true for the class yC for all y E F. Thus it
suffices to verify that for each z E A, the equivalence class generated by (t,z) has
the local tesselation property. Let us investigate a number of special cases.
The full equivalence class of a point of the form (t,z), z E Into, consists of
that point alone. Any subset of {t} x A is stable in r x, and the open subset
t.IntA of X is mapped by f:X-+H2 onto InI A.
The full equivalence class in r x A generated by a point of the form (t,a),
where a is an interior point of an edge s, consists of two points
(t,a) , (os,b) ; b = vs 1(a)
206 POINCARE'S THEOREM
Let us show that the open subsets of X exhibited in 1.9 form a basis for
the topology on X: with the notation above, an open subset U of X which contains
the point x representing the equivalence class C can be thought of as a - stable
open subset V of r x A containing (rj1,k1),...,(r11,,,k1 ). This implies that V contains
{r1j} x A(k;;e) for i = 1,...,m and e > 0 sufficiently small as required. As a conse-
quence of the local tesselation property, we find that f:X-*H2 maps open subsets of
X to open subsets of H2 . Indeed f is a local homeomorphism.
It is time to show that X is a Hausdorff space: let 10 C be equivalence
classes where C is given by 1.9 and `D is given by
; j E I' , hl E A
Let us consider a fixed pair ij and show that we can find (ij > 0, such that
{t1i} x 0(k,;E) n x A(hj;E) = 0 ; E E }O,Ejj]
If rh # tj any eij > 0 will do. If T1, = j then k, # hj and it suffices to make sure
that A(k1;e) and A(hj;e) are disjoint.
As another consequence of the local tesselation property, we conclude that
the r-translates of t.0 form a locally finite covering of X: let us consider the
point x E X given by the equivalence class C from 1.9. If we let U denote an open
neighbourhood of x in X of the form displayed in 1.9, we have that
o(t.-A) n u = 0 a7 711,...,17m
Since the angle sum of ABC is < 7r we must of course assume that
2.2 ; p,q,r E N
Conversely, for any three integers p,q,r satisfying this condition, we can find
DABC with angles of these measures; this triangle is unique up to congruence,
compare 111.5. The subgroup of Isom(H2) generated by the reflections a,(3,y is
denoted G(p,q,r).
The isometry yQ is a rotation around A of angle 2a/p which makes y/3 a
rotation of order p, similar remarks apply to the vertices B and C. We can now
write down three reflection relations and three cycle relations for our three
generators
c b-1 , a c-1, a1 b
This realises the relations 2.3 as reflection relations and cycle relations.
A picture of the tesselation of the Poincare disc by the fundamental
domain for the triangle group D(2,3,7) is given in the introduction.
210 POINCARE'S THEOREM
Let us work out generators and relations for the even part G+(p,q,r) of the
triangle group. It is quite clear that G+(p,q,r) is generated by
P2A=N7, P2B=rya ,P2C=a13
From 2.3 we derive the following relations for the ps
Let us show that the list 2.4 is a complete set of relations for G+(p,q,r). To this
end we apply Poincare's theorem to the quadrangle ABCD where the point D is
reflection of B in the geodesic through A and C.
Edge cycles: a*c 1, c 1*a, a-, *a 1, c, *c. The side transformations are oa=P2C
and o*c = P2A as follows from
P2C(D) = aa(D) = a(B) = B , P2A(B) = /37(B) =,3(B) = D
Thus Poincare's theorem gives us the following complete set of relations
(P2CP2A)q = P2C' = P2AP =
which is the same as 2.4.
We shall continue to give some applications of Poincare's theorem to
compact polygons. The following notation is useful and will be exploited in VII.6.
pie
=t , P,Pi+l = Pi+1Pi
; i = 1,...,6 (P7 = Pl)
Consider a second copy of the same hexagon and sew the two hexagons together
along the seems a2,a4,a6. The result is a pair of hyperbolic pants. In order to
realise this as an orbit space, consider the rectangular octagon A = U p4(A)
with the side pairing given below (*a = a, *c = c, *e = e, *f = f).
Torus with a hole Let us start with a right angled hexagon with
boundary cycle a1a2a3a,1a5a6 as before. This time we require a2 and a6 to have
the same length; formula III.8.1 implies that a3 and a5 have the same length as
well. Reflect this in the side a1 to get an octagon with the side pairing as
indicated
Example 2.8 Let us fix a pentagon A with angle sum inn and consider the
side pairing with boundary cycle
a*b"1*a1 be
The free edges are supposed to be fixed points under * while the non-free edges are
subject to the following conditions: if the initial, resp. final vertex of the edge s is
finite we require the initial, resp. final vertex of *s to be finite as well ; if both
ends of s are finite, we require s and *s to have the same length.
We shall assume that for each edge s E 9 there is given an isometry as of
H2 subject to the conditions: for a non-free edge s the map as maps the edge *s
onto s in such a way that locally the half-plane determined by *s and A is mapped
to the half-plane bounded by s but opposite A. For a free edge s we require as = t.
Observe that unless both ends of the non-free edge s E S are infinite then
the isometry as is uniquely determined by these requirements. In general the
precise determination of as is part of the structure.
The isometry as is called the side transformation determined by s. It is
required that these data satisfy the
For an edge s of A we let is denote edge with the same initial vertex as s but
different final vertex. This defines for us a second involution on g, s I f s. The
214 POINCARE'S THEOREM
composite of these two operators defines the edge operator 41: given by
Let us introduce the set 1900 of edges with initial point on OH2. Observe that goo
is stable under * and 1 and conclude that goo is stable under the edge operator W.
The cycles sl,.... sr of 'Y on go, are of two sorts: free cycles which contain free
edges and parabolic cycles which do not contain free edges. The cycle man
a= a parabolic cycle is even, as follows from the proof of 1.4. Let us
remark that o is a parabolic transformation if the side transformations generate a
discrete group with fundamental domain .A, compare V.2.4. The cycle map of a
free cycle is a reflection or a as follows from the general
Lemma 3.4 Let sl,...,sr be an edge cycle with side transformations o1,...,0r
and cycle map o = 0102... or. If none of the edges of the cycle are fixed by * then
the vertex cycle P1,...,Pr has no repetitions. Conversely, if the vertex cycle
P1,...,Pr contains a repetition, P1 = P11 say, then the edge cycle s1,...,Sr contains
exactly two edges fixed by *, and the "half cycle maps"
01...0n-1 , 0n...or
are conjugates of the side transformations relative to the two edges fixed by *.
Let us assume for a second that n is odd, say n = 21n+1. Put i = n1 in the formula
above to get that sin = *snl+1 or *sm = Sm+1 = !*s., which is absurd. In
conclusion the number in must be even, n = 2m. From the boxed formula with
n = 2m and i = m we get that sm = *sm. From the same source we get that
sl,...,sm,...,s2m-1 = 511 ...1sii-1'sm'*sin-1,...,*Sl
and we conclude that
VII.3 PARABOLIC CYCLE CONDITION 215
0'1 ...on_1 - 1
Boundary cycle :
a*b"l*al
be
Edge cycle :
a b *al*b-lc b-la l*b *a c-1
Cycle transformation :
aQa la-1y )3a,6-'al'Y
C
Edge cycle : a c *a b
Vertex cycle: ACCA
Cycle map: reflection fi in b
Theorem 3.7 The conclusion of Poincare's theorem 1.7 remains valid for
polygons with vertices at infinity, provided that each parabolic cycle s1,...,sr
generates a parabolic cycle map
Proof Let us return to the proof of 1.7 and observe that it suffices to prove
that X/r is complete. According to VI.7.4 this space is isometric to the space
t.0/g where g is the equivalence relation on t.0 induced by the action of I' on X.
