This action might not be possible to undo. Are you sure you want to continue?
University of Cambridge
GEOMETRY AND GROUPS
Albrecht D¨ urer’s engraving Melencolia I (1514)
Notes Lent 2011
T. K. Carne.
t.k.carne@dpmms.cam.ac.uk
c ( Copyright. Not for distribution outside Cambridge University.
CONTENTS
1 INTRODUCTION 1
1.1 Symmetry 1
1.2 Group Actions 4
Proposition 1.1 Orbit – Stabilizer theorem 5
2 ISOMETRIES OF EUCLIDEAN SPACE 7
2.1 Deﬁnitions 7
Lemma 2.1 Orthogonal maps as Euclidean isometries 8
Proposition 2.2 Euclidean isometries are aﬃne 8
2.2 Isometries of the Euclidean Plane 9
Proposition 2.3 Isometries of E
2
10
Proposition 2.4 Isometries of E
3
10
3 THE ISOMETRY GROUP OF EUCLIDEAN SPACE 11
3.1 Quotients of the Isometry group 11
Proposition 3.1 11
3.2 Matrices for Isometries 12
3.3 Finite Groups of Isometries of the Plane 13
Proposition 3.2 Finite subgroups of Isom(E
2
). 13
3.4 Compositions of Reﬂections 14
4 FINITE SYMMETRY GROUPS OF EUCLIDEAN SPACE 15
4.1 Examples of Finite Symmetry Groups 15
4.2 Finite Subgroups of Isom(E
3
). 16
Theorem 4.1 Finite symmetry groups in Isom
+
(E
3
) 17
5 THE PLATONIC SOLIDS 19
5.1 History 19
5.2 Regularity 21
5.3 Convex Regular Polyhedra 22
5.4 The Symmetry Groups 22
5.5 Fundamental sets 23
6 LATTICES 24
Proposition 6.1 Lattices in R 25
Proposition 6.2 Lattices in R
2
25
7 EUCLIDEAN CRYSTALLOGRAPHIC GROUPS 26
Lemma 7.1 The point group acts on the lattice 26
7.1 Rank 0 : Finite Groups 26
7.2 Rank 1: Frieze Patterns 27
7.3 Rank 2: Wallpaper patterns 28
Lemma 7.2 The crystallographic restriction 28
Corollary 7.3 Point groups 29
8 M
¨
OBIUS TRANSFORMATIONS 31
8.1 The Riemann Sphere 31
Proposition 8.1 Stereographic projection preserves circles. 33
8.2 M¨obius Transformations 33
Proposition 8.2 M¨obius transformations map circles to circles 34
Proposition 8.3 34
Proposition 8.4 Isometries of the Riemann sphere. 35
9 VISUALISING M
¨
OBIUS TRANSFORMATIONS 36
9.1 Fixed Points 36
Theorem 9.1 Fixed points of M¨obius transformations 36
Corollary 9.2 Trace determines conjugacy class of a M¨obius transformation 37
9.2 Inversion 38
Lemma 9.3 M¨obius transformations preserve inverse points 38
i
Proposition 9.4 Inversion 39
Proposition 9.5 The composition of two inversions is M¨obius. 39
10 THE HYPERBOLIC PLANE, I 41
10.1 M¨obius Transformations of the unit disc 41
Proposition 10.1 M¨obius transformations of the unit disc 41
10.2 The Hyperbolic Metric on D 42
Lemma 10.2 Hyperbolic geodesics from the origin. 44
Theorem 10.3 Hyperbolic metric on the unit disc. 45
11 THE HYPERBOLIC PLANE, II 46
11.1 The Hyperbolic Metric on a Half Plane. 46
11.2 Inversions 46
Proposition 11.1 Inversions preserve the hyperbolic metric 46
12 FUCHSIAN GROUPS 47
12.1 Single generator Fuchsian groups 47
Proposition 12.1 47
12.2 Triangle Groups 48
13 THE MODULAR GROUP 51
Proposition 13.1 Fundamental set for the modular group. 52
14 HYPERBOLIC 3SPACE 54
14.1 The Hyperbolic Metric 54
Proposition 14.1 Hyperbolic metric on B
3
54
14.2 Inversion 54
Proposition 14.2 Inversion in spheres 55
Proposition 14.3 Inversion preserves spheres. 55
Corollary 14.4 Inversion preserves circles. 55
Proposition 14.5 Inversion preserves angles. 56
15 EXTENDING M
¨
OBIUS TRANSFORMATIONS TO HYPERBOLIC SPACE 57
15.1 Inversions and the hyperbolic metric 57
Proposition 15.1 Inversions are hyperbolic isometries. 57
Lemma 15.2 57
Proposition 15.3 Hyperbolic geodesics 58
Proposition 15.4 Extensions of M¨obius transformations. 58
Theorem 15.5 M¨obius transformations as isometries of hyperbolic 3space. 58
15.2 The upper halfspace. 59
16 ISOMETRIES OF H
3
60
16.1 Examples in Hyperbolic Geometry 60
16.2 Axes of Isometries 62
17 INVOLUTIONS 64
Proposition 17.1 Isometries of H
3
are compositions of two involutions. 64
Proposition 17.2 65
18 KLEINIAN GROUPS 67
18.1 Finite Kleinian Groups 67
Lemma 18.1 67
Proposition 18.2 Finite Kleinian groups are conjugate to subgroups of SO(3). 68
18.2 Discontinuous Action 68
Lemma 18.3 68
Theorem 18.4 Discrete if and only if acts discontinuously. 69
19 LIMITS OF ORBITS 71
Proposition 19.1 The limit set is closed and Ginvariant. 71
Lemma 19.2 71
Proposition 19.3 Limit set is independent of the base point. 72
Lemma 19.4 72
Proposition 19.5 73
ii
Corollary 19.6 Limit sets are perfect. 73
Corollary 19.7 Limit sets are ﬁnite or uncountable. 74
20 HAUSDORFF DIMENSION 75
20.1 Cantor Sets 75
20.2 Hausdorﬀ Dimension 76
Proposition 20.1 77
Corollary 20.2 Lipschitz maps preserve Hausdorﬀ dimension 78
21 CALCULATING THE HAUSDORFF DIMENSION 80
Proposition 21.1 80
21.1 Invariant Sets 80
Lemma 21.2 Hausdorﬀ distance 81
Proposition 21.3 Invariant sets 81
Proposition 21.4 82
Theorem 21.5 83
22 EXAMPLES OF HAUSDORFF DIMENSION 84
22.1 The Cantor Set 84
22.2 The von Koch snowﬂake 85
22.3 The Sierpi´ nski Gasket 87
23 SCHOTTKY GROUPS 88
23.1 Fuchsian Groups 88
Proposition 23.1 88
23.2 Schottky Groups 90
Proposition 23.2 90
Theorem 23.3 Schottky groups are free groups. 92
23.2 The Limit Set for a Schottky Group 94
Lemma 23.4 95
24 DEGENERATE SCHOTTKY GROUPS 97
24.1 How Schottky Groups Degenerate 97
*24.2 Riemann Surfaces * 103
iii
1 INTRODUCTION
1.1 Symmetry
This course will explore symmetry groups. We will look at examples of symmetry groups acting
on various diﬀerent geometric spaces. We wish to have a large variety of symmetries, so we will look at
the simplest and most symmetric spaces, beginning with the Euclidean spaces R
2
, R
3
and the spheres
S
1
, S
2
. There is also a even more important example, the hyperbolic spaces, that we will look at later in
the course. Each of these geometric spaces has a metric and we will study the isometries that preserve
the distances between any pair of points.
These isometries are represented by matrices, so we will usually be considering straightforward
matrix groups such as O(2), O(3), SL(2, R), SL(2, C). These groups act on the geometric space and we
will want to study how they do so. We will look at various patterns on the geometric spaces and the
subgroups that preserve these patterns and so are symmetries of the patterns.
A ﬁrst example is the “Platonic Solids”:
These are the ﬁve convex regular solids. We will want to consider what makes them regular and prove
that there are only ﬁve.
Lecture 1 1
A second is the “wallpaper patterns”:
Andrew Crompton
There are 17 diﬀerent groups of symmetries for wallpaper patterns.
Lecture 1 2
Then there are tessellations of the hyperbolic plane:
M.C. Escher, Circle Limit IV (1960)
Lecture 1 3
Studying symmetry groups of the hyperbolic plane will lead us to consider limit sets which are
fractal. For example:
Sierpi´ nski Gasket Kleinian Limit Set
and show how to calculate their dimension.
Although many of the earlier courses in the Tripos are relevant to this one, there is rather little that
is required in the way of background. The most important background is from the “Algebra and Geom
etry” course. You will need to recall the deﬁnition of a group and how it acts on a set. We will review
this brieﬂy. You also need to know about groups of matrices such as GL(N, R), SL(N, C), SO(N), SU(N)
and the group of M¨obius transformations:
z →
az +b
cz +d
ad −bc = 1
acting on the Riemann sphere. We will also use the notion of a metric space and compactness from the
“Metric and Topological Spaces” course.
1.2 Group Actions
We will be interested in groups G that act as symmetries of a space X. So, each group element
g ∈ G gives us a symmetry X → X. More formally we say that a group G acts on a set X if there is a
map:
GX → X ; (g, x) → g x
which satisﬁes:
(a) e x = x for the identity e of G and any point x ∈ X;
(b) g (h x) = (gh) x for g, h ∈ G and any point x ∈ X.
For example, the group GL(n, R) of invertible n n real matrices acts on R
n
by matrix multipli
cation: (M, x) → Mx. The group SL(2, C) of 2 2 complex matrices with determinant 1 acts on the
Riemann sphere by:
a b
c d
, z
→
az +b
cz +d
.
Lecture 1 4
Note that the set Bij(X) of all bijections from X to itself forms a group under composition. If the
group G acts on X then the map θ : G → Bij(X) with
θ(g) : X → X ; x → g x
is a group homomorphism. For condition (a) shows that θ(e) = I
X
and condition (b) shows that
θ(g) ◦ θ(h)x = θ(gh)x.
Exercise:
1. Show that, for any group homomorphism θ : G → Bij(X), the group G acts on X by
(g, x) → θ(g)x .
The group action is faithful or eﬀective when θ : G → Bij(X) is injective.
Suppose that the group G acts on the set X. For each x ∈ X, the orbit Orb(x) is the set ¦g x :
g ∈ G¦ of points that x is mapped to by G. It is a subset of the space X. The stabilizer Stab(x) is
¦g ∈ G : g x = x¦. This is a subgroup of G.
Proposition 1.1 Orbit – Stabilizer theorem
If a group G acts on a set X, then the map
α : G/Stab(x) → Orb(x) ; gStab(x) → g x
is a bijection. When G is a ﬁnite group, this shows that
[G[ = [Stab(x)[[Orb(x)[
for each x ∈ X.
Proof:
Note that
g x = h x ⇔ (h
−1
g) x = x
⇔ h
−1
g ∈ Stab(x)
⇔ g ∈ hStab(x)
So α maps each coset gStab(x) onto the element g x in the orbit of x. Consequently, α is well deﬁned
and a bijection.
We can also consider the quotient of X by the group action. The relation
x ∼ y ⇔ y = g x for some g ∈ G
is an equivalence relation on X with the orbits as the equivalence classes. The quotient X/G is the set
of equivalence classes. A fundamental set is a subset F of X that contains exactly one element from
each orbit.
For example, let Z be the additive group of integers acting on the plane R
2
by translations:
n x = x +ni .
The stabiliser of each point x ∈ R
2
is the identity alone, while the orbit of x is the set ¦x+ni : n ∈ Z¦
of all translations of x by integer multiples of the unit vector i. Let F be the strip
x
1
x
2
∈ R
2
: 0 x
1
< 1
.
Lecture 1 5
Each point y =
y
1
y
2
∈ R
2
is equivalent to a point
y
1

y
2
in F, so F is a fundamental set. Two
points in closed strip F are only equivalent to one another when they are points
0
y
2
and
1
y
2
on the opposite edges. If we identify these points on the opposite edges then we obtain a cylinder. Each
orbit corresponds to a unique point of this cylinder, so we can identify this cylinder with the quotient
X/G.
Exercise:
2. Show that additive the group Z Z acts on the plane R
2
by
n
1
n
2
x
1
x
2
=
x
1
+n
1
x
2
+n
2
and that the unit square S =
x
1
x
2
: 0 x
1
< 1 and 0 x
2
< 1
is a fundamental set. Hence
show that we can identify the quotient R
2
/Z Z with a torus.
Let u, v be two vectors in R
2
and let P be the parallelogram:
¦λu +µv : 0 λ < 1 and 0 µ < 1¦ .
Suppose that P is also a fundamental set for the action of Z Z on R
2
. Show that
u =
a
c
, v =
b
d
for some integers a, b, c, d with ad −bc = ±1.
3. Consider the two maps:
A :
x
1
x
2
→
x
1
x
2
+ 1
; B :
x
1
x
2
→
x
1
+ 1
−x
2
acting on the plane R
2
. Let G be the group they generate. Is G Abelian? Find the orbit of a point
x under this group. Find a fundamental set and hence describe the quotient R
2
/G.
Lecture 1 6
2 ISOMETRIES OF EUCLIDEAN SPACE
2.1 Deﬁnitions
Let M be a metric space with d as its metric. A map T : M → M is an isometry if it is invertible
and preserves distances, so
d(T(x), T(y)) = d(x, y) for all x, y ∈ M .
The set of isometries of M form a group Isom(M) under composition.
In this section we want to study the isometries of the Euclidean space E
N
and especially of the
Euclidean plane. Euclidean space consists of R
N
with the Euclidean metric, which is deﬁned in terms
of the inner product (or scalar product). The inner product of two vectors x, y ∈ R
N
is given by
x y = x
1
y
1
+x
2
y
2
+. . . +x
N
y
N
.
We can use this to deﬁne the norm or length of a vector x as
[[x[[ =
√
(x x) =
√
(x
1
x
1
+x
2
x
2
+. . . +x
N
x
N
) .
The Euclidean metric is given by
d(x, y) = [[x −y[[ .
For any vector a ∈ R
N
the map T : x → x +a is clearly an isometry and is called the translation
by a. Let u be a unit vector. For each real number λ, the set π = ¦x : x u = λ¦ is a translation
of a vector subspace of dimension N − 1. This is called a hyperplane. Any vector x can be written as
(x u)u+x
⊥
with x
⊥
perpendicular to u. Then reﬂection in π should leave x
⊥
unaltered but map tu
to (2λ −t)u. Hence we deﬁne reﬂection in the hyperplane π to be
R : x → x −2(x u −λ)u .
It is straightforward to check that
[[R(x) −R(y)[[ = [[(x −y) −2((x −y) u)u[[ = [[x −y[[
so the reﬂection is an Euclidean isometry.
We wish to describe all of the isometries of Euclidean space. It is easiest to do this by ﬁrst
considering those isometries that ﬁx the origin. Our ﬁrst lemma shows that any such isometry is an
orthogonal linear map.
Recall that a linear map T : R
N
→R
N
is orthogonal when it preserves the inner product, so
T(x) T(y) = x y for all x, y ∈ R
N
.
Since the inner product is given by x y = x
t
y, this is equivalent to the NNmatrix M that represents
T satisfying
x
t
M
t
My = x
t
y for all x, y ∈ R
N
.
Hence, M
t
M = I. These matrices give us the orthogonal group
O(N) = ¦M : M is an N N real matrix with M
t
M = I¦ .
Lemma 2.1 Orthogonal maps as Euclidean isometries
Every orthogonal linear map B : R
N
→ R
N
is an isometry for the Euclidean metric. Conversely, if
B : R
N
→R
N
is a Euclidean isometry that ﬁxes the origin, the B is an orthogonal linear map.
Lecture 2 7
Proof:
Let B be an orthogonal linear map. Then
d(Bx, By)
2
= [[Bx −By[[
2
= B(x −y) B(x −y) = (x −y) (x −y) = d(x, y)
2
so B is an isometry.
Now suppose that B is an isometry of E
N
that ﬁxes the origin. The polarization identity:
2x y = [[x[[
2
+[[y[[
2
−[[x −y[[
2
= d(x, 0)
2
+d(y, 0)
2
−d(x, y)
2
shows that
Bx By = x y
for any vectors x, y ∈ R
N
. This shows that B preserves the inner product. We need to show that it is
linear.
Let (e
n
) be the standard orthonormal basis for R
N
. Because B preserves the inner product, (Be
n
)
is another orthonormal basis. For any vector x we have
x =
¸
(x e
n
)e
n
.
Also,
Bx =
¸
(Bx Be
n
)Be
n
=
¸
(x e
n
)Be
n
because B preserves the inner product. Hence
B :
¸
x
n
e
n
→
¸
x
n
Be
n
and so B is a linear map.
Consequently, B is an orthogonal linear map.
Now let A be any isometry of Euclidean space. The translation T by the vector A0 is an isometry
so B = T
−1
◦ A will be an isometry that ﬁxes the origin. Hence we obtain:
Proposition 2.2 Euclidean isometries are aﬃne
If A : E
N
→ E
N
is an isometry of the Euclidean space E
N
, then there is a vector v ∈ R
N
and an
orthogonal matrix B with
A(x) = Bx +v for x ∈ R
N
.
Conversely, this map is an isometry for any vector v and any orthogonal matrix B.
Proof:
Let A be an isometry of E
N
and set v = A0. Then the translation T by v is an isometry and
B = T
−1
◦ A : x → Ax −A0
is an isometry that ﬁxes the origin. Hence Lemma 2.1 shows that B is an orthogonal linear map with
Ax = Bx +v.
Conversely, suppose that B ∈ O(N), v ∈ R
N
and Ax = Bx+v. Then A is the composition of the
isometries B and translation by v, so it is also an isometry.
Lecture 2 8
2.2 Isometries of the Euclidean Plane
We can use the previous results to describe all of the isometries of the Euclidean plane: E
2
. First
suppose that B is an isometry that ﬁxes the origin. Then
B =
a b
c d
∈ O(2) .
This means that the columns of B are unit vectors orthogonal to one another. Hence we can choose an
angle θ ∈ [0, 2π) with
a
c
=
cos θ
sin θ
.
The vector
b
d
must be a unit vector orthogonal to
a
c
, so there are just two possibilities:
b
d
=
−sin θ
cos θ
or
sin θ
−cos θ
.
The ﬁrst case gives a rotation about the origin through an angle θ (or the identity when θ = 0). The
second gives reﬂection in the line ¦(x, y) : y = (tan
1
2
θ)x¦ at an angle
1
2
θ from the xaxis.
Note that the determinant of the orthogonal matrix B must be either +1 or −1. It is +1 for the
identity or rotations that preserve the orientation of the plane and −1 for reﬂections that reverse the
orientation.
If we conjugate B by a translation T : x → x + p, then we obtain another isometry T ◦ B ◦ T
−1
.
This ﬁrst translates p back to the origin, then applies B, and then translates the origin back to p. When
B is a rotation through an angle θ, then T ◦ B ◦ T
−1
is a rotation about T0 = p through an angle θ.
When B is a reﬂection in a line , then T ◦ B ◦ T
−1
is reﬂection in the line T().
Now consider an isometry A of the Euclidean plane that does not ﬁx the origin. Proposition 2.2
shows that Ax = Bx +v for some B ∈ O(2) and some vector v ∈ R
2
. When B is the identity, then A
is a translation. When B is a rotation, we can always choose a vector p with (I −B)p = v. This means
that
T ◦ B ◦ T
−1
(x) = Bx + (I −B)p = Bx +v
so the isometry A is a rotation about the point p. When B is a reﬂection in a line through the origin,
then we can split the vector v into a part v
1
perpendicular to and a part v
2
parallel to . The linear
map I − B maps onto the vector subspace of vectors perpendicular to , so we can choose a vector p
with (I −B)p = v
1
. This means that
T ◦ B ◦ T
−1
(x) = Bx + (I −B)p = Bx +v
1
so Ax = T ◦ B ◦ T
−1
(x) +v
2
. When v
2
= 0, this shows that A is reﬂection in the line translated by
p. However, when v
2
= 0, then A is a glide reﬂection, that is reﬂection in the line translated by p
followed by a translation parallel to .
Proposition 2.3 Isometries of E
2
An orientation preserving isometry of the Euclidean plane E
2
is:
(a) The identity.
(b) A translation.
(c) A rotation about some point c ∈ E
2
.
An orientation reversing isometry of E
2
is:
(d) A reﬂection.
(e) A glide reﬂections, that is a reﬂection in a line followed by a translation parallel to .
Lecture 2 9
A similar, but slightly more involved, argument gives the corresponding result for Euclidean space
E
3
.
Proposition 2.4 Isometries of E
3
An orientation preserving isometry of Euclidean 3space E
3
is:
(a) The identity.
(b) A translation.
(c) A rotation about some line .
(d) A screw rotation, that is a rotation about some line followed by a translation parallel to .
An orientation reversing isometry of E
3
is:
(e) A reﬂection in some plane Π.
(f) A glide reﬂection, that is a reﬂection in a plane Π followed by a translation parallel to Π.
(g) A rotatory reﬂection, that is a rotation about some axis followed by reﬂection in a plane perpen
dicular to .
Lecture 2 10
3 THE ISOMETRY GROUP OF EUCLIDEAN SPACE
3.1 Quotients of the Isometry group
Recall from Proposition 2.2 that every isometry A of E
N
can be written as
A : x → Bx +v
with B ∈ O(N) and v ∈ R
N
. Hence we can deﬁne a map
φ : Isom(E
N
) → O(N) by A → B .
Proposition 3.1
The map φ : Isom(E
N
) → O(N) given above is a group homomorphism with kernel equal to the group
Trans(E
N
) of all translations of E
N
.
Proof:
Let A
1
, A
2
be two isometries with A
k
x = B
k
x +v
k
. Then
A
2
◦ A
1
(x) = A
2
(B
1
x +v
1
) = B
2
(B
1
x +v
1
) +v
2
= (B
2
B
1
)x + (B
2
v
1
+v
2
) .
Hence,
φ(A
2
◦ A
1
) = B
2
B
1
= φ(A
2
)φ(A
1
)
which shows that φ is a group homomorphism.
The isometry A : x → Bx + v is in the kernel of φ when φ(A) = B = I. This means that A is a
translation.
The group Trans(E
N
) of translations is a normal subgroup of the isometry group, since it is the
kernel of the homomorphism φ. This means that an isometry A acts on the translations by conjugation.
If A : x → Bx +v and T is the translation T : x → x +t, then
A◦ T ◦ A
−1
(x) = x +Bt .
This action will be very important to us later when we look at crystallographic groups.
There is another important homomorphism from Isom(E
N
) that tells us whether an isometry pre
serves or reverses orientation. When A : x → Bx +v, we deﬁne ε(A) to be det B. Then
ε : Isom(E
N
) → ¦−1, +1¦
is a group homomorphism. The kernel of ε is the group Isom
+
(E
N
) of orientation preserving isometries
of E
N
. This is a normal subgroup of Isom(E
N
).
Recall that, for any surjective group homomorphism α : G → H, the inverse images α
−1
(h) are the
cosets of ker α in G. The number of cosets is equal to the number of elements in H and is called the index
of ker α in G. Since ε : Isom(E
N
) → ¦−1, +1¦ is a group homomorphism onto ¦−1, +1¦, the subgroup
Isom
+
(E
N
) has index 2 in Isom(E
N
). This means that it has just two cosets Isom
+
(E
N
) = ε
−1
(+1) and
the complement Isom
−
(E
N
) = ε
−1
(−1). For any orientation reversing isometry J, this complement is
equal to the coset JIsom
+
(E
N
).
If G is any subgroup of Isom(E
N
), then the restriction
ε[
G
: G → ¦−1, +1¦
is a group homomorphism. Let G
+
= G ∩ Isom
+
(E
N
) be the orientation preserving isometries in
G. These form the normal subgroup ker ε[
G
of G. The image of ε[
G
is either ¦+1¦ or ¦−1, +1¦.
Consequently, either G = G
+
contains only orientation preserving isometries or else G
+
is of index 2 in
G. When we study subgroups like G we often begin by looking at G
+
and then consider how we can
add orientation reversing isometries.
Lecture 3 11
3.2 Matrices for Isometries
The Euclidean space E
N
is the same set as R
N
but it has diﬀerent properties. In the vector space
R
N
the origin 0 is a special point. However, in Euclidean space it is not since translations can be used
to send 0 to any other point. Rather than thinking of E
N
as equal to R
N
, it is often more convenient
to identify it with a hyperplane in R
N+1
that does not go through the origin. Then we can represent
every isometry as an (N + 1) (N + 1) matrix.
Set E
N
to be the hyperplane ¦y = (y
n
)
N+1
n=1
∈ R
N+1
: y
N+1
= 1¦in R
N+1
. Any vector x ∈ R
N
then
corresponds to a point
x
1
∈ E
N
. The Euclidean metric on E
N
is just the restriction of the usual
metric on R
N+1
:
d(y, z) = [[y −z[[ .
Let A be an isometry of E
N
. Then Proposition 2.2 shows that A maps a vector x ∈ R
N
to the
vector Bx+v for some B ∈ O(N) and v ∈ R
N
. When we think of E
N
as the hyperplane ¦y : y
N+1
= 1¦
we see that A sends the vector
x
1
to
B v
0 1
x
1
=
Bx +v
1
.
So A is represented by the (N + 1) (N + 1) matrix
B v
0 1
.
Exercise:
4. Show that an (N + 1) (N + 1) matrix maps the hyperplane E
N
isometrically onto itself if and
only if
M =
B v
0 1
for some matrix B ∈ O(N) and some vector v ∈ R
N
.
This means that we can regard the group Isom(E
N
) as a group of matrices. This can make compu
tations simpler and it also gives us a natural metric on the group. For the (N + 1) (N + 1) matrices
are a vector space of dimension (N + 1)
2
and have a inner product given by
K M = tr K
t
M =
N+1
¸
i=1
N+1
¸
j=1
k
ij
m
ij
.
We can then deﬁne the distance between two matrices K and M as
d(K, M)
2
= (K −M) (K −M) =
¸
i,j
(k
ij
−m
ij
)
2
.
This metric behaves as you would expect. A sequence (M(k)) of matrices converges to a limit matrix
M if and only if the sequence (M(k)
ij
) of ij entries converges to M
ij
for each pair ij.
Lecture 3 12
3.3 Finite Groups of Isometries of the Plane
In Proposition 2.2 we showed that each isometry A of E
N
is of the form x → Bx + v with
B ∈ O(N), v ∈ R
N
. This means that A is aﬃne, so
A
¸
λ
j
x
j
=
¸
λ
j
A(x
j
) provided that
¸
λ
j
= 1 .
Let G be a ﬁnite subgroup of Isom(E
N
). Choose any point a ∈ E
N
. The centroid of the orbit of a:
c =
1
[G[
¸
T∈G
T(a)
then satisﬁes A(c) = c for each A ∈ G. Therefore all of the elements of G ﬁx the point c.
Proposition 3.2 Finite subgroups of Isom(E
2
).
A ﬁnite subgroup G of Isom(E
2
) is either a cyclic group consisting of N rotations through angles 2πk/N
(k = 0, 1, 2, . . . , N −1) about some point c, or else a dihedral group consisting of N rotations through
angles 2πk/N (k = 0, 1, 2, . . . , N −1) about some point c and N reﬂections in lines through c.
Proof:
The centroid c is ﬁxed by all of the isometries in G.
The subgroup G
+
= G∩Isom
+
(E
2
) is also a ﬁnite group, with order N say. Then each transforma
tion A ∈ G
+
is a rotation about c with A
N
= I. Hence A must be a rotation through an angle 2πk/N
for some integer k ∈ ¦0, 1, 2, . . . , N −1¦. There are only N such rotations about c so all of them must
lie in G
+
. Hence G
+
must be the cyclic group of order N generated by a rotation R about c through
an angle 2π/N.
If G consists only of orientation preserving isometries, then G = G
+
is cyclic. Otherwise, there
must be an orientation reversing isometry M in G` G
+
. This ﬁxes c so it must be a reﬂection in a line
through c. The homomorphism ε : G → ¦−1, +1¦ maps G
+
onto +1 and the coset G
+
M onto −1, so
[G[ = 2N.
The products M, RM, R
2
M, . . . , R
N−1
M are all distinct and are reﬂections in the line obtained by
rotating about c through angles πk/N for k = 0, 1, 2, . . . , N − 1 respectively. So we see that G is
dihedral of order 2N.
The groups described in this proposition are the symmetries of a regular Ngon centred on c, either
the orientation preserving isometries for the cyclic group or all the isometries for the dihedral group.
This is illustrated in the picture below. The orientation preserving isometries permute the shaded
regions while the orientation reversing isometries interchange the shaded and unshaded regions.
Lecture 3 13
Regular polygon of order N = 7.
3.4 Compositions of Reﬂections
We wish to ﬁnd the result of composing two reﬂections. The simplest way to do this is to choose
coordinates so that one of the reﬂections is, say, reﬂection in the xaxis and then simply use matrices.
You should do this. We will adopt a diﬀerent approach.
Let M be the reﬂection of E
2
in the line . Let T be a translation perpendicular to . Then the
conjugate MTM
−1
is translation in the reverse direction, that is T
−1
. So MTM
−1
= T
−1
and hence
MT = T
−1
M. Similarly, if R is a rotation about a point on , then the conjugate MRM
−1
is rotation
about the same point in the opposite direction, so MRM
−1
= R
−1
.
Now suppose that M
is reﬂection in a second line
. If the two lines and
are parallel, then
we can ﬁnd a translation T perpendicular to both line that maps onto
. Hence M
= TMT
−1
.
Consequently
M
M = TMT
−1
M = TMMT = T
2
.
So the composition of two reﬂections in parallel lines is a translation perpendicular to those lines by
twice the distance between them.
If the two lines and
meet at a point P, then there is a rotation R about P that maps onto
.
Hence M
= RMR
−1
. Consequently
M
M = RMR
−1
M = RMMR = R
2
.
So the composition of the two reﬂections is a rotation about the point P of intersection through twice
the angle from to
Lecture 3 14
4 FINITE SYMMETRY GROUPS OF EUCLIDEAN SPACE
If P is an object in E
3
, then the isometries of E
3
that are symmetries of P form a subgroup. By
choosing P to be highly symmetric, such as one of the Platonic solids, we obtain a nontrivial, ﬁnite
group of symmetries. The aim of this lecture is to prove that these give us all of the ﬁnite subgroups of
Isom(E
3
). We will begin by looking only at the orientation preserving symmetries.
Recall from '3 that every ﬁnite subgroup G of Isom(E
N
) must ﬁx a point c ∈ E
N
. By translating,
we can ensure that this point c is at the origin. Then each symmetry A ∈ G must be in O(N). This
means that A maps the unit sphere S
N−1
= ¦x ∈ R
N
: [[x[[ = 1¦ to itself isometrically. Hence we can
equally well look at the ﬁnite subgroups of Isom(S
N
).
The only orientation preserving symmetries of E
3
that ﬁx the origin are the identity and rotations
about an axis through the origin. Hence we need only consider which rotations are symmetries.
4.1 Examples of Finite Symmetry Groups
We begin by giving examples of objects with ﬁnite symmetry groups. It is usually easy to see what
the symmetries are and to verify that we have them all by using the orbit – stabilizer theorem.
Let v be a unit vector in R
3
and R the rotation about this vector through an angle 2π/N for some
N = 2, 3, 4, . . .. Let u be a unit vector orthogonal to v. Then the points u, Ru, R
2
u, . . . , R
N−1
u are
the vertices of a regular Ngon P. By joining each vertex of P to 2v we get the cone on P. By joining
each vertex of P to both 2v and −2v we get the double cone on P.
0
2v
u
Ru
R
2
u
R
N−1
u
0
2v
−2v
u
Ru
R
2
u
R
N−1
u
Regular cone Regular double cone
It is clear that each of maps I, R, R
2
, . . . , R
N−1
is a symmetry of the cone or double cone on P. For
the single cone, these are the only orientation preserving symmetries. For the group of symmetries acts
on the vertices of the polygon P with the orbit of u being all N vertices and the stabilizer of u being
only the identity. So the orbit – stabilizer theorem shows that the orientation preserving symmetry
group of the cone on P has N elements and so is the cyclic group ¦I, R, R
2
, . . . , R
N−1
¦
∼
= C
N
.
For the double cone on P, there are other symmetries. The rotation S about u through an angle
π is one and SR, SR
2
, . . . , SR
N−1
are the others. Each of these is a rotation through an angle π. Once
again, the orbit – stabilizer theorem shows that this is all of the orientation preserving symmetries.
Lecture 4 15
They form a group isomorphic to the dihedral group D
2N
. (Note that we usually think of the dihedral
group as the plane symmetries of P. Then it has N reﬂections. The rotations through angle π act of
the plane of P in the same way as the reﬂections but interchange the halfspaces above and below P.)
There are also orientation reversing symmetries. For the cone, there are N reﬂections in planes
through v. So the full symmetry group is isomorphic to D
2N
. For the double cone, the reﬂection J in
the plane of the polygon P is one symmetry. The others are JR, JR
2
, . . . , JR
N−1
, which are rotatory
reﬂections, and JS, JSR, . . . , JSR
N−1
, which are reﬂections in planes through v. Since J commutes
with all the other symmetries, we see that the full symmetry group is isomorphic to D
2N
C
2
.
We can do a similar analysis when P is one of the Platonic solids. Recall that the Platonic solids
are the regular tetrahedron, octahedron, cube, dodecahedron and icosahedron. Let G be the group
of symmetries of one of these, say P, centred on the origin. Then G
+
is the subgroup of orientation
preserving symmetries. These groups act on the vertices. The regularity of the Platonic solids shows
that there are symmetries that move any vertex to any other, so there is one orbit consisting of all the
vertices. The stabilizer in G
+
consists of all the rotations that ﬁx that vertex. These must permute the
faces that meet at the vertex, so the stabilizer in G
+
is cyclic with order equal to the number of faces
meeting at each vertex. The various numbers are:
Orbit Stabilizer [G
+
[
(vertices) (faces at each vertex)
Tetrahedron 4 3 12
Cube 8 3 24
Octahedron 6 4 24
Dodecahedron 20 3 60
Icosahedron 12 5 60
These give ﬁnite groups of orientation preserving symmetries. (We will see later that the groups are
not all distinct. A Platonic solid and its dual have the same symmetries, so the groups for the cube and
octahedron are the same, as are the groups for the dodecahedron and the icosahedron.) If we include
the orientation reversing symmetries we obtain ﬁnite groups with twice as many elements.
We have not shown that these ﬁnite symmetry groups really exist. To do so, we would need to
show that the Platonic solids exist and their symmetry groups act as claimed on the vertices. This was
done in the Geometry course and various approaches to it are outlined in the example sheets.
4.2 Finite Subgroups of Isom(E
3
).
Now we aim to prove that the ﬁnite groups described in the previous section are the only ones that
exist. We will concentrate on the groups of orientation preserving symmetries. The groups are then
cyclic of order N for N = 1, 2, 3, . . .; dihedral of order 2N; and the rotational symmetry groups of the
Platonic solids, called the tetrahedral, octahedral and icosahedral groups.
Theorem 4.1 Finite symmetry groups in Isom
+
(E
3
)
Let G be a ﬁnite subgroup of Isom
+
(E
3
) consisting of orientation preserving isometries. Then G is is
the orientation preserving symmetry group of one of the following:
(a) A cone on a regular plane polygon.
(b) A double cone on a regular plane polygon.
(c) A regular tetrahedron.
(d) A regular octahedron.
(e) A regular icosahedron.
Lecture 4 16
Proof:
First translate so that G ﬁxes the origin. Then each nonidentity element of G is a rotation
about an axis through 0. Let Ω be the set of unit vectors that are ﬁxed by by some nonidentity element
of G. Then Ω is a ﬁnite set and G acts on it. Let Ω
1
, Ω
2
, . . . , Ω
J
be the diﬀerent orbits in Ω. The orbit
– stabilizer theorem shows that each vector u ∈ Ω
j
has a stabilizer of order S
j
= [G[/[Ω
j
[.
Now we count the number of pairs in the set
X = ¦(A, u) : A ∈ G` ¦I¦, u ∈ S
2
and Au = u¦ .
Each A ∈ G ` ¦I¦ is a rotation and so ﬁxes exactly two unit vectors. Therefore [X[ = 2([G[ − 1).
Alternatively, each u ∈ Ω gives rise to [Stab(u)[ −1 pairs in X. So
[X[ =
J
¸
j=1
(S
j
−1)[Ω
j
[ =
J
¸
j=1
[G[ −[Ω
j
[ .
Dividing by [G[ we see that
2 −
2
[G[
=
J
¸
j=1
1 −
1
S
j
. ∗
Each stabilizer of u ∈ Ω has order at least 2, so
1 −
1
S
j
1
2
.
Hence,
2 > 2 −
2
[G[
=
J
¸
j=1
1 −
1
S
j
1
2
J
and so J is 1, 2 or 3.
Order the orbits so that S
1
S
2
S
3
2. Clearly there are no solutions to (∗) with J = 1. If
J = 2 then (∗) gives
2 −
2
[G[
= 2 −
1
S
1
−
1
S
2
2 −
2
S
1
so S
1
[G[. This implies that S
1
= [G[ and [Ω
1
[ = 1. Hence,
2 −
2
[G[
= 2 −
1
S
1
−
1
S
2
= 2 −
1
[G[
−
1
S
2
and so S
2
= [G[ and [Ω
2
[ = 1. Hence Ω consists of two unit vectors v and −v which are ﬁxed by each
isometry in G. The group G is then a ﬁnite isometry group of the plane orthogonal to v so it is cyclic
by Proposition 3.2 . This shows that G is the symmetry group of a cone as in (a).
When J = 3, equation (∗) gives
1
S
1
+
1
S
2
+
1
S
3
= 1 +
2
[G[
.
This implies that
3
S
3
1 +
2
[G[
> 1
so S
3
= 2.
Equation (∗) now yields
2
S
2
1
S
1
+
1
S
2
=
1
2
+
2
[G[
>
1
2
Lecture 4 17
which implies that S
2
= 2 or 3.
When S
2
= 2 we have
1
S
1
=
2
[G[
which gives S
1
= N, S
2
= 2, S
3
= 2 and [G[ = 2N. The orbit Ω
1
has has just two points. Let v be
one of them. The stabilizer Stab(v) is a ﬁnite group of N rotations about v, so it is cyclic generated
by a rotation R. Choose u as one of the points in Ω
3
. Then the others are Ru, R
2
u, . . . , R
N−1
u. The
stabilizer of u has order 2, so it contains a rotation S of order 2. This maps v to −v. It is now apparent
that each element of G is a symmetry of a double cone as in (b).
When S
2
= 3 we have
1
S
1
=
2
[G[
+
1
6
>
1
6
so S
1
= 3, 4 or 5. The possibilities are:
S
1
[Ω
1
[ S
2
[Ω
2
[ S
3
[Ω
3
[ [G[
3 4 3 4 2 6 12
4 6 3 8 2 12 24
5 12 3 20 2 30 60
We need to show that these correspond to the symmetry groups of the tetrahedron, octahedron and
icosahedron respectively.
We will consider the middle case as an example. Here Ω
1
has 6 points. The stabiliser of each is a
cyclic group of order S
1
= 4. Choose one point v ∈ Ω
1
. The stabiliser of v is a cyclic group of order
4; let R be a generator. Now −v is also ﬁxed by R and has the same stabiliser. So it must be in Ω
1
.
There remain 4 other points in Ω
1
and these must be w, Rw, R
2
w, R
3
w all lying in the plane through
0 orthogonal to v. Hence the points of Ω
1
are the 6 vertices of a regular octahedron.
Note that the points of Ω
2
are the midpoints of the faces of this octahedron and the points of Ω
3
are the midpoints of the edges. The points of Ω
2
are the vertices of a cube. This is the dual of the
octahedron. The polyhedron and its dual have the same symmetry group.
Ω
1
Ω
2
Ω
3
In a similar way, the ﬁrst row in the table above gives us a regular tetrahedron. Ω
1
is the set of
vertices; Ω
2
the centres of the faces; Ω
3
the midpoints of edges. The tetrahedron is dual to another
tetrahedron.
The ﬁnal row gives a icosahedron. Ω
1
is the set of vertices; Ω
2
the centres of the faces; Ω
3
the
midpoints of edges. The dual is a dodecahedron.
Lecture 4 18
5 THE PLATONIC SOLIDS
5.1 History
The Platonic solids have been known and studied for a very long time. Stones carved into polyhedral
shapes date from about 2000BC in Scotland.
(See http://www.georgehart.com/virtualpolyhedra/neolithic.html.)
The Pythagoreans were aware of at least some of the solids and endowed them with a mystical
signiﬁcance. Theaetetus (c. 417BC  369BC) was the ﬁrst to prove that there were only ﬁve convex
regular polyhedra. Plato refers to the solids in the Timaeus c. 360BC and follows the Pythagoreans in
giving them mystical signiﬁcance. Four of them represented the four elements: the tetrahedron for ﬁre,
the cube for earth, the octahedron for air and the icosahedron for water. This association was justiﬁed
on the grounds that the icosahedron is the smoothest of the polyhedra while the tetrahedron is the
sharpest. The dodecahedron represented the entire universe with the twelve faces showing the twelve
signs of the zodiac. Euclid devoted the 13th and last book of his Elements to the Platonic solids.
In the 18th Century Kepler pursued Plato’s mystical interest but also tried to use the Platonic
solids to describe the known universe. In the Mysterium Cosmographicum (1596) he suggested that the
radii of the orbits of the ﬁve known planets could be found by inscribing the solids one inside another.
Lecture 5 19
Lecture 5 20
5.2 Regularity
The Platonic solids are the only regular, convex polyhedra. We need to consider what regularity
means. For a polygon, regularity means that each edge looks the same and each vertex looks the same.
This means that there are symmetries of the polygon that map each edge to any other and each vertex
to any other. Note that there are nonconvex polygons that also have this property. For example:
We will consider polyhedra that have a ﬁnite number of vertices, edges and faces. For the polyhedron
to be regular we want to there to be symmetries that map any vertex, edge or face to any other. A ﬂag
for the polyhedron P is a triple (v, e, F) consisting of a vertex v, an edge e and a face F such that v is
one end of e, and e is one of the boundary edges of F. We will say that the polyhedron is regular if, for
any two ﬂags (v, e, F) and (v
, e
, F
) there is a symmetry of P that maps one ﬂag to the other.
This condition certainly implies that the group of symmetries of P acts on the set V of vertices
and V is a single orbit. Once we know where a symmetry sends each vertex, we can determine where
each edge and face goes and hence ﬁnd the symmetry completely. This means that the symmetry group
G of P is a ﬁnite group of isometries. Hence it is one of the groups found in '4 . Furthermore, the
group G will contain symmetries that ﬁx a vertex v and maps any one edge at v to any other. Hence,
each vertex has a nontrivial stabilizer. This means that the vertices of P must be one of the orbits
Ω
1
, Ω
2
, Ω
3
found in the proof of Theorem 4.1 . Indeed, if we are not in the trivial situation where only
two edges meet at each vertex, then the vertices can not be Ω
3
which has stabilizers of order 2. Hence
the set of vertices is one of the orbits we considered in '4 .
It is not, however, necessary that we join up the vertices in the expected way. See, for example, the
nonconvex regular polygon above. Similar possibilities arise for polyhedra. For example, consider the
vertices of a cube with side length 1. We can join each vertex to the 3 other vertices at a distance
√
2
rather than those at unit distance. This gives two tetrahedra which together form a nonconvex regular
polyhedron.
Note that the intersection of these two tetrahedra is the octahedron dual to the original cube.
Lecture 5 21
5.3 Convex Regular Polyhedra
Suppose that P is a convex polyhedron that is regular. The regularity certainly implies that each
face is a regular polygon and an equal number of these faces meet at each vertex. Say the faces are
pgons and q meet at each vertex. Here p and q are at least 3. The pair ¦p, q¦ is called the Schl¨aﬁ
symbol for the polyhedron.
Choose one vertex v. The q vertices adjacent to v are the vertices of a regular qgon. The angle
between two edges that meet at a vertex is π −2π/p. At each vertex q of these meet, so
q
π −
2π
p
< 2π .
This simpliﬁes to
(p −2)(q −2) < 4
so the only possible solutions are ¦3, 3¦, ¦3, 4¦, ¦4, 3¦, ¦3, 5¦, ¦5, 3¦. These give the ﬁve Platonic solids.
5.4 The Symmetry Groups
In Theorem 4.1 showed which groups could arise as the ﬁnite subgroups of Isom(E
3
). However,
we have not identiﬁed which groups these are. By using some of the nonconvex polyhedra with these
symmetry groups, we can easily do so.
Consider ﬁrst the tetrahedral group Sym(T). This permutes the four vertices of a tetrahedron T,
and so we obtain a group homomorphism θ : Sym(T) → S
4
into the symmetric group S
4
on the vertices.
If a symmetry ﬁxes all four vertices, then it is the identity. Hence θ is injective. We already know,
from the orbit – stabilizer theorem, that Sym(T) has 24 elements, so θ must be an isomorphism. Thus
the full symmetry group of the regular tetrahedron is isomorphic to S
4
. By looking at the individual
symmetries we see that the orientation preserving symmetries correspond to the even permutations of
the vertices. So the group Sym
+
(T) is isomorphic to the alternating group A
4
.
Now consider the symmetry group of a cube centred on the origin. This is the same as the symmetry
group of the dual octahedron which has vertices at the centres of each face of the cube. There are two
tetrahedra embedded in this cube with vertices at the vertices of the cube, denote these by T
+
and
T
−
. The isometry Jx → −x interchanges T
+
and T
−
. It also commutes with every other symmetry of
the cube. A symmetry S of the cube either maps each tetrahedron onto itself or else interchanges the
tetrahedra. In the latter case, J ◦ S is another symmetry that maps each tetrahedron onto itself. Hence
we see that the homomorphism
θ : Sym(C) → Sym(T
+
) ¦1, −1¦ ; S →
(S, 1) when S(T
+
) = T
+
;
(J ◦ S, −1) when S(T
+
) = T
−
.
is injective. Counting elements shows that the full symmetry group Sym(C) is isomorphic to S
4
C
2
.
The subgroup Sym
+
(C) is isomorphic to S
4
. (We can also prove this by considering how a symmetry
of the cube acts on the four long diagonals joining a vertex of V to its antipodal vertex.)
Finally, consider the symmetry group of the dodecahedron D, or the dual icosahedron. There are
ﬁve cubes embedded inside the dodecahedron (or equivalently ﬁve octahedra within which an icosahedron
is embedded). A symmetry of the dodecahedron permutes these 5 embedded cubes, so we get a group
homomorphism θ : Sym(D) → S
5
into the symmetric group on the cubes. By looking at each rotational
symmetry of the dodecahedron we can check that θ actually maps into the alternating group A
5
. Suppose
that the symmetry T is in the kernel of this homomorphism. Then T maps each cube to itself. A vertex
v of the dodecahedron is a vertex of exactly two of the cubes and these two cubes have only the vertices
v and J(v) = −v in common. Hence, T must map each vertex either to itself or to the antipodal vertex
J(v). Since T is an isometry, we must have T = I or T = J. This shows that the map
Sym(D) → A
5
C
2
; T → (θ(T), ε(T))
is an injective group homomorphism. The orbit – stabilizer theorem shows that there are an equal
number 120 of elements in Sym(D) as in A
5
C
2
, so this is an isomorphism. The orientation preserving
symmetries of the dodecahedron form a group isomorphic to A
5
.
Lecture 5 22
Five cubes embedded in a regular dodecahedron
5.5 Fundamental sets
The symmetry group G of a Platonic solid P acts on the faces of P so we can look for a fundamental
set. Since G permutes the faces, we can restrict our attention to one face F and look for a fundamental
set of Stab(F). This stabilizer is the symmetry group of the polygonal face. It is easy to see that a
fundamental set for this dihedral group is a closed triangle as shown in the diagram below.
The corners of this triangle are a vertex of P, the midpoint of an edge of P and the centre of the
face F. The triangle thus corresponds to a ﬂag for P. For each ﬂag we obtain a copy of the fundamental
triangle and these tessellate the surface of the polyhedron.
We can also look for a fundamental set for the symmetry group acting on all of E
3
. For this, we
can take the cone on the triangle found above with its vertex at the centroid of P.
Lecture 5 23
6 LATTICES
Let M be a metric space with metric d. A point x ∈ M is isolated if there is some neighbourhood
V of x that contains no point of M except x. This means that there is a δ > 0 with
d(x, y) > δ for all y ∈ M ` ¦x¦ .
For example, each point of Z is isolated. The metric space M is discrete if each point of M is isolated.
Note that the particular metric on M is not important only the topology.
Let G be a matrix group. Then G has a natural (Euclidean) metric. We say that G is a discrete
group if it is discrete for this metric. If the identity I is isolated in G, then every other point T ∈ G is
also isolated. For the multiplication
G → G ; A → TA
is continuous and has a continuous inverse A → T
−1
A. Hence, to check that a group is discrete we need
only check that I is isolated.
Exercise:
5. Show that G is a discrete matrix group if and only if there is no sequence of nonidentity elements
elements g
n
∈ G that converge to the identity.
For example, the group SL(2, Z) of 2 2 matrices M =
a b
c d
with a, b, c, d ∈ Z and ad −bc = 1
is discrete, because any matrix M = I satisﬁes
d(M, I)
2
= (a −1)
2
+b
2
+c
2
+ (d −1)
2
1 .
Every ﬁnite group is certainly discrete. Discrete groups can be inﬁnite but they can not be too
large.
First we will look at the discrete groups of translations. These groups correspond to additive
subgroups of R
N
. Let G be a subgroup of Trans(E
N
), then the set
Λ = ¦T(0) : T ∈ G¦
is an additive subgroup of R
N
. Conversely, if Λ is any additive subgroup of R
N
, then
G = ¦x → x +v : v ∈ Λ¦
is a subgroup of Trans(E
N
). Note that d(T
1
, T
2
) = [[v
1
−v
2
[[, so G is a discrete group if and only if Λ
is a discrete subset of R
N
. We call a discrete additive subgroup of R
N
a lattice in R
N
.
For example, the set Zw
1
for any nonzero vector w
1
is a lattice in R
2
. Similarly, Zw
1
+ Zw
2
is
a lattice in R
2
for any two linearly independent vectors w
1
, w
2
. We will show that these are the only
lattices in R
2
.
Proposition 6.1 Lattices in R
Each lattice in R is of the form Zw for some w ∈ R.
Lecture 6 24
Proof:
Since Λ is discrete, there is a δ > 0 with [λ −0[ > δ for each λ ∈ Λ` ¦0¦. Hence, [λ
1
−λ
2
[ > δ
for each pair of distinct points λ
1
, λ
2
∈ Λ. Consequently, there can be no more than a ﬁnite number of
points of Λ in any ball B(0, r). This implies that either Λ = ¦0¦ or else there is a point a ∈ Λ ` ¦0¦
closest to 0.
In the ﬁrst case we have Λ = Zw for w = 0. In the second case we will see that Λ = Za. We
certainly have a ∈ Λ, so Za ⊂ Λ. Suppose that b ∈ Λ. Then b is a scalar multiple of a, say b = ta with
t ∈ R. Now t = k +t
with k ∈ Z and 0 t
< 1. So b
= b −ka is also in Λ and has
[b
[ = t
[a[ < [a[ .
The choice of a tells us that b
must be 0, so b = ka ∈ Za as required.
We can extend the argument used above to ﬁnd the lattices in R
2
and, indeed, in R
N
for any N.
Proposition 6.2 Lattices in R
2
Each lattice in R
2
is either ¦0¦, or Zw
1
, or Zw
1
+ Zw
2
for a pair of linearly independent vectors
w
1
, w
2
∈ R
2
.
Proof:
Let Λ be a lattice in R
2
with Λ = ¦0¦. As in the previous proposition, we can ﬁnd a δ > 0
with d(λ
1
, λ
2
) > δ for each pair of distinct points λ
1
, λ
2
∈ Λ. Hence, only a ﬁnite number of points of Λ
can lie within a ball B(0, r). Choose a vector w
1
∈ Λ` ¦0¦ with d(w
1
, 0) minimal. The argument used
in the proof of the previous proposition shows that a scalar multiple tw
1
is in Λ if and only if t ∈ Z.
If there are no elements of Λ`Zw
1
, then we are ﬁnished. Otherwise, there are vectors v ∈ Λ`Zw
1
.
Each such vector can be written as
v = v
⊥
+tw
1
with v
⊥
orthogonal to w
1
and t ∈ R .
Observe that
d(v, Rw
1
) = [[v
⊥
[[ and d(v, 0) =
[[v
⊥
[[
2
+t
2
[[w
1
[[
2
.
Since only a ﬁnite number of points of Λ lie within any ball about the origin, we can choose a vector
v ∈ Λ ` Zw
1
with [[v
⊥
[[ = d(v, Rw
1
) minimal. Call this vector w
2
.
The vectors w
1
, w
2
are certainly linearly independent, so any vector v ∈ Λ can be written as a
linear combination v = t
1
w
1
+ t
2
w
2
. The real numbers t
j
can be written as t
j
= k
j
+ t
j
with k
j
∈ Z
and 0 t
j
< 1. Then
v
= v −(k
1
w
1
+k
2
w
2
) = t
1
w
1
+t
2
w
2
∈ Λ
and has
d(v
, Rw
1
) = d(t
2
w
2
, Rw
1
) = t
2
d(w
2
, Rw
1
) < d(w
2
, Rw
1
) .
The choice of w
2
ensures that v
must be in Rw
1
and in Λ. We showed above that such a vector must
be an integer multiple of w
1
, so v ∈ Zw
1
+Zw
2
.
We say that the lattice Λ has rank 0, 1 or 2 according as Λ = ¦0¦, Zw
1
or Zw
1
+Zw
2
.
The vectors w
1
, w
2
in this proposition are not unique. For an example, consider the hexagonal
lattice Z
1
0
+Z
1
2
1
2
√
3
.
The proof also shows that the parallelogram
¦t
1
w
1
+t
2
w
2
: 0 t
1
, t
2
< 1¦
is a fundamental set for the lattice Λ acting on E
2
. The quotient E
2
/Λ is obtained by identifying the
parallel sides of this parallelogram to give a torus.
Lecture 6 25
7 EUCLIDEAN CRYSTALLOGRAPHIC GROUPS
A (2dimensional Euclidean) crystallographic group is a discrete subgroup of Isom(E
2
). Let G be
such a group. The homomorphism φ : Isom(E
2
) → O(2) deﬁned in '3 has all the translations Trans(E
2
)
as its kernel. So, when we restrict it to G we get a group homomorphism
φ : G → O(2)
with kernel equal to G ∩ Trans(E
2
). This is a discrete group of translations, so gives a lattice Λ. We
call the rank of Λ the rank of the original crystallographic group G. The image φ(G) is called the point
group G of G. It is a subgroup of the orthogonal group O(2). We think of the group G as being made
by combining the lattice of translations with the point group.
Lemma 7.1 The point group acts on the lattice
Each isometry in the point group G of a discrete subgroup G of Isom(E
2
) maps the lattice of G onto
itself.
Proof:
Let A : x → Bx +v be an isometry in G. Then φ(A) = B is in the point group G. If w is in
the lattice Λ for G, then the translation T : x → x +w is in G. Hence, the composite:
A◦ T ◦ A
−1
: x → B(B
−1
(x −v) +w) +v = x +Bw
is in G. This shows that Bw ∈ Λ.
We are now in a position to describe all of the 2dimensional Euclidean crystallographic groups.
However, the entire program is rather tedious so we will only explain the main themes and illustrate
the results.
Let G be the crystallographic group, Λ its lattice, and G its point group.
7.1 Rank 0 : Finite Groups
If G is of rank 0, then Λ = ¦0¦. This means that G can contain no translations and no glide
reﬂections (since the square of a glide reﬂection is a translation). Hence G contains only the identity,
rotations and reﬂections. Let us ﬁrst consider the orientation preserving subgroup G
+
of G. Suppose
that G contains two rotations R
1
, R
2
. Then their commutator R
1
R
2
R
−1
1
R
−1
2
is a translation since
φ(R
1
R
2
R
−1
1
R
−1
2
) = φ(R
1
)φ(R
2
)φ(R
1
)
−1
φ(R
2
)
−1
and SO(2) is commutative. Since this translation is
by a vector in the lattice Λ = ¦0¦, it must be the identity and so R
1
and R
2
commute. This implies
that they are rotations about the same centre. Since G
+
is a discrete group of rotations, it must be
a ﬁnite cyclic group. Since G
+
is a normal subgroup of G with index 1 or 2, it must also be ﬁnite.
Proposition 3.2 shows that G must then be a cyclic or dihedral group.
Lecture 7 26
7.2 Rank 1: Frieze Patterns
If G is of rank 1, then Λ = Zw for some nonzero vector w ∈ R
2
. We will denote the translation
x → x +w by T. Each orthogonal map B in the point group G maps Λ onto itself. So B must be the
identity, a rotation R through angle π about the origin, a reﬂection M in the line Rw, or a reﬂection
N in the line through the origin orthogonal to w. These four maps form a group D
4
. Hence G must be
a subgroup of ¦I, R, M, N¦. The possible subgroups are:
¦I¦ ¦I, R¦ ¦I, M¦ ¦I, N¦ ¦I, R, M, N¦
For each of these we can work out what the possibilities are for G.
(a) G = ¦I¦.
Then G = Λ and consists entirely of translations and is cyclic of inﬁnite order. This G is the
symmetry group of a pattern such as:
We call such patterns whose symmetry group is a rank 1 crystallographic group frieze patterns.
(b) G = ¦I, R¦.
Then G must contain a rotation A with φ(A) = R. So A is a rotation about some centre c through
an angle π. Choose coordinates in the plane so that this point c is the origin. Then A = R. Note that
RTR
−1
= T
−1
. Hence the group G consists of the translations T
k
and the rotations T
k
R for k ∈ Z.
The rotation T
k
R is through angle π with centre
1
2
kw. The group G is an inﬁnite dihedral group D
∞
.
Such a G is the symmetry group of a frieze pattern such as:
(c) G = ¦I, M¦.
G must contain at least one isometry A with φ(A) = M. This means that A is either a reﬂection
in a line parallel to w or a glide reﬂection parallel to w. In either case, the other isometries of G that
φ maps to M are T
k
A. Choose coordinates so that the origin is on the mirror.
In the ﬁrst case G contains the translations T
k
, the reﬂection A, and the glide reﬂections T
k
A.
Note that ATA = T, so G
∼
= C
∞
C
2
.
In the second case, A
2
is a translation, so A
2
= T
r
for some r ∈ Z ` ¦0¦. Then A is reﬂection
followed by translation parallel by
1
2
rw. If r is even, say r = 2k, then T
−k
A is a reﬂection and we are
in the previous case. If r = 2k + 1, then C = T
−k
A is reﬂection followed by translation by
1
2
w. This
generates the cyclic group G and gives the pattern:
Lecture 7 27
(d) G = ¦I, N¦.
G must contain at least one isometry A with φ(A) = N. This means that A is either a reﬂection
in a line orthogonal to w or a glide reﬂection orthogonal to w. In the second case A
2
would be a
translation orthogonal to w, which is impossible. So A is a reﬂection in a line orthogonal to w.
Choose coordinates so that the origin is on the mirror. Note that ATA = T
−1
so the group G is an
inﬁnite dihedral group.
(e) G = ¦I, R, M, N¦.
As in (b) we can choose coordinates so that R ∈ G. There must be an isometry A ∈ G that maps
to M under φ. There are two cases, as in (b). If we can choose A to be a reﬂection in a line parallel to
w, then ATA
−1
= T and ARA
−1
= R so G
∼
= D
∞
C
2
:
If we can choose A to be reﬂection followed by translation by
1
2
w then A generates an inﬁnite cyclic
group and RAR
−1
= A
−1
, so G is an inﬁnite dihedral group.
Exercise:
6. Find fundamental sets for each of the frieze groups.
7.3 Rank 2: Wallpaper patterns
If G is of rank 2, then Λ = Zw
1
+ Zw
2
for two linearly independent vectors w
1
, w
2
∈ R
2
. Each
orthogonal map B in the point group G maps Λ onto itself. The rotation about the origin through an
angle π always maps Λ onto itself. For most lattices there are no other symmetries ﬁxing the origin as
the following result shows.
Lemma 7.2 The crystallographic restriction
A rotation in the point group G of a crystallographic group G must be of order 2, 3, 4 or 6.
Proof:
Let w
1
be an element of Λ ` ¦0¦ with [[w
1
[[ = r minimal. Since Λ is discrete, the set S =
Λ ∩ ¦v : [[v[[ = r¦ is ﬁnite and the point group G must map this isometrically to itself. This certainly
implies that any rotation R ∈ G is of ﬁnite order.
Choose R as the rotation in G through the smallest angle, say θ. The vector Rw
1
− w
1
is also in
Λ so it must have length at least [[w
1
[[ unless it is 0. This means that θ π/3. Consequently R has
order 2, 3, 4, 5 or 6.
Lecture 7 28
0
w
1
Rw
1
θ
Suppose that R had order 5. Then, w
1
+ R
2
w
1
∈ Λ would be at a distance less than r from the
origin, which is forbidden.
0
w
1
Rw
1
R
2
w
1
R
2
w
1
+w
1
2π/5
2π/5
Corollary 7.3 Point groups
The point group of a 2dimensional Euclidean crystallographic group is either cyclic C
1
, C
2
, C
3
, C
4
, C
6
or dihedral D
2
, D
4
, D
6
, D
8
, D
12
.
Proof:
Take w
1
as in the previous proof and let R ∈ G be the rotation through the smallest angle θ.
If θ = π/2, then w
1
and w
2
= Rw
1
generate the lattice. This is the square lattice.
If θ = 2π/3 or θ = π/3, then w
1
and w
2
= Rw
1
generate the lattice. This is the hexagonal lattice.
Lecture 7 29
Square lattice Hexagonal lattice
We can now proceed as in the previous part to ﬁnd all the possible crystallographic groups with
rank 2. These are the wallpaper groups. There are 17 of them, as illustrated below.
http://www.clarku.edu/∼djoyce/wallpaper/
Lecture 7 30
8 M
¨
OBIUS TRANSFORMATIONS
8.1 The Riemann Sphere
It is useful to add an extra element ∞ to the complex plane to form the extended complex plane
C∪ ¦∞¦. It seems that the point at inﬁnity is very diﬀerent from the other, ﬁnite points but Riemann
showed that this is not really the case. He did this by representing all of the points of the extended
complex plane by points of the unit sphere S
2
in R
3
. This sphere is called the Riemann sphere.
Let P be the unit sphere
P = ¦(z, t) ∈ C R : [z[
2
+t
2
= 1¦
in the three dimensional real vector space C R. The “North pole” of this sphere will be denoted by
N = (0, 1). Stereographic projection maps points of the complex plane to points of the Riemann sphere
P and vice versa. Let z ∈ C. Then the straight line through N and (z, 0) crosses the sphere at N
and another point (w, t) ∈ P. We write π(z) = (w, t) and deﬁne π(∞) = N. Then π gives us a map
π : C ∪ ¦∞¦ →P. This map is invertible, for if (t, w) is any point of P except N, then the straight line
through N and (w, t) will cross the plane ¦(z, s) : s = 0¦ at a single point (z, 0) with π(z) = (w, t).
It is easy to give a formula for stereographic projection. The points on the line from z ∈ C to N
are ¦s(z, 0) + (1 − s)(0, 1) : s ∈ R¦. This line crosses the sphere P when s = 0, giving the North pole,
and when s =
2
1−z
2
, giving
π(z) =
2z
1 +[z[
2
,
−1 +[z[
2
1 +[z[
2
.
Note from the diagram below that the triangles ´z0N, ´0AN and ´0Aπ(z) are all similar. Pythagoras’
theorem shows that d(N, z) =
1 +[z[
2
. Hence,
d(N, A) = d(A, π(z)) =
1
1 +[z[
2
and d(N, π(z)) =
2
1 +[z[
2
.
0
N
C
z
π(z)
A
Now let us consider two points z
1
, z
2
∈ C. The chordal distance κ(z
1
, z
2
) is the Euclidean distance
between the stereographic projections π(z
1
) and π(z
2
). In the diagram below, we show the triangle with
vertices N, z
1
and z
2
.