Projection t.0--* decreases distances2 and induces a distance decreasing bijection
t.0/g -A/%
where % denotes the equivalence relation on A generated by the side pairing. On
the other hand, the presence of the continuous section, zHt.z , 0->t.A shows that
the bijection above has a continuous inverse. Thus it suffices to prove that A/96
is complete.
Let us consider a sequence (wn)n E N
of points of A representing a Cauchy
sequence in A/%. We intend to show that this sequence is bounded in H2. Once
this is done, we conclude that our Cauchy sequence is contained in a compact
subset of A/`3U, which makes it convergent in -A/%.
Consider first a free edge AB of A. Let us pick a hypercycle C with ends
A and B, which does not meet OA. The function f f:H2--*R introduced in the proof
of VI.7.7 is distance decreasing and induces a distance decreasing function on
Proof Let us assume that P E 112 is an inner point of the edge s. Observe that
A U arA is a neighbourhood of P and conclude that a y E I'p must equal t or os.
When y = os, we conclude that P is a fixed point for os and we get that *s = s or
*s =s-1. In the first case as is a reflection in s, in the second case os is a half-turn
in P.
Let us assume that P E 0h0. Observe first that an even y E rp is
parabolic as follows from V.2.4. It follows that an odd y E rp must be a reflection
in a geodesic through P since y2 can't be proper hyperbolic. It also follows that
horocycles with centre P are stable under r. When P is an interior point of a free
side s of A we conclude that A contains a full horocycle A with centre P. For
y E rp we conclude that hat A meets y hat A, thus y = t. 11
VILA CONJUGACY CLASSES 219
Theorem 4.2 Let P be a vertex of the polygon A and let sl,.,.,sr be a full
edge cycle with vertex cycle P = P11 ...,Pr. If the vertex cycle is without repetitions
then F is generated by the cycle map a,...a,. If the vertex cycle has a repetition,
P1 = Pn say, then 1'p is generated by the two "half-cycle maps"
Q1...an-i +
an...61.
Proof Let us first consider the case where P is a point of H2. Let us inves-
tigate the following union of translates of I'
3=AU0_1AUa10_2AU ...Ua1a2...ar-1L
From the formula ai(Pi+1) = Pi it follows that each of these translates passes
through the point P1; in fact P1 = °1...oi(Pi+1) Observe that
Q1...ai-10 fl ul...oi(A) D o1...vi-1(si)
as follows from si = vi(*si). Let us consider a small circle A with centre Pt and
observe that 3, = j fl A is a connected with J, fl vJ, # 0. More precise information
is available: the arc 2. subtends an angle 27r/n where n = ord(o ). It follows that
translates of J of the form Qsj, s E Z, cover an open neighbourhood of P. For
'Y E 1'p we conclude that y Int A meets one of these translates ; in fact y must have
the form
y=as or yasQ1...Qn.1 ;n=2,...,r,sEZ
In the second case, we must have PI = u1...an-1(P1), i.e. P1 = P. It follows that
0" is a "half-cycle map".
Let us consider a parabolic cycle sl,...,sr. The main point is to show that
the cycle map Q1...an is non-trivial. Pick a 1lorocycle A at the initial point P1 of
sl and let A1,...,A,. be defined recursively by
Ai = °iAi+1 ,
A,. = O- A 1,...,n
Notice that A = Al since o = al.... ar is either a horolation or L. If repetitions
occur in the sequence P1,...,Pr, the same repetitions occur in the sequence Al,...,Ar
as follows from 3.4: "half cycle maps" are reflections. We can now measure the
the distance from st to a(s1) along the horocycle A and show that this equals the
sum E i=1 r 0i where ©i is the length of the horocyclic arc cut by (si,1si) in A.
The detailed argument is like the proof of 1.4
220 POINCARE'S THEOREM
We can introduce the horocycle arc L = .A n J and observe that I. n o3. 0 0 implies
that .A is covered by translate o-5L, s E Z. We can proceed as above taking the
second half of the proof of 4.2 into account.
Let us now turn to a free cycle. Let us treat the case where *sr = Is1 is a
free edge. The full cycle can be written, compare the proof of 3.4
s1...s11_1sn*sii_1...*sls1- ; r = 2n
with *sn = s, . In the case where the edge sn is non-free we have found that sn is a
reflection and it follows that o = 01...oi_1 is a reflection. Observe that the
horocyclic arc L _ .A n J contains .A n A and .A n o0. Since L is connected we
find that 3, = .A and we can proceed as before. In the case where the edge sn is
free, let us observe that Int 0 and o1...on-1Int 0 are disjoint. It follows that A and
cover the two ends of A and we can conclude that k = A. We can
finally conclude the proof as above. 0
VII.5 SUBGROUPS OF THE MODULAR GROUP 221
5.1
-1/2 1/2
of=p=TQ= 3a7O f
0'r=ay=a)30-y=o-p-10"
Boundary cycle : r*r 1*flf -C
It follows that [G,G] is generated by the third order rotations p and opo and that
there is no relation between these two generators. For further reference let us note
that the parabolic cycle transformations are
(oP 1.Y) P 1 = r-2 , (op 10) 1P = OT2o
It follows from our investigations that G has three subgroups of index 2.
3
7
Side transformations :
r3, 7 _1C 7 , U, 7-?r 1
Cycle relation :
(r1oT) o (ror1)T3=t
T
1cr a TOT-1
The cycle relation, which may be read 7-2p37 2 = t , eliminates the generator T3.
The remaining cycle conditions simply express the fact that r1or, 0-, TUT- I all
have order 2. We may express this by saying that r3 is the free product of three
cyclic groups of order 2.
VII.5 SUBGROUPS OF THE MODULAR GROUP 223
Proposition 5.2 The commutator group [r,r] has index 6 in r and the
group mr,r] is cyclic. The group [F,I'] is a free group on the two generators
?T IOT , QrUT 1
*k
*n
n m
1 R fk
Let us conclude this section with a general formula for the genus of the
compact Riemann surface, IV.5.8, associated with a normal subgroup II of finite
index in the modular group F. The important invariants are the indices [Fx:IIx] of
the stabiliser IIx of the point x, where x = i, (, oo (( = exp(27ri/3)).
In particular we find that the group [r,r] has a branch scheme (2,3,6) and the
genus formula gives g = 1. Observe also that the branch scheme (2,3,6) gives
genus 1 independent of the index The group r2 has a branch scheme (2,1,2)
while the group r3 has a branch scheme (1,3,3).
Let us conclude this section with the introduction of a series of important
normal subgroups of the modular group r. Let us consider a positive integer n
and introduce F(n), modular rou of level n, by the exact sequence
For more information on this topic refer the reader to the books of
[Magnus] and [Newman] and the references given there. In particular [Newman]
p.143: apart from the three exceptions r, r2 and r3, every normal subgroup of r
is free.
226 POINCARE'S THEOREM
6.1
6.2
[A p QB*pW H ApB*pQ-1W
6.3
IAQPB*p'1W ti ApB*p'1QW
6.4
AQpB*pW H ApBQ-1*pW
228 POINCARE'S THEOREM
fI ldk*dk-1) A (albl*al'1*bl I)
fI (ci*ci) (fI ehUh*eh 1) V
k 1=1 i=1 h=1
Proof Let us depart from the boundary cycle of any fundamental polygon and
apply the rules 6.1 through 6.4. Let us first observe that if the boundary cycle
contains p and *p, then we can use 6.2 with B = 0 to bring these symbols together
in such a way that the boundary cycle contains the syllable p*p. This syllable can
be moved freely around as follows from 6.4 and 6.2 with B = 0
Qp*p '-+ PQ-1*P '-+ P*PQ
Let us now assume that the boundary cycle doesn't contain pairs of the form p,*p.
Observe that a syllable of the form a*a 1 can be brought up front in virtue of 6.3
with B = 0
Qa*a 1 . a*a 1Q
Let us suppose that the boundary cycle contains the letters a,*a 1,b,*b-1. The
heart of the matter is the relative position of these four symbols. To get the
general idea we ask the reader to work out the two alternative reduction steps
aXbY*a1 aZYXb*a1*b-1Hab*a1*b"1ZYX
aXbY*b"1Z*a1F-+ aZXbY*b"1*a1 aZX*a"1 by*b"1
Remark 6.6 The standard form of the boundary cycle as given in 1.5 is
not unique. The following reductions can easily be deduced from our four rules
ab*a1*b"1c*c u-+ ab*a1c*b*c H abc*b*a*c u-+
The number 2g + s is unique, being the epos of the surface H2/I'. Let us also
mention that the surface is orientable if s = 0 and unorientable if s > 0. The
number of non-empty words on the list U1,...,Ur,V equals the number of boundary
components of H2/17. These facts follow from Reidemeister's theory of surfaces: a
boundary cycle for the surface H2/1' is obtained by deleting the symbol * in the
boundary cycle 6.5 for A, see for example [Zieschang, Vogt, Coldewey] p.114.
Let us conclude this section with some standard notations, see [Griffiths].
Sphere a*a1b*1)1
Mobius band a *a c
The first presentation of the Klein bottle is the result of gluing two Mobius bands
together along their boundaries. The second presentation alludes to a cylinder
whose boundary circles have been given different orientations and glued together
accordingly.
230 POINCARE'S THEOREM
7.1 x(H2/r)=2-22g
Moreover, we let nl,...,nr, denote the order of the conjugacy classes of elliptic
elements of r. The fact that the number of conjugacy classes is finite follows from
the presence of a compact Dirichlet polygon. The signature of f is
Thus the signature is collection of integers with g > 0 and n1 > 2 , i = 1,...,r. In
order to investigate which data can occur as the signature we consider a Dirichlet
polygon A for r. We shall show that
i=1
7.3 -
Area(O) = 2a[ x(H2/r) + E (1-
Proof Let the polygon A have v vertices. By the definition of area, 111.9.9
Area(s) = 7r (v - 2) - E p L;ncP
where the sum is over all vertices P of A. Let us recall from VII.3.5 that the
vertex cycles of a Fuchsian group are without repetitions. This allows us to
partition the sum above into vertex cycles and use V.3.15 to rewrite the formula
VII.7 FRJCKE-KLEIN SIGNATURES 231
r
2g-2+n.)?42
Let us first remark that this is clear for g > 1. For g = 1 we must have r > 1 and
the result follows since (1 -
j ) > 2. When g = 0 we have r > 3 since the area is
positive. When r > 5, the estimate follows by observing that each term in the sum
is at least 2. When r = 4, the minimum is taken for signature (0;2,2,2,3), but
0-2+2+2+2+1-3-G
It remains to treat the case g = 0 and r = 3. We must prove the inequality
f(k,m,n) = I 42 ;2<k<m<n
for all strictly positive values of f(k,m,n). Observe that f(3,3,4) = i2 and conclude
that we only have to worry about the case k = 2. Observe that
f(2,2,n) < 0 , f(2,4,4) = 0 , f(2,4,5) = 120
and conclude that we only have to worry about k = 2 and m = 3. We have
f(2,3,n) = s -
In the end, the result follows from the evaluation f(2,3,7) _ 2.
232 POINCARE'S THEOREM
Proof The first part of the theorem follows from 7.3. Conversely, for any
data as above, we shall construct a suitable polygon and apply Poincare's theorem.
We fix a point 0 of H2 and consider a circle with centre 0 and radius r, say. Let
us introduce the angle
0- 8g+2i.
21r
and a sequence of vertices Al,...,Ag,B1,.... B1. (B1 = Ag+11 Br+1 = A1) on the circle
subject to the conditions
LA;OAi+1 = 80 , LBiOBi+1 = 20 ; i = 1,...,g, j = 1,...,r
For the remaining specifications we refer to the drawing
VII.7 FRICKE-KLEIN SIGNATURES 233
The sum of the inner angles for the cycle of initial vertices is
r
f(r) = 8ga + 2 /3i
If f(r) = 2r, then Poincare's theorem assures us that the side transformations
generate a Fuchsian group with the given signature. Let vary r through the
interval ]O,oo[ to see that this can be achieved. The angles a and (3j, j = 1,...,r can
be analysed through the "alternative cosine formula" 111.5.2
Theorem 7.6 Let I' be a Fuchsian group with H2/I' a compact surface of
genus g. If n,,...,nr denotes the orders of the conjugacy classes of elliptic
subgroups of r, then I' admits a presentation with generators
0(l,...,a91/31,...,0g161,...,6r
and relations ( [a 6] = a0a-1,0-1)
b1...br11
rf [ai,,3i] = t , bn1 = 1 ,..., bn' = c
i=1
Proof Let us use 6.5 to chooses fundamental domain with boundary cycle
11 (dk*dk 1) fI (ai*bi 1*a; 1 bi)
k=2 i=1
The edge cycles are
g
d2....d,. fl (aibi*a; 1*b;'1) , d2 1,..., dr 1
Let us use Greek letters for the corresponding side transformations to get the
presentation of I' with a complete set of relations
g
[ 13 ])111
112_ nr-
1=1
Upon introducing the auxiliary transformation bl given by
g
bl = 62...b,, n [-i,/3 ]
i=1
we can bring the presentation of 1' into the announced form 0
Analogous results can be derived by the same method for the general
group I' with H2/1' compact, see [Wilkie] and [Macbeath]. For a different
approach see [Zieschang, Vogt, Caldeway].
VII EXERCISES 235
VII EXERCISES
EXERCISE 2.1 Let p and q be ordinary integers which satisfy the relation
(p-2)(q-2) > 4
1° Show that there exists a regular p-gon 0 with angles 2ir/q.
2° Show that there exists a tesselation `JI, q of H2 consisting of regular p-gons
with angles of measure 2ir/q, such that any two of the p-gons have empty
intersection or have an edge in common. Hint: Consider the triangle group
G(2,p,q) and its canonical tesselation.
3° Let p denote a rotation of order p about the centre of 0 and let o denote a
rotation of order q with respect to a vertex of A. Show, that the group of even
symmetries of the tesselation Tp q is generated by p and o, and that
(PT)2
Pp = oq = = t
EXERCISE 2.2 Let 0 denote a regular p-gon with angles of measure 2a/q with
the notation of the previous exercise.
1° Show that the group D(0) generated by reflections in the sides of 0 is
discrete if and only if q is an even integer.
2° Suppose q is even. Show that the group D(0) is a normal subgroup of the
symmetry group of `fp q, and identify D(0) with a normal subgroup of G(p,q,2) of
index 2p.
RP=Sq =(RS)2=c
is infinite. Hint: When (p - 2)(q - 2) > 4 use hyperbolic geometry, but when
(p - 2)(q - 2) use Euclidean geometry.
EXERCISE 2.4 Let A be an r-gon in the hyperbolic plane whose vertices has
angles of measures
LPi = 7r/n; ; n, E N, n, > 2, i=1,...,r
10 Show that reflections in the sides of A generate a discrete group D(0)
with A as fundamental domain.