Lecture 8 31
N
z
1
z
2
π(z
1
)
π(z
2
)
We know that
d(N, z
j
) =
1 +[z
j
[
2
and d(N, π(z
j
)) =
2
1 +[z
j
[
2
.
So the triangles ´Nz
1
z
2
and ´Nπ(z
2
)π(z
1
) are similar with scale factor
2
√
1+z
1

2
√
1+z
2

2
. Conse
quently,
κ(z
1
, z
2
) = d(π(z
1
), π(z
2
)) =
2[z
1
−z
2
[
1 +[z
1
[
2
1 +[z
2
[
2
.
When one of the points, say z
2
, is ∞ then we interpret this as
κ(z
1
, ∞) =
2
1 +[z
1
[
2
.
Exercise:
7. Prove the formula for the chordal distance between two points z
1
, z
2
∈ C ∪ ¦∞¦ algebraically by
using the formula for stereographic projection.
8. Prove that the chordal metric is a metric on the Riemann sphere.
9. Two points z, z
∈ C ∪ ¦∞¦ are antipodal if their stereographic projections satisfy π(z
) = −π(z),
so they are at the opposite ends of a diameter of the Riemann sphere. Show that z, z
are antipodal
if and only if
z
= −1/z .
A similar argument to that used above to ﬁnd the chordal metric shows that stereographic projection
is conformal: it preserves the angle between curves. For, in the diagram below, the straight line from N
to z crosses the tangent plane at π(z) and the complex plane C at the same angle θ. Hence projection
with centre N from the tangent plane to C preserves angles. Consequently, π also preserves the angle
between two curves that meet at z.
Lecture 8 32
0
N
C
z
π(z)
Tangent plane at z
θ
θ
It is convenient to deﬁne circles in C∪ ¦∞¦ to mean both straight lines and circles. The following
result explains why this is so.
Proposition 8.1 Stereographic projection preserves circles.
A curve Γ in C∪ ¦∞¦ is a circle or a straight line if, and only if, the stereographic projection π(Γ) is a
circle on the Riemann sphere.
Proof:
We can write any circle or straight line in C ∪ ¦∞¦ as
a
o
[z[
2
+az +az +a
∞
= 0 (1)
where a
0
, a
∞
∈ R and a ∈ C. The stereographic projection π(z) is
(w, t) =
2z
1 +[z[
2
,
−1 +[z[
2
1 +[z[
2
.
So (1) is equivalent to
(a
0
+a
∞
) +aw +aw + (a
0
−a
∞
)t = 0 . (2)
This is the intersection with the Riemann sphere of a plane, so it is a circle on the sphere.
Note that the plane intersects the sphere if, and only if, [a[
2
− a
0
a
∞
0. This same condition
ensures that (1) does describe a circle or straight line rather than the empty set.
8.2 M¨obius Transformations
Let a, b, c, d be complex numbers with ad−bc = 0. Then we can deﬁne a map T : C∪¦∞¦ →C∪¦∞¦
by
T : z →
az +b
cz +d
.
Note that T(∞) = a/c and T(−d/c) = ∞. These maps are called M¨obius transformations and form a
group M¨ob under composition of maps. The map
φ : GL(2, C) → M¨ob ;
a b
c d
→ T
Lecture 8 33
is a group homomorphism.
A matrix M =
a b
c d
is in the kernel of this homomorphism when
az +b
cz +d
= z for all z ∈ C ∪ ¦∞¦ .
This occurs if and only if a = d and b = c = 0, so
a b
c d
= λI for some scalar λ ∈ C ` ¦0¦. This
shows that a M¨obius transformation is unaltered when we multiply each of the coeﬃcients a, b, c, d by a
nonzero scalar λ. Usually we choose the scalar λ so that the determinant ad−bc is 1. Then the matrix
a b
c d
is in SL(2, C). Now
φ : SL(2, C) → M¨ob ;
a b
c d
→ T
is a group homomorphism whose kernel consists of the two matrices I and −I. Consequently, the M¨obius
group is the quotient SL(2, C)/¦I, −I¦. We denote this quotient by PSL(2, C) and call it the projective
special linear group.
Recall that M¨obius transformations map circles to circles.
Proposition 8.2 M¨obius transformations map circles to circles
A M¨obius transformation maps any circle on the Riemann sphere to a circle on the Riemann sphere.
Proof:
Let Γ be the circle ¦z : p
0
[z[
2
+ pz + pz + p
∞
= 0¦. Let T be a M¨obius transformation with
inverse
S : z →
az +b
cz +d
.
Then
T(Γ) = ¦z : S(z) ∈ Γ¦ =
z : p
0
[az +b[
2
+p(az +b)(cz +d) +p(az +b)(cz +d) +p
∞
[cz +d[
2
= 0
¸
.
Expanding this gives an expression of the form:
¸
z : q
0
[z[
2
+qz +qz +q
∞
= 0
¸
which is clearly another circle.
Proposition 8.3
For any triples of distinct points in the Riemann sphere, (a
0
, a
1
, a
∞
) and (b
0
, b
1
, b
∞
) , there is a unique
M¨obius transformation T with T(a
0
) = b
0
, T(a
1
) = b
1
and T(a
∞
) = b
∞
.
Proof:
The M¨obius transformation
S
a
: z →
a
1
−a
∞
a
1
−a
0
z −a
0
z −a
∞
has S
a
(a
0
) = 0, S
a
(a
1
) = 1 and S
a
(∞) = ∞. Hence, T = S
−1
b
◦ S
a
has the required properties. If T
is
another M¨obius transformation with the same properties, then S
b
◦ T
◦ S
−1
a
ﬁxes 0, 1 and ∞ so it must
be the identity.
It follows from this that we can ﬁnd a M¨obius transformation that maps any circle in P onto any
other circle. For we choose three points on the ﬁrst and ﬁnd a M¨obius transformation that maps them
to three points on the second.
Exercise:
10. Let Γ
1
, Γ
2
be two disjoint circles on the Riemann sphere. Show that there is a M¨obius transforma
tion that maps them to two circles in C centred on 0.
Lecture 8 34
Proposition 8.4 Isometries of the Riemann sphere.
A M¨obius transformation is an isometry of the Riemann sphere (for the chordal metric) if, and only if,
it is represented by a matrix
M =
a b
c d
∈ SU(2) .
M ∈ SU(2) means that the matrix M preserves the complex inner product ', ` on C
2
and has deter
minant 1. This means that M
∗
M = I or, equivalently, that d = a, c = −b and [a[
2
+[b[
2
= 1.
Proof:
Let J : P → P be the map z → −1/z that maps a point to the antipodal point. Then J(z)
is the unique point of P that is at a distance 2 from z. Hence any M¨obius isometry T must satisfy
J ◦ T ◦ J = T. Let T be represented by the matrix M =
a b
c d
with det M = 1. We then have
J ◦ T ◦ J : z →
−dz +c
bz −a
.
This shows that J ◦ T ◦ J is itself a M¨obius transformation represented by the matrix
−d c
b −a
.
Therefore, if T is an isometry, then the two matrices
a b
c d
and
−d c
b −a
both represent T.
This implies that
−d c
b −a
= ±
a b
c d
.
The + sign requires that d = −a and c = b so 1 = ad −bc = −[a[
2
−[b[
2
, which is impossible. Hence we
must have the − sign and this shows that M is unitary.
Conversely, suppose that the M¨obius transformation T is represented by the matrix M ∈ SU(2).
Each z ∈ P can be written as z = z
1
/z
2
with z =
z
1
z
2
∈ C
2
. Similarly, write w = w
1
/w
2
with
w =
w
1
w
2
. Write w
∗
=
w
∗
1
w
∗
2
=
−w
2
w
1
, so that Jw = w
∗
1
/w
∗
2
= −w
2
/w
1
. Then we have
κ(Jw, z) =
2[Jw −z[
1 +[Jw[
2
1 +[z[
2
=
2[ −w
2
z
2
−w
1
z
1
[
[w
1
[
2
+[w
2
[
2
[z
1
[
2
+[z
2
[
2
=
2['w, z`[
[[w[[ [[z[[
.
Since J ◦ T = T ◦ J and 'Ma, Mb` = 'a, b` we see that
κ(T(Jw), T(z)) = κ(J(Tw), T(z)) =
2['M(w), M(z)`[
[[M(w)[[ [[M(z)[[
=
2['w, z`[
[[w[[ [[z[[
= κ(Jw, z)
and so T is an isometry.
Lecture 8 35
9 VISUALISING M
¨
OBIUS TRANSFORMATIONS
9.1 Fixed Points
Let T be the M¨obius transformation represented by the matrix M =
a b
c d
with ad − bc = 1.
The vector
z
1
z
2
is an eigenvector of M precisely when T(z
1
/z
2
) = z
1
/z
2
. This means that z
1
/z
2
is
a ﬁxed point of T. We know, from Linear Algebra, that the matrix M is conjugate to
λ 0
0 λ
−1
for
some λ = 0, or to
1 1
0 1
. This gives corresponding results for the M¨obius transformations. Let us
prove this directly using the transformations.
Theorem 9.1 Fixed points of M¨obius transformations
A nonidentity M¨obius transformation has either 1 or 2 ﬁxed points in P. If it has 1, then it is conjugate
in M¨ob to P : z → z + 1. If it has 2, then it is conjugate in M¨ob to M
k
: z → kz for some k = 0, 1.
Proof:
Suppose that the M¨obius transformation T has two ﬁxed points z
0
and z
∞
. Choose a M¨obius
transformation A with A(z
0
) = 0 and A(z
∞
) = ∞. Then the conjugate A◦T ◦A
−1
ﬁxes 0 and ∞. This
implies that A◦ T ◦ A
−1
is represented by a matrix
λ 0
0 λ
−1
for some λ = −1, 0, +1. Consequently,
A◦ T ◦ A
−1
= M
λ
2.
Suppose that T has 1 ﬁxed point z
0
only. Choose A with A(z
0
) = ∞. Then A ◦ T ◦ A
−1
ﬁxes ∞
alone. This means that it is z → z +b for some b = 0. By replacing A by b
−1
A we get A◦ T ◦ A
−1
= P.
Exercise:
11. When are two of the M¨obius transformations M
k
and P conjugate?
12. Find all of the M¨obius transformations that commute with M
k
for a ﬁxed k. Hence describe the
group
Z(T) = ¦A ∈ M¨ob : A◦ T = T ◦ A¦
for an arbitrary M¨obius transformation T. Describe the set ¦A(z
o
) : A ∈ Z(T)¦ for z
o
a point in
P.
A nonidentity M¨obius transformation is said to be:
parabolic if it is conjugate to P;
elliptic if it is conjugate to M
k
for [k[ = 1 (k = 1);
hyperbolic if it is conjugate to M
k
for k ∈ R (k = −1, 0, +1);
loxodromic if it is conjugate to M
k
for k ∈ C with [k[ = 1 and k / ∈ R.
Theorem 9.1 shows that every nonidentity transformation falls into one of these classes. It is simple
to tell which by considering the trace.
Lecture 9 36
Corollary 9.2 Trace determines conjugacy class of a M¨obius transformation
Let T be a nonidentity M¨obius transformation represented by a matrix M =
a b
c d
with determinant
1. Then T is
parabolic ⇔ tr M = ±2;
elliptic ⇔ −2 < tr M < 2;
hyperbolic ⇔ tr M < −2 or tr M > 2;
loxodromic ⇔ tr M / ∈ R.
Proof:
We know that T is conjugate to M
k
or to P. This means that the matrix M is conjugate to
λ 0
0 λ
−1
for λ a square root of k, or to ±
1 1
0 1
. Now we simply note that
tr
λ 0
0 λ
−1
= λ +λ
−1
and tr
1 1
0 1
= 2 .
Exercise:
13. If M is a 2 2 matrix with determinant 1, show that the characteristic equation for M is
t
2
−(tr M)t + 1 = 0 .
Deduce that the trace determines the eigenvalues of M.
14. Suppose that the M¨obius transformation T is represented by the matrix M but that det M = 1.
Show that T is parabolic if and only if (tr M)
2
= 4 det M. Establish similar conditions for T to be
elliptic, hyperbolic or loxodromic.
It is now fairly simple to visualise how M¨obius transformations act on the Riemann sphere. An
elliptic transformation is conjugate to z → e
iθ
z. This rotates the sphere ﬁxing 0 and ∞. Each point is
moved along a circle. A hyperbolic transformation is conjugate to z → kz for k > 1. This moves points
along arcs of circles from one ﬁxed point towards the other. A loxodromic transformation is conjugate
to z → kz for k / ∈ R. This moves points along logarithmic spirals away from one ﬁxed point and towards
the other. Finally, a parabolic transformation is conjugate to z → z +1. This maps points along a circle
through the single ﬁxed point. The points are mapped away on one side and towards on the other. The
pictures below illustrate this.
Elliptic Hyperbolic
Lecture 9 37
Loxodromic Parabolic
9.2 Inversion
Let Γ be a circle on the Riemann sphere. Two distinct points z, z
are inverse points for Γ if every
circle orthogonal to Γ through z also passes through z
. Also, when z ∈ Γ, we say that z and z itself
are inverse points. We will prove later that, for every point z there is a unique point z
so that z, z
are
inverse points for Γ.
For example, two points are inverse for the real axis R∪¦∞¦ when they are complex conjugates of
one another.
Lemma 9.3 M¨obius transformations preserve inverse points
Let T be a M¨obius transformation and Γ a circle in P. If z and z
are inverse points for Γ then T(z),
T(z
) are inverse points for T(Γ).
Proof:
We know that the M¨obius transformation T preserves angles and maps circles to circles. Any
circle through z orthogonal to Γ is therefore mapped to a circle through T(z) orthogonal to T(Γ). The
original circle passes through z
so the the image passes through T(z
) as required.
Proposition 9.4 Inversion
For each circle Γ in the Riemann sphere and each point z ∈ P there is a unique point J(z) with z, J(z)
inverse points for Γ.
Lecture 9 38
The map J is called inversion in Γ. It is an involution: J
2
= I and reverses orientation, so it is certainly
not a M¨obius transformation. J ﬁxes every point of the circle Γ.
Proof:
It is clear that C(z) = z is the unique point with z, C(z) inverse points for R ∪ ¦∞¦.
For any circle Γ we can ﬁnd a M¨obius transformation T that maps R∪¦∞¦ onto Γ. Now the lemma
shows that J(z) = T(C(T
−1
(z))) is the unique point with z, J(z) inverse points for Γ.
.
Example: Inversion in the unit circle.
The M¨obius transformation
T : z → −i
z +i
z −i
maps the unit circle T onto the real axis. (It is a rotation of the Riemann sphere about an axis through
1 and −1.) Hence, inversion in T is given by
J(z) = T
−1
(T(z)) =
1
z
.
More generally, when Γ = ¦z ∈ C : [z −c[ = r¦, then two points z, z
are inverse points for Γ when they
lie on the same halfline from c to ∞ and [z −c[ [z
−c[ = r
2
. Hence
J(z) = c +
r
2
z −c
.
Exercise:
15. Show that inversion maps any circle to another circle. Show that inversion preserves the magnitude
of angles but reverses their orientation.
Inversions in circles are analogous to reﬂections. Our ultimate aim is to ﬁnd a metric on the ball
B
3
= ¦x ∈ R
3
: [[x[[ < 1, so that the M¨obius transformations form the orientation preserving isometry
group of B
3
for this metric. The inversions will then be reﬂections in hyperbolic planes. Before achieving
this goal we will study the simpler case of those M¨obius transformations that map a disc onto itself.
Proposition 9.5 The composition of two inversions is M¨obius.
The composition of an even number of inversions is a M¨obius transformation.
Proof:
Let C be the particular inversion z → z. If Γ is a circle, we can ﬁnd a M¨obius transformation
T that maps R ∪ ¦∞¦ onto Γ. Then inversion in Γ is given by J = T ◦ C ◦ T
−1
.
If T(z) =
az +b
cz +d
, then C ◦ T ◦ C(z) =
az +b
cz +d
. This shows that C◦T ◦C is a M¨obius transformation.
Hence,
C ◦ J = C ◦ T ◦ C ◦ T
−1
= (C ◦ T ◦ C) ◦ T
−1
is a M¨obius transformation, as is its inverse J ◦ C.
Now, if J
1
, J
2
are two inversions we can write
J
2
◦ J
1
= (J
2
◦ C) ◦ (C ◦ J
1
)
to show that J
2
◦ J
1
is a M¨obius transformation.
Lecture 9 39
It is quite easy to identify the M¨obius transformations that we obtain by composing two inversions.
Suppose that J
1
, J
2
are inversions in the circles Γ
1
, Γ
2
respectively. If Γ
1
and Γ
2
cross at two points
w
1
, w
2
, then we can conjugate by a M¨obius transformation that sends these points to 0 and ∞ respec
tively. This transforms Γ
1
and Γ
2
to two straight lines through 0 (and ∞). The inversions become
reﬂections in these lines. So J
2
◦ J
1
is the elliptic transformation rotating about an axis through 0 and
∞ through twice the angle between the lines. Clearly we can obtain any elliptic transformation in this
way.
Similarly, if Γ
1
and Γ
2
touch at a single point w, then we can transform w to ∞ and see that J
2
◦J
1
is a parabolic transformation ﬁxing ∞.
Finally, if Γ
1
and Γ
2
are disjoint, then we can conjugate them to get two concentric circles
¦z : [z[ = 1¦ and ¦z : [z[ = R¦. Then
J
1
(z) =
1
z
; J
2
(z) =
R
2
z
; so J
2
◦ J
1
(z) = R
2
z .
Hence J
2
◦ J
1
is a hyperbolic transformation.
It is clear that we can obtain any elliptic, parabolic or hyperbolic transformation in this way.
There are no other ways that two circles can intersect, so loxodromic transformations can not be the
composition of two inversions.
Exercise:
16. Show that any loxodromic transformation is the composite of 4 inversions.
Lecture 9 40
10 THE HYPERBOLIC PLANE, I
10.1 M¨obius Transformations of the unit disc
Inversion in the unit circle is J : z → 1/z. This ﬁxes points of the unit circle T and interchanges
the unit disc D and the complementary disc D
= ¦z ∈ P : [z[ > 1¦.
Proposition 10.1 M¨obius transformations of the unit disc
A M¨obius transformation T maps the unit disc onto itself if, and only if, it is of the form:
z →
az +b
bz +a
with [a[
2
−[b[
2
= 1 .
Proof:
Let T be the M¨obius transformation T : z →
az+b
cz+d
with ad − bc = 1. If T maps the unit disc
D onto itself, then it must also map the unit circle onto itself. Hence, it must map any pair of inverse
points for T to another pair of inverse points. This implies that J ◦ T ◦ J = T. Now
J(T(J(z))) =
dz +c
bz +a
so the matrices
d c
b a
and
a b
c d
both represent the same transformation. This means that
d c
b a
= ±
a b
c d
.
If we have the + sign, then the matrix is
a b
b a
with [a[
2
−[b[
2
= 1
as required. If we have the − sign, then the matrix is
a b
−b −a
with −[a[
2
+[b[
2
= 1. So [T(0)[ = [ −b/a[ > 1, which is impossible if T maps D onto itself.
Conversely, suppose that
T : z →
az +b
bz +a
with [a[
2
−[b[
2
= 1 .
Then J ◦ T ◦ J = T. Any point z ∈ T satisﬁes J(T(z)) = T(J(z)) = T(z) so T maps the unit circle onto
itself. Consequently, T must map the unit disc onto either D or the complementary disc D
. However,
[T(0)[ = [b/a[ < 1 so it must map onto D.
The M¨obius transformations that map D onto itself form a subgroup
M¨ob(D) =
z →
az +b
bz +a
: [a[
2
−[b[
2
= 1
of the M¨obius group. This is the M¨obius group of the disc D.
Lecture 10 41
Example: For each ω with [ω[ = 1, the map z → ωz is clearly in M¨ob(D). Also, for each z
o
∈ D the
map
z →
z +z
o
1 +z
o
z
is represented by the matrix
¸
1
√
1−z
o

2
z
o
√
1−z
o

2
z
o
√
1−z
o

2
1
√
1−z
o

2
¸
and so is in M¨ob(D).
Exercise:
17. Show directly that the map
z →
ωz +z
o
1 +z
o
ωz
maps the unit disc D onto itself when [ω[ = 1 and z
o
∈ D. Show conversely, that every transforma
tion in M¨ob(D) is of this form.
.
For any other disc ∆ in the Riemann sphere, we can do the same argument as above to ﬁnd the
group M¨ob(∆) of M¨obius transformations that map ∆ onto itself. This is a subgroup of the full M¨obius
group conjugate to M¨ob(D). For, we can ﬁnd a transformation T with T(D) = ∆. Then
A ∈ M¨ob(D) if, and only if, T ◦ A◦ T
−1
∈ M¨ob(∆) .
A particularly important example is when ∆ is the upper halfplane: R
2
+
= ¦x + iy ∈ C : y > 0¦.
Inversion in the boundary of this is complex conjugation J : z → z. So a M¨obius transformation
T : z →
az+b
cz+d
with ad −bc = 1 maps R
2
+
onto itself when J ◦ T ◦ J = T and T(i) ∈ R
2
+
. Now
J ◦ T ◦ J : z →
az +b
cz +d
so we need
a b
c d
= ±
a b
c d
.
The + sign gives transformations z →
az+b
cz+d
with a, b, c, d ∈ R and ad −bc = 1. These do map the upper
halfplane onto itself. The − sign gives z →
az+b
cz+d
with a, b, c, d ∈ iR and ad − bc = 1. These map the
upper halfplane onto the lower halfplane. This shows that every M¨obius transformation mapping the
upper halfplane onto itself is represented by a matrix in SL(2, R). So M¨ob(R
2
+
)
∼
= SL(2, R)/¦I, −I¦.
10.2 The Hyperbolic Metric on D
We wish to deﬁne a new metric, the hyperbolic metric, on the unit disc D for which each of the
M¨obius transformations in M¨ob(D) will be an isometry.
We begin by deﬁning this metric for an inﬁnitesimal displacement dz. This should have length
ds = λ(z)[dz[ for some density function λ : D → (0, ∞). More formally, this means that a smooth curve
γ : [a, b] →D should have length
L(γ) =
b
a
λ(γ(t)) [γ
(t)[ dt .
If such a metric is to have each transformation in M¨ob(D) as an isometry, then the function λ is
almost completely determined. For
T : z →
z +z
o
1 +z
o
z
Lecture 10 42
is a M¨obius transformation mapping D onto D for each z
o
∈ D and so must be an isometry. An
inﬁnitesimal displacement dz from 0 is mapped by T to the T
(0)dz at z
o
. So we must have
λ(0)[dz[ = λ(z
o
)[T
(0)dz[ .
This means that
λ(0) = λ(z
o
)[T
(0)[ = λ(z
o
)(1 −[z
o
[
2
) .
So the metric must be given by ds =
K
1−z
2
[dz[ at the point z ∈ D. It is usual to set the constant K to
2. So we deﬁne the hyperbolic density on D to be
λ(z) =
2
1 −[z[
2
for z ∈ D .
The hyperbolic density increases as we approach the boundary of the disc. This means that a
constant Euclidean displacement increases in hyperbolic length as we approach the boundary. Simi
larly, a displacement of constant hyperbolic length has decreasing Euclidean length as we approach the
boundary.
Lecture 10 43
M.C. Escher, Circle Limit IV (1960)
In Escher’s picture above, all of the angels are of the same hyperbolic size even though, to our
Euclidean eyes, they appear to get smaller as we approach the boundary.
Having deﬁned the hyperbolic density that gives the length of inﬁnitesimal displacements, it is
simple to deﬁne the hyperbolic metric. The hyperbolic distance ρ(z
0
, z
1
) is the inﬁmum of the lengths
L(γ) =
b
a
λ(γ(t)) [γ
(t)[ dt =
b
a
2[γ
(t)[
1 −[γ(t)[
2
dt
over all smooth curves γ : [a, b] →D that have γ(a) = z
0
and γ(b) = z
1
. We will see that there is a path
γ from z
0
to z
1
that has shortest hyperbolic length. This is the hyperbolic geodesic from z
0
to z
1
.
Lemma 10.2 Hyperbolic geodesics from the origin.
The hyperbolic geodesic from 0 to z ∈ D is a radial line and it has hyperbolic length log
1 +[z[
1 −[z[
.
Proof:
It is clear that rotation about 0 preserves the hyperbolic length of any curve, so we may assume
that z = R ∈ [0, 1).
Lecture 10 44
Use polar coordinates to write a curve γ : [a, b] → D as γ(t) = r(t)e
iθ(t)
. We will assume that
γ(a) = 0 and γ(b) = z = R 0. Then the hyperbolic length of γ is more than the hyperbolic length of
its radial part, for
L(γ) =
b
a
2[γ
(t)[
1 −[γ(t)[
2
dt =
b
a
2[r
(t) +iθ
(t)r(t)[
1 −r(t)
2
dt
b
a
2[r
(t)[
1 −r(t)
2
dt = L(r) .
Moreover, [r
(t) +iθ
(t)r(t)[ =
r
(t)
2
+r(t)
2
θ
(t)
2
so we have equality only when θ
(t) is identically 0.
This shows that the radial path from 0 to z is the shortest path from 0 to z.
The length of this radial path is:
R
0
2
1 −r
2
dr = log
1 +R
1 −R
so we see that
ρ(0, z) = log
1 +[z[
1 −[z[
.
Theorem 10.3 Hyperbolic metric on the unit disc.
The expression
ρ(z
0
, z
1
) = inf ¦L(γ) : γ is a smooth curve in D from z
0
to z
1
¦
gives a metric on D for which each M¨obius transformation that maps D onto itself is an isometry. A
geodesic for this metric is the arc of a circle orthogonal to the unit circle. This is the hyperbolic metric
on D.
Proof:
Proposition 10.1 shows that any M¨obius transformation from D onto D must be of the form
T : z →
az +b
bz +a
with [a[
2
−[b[
2
= 1. For this we have
T
(z) =
1
(bz +a)
2
.
A straightforward calculation gives:
1
1 −[T(z)[
2
=
[bz +a[
2
[bz +a[
2
−[az +b[
2
=
[bz +a[
2
1 −[z[
2
.
So we see that
λ(T(z))[T
(z)[ =
2
1 −[T(z)[
2
[T
(z)[ =
2
1 −[z[
2
= λ(z) .
This shows that ρ(T(z
0
), T(z
1
)) = ρ(z
0
, z
1
).
The previous lemma certainly shows that ρ(0, z) > 0 whenever z = 0. Since we can always ﬁnd a
T ∈ M¨ob(D) with T(z
0
) = 0, it follows that ρ(z
0
, z
1
) > 0 whenever z
0
= z
1
. The symmetry and triangle
inequality for ρ are obvious, so ρ is a metric.
The lemma shows that the geodesic from 0 = T(z
0
) to T(z
1
) is a radial line segment. This is a
segment of a straight line through 0 orthogonal to the unit circle. The M¨obius transformation T maps
T onto itself and preserves angles, so the geodesic from z
0
to z
1
is an arc of the circle through z
0
and
z
1
that is orthogonal to T.
Lecture 10 45
11 THE HYPERBOLIC PLANE, II
11.1 The Hyperbolic Metric on a Half Plane.
Rather than working with the unit disc, we could work with any other disc ∆. We know that there
is a M¨obius transformation T that maps ∆ onto D and we deﬁne the hyperbolic metric on ∆ so that
this is an isometry from ∆ with this metric to D with the hyperbolic metric. This deﬁnition does not
depend on which transformation T we choose. For, if S is another M¨obius transformation that maps ∆
onto D, then R = S ◦ T
−1
will be a M¨obius transformation that maps D onto itself. This means that R
is an isometry for the hyperbolic metric on D. Hence T and S give the same metric on ∆.
An important example is when ∆ is the upper halfplane R
2
+
. For this we can take T to be the map
T : z → i
z −i
z +i
.
(This is rotation of the Riemann sphere about an axis through ±1.) The hyperbolic metric on R
2
+
will
then be ds = µ(z)[dz[ for some density function µ. For T to be an isometry we must have
µ(z)[dz[ = λ(Tz)[T
(z)[[dz[ .
So
µ(z) =
2[T
(z)[
1 −[T(z)[
2
=
2
−2
(z+i)
2
1 −
z−i
z+i
2
=
1
Im(z)
.
The hyperbolic geodesics on the upper halfplane are half circles orthogonal to R, together with halflines
parallel to the imaginary axis.
11.2 Inversions
Let γ be a hyperbolic geodesic for D. Then γ is an arc of a circle Γ that is orthogonal to T.
Therefore inversion J in Γ maps D onto itself. We will call this inversion in γ.
Proposition 11.1 Inversions preserve the hyperbolic metric
Inversion in a hyperbolic geodesic preserves the hyperbolic metric.
Proof:
It is clear that the inversion C : z → z preserves the hyperbolic metric. Suppose that J is
inversion in a geodesic γ. Then Proposition 9.5 shows that C ◦ J is a M¨obius transformation. Both C
and J maps the unit disc D onto itself, so C◦J will also do so. Hence, it must be a hyperbolic isometry.
Therefore, J = C ◦ (C ◦ J) preserves hyperbolic lengths.
We can now consider the M¨obius transformations that we get by composing two inversions in
hyperbolic geodesics. The two geodesics may either cross at a point w ∈ D, or meet at a point w ∈ T,
or not meet at all in the closure of D. As at the end of '8 we see that the composition of these two
inversions is elliptic, parabolic or hyperbolic in these three cases.
Let T be a M¨obius transformation that maps D onto itself. Then Proposition 10.1 shows that
T : z →
az +b
az +b
for some a, b ∈ C with [a[
2
−[b[
2
= 1. This ﬁxes the points
iIm(a) ±
Re(a)
2
−1
b
.
The trace of the matrix is 2Re(a), so Corollary 9.2 shows that T can not be loxodromic. If T is elliptic,
then −1 < Re(a) < 1, one ﬁxed point w is in D and the other is 1/w. If T is parabolic, then Re(a) = ±1
and there is a single ﬁxed point on T. If T is hyperbolic, then Re(a) > 1 or Re(a) < −1 and there are
two ﬁxed points both on T.
Lecture 11 46
12 FUCHSIAN GROUPS
We can give any disc ∆ ⊂ P the hyperbolic metric and take it as a model for the hyperbolic plane.
The orientation preserving isometries for this metric are the M¨obius transformations that map ∆ onto
itself. A subgroup G of M¨ob(∆) is a Fuchsian group if it is discrete. In this section we will look at some
examples of Fuchsian groups and see that there is a very great variety.
We will study Fuchsian groups by looking at the orbits of points in the hyperbolic plane. These
orbits must be discrete, so their limit points will lie on the unit circle. These limit points form the limit
set of the group. We will also try to ﬁnd fundamental sets for the group acting on ∆. The images of
the fundamental set under the group G cover all of the hyperbolic plane and form a tessellation of the
plane. Often exhibiting this tessellation will be the simplest way to prove that the group is discrete.
12.1 Single generator Fuchsian groups
Proposition 12.1
A nonidentity M¨obius transformation T that maps a disc ∆ onto itself is one of the following:
(a) Elliptic with two ﬁxed points, one in ∆ and one in the complementary disc;
(b) Hyperbolic with two ﬁxed points, both on the boundary ∂∆ of ∆ in P;
(c) Parabolic with one ﬁxed point, which lies on ∂∆.
Proof:
We know, from Theorem 9.1 , that T has either one or two ﬁxed points in P.
Suppose that T has two ﬁxed points. Then T is conjugate to M
k
: z → kz for some k = 0, 1.
When [k[ = 1, the only discs M
k
maps onto themselves are ¦z ∈ P : [z[ < r¦ and ¦z ∈ P : [z[ > r¦.
Hence we are in case (a). When k > 0 (k = 1), the only discs mapped onto themselves are the half
planes ¦z ∈ P : Re(e
iθ
)z) > 0¦. Hence we are in case (b). (All other values for k give loxodromic
transformations and these map no disc onto itself.)
Suppose that T has only one ﬁxed point. Then T is conjugate to P : z → z + 1. The only discs
mapped onto themselves by P are the halfplanes ¦z ∈ P : Im(z) > c¦. Hence we are in case (c).
By using the conjugates M
k
for [k[ = 1, M
k
for k > 1 and P, it is simple to see when the group
G = 'T` generated by a transformation T is discrete.