2° Show that "Dyck's group" D+(0) can be generated by pl,...,pr subject to
nl = n1,
pl ... =P =P1P2...p1.=c
EXERCISE 2.5 Let n > 2 be an integer and let A be a regular 2n-gon with
angles of measure a/n. Let us consider a side pairing on A with boundary cycle
al*al a2 *a2 ... an *an
For i = 1,...,n we let Mi resp. *M1 denote the midpoint of the edge a; resp. *Mi.
Show that the side transformation ai is glide transformation along M;*M4.
2° Show that al,...,an generate a discrete group G with fundamental domain
A and that a complete set of relations is
anon-1
2 2
... al2 =t
3° Identify the group G with a subgroup of index 4n in the triangle group
G(2,2n,2n).
EXERCISE 2.6 Let g > 2 be an integer and let A be a regular 4g-gon with
angles of measure 7r/2g. Let us consider the side pairing on .A which identifies
each side with the opposite side of the polygon by means of parallel translation
along the midpoints of the two sides. Let us enumerate the sides of A such that
the boundary cycle reads
al bl a2 b2 ... ag bg *al-1*bl... *ag 1*bg 1
1° Show that a discrete group G with fundamental
domain A and that a complete set of relations is
VII EXERCISES 237
1 _1 _1 _1
l(3» = Q» a» ... 01-1-1
2p2 ... an/3n_1:_
x101 a2)32-
2° Show that the Riemann surface H2/G has genus g.
3° Identify the group G with a subgroup of index 8g in the triangle group
G(2,4g,4g).
EXERCISE 2.7 Let 0 be the regular 8-gon with angles of measure it/4.
Consider the side pairing with boundary cycle
a b *a 1*c 1*d-1 c *b"1d
Show that the corresponding Riemann surface has genus 2 and a fundamental
group generated by a, 13, y, 6 subject to the relation
6 -Y60a-' 0-1 y1 a=t
EXERCISE 5.1 Show that the commutator subgroup [r,r] of the modular
group r has the fundamental domain A U r/, U r20 U r30 U r40 U r50 with side
transformations r6, r3o, r4or 1, r5or 2. Work out the edge cycles and apply
Poincare's theorem.
EXERCISE 5.2 1° Show that the subgroup r2 of index 2 in the modular group
1' has a branch scheme (2,1,2), and that this is the only normal subgroup of I'
with a branch scheme of the form (2,1,n).
2° Show that the normal subgroup r3 of r of index 3 has a branch scheme
(1,3,3) and that this is the only normal subgroup of finite index with a branch
scheme of the form (1,3,n).
30 Show that no normal subgroup of finite index of r has a branch scheme of
the form (1,1,n), except for 1' itself.
EXERCISE 5.3 Let i denote the hyperbolic triangle oo,0, l with side trans-
formations o, p-1 ap, pop 1 where p = ro. Show that Poincare's theorem applies
and deduce that r3 is the free product of the three cyclic subgroups generated by
0',p 1l7P, P7p1
238 POINCARE'S THEOREM
EXERCISE 5.4 I 'Show that 1.3 is the smallest normal subgroup of r containing p.
2° Show that the subgroup I'2 of I' of index 2 is the smallest normal subgroup
of IF containing o.
3° Show that 1'2 and 1'3 are the only non-trivial normal subgroups of r which
contains elements of finite orders.
EXERCISE 5.5 Let 11 denote a normal subgroup of finite index p of I' with a
branch scheme of the the form (2,3,n).
10 Show that the corresponding compact Riemann surface has genus 0 for the
following values of (µ,n) only
(6,2) (12,3) (24,4) (60,5)
2° Show that the table above is realised by 1'(2), I'(3), r(4), 1'(5)
genus 0 0 0 0 0 1 3 5 10 13 26 25
VIII HYPERBOLIC 3-SPACE
Observe that the operator La on A E has square 0. It follows that we can extend
the operator L to all of A E making A E into a r ht module over the ring A E
where we have used expansion of the determinant after the first row. 0
1.8 *or= 1
1.9 ** V = (-l)P(I1-P)+sv
; vE APE
Let us remark that the operator * changes sign if the orientation is changed; the
orientation must be changed (n - p)p times to make eP+j.... jen,el,...ep positively
(-1)n(n-p)
It
(-1)s* eP+1 A ... hen = fi <ei,ei> el h ... A ep
i=P+1
n
Apply 1.7 once more and observe that r[ <ei,ei> = (-1)5. 11
i=1
The reader is referred to [Bourbaki] for the exterior algebra. Note that we
have used the sign conventions which makes A E into a right module over itself, as
opposed to later editions of Bourbaki.
VIII.2 GRASSMANN RELATIONS 245
where H runs through all strictly increasing maps in the range indicated above and
eH = eH(1) A ... A eH(p). The relation vLe = 0 can be written
0= H aH <eH(1),e> eH(2) A ... A eH(p)
which gives aH = 0 whenever 11(1) = 1. Thus Ev C e 1 as required.
246 HYPERBOLIC 3-SPACE
Proof From 1.3 it follows that for u E A rE and v E ASE we have the formula
Let us consider a pure vector w = el A ... A ep and let E, denote the space spanned
by el,...,ep. Pick a basis f1,...,fn for E with fp+l,...,fn E Eµ, 1 and apply 2.5 to get
(f1 A ... A fn)Le1 A... A ep = <f1 A ... A fp,el A ... A ep>fp+l A ... A fn
Observe that we can arrange for or = f1 A ... A fn and the result follows.
Theorem 2.6 A non-zero z E APE is pure if and only if the composite map
AP-1E zL E _zA Ap+1E
is zero. When z is pure, the sequence is exact.
to see that z A e = 0 if and only if e c Ker(*zL). We conclude from 2.2 that the
image of the first map is Mz while the kernel of the second map is M*Z 1 . Thus
(z A ) o (zL) = 0 a Ez is orthogonal to E*Z
If z is pure, we have seen in 2.3 that EZ is orthogonal to E*z. Conversely, if EZ is
orthogonal to E*,L we conclude from 1.1.7 that
dim Ez + dim E*Z < n
Since in general dim EZ > p and dim. E*z > n - p we get dim EZ = p. This means
that z is a pure p-vector.
VIII.2 GRASSMANN RELATIONS 247
Let us now specify our theory to a Minkowski space M. Our main concern
will be the classification of two-dimensional subspaces of M by 2-vectors.
M
awl M -+*wl M
For a pure 2-vector w the plane Mw = Ker(w A: M-> A 3M) has Sylvester type
(-1,1), (-1,0),(-2,0) depending on whether <w,w> is negative, zero or positive.
Proof The composite of the two maps we have just written is zero if and only
if w is pure as follows from 2.6 and 2.7. The first statement will follow once we
have established the formula
We shall be content with a particular case, namely, with the notation of 1.16
w_(x+iy)Q1Av2=xo'1Aa2+Ya3AQo, *w=iw= -yo1Aa2+xc3Aco
observe that Im[<w,w>,] = Im[(x + iy)2] = 2xy. Finally check through the
formula as e varies through ao,al,a2,a3. An alternative proof can be given using
the technique from VIII.7.
Let us investigate the general subspace E of M of dimension 2. Pick an
orthonormal basis e,f for E and calculate the norm of the 2-vector e A f
<e A f,e A f> = <e,e><f,f> - <e,f>2
The discriminant lemma 1.6.3 shows that the norm is negative, zero, positive
depending on the Sylvester type as stated. 0
Let us say that two geodesics h and k are perpendicular if they meet in a
point A and their tangent vectors at A are orthogonal.
Proof If <H,K>c = 0 then h and k are coplanar by 3.2. Observe also that if h
and k are perpendicular then they are contained in a hyperbolic plane. Thus the
result follows from 111.2.
Proof The group S03(C) = SO( A 2M) acts transitively on orthonormal pairs
H,K in A 2M, while the group Lor+(M) acts transitively on pairs h,k of oriented
perpendicular geodesics.
250 HYPERBOLIC SPACE
Let OH3 denote the set of isotropic lines in M (on ints at infinity). We say
that P E OH3 is an end for a geodesic h in 113 if the line S is contained in the linear
span of h. It is clear that a geodesic h has two ends, and conversely, given two
distinct points A and B of 8H3, there exists a unique geodesic in H3 with ends A
and B. Observe also that two geodesics h and k with a common end are coplanar.
Theorem 3.5 Let h and k be two geodesics in H3. There exists a geode-
sic g perpendicular to h and k if and only if h and k don't have a common end.
Proof Let H and K be normal vectors for h and k respectively. The vectors H
and K are linearly independent over C: from H = AK, A E C, we get, taking
norms, that A2 = 1 and therefore H = K or H = -K, which means that H and K
represent the same unoriented geodesic. Let us introduce the complex vector space
E C A 2M spanned by H and K and let us form the discriminant
A = <H,H>c<K,K>c - <H,K>c2 = 1 - <H,K>c2
The complex line E 1 is isotropic if and only if A = 0.
If A # 0, we can find a 2-vector G with norm -1 spanning E 1 . The geo-
desic g with normal vector G is perpendicular to both h and k as follows from 3.3.
Conversely, if the geodesic g is perpendicular to h and k, then the normal vector G
for g spans the complex line E 1 and it follows that A # 0
If A = 0, then we have <H,K>,, = ± 1, and we conclude that h and k
coplanar. Thus the result follows from the discussion in 111.2.
VIII.4 PENCILS OF PLANES 251
Proof With the notation of 4.1, pick a positively oriented orthonormal basis
el,e2,e3 with norms -1,-1,+1. This gives
<S,S> = det1j<e1,ei> = 1
The main body of the proposition follows from VIII.2. 0
Proof From the assumption that P $ Q it follows that *S and *T are linearly
independent which implies that *S A *T # 0. Observe that *K A *S=0 to conclude
that *S belongs to M.,<, compare 2.8. . Since M*I< = MI< 1 by 2.3, we conclude
that *S is orthogonal to MK or MK C (*S) 1 . Similarly, we find that
MI< C (*T) 1 . For dimension reasons we must have that MK = (*S) 1 n (*T) 1 .lJ
Proof Let us consider a pencil 9 of type (0,1). The condition for the
hyperbolic plane P with normal vector S to pass through the point A is
<*S,A> = 0. Thus we are lead to consider the intersection A 1 n *9. The plane
*9 in M has type (- 1,0) while A 1 has type ( - 3,0). It follows that the
intersection is one-dimensional, generated by a vector S of norm - I. We leave it
to the reader to treat the remaining two types of pencil.
VIII.5 BUNDLES OF GEODESICS 255
Proof Let us pick two distinct geodesics h and k from the pencil and let H and
K be normal vectors for these. Observe that the formula for polarisation 1.1.2
shows that <H,K> E R since 11, K and H + K all have real norms; it follows from
3.2 that h and k are contained in the same linear hyperplane N of M of type
(-2,1). It follows that 9 C A 2N.
5.3 0 -a A °M -* A 1 M A 2M -> A 3M - A 4M _ 0
256 HYPERBOLIC 3-SPACE
From this it follows that .A = Im( A A: A 1M--> A 2M). This description reveals
that A has dimension 3 and consists of 2-vectors with real norm, 1.18. From
<A A X,A A Y> = <X,Y> ; X,Y E TA(H)
it follows that A is isometric to the tangent space TA(H3), in particular the
Sylvester type is (-3,0). An oriented geodesic curve k has normal vector K E A if
and only if k passes through A, compare 2.8. In resume, we have realised the set
of geodesics through A as the geodesics in the bundle A.
Let us summarise our findings in table 5.4 below. The heading "Pencils"
indicates the possible Sylvester types of pencil in the bundle.
Let us notice that the Hodge * transforms spaces of type (-3,0) or (-1,0) into
spaces of type (0,3) or (0,1); however, such spaces will not contain 2-vectors of
norm -1. We have, in fact, exhausted the possible Sylvester types:
Proof We must rule out the four types (0,2), (-2,0), (-1,1) and (0,0). In
the first three cases it is easily checked that an orthonormal real basis forms a
complex basis as well. These three types can now be ruled out on the grounds that
the complex inner product on A 2M is non-singular.
258 HYPERBOLIC 3-SPACE
In order to rule out the fourth case, observe that a hypothetical three-
dimensional subspace consisting of 2-vectors of norm 0 would span a complex
subspace E isotropic for the complex inner product, i.e E C E 1 . Since the inner
product on A M2 is non-singular we get that dimCE = 1, 0 , which is a
contradiction.
Proof By theorem 3.4 it suffices to prove that SO3 =SO( A 2M) acts
transitively on the bundles of given type.
Let us treat type (-2,1) separately. A bundle is specified by a sequence
el,e2,e3 of orthogonal 2-vectors of norms -1,-1, 1. It is clear that 03 = O( A 2M)
acts transitively on such sequences. An orthogonal transformation has del = ± 1,
but the sign can be changed by permuting the first two vectors leaving the third
invariant. Types (-3,0) and (-1,2) can be treated in a similar fashion.
Let us consider a bundle E of type (-1,0) and let us prove that E fl iE # 0.
Pick an orthogonal basis el,e2,e3 with norms -1, 0, 0 to see that E n iE = 0 is
equivalent to E + iE = M, which implies that el,e2,e3 are linearly independent over
C; in particular we find that (Ce2 + Ce3) 1 = Ce2 + Ce3 contradicting that our
complex form is non-singular. Let us show that E fl iE C E 1 : starting with
<e,z>_-i<ie,z> ;zEE,e E EniE
and the observation ie E E we conclude that <e,z> = 0 as required. This shows
that we can construct a real basis for E as follows: pick any vector e E E with
norm -1 and a non-zero vector f E E fl iE to get the basis e, f, if. This shows
how to recover E from a pair of orthogonal vectors e,f of norms 1, 0. It follows
from Witt's theorem 1.2.4 that 03 acts transitively on such data. To see that S03
acts transitively, let us consider an orthonormal basis el, e2, e3 and put e = el,
f = e2 + ie3 ; observe that we can permute e2 and e3 keeping el fixed.
VIH.5 BUNDLES OF GEODESICS 259
Let us emphasise that the fact that geometrically different pencils have dif-
ferent Sylvester types amplifies statements like "through a given point there passes
at most one geodesic perpendicular to a given geodesic". As an illustration let us
apply this to the theorem of Whitehead, Smid and Van der Waerden, see
[Coxeterl) § 4.
Proof Let the four geodesics be a,b,c,d. In the graph below we have
represented each of the five hyperbolic planes by a bar. The problem is to show
that the geodesics b and d are coplanar.
F a-
b d
Lemma 5.8 Let `.B and `.b be bundles of geodesics both of type # (-2,1). If
`B and 9 have a common pencil, then `B = `D.
Proof Inspection of 5.4 reveals that `B and `D must be of the same type in
order to have a common pencil. We ask the reader to go through the three cases.