Elliptic:
By conjugating we get M
k
: z → kz for some k with [k[ = 1. This is a rotation about 0 so the group it
generates is discrete when it is of ﬁnite order. The orbits in this case are ﬁnite and a fundamental set is
a sector of the unit disc. Consequently, the group generated by the elliptic transformation T is discrete
when T is of ﬁnite order and a fundamental set is a sector from the ﬁxed point bounded by two half
geodesics. The tessellation by images of this fundamental set are shown in the ﬁrst diagram below.
Hyperbolic:
By conjugating we get M
k
: z → kz for some k > 0 (k = 1). This generates an inﬁnite cyclic group
which is always discrete with the set ¦z : 1 [z[ < k¦ as a fundamental set. Hence the group generated
by the hyperbolic transformation T is discrete; the orbits are doubly inﬁnite sequences (T
n
z
o
)
n=∞
n=−∞
that converge to the two ﬁxed points of T; and a fundamental set is the region between two suitable
disjoint geodesics.
Parabolic:
By conjugating we get P : z → z + 1. This generates an inﬁnite cyclic group which is always discrete
Lecture 12 47
with the set ¦z : 0 Re(z) < 1¦ as a fundamental set. Hence the group generated by the parabolic
transformation T is discrete; the orbits are doubly inﬁnite sequences (T
n
z
o
)
n=∞
n=−∞
that converge to the
single ﬁxed point of T; and a fundamental set is the region between two suitable geodesics that both
end at the ﬁxed point.
Elliptic Hyperbolic Parabolic
In the case of a hyperbolic transformation T we call the hyperbolic geodesic joining the two ﬁxed
points the axis of T. It is clear that T maps its axis onto itself.
Exercise:
18. Show that a hyperbolic transformation T moves any point on its axis by a ﬁxed hyperbolic distance
along the axis. This distance is called the translation length of T. Show how to calculate the
translation length from the trace of a matrix that represents T.
It is also useful to use Proposition 9.5 and consider the M¨obius transformation as a composition
of two inversions in hyperbolic geodesics.
12.2 Triangle Groups
Let U be a “triangular” region bounded by three disjoint geodesics γ
1
, γ
2
, γ
3
. Let J
k
be inversion
in the geodesic γ
k
. Then the M¨obius transformations: A = J
2
◦ J
1
, B = J
3
◦ J
2
generate a group G.
Note that
J
1
◦ J
2
= A
−1
; J
2
◦ J
3
= B
−1
; J
3
◦ J
1
= B ◦ A ; J
1
◦ J
3
= A
−1
◦ B
−1
so any product of an even number of the inversions is an element of G. The set F that is the union of
the closure of U in D and J
2
(U) is a fundamental set for G. (The closure of U is a fundamental set
for the larger group generated by J
1
, J
2
and J
3
.) The quotient D/G is homeomorphic to a sphere with
three holes: a “pair of pants”.
G is a free group generated by A and B. For suppose that we can write g ∈ G as a product
g = A
k(1)
B
k(2)
A
k(3)
. . . B
k(N)
with each k(j) = 0 .
If k(1) > 0 (so the product for g begins with an A) then the copy g(F) of the fundamental set is separated
from F by the geodesic γ
1
. Similarly, if g begins with A
−1
(or B or B
−1
, then g(F) is separated from
F by J
2
(γ
1
) (or γ
3
or J
2
(γ
3
)). Hence we can only have g = I when the product for g is trivial and does
not begin with any of A, A
−1
, B, B
−1
.
Any orbit of G contains exactly one point in each copy g(F) of the fundamental set. The ﬁxed
points of each nonidentity element of G clearly give points in the limit set. The limit set is like a Cantor
set: It is a subset of ∂D but omits points in the open intervals where F meets ∂D. Similarly, it omits
points where A(F), B(F), AB(F), etc meet ∂D.
Lecture 12 48
F A(F) B(F)
B
−1
(F)
A
−1
(F)
AB
−1
(F)
AB(F)
A
2
(F)
BA
−1
(F)
B
2
(F)
BA(F)
B
−1
A(F)
B
−1
A
−1
(F) B
−2
(F)
A
−1
B(F)
A
−2
(F) A
−1
B
−1
(F)
A
B
The group G is certainly discrete. For suppose that (g
n
) is a sequence of nonidentity transforma
tions in G that converge to I. Then g
n
(z
o
) → z
o
as n → ∞. When we choose z
o
to be an interior point
of the fundamental set F we see that this is impossible.
Lecture 12 49
We can also consider the groups generated by inversions in the sides of a triangle where the sides
cross. Then the M¨obius transformations A, B are elliptic rather than hyperbolic. To get a Fuchsian
group these elliptic elements must be of ﬁnite order, so the angles of the triangle must be π/k for some
integers k. The sum of the angles of a hyperbolic triangle is strictly less than π, so there are many
triangles where these conditions are satisﬁed. For example we may construct a triangle with all angles
π/4 and obtain a tessellation as shown below.
The limit set in these cases is all of the unit circle.
Lecture 12 50
13 THE MODULAR GROUP
In many ways the most interesting triangle group is when the triangle has all of its angles 0. The
three vertices all lie on the boundary ∂D and the compositions of two inversions is parabolic. This group
is important in may branches of Mathematics. Rather than studying it directly, we will look at a closely
related group ﬁrst: the Modular group.
The Modular group M is the group of M¨obius transformations of the form
z →
az +b
cz +d
with a, b, c, d ∈ Z and ad −bc = 1. Hence M is the quotient PSL(2, Z) of the matrix group SL(2, Z) by
¦I, −I¦. The modular group acts as a group of hyperbolic isometries on the upper halfplane R
2
+
. This
group is certainly discrete because the integers are discrete. We wish to ﬁnd a fundamental set for the
modular group and the corresponding tessellation.
Consider the orbit Ω of a point w ∈ R
2
+
. For a transformation T : z →
az+b
cz+d
in the modular group
we have
Im(T(w)) = Im
(aw +b)(cw +d)
[cw +d[
2
=
1
[cw +d[
2
Im(w) .
Hence,
Im(T(w)) > Im(w) ⇔ 1 > [cw +d[ .
The points cw + d all lie in the lattice Zw + Z, so only a ﬁnite number can lie within the unit disc.
Hence there are only a ﬁnite number of pairs (c, d) with Im(T(w)) > Im(w). In particular, there are
only ﬁnitely many values of Im(z) for z ∈ Ω which are greater than Im(w). Consequently, there is a
point w
o
∈ Ω where Im(w
o
) is maximal.
There are always inﬁnitely many points in Ω where the imaginary part is greatest. For the matrix
1 1
0 1
is in SL(2, Z) and so z → z +1 is in the modular group. Consequently, w
o
+k is in Ω for each
k ∈ Z.
For each pair of coprime integers c, d we can ﬁnd integers a, b with ad − bc = 1, so the matrix
a b
c d
is in the modular group. Note that the point w
o
must satisfy [cw
o
+d[ 1 for each such pair
(c, d). This is the same as
w
o
+
d
c
1
c
so we see that w
o
must lie outside the shaded region below.
2 1 0 1 2
i
x
iy
S = {z : z − k 1 for all k ∈ Z}
This means that, for any point w ∈ R
2
+
, we can ﬁnd a point w
o
in the orbit of w that lies in the
unshaded region or on its boundary. Since z → z + 1 is in the modular group, we can choose w
o
to lie
in the region ¦z = x +iy ∈ R
2
+
: −
1
2
< x
1
2
, [z[ 1¦ shaded below.
Lecture 13 51
x
iy
2 1 0 1 2
i
This set: ¦z = x + iy ∈ R
2
+
: −
1
2
< x <
1
2
, [z[ > 1¦, together with part of the boundary is a
fundamental set for the modular group.
Proposition 13.1 Fundamental set for the modular group.
For every point z in R
2
+
, there is a transformation T in the modular group with T(z) lying in the set
F = ¦z = x +iy ∈ R
2
+
: −
1
2
x
1
2
, [z[ 1¦ .
Moreover, two points in F are in the same orbit for the modular group if and only if they are either
−
1
2
+iy and
1
2
+iy with y
1
2
√
3 or
ie
iθ
and ie
−iθ
with 0 θ
1
3
π
Proof:
Let A : z → z + 1 and B : z → −1/z. Both of these are in the modular group.
If z lies in the set
S = ¦w : [w −k[ 1 for all k ∈ Z¦
then there is an integer k with A
k
(z) = z + k ∈ F. If z lies outside this set, then we have shown that
there is an element z
in the orbit of z that lies within S.
Suppose that z and z
are in the same orbit and both lie within F. Then both have maximal
imaginary part for that orbit, so their imaginary parts are equal. Since they are in the same orbit, we
must have z
=
az +b
cz +d
for
a b
c d
∈ SL(2, Z). Since the imaginary parts of z and z
are equal, we
must have [cz + d[ = 1. Hence, either c = 0 and d = ±1 or c = ±1 and d = 0. In the ﬁrst case,
z
= z +k = A
k
(z) for some integer k and we are in the ﬁrst case of the proposition. In the second case,
z
= k −
1
z
= A
k
B(z) and we are in the second case.
It is useful to divide the fundamental set for the modular group into two. In the diagram below,
the set is divided into two triangles, one shaded and the other unshaded. The modular group permutes
Lecture 13 52
the shaded triangles and also permutes the unshaded ones. The inversions in the three sides of one of
these triangles gives a larger group with the modular group as an index two subgroup.
x
2 1 0 1 2
iy
Finally, the diagram below shows that there we can join together 6 of these triangles to form an
ideal triangle which has all three vertices on the boundary of R
2
+
. Inversions in the sides of this ideal
triangle give a subgroup of the group generated by inversions in sides of the smaller triangles.
x
2 1 0 1 2
iy
Lecture 13 53
14 HYPERBOLIC 3SPACE
In this section we will deﬁne a hyperbolic metric on the unit ball in R
3
so that the orientation
preserving isometries are the full group of M¨obius transformations. To do this, we ﬁrst deﬁne the
metric, then consider inversions, and ﬁnally show that every M¨obius transformation is the composition
of inversions.
14.1 The Hyperbolic Metric
Let B
3
be the unit ball ¦x ∈ R
3
: [[x[[ < 1¦ in Euclidean 3space. The hyperbolic density on B
3
is
λ(x) =
2
1 −[[x[[
2
.
The hyperbolic length of a smooth curve γ : [a, b] → B
3
is then
L(γ) =
b
a
λ(γ(t)) [[γ
(t)[[ dt =
b
a
2[[γ
(t)[[
1 −[[γ(t)[[
2
dt .
The hyperbolic metric ρ on B
3
is deﬁned by
ρ(x
o
, x
1
) = inf
¸
L(γ) : γ is a smooth curve in B
3
from x
o
to x
1
¸
.
A curve that attains this inﬁmum is a hyperbolic geodesic from x
o
to x
1
. The arguments used for the
hyperbolic metric on the unit disc ( Lemma 10.2 and Theorem 10.3 ) show that:
Proposition 14.1 Hyperbolic metric on B
3
The hyperbolic metric ρ is a metric on the unit ball B
3
. Moreover, the hyperbolic geodesic from the
origin 0 to any point x ∈ B
3
is a radial path with hyperbolic length log
1 +[[x[[
1 −[[x[[
.
The disc B
3
∩ ¦x : x
3
= 0¦ = ¦(x
1
, x
2
, 0) : [x
1
[
2
+[x
2
[
2
< 1¦ can be identiﬁed with the unit disc D
in C by letting (x
1
, x
2
, 0) correspond to x
1
+ix
2
. Then the hyperbolic on B
3
restricts to give the plane
hyperbolic metric on the D. We will see shortly that much more is true: the restriction of the hyperbolic
metric to the intersection of B
3
with a sphere orthogonal to the unit sphere gives a hyperbolic plane
metric.
14.2 Inversion
We are used to thinking of extending the complex plane C by adding a point ∞ and identify
ing the resulting space with the Riemann sphere P. We can also do this in higher dimensions. The
Ndimensional Euclidean space R
N
is extended by adjoining a point ∞ to obtain R
N
∞
. This is homeo
morphic to the unit sphere S
N
in R
N+1
and we can use stereographic projection to identify R
N
∞
with
S
N
. We will only consider the 3dimensional case although the results apply in higher dimensions.
As in the case of the Riemann sphere we will deﬁne circles to include straight lines, and spheres in
R
3
∞
to include planes. So a sphere in R
3
∞
is either of the form
S(c, r) = ¦x ∈ R
3
: [[x −c[[ = r¦
for c ∈ R
3
and r > 0, or else
Π(u, t) = ¦x ∈ R
N
: x u = t¦ ∪ ¦∞¦
for a unit vector u and t ∈ R. We call c and r the Euclidean centre and Euclidean radius of the sphere
S(c, r), while u is a unit normal to the plane Π(u, t).
We deﬁne inversion is a sphere Σ in R
3
∞
exactly as in '9.2 . Two distinct points x, x
are inverse
points for Σ if every circle orthogonal to Σ through x also passes through x
. Also, when x ∈ Σ, we say
that x and x itself are inverse points.
Lecture 14 54
Proposition 14.2 Inversion in spheres
For each sphere Σ in R
3
∞
and each point x ∈ R
3
∞
there is a unique point J(x) with x and J(x) inverse
points for Σ.
The map J is called inversion in Σ. It is an involution that reverses orientation.
Proof:
We already know from Proposition 9.4 that inversion in a circle maps the plane to itself. We
will use this to extend the result to higher dimensions.
Suppose ﬁrst that Σ = S(c, r). Any plane π through x and c cuts Σ in a circle σ and we know
from Proposition 9.4 that there is a point x
in π with x and x
inverse points for σ. Indeed we know
that the inverse point is given by
J(x) = c +
r
2
[[x −c[[
2
(x −c) .
This expression makes sense for any point x ∈ R
3
∞
, so we only need to show that it does have the
properties we require. (Note that we should interpret this formula as saying that J(c) = ∞ and
J(∞) = c.)
If γ is any circle through x that crosses Σ orthogonally, then there is a plane π through γ. In this
plane, we know that x and J(x) are inverse points for the circle σ = Σ∩π in the plane π. Consequently,
γ must pass through J(x).
An entirely similar argument applies when Σ is a plane Π(u, t). Then we have
J(x) = x + 2(t −x u)u .
We already know a lot about inversion in 2dimensions. Since inversion in higher dimensions is
deﬁned in terms of inversions in 2dimensions, we readily obtain a variety of results about inversion.
Proposition 14.3 Inversion preserves spheres.
Let J be inversion in a sphere Σ. Then J maps any sphere U in R
N
∞
onto another sphere.
Proof:
Let Σ be the sphere S(c, r) and U the sphere S(d, s). Let be the straight line through c and
d. Then any plane π through cuts Σ is a circle σ and U in a circle u. We know that inversion in the
circle σ sends u to another circle, so we see that J(U) cuts π in a circle. This is true for every plane π
through , so J(U) must be a sphere.
A similar but simpler argument applies when Σ or U are planes.
Corollary 14.4 Inversion preserves circles.
Let J be inversion in a sphere Σ. Then J maps any circle γ in R
N
∞
onto another circle.
Proof:
The circle γ is the intersection of two spheres. Inversion maps each of these to another sphere
and the intersection of these two spheres is again a circle.
Proposition 14.5 Inversion preserves angles.
Let J be inversion in a sphere Σ. If two curves α and β in R
3
∞
cross at an angle θ, then J(α) and J(β)
also cross at an angle θ.
Lecture 14 55
Proof:
Let J be inversion in the sphere S(c, r). We wish to calculate the derivative J
(x). To simplify
the algebra, translate and enlarge the sphere so that it becomes the unit sphere S(0, 1). Note that this
does not alter any angles.
Now J : x →
1
x
2
x satisﬁes
J(x +h) =
1
[[x[[
2
+ 2x h +[[h[[
2
(x +h)
= J(x) +
1
[[x[[
2
h −
2x h
[[x[[
2
x
+o(h) .
So we see that J is diﬀerentiable at x with
J
(x) : h →
1
[[x[[
2
h −
2x h
[[x[[
2
x
.
This is a scalar multiple of reﬂection in the plane orthogonal to x and so preserves angles (and reverses
orientation).
Exercise:
19. Show from the formula
J(x) = c +
r
2
[[x −c[[
2
(x −c) .
for inversion in the sphere S(c, r) that inversion maps a sphere to another sphere.
Lecture 14 56
15 EXTENDING M
¨
OBIUS TRANSFORMATIONS TO HYPERBOLIC SPACE
15.1 Inversions and the hyperbolic metric
Let Σ be a sphere that is orthogonal to the unit sphere S
2
in R
3
. Then Σ meets S
2
in a circle σ.
Proposition 14.3 shows that inversion J in Σ ﬁxes σ and maps the unit sphere to another sphere.
Proposition 14.5 shows that this image sphere is also orthogonal to Σ. Hence J maps the unit sphere
onto itself. Therefore, J maps the ball B
3
onto itself.
Conversely, any circle σ on the unit sphere S
2
is the intersection of S
2
with a sphere Σ orthogonal
to S
2
.
Throughout this section we will only be concerned with spheres Σ orthogonal to the unit sphere,
and the circle σ = Σ ∩ S
2
where it meets the unit sphere.
Proposition 15.1 Inversions are hyperbolic isometries.
Let J be inversion in a sphere Σ orthogonal to the unit sphere S
2
. Then J is an isometry for the
hyperbolic metric but reverses orientation.
Proof:
Let x be a point of B
3
and choose any plane π through the origin and x. Then π intersects Σ
in a circle γ and the map J acts on π as inversion in γ. Proposition 11.1 shows that inversion in γ is
an isometry for the hyperbolic metric on the disc π ∩ B
3
. Hence J must be an isometry at x.
We have already seen that the derivative of J is orientation reversing.
Exercise:
20. Let J be inversion in a sphere Σ and Q inversion in the unit sphere S
2
. Show that Σ is orthogonal
to S
2
if and only if J ◦ Q = Q◦ J.
Lemma 15.2
For each point a ∈ B
3
there is an inversion J in a sphere orthogonal to S
2
that interchanges the origin
and a.
Proof:
If a = 0, then inversion in any plane through the origin (that is reﬂection in such a plane) will
do.
Otherwise, let Q(a) =
a
[[a[[
2
and set
Σ =
x : [[x −Q(a)[[ =
1 −[[a[[
2
[[a[[
¸
.
Then it is simple to check that inversion in Σ maps 0 to a.
Proposition 15.3 Hyperbolic geodesics
Between any two distinct points of B
3
there is an unique path with shortest hyperbolic length. This is
a section of a circle orthogonal to the unit sphere S
2
.
Lecture 15 57
Proof:
Suppose that a and b are two distinct points in the ball B
3
. We already know that the radial
path is the shortest path from a to b when a = 0 because of Proposition 14.1 .
For any other value of a, the lemma shows that there is an inversion J in a sphere orthogonal to
S
2
with J(a) = 0. Since this inversion is a hyperbolic isometry, we see that the shortest path from a
to b must be the image of a radial path under J. Now Corollary 14.4 completes the proof.
Suppose now that T is a M¨obius transformation. This maps the unit sphere S
2
to itself. We
know from '9 that T can be written as the composition of inversions in an even number of circles, say
σ
1
, σ
2
, . . . , σ
N
. For each of these circles, there is a sphere Σ
n
orthogonal to S
2
which intersects S
2
in the
circle σ
n
. Let J
n
be inversion in the sphere Σ
n
. Then the composition J
N
◦ J
N−1
◦ . . . ◦ J
3
◦ J
2
◦ J
1
acts
on the sphere S
2
as the M¨obius transformation T. However, this composition gives a map
¯
T from all
of R
3
∞
to itself. This map agrees with T on the sphere S
2
and maps the ball B
3
onto itself. So we have
a way to extend the M¨obius transformation T to the ball. This extension was introduced by Poincar´e
and is often called the Poincar´e extension of the M¨obius transformation.
This extension is unique. For suppose that x is any point of B
3
. Choose a circle γ through x that
crosses S
2
orthogonally at two points a and b. Then each inversion J in a sphere orthogonal to S
2
maps γ to another circle orthogonal to S
2
. Consequently,
¯
T(x) lies on the circle orthogonal to S
2
that
joins the two points T(a) and T(b) in S
2
. This is true for every choice of γ so we see that the point
¯
T(x) is completely determined by the map T acting on the sphere. In particular, if we write T as a
composition of inversions in two diﬀerent ways we must obtain the same extension
¯
T.
Proposition 15.4 Extensions of M¨obius transformations.
For every M¨ obius transformation T : S
2
→ S
2
the extension
¯
T : R
3
∞
→R
3
∞
maps the unit ball B
3
onto
itself and is an orientation preserving isometry for the hyperbolic metric.
Proof:
Proposition 15.1 shows that each inversion J
n
is an isometry for the hyperbolic metric, hence
the composition
¯
T is also. Since T is the composition of an even number of inversions, we see that
¯
T
is the composition of an even number of the orientation reversing isometries J
n
. Hence
¯
T is orientation
preserving.
Now we will prove that every orientation preserving isometry of B
3
for the hyperbolic metric is an
extension of a M¨obius transformation.
Theorem 15.5 M¨obius transformations as isometries of hyperbolic 3space.
Every orientation preserving isometry of hyperbolic 3space B
3
is
¯
T for some M¨obius transformation
T : S
2
→ S
2
.
Proof:
Suppose that A : B
3
→ B
3
is an orientation preserving isometry for the hyperbolic metric
on B
3
. Then A(0) ∈ B
3
so Lemma 15.2 gives an inversion J in a sphere orthogonal to S
2
with
J(A(0)) = 0. Hence, A
= J ◦ A is an isometry of B
3
that ﬁxes 0.
For each unit vector u ∈ S
2
we know that the path
γ
u
: [0, 1) → tu
is a hyperbolic geodesic with ρ(0, γ
u
(t)) = log(1 +t)/(1 −t). Hence, A
◦ γ
u
must also be a hyperbolic
geodesic. Since it starts at the origin, we must have
A
(tu) = tv
Lecture 15 58
for some unit vector v. Write v = α(u).
Now observe that
lim
t→0+
ρ(tu
1
, tu
2
)
t
= 2[[u
1
−u
2
[[
for any unit vectors u
k
. Since A
is an isometry, this shows that
[[α(u
1
) −α(u
2
)[[ = [[u
1
−u
2
[[ .
It follows that α is an orthogonal linear map in O(3). This shows that A
: R
3
∞
→R
3
∞
is orthogonal.
Finally we know that every orthogonal linear map is the composition of reﬂections in plane through
the origin. These are inversions, so we see that A
, and hence A, is the composition of inversions in
spheres orthogonal to S
2
.
We have now proved that the groups M¨ob and Isom
+
(B
3
) are isomorphic with a M¨obius transfor
mation T corresponding to the extension
¯
T. This means that the M¨obius group M¨ob acts on hyperbolic
3space B
3
. Usually we will not distinguish between T and its extension
¯
T.
15.2 The upper halfspace.
We can redo the arguments above for any ball in R
3
∞
and obtain a hyperbolic metric on the ball
for which the orientation preserving isometries are the M¨obius transformations. The most important
example is when the ball is the upper halfspace:
R
3
+
= ¦(x
1
, x
2
, x
3
) ∈ R
3
: x
3
> 0¦ .
The boundary of this is the extended complex plane C
∞
= R
2
∞
. We can show that any M¨obius
transformation acting on this boundary extends to an orientation preserving isometry of the upper
halfspace for the hyperbolic metric with density:
λ(x) =
1
x
3
.
We can also deduce the results for the upper halfspace directly from those for the ball B
3
for
inversion in the sphere
Σ =
x : [[x +e
3
[[ =
√
2
¸
where e
3
= (0, 0, 1)
maps the upper halfspace onto the ball and vice versa.
Lecture 15 59
16 ISOMETRIES OF H
3
We have seen how to put a hyperbolic metric on the unit ball B
3
in R
3
or the upper halfspace
R
3
+
. We will denote both of these by H
3
and call them hyperbolic 3space. The orientation preserving
isometries for hyperbolic 3space have been identiﬁed with the group of M¨obius transformations acting
on the boundary ∂H
3
. In this section we wish to study these isometries in more detail.
16.1 Examples in Hyperbolic Geometry
The hyperbolic metric is given by:
ds =
2
1 −[[x[[
2
[[dx[[ on B
3
and ds =
1
x
3
[[dx[[ on R
3
+
.
The hyperbolic geodesics are the arcs of circles orthogonal to the boundary ∂H
3
.
The plane ¦x : x
3
= 0¦ meets the ball B
3
in the unit disc and the hyperbolic metric on B
3
restricts
to the hyperbolic plane metric on this disc. A similar result holds for the intersection of any other sphere
orthogonal to ∂H
3
with H
3
. For suppose that Σ is the intersection with B
3
of a sphere orthogonal to
∂B
3
. Then the boundary of Σ is a circle σ ⊂ ∂B
3
. We know that there is a M¨obius transformation T
that maps this circle σ to the unit circle T. Since T acts isometrically on B
3
, it must map Σ to the
unit disc. Hence Σ with the hyperbolic metric is isometric to the hyperbolic plane. We call such an
intersection of a sphere orthogonal to ∂H
3
with H
3
a hyperbolic plane in H
3
.
Two hyperbolic planes that meet in H
3
intersect in a hyperbolic geodesic. We think of geodesics
as the straight lines for hyperbolic geometry.
In order to develop our sense of what hyperbolic 3space is like, we will prove a series of simple
results. In all of these we choose whichever model (B
3
or R
3
+
) is easiest and apply isometries to make
the calculations simple.
A ball ¦x ∈ H
3
: ρ(x, c) < ρ
o
¦ is a Euclidean ball which lies entirely within H
3
. Note that the
Euclidean centre will not normally be c nor the Euclidean radius ρ
o
.
Take H
3
as B
3
and consider ﬁrst the case where c = 0. Then Proposition 14.1 shows that
ρ(x, 0) = log
1 +[[x[[
1 −[[x[[
.
Hence the ball is the set ¦x : [[x[[ < tanh
1
2
ρ
o
¦. Each M¨obius transformation is a composition of
inversions and these map balls to balls, so the result continues to hold for any c ∈ B
3
.
There is a unique point of a geodesic γ closest to a point c in H
3
.
c a
γ
Lecture 16 60
Take H
3
to be B
3
and c = 0. The geodesic γ is then the arc of a circle orthogonal to the unit sphere
∂B
3
. This circle has a unique point a with smallest norm and, by the previous remark, this is closest
hyperbolically to 0. Note that the shortest hyperbolic path from 0 to a is a radial line. So we see that
the hyperbolic geodesic from a point c to the closest point of a hyperbolic geodesic γ is orthogonal to γ.
For two geodesics α, β in H
3
which do not have a common endpoint on ∂H
3
there are
unique points a ∈ α and b ∈ β with ρ(a, b) minimal.
By applying an isometry we may assume that H
3
= R
3
+
and α is the geodesic ¦(0, 0, x
3
) : x
3
> 0¦.
Consider the point x = (sin θ, 0, cos θ) ∈ R
3
+
. The shortest path from x to α is the arc
¦(sin φ, 0 cos φ) : 0 φ θ¦ .
This has hyperbolic length
ρ
o
=
θ
0
1
cos φ
dφ .
Now make the change of variables t = tan
1
2
φ, so that cos φ =
1−t
2
1+t
2
and
dt
dφ
=
1
2
(1 +t
2
). This gives:
ρ
o
=
tan
1
2
θ
0
1 +t
2
1 −t
2
2
1 +t
2
dt =
tan
1
2
θ
0
2
1 −t
2
dt = tanh
−1
(tan
1
2
θ) .
Now the M¨obius transformation T : z → kz (k > 0) acts on R
3
+
as
T : (x
1
, x
2
, x
3
) → (kx
1
, kx
2
, kx
3
)
and this must be an isometry. So all of the points (k sin θ, 0, k cos θ) are at a hyperbolic distance ρ
o
from
α. Similarly, T : z → e
iψ
z acts on R
3
+
as
T : (x
1
+ix
2
, x
3
) → (e
iψ
(x
1
+ix
2
), x
3
)
and this must be an isometry. Hence all of the points
¦x ∈ R
3
+
: x
3
= [[x[[ cos θ¦
are at a hyperbolic distance ρ
o
= tanh
−1
tan
1
2
θ
from α. This is a cone about the axis α.
0
α
a
θ
x
Exercise:
21. Draw the set of points that lie within a ﬁxed hyperbolic distance ρ
o
of a geodesic α in the unit disc
D and in the unit ball B
3
.
Lecture 16 61
Now suppose that β is a geodesic in R
3
+
joining points of ∂R
3
+
neither of which is 0 or ∞. Then
there will be a unique point b ∈ β with the angle
sin
−1
x
3
[[x[[
minimal. This is the point closest to α. It is clear that the shortest path from α to β is a geodesic that
crosses both α and β orthogonally.
α
β
γ
If two geodesics have a common endpoint, then they contain points that are arbitrarily
close.
Let α and β be geodesics in R
3
+
with ∞ as their common endpoint. Then
α = ¦(a
1
, a
2
, x
3
) : x
3
> 0¦ ; β = ¦(b
1
, b
2
, x
3
) : x
3
> 0¦ .
Now it is clear that
ρ((a
1
, a
2
, x
3
), (b
1
, b
2
, x
3
))
[[(a
1
, a
2
) −(b
1
, b
2
)[[
x
3
so there are points of the two geodesics that are arbitrarily close together.
Two geodesics α and β in H
3
are either:
(a) Parallel: when they have a common endpoint.
(b) Cross: when they intersect at a point of H
3
.
(c) Skew: when they have no common endpoint and do not intersect.
We have shown that two skew lines have a common normal joining the closest points of the two lines.
In case (b) there is also a common normal through the point of intersection. However, there is no such
normal for parallel lines unless the lines are identical.
16.2 Axes of Isometries
Let T be a M¨obius transformation. When we think of T acting on the Riemann sphere, it has two
ﬁxed points, say a and b. When we think of T acting on the hyperbolic 3space H
3
, then there is a
geodesic α joining a and b. T maps this geodesic to itself since it ﬁxes the endpoints. We call α the axis
of the M¨obius transformation.
The M¨obius transformation T acts isometrically on H
3
, so T must move the points of the axis a
ﬁxed hyperbolic distance. This is called the translation length of T. It is 0 for elliptic transformations
but nonzero for hyperbolic and loxodromic transformations.
Lecture 16 62
Exercise:
22. Show that the translation length of the transformation M
k
: z → kz is log [k[. Hence show how to
ﬁnd the translation length of the M¨obius transformation
z →
2z + 1
5z + 3
.
A parabolic transformation does not have an axis. If there were any geodesic that were mapped
onto itself, preserving direction, then both endpoints would be ﬁxed. This can not occur if there is just
one ﬁxed point on ∂H
3
.
Lecture 16 63
17 INVOLUTIONS
A M¨obius transformation R which has ﬁnite order must be elliptic. In particular, an involution
which has R
2
= I must be elliptic. We can conjugate R so that its ﬁxed points are at 0, ∞ ∈ R
3
+
. The
axis α of R is then the positive x
3
axisjoining 0 to ∞. Then R : z → −z on P and
R : (x
1
, x
2
, x
3
) → (−x
1
, −x
2
, x
3
) for x ∈ R
3
+
.
This ﬁxes every point of the axis.
Note that, if π is a hyperbolic plane that contains the axis α of the involution R, then R maps this
plane to itself but interchanges the halfspaces on either side of it. On the plane π, the involution acts
as inversion in the geodesic α. In '11 we saw that every M¨obius transformation of the disc could be
written as the composite of two inversions. Here we will prove that each M¨obius transformation on P
can be written as the composite of two involutions.
Proposition 17.1 Isometries of H
3
are compositions of two involutions.
Every M¨obius transformation can be expressed as R
2
◦ R
1
for two elliptic involutions R
1
, R
2
.
Proof:
The identity is R
2
for every involution R.
Suppose that P is parabolic. Then we may assume that it is P : z → z + 1. This acts as
P : (x
1
, x
2
, x
3
) → (x
1
+ 1, x
2
, x
3
)
on R
3
+
. For this, take
R
1
: z → −z and R
2
: z → 1 −z .