260 HYPERBOLIC 3-SPACE
d -b I
6.1 ; a,b,c,d E C
-c a
(AB)"=B-A"
6.3 ;A,BEM2(C)
; a,dER,bEC
From this description it follows that the inner product takes real values on M. In
particular, the determinant defines a real quadratic form on M. An orthonormal
basis for M is formed by the
0 1
0.1
1 0 '
0'2
i 0
'73 = [ 0 -1
]
These transformations are orthogonal and we conclude from the fact that S12(C) is
connected that a transformation of the form 6.8 preserves orientation and the
connected components of the unit hyperbola in M. Thus we deduce a morphisml
S12(C) - Lor+(M)
which we shall demonstrate to be surjective, see 8.5. It is this fine action of S12(C)
that makes the Hermitian model of H3 extremely useful.
'A different proof of this will be presented during the proof of 8.5. As it
happens, it follows from the same source that S12(C) is connected. See also 8.9.
262 IIYPERBOLIC 3-SPACE
We ask the reader to verify that the Dirac algebra D is associative with multipli-
cative unit (i,0). Let us introduce the element %P = (O,t) and let us identify a
matrix A E M2(C) with the element (A,0) of the Dirac algebra. Then we can write
(A,B) = A + BAY ; A,B E M2(C)
The multiplication is ruled by the formulas
We ask the reader to check that indeed (XY)' = YX , X,Y E D. Our next
The Minkowski space M is oriented by means of the Pauli matrices 6.7. This has
the effect of making A 2M into a complex vector space with a complex quadratic
form, 1.15. Let us introduce the D-valued alternating bilinear form on M
Observe that this symbol is invariant under the principal autornorphism but anti-
invariant under the principal anti-automorphism. As a consequence we conclude
that the symbol takes values in the following subspace of D+
s12(C) = {X E M2(C) I Ir X = 0 }
The space s12(C) has a natural complex structure and the determinant defines a
complex quadratic form on s12(C).
Let us prove that the map R A S f-+ R x S is C-linear. According to the formulas
1.16 this is equivalent to the formulas
i(o1XO2)=0.3Xo0, i(a2X0-3)=al xa0' t(o3xol)=0,2X00
These can be verified by means of 7.7 and 7.10. Let us be content with the first
00 471*aAo3" (0,1' 2W) = - 00"P02P03' A = 03*00%p
It remains to verify that of X o2 , a2 X 03, 0-3 x a1 form an orthogonal basis for
s12(C). Let us,/ verify that the first two vectors are orthogonal:
-2t'1(a1 X a2)(a2 X 03)) 21T(o1%Yo3'Y) = 0
The remaining details are left to the reader.
Let us end this section with some information for readers familiar with the
general formalism of Clifford algebras [Deheuvels].
For the benefit of the reader we should mention that the Dirac matrices
refer to the following eight-dimensional real representation of Dirac's algebra
A B
A+BW ; D -' M4(C)
B A
It is known that D M2(H), compare execise 7.1. It follows that the eight-
dimensional representation above is simple.
VIII.8 CLIFFORD GROUP 265
8.1 G# MW G-1 = MW
Let us identify G12(C) with the group of multiplicative units in D+. For a
matrix R E G12(C) with real determinant we have i0 = R*(det R)'1 , thus
RXIR'1 = RXR-1 F
= RXR*,P(det R)-1 ;XEM
which shows that R E r+, where I'+ = D+ fl r. We ask the reader to show that
r+={REG12(C) I MR ER)
8.4
R.X=RXR''=RXR*(detR)-' ;REr+,XEM
it
a b 1 0
][ a c ][ 1 0
] =as+bb+cc+dd
c d
Let us also remark that R E r+ acts trivially on M if and only if R is a real scalar
matrix: this follows from the proof of 7.10. Consider RE r+ with del R > 0 and
decompose the action of R on M into a product of reflections T1T2...Td along
vectors Sl,...,Sd of norm -1. It follows that we can find k E R* such that
R = k SOP S2I'...Sd
Let us make use of the principal anti-involution 7.4 to get that
k2(-1)d
R R- = k2 S1S1- S2S2"... SdSd. =
Using that delR = RR' is positive we conclude that d is even.
An even Lorentz transformation of Al can be written TATBTCTD, where
A,B,C,D E M, each of determinant -1. According to 7.5 this is the action of
ABCD
This allows us to conclude that S12(C)-Lor+(Af) is surjective. 0
Vffl.8 CLIFFORD GROUP 267
It follows that (iH)(iK) E S12(C) acts as a/3 on M which gives S = f iH iK. Let
us recall that K" = -K, since K E s12(C). This gives us
lr(ill iK) = - tr(HK) = tr(HK-) = 2<H,K>
The required formula follows by squaring this relation. Conclusion by 8.9. 0
Lemma 8.9 Any S E S12(C) can be written S = HK where H,K E s12(C) with
detH = de1K = -1.
Proof We ask the reader to use the theory of quadratic forms to find vectors
H,K E s12(C) with <11,H> = <K,K> = -1 and 2<H,K> = tr(S). Observe that
1r2S = 1r2HK and conclude from 1.9.15 that S and IIK are conjugated in PSI2(C).
Upon changing the sign of H we may assume that S and HK are conjugated in
S12(C). 0
268 HYPERBOLIC 3-SPACE
In this section we shall work with the Hermitian model for H3 and identify
its boundary OH3 with t keeping track of the action of S12(C). Once this is done
we write down the normal vector L(A,B) for the oriented geodesic with ends
A,B E C and evaluate <L(A,B),L(P,Q)> in terms of the cross ratio [A,B,P,Q].
To begin let us assign to e E C2 the following Hermitian matrix
T
:]=[:]
Zz
9.1 L(e) _
w
;e=[]
Notice that L(Ae) = ).L(e) , A E C, and observe the pleasing transformation rule
Proof The equivariant part of the statement is formula 9.2. The verification
of bijectivity is straightforward. Let me take the opportunity to point out that the
construction performed here is a special case of the general treatment in 1.7. It
suffices to identify C ® R2 with M through the assignment
; zEC,x,yER
;bEC,a,dEl,ad-bb=-1
=0
9.5 L( ; D = det I zu
wv J
zv-}-wu -2zu
9.6 L ( P ,Q) = zv lwu
2wv -zv-wu
;P=[z1,Q=[°I
This matrix is easily seen to have determinant -1. Let us recall from 8.11 that
270 HYPERBOLIC 3-SPACE
We shall evaluate the complex inner products of 2-vectors of the type 9.5 in terms
of the cross ratio on C. Recall from 1.9.8 that for four distinct points A,B,P,Q of
C the cross ratio [A,B,C,D] $ 0, 1, oo.
1
9.8 <L(A,B),L(P,Q)> =
[A,B,P,Q]+1
Proof According to 9.7 and 1.9.7 both sides of the formula are invariant under
S12(C). From the fact, 1.9.3, that the action of S12(C) on C is triply transitive it
follows that it suffices to verify the formula in the case where A = oo, B = 0,
P = 1, Q = q with q E C, q 0, 1. Recall from 1.9.9 that [oo,0,1,Q] = q. We find
that
+q -q
2
L(00,0) L(1,Q) = 1_' q
0 -1 2 -1-q
Corollary 9.9 Let A,B,P,Q be four distinct points of the Riemann sphere C.
The geodesics h and k in h3 with ends A,B and P,Q are coplanar if and only if
[A,B,P,Q] E R. The two geodesics h and k will intersect if and only if [A,B,P,Q] is
negative; h and k are perpendicular if and only if [A,B,P,Q] = - 1.