Then R
1
, R
2
are involutions with P = R
2
◦ R
1
. Note that the axes of R
1
, R
2
are
¦(0, 0, x
3
) : x
3
> 0¦ and ¦(
1
2
, 0, x
3
) : x
3
> 0¦
which have a common endpoint at the ﬁxed point of P.
Suppose that T is a M¨obius transformation with 2 ﬁxed points. We may assume that these are 0
and ∞. So T : z → λ
2
z for some λ = 0. Then T acts on R
3
+
as
T : (x
1
+ix
2
, x
3
) → (λ
2
(x
1
+ix
2
), [λ
2
[x
3
) .
For this, take
R
1
: z →
1
z
and R
2
: z →
λ
2
z
.
These are involutions with T = R
2
◦ R
1
. Note that the axes of R
1
, R
2
are
¦(cos θ, 0, sin θ) : 0 < θ < π¦ and ¦(λcos θ, 0, [λ[ sin θ) : 0 < θ < π¦ .
These are identical when λ = ±1 and T = I. They cross at (0, 0, 1) when [λ[ = 1 and T is elliptic.
Otherwise, they are skew and T is loxodromic or hyperbolic with its axis normal to both.
Proposition 17.2
Let R
1
, R
2
be involutions with axes α
1
, α
2
in H
3
. Then
(a) If α
1
= α
2
then R
2
◦ R
1
= I.
(b) If α
1
and α
2
are parallel, then R
2
◦ R
1
is parabolic with the common endpoint of α
1
and α
2
as its
ﬁxed point on ∂H
3
.
(c) If α
1
, α
2
are skew, then R
2
◦ R
1
is loxodromic or hyperbolic with axis normal to both α
1
and α
2
.
Lecture 17 64
Proof:
(a) is obvious. For (b) conjugate so that in R
3
+
we have:
α
1
= ¦(a
1
, x
3
) : x
3
> 0¦ and α
1
= ¦(a
2
, x
3
) : x
3
> 0¦
for some a
1
, a
2
∈ C. Then
R
1
: z → 2a
1
−z and R
2
: z → 2a
2
−z
and R
2
◦ R
1
: z → 2(a
2
−a
1
) +z is parabolic with its single ﬁxed point at ∞.
For (c), we know that there is a unique geodesic γ normal to both α
1
and α
2
. Conjugate so this is
the x
3
axis in R
3
+
. The geodesic α
j
is then a halfcircle that crosses the x
3
axis orthogonally, so it must
join points ±w
j
. Consequently,
R
j
: z →
w
2
j
z
and R
2
◦ R
1
: z → (w
2
2
/w
2
1
)z is either hyperbolic or loxodromic.
Exercise:
23. Let R
1
, R
2
be involutions with axes α
1
, α
2
in H
3
. Show that R
2
◦ R
1
is hyperbolic when both α
1
and α
2
lie in a hyperbolic plane.
Finally, let us look at the group G = 'T` generated by a single orientation preserving isometry T
of H
3
. The last proposition shows that we can write T as the composite R
2
◦ R
1
of two involutions with
axes α
1
, and α
2
. We will concentrate on the case where T is loxodromic or hyperbolic, since the other
cases are simpler. There is then a common normal γ to α
1
and α
2
and this is the axis for T. If we
parametrise γ by hyperbolic length, then we can assume that γ(0) is the point where α
1
meets γ and
γ(τ) is the point where α
2
meets γ. This means that
R
1
(γ(t)) = γ(−t) ; R
2
(γ(t)) = γ(2τ −t)) ; T(γ(t)) = R
2
(R
1
(γ(t))) = γ(2τ +t) .
So the translation length of T is 2τ: twice the distance between the axes α
1
and α
2
.
Each point x ∈ H
3
is closest to some point of γ and those points which are closest to γ(t) ﬁll a
hyperbolic plane π(t). Since T is an isometry, it maps π(t) onto π(2τ +t). Hence the set
F =
¸
¦π(t) : 0 t < 2τ¦
is a fundamental set for the group G = 'T`. T maps the plane π(0) that bounds one side onto the plane
π(2τ) that bounds the other.
α
1
α
2
γ
Lecture 17 65
Note also that half of F:
¸
¦π(t) : 0 t τ¦
is a fundamental set of the larger group 'R
1
, R
2
`.
When T is parabolic, there is a fundamental set bounded by two hyperbolic planes that touch at
the ﬁxed point of T on ∂H
3
. When T is elliptic of ﬁnite order, there is a fundamental set bounded by
two hyperbolic halfplanes that meet on the axis of T.
Exercise:
24. Suppose that T is a M¨obius transformation that maps the unit disc D onto itself. Then T also acts
as an isometry of the hyperbolic 3space B
3
. How are fundamental sets for G = 'T` acting on D
related to fundamental sets for G acting on B
3
?
Lecture 17 66
18 KLEINIAN GROUPS
M¨obius transformations are represented by 2 2 complex matrices so a group of M¨obius transfor
mations is discrete if it is a discrete subset of the set of all 22 matrices. A Kleinian group is a discrete
subgroup of M¨ob. We will think of these groups acting as isometries of the hyperbolic 3space H
3
.
Every Fuchsian group is certainly discrete when we think of it as a subgroup of M¨ob rather than a
subgroup of M¨ob(D). We will see many more examples later.
18.1 Finite Kleinian Groups
Any ﬁnite subgroup of M¨ob is certainly a Kleinian group. However, we will show that these ﬁnite
groups are all conjugate to ﬁnite subgroups of SO(3). So we already know which groups can arise: cyclic
and dihedral groups together with the tetrahedral, octahedral and icosahedral groups.
Lemma 18.1
Let S be a nonempty ﬁnite subset of H
3
. Then there is a unique closed hyperbolic ball B(c, ρ) of
smallest hyperbolic radius that contains S.
Proof:
Set ρ
o
be the inﬁmum of the radii ρ for which there is a centre c ∈ H
3
with S ⊂ B(c, ρ).
Then there is a sequence ρ
n
that decreases to ρ
0
and a sequence of centres c
n
with S ⊂ B(c
n
, ρ
n
). The
centres c
n
all lie within ρ
1
of any chosen point of S, so we can ﬁnd a convergent subsequence using the
Bolzano – Weierstrass Theorem. We will assume that c
n
→ c as n → ∞. Then S ⊂ B(c, ρ
o
). This
shows that there is at least one closed ball of smallest radius containing S.
Now suppose that there are two diﬀerent closed balls with smallest radius ρ
0
which contain S, say
B(c
+
, ρ
o
) and B(c
−
, ρ
o
). Let m be the midpoint of the hyperbolic geodesic from c
+
to c
−
. We can
apply an isometry to H
3
to move m to the origin. Then c
+
= −c
−
. In the picture below, S must lie
in the intersection of the two balls B(c
+
, ρ
o
) and B(c
−
, ρ
o
), so it must lie in the shaded region. This is
contained in the closed ball centred on m with radius ρ(m, A). This radius is less than ρ
o
, which is a
contradiction.
c
+
c
−
m
A
Proposition 18.2 Finite Kleinian groups are conjugate to subgroups of SO(3).
Every ﬁnite subgroup of M¨ob is conjugate in the M¨obius group to a subgroup of SO(3).
Lecture 18 67
Proof:
We will consider the ﬁnite group G acting on the unit ball B
3
. It has a ﬁnite orbit Ω = G(x) for
any point x ∈ B
3
. The lemma shows that this is contained within a unique smallest closed hyperbolic
ball, say B(c, ρ
o
).
Each T ∈ G acts isometrically on B
3
and permutes the elements of the orbit Ω. So
Ω = T(Ω) ⊂ T(B(c, ρ
o
)) = B(T(c), ρ
o
) .
This implies that T(c) = c, so every element of G ﬁxes c.
Now conjugate by a M¨obius transformation that maps c to the origin. Then G becomes a group of
hyperbolic isometries that ﬁx the origin. These must be elements of SO(3).
This proposition shows that we need to consider inﬁnite Kleinian groups in order to obtain new
and interesting examples of such groups. To do this we need to think more carefully about the action
of a Kleinian group on hyperbolic 3space.
18.2 Discontinuous Action
Let G be a subgroup of M¨ob = Isom
+
(H
3
). The group G acts discontinuously at x
o
∈ H
3
if there
is some δ > 0 for which ¦T ∈ G : ρ(x
o
, T(x
o
)) < δ¦ is ﬁnite. The group G acts discontinuously on H
3
if it acts discontinuously at each point of H
3
.
If G acts discontinuously at x
o
, then the stabilizer: Stab(x
o
) = ¦T ∈ G : T(x
o
) = x
0
¦ is a ﬁnite
group and so conjugate to a ﬁnite subgroup of SO(3).
Lemma 18.3
Let G act discontinuously at a point x
0
∈ H
3
. Then, for any compact set K ⊂ H
3
, the set ¦T ∈ G :
T(K) ∩ K = ∅¦ is ﬁnite.
Proof:
Since K is compact, it is certainly bounded, so there is a ρ
o
with K ⊂ B(x
o
, ρ
o
). If K∩T(K) =
∅, then we can ﬁnd a ∈ K with T(a) ∈ K. So
ρ(T(x
o
), x
o
) ρ(T(x
o
), T(a)) +ρ(T(a), x
o
) = ρ(x
o
, a) +ρ(T(a), x
o
) < 2ρ
o
.
Suppose that there were inﬁnitely many such elements T of G. Then the Bolzano – Weierstrass theorem
shows that we can ﬁnd a sequence of distinct elements (T
n
) with T
n
(x
o
) converging to some point
y ∈ H
3
as n → ∞.
In particular, there is a natural number N with ρ(T
n
(x
o
), y) <
1
2
δ for n N. Consequently the
elements S
n
= T
−1
N
◦ T
n
of G satisfy
ρ(S
n
(x
o
), x
o
) = ρ(T
n
(x
o
), T
N
(x
o
)) <
1
2
δ +
1
2
δ = δ
for n N. This means that there are inﬁnitely many S
n
∈ G with ρ(S
n
(x
o
), x
o
) < δ which contradicts
G acting discontinuously at x
o
.
Suppose that F is a fundamental set for G acting on H
3
. Then the copies T(F) for T ∈ G tessellate
all of H
3
. If a neighbourhood of x
o
meets only ﬁnitely many copies of F, then G acts discontinuously at
x
o
. This means it is often fairly straightforward to show that a group acts discontinuously by exhibiting
a suitable fundamental set.
Lecture 18 68
For example, we can construct a regular dodecahedron in H
3
with each pentagonal face having each
angle a rightangle and with the angles between faces being rightangles. The copies of this dodecahedron
obtained by reﬂecting (inverting) in the faces tessellates all of H
3
. The group of M¨obius transformations
that are symmetries of this tessellation clearly acts discontinuously on H
3
.
Dodecahedral tessellation of H
3
. (See http://www.geom.uiuc.edu/graphics/pix/Special Topics/ .)
We will now show that acting discontinuously on H
3
is equivalent to being a discrete group of
M¨obius transformations.
Theorem 18.4 Discrete if and only if acts discontinuously.
A group of M¨obius transformations is discrete if and only if it acts discontinuously on H
3
.
Proof:
Suppose ﬁrst that G is not discrete. Then there is a sequence (T
n
) of nonidentity transforma
tions in G with T
n
→ I as n → ∞. This implies that T
n
(x
o
) → x
o
for any point x
o
∈ H
3
. So G does
not act discontinuously at x
o
.
Suppose now that G does not act discontinuously at some point x
o
∈ H
3
. Then there is a δ > 0
with inﬁnitely many T ∈ G satisfying ρ(x
o
, T(x
o
)) < δ. It is more convenient to work on the upper
halfspace, so conjugate so that x
o
is mapped to k = (0, 0, 1) ∈ R
3
+
.
For each a ∈ R
3
+
deﬁne S
a
: x → a
3
x + (a
1
, a
2
, 0). Then S
a
(k) = a. (This corresponds to the
M¨obius transformation on P given by z → a
3
z +(a
1
+ia
2
).) Use the Bolzano – Weierstrass theorem to
ﬁnd a sequence of distinct transformations T
n
∈ G with T
n
(k) → y as n → ∞. Set
S
n
= S
T
n
(k)
and R
n
= S
−1
n
◦ T
n
.
Lecture 18 69
(Note that S
n
and R
n
need not be in G.) Then we have R
n
(k) = k so each R
n
is in the group Stab(k).
This stabilizer is conjugate to the group SO(3), which is compact. Hence we can ﬁnd a subsequence
R
n
which converges to a transformation R
∞
as n
→ ∞.
Now we have T
n
(k) → y, so S
n
→ S
y
as n → ∞. Hence, on the subsequence, we have
T
n
= S
n
◦ R
n
→ S
y
◦ R
∞
as n → ∞ .
This shows that G is not discrete.
We can also consider groups acting discontinuously on other spaces and prove similar results.
Look at: http://www.plunk.org/∼hatch/HyperbolicApplet/ .
Lecture 18 70
19 LIMITS OF ORBITS
Let G be a Kleinian group acting on the ball B
3
and choose a point x
o
∈ B
3
. Then we have shown
that G acts discontinuously on H
3
= B
3
. So only ﬁnitely many points of the orbit G(x
o
) lie within
any hyperbolic ball. This implies that [[T(x
o
)[[ tends to 1 as T runs through G. We can think of the
points G(x
o
) as lying inside the space R
3
with the Euclidean metric (or, better still, inside R
3
∞
with
the chordal metric) and consider the limit points of the orbit.
Recall that a limit point of a set Ω ⊂ R
3
is a point u for which there is a sequence (w
n
) of distinct
points w
n
∈ Ω which converge to u, so [[w
n
− u[[ → 0 as n → ∞. The set of limit points of the orbit
G(x
o
) is called the limit set of G and will be denoted by Λ(x
o
). We will see shortly that this limit set
is independent of the point x
o
we choose.
Note that when we talk about limit points of the orbit we are using the Euclidean (or, chordal)
metric. The hyperbolic metric is not deﬁned on the boundary so it does not make sense to ask for limit
points in the hyperbolic metric.
Proposition 19.1 The limit set is closed and Ginvariant.
For any Kleinian group acting on B
3
the limit set Λ(x
o
) is a closed, Ginvariant subset of ∂H
3
= S
2
.
Proof:
Λ(x
o
) is obviously closed since it is G(x
o
) ` B
3
. Since the orbit Ω = G(x
o
) satisﬁes T(Ω) = Ω
for each T ∈ G, and T is continuous on all of R
3
∞
, we see that T(Λ(x
o
)) = Λ(x
o
). So the limit set is
Ginvariant.
Suppose that T is a loxodromic or hyperbolic transformation in the group G. Then T
n
(x
o
) tends
to one of the ﬁxed points of T as n → ∞ and T
−n
(x
o
) tends to the other. Hence, both the ﬁxed points
of T are in the limit set for G. Similarly the ﬁxed point of a parabolic transformation is in the limit set.
However the ﬁxed points of an elliptic transformation need not lie in the limit set.
Exercise:
25. Give an example of an elliptic element of a Kleinian group with ﬁxed points that do not lie in the
limit set. Give an example of a Kleinian group for which the limit set is empty.
26. Let G be a Kleinian group with an invariant disc ∆ ⊂ P. Show that the limit set of G is a subset
of ∂∆.
We will now show that Λ(x
o
) is independent of the point x
o
. For suppose that x
1
is another
point of B
3
. Then ρ(x
o
, x
1
) is ﬁnite. Each T ∈ G is an isometry for the hyperbolic metric, so
ρ(T(x
o
), T(x
1
)) = ρ(x
o
, x
1
). Although these points T
n
(x
o
), and T
n
(x
1
) stay the same hyperbolic
distance apart they get closer together for the Euclidean metric as the points get closer to the boundary.
Hence T
n
(x
o
) and T
n
(x
1
) will converge to the same point of ∂B
3
.
Lemma 19.2
Let (x
n
) and (y
n
) be two sequences of points in B
3
with ρ(x
n
, y
n
) K for each n ∈ N. If the sequence
of points (x
n
) in B
3
converges in the Euclidean metric to a limit point u ∈ ∂B
3
, then the sequence
(y
n
) will also converge to u for the Euclidean metric.
Proof:
The hyperbolic density at a point x is
λ(x) =
2
1 −[[x[[
2
= 2 cosh
2 1
2
ρ(0, x) .
So, for any given ε > 0, we can ﬁnd ρ
o
with
λ(x) >
1
ε
for ρ(0, x) > ρ
o
−K .
Lecture 19 71
Suppose that γ is a hyperbolic geodesic joining the points a, b. Then
ρ(a, b) = L(γ) =
γ
λ(x) [[dx[[ inf ¦λ(x) : x ∈ γ¦ [[b −a[[ .
Hence, if ρ(a, b) K and ρ(0, a) > ρ
o
, then
K ρ(a, b)
1
ε
[[b −a[[ .
Applying this to the pairs of points x
n
, y
n
gives the result.
Proposition 19.3 Limit set is independent of the base point.
The limit sets Λ(x
o
) and Λ(x
1
) are equal for any Kleinian group G and any points x
o
, x
1
∈ B
3
.
Proof:
A point u is in Λ(x
o
) if there is a sequence (T
n
) in G with [[T
n
(x
o
) −u[[ → 0 as N → ∞. Now
each T
n
acts isometrically on B
3
, so
ρ(T
n
(x
o
), T
n
(x
1
)) = ρ(x
o
, x
1
) .
The lemma shows that [[T
n
(x
1
) −u[[ → 0 as N → ∞. So u ∈ Λ(x
1
).
We already know that every ﬁxed point of a hyperbolic or loxodromic transformation does lie in the
limit set. For almost all Kleinian groups the ﬁxed points of loxodromic and hyperbolic transformations
are dense in the limit set. The exceptional groups are ones with very simple structure. They are called
elementary groups and we will not be concerned with them.
Exercise:
27. Let G be the group generated by the single parabolic transformation P : z → z +1. Show that the
limit set is ¦∞¦ but that there are no hyperbolic or loxodromic transformations in G.
Lemma 19.4
Let α be a hyperbolic geodesic that passes through a point c ∈ B
3
. At least one of the endpoints u of
α satisﬁes
[[u −c[[
1 −[[c[[
2
√
2[[c[[
=
√
2
sinh ρ(0, c)
.
Proof:
Draw the Euclidean sphere π that passes through c, is orthogonal to ∂B
3
and crosses the radius
from 0 to c normally. This Euclidean sphere has radius r and centre k. It meets B
3
in a hyperbolic
plane.
Since π cuts the unit sphere orthogonally, we must have
[[k[[
2
= ([[c[[ +r)
2
= 1
2
+r
2
.
So
r =
1 −[[c[[
2
2[[c[[
.
Any hyperbolic geodesic through c goes inside π in one direction, so one endpoint will lie within the
sphere π. This endpoint u satisﬁes
[[u −c[[
√
2r .
Lecture 19 72
0 k
c
π
r
1
Proposition 19.5
Let G be a Kleinian group that contains a hyperbolic or loxodromic transformation. Then the ﬁxed
points of all the hyperbolic and loxodromic transformations in G are dense in the limit set for G.
Proof:
Let A be the hyperbolic or loxodromic transformation in G. Let α be its axis joining the two
ﬁxed points a
+
, a
−
. We know that both these ﬁxed points lie in the limit set. Choose a point x
o
∈ B
3
that lies on the axis α. Then Proposition 19.3 shows that it suﬃces to prove that the ﬁxed points are
dense in Λ(x
o
).
Let u ∈ Λ(x
o
). Then there is a sequence (T
n
) in G with [[T
n
(x
o
) − u[[ → 0 as n → ∞. Now the
conjugate T
n
◦ A ◦ T
−1
n
is hyperbolic or loxodromic and has axis T
n
(α) which passes through T
n
(x
o
).
The lemma shows that one of the endpoints T
n
(a
+
) or T
n
(a
−
) satisﬁes
[[T
n
(a
±
) −u[[
√
2
sinh ρ(0, T
n
(x
o
))
.
As n → ∞, so ρ(0, T
n
(x
o
)) → ∞. Hence, some sequence of endpoints T
n
(a
±
) converges to u.
Recall that a subset Q of a metric space is perfect if no point of Q is isolated.
Corollary 19.6 Limit sets are perfect.
The limit set of a Kleinian group G that contains a hyperbolic or loxodromic transformation is either
ﬁnite or perfect.
Proof:
Suppose that the limit set is not ﬁnite. Let A be a hyperbolic or loxodromic transformation
in G with axis α. For each point v ∈ ∂B
3
= S
2
, except the endpoints of α, we know that A
n
(v) tends
to one endpoint u
+
of the axis α as n → +∞ and to the other u
−
as n → −∞.
Since the limit set is inﬁnite, Lemma 19.4 shows that there is another ﬁxed point v of a hyperbolic
or loxodromic transformation that is not ﬁxed by A. Then A
n
(v) → u
±
as n → ±∞. Hence both the
ﬁxed points u
±
of A are not isolated in the limit set.
Lecture 19 73
Every point of the limit set is a limit of ﬁxed points of hyperbolic or loxodromic transformations
in G, so no point in the limit set is isolated.
Corollary 19.7 Limit sets are ﬁnite or uncountable.
The limit set of a Kleinian group G that contains a hyperbolic or loxodromic transformation is either
ﬁnite or uncountable.
Proof:
For the limit set Λ is a closed subset of the sphere S
2
, so it is compact. The previous proposition
shows that it is perfect. Now the result follows from Cantor’s theorem that a perfect, compact, metric
space is uncountable.
Suppose that Λ were countable and enumerate its points as (x
n
)
n∈N
. We will construct a decreasing
sequence of nonempty, perfect, closed subsets K
n
of Λ with x
n
/ ∈ K
n
. Then the intersection
¸
K
n
can
not be empty, since Λ is compact. However, it does not contain any of the points x
n
so it must be
empty.
Set K
0
= Λ. Suppose that K
n
has been deﬁned and is a nonempty, perfect, closed subset of Λ. If
x
n+1
/ ∈ K
n
then take K
n+1
= K
n
. Otherwise, x
n+1
∈ K
n
. Since K
n
is perfect, there must be another
point, say y, in K
n
. Set K
n+1
to be the closure of
K
n
∩ B(y,
1
2
ρ(x
n+1
, y)) .
No point of this closure is isolated, so K
n+1
is a nonempty, perfect, closed set with x
n+1
/ ∈ K
n+1
⊂ K
n
.
This completes the inductive construction.
Lecture 19 74
20 HAUSDORFF DIMENSION
20.1 Cantor Sets
Deﬁne a sequence of sets C
n
⊂ [0, 1] as follows:
C
0
= [0, 1]
C
1
= [0,
1
3
] ∪ [
2
3
, 1]
C
2
= [0,
1
9
] ∪ [
2
9
,
1
3
] ∪ [
2
3
,
7
9
] ∪ [
8
9
, 1]
C
3
= [0,
1
27
] ∪ [
2
27
,
1
9
] ∪ [
2
9
,
7
27
] ∪ [
8
27
,
1
3
] ∪ [
2
3
,
19
27
] ∪ [
20
27
,
7
9
] ∪ [
8
9
,
25
27
] ∪ [
26
27
, 1]
.
.
.
.
.
.
At each stage, C
n
consists of 2
n
disjoint intervals each of length 3
−n
. C
n+1
is obtained from C
n
by
removing the open middle third from each of these intervals. The Cantor set C is the intersection
C =
¸
n∈N
C
n
.
The Cantor set is clearly closed and bounded in R, so it is compact. It is nonempty, since the
endpoints of each component interval of C
n
lie in C. More carefully, we can write any x ∈ [0, 1] in base
3 as
x = (0.x
1
x
2
x
3
. . .)
3
=
x
1
3
+
x
2
3
2
+
x
3
3
3
+. . .
where the digits x
1
, x
2
, x
3
, . . . are each 0 or 1 or 2. Then x ∈ C
n
if and only if it can be written as
x =
x
1
3
+
x
2
3
2
+
x
3
3
3
+. . . with x
1
, x
2
, . . . , x
n
∈ ¦0, 2¦ .
Note that this is true even for the endpoints of C
n
. For example,
1
3
= (0.022222 . . .)
3
and
2
3
= (0.200000 . . .)
3
.
Consequently, the Cantor set consists of all x ∈ [0, 1] that can be written as
x =
∞
¸
k=1
x
k
3
k
with each x
k
∈ ¦0, 2¦ .
This shows that the Cantor set has the same cardinality as the unit interval, for the map
C → [0, 1] ;
¸
x
k
3
k
→
¸
x
k
/2
2
k
is a bijection. Hence the Cantor set is uncountable. It also shows that the map
φ : Z
N
2
→ C ; (y
k
) →
¸
2y
k
3
k
is a bijection, so we can identify C with the inﬁnite group Z
N
2
.
Exercise:
28. Let y = (y
k
) and z = (z
k
) be sequences in Z
N
2
. Show that
d(y, z) =
0 when y = z;
2
−n
when n = min¦k : y
k
= z
k
¦
is a metric on Z
N
2
. Show that the map φ : Z
N
2
→ C deﬁned above is a homeomorphism from Z
N
2
with this metric.
Lecture 20 75
The Cantor set is easily seen to be perfect. This gives another proof that the set C is uncountable.
The sets C
n
for N 1 are disconnected, so the Cantor set is disconnected. Indeed, the Cantor set
is totally disconnected: the only connected subsets of C are the singletons. For suppose that x, y were
two diﬀerent points of C. Then [x − y[ > 3
−n
for some n ∈ N, so x and y are in diﬀerent components
of C
n
. Any nonempty, compact, metric space which is both perfect and totally disconnected is called
a Cantor set. All such sets are, in fact, homeomorphic to the Cantor set (but we will not prove this).
The most important property of the Cantor set is its selfsimilarity. The maps
s
0
: C → C ; x →
1
3
x and s
1
: C → C ; x →
1
3
x +
2
3
send C homeomorphically onto the subsets C ∩ [0,
1
3
] and C ∩ [
2
3
, 1]. Each of them is an contraction
with scale factor
1
3
.
Exercise:
29. How do these selfsimilarities act on Z
N
2
?
We can use these selfsimilarities to ﬁnd the “dimension” of the Cantor set. For now we will do
this informally. Later we will deﬁne the Hausdorﬀ dimension of a set and prove the results properly.
Consider a subset X of R
N
and let V
d
(X) denote the ddimensional volume of X. So V
1
(X) is the
length of X; V
2
(X) the area of X; V
3
(X) the volume of X. If X is a set of dimension k = 1, 2, 3, . . .,
then we expect that
V
d
(X) =
0 for k < d;
∞ for k > d.
If s is a contraction with scale factor λ then
V
d
(s(X)) = λ
d
V
d
(X) .
For the Cantor set C we know that C is the disjoint union of s
0
(C) and s
1
(C). Hence, we would
expect
V
d
(C) = V
d
(s
0
(C)) +V
d
(s
1
(C))
=
1
3
d
V
d
(C) +
1
3
d
V
d
(C) = 2
1
3
d
V
d
(C) .
So the ddimensional measure can only be ﬁnite and nonzero when
1 = 2
1
3
d
, that is d =
log 2
log 3
= 0.63093 . . . .
Hence the Cantor set has fractional dimension 0.63093 . . ..
20.2 Hausdorﬀ Dimension
Let M be a metric space with metric d. We will be particularly interested in subsets of R
N
with
the Euclidean metric or R
N
∞
with the chordal metric, for example the Cantor set or limit sets of Kleinian
groups. The diameter of M is
diam(M) = sup¦d(x, y) : x, y ∈ M¦ .
A collection ¦U
1
, U
2
, U
3
, . . .¦ of subsets of M is a δcover for M when M =
¸
n∈N
U
n
and each set
U
n
has diameter at most δ. We set
H
d
δ
(M) = inf
¸
n∈N
diam(U
n
)
d
: (U
n
)
n∈N
is a δcover for M
¸
.
Lecture 20 76
As we decrease δ, so the class of allowed δcovers is reduced. Hence, H
d
s
(M) increases as δ ` 0. Now
we deﬁne the ddimensional Hausdorﬀ measure of M as:
H
d
(M) = lim
δ→0
H
d
δ
(M) = sup
¸
H
d
δ
(M) : δ > 0
¸
.
This measures the ddimensional size of the set M. It is usually either 0 or ∞. Indeed, we will show
that, for each set M, there is at most one dimension d for which H
d
(M) is ﬁnite and nonzero.
Suppose that M is a metric space and 0 s < d. For each 0 < δ < 1 and any δcover (U
n
) of M
we have
¸
diam(U
n
)
d
δ
d−s
¸
diam(U
n
)
s
.
So H
d
δ
(M) δ
d−s
H
s
δ
(M). This shows that
if H
s
(M) < ∞ then H
d
(M) = 0 for s < d ;
if H
d
(M) > 0 then H
s
(M) = ∞ for s < d .
So there is a critical value at which the Hausdorﬀ dimension H
s
(M) jumps from ∞ to 0. This value is
called the Hausdorﬀ dimension dim
H
M of M. Note that
H
s
(M) =
∞ for s < dim
H
M;
0 for s > dim
H
M.
but the Hausdorﬀ measure at the critical value dim
H
M may be any number between 0 and ∞ including
both endpoints. It is usually very hard to calculate the Hausdorﬀ measure at this critical value and
only a little easier to ﬁnd the Hausdorﬀ dimension.
A map f : M → N between two metric spaces is KLipschitz if
d(f(x), f(y)) Kd(x, y) for all x, y ∈ M .
Such a map is certainly uniformly continuous. A map f : M → N is Lipschitz if it is KLipschitz for
some ﬁnite K. A map f : M → N is biLipschitz if it is Lipschitz and it has an inverse which is also
Lipschitz. Any map f : R
N
→ R
N
that is diﬀerentiable with bounded derivative is certainly Lipschitz
by the mean value theorem. So, for example, every M¨obius transformation is biLipschitz.
Proposition 20.1
If f : M → N is a KLipschitz map, then
H
d
(f(M)) K
d
H
d
(M) .
Proof:
Suppose that (U
n
) is a δcover for M. Then each f(U
n
) has diameter at most Kdiam(U
n
), so
(f(U
n
)) is a Kδcover for f(M). Furthermore,
¸
diam(f(U
n
))
d
K
d
¸
diam(U
n
)
d
.
So, H
d
Kδ
(f(M)) K
d
H
d
δ
(M). Taking the limits as δ ` 0 gives the result.
Corollary 20.2 Lipschitz maps preserve Hausdorﬀ dimension
If f : M → N is a Lipschitz map, then
dim
H
f(M) dim
H
M .
Lecture 20 77
Proof:
Suppose that f is KLipschitz. The Proposition shows that H
d
(f(M)) K
d
H
d
(M) for every
value of the dimension d. Hence dim
H
f(M) dim
H
M.
This Corollary shows that the Hausdorﬀ dimension of a set is unchanged by biLipschitz maps.
In particular, the Hausdorﬀ dimension of the limit set Λ(G) of a Kleinian group G is unchanged by a
M¨obius transformation. So Λ(TGT
−1
) and Λ(G) have the same Hausdorﬀ dimension.
Exercise:
30. A map f : M → N is αH¨older continuous if there is a constant C < ∞ with
d(f(x), f(y)) Cd(x, y)
α
for all x, y ∈ M .
Show that, for such a map,
dim
H
f(M)
1
α
dim
H
M .
The Hausdorﬀ measure H
d
is an (outer) measure. We will not use measure theory but we will need
the following very simple consequences:
If M ⊂ N then H
d
(M) H
d
(N).
If M is the disjoint union of two subsets M
1
, M
2
, then H
d
(M) = H
d
(M
1
) +H
d
(M
2
).
We are now in a position to prove properly that the Cantor set C has Hausdorﬀ dimension
log 2/ log 3. Let C
0
= C ∩ [0,
1
3
] and C
1
= C ∩ [
2
3
, 1]. The contraction maps
s
0
: C → C
0
; x →
1
3
x and s
1
: C → C
1
; x →
1
3
x +
2
3
are each Lipschitz maps with Lipschitz constant
1
3
. Their inverses are also Lipschitz with constants 3.
So
H
d
(C
0
) = H
d
(C
1
) =
1
3
d
H
d
(C) .
This shows that
H
d
(C) = H
d
(C
0
) +H
d
(C
1
) = 2
1
3
d
H
d
(C)
for every dimension d. If, for some value d, we have 0 < H
d
(C) < ∞ then we see that 2
1
3
d
= 1 and
so the Hausdorﬀ dimension d must satisfy
2
1
3
d
= 1 that is d =
log 2
log 3
.