VIII.10 QUATERNIONS
2
10.1 cosh p(R,S) = 1 + IR2rs I R = (z,r), S = (w,s)
The group S12(C) acts as isometries on this model: if we agree to view the point R
above as a quaternion R = x + iy +j r, then the action is given by the formula
z + r2 z
10.3 F(R) = ; R=z+jr
z 1
2
trF(R)°F(S)=1+JS,R1
rs ;R=z+jr,S=w+js
which shows that F is an isometry as required, compare 11.4.1, 7.5. and 10.1. The
relevant formula for S12(C)-equivariance
In this section we shall apply geometry to prove some fixed point theorems
for subgroups of PS12(C). Let us first recall that or E PS12(C), o # t, has one or
two fixed points on OH3. In the case of two fixed points, the geodesic through
these is stable under o and is called the axis of o. We shall analyse o in terms of
theorem 8.7 on half-turns. Details of some of the arguments can be found in the
exercises.
Proof Let us consider two distinct non-trivial members o and r of G and show
that their axes h and k intersect. To this end we shall rule out two alternatives:
common end and common perpendicular.
When h and k have a common end, the commutator oro 1r 1 is parabolic
and we conclude that o and r commute. The fixed point set h for c on H3 is
stable under r as follows from the "normaliser principle" IV.2.6. It follows that
h=k or ( when r is a half-turn) It is perpendicular to k.
When It and k have a common perpendicular, let A E It and B E k be
points such that the geodesic I through A and B is perpendicular to h and k. Pick
geodesics a through A and b through B such that a = a\ and r = A,3 where a,3,\
are half-turns in the geodesics a,b,l. Observe that or = a# and conclude from 8.7
that a and b intersect. This implies that the plane P spanned by a and b is perpen-
274 HYPERBOLIC 3-SPACE
dicular to the axes h and k for a and r. We conclude from the proof of IV.2.8 that
A = B and therefore h = k.
Since any two axes of rotation are coplanar, we conclude that the axes of
rotation are members of some bundle X. Since any two axes intersect, we
conclude from 5.4 that all axes pass through the same point of H3. 0
Proof Suppose that G has a finite orbit on H3 of order n > 1. Any non-
trivial y E G acts as a permutation on the set 0 on n elements and we find that
y"! has a fixed point on H3. From this it follows that y" is a elliptic (or t) and
we conclude that y is a elliptic: the class of loxodromic (resp. parabolic elements)
is stable under n!-powers. It follows from 11.1 that G has a fixed point on H3.
Suppose that G has an orbit on OOH3 consisting of n > 1 elements. In case
n = 1, the group has a fixed point on 8H3. When n = 2, the geodesic through the
two point orbit is stable. When n > 3, we find that all elements of G have finite
order and the elliptic fixed point theorem tells us that G has a fixed point on H3.
Proof Let us assume that G does not contain a loxodromic element. According
to the elliptic fixed point theorem, the group G contains a parabolic element y,
fixing oo say. For any element a E G we have that
1r2(y"c)E[0,4] ;nEN
VIII.11 FIXED POINT THEOREMS 275
1 1 a b
; a,b,c,d E C, ad-bc = 1
0 1 c d
Proof Suppose first that G doesn't contain strictly loxodromic elements. Let
us show that any two hyperbolic transformations u,v E G have coplanar axes. To
this end we may assume that the axes m and it have a common perpendicular c.
Let us write p = ay and v = y,3 where y is a half-turn with respect to c, a is a
half-turn with respect to a geodesic a perpendicular to m and a is a half-turn with
respect to a geodesic b perpendicular to it. From the fact that none of the
transformations ay, 'y,(3 , a,6 are strictly loxodromic it follows that a,b,c are
pairwise coplanar, compare 8.7. It follows from 5.4 that we can find a plane P
such that either a,b,c are perpendicular to P or all lie in the same plane P. In
both cases we conclude that m and n lie in P.
Let us remark that G contains a hyperbolic transformation p with axis m
say, 11.4. For o E G we find that opo l is hyperbolic with axis o(m). It follows
that we can find another hyperbolic transformation v E G with axis n # m. At
this point we ask the reader to use 5.4 to find a plane R containing the axes of all
hyperbolic elements of G. Observe that R is stable under G.
It remains to prove that G preserves the orientation of R. To this end we
must prove that the group I' C Isom(R) induced by G consists of even transfor-
mations only. Assume for a moment that this is not the case. Taking into
276 HYPERBOLIC 3-SPACE
account that r is non-elementary, it follows from IV.2.9 that we can find a glide
reflection y E r with axes g, say. Let us write y = rRKR where r is restriction to
R of a translation r in H3 along g and KR is restriction to R of reflection 1s in the
hyperbolic plane K perpendicular to R through g. Let p denote reflection in R and
conclude that rtzp E 1'. This is a contradiction since rKp is strictly loxodromic.
Conversely, let R be an oriented plane fixed by G. An element y E G
induces a transformation of the plane R which is even. It follows that y is the
product of two reflections in planes perpendicular to R. 0
Theorem 11.7 The group PS12(C) is simple, i.e. the group contains no
non-trivial normal subgroup.
VIII EXERCISES
EXERCISE 4.1 Let P and Q be distinct hyperbolic planes in H3. Show that
there exists at most one geodesic h perpendicular to both P and Q. Hint: Use
proposition 4.6.
EXERCISE 6.3 For matrices A,B E M2(C) put A x B = BA'). With the
2(AB
notation of exercise 6.2. show that for all A,B,C,D E M2(C).
<A x B,C x D> = <A,C><B,D> - <A,D><B,C> - i vol(A,B,C,D)
Hint: First observe that the formula is valid for A,B,C,D E M. Second, observe
that the formula is C-linear in all variables. Third, observe that the complex
vector space M2(C) is spanned by M.
; a,dER,bEC
10
Show that the determinant makes this into a Minkowski space. Hint: Use
the transformation T.A-M given by
b iI
-a I
can be decomposed into two simple representations. Hint: Show that vectors of
the form (z1,z2,z1,z2), z1,z2 E C make up a D-stable subspace of C4.
6° Conclude that the Clifford algebra of a form of type (-1,3) is M4(R).
EXERCISE 8.1 1° Show that the only involutions in Lor+(M) are the half-
turns. Hint: See exercise I.1.1.
2° Find all involutions in the group Lor(M).
VIII EXERCISES 281
EXERCISE 8.3 For U E r show that UU' E ode and UU#' E Ei. Verify that the
spinor norms N:r-R and S:r-*R defined by
N(U)i = UU' , S(U)t = UU#'c ;UEF
are morphisms of groups.
20 Explain how the spinor norm N was used in the proof of Theorem 8.5.
3° Show that U E r acts on M as a Lorentz transformation if and only if the
spinor norm is S(U) is positive.
EXERCISE 8.4 Let r denote an oriented geodesic with normal vector R. Show
that a half-turn p with respect to r acts on M through the formula
p = RL RL + *RL *RL
Hint: Consider a positively oriented orthonormal basis A,T,U,V for M where A E r
and T is a positively oriented unit tangent vector to r at A.
EXERCISE 9.3 Consider two non-intersecting discs in the complex plane and let
the Euclidean line through their centres intersect the bounding circles in the points
A,B,C,D in this order.
10 Show that the hyperbolic distance distance d between the corresponding
hyperbolic planes is given by
tanh.2 2cl =AAD BC
C BD
20 In the case where the two circles are concentric, show that d is the
logarithm of the ratio of their radii.
eZ 0
1
2
tr 0 ez = cosh x cos y+ i sinh x sin. y
components of H3-P.
EXERCISE 11.3 Let 9 denote a pencil of hyperbolic planes. Show that the
product of three reflections in planes from 9 is itself a reflection in a plane from 9.
Hint: Show that there exists a hyperbolic plane Q whose normal vector is
orthogonal to the normal vector of any plane from 9.
VIII EXERCISES 283
The aim of this appendix is to prove that a metric space which satisfies
the incidence axiom and the reflection axiom is isometric to the Euclidean plane or
(after rescaling) the hyperbolic plane. For a precise statement see the first page of
the introduction. The need for a proof comes from the fact that our axioms are
somewhat untraditional. The proof is presented as a systematic series of
deductions from the axioms. This will be carried out to the point where we have
proved the "Pasch's axiom", see the discussion at the end of the appendix.
Let us agree that a straight geodesic curve in a metric space X is a
distance preserving map y: R-X and recall that a line in X is the image of a
straight geodesic curve, compare 11.1. In the following we let X be a plane i.e. a
metric space X which satisfies the following two axioms
Sharp triangle inequality A.1 For a point B of the plane not on the
line through two distinct points A and C we have that
d(A,C) < d(A,B) + d(B,C)
Proof We shall turn the problem upside down and assume that we are given
three distinct points A,B,C with d(A,C) = d(A,B) + d(B,C) and proceed to
construct a line through all three points. Pick a,b,c E R with b - a = d(A,B) and
c - b = d(B,C) and choose three straight geodesic curves 0, A, It in X with
0(a) = A , q(c) = C, A(a) = A, .\(b) = B, µ(b) = B, p(c) = C
Let o:R-X denote the curve with restriction 0 to ]-oo,a], restriction A to [a,b],
restriction it to [b,c] and restriction 0 to [c,+oo[. We shall prove that o:IR-+X is a
APPENDIX: AXIOMS FOR PLANE GEOMETRY 285
We proceed using the reflection axiom. For convenience we shall use the word
sides for the connected components of the complement of a line k.
Proposition A.2 Given a point P in the plane outside the line k. There
exists a unique point Pk E k such that
d(P,Pk) < d(P,K) ; K E k, K o Pk
Proof Let tc be an isometry of the plane which fixes k pointwise but inter-
changes the sides of k. For a P 0 k, the points K(P) and P lie on different sides of
k. It follows that the line n through P and h(P) will meet k in a point Pk say.
11
K Pk k
1c(P)
Notice, that the point Pk is fixed by K. It follows that the line n is stable under k
since the points P and Pk are mapped into n. By examination of the restriction of
1c to n we conclude that t4tc(P)) = P and that Pk is the midpoint of P and tc(P).
Let us now investigate the distance from P to an arbitrary point K E k,
286 APPENDIX: AXIOMS FOR PLANE GEOMETRY
The isometry c whose existence is postulated in the reflection axiom is unique and
is called reflection in the line k.
A0 Al A2 A3 A2n-1 A2n
P2 P2n-1 P2n
Proposition A.7 For any point K on the line k there exists a point P off k
with Pk = K.
Proof Let us fix one of the sides S of k and observe that perpendicular
projection p:S-*k is continuous as a consequence of Saccheri's inequality. Using
that S is connected, we conclude the image p(S) is a connected subset of k. It
remains to prove that p(S) is dense in k, i.e. for K E k and e > 0 given, there exists
Q E S with d(K,Qk) < c. Observe that for K E k and Q E S we get from A.5 that
k is perpendicular to the line n. through Q, Qk. It follows that d(K,Qk) < d(K,Q).
Finally, observe that for K E k and e > 0 given there exists a point Q E S with
d(Q,K) < e : such a point can be found on any line through K which meets S.
288 APPENDIX: AXIOMS FOR PLANE GEOMETRY
Proof Let us use A.7 to pick two perpendicular lines in and it through S and
let p and v be reflections in in and it respectively. We intend to show that pv is a
half-turn with respect to S. It is easy to check that the isometry pvpv is the
identity on in and on it. We conclude from A.4 that pvpv = t, the identity. It
follows that o = pv is an involution with fixed point S. We proceed to show that
S is the only fixed point for o. It is easily checked that S is the only fixed point
for o on m U n. Let us turn to the complement of in U it. Reflection v in n leaves
the two sides M1 and M2 of in invariant: from it 1 in it follows that M1 and M2
both meet it, and we conclude that v has a fixed point on both of them. In the
same way we conclude that p leaves the two sides N1 and N2 of n invariant. The
complement of in U it is decomposed into the union of four disjoint "quadrants"
M1 n N1, Min N2, M2 n N1, M2 n N2
Observe that o interchanges the first quadrant with the third and the second
quadrant with the fourth. It follows that S is the only fixed point for o.
From the fact that o is an involution with S as the only fixed point it
follows that o is a half-turn with respect to S: For a given point P # S the two
points P and o(P) are interchanged by or. It follows that the line h through P and
a(P) is stable under a. Second, we deduce that the midpoint of P and o(P) is a
fixed point for o. In conclusion, the point S is the midpoint for P and o(P). D
APPENDIX: AXIOMS FOR PLANE GEOMETRY 289
Theorem A.9 A line h in the plane which meets a second line k must meet
both sides of k.
Corollary A.10 Let A and B be two points of the plane outside a line k.
Then A and B lie on different sides of k if and only if k meets [A,B].1
Proof If k does not meet [A,B], then the connected set [A,B] is contained in
the complement of k and must be contained in one of its connected components. If
k meets [A,B] in S, then S divides the line it through A and B into two open half-
lines a and b with A E a and B E b. Since a and b are connected, they are entirely
contained in the connected components of the complement of k. It follows from
A.9 that they are contained in different components.
Corollary A.11 Through a point P of the plane there passes a unique line
perpendicular to a given line k.
Let A and B be two distinct points of the plane. The line n through the
midpoint of A and B and perpendicular to the line through A and B is called the
uernendicular bisector for A and B.
n.
To this end, let N be the point of intersection for [A,P] and n. Since P and B are
on the same side of n we have N 0 [P,B] and the sharp triangle inequality gives
d(P,B) < d(P,N) + d(N,B) = d(P,N) + d(N,A) = d(P,A)
The last equality follows from N E [P,A]. 0
This is a good place to stop. Readers with a sense of humour could once
again consult the opening chapter of [Coxeter] on Pons Asinorum. Now we have
proved "Pasch's axiom", a proof of the classification theorem, presented in the
introduction to the book, can be found in [Borsuk, Szmielew] or [Martin].
SOURCES
Let me try to make a list of the sources which have been most useful to
me in preparing the book. At the same time I hope to give an introduction to the
bibliography and some suggestions to further reading.
Chapter VII The proof of Poincare's theorem leans heavily on its prede-
cessors [de Rham], [Maskitl], [Beardon]. For the organization of the proof in case
of non orientation preserving transformations, it has been beneficial to conform to
the principles of combinatorial topology. Here the book of [Zieschang, Vogt,
Coldewey] have been very useful.
Chapter VIII The book [Fenchel] and a paper by [Coxeter']. For the
section on fixed points [Lyndon] and [Beardon].
292 BIBLIOGRAPHY
BIBLIOGRAPHY
SYMBOLS
K1 2 vol(K,L,M) 89
C ,0 44 goo 214
0 62 ej v 240
TV 62 *w 242
sl2(R) 88 A 262
INDEX
area, hyperbolic 117, 230, 231 elliptic transformation 82, 106, 272
hyperbolic transformation 83, 106, 272 normal vector 92, 248, 251
hypercycle 83, 102 normaliser of subgroup 134, 190
indecomposable matrix 21 odd Lorentz transformation 25
infinity, line at 18 orientable surface 229
interior angle of polygon 94 orientation of vector space 241