However, we need to show that for this dimension we do indeed have 0 < H
d
(C) < ∞.
First observe that the set C
k
consists of 2
k
intervals each of length 3
−k
. These intervals form a
δcover of C for δ = 3
−k
. So
H
d
δ
(C) 2
k
(3
−k
)
d
= (2 3
−d
)
k
= 1
because d = log 3/ log 2. Letting k ∞ this gives H
d
(C) 1.
For the converse, suppose that (U
n
) is a δcover of C. We will show that
¸
diam(U
n
)
d
1
2
.
This will prove that H
d
(C)
1
2
as required.
Lecture 20 78
First observe that, by expanding the sets U
n
by a small amount, we can ensure that they are open.
Since C is compact, there is then a ﬁnite subcover. So we can assume that (U
n
) is a ﬁnite collection of
open sets that cover C. Choose an integer K so that each U
n
has diameter greater than 3
−K
. Suppose
that U
n
is one of these sets, with
3
−k−1
diam(U
n
) < 3
−k
for some natural number k < K. Now any two components of C
k
are distance at least 3
−k
apart, so
U
n
can not meet more than one of them. Let V
n
be this component. It has length 3
−k
, so diam(U
n
)
1
3
diam(V
n
). Then (V
n
) is a cover for C and
¸
diam(U
n
)
d
¸
1
3
diam(V
n
)
d
=
1
3
d
¸
diam(V
n
)
d
=
1
2
¸
diam(V
n
)
d
.
Hence it will suﬃce to prove that
¸
diam(V
n
)
d
1
for a ﬁnite cover of C by the intervals V
n
.
Now consider one of the intervals V
n
. The intersection C
K
∩ V
n
consists of 2
K−k
intervals each of
length 3
−K
, which we will denote by (W
j
)
2
K−k
j=1
. Then
diam(V
n
)
d
= (3
−k
)
d
= 2
−k
and
2
K−k
¸
j=1
diam(W
j
)
d
= 2
K−k
(3
−K
)
d
= 2
−k
(2 3
−d
)
K
= 2
−k
are equal. So it suﬃces to prove that
¸
diam(W
j
)
d
1
when (W
j
) is a cover of C by the component intervals of C
K
. Since these are a cover, all 2
K
intervals
must appear in the sum and the result is clear.
Lecture 20 79
21 CALCULATING THE HAUSDORFF DIMENSION
Proposition 21.1
If a metric space M has dim
H
M < 1, then M is totally disconnected.
Proof:
Consider ﬁrst a subset X of R with dim
H
X < 1. Any nonempty open interval U has H
1
(U) =
1, so X can not contain any such interval. This means that between any two points x, y ∈ X there is a
point c / ∈ X.
Now suppose that x, y are two distinct points of the metric space M. Then
f : M →R ; z → d(x, z)
is 1Lipschitz, so Corollary 20.2 shows that dim
H
f(M) < 1. Hence there is a c / ∈ f(M) with 0 =
f(x) < c < f(y). The sets f
−1
(−∞, c) and f
−1
(c, ∞) are then disjoint open sets containing x and y
respectively with union M.
The converse fails as the example of Cantor dust shows. We construct Cantor dust as follows:
D
0
= [0, 1] [0, 1] ⊂ R
2
;
If D
n
has been deﬁned and consists of a ﬁnite number of disjoint closed squares, then D
n+1
is
obtained by dividing each square into 16 smaller squares and keeping only the 4 shown in the
diagram below.
Cantor dust is the intersection
¸
D
n
.
D
0
D
1
D
2
D
3
D
4
Each D
n
consists of 4
n
squares with side length 4
−k
and hence diameter 4
−k
√
2. Covering D by these
squares shows that
H
1
(D)
√
2 .
The projection π
1
onto the ﬁrst coordinate is a 1Lipschitz map, with π
1
(D) = [0, 1], so
H
1
(D) H
1
([0, 1]) = 1 .
Therefore we see that D has Hausdorﬀ dimension 1. However, it is clear that D is totally disconnected.
21.1 Invariant Sets
Let M be a metric space and /(M) the collection of nonempty, compact subsets of M. If K is
one of these compact sets then
K
δ
=
¸
¦B(x, δ) : x ∈ K¦
is an open set containing K for each δ > 0. It is called the δneighbourhood of K. The Hausdorﬀ
distance D on /(M) is
D(K, L) = inf¦δ > 0 : K ⊂ L
δ
and L ⊂ K
δ
¦ .
Lemma 21.2 Hausdorﬀ distance
The Hausdorﬀ distance D is a metric on /(M).
Lecture 21 80
Proof:
Let K, L ∈ /(M). Choose a point x ∈ L. The compact set K is bounded, so there is some
δ > 0 with K ⊂ B(x, δ) ⊂ L
δ
. A similar argument with K and L interchanged shows that the Hausdorﬀ
distance D(K, L) is ﬁnite.
If D(K, L) = 0, then K ⊂
¸
¦L
δ
: δ > 0¦ = L and L ⊂ K, so K = L.
It is clear that D(K, L) = D(L, K).
For a third set M ∈ /(M) suppose that M ⊂ L
ε
and L ⊂ K
δ
. Then
M ⊂ L
ε
⊂ (K
δ
)
ε
⊂ K
δ+ε
.
So we see that M ⊂ K
D(M,L)+D(L,K)
. Hence we obtain the triangle inequality:
D(M, K) D(M, L) +D(L, K) .
A contraction on the metric space M is a map C : M → M for which there is a constant c with
0 c < 1 and
d(C(x), C(y)) c d(x, y) for all x, y ∈ M .
The contraction mapping theorem shows that such a map has a unique ﬁxed point in M, provided that
M is complete.
Let C
1
, C
2
, . . . , C
N
be a ﬁnite collection of contractions with constants c
1
, c
2
, . . . , c
N
respectively.
A subset F of M is an invariant set for C
1
, C
2
, . . . , C
N
when
F =
N
¸
n=1
C
n
(F) .
Proposition 21.3 Invariant sets
A ﬁnite set C
1
, C
2
, . . . , C
N
of contractions on the Euclidean space E
M
have a nonempty, compact,
invariant set. There is only one such nonempty, compact, invariant set.
Proof:
For each compact subset K ∈ /(M) deﬁne ((K) =
N
¸
n=1
C
n
(K). Each map C
n
is continuous,
so C
n
(K) is compact and hence ((K) ∈ /(M). We are looking for a set F with ((F) = F.
For two sets K, L ∈ /(M) with K ⊂ L
δ
we have C
n
(K) ⊂ C
n
(L
δ
) ⊂ C
n
(L)
c
n
δ
. Hence,
((K) ⊂ ((L)
cδ
where c = max¦c
1
, c
2
, . . . , c
N
¦ .
This implies that
D(((K), ((L)) c D(K, L) .
Hence ( is a contraction.
If K and L were both nonempty, compact, invariant sets, then D(((K), ((L)) cD(K, L), so
K = L. It remains to show that there is at least one invariant set.
Consider the closed ball Q = B(0, R). Its image under ( lies within a distance max¦d(0, C
n
(0))¦+
cR of the origin. Hence, if we choose R large enough we will have ((Q) ⊂ Q. This implies that
Q ⊃ ((Q) ⊃ (
2
(Q) ⊃ (
3
(Q) ⊃ (
4
(Q) ⊃ . . . .
Lecture 21 81
The intersection F =
¸
(
n
(Q) is therefore a nonempty compact set. It is clear that ((F) = F, so F is
invariant.
For a contraction mapping C : M → M we know that the sequence x, C(x), C
2
(x), C
3
(x), . . .
converges to the ﬁxed point. Hence the proof above show that the sets
K, ((K), (
2
(K), (
3
(K), . . .
converge to the invariant set in the Hausdorﬀ metric for any starting set K ∈ /(M). If we choose K as
a compact set with ((K) ⊂ K, then
K ⊃ ((K) ⊃ (
2
(K) ⊃ (
3
(K) ⊃ . . .
and the intersection
¸
(
n
(K) is the invariant set.
Exercise:
31. What is the unique nonempty, compact, invariant set for a single contraction C : M → M?
Let C
0
, C
1
be the contractions on R given by
C
0
: x →
1
3
x and C
1
: x →
1
3
x +
2
3
.
Find the unique nonempty, compact, invariant set for C
0
, C
1
. Show that there are other nonempty
invariant sets.
We can also use the argument in the last proposition to give an upper bound on the Hausdorﬀ
dimension of the invariant set.
Proposition 21.4
Let C
1
, C
2
, . . . , C
N
be contractions for the Euclidean metric on E
M
with F as the nonempty, compact,
invariant set. Let c
n
< 1 be constants with
d(C
n
(x), C
n
(y)) c
n
d(x, y) for all x, y ∈ E
M
.
Then dim
H
F d where d is the unique solution to the equation
N
¸
n=1
c
d
n
= 1 .
Proof:
First note that the function f : s →
¸
c
s
n
is strictly decreasing, so there is a unique positive
number d with
¸
N
n=1
c
d
n
= 1.
The set F is bounded and so of ﬁnite diameter. Fix a natural number K and consider sequences
n = (n(1), n(2), . . . , n(K)) of K integers with 1 n(k) N. For any such sequence n deﬁne a set
(
n
(F) as
(
n
(F) := C
n(1)
◦ C
n(2)
◦ . . . ◦ C
n(K)
(F) .
This has diameter at most c
n(1)
c
n(2)
. . . c
n(K)
diam(F). If c = max¦c
n
: n = 1, 2, . . . , N¦ < 1 then this
diameter is at most c
K
diam(F).
Since F is invariant, we have
F =
¸
n
(
n
(F)
Lecture 21 82
where the union is over all sequences of K integers. Hence the sets (
n
(F) form a δcover for F provided
that c
K
diam(F) δ. For this cover we have
¸
n
diam((
n
(F))
d
¸
n
(c
n(1)
c
n(2)
. . . c
n(K)
diam(F))
d
= diam(F)
d
.
The last equality follows from the choice of d to satisfy
¸
c
d
n
= 1. So H
d
δ
(F) diam(F)
d
for every
δ > 0. Consequently, H
d
(F) diam(F)
d
and the Hausdorﬀ dimension of F is at most d.
The proof above should be compared with the proof that the Hausdorﬀ dmeasure of the Cantor
set is ﬁnite when d = log 2/ log 3. The other part of the proof, showing that the Hausdorﬀ dmeasure
is larger than 0, can also be generalised, although we need stronger restrictions on the contractions and
the metric space M. The Theorem below gives a useful result in this direction. We will not prove it
since it would involve a little probability.
A map C : M → M is a similarity with scale factor c ∈ [0, 1) if
d(C(x), C(y)) = c d(x, y) for all x, y ∈ M .
Such a similarity is certainly a contraction map.
Theorem 21.5
Let C
1
, C
2
, . . . , C
N
be similarities on the Euclidean space E
M
with scale factors c
1
, c
2
, . . . , c
N
. These
similarities have a unique nonempty, compact invariant set F. Let d be the unique number with
¸
c
d
n
= 1. Suppose that there is a bounded open set V with C
1
(V ), C
2
(V ), . . . , C
N
(V ) disjoint and
¸
C
n
(V ) ⊂ V . Then the Hausdorﬀ dmeasure H
d
(F) is greater than 0 and so the Hausdorﬀ dimension
of F is at least d.
Combining this with the previous Proposition shows that the Hausdorﬀ dimension is exactly d and that
0 < H
d
(F) < ∞.
For a proof of this result, which is not examinable, see “Fractal Geometry”, by K. Falconer, pp,
119120.
V
C
1
(V )
C
2
(V )
Lecture 21 83
22 EXAMPLES OF HAUSDORFF DIMENSION
We can now determine the Hausdorﬀ dimension of many selfsimilar sets.
22.1 The Cantor Set
The two similarities
C
0
: R →R ; x →
1
3
x and C
1
: R →R ; x →
1
3
x +
2
3
have the Cantor set as an invariant set. For we see that
(([0, 1]) = C
1
= [0,
1
3
] ∪ [
2
3
, 1]
(
2
([0, 1]) = C
2
(
3
([0, 1]) = C
3
etc.
and the invariant set is the limit of this sequence. We can take the open interval (0, 1) as the set V in
Theorem 21.5 . Hence the dimension d of the Cantor set is the unique solution of
¸
c
d
n
= 1. This is
2 (
1
3
)
d
= 1, so d = log 2/ log 3.
Lecture 22 84
22.2 The von Koch snowﬂake
This is the curve constructed by an iterative process starting from the unit interval [0, 1] ⊂ R
2
. At
each stage, each straight line segment is replaced by 4 line segments each
1
3
as long, as in the diagram.
For this we have 4 similarities each with scale factor
1
3
. We can take the open isosceles triangle with
base (0, 1) and height
1
2
√
3 as the set V . Then 4(
1
3
)
d
= 1, so the Hausdorﬀ dimension of the von Koch
snowﬂake is log 4/ log 3.
V
The von Koch snowﬂake Images of V .
Lecture 22 85
Three copies of the von Koch curve making a snowﬂake.
Lecture 22 86
22.3 The Sierpi´ nski Gasket
This is the set constructed as follows. We begin with a ﬁlled
in equilateral triangle. At each stage we replace each equilateral
triangle by 3 triangles each
1
3
the size. So we have 3 similarities
each with scale factor
1
2
. We can take the interior of the initial
triangle as the set V . Then 3(
1
2
)
d
= 1, so the Hausdorﬀ dimension
of Sierpi´ nski’s gasket is log 3/ log 2.
Lecture 22 87
23 SCHOTTKY GROUPS
23.1 Fuchsian Groups
We begin with the following simple exercise.
Proposition 23.1
Let γ
−
, γ
+
be two hyperbolic geodesics in D that do not meet either in D or on its boundary. These
bound two disjoint halfplanes δ
−
, δ
+
. There is a M¨obius transformation T that maps γ
−
onto γ
+
and
δ
−
onto D ` δ
+
.
Proof:
There is a hyperbolic geodesic ν normal to both γ
−
and γ
+
. This has endpoints w
−
, w
+
where
we may assume that w
−
is in the boundary of δ
−
and w
+
in the boundary of δ
+
. Conjugate by a M¨obius
transformation A chosen to map D onto R
2
+
, w
−
to 0 and w
+
to ∞. Then ν is mapped to the imaginary
axis. So the geodesics A(γ
±
) must be halfcircles perpendicular to this: A(γ
±
) = ¦z ∈ R
2
∞
: [z[ = R
±
¦.
Now the map
U : z →
R
+
R
−
z
is a hyperbolic M¨obius transformation that maps A(γ
−
) onto A(γ
+
). So we may take T = A
−1
◦U ◦A.
δ
−
δ
+
γ
−
γ
+
φ
T
φ
γ
−
= γ
+
D/G
The region f = D ` (δ
−
∪ δ
+
) between γ
−
and γ
+
is a fundamental set for the group G generated
by the single M¨obius transformation T. The quotient D/G is obtained by identifying the two sides γ
−
and γ
+
of f to get an annulus (ring) as shown above.
We can also think of D as a subset of the Riemann sphere P. The geodesics γ
±
are parts of
circles Γ
±
orthogonal to the unit circle ∂D. These circles enclose two disjoint discs ∆
±
. The M¨obius
transformation T acts on all of the Riemann sphere and maps ∆
−
onto P ` ∆
+
. Let G be the group
generated by T. Then the limit set Λ(G) consists of just the two ﬁxed points of T. The group G acts
discontinuously on all of the remainder of P and the set Φ = P ` (∆
−
∪ ∆
+
) is a fundamental set. The
quotient (P ` Λ(G))/G is obtained from Φ by identifying the two circles Γ
−
and Γ
+
to obtain a torus.
We will sometimes abuse the notation by writing (P ` Λ(G))/G as P/G.
Lecture 23 88
∆
−
∆
+
Γ
−
Γ
+
Φ
T
Γ
−
= Γ
+
(P \ Λ(G))/G
Note that inversion J in the unit circle ∂D maps D to the complementary disc J(D) and sends Φ
to itself with the part φ inside D going to the part J(φ) outside D. The torus (P ` Λ(G))/G is obtained
by taking the two annuli D/G and J(D)/G and joining them along their boundaries.
We can also think of the M¨obius transformation acting on hyperbolic 3space H
3
. The circles Γ
±
are the boundaries of hyperbolic planes (
±
in B
3
= H
3
. These bound disjoint halfspaces H
±
. The
M¨obius transformation T maps T
−
onto H
3
` T
+
. The set T = H
3
` (T
−
∪ T
+
) is a fundamental set
for G. The quotient H
3
/G is obtained from T by identifying the two planes (
−
and (
+
to get a solid
torus. Then (P ` Λ(G))/G is the boundary of H
3
/G.
We can do the same for Fuchsian groups G generated by more than one element. The group G
of M¨obius transformations acts on D and we obtain a surface D/G for the quotient. This surface is
orientable since each M¨obius transformation is orientation preserving. Similarly, G acts on J(D) = P`D.
The quotient J(D)/G is another surface, called the dual of D/G. The inversion J in ∂D induces an
orientation reversing bijection from D/G to J(D)/G.
When we think of G acting on all of the Riemann sphere we get a quotient (P`Λ(G))/G that consists
of D/G and J(D)/G joined together along their boundaries. Similarly, when we think of G acting on H
3
we get a quotient H
3
/G which is a 3dimensional solid that has (P ` Λ(G))/G as its boundary. (Recall
that we have seen examples of Fuchsian groups which have all of the unit circle in the limit set Λ(G).
In this case, the two parts D/G and J(D)/G are separated by the quotient of the limit set.)
Lecture 23 89
23.2 Schottky Groups
In the previous section all of our circles were orthogonal to the unit circle. In this section we do not
insist on this and produce Kleinian groups rather than Fuchsian groups. These are the Schottky groups.
As in the last section, we will begin by considering groups generated by a single M¨obius transformation.
Proposition 23.2
Let Γ
−
, Γ
+
be two disjoint circles bounding two disjoint discs ∆
−
, ∆
+
in P. Then there is a M¨obius
transformation T that maps Γ
−
onto Γ
+
and ∆
−
onto P ` ∆
+
.
Proof:
We could prove this by adapting the proof of Proposition 23.1 for hyperbolic 3space. However,
for variety, we will give a diﬀerent argument that works entirely on the Riemann sphere.
Let J
±
be inversion in the circle Γ
±
. Then S = J
+
◦ J
−
is a nonidentity M¨obius transformation.
Let z
o
be one ﬁxed point. Then J
+
(J
−
(z
o
)) = S(z
o
) = z
o
so J
−
(z
o
) = J
+
(z
o
). Also,
S(J
−
(z
o
)) = J
+
◦ J
−
◦ J
−
(z
o
) = J
+
(z
o
) = J
−
(z
o
)
so J
−
(z
o
) is also ﬁxed by S. Since Γ
−
and Γ
+
are disjoint, the points z
o
and J
−
(z
o
) must be distinct.
So S has 2 ﬁxed points. Note that J
−
and J
+
both interchange these two ﬁxed points.
Conjugate by a M¨obius transformation A that sends z
o
to 0 and J
−
(z
o
) to ∞. Then A◦J
±
◦A
−1
is
inversion in the circle A(Γ
±
) and must interchange 0 and ∞. So A(Γ
±
) must be a circle ¦z ∈ P : [z[ = r
±
¦
for some 0 < r
±
< ∞. Consequently the map
U : z → e
iθ
r
+
r
−
z
maps A(Γ
−
) onto A(Γ
+
). (It is hyperbolic if e
iθ
= 1 and loxodromic otherwise.) The map T = A◦U◦A
−1
now has the required properties.
Let Γ
−
, Γ
+
, ∆
−
, ∆
+
and T be as in the proposition. The group G generated by T has a limit set
Λ(G) consisting of the two ﬁxed points of T. The set Φ = P ` (∆
−
∪ ∆
+
) is a fundamental set for the
group G acting on P ` Λ(G). The quotient P/G (or, more accurately, (P ` Λ(G))/G) is obtained from
this fundamental set by identifying the two circles Γ
−
and Γ
+
. So the quotient is a torus (the surface
of a ring doughnut).
We can also think of the M¨obius transformation acting on hyperbolic 3space H
3
. The circles Γ
±
are the boundaries of hyperbolic planes (
±
in B
3
= H
3
. These bound disjoint halfspaces T
±
. The
M¨obius transformation T maps T
−
onto H
3
` T
+
. The set T = H
3
` (T
−
∪ T
+
) is a fundamental set
for G. The quotient H
3
/G is obtained from T by identifying the two planes (
−
and (
+
to get a solid
torus (the body of a ring doughnut). Then P/G is the boundary of H
3
/G.
The special case when the circles Γ
n
are orthogonal to the unit circle was the Fuchsian case dealt
with in the previous section.
We can do the same when we have more than one pair of circles. In this case we get a Schottky group
generated by several M¨obius transformations. For k = 1, 2, . . . , K, let Γ
−
k
and Γ
+
k
be disjoint circles
bounding 2K disjoint discs ∆
−
k
and ∆
+
k
. The proposition shows that there are M¨obius transformations
T
k
that map Γ
−
k
onto Γ
+
k
and ∆
−
k
onto P ` ∆
+
k
. The group G generated by T
1
, T
2
, . . . , T
K
is called the
(classical) Schottky group for the discs. We have already considered the Schottky groups for K = 1 but
Schottky groups with more than one generator are more interesting.
Let Φ be the set
Φ = P `
K
¸
k=1
∆
−
k
∪ ∆
+
k
.
Lecture 23 90
This is the region outside all of the discs ∆
±
k
except that we have included half of the bounding circles.
We will show that Φ is a fundamental set for the Schottky group G acting on P ` Λ(G).
The image T
k
(Φ) lies inside ∆
+
k
and is bounded by the 2K circles T
k
(Γ
±
j
). One of these is T
k
(Γ
−
k
) =
Γ
+
k
but all of the others are strictly inside ∆
+
k
. Now apply another one of the generators, say T
j
. The
image T
j
(T
k
(Φ)) is a subset of T
j
(∆
+
k
), which is itself a disc inside ∆
+
j
. Hence we get a pattern of
nested discs as shown below.
Φ
T
1
T
2
T
1
(Φ)
T
−1
1
(Φ)
T
2
(Φ) T
−1
2
(Φ)
The Tessellation for the Schottky Group.
We will see shortly that the intersection of any chain of nested discs is a point and the closure of
these points is the limit set Λ(G). The remainder of the Riemann sphere: P ` Λ(G), is tessellated by
the images g(Φ) for g ∈ G. These are clearly locally ﬁnite so we see that G acts discontinuously on
P ` Λ(G).
Lecture 23 91
T
1
(Φ)
T
1
T
2
(Φ)
T
1
T
−1
2
(Φ)
T
2
1
(Φ)
Enlargement of the region ∆
+
1
.
Theorem 23.3 Schottky groups are free groups.
The Schottky group generated by the M¨obius transformations pairing discs ∆
−
k
and ∆
+
k
for k =
1, 2, 3 . . . , K with the closure of all these 2K discs disjoint, is a free group on the generators (T
k
)
K
k=1
.
Proof:
To see this we need to look more closely at the tessellation shown above. It is simplest to see
the structure of this tessellation if we simplify the diagram. So draw a graph — the Cayley graph for G
— by putting a vertex for each image g(Φ) and labelling it with the element g of G. Join the vertices
labelled g and g ◦ T
k
for k = 1, 2, . . . , K. We have done this below for the case of two pairs of discs and
coloured the edges from g to g ◦ T
1
in blue and the edges from g to g ◦ T
2
in red. Two vertices g and h
in this graph are adjacent when g = hA for A one of the generators T
k
or their inverses T
−1
k
.
Lecture 23 92
I
T
1
T
2
T
−1
2
T
−1
1
T
2
1
T
1
T
2
T
1
T
−1
2
T
2
T
1
T
2
T
−1
1
T
2
2
T
−2
2
T
−1
2
T
1
T
−1
2
T
−1
1
T
−1
1
T
2
T
−1
1
T
−1
2
T
−2
1
The Cayley graph for a Schottky Group.
Consider a product
A
1
◦ A
2
◦ A
3
◦ . . . ◦ A
N
where N 0 and each A
n
is one of T
k
or T
−1
k
. If two successive terms A
n
, A
n+1
have A
n
◦ A
n+1
= I,
then we can cancel them and reduce the length N by 2. Repeat this until there are no such pairs. We
need to show that no such product is the identity I (except for the trivial product of no elements). In
the graph, the path
I, A
1
, A
1
◦ A
2
, A
1
◦ A
2
◦ A
3
, . . . , A
1
◦ A
2
◦ A
3
◦ . . . ◦ A
N
= g
follows edges from I to g and never turns back on itself. Since there are no loops in the graph, this
implies that the path can not return to its starting point. So g = I unless the product is trivial. Hence
G is a free group.
Lecture 23 93
The quotient (P`Λ(G))/G is obtained from the fundamental set Φ by identifying the pairs of circles
∆
−
k
and ∆
+
k
for k = 1, 2, . . . , K. This gives a sphere with K handles.
Γ
−
1
= Γ
+
1
Γ
−
2
= Γ
+
2
We can also think of the Schottky group acting on the hyperbolic 3space H
3
. The circles Γ
±
k
are
the boundaries of hyperbolic planes (
±
k
in B
3
= H
3
. These bound disjoint halfspaces T
±
k
. The M¨obius
transformations T
k
maps T
−
k
onto H
3
` T
+
k
. The set T = H
3
` (T
−
∪ T
+
) is a fundamental set for G.
The quotient H
3
/G is obtained from T by identifying the two planes (
−
k
and (
+
k
for k = 1, 2, . . . , K.
This gives a solid ball with K handles. Then (P ` Λ(G))/G is the boundary of H
3
/G.
D
−
1
D
+
1
D
−
2
D
+
2
F
23.2 The Limit Set for a Schottky Group
The pictures above suggest that nested sequences of the discs obtained as images of the basic discs
∆
±
k
under the group G are single points and that these points are dense in the limit set. We will prove
this and also obtain a bound on the Hausdorﬀ dimension of the limit set.
It is convenient to work in H
3
. The hyperbolic planes T
−
k
and T
+
k
do not meet either in H
3
nor on
its boundary, so they are all at least a nonzero hyperbolic distance t apart. This means that any curve
in the fundamental set T that joins one of these planes to another must have length at least t.
Lecture 23 94
We can write any g ∈ G as a product
g = A
1
◦ A
2
◦ . . . ◦ A
N
where N 0, each A
n
is one of T
k
or T
−1
k
and A
n
◦ A
n+1
= I for n = 1, 2, . . . , N −1. Since G is a free
group, the length N of such a product is determined by g. We will call it the length of g. There is 1
element of length 0, 2K of length 1, and 2K(2K −1)
N−1
of length N.
Choose a base point in T, say the origin 0. Let γ be the shortest hyperbolic path from this base
point to the set g(T). This path begins in T, then crosses into A
1
(T), then A
1
◦ A
2
(T) and so on until
it crosses into A
1
◦ A
2
◦ . . . ◦ A
N−1
(T) and ﬁnally meets g(T). This means that it must cross each of
the regions A
1
(T), A
1
◦ A
2
(T), . . . , A
1
◦ A
2
◦ . . . ◦ A
N
(T) from one of the bounding planes to another.
Hence the hyperbolic length of γ must be at least Nt.
Lemma 23.4
Let T be a hyperbolic plane at a hyperbolic distance ρ from the origin in B
3
= H
3
. Then the Euclidean
diameter of T is at most 2/ sinh ρ.
This is essentially the same result as Lemma 19.4 . The inequality is only useful when ρ is small. For
large ρ the observation that diam(T) 2 is better.
Proof:
The hyperbolic plane T is part of a Euclidean sphere with centre k and Euclidean radius r.
The shortest path from the origin to T is a radial line from 0 to k that crosses the plane T at a point
c with [[c[[ = tanh
1
2
ρ. Since T is orthogonal to the unit sphere, Pythagoras’ Theorem shows that
1 +r
2
= [[k[[
2
= ([[c[[ +r)
2
.
So
r =
1 −[[c[[
2
2[[c[[
=
1 −tanh
2 1
2
ρ
2 tanh
1
2
ρ
=
1
sinh ρ
.
k 0
D
c
1 r
The Euclidean diameter of T is at most 2r.
It follows from this lemma that an element g of length N in G has g(T) at a hyperbolic distance
ρ Nt from 0 and hence the Euclidean diameter of g(T) is at most 2/sinh Nt. When we look at this
on the Riemann sphere it shows that diam(g(Φ)) 2/ sinh Nt.
Lecture 23 95
Consider the tessellation given by the images g(Φ) for g in the Schottky group G. We wish to study
a chain of these images g
n
(Φ) where every successive pair g
n
(Φ) and g
n+1
(Φ) meet along a circle. Now
the fundamental domain Φ only meets the images A(Φ) where A is one of T
k
or T
−1
k
. Hence g
n
(Φ) and
g
n+1
(Φ) only meet when g
n+1
= g
n
◦ A
n
where A
n
∈ ¦T
k
, T
−1
k
: k = 1, 2, . . . , K¦. Thus we have
g
n
= A
1
◦ A
2
◦ . . . ◦ A
n
.
Note that we can not have A
n
◦A
n+1
= I or else g
n−1
= g
n+1
and the two images g
n−1
(Φ) and g
n+1
(Φ)
are the same. The images g
n
(Φ) touch each other along a sequence of circles that are nested inside one
another. (Look at the diagram of the Tessellation for the Schottky Group.) We can use the previous
lemma to estimate the size of these circles as
diam(g
n
(Φ))
2
sinh nt
.
This certainly shows that their Euclidean diameter tends to 0 as n → ∞.
We can also use this idea to ﬁnd an upper estimate for the Hausdorﬀ dimension of the limit set Λ(G).
Suppose that C
n
is the circle separating g
n
(Φ) from g
n+1
(Φ). Then C
n
surrounds a disc containing all
of the images g
m
(Φ) for m > n. Hence this disc must contain a limit point of G. So every sequence
(g
n
) as above gives us a point of the limit set. However, no point of any of the images g(Φ) can be in
the limit set since there are only ﬁnitely many copies h(Φ) for h ∈ G that border it. Therefore the limit
set is the complement of all these images g(Φ).
For any natural number N, the complement of the union:
¸
¦g(Φ) : g ∈ G has length at most N¦
certainly contains the limit set. Our argument shows that the complement consists of 2K(2K −1)
N−1
discs each with Euclidean diameter at most
2
sinh Nt
. So the Hausdorﬀ ddimensional measure of Λ(G) is
at most
H
d
δ
(Λ(G)) 2K(2K −1)
N−1
2
sinh Nt
d
for δ >
2
sinh Nt
. When N is large this is approximately
2K(2K −1)
N−1
4e
−Nt
d
=
2K 4
d
2K −1
(2K −1)e
−td
N
.
So we see that the ddimensional Hausdorﬀ measure is 0 for
d >
log(2K −1)
t
.
This proves that the Hausdorﬀ dimension of the limit set is at most
log(2K −1)
t
.
(Much more delicate arguments show that the Hausdorﬀ dimension of the limit set is actually equal to
the inﬁmum of those d for which the series
¸
g∈G
exp(−dρ(0, g(0))) converges.)
Lecture 23 96
24 DEGENERATE SCHOTTKY GROUPS
24.1 How Schottky Groups Degenerate
So far we have insisted that the discs ∆
±
k
for our Schottky groups are disjoint. It is interesting to
ask what happens as they move closer to one another and, eventually touch or cross. There are still
M¨obius transformations that pair the circles Γ
±
k
. However, the group generated by them need not be
discrete. For example, if the circles crossed and the map T
k
were elliptic, then G could only be discrete
if T
k
were of ﬁnite order.
Consider the Schottky group for the disjoint discs (∆
±
k
)
K
k=1
. Take one pair, say ∆
−
1
and ∆
+
1
, and
allow them to move closer. This means that the generator T
1
also varies. In the picture below the
shortest (chordal) path between the two discs is marked in green. As the discs move closer together, so
this green curve becomes shorter. Ultimately, the green curve reduce to a point and the two discs ∆
−
1
and ∆
+
1
touch at a point P. In the quotient, the point P corresponds to a singular point of the surface.
∆
−
1
∆
+
1
∆
−
2
∆
+
2
Γ
−
1
= Γ
+
1
Γ
−
2
= Γ
+
2
∆
−
1
∆
+
1
∆
−
2
∆
+
2
P
Γ
−
1
= Γ
+
1
Γ
−
2
= Γ
+
2
Lecture 24 97
The transformation T
1
also changes as we vary the discs. The most interesting case is when, in the
limit, T
1
becomes a parabolic transformation that ﬁxes P. Then we get a tessellation as shown below.
Lecture 24 98
We can also let other pairs of circles touch. Let us consider the case where we have 2 pairs of circles
Γ
±
1
and Γ
±
2
as shown below. We will also assume that the transformations T
1
, T
2
are parabolic with
T
1
: A → A ; T
2
: C → C ;
T
1
: B → D ; T
2
: B → D .
The fundamental set then breaks into two parts, both quadrilaterals bounded by arcs of each of the
four circles. When we identify the edges using T
1
and T
2
, each of these quadrilaterals becomes a sphere
with 3 punctures.
A
B
C
D
The limit set in this case is a closed Jordan curve contained in the chains of circles shown below.
Exercise:
32. Show that, in the special case where the 4 points A, B, C, D lie on a circle that crosses the 4 circles
Γ
±
k
orthogonally, the group is a Fuchsian group.
Lecture 24 99
A special case of this is when our discs are:
∆
−
1
= ¦z ∈ C
∞
: Re(z) < −1¦ ∆
+
1
= ¦z ∈ C
∞
: Re(z) > 1¦
∆
−
2
= ¦z ∈ C
∞
: [z +
1
2
[ <
1
2
¦ ∆
+
2
= ¦z ∈ C
∞
: [z −
1
2
[ <
1
2
¦
The transformations are
T
1
: z → z + 2 and T
2
: z → −
1
z
.
This is then a Fuchsian group preserving the upper and lower halfplanes. It is a subgroup of the
modular group we looked at earlier. Its limit set is the circle R ∪ ¦∞¦.
∆
−
1
∆
+
1
∆
−
2
∆
+
2
2 1 0 1 2
∆
−
1
∆
+
1
∆
−
2
∆
+
2
Lecture 24 100
An even more degenerate case is when each disc touches the 3 others. This is illustrated below on
the left. The group is generated by two parabolic transformations ﬁxing points where two circles touch.
The limit set is the Apollonian gasket. This is the pattern obtained as follows:
Given any 3 mutually tangent circles, there are 2 further circles each of which touches
each of the original 3. Start with 3 mutually tangent circles and use this process to add further
circles recursively.
Exercise:
33. Show that we can choose 4 circles each pair of which touch so that they have tetrahedral
symmetry. Show that the limit set, the Apollonian gasket also has tetrahedral symmetry.
Four circles with each pair tangent. Limit set — The Apollonian Gasket
Lecture 24 101
We ought really to draw this picture on the Riemann sphere, as illustrated below.
By allowing these sorts of degenration for Schottky groups we can create very complicated,
fractal limit sets. An example is given below.
Lecture 24 102
*24.2 Riemann Surfaces *
Let G be a group of M¨obius transformations that acts discontinuously on the unit disc D.
Then the quotient D/G is an orientable surface. Each of the M¨obius transformations in G is
analytic and has an analytic inverse. Hence we can do complex analysis on the quotient D/G.
We call such a surface which looks locally like a piece of the complex plane a Riemann surface.
Similar arguments apply to groups acting discontinuously on the complex plane. For
example, consider the group of translations G = ¦z → z + λ : λ ∈ Λ¦. When Λ = ¦0¦, the
quotient C/G is just C; when Λ = Zω
1
, the quotient C/G = C/Zω
1
is a cylinder; when G =
Zω
1
+Zω
2
(with ω
1
and ω
2
linearly independent over R), the quotient C/G = C/(Zω
1
+Zω
2
)
is a torus.
The Riemann Mapping Theorem (and the Uniformization Theorem) states that every
Riemann surface is either the quotient of the unit disc D by a discrete group or else on of the
special surfaces: P, C, a cylinder or a torus. This means that almost all Riemann surfaces
are quotients of the unit disc. We can study them by studying the discrete groups of M¨obius
transformations that maps the unit disc to itself, that is, by studying Fuchsian groups.
Compact Riemann surfaces are compact orientable surfaces. It can be shown that these
are homeomorphic to spheres with K handles for some K 0. Hence every such compact
orientable surface is the quotient of the Riemann sphere by a Schottky group. Indeed, more is
true, by choosing the discs ∆
±
k
appropriately, we can obtain every compact Riemann surface
from a Schottky group, including the complex structure on the surface. This leads us to think
about classifying all compact Riemann surfaces by using the Schottky groups. This is a very
active area of research known as the study of Teichm¨ uller space.
For further study I recommend the book Indra’s Pearls: The Vision of Felix Klein by
David Mumford, Caroline Series, and David Wright. Cambridge University Press, 2002 (ISBN
0521352533). This has much more information, beautifully presented.
Lecture 24 103
CONTENTS 1 INTRODUCTION 1.1 Symmetry 1.2 Group Actions Proposition 1.1 1 1 4 5 7 7 8 8 9 10 10 11 11 11 12 13 13 14 15 15 16 17 19 19 21 22 22 23 24 25 25 26 26 26 27 28 28 29 31 31 33 33 34 34 35
Orbit – Stabilizer theorem
2 ISOMETRIES OF EUCLIDEAN SPACE 2.1 Deﬁnitions Lemma 2.1 Orthogonal maps as Euclidean isometries Proposition 2.2 Euclidean isometries are aﬃne 2.2 Isometries of the Euclidean Plane Proposition 2.3 Isometries of E2 Proposition 2.4 Isometries of E3 3 THE ISOMETRY GROUP OF EUCLIDEAN SPACE 3.1 Quotients of the Isometry group Proposition 3.1 3.2 Matrices for Isometries 3.3 Finite Groups of Isometries of the Plane Proposition 3.2 Finite subgroups of Isom(E2 ). 3.4 Compositions of Reﬂections 4 FINITE SYMMETRY GROUPS OF EUCLIDEAN SPACE 4.1 Examples of Finite Symmetry Groups 4.2 Finite Subgroups of Isom(E3 ). Theorem 4.1 Finite symmetry groups in Isom+ (E3 ) 5 THE PLATONIC SOLIDS 5.1 History 5.2 Regularity 5.3 Convex Regular Polyhedra 5.4 The Symmetry Groups 5.5 Fundamental sets 6 LATTICES Proposition 6.1 Proposition 6.2 Lattices in R Lattices in R2
7 EUCLIDEAN CRYSTALLOGRAPHIC GROUPS Lemma 7.1 The point group acts on the lattice 7.1 Rank 0 : Finite Groups 7.2 Rank 1: Frieze Patterns 7.3 Rank 2: Wallpaper patterns Lemma 7.2 The crystallographic restriction Corollary 7.3 Point groups ¨ 8 MOBIUS TRANSFORMATIONS 8.1 The Riemann Sphere Proposition 8.1 Stereographic projection preserves circles. 8.2 M¨bius Transformations o Proposition 8.2 M¨bius transformations map circles to circles o Proposition 8.3 Proposition 8.4 Isometries of the Riemann sphere.
¨ 9 VISUALISING MOBIUS TRANSFORMATIONS 36 9.1 Fixed Points 36 Theorem 9.1 Fixed points of M¨bius transformations o 36 Corollary 9.2 Trace determines conjugacy class of a M¨bius transformation 37 o 9.2 Inversion 38 Lemma 9.3 M¨bius transformations preserve inverse points o 38 i
Proposition 9.4 Proposition 9.5
Inversion The composition of two inversions is M¨bius. o
39 39 41 41 41 42 44 45 46 46 46 46 47 47 47 48 51 52 54 54 54 54 55 55 55 56
10 THE HYPERBOLIC PLANE, I 10.1 M¨bius Transformations of the unit disc o Proposition 10.1 M¨bius transformations of the unit disc o 10.2 The Hyperbolic Metric on D Lemma 10.2 Hyperbolic geodesics from the origin. Theorem 10.3 Hyperbolic metric on the unit disc. 11 THE HYPERBOLIC PLANE, II 11.1 The Hyperbolic Metric on a Half Plane. 11.2 Inversions Proposition 11.1 Inversions preserve the hyperbolic metric 12 FUCHSIAN GROUPS 12.1 Single generator Fuchsian groups Proposition 12.1 12.2 Triangle Groups 13 THE MODULAR GROUP Proposition 13.1 Fundamental set for the modular group.
14 HYPERBOLIC 3SPACE 14.1 The Hyperbolic Metric Proposition 14.1 Hyperbolic metric on B 3 14.2 Inversion Proposition 14.2 Inversion in spheres Proposition 14.3 Inversion preserves spheres. Corollary 14.4 Inversion preserves circles. Proposition 14.5 Inversion preserves angles.
¨ 15 EXTENDING MOBIUS TRANSFORMATIONS TO HYPERBOLIC SPACE 57 15.1 Inversions and the hyperbolic metric 57 Proposition 15.1 Inversions are hyperbolic isometries. 57 Lemma 15.2 57 Proposition 15.3 Hyperbolic geodesics 58 Proposition 15.4 Extensions of M¨bius transformations. o 58 Theorem 15.5 M¨bius transformations as isometries of hyperbolic 3space. 58 o 15.2 The upper halfspace. 59 16 ISOMETRIES OF H3 16.1 Examples in Hyperbolic Geometry 16.2 Axes of Isometries 17 INVOLUTIONS Proposition 17.1 Proposition 17.2 Isometries of H3 are compositions of two involutions. 60 60 62 64 64 65
18 KLEINIAN GROUPS 67 18.1 Finite Kleinian Groups 67 Lemma 18.1 67 Proposition 18.2 Finite Kleinian groups are conjugate to subgroups of SO(3). 68 18.2 Discontinuous Action 68 Lemma 18.3 68 Theorem 18.4 Discrete if and only if acts discontinuously. 69 19 LIMITS OF ORBITS Proposition 19.1 Lemma 19.2 Proposition 19.3 Lemma 19.4 Proposition 19.5 The limit set is closed and Ginvariant. Limit set is independent of the base point. 71 71 71 72 72 73
ii
2 The Limit Set for a Schottky Group Lemma 23.4 Theorem 21.2 Hausdorﬀ Dimension Proposition 20.1 Corollary 20.2 Theorem 23.1 21.1 Fuchsian Groups Proposition 23.3 The Sierpi´ ski Gasket n 23 SCHOTTKY GROUPS 23.1 Cantor Sets 20.6 Corollary 19.4 24 DEGENERATE SCHOTTKY GROUPS 24.2 Lipschitz maps preserve Hausdorﬀ dimension 21 CALCULATING THE HAUSDORFF DIMENSION Proposition 21.2 Schottky Groups Proposition 23.3 Schottky groups are free groups.7 Limit sets are perfect. 73 74 75 75 76 77 78 80 80 80 81 81 82 83 84 84 85 87 88 88 88 90 90 92 94 95 97 97 103 20 HAUSDORFF DIMENSION 20.1 The Cantor Set 22.5 22 EXAMPLES OF HAUSDORFF DIMENSION 22.1 23.1 Invariant Sets Lemma 21.1 How Schottky Groups Degenerate *24.Corollary 19. Limit sets are ﬁnite or uncountable.2 Hausdorﬀ distance Proposition 21.2 The von Koch snowﬂake 22.3 Invariant sets Proposition 21.2 Riemann Surfaces * iii . 23.
beginning with the Euclidean spaces R2 . so we will usually be considering straightforward matrix groups such as O(2). These groups act on the geometric space and we will want to study how they do so. We will look at examples of symmetry groups acting on various diﬀerent geometric spaces. C). S 2 . that we will look at later in the course. R3 and the spheres S 1 . There is also a even more important example. O(3).1 INTRODUCTION 1. SL(2. so we will look at the simplest and most symmetric spaces. the hyperbolic spaces. We will want to consider what makes them regular and prove that there are only ﬁve. Lecture 1 1 . Each of these geometric spaces has a metric and we will study the isometries that preserve the distances between any pair of points. We wish to have a large variety of symmetries. We will look at various patterns on the geometric spaces and the subgroups that preserve these patterns and so are symmetries of the patterns. A ﬁrst example is the “Platonic Solids”: These are the ﬁve convex regular solids.1 Symmetry This course will explore symmetry groups. These isometries are represented by matrices. SL(2. R).
A second is the “wallpaper patterns”: Andrew Crompton There are 17 diﬀerent groups of symmetries for wallpaper patterns. Lecture 1 2 .
Escher.Then there are tessellations of the hyperbolic plane: M. Circle Limit IV (1960) Lecture 1 3 .C.
c d cz + d Lecture 1 4 . The most important background is from the “Algebra and Geometry” course. x) → g · x which satisﬁes: (a) e · x = x for the identity e of G and any point x ∈ X. 1. The group SL(2. You also need to know about groups of matrices such as GL(N. the group GL(n. C). Although many of the earlier courses in the Tripos are relevant to this one. More formally we say that a group G acts on a set X if there is a map: G × X → X . We will review this brieﬂy. R) of invertible n × n real matrices acts on Rn by matrix multiplication: (M. So.Studying symmetry groups of the hyperbolic plane will lead us to consider limit sets which are fractal. there is rather little that is required in the way of background. each group element g ∈ G gives us a symmetry X → X. You will need to recall the deﬁnition of a group and how it acts on a set. x) → M x. For example: Sierpi´ski Gasket n Kleinian Limit Set and show how to calculate their dimension. R). We will also use the notion of a metric space and compactness from the “Metric and Topological Spaces” course.2 Group Actions We will be interested in groups G that act as symmetries of a space X. (b) g · (h · x) = (gh) · x for g.z → . For example. h ∈ G and any point x ∈ X. SO(N ). SU(N ) and the group of M¨bius transformations: o z→ az + b cz + d ad − bc = 1 acting on the Riemann sphere. (g. SL(N. C) of 2 × 2 complex matrices with determinant 1 acts on the Riemann sphere by: az + b a b .
Note that the set Bij(X) of all bijections from X to itself forms a group under composition. A fundamental set is a subset F of X that contains exactly one element from each orbit. the group G acts on X by (g. this shows that G = Stab(x)Orb(x) for each x ∈ X. This is a subgroup of G. The stabilizer Stab(x) is {g ∈ G : g · x = x}. gStab(x) → g · x is a bijection. Proof: Note that g · x = h · x ⇔ (h−1 g) · x = x ⇔ h−1 g ∈ Stab(x) ⇔ g ∈ hStab(x) So α maps each coset gStab(x) onto the element g · x in the orbit of x. The relation x ∼ y ⇔ y = g · x for some g ∈ G is an equivalence relation on X with the orbits as the equivalence classes. α is well deﬁned and a bijection. 5 .1 Orbit – Stabilizer theorem If a group G acts on a set X. The stabiliser of each point x ∈ R2 is the identity alone. The quotient X/G is the set of equivalence classes. For example. Let F be the strip x1 x2 Lecture 1 ∈ R2 : 0 x1 < 1 . x → g · x is a group homomorphism. Show that. The group action is faithful or eﬀective when θ : G → Bij(X) is injective. Exercise: 1. For condition (a) shows that θ(e) = IX and condition (b) shows that θ(g) ◦ θ(h)x = θ(gh)x. x) → θ(g)x . Proposition 1. then the map α : G/Stab(x) → Orb(x) . If the group G acts on X then the map θ : G → Bij(X) with θ(g) : X → X . It is a subset of the space X. We can also consider the quotient of X by the group action. let Z be the additive group of integers acting on the plane R2 by translations: n · x = x + ni . the orbit Orb(x) is the set {g · x : g ∈ G} of points that x is mapped to by G. Suppose that the group G acts on the set X. For each x ∈ X. while the orbit of x is the set {x + ni : n ∈ Z} of all translations of x by integer multiples of the unit vector i. Consequently. for any group homomorphism θ : G → Bij(X). When G is a ﬁnite group.
so we can identify this cylinder with the quotient X/G. Exercise: 2. Suppose that P is also a fundamental set for the action of Z × Z on R2 . v be two vectors in R2 and let P be the parallelogram: {λu + µv : 0 λ < 1 and 0 µ < 1} . Each orbit corresponds to a unique point of this cylinder. Let G be the group they generate. 3. If we identify these points on the opposite edges then we obtain a cylinder. Show that additive the group Z × Z acts on the plane R2 by n1 n2 and that the unit square S = · x1 x2 = x1 + n1 x2 + n2 is a fundamental set. Consider the two maps: A: x1 x2 → x1 x2 + 1 . Lecture 1 6 . Show that u= a c . Let u. Two y2 y2 points in closed strip F are only equivalent to one another when they are points Each point y = 0 y2 and 1 y2 on the opposite edges. Hence x1 : 0 x1 < 1 and 0 x2 < 1 x2 show that we can identify the quotient R2 /Z × Z with a torus.y1 y1 ∈ R2 is equivalent to a point in F . Find a fundamental set and hence describe the quotient R2 /G. so F is a fundamental set. c. B: x1 x2 → x1 + 1 −x2 acting on the plane R2 . b. d with ad − bc = ±1. Is G Abelian? Find the orbit of a point x under this group. v= b d for some integers a.
if B : RN → RN is a Euclidean isometry that ﬁxes the origin. T (y)) = d(x. M t M = I. This is called a hyperplane. Hence. y) for all x. It is straightforward to check that R(x) − R(y) = (x − y) − 2((x − y) · u)u = x − y so the reﬂection is an Euclidean isometry. the set π = {x : x · u = λ} is a translation of a vector subspace of dimension N − 1. Lemma 2. y ∈ RN is given by x · y = x1 y1 + x2 y2 + . The Euclidean metric is given by d(x. A map T : M → M is an isometry if it is invertible and preserves distances. Our ﬁrst lemma shows that any such isometry is an orthogonal linear map.2 ISOMETRIES OF EUCLIDEAN SPACE 2. .1 Deﬁnitions Let M be a metric space with d as its metric. Lecture 2 7 . For any vector a ∈ RN the map T : x → x + a is clearly an isometry and is called the translation by a. Let u be a unit vector. y) = x − y . y ∈ M . Any vector x can be written as (x · u)u + x⊥ with x⊥ perpendicular to u. These matrices give us the orthogonal group O(N ) = {M : M is an N × N real matrix with M t M = I} . + xN yN . so T (x) · T (y) = x · y for all x. The set of isometries of M form a group Isom(M ) under composition. Since the inner product is given by x·y = xt y. We can use this to deﬁne the norm or length of a vector x as √ √ x = (x · x) = (x1 x1 + x2 x2 + . We wish to describe all of the isometries of Euclidean space. + xN xN ) . . Recall that a linear map T : RN → RN is orthogonal when it preserves the inner product. the B is an orthogonal linear map. y ∈ RN .1 Orthogonal maps as Euclidean isometries Every orthogonal linear map B : RN → RN is an isometry for the Euclidean metric. Euclidean space consists of RN with the Euclidean metric. so d(T (x). Conversely. Hence we deﬁne reﬂection in the hyperplane π to be R : x → x − 2(x · u − λ)u . In this section we want to study the isometries of the Euclidean space EN and especially of the Euclidean plane. . this is equivalent to the N ×N matrix M that represents T satisfying x t M t M y = xt y for all x. The inner product of two vectors x. Then reﬂection in π should leave x⊥ unaltered but map tu to (2λ − t)u. For each real number λ. . which is deﬁned in terms of the inner product (or scalar product). y ∈ RN . It is easiest to do this by ﬁrst considering those isometries that ﬁx the origin.
suppose that B ∈ O(N ). y ∈ RN .Proof: Let B be an orthogonal linear map. Then the translation T by v is an isometry and B = T −1 ◦ A : x → Ax − A0 is an isometry that ﬁxes the origin. (Ben ) is another orthonormal basis.1 shows that B is an orthogonal linear map with Ax = Bx + v. Hence Lemma 2. Let (en ) be the standard orthonormal basis for RN . We need to show that it is linear. Then A is the composition of the isometries B and translation by v. 0)2 − d(x. By)2 = Bx − By2 = B(x − y) · B(x − y) = (x − y) · (x − y) = d(x. Hence B: and so B is a linear map. Now suppose that B is an isometry of EN that ﬁxes the origin. this map is an isometry for any vector v and any orthogonal matrix B. y)2 shows that Bx · By = x · y for any vectors x. For any vector x we have x= Also. Conversely. y)2 so B is an isometry. Now let A be any isometry of Euclidean space. Bx = = because B preserves the inner product. B is an orthogonal linear map. Lecture 2 8 . Because B preserves the inner product. The polarization identity: 2x · y = x2 + y2 − x − y2 = d(x.2 Euclidean isometries are aﬃne If A : EN → EN is an isometry of the Euclidean space EN . v ∈ RN and Ax = Bx + v. Proof: Let A be an isometry of EN and set v = A0. This shows that B preserves the inner product. The translation T by the vector A0 is an isometry so B = T −1 ◦ A will be an isometry that ﬁxes the origin. Consequently. Conversely. Hence we obtain: Proposition 2. 0)2 + d(y. Then d(Bx. so it is also an isometry. xn en → xn Ben (Bx · Ben )Ben (x · en )Ben (x · en )en . then there is a vector v ∈ RN and an orthogonal matrix B with A(x) = Bx + v for x ∈ RN .
this shows that A is reﬂection in the line translated by p. then we can split the vector v into a part v1 perpendicular to and a part v2 parallel to . then A is a translation. The ﬁrst case gives a rotation about the origin through an angle θ (or the identity when θ = 0). Now consider an isometry A of the Euclidean plane that does not ﬁx the origin. This ﬁrst translates p back to the origin. 2 Note that the determinant of the orthogonal matrix B must be either +1 or −1. When B is the identity.2. (c) A rotation about some point c ∈ E2 . Hence we can choose an angle θ ∈ [0. 9 . First suppose that B is an isometry that ﬁxes the origin. An orientation reversing isometry of E2 is: (d) A reﬂection. When B is a rotation through an angle θ. c sin θ The vector b d must be a unit vector orthogonal to b d = − sin θ cos θ or a . 2π) with a cos θ = . (e) A glide reﬂections. However. then applies B. then we obtain another isometry T ◦ B ◦ T −1 . (b) A translation. The linear map I − B maps onto the vector subspace of vectors perpendicular to . that is reﬂection in the line translated by p followed by a translation parallel to . If we conjugate B by a translation T : x → x + p. we can always choose a vector p with (I − B)p = v. Proposition 2. when v2 = 0.3 Isometries of E2 An orientation preserving isometry of the Euclidean plane E2 is: (a) The identity. When B is a rotation. that is a reﬂection in a line Lecture 2 followed by a translation parallel to .2 Isometries of the Euclidean Plane We can use the previous results to describe all of the isometries of the Euclidean plane: E2 . This means that the columns of B are unit vectors orthogonal to one another. so there are just two possibilities: c sin θ − cos θ . This means that T ◦ B ◦ T −1 (x) = Bx + (I − B)p = Bx + v so the isometry A is a rotation about the point p. then T ◦ B ◦ T −1 is reﬂection in the line T ( ). When v2 = 0. then A is a glide reﬂection. y) : y = (tan 2 θ)x} at an angle 1 θ from the xaxis. When B is a reﬂection in a line .2 shows that Ax = Bx + v for some B ∈ O(2) and some vector v ∈ R2 . then T ◦ B ◦ T −1 is a rotation about T 0 = p through an angle θ. so we can choose a vector p with (I − B)p = v1 . Then B= a b c d ∈ O(2) . Proposition 2. This means that T ◦ B ◦ T −1 (x) = Bx + (I − B)p = Bx + v1 so Ax = T ◦ B ◦ T −1 (x) + v2 . The 1 second gives reﬂection in the line {(x. It is +1 for the identity or rotations that preserve the orientation of the plane and −1 for reﬂections that reverse the orientation. When B is a reﬂection in a line through the origin. and then translates the origin back to p.
(b) A translation. (c) A rotation about some line . (g) A rotatory reﬂection.4 Isometries of E3 An orientation preserving isometry of Euclidean 3space E3 is: (a) The identity.A similar. Proposition 2. followed by reﬂection in a plane perpenfollowed by a translation parallel to . (f) A glide reﬂection. that is a rotation about some axis dicular to . (d) A screw rotation. but slightly more involved. argument gives the corresponding result for Euclidean space E3 . that is a reﬂection in a plane Π followed by a translation parallel to Π. that is a rotation about some line An orientation reversing isometry of E3 is: (e) A reﬂection in some plane Π. Lecture 2 10 .
1 The map φ : Isom(EN ) → O(N ) given above is a group homomorphism with kernel equal to the group Trans(EN ) of all translations of EN . this complement is equal to the coset JIsom+ (EN ). either G = G+ contains only orientation preserving isometries or else G+ is of index 2 in G. Proposition 3. When A : x → Bx + v. Consequently. Then ε : Isom(EN ) → {−1. +1} is a group homomorphism. These form the normal subgroup ker εG of G.2 that every isometry A of EN can be written as A : x → Bx + v with B ∈ O(N ) and v ∈ RN . Hence we can deﬁne a map φ : Isom(EN ) → O(N ) by A→B . If G is any subgroup of Isom(EN ). Recall that. A2 be two isometries with Ak x = Bk x + vk . then the restriction εG : G → {−1. Since ε : Isom(EN ) → {−1. When we study subgroups like G we often begin by looking at G+ and then consider how we can add orientation reversing isometries. Then A2 ◦ A1 (x) = A2 (B1 x + v1 ) = B2 (B1 x + v1 ) + v2 = (B2 B1 )x + (B2 v1 + v2 ) . then A ◦ T ◦ A−1 (x) = x + Bt . the subgroup Isom+ (EN ) has index 2 in Isom(EN ). For any orientation reversing isometry J. This means that an isometry A acts on the translations by conjugation. +1} is a group homomorphism. This means that it has just two cosets Isom+ (EN ) = ε−1 (+1) and the complement Isom− (EN ) = ε−1 (−1).1 Quotients of the Isometry group Recall from Proposition 2. +1}. If A : x → Bx + v and T is the translation T : x → x + t. Hence. The isometry A : x → Bx + v is in the kernel of φ when φ(A) = B = I. The kernel of ε is the group Isom+ (EN ) of orientation preserving isometries of EN . +1}. +1} is a group homomorphism onto {−1. This means that A is a translation. the inverse images α−1 (h) are the cosets of ker α in G. Lecture 3 11 . There is another important homomorphism from Isom(EN ) that tells us whether an isometry preserves or reverses orientation. The group Trans(EN ) of translations is a normal subgroup of the isometry group. This action will be very important to us later when we look at crystallographic groups. The number of cosets is equal to the number of elements in H and is called the index of ker α in G. for any surjective group homomorphism α : G → H. Proof: Let A1 . φ(A2 ◦ A1 ) = B2 B1 = φ(A2 )φ(A1 ) which shows that φ is a group homomorphism. This is a normal subgroup of Isom(EN ).3 THE ISOMETRY GROUP OF EUCLIDEAN SPACE 3. since it is the kernel of the homomorphism φ. Let G+ = G ∩ Isom+ (EN ) be the orientation preserving isometries in G. The image of εG is either {+1} or {−1. we deﬁne ε(A) to be det B.
Let A be an isometry of EN .2 Matrices for Isometries The Euclidean space EN is the same set as RN but it has diﬀerent properties. z) = y − z . This metric behaves as you would expect. Any vector x ∈ RN then n=1 x corresponds to a point ∈ EN . Lecture 3 12 . M )2 = (K − M ) · (K − M ) = i. it is often more convenient to identify it with a hyperplane in RN +1 that does not go through the origin. When we think of EN as the hyperplane {y : yN +1 = 1} we see that A sends the vector x 1 to B 0 v 1 x 1 = Bx + v 1 .j (kij − mij )2 . In the vector space RN the origin 0 is a special point. So A is represented by the (N + 1) × (N + 1) matrix B 0 v 1 . For the (N + 1) × (N + 1) matrices are a vector space of dimension (N + 1)2 and have a inner product given by N +1 N +1 K · M = tr K t M = i=1 j=1 kij mij . Rather than thinking of EN as equal to RN . This can make computations simpler and it also gives us a natural metric on the group.2 shows that A maps a vector x ∈ RN to the vector Bx + v for some B ∈ O(N ) and v ∈ RN . The Euclidean metric on EN is just the restriction of the usual 1 metric on RN +1 : d(y.3. Show that an (N + 1) × (N + 1) matrix maps the hyperplane EN isometrically onto itself if and only if B v M= 0 1 for some matrix B ∈ O(N ) and some vector v ∈ RN . However. This means that we can regard the group Isom(EN ) as a group of matrices. in Euclidean space it is not since translations can be used to send 0 to any other point. Then we can represent every isometry as an (N + 1) × (N + 1) matrix. Exercise: 4. We can then deﬁne the distance between two matrices K and M as d(K. A sequence (M (k)) of matrices converges to a limit matrix M if and only if the sequence (M (k)ij ) of ij entries converges to Mij for each pair ij. Then Proposition 2. Set EN to be the hyperplane {y = (yn )N +1 ∈ RN +1 : yN +1 = 1}in RN +1 .
with order N say.3 Finite Groups of Isometries of the Plane In Proposition 2. . Choose any point a ∈ EN . v ∈ RN . . So we see that G is dihedral of order 2N . N − 1) about some point c and N reﬂections in lines through c. . N − 1}. If G consists only of orientation preserving isometries. 2. There are only N such rotations about c so all of them must lie in G+ . Hence A must be a rotation through an angle 2πk/N for some integer k ∈ {0. 1. . R2 M. . . The orientation preserving isometries permute the shaded regions while the orientation reversing isometries interchange the shaded and unshaded regions. .3.2 Finite subgroups of Isom(E2 ). . This ﬁxes c so it must be a reﬂection in a line through c. The groups described in this proposition are the symmetries of a regular N gon centred on c. This means that A is aﬃne. . RN −1 M are all distinct and are reﬂections in the line obtained by rotating about c through angles πk/N for k = 0. . The homomorphism ε : G → {−1. Proof: The centroid c is ﬁxed by all of the isometries in G. . . either the orientation preserving isometries for the cyclic group or all the isometries for the dihedral group. The centroid of the orbit of a: c= 1 G T (a) T ∈G then satisﬁes A(c) = c for each A ∈ G. . 2. N − 1) about some point c. then G = G+ is cyclic. The products M. Therefore all of the elements of G ﬁx the point c. This is illustrated in the picture below. A ﬁnite subgroup G of Isom(E2 ) is either a cyclic group consisting of N rotations through angles 2πk/N (k = 0. so A λ j xj = λj A(xj ) provided that λj = 1 . Let G be a ﬁnite subgroup of Isom(EN ). . RM. so G = 2N . Then each transformation A ∈ G+ is a rotation about c with AN = I. 2. +1} maps G+ onto +1 and the coset G+ M onto −1. there must be an orientation reversing isometry M in G \ G+ . Lecture 3 13 . . 1. N − 1 respectively. 1. Proposition 3. . Hence G+ must be the cyclic group of order N generated by a rotation R about c through an angle 2π/N . .2 we showed that each isometry A of EN is of the form x → Bx + v with B ∈ O(N ). The subgroup G+ = G ∩ Isom+ (E2 ) is also a ﬁnite group. . 1. 2. or else a dihedral group consisting of N rotations through angles 2πk/N (k = 0. Otherwise. . .
if R is a rotation about a point on . Let M be the reﬂection of E2 in the line . So the composition of the two reﬂections is a rotation about the point P of intersection through twice the angle from to onto . so M RM −1 = R−1 . reﬂection in the xaxis and then simply use matrices. Similarly. Consequently M M = RM R−1 M = RM M R = R2 . then there is a rotation R about P that maps Hence M = RM R−1 . that is T −1 . then we can ﬁnd a translation T perpendicular to both line that maps onto . say. 3.4 Compositions of Reﬂections We wish to ﬁnd the result of composing two reﬂections. Then the conjugate M T M −1 is translation in the reverse direction. We will adopt a diﬀerent approach. then the conjugate M RM −1 is rotation about the same point in the opposite direction. The simplest way to do this is to choose coordinates so that one of the reﬂections is. Lecture 3 14 .Regular polygon of order N = 7. Let T be a translation perpendicular to . You should do this. Now suppose that M is reﬂection in a second line . Consequently M M = T M T −1 M = T M M T = T 2 . Hence M = T M T −1 . If the two lines and meet at a point P . So the composition of two reﬂections in parallel lines is a translation perpendicular to those lines by twice the distance between them. If the two lines and are parallel. So M T M −1 = T −1 and hence M T = T −1 M .
. ﬁnite group of symmetries.4 FINITE SYMMETRY GROUPS OF EUCLIDEAN SPACE If P is an object in E3 . Hence we need only consider which rotations are symmetries. For the group of symmetries acts on the vertices of the polygon P with the orbit of u being all N vertices and the stabilizer of u being only the identity. By joining each vertex of P to both 2v and −2v we get the double cone on P . By translating. . . . . Then each symmetry A ∈ G must be in O(N ). . Lecture 4 15 . . RN −1 u are the vertices of a regular N gon P . Each of these is a rotation through an angle π. R. RN −1 is a symmetry of the cone or double cone on P . Once again. SR2 . Hence we can equally well look at the ﬁnite subgroups of Isom(S N ). It is usually easy to see what the symmetries are and to verify that we have them all by using the orbit – stabilizer theorem. For the single cone. 4. such as one of the Platonic solids. . 3.. R2 . By joining each vertex of P to 2v we get the cone on P . = For the double cone on P . Then the points u. . 4. The only orientation preserving symmetries of E3 that ﬁx the origin are the identity and rotations about an axis through the origin. . This means that A maps the unit sphere S N −1 = {x ∈ RN : x = 1} to itself isometrically. Let u be a unit vector orthogonal to v. R2 . So the orbit – stabilizer theorem shows that the orientation preserving symmetry group of the cone on P has N elements and so is the cyclic group {I. these are the only orientation preserving symmetries. . . we obtain a nontrivial. . we can ensure that this point c is at the origin. SRN −1 are the others. Let v be a unit vector in R3 and R the rotation about this vector through an angle 2π/N for some N = 2. there are other symmetries. The aim of this lecture is to prove that these give us all of the ﬁnite subgroups of Isom(E3 ). RN −1 } ∼ CN . R2 u. the orbit – stabilizer theorem shows that this is all of the orientation preserving symmetries. . We will begin by looking only at the orientation preserving symmetries. . Ru. Recall from §3 that every ﬁnite subgroup G of Isom(EN ) must ﬁx a point c ∈ EN . 2v 2v 0 RN −1 u R2 u RN −1 u 0 R2 u Ru Ru u u −2v Regular cone Regular double cone It is clear that each of maps I. .1 Examples of Finite Symmetry Groups We begin by giving examples of objects with ﬁnite symmetry groups. . The rotation S about u through an angle π is one and SR. then the isometries of E3 that are symmetries of P form a subgroup. . R. By choosing P to be highly symmetric. .
as are the groups for the dodecahedron and the icosahedron. 4. JSRN −1 . The others are JR. dodecahedron and icosahedron. The groups are then cyclic of order N for N = 1. so the stabilizer in G+ is cyclic with order equal to the number of faces meeting at each vertex. We will concentrate on the groups of orientation preserving symmetries. and the rotational symmetry groups of the Platonic solids. . Since J commutes with all the other symmetries. 2. .. (b) A double cone on a regular plane polygon.1 Finite symmetry groups in Isom+ (E3 ) Let G be a ﬁnite subgroup of Isom+ (E3 ) consisting of orientation preserving isometries.2 Finite Subgroups of Isom(E3 ). So the full symmetry group is isomorphic to D2N . Now we aim to prove that the ﬁnite groups described in the previous section are the only ones that exist. octahedral and icosahedral groups. The regularity of the Platonic solids shows that there are symmetries that move any vertex to any other. .) If we include the orientation reversing symmetries we obtain ﬁnite groups with twice as many elements. (e) A regular icosahedron. we see that the full symmetry group is isomorphic to D2N × C2 . dihedral of order 2N . Theorem 4. We can do a similar analysis when P is one of the Platonic solids. . The stabilizer in G+ consists of all the rotations that ﬁx that vertex. These groups act on the vertices. say P . Recall that the Platonic solids are the regular tetrahedron. which are reﬂections in planes through v. For the cone. These must permute the faces that meet at the vertex. there are N reﬂections in planes through v. centred on the origin. We have not shown that these ﬁnite symmetry groups really exist.They form a group isomorphic to the dihedral group D2N . . Let G be the group of symmetries of one of these. octahedron. Lecture 4 16 .) There are also orientation reversing symmetries. we would need to show that the Platonic solids exist and their symmetry groups act as claimed on the vertices. Then it has N reﬂections. This was done in the Geometry course and various approaches to it are outlined in the example sheets. For the double cone. A Platonic solid and its dual have the same symmetries. Then G is is the orientation preserving symmetry group of one of the following: (a) A cone on a regular plane polygon. The rotations through angle π act of the plane of P in the same way as the reﬂections but interchange the halfspaces above and below P . (d) A regular octahedron. . and JS. . (We will see later that the groups are not all distinct. The various numbers are: Orbit (vertices) Tetrahedron Cube Octahedron Dodecahedron Icosahedron 4 8 6 20 12 Stabilizer (faces at each vertex) 3 3 4 3 5 G+  12 24 24 60 60 These give ﬁnite groups of orientation preserving symmetries. . which are rotatory reﬂections. . . JRN −1 . (Note that we usually think of the dihedral group as the plane symmetries of P . . called the tetrahedral. JSR. To do so. 3. the reﬂection J in the plane of the polygon P is one symmetry. (c) A regular tetrahedron. so the groups for the cube and octahedron are the same. JR2 . cube. so there is one orbit consisting of all the vertices. Then G+ is the subgroup of orientation preserving symmetries.
2− 2 1 1 1 1 − =2− =2− − G S1 S2 G S2 and so S2 = G and Ω2  = 1. Therefore X = 2(G − 1). The orbit – stabilizer theorem shows that each vector u ∈ Ωj has a stabilizer of order Sj = G/Ωj . Hence Ω consists of two unit vectors v and −v which are ﬁxed by each isometry in G. 2>2− and so J is 1. so 1− Hence. Hence. ΩJ be the diﬀerent orbits in Ω. Then each nonidentity element of G is a rotation about an axis through 0. Equation (∗) now yields 2 S2 Lecture 4 1 1 1 2 1 + = + > S1 S2 2 G 2 17 1+ 2 >1 G . This implies that S1 = G and Ω1  = 1. Ω2 . Order the orbits so that S1 J = 2 then (∗) gives S2 2− so S1 S3 2. . If 2− 2 S1 2 = G J 1 Sj 1 . Dividing by G we see that 2− 2 = G J 1− j=1 1 .Proof: First translate so that G ﬁxes the origin. u ∈ S 2 and Au = u} .2 . Let Ω1 . Each A ∈ G \ {I} is a rotation and so ﬁxes exactly two unit vectors. S1 S2 S3 G This implies that 3 S3 so S3 = 2. This shows that G is the symmetry group of a cone as in (a). u) : A ∈ G \ {I}. equation (∗) gives 1 1 2 1 + + =1+ . 2 1− j=1 1 Sj 1 2J 1 1 2 =2− − G S1 S2 G. Clearly there are no solutions to (∗) with J = 1. . So J J X = j=1 (Sj − 1)Ωj  = j=1 G − Ωj  . The group G is then a ﬁnite isometry group of the plane orthogonal to v so it is cyclic by Proposition 3. . Now we count the number of pairs in the set X = {(A. . When J = 3. Sj ∗ Each stabilizer of u ∈ Ω has order at least 2. each u ∈ Ω gives rise to Stab(u) − 1 pairs in X. Alternatively. 2 or 3. Then Ω is a ﬁnite set and G acts on it. Let Ω be the set of unit vectors that are ﬁxed by by some nonidentity element of G.
octahedron and icosahedron respectively. The stabiliser of v is a cyclic group of order 4. . So it must be in Ω1 . Lecture 4 18 . R3 w all lying in the plane through 0 orthogonal to v. Hence the points of Ω1 are the 6 vertices of a regular octahedron. Rw. The possibilities are: S1 3 4 5 Ω1  4 6 12 S2 3 3 3 Ω2  4 8 20 S3 2 2 2 Ω3  6 12 30 G 12 24 60 We need to show that these correspond to the symmetry groups of the tetrahedron. Ω3 the midpoints of edges. We will consider the middle case as an example. Choose u as one of the points in Ω3 . Ω2 the centres of the faces. The orbit Ω1 has has just two points. Ω1 is the set of vertices. Here Ω1 has 6 points. 4 or 5. The stabiliser of each is a cyclic group of order S1 = 4. . so it is cyclic generated by a rotation R. . When S2 = 2 we have 1 2 = S1 G which gives S1 = N. The stabilizer of u has order 2. . S2 = 2. Ω1 is the set of vertices. Then the others are Ru. the ﬁrst row in the table above gives us a regular tetrahedron. Ω3 the midpoints of edges. The ﬁnal row gives a icosahedron. RN −1 u. This maps v to −v. so it contains a rotation S of order 2. The points of Ω2 are the vertices of a cube. Ω2 the centres of the faces. There remain 4 other points in Ω1 and these must be w. This is the dual of the octahedron. The stabilizer Stab(v) is a ﬁnite group of N rotations about v. The polyhedron and its dual have the same symmetry group. The tetrahedron is dual to another tetrahedron. S3 = 2 and G = 2N . R2 u. Note that the points of Ω2 are the midpoints of the faces of this octahedron and the points of Ω3 are the midpoints of the edges. Let v be one of them. R2 w. Now −v is also ﬁxed by R and has the same stabiliser. It is now apparent that each element of G is a symmetry of a double cone as in (b). Ω1 Ω2 Ω3 In a similar way. let R be a generator. When S2 = 3 we have 1 2 1 1 = + > S1 G 6 6 so S1 = 3.which implies that S2 = 2 or 3. Choose one point v ∈ Ω1 . The dual is a dodecahedron.
5 THE PLATONIC SOLIDS 5.1 History The Platonic solids have been known and studied for a very long time.369BC) was the ﬁrst to prove that there were only ﬁve convex regular polyhedra. the cube for earth. Stones carved into polyhedral shapes date from about 2000BC in Scotland. the octahedron for air and the icosahedron for water.com/virtualpolyhedra/neolithic. Theaetetus (c. 360BC and follows the Pythagoreans in giving them mystical signiﬁcance. 417BC . Euclid devoted the 13th and last book of his Elements to the Platonic solids.html. In the 18th Century Kepler pursued Plato’s mystical interest but also tried to use the Platonic solids to describe the known universe.georgehart. Lecture 5 19 . Four of them represented the four elements: the tetrahedron for ﬁre. (See http://www. This association was justiﬁed on the grounds that the icosahedron is the smoothest of the polyhedra while the tetrahedron is the sharpest. In the Mysterium Cosmographicum (1596) he suggested that the radii of the orbits of the ﬁve known planets could be found by inscribing the solids one inside another. The dodecahedron represented the entire universe with the twelve faces showing the twelve signs of the zodiac. Plato refers to the solids in the Timaeus c.) The Pythagoreans were aware of at least some of the solids and endowed them with a mystical signiﬁcance.
Lecture 5 20 .
e. regularity means that each edge looks the same and each vertex looks the same. We need to consider what regularity means. e . the group G will contain symmetries that ﬁx a vertex v and maps any one edge at v to any other. Hence it is one of the groups found in §4 . convex polyhedra. Indeed. This means that the vertices of P must be one of the orbits Ω1 . e. for example. edge or face to any other. Note that there are nonconvex polygons that also have this property. necessary that we join up the vertices in the expected way. Once we know where a symmetry sends each vertex. See. For example. It is not. This gives two tetrahedra which together form a nonconvex regular polyhedron. This condition certainly implies that the group of symmetries of P acts on the set V of vertices and V is a single orbit. for any two ﬂags (v. Ω3 found in the proof of Theorem 4. and e is one of the boundary edges of F . the nonconvex regular polygon above. Furthermore. For a polygon. This means that there are symmetries of the polygon that map each edge to any other and each vertex to any other. F ) consisting of a vertex v. We can join each vertex to the 3 other vertices at a distance 2 rather than those at unit distance. Hence. Note that the intersection of these two tetrahedra is the octahedron dual to the original cube. We will say that the polyhedron is regular if. if we are not in the trivial situation where only two edges meet at each vertex. A ﬂag for the polyhedron P is a triple (v. F ) and (v . Hence the set of vertices is one of the orbits we considered in §4 . an edge e and a face F such that v is one end of e.5. Ω2 . F ) there is a symmetry of P that maps one ﬂag to the other.2 Regularity The Platonic solids are the only regular.1 . consider the √ vertices of a cube with side length 1. This means that the symmetry group G of P is a ﬁnite group of isometries. however. edges and faces. For example: We will consider polyhedra that have a ﬁnite number of vertices. Similar possibilities arise for polyhedra. each vertex has a nontrivial stabilizer. we can determine where each edge and face goes and hence ﬁnd the symmetry completely. For the polyhedron to be regular we want to there to be symmetries that map any vertex. then the vertices can not be Ω3 which has stabilizers of order 2. Lecture 5 21 .
) Finally. then it is the identity. If a symmetry ﬁxes all four vertices. 3}. The subgroup Sym+ (C) is isomorphic to S4 . The pair {p. This permutes the four vertices of a tetrahedron T . −1} . 5}. J ◦ S is another symmetry that maps each tetrahedron onto itself. that Sym(T ) has 24 elements. {3. Then T maps each cube to itself. we can easily do so. A vertex v of the dodecahedron is a vertex of exactly two of the cubes and these two cubes have only the vertices v and J(v) = −v in common. Consider ﬁrst the tetrahedral group Sym(T ). In the latter case. By looking at each rotational symmetry of the dodecahedron we can check that θ actually maps into the alternating group A5 . It also commutes with every other symmetry of the cube. Hence θ is injective. so q π− This simpliﬁes to (p − 2)(q − 2) < 4 so the only possible solutions are {3. when S(T + ) = T − . 1) (J ◦ S. Hence we see that the homomorphism θ : Sym(C) → Sym(T + ) × {1. (We can also prove this by considering how a symmetry of the cube acts on the four long diagonals joining a vertex of V to its antipodal vertex. denote these by T + and T − .5. Now consider the symmetry group of a cube centred on the origin. Since T is an isometry. The isometry Jx → −x interchanges T + and T − . ε(T )) is an injective group homomorphism. The orientation preserving symmetries of the dodecahedron form a group isomorphic to A5 . so this is an isomorphism. Thus the full symmetry group of the regular tetrahedron is isomorphic to S4 . so θ must be an isomorphism. we must have T = I or T = J. or the dual icosahedron. We already know.1 showed which groups could arise as the ﬁnite subgroups of Isom(E3 ). is injective. from the orbit – stabilizer theorem. The regularity certainly implies that each face is a regular polygon and an equal number of these faces meet at each vertex. we have not identiﬁed which groups these are. There are ﬁve cubes embedded inside the dodecahedron (or equivalently ﬁve octahedra within which an icosahedron is embedded). By using some of the nonconvex polyhedra with these symmetry groups. Here p and q are at least 3. 3}. Choose one vertex v. 4}. This is the same as the symmetry group of the dual octahedron which has vertices at the centres of each face of the cube. 5. The orbit – stabilizer theorem shows that there are an equal number 120 of elements in Sym(D) as in A5 × C2 . T must map each vertex either to itself or to the antipodal vertex J(v). T → (θ(T ). {4. These give the ﬁve Platonic solids. Say the faces are pgons and q meet at each vertex. {3. The angle between two edges that meet at a vertex is π − 2π/p. At each vertex q of these meet. There are two tetrahedra embedded in this cube with vertices at the vertices of the cube. This shows that the map Sym(D) → A5 × C2 . S → (S. consider the symmetry group of the dodecahedron D. 2π p < 2π . −1) when S(T + ) = T + . {5.3 Convex Regular Polyhedra Suppose that P is a convex polyhedron that is regular. Hence. So the group Sym+ (T ) is isomorphic to the alternating group A4 . so we get a group homomorphism θ : Sym(D) → S5 into the symmetric group on the cubes. and so we obtain a group homomorphism θ : Sym(T ) → S4 into the symmetric group S4 on the vertices.4 The Symmetry Groups In Theorem 4. A symmetry S of the cube either maps each tetrahedron onto itself or else interchanges the tetrahedra. However. q} is called the Schl¨ﬁ a symbol for the polyhedron. A symmetry of the dodecahedron permutes these 5 embedded cubes. Suppose that the symmetry T is in the kernel of this homomorphism. The q vertices adjacent to v are the vertices of a regular qgon. 3}. Counting elements shows that the full symmetry group Sym(C) is isomorphic to S4 × C2 . By looking at the individual symmetries we see that the orientation preserving symmetries correspond to the even permutations of the vertices. Lecture 5 22 .
We can also look for a fundamental set for the symmetry group acting on all of E3 . the midpoint of an edge of P and the centre of the face F . we can take the cone on the triangle found above with its vertex at the centroid of P . For this. Lecture 5 23 . The corners of this triangle are a vertex of P . It is easy to see that a fundamental set for this dihedral group is a closed triangle as shown in the diagram below. The triangle thus corresponds to a ﬂag for P . we can restrict our attention to one face F and look for a fundamental set of Stab(F ).5 Fundamental sets The symmetry group G of a Platonic solid P acts on the faces of P so we can look for a fundamental set. Since G permutes the faces. This stabilizer is the symmetry group of the polygonal face.Five cubes embedded in a regular dodecahedron 5. For each ﬂag we obtain a copy of the fundamental triangle and these tessellate the surface of the polyhedron.
Let G be a matrix group. Proposition 6. A point x ∈ M is isolated if there is some neighbourhood V of x that contains no point of M except x. This means that there is a δ > 0 with d(x. We say that G is a discrete group if it is discrete for this metric. These groups correspond to additive subgroups of RN . Lecture 6 24 . Z) of 2 × 2 matrices M = is discrete. d ∈ Z and ad − bc = 1 d(M. so G is a discrete group if and only if Λ is a discrete subset of RN . Every ﬁnite group is certainly discrete. Similarly.6 LATTICES Let M be a metric space with metric d. For example. then the set Λ = {T (0) : T ∈ G} is an additive subgroup of RN . Show that G is a discrete matrix group if and only if there is no sequence of nonidentity elements elements gn ∈ G that converge to the identity. For example. Note that d(T1 . Zw1 + Zw2 is a lattice in R2 for any two linearly independent vectors w1 . because any matrix M = I satisﬁes a b c d with a. the group SL(2. We will show that these are the only lattices in R2 . c. Then G has a natural (Euclidean) metric. First we will look at the discrete groups of translations. the set Zw1 for any nonzero vector w1 is a lattice in R2 . Conversely. We call a discrete additive subgroup of RN a lattice in RN . I)2 = (a − 1)2 + b2 + c2 + (d − 1)2 1. then G = {x → x + v : v ∈ Λ} is a subgroup of Trans(EN ). then every other point T ∈ G is also isolated. Let G be a subgroup of Trans(EN ). For example. each point of Z is isolated. to check that a group is discrete we need only check that I is isolated. b. For the multiplication G → G . Note that the particular metric on M is not important only the topology. T2 ) = v1 − v2 . w2 . The metric space M is discrete if each point of M is isolated.1 Lattices in R Each lattice in R is of the form Zw for some w ∈ R. Hence. if Λ is any additive subgroup of RN . If the identity I is isolated in G. Exercise: 5. A → TA is continuous and has a continuous inverse A → T −1 A. Discrete groups can be inﬁnite but they can not be too large. y) > δ for all y ∈ M \ {x} .
r). The vectors w1 . λ1 − λ2  > δ for each pair of distinct points λ1 . there are vectors v ∈ Λ \ Zw1 . is a fundamental set for the lattice Λ acting on E2 . 0) = v ⊥ 2 + t2 w1 2 . So b = b − ka is also in Λ and has b  = t a < a . so any vector v ∈ Λ can be written as a linear combination v = t1 w1 + t2 w2 . or Zw1 + Zw2 for a pair of linearly independent vectors w 1 . w 2 ∈ R2 . The choice of a tells us that b must be 0. we can choose a vector v ∈ Λ \ Zw1 with v ⊥  = d(v. The vectors w1 . Rw1 ) = t2 d(w2 . lattice Z 0 2 3 The proof also shows that the parallelogram {t1 w1 + t2 w2 : 0 t1 . Otherwise. We certainly have a ∈ Λ. In the ﬁrst case we have Λ = Zw for w = 0. The quotient E2 /Λ is obtained by identifying the parallel sides of this parallelogram to give a torus. Rw1 ) = v ⊥  and d(v. Now t = k + t with k ∈ Z and 0 t < 1. Lecture 6 25 . Proposition 6. in RN for any N . If there are no elements of Λ \ Zw1 . The choice of w2 ensures that v must be in Rw1 and in Λ. w2 are certainly linearly independent. so b = ka ∈ Za as required. 1 or 2 according as Λ = {0}. The real numbers tj can be written as tj = kj + tj with kj ∈ Z and 0 tj < 1.Proof: Since Λ is discrete. Then b is a scalar multiple of a. Then v = v − (k1 w1 + k2 w2 ) = t1 w1 + t2 w2 ∈ Λ and has d(v . Suppose that b ∈ Λ. Consequently. λ2 ∈ Λ. Hence. Rw1 ) minimal. Call this vector w2 . For an example. Zw1 or Zw1 + Zw2 . w2 in this proposition are not unique. Proof: Let Λ be a lattice in R2 with Λ = {0}. t2 < 1} with v ⊥ orthogonal to w1 and t ∈ R . Hence. we can ﬁnd a δ > 0 with d(λ1 . We showed above that such a vector must be an integer multiple of w1 . λ2 ∈ Λ. We say that the lattice Λ has rank 0. As in the previous proposition. so v ∈ Zw1 + Zw2 . so Za ⊂ Λ. Rw1 ) < d(w2 . We can extend the argument used above to ﬁnd the lattices in R2 and. The argument used in the proof of the previous proposition shows that a scalar multiple tw1 is in Λ if and only if t ∈ Z. only a ﬁnite number of points of Λ can lie within a ball B(0. there can be no more than a ﬁnite number of points of Λ in any ball B(0. there is a δ > 0 with λ − 0 > δ for each λ ∈ Λ \ {0}. say b = ta with t ∈ R. In the second case we will see that Λ = Za. consider the hexagonal 1 1 2 + Z 1√ . Rw1 ) = d(t2 w2 . indeed. 0) minimal. r). Since only a ﬁnite number of points of Λ lie within any ball about the origin. or Zw1 . λ2 ) > δ for each pair of distinct points λ1 . Rw1 ) . then we are ﬁnished.2 Lattices in R2 2 Each lattice in R is either {0}. This implies that either Λ = {0} or else there is a point a ∈ Λ \ {0} closest to 0. Each such vector can be written as v = v ⊥ + tw1 Observe that d(v. Choose a vector w1 ∈ Λ \ {0} with d(w1 .
Let G be such a group.1 The point group acts on the lattice Each isometry in the point group G of a discrete subgroup G of Isom(E2 ) maps the lattice of G onto itself. Lemma 7. We call the rank of Λ the rank of the original crystallographic group G. 7. Lecture 7 26 . Let G be the crystallographic group. So. it must be a ﬁnite cyclic group. If w is in the lattice Λ for G. Since this translation is by a vector in the lattice Λ = {0}. Then their commutator R1 R2 R1 R2 is a translation since −1 −1 φ(R1 R2 R1 R2 ) = φ(R1 )φ(R2 )φ(R1 )−1 φ(R2 )−1 and SO(2) is commutative.7 EUCLIDEAN CRYSTALLOGRAPHIC GROUPS A (2dimensional Euclidean) crystallographic group is a discrete subgroup of Isom(E2 ). the composite: A ◦ T ◦ A−1 : x → B(B −1 (x − v) + w) + v = x + Bw is in G. This implies that they are rotations about the same centre.2 shows that G must then be a cyclic or dihedral group. Hence G contains only the identity. The image φ(G) is called the point group G of G. then Λ = {0}. This is a discrete group of translations. Then φ(A) = B is in the point group G. This means that G can contain no translations and no glide reﬂections (since the square of a glide reﬂection is a translation). Λ its lattice. then the translation T : x → x + w is in G. We are now in a position to describe all of the 2dimensional Euclidean crystallographic groups. and G its point group. We think of the group G as being made by combining the lattice of translations with the point group. Since G+ is a discrete group of rotations. the entire program is rather tedious so we will only explain the main themes and illustrate the results.1 Rank 0 : Finite Groups If G is of rank 0. Proposition 3. rotations and reﬂections. Since G+ is a normal subgroup of G with index 1 or 2. so gives a lattice Λ. it must also be ﬁnite. Suppose −1 −1 that G contains two rotations R1 . However. It is a subgroup of the orthogonal group O(2). Hence. it must be the identity and so R1 and R2 commute. The homomorphism φ : Isom(E2 ) → O(2) deﬁned in §3 has all the translations Trans(E2 ) as its kernel. R2 . when we restrict it to G we get a group homomorphism φ : G → O(2) with kernel equal to G ∩ Trans(E2 ). This shows that Bw ∈ Λ. Proof: Let A : x → Bx + v be an isometry in G. Let us ﬁrst consider the orientation preserving subgroup G+ of G.
a reﬂection M in the line Rw. Each orthogonal map B in the point group G maps Λ onto itself.2 Rank 1: Frieze Patterns If G is of rank 1.7. (b) G = {I. say r = 2k. = In the second case. If r = 2k + 1. In the ﬁrst case G contains the translations T k . Then A = R. R. so A2 = T r for some r ∈ Z \ {0}. then C = T −k A is reﬂection followed by translation by 1 w. These four maps form a group D4 . Then G must contain a rotation A with φ(A) = R. A2 is a translation. We will denote the translation x → x + w by T . So B must be the identity. If r is even. R}. G must contain at least one isometry A with φ(A) = M . Choose coordinates in the plane so that this point c is the origin. Then A is reﬂection followed by translation parallel by 1 rw. or a reﬂection N in the line through the origin orthogonal to w. The possible subgroups are: {I} {I. the other isometries of G that φ maps to M are T k A. The group G is an inﬁnite dihedral group D∞ . Note that RT R−1 = T −1 . N } For each of these we can work out what the possibilities are for G. Hence G must be a subgroup of {I. In either case. the reﬂection A. a rotation R through angle π about the origin. This G is the symmetry group of a pattern such as: We call such patterns whose symmetry group is a rank 1 crystallographic group frieze patterns. M }. 2 Such a G is the symmetry group of a frieze pattern such as: (c) G = {I. Then G = Λ and consists entirely of translations and is cyclic of inﬁnite order. Hence the group G consists of the translations T k and the rotations T k R for k ∈ Z. The rotation T k R is through angle π with centre 1 kw. This means that A is either a reﬂection in a line parallel to w or a glide reﬂection parallel to w. M } {I. M. N }. So A is a rotation about some centre c through an angle π. R. R} {I. and the glide reﬂections T k A. then T −k A is a reﬂection and we are 2 in the previous case. M. N } {I. (a) G = {I}. Note that AT A = T . This 2 generates the cyclic group G and gives the pattern: Lecture 7 27 . so G ∼ C∞ × C2 . then Λ = Zw for some nonzero vector w ∈ R2 . Choose coordinates so that the origin is on the mirror.
5 or 6. The rotation about the origin through an angle π always maps Λ onto itself. R. M. (e) G = {I. the set S = Λ ∩ {v : v = r} is ﬁnite and the point group G must map this isometrically to itself. Lecture 7 28 . Lemma 7.2 The crystallographic restriction A rotation in the point group G of a crystallographic group G must be of order 2. say θ. Choose coordinates so that the origin is on the mirror. N }. Each orthogonal map B in the point group G maps Λ onto itself. This certainly implies that any rotation R ∈ G is of ﬁnite order. N }. Consequently R has order 2. This means that θ π/3. Choose R as the rotation in G through the smallest angle. Since Λ is discrete. There must be an isometry A ∈ G that maps to M under φ. In the second case A2 would be a translation orthogonal to w. 3. There are two cases. G must contain at least one isometry A with φ(A) = N . This means that A is either a reﬂection in a line orthogonal to w or a glide reﬂection orthogonal to w. which is impossible. So A is a reﬂection in a line orthogonal to w. Find fundamental sets for each of the frieze groups. Note that AT A = T −1 so the group G is an inﬁnite dihedral group. As in (b) we can choose coordinates so that R ∈ G. as in (b).(d) G = {I.3 Rank 2: Wallpaper patterns If G is of rank 2. then Λ = Zw1 + Zw2 for two linearly independent vectors w1 . If we can choose A to be a reﬂection in a line parallel to w. 4. The vector Rw1 − w1 is also in Λ so it must have length at least w1  unless it is 0. For most lattices there are no other symmetries ﬁxing the origin as the following result shows. Proof: Let w1 be an element of Λ \ {0} with w1  = r minimal. Exercise: 6. 4 or 6. 3. w2 ∈ R2 . so G is an inﬁnite dihedral group. then AT A−1 = T and ARA−1 = R so G ∼ D∞ × C2 : = If we can choose A to be reﬂection followed by translation by 1 w then A generates an inﬁnite cyclic 2 group and RAR−1 = A−1 . 7.
Proof: Take w1 as in the previous proof and let R ∈ G be the rotation through the smallest angle θ. If θ = 2π/3 or θ = π/3. This is the hexagonal lattice. then w1 and w2 = Rw1 generate the lattice. D12 . If θ = π/2. w1 + R2 w1 ∈ Λ would be at a distance less than r from the origin.3 Point groups The point group of a 2dimensional Euclidean crystallographic group is either cyclic C1 . which is forbidden. Then. C3 . D8 . Lecture 7 29 . then w1 and w2 = Rw1 generate the lattice. D6 .Rw1 θ 0 w1 Suppose that R had order 5. C2 . This is the square lattice. D4 . C6 or dihedral D2 . Rw1 R 2 w1 R 2 w1 + w1 2π/5 0 2π/5 w1 Corollary 7. C4 .
Square lattice Hexagonal lattice We can now proceed as in the previous part to ﬁnd all the possible crystallographic groups with rank 2. http://www.clarku. as illustrated below. These are the wallpaper groups. There are 17 of them.edu/∼djoyce/wallpaper/ Lecture 7 30 .
t) will cross the plane {(z. 2 1 + z 1 + z2 0AN and . 1). Then π gives us a map π : C ∪ {∞} → P. N A π(z) 0 z C Now let us consider two points z1 . 0Aπ(z) are all similar. z2 ) is the Euclidean distance between the stereographic projections π(z1 ) and π(z2 ). 0) with π(z) = (w. giving the North pole.¨ 8 MOBIUS TRANSFORMATIONS 8. Stereographic projection maps points of the complex plane to points of the Riemann sphere P and vice versa. 2 and when s = 1−z2 . We write π(z) = (w. ﬁnite points but Riemann showed that this is not really the case. d(N. t) ∈ C × R : z2 + t2 = 1} in the three dimensional real vector space C × R. t) and deﬁne π(∞) = N . t) ∈ P. Pythagoras’ Note from the diagram below that the triangles z0N . He did this by representing all of the points of the extended complex plane by points of the unit sphere S 2 in R3 . It seems that the point at inﬁnity is very diﬀerent from the other. s) : s = 0} at a single point (z. then the straight line through N and (w. Hence. z) = 1 + z2 . Let P be the unit sphere P = {(z. π(z)) = 2 1 + z2 . giving π(z) = 2z −1 + z2 . w) is any point of P except N . 0) crosses the sphere at N and another point (w. π(z)) = 1 1+ z2 and d(N. The “North pole” of this sphere will be denoted by N = (0. 1) : s ∈ R}. It is easy to give a formula for stereographic projection. we show the triangle with vertices N . This sphere is called the Riemann sphere. In the diagram below. Then the straight line through N and (z. The chordal distance κ(z1 . This map is invertible. Let z ∈ C. t). This line crosses the sphere P when s = 0. Lecture 8 31 . z2 ∈ C. z1 and z2 . A) = d(A. for if (t. 0) + (1 − s)(0.1 The Riemann Sphere It is useful to add an extra element ∞ to the complex plane to form the extended complex plane C ∪ {∞}. theorem shows that d(N. The points on the line from z ∈ C to N are {s(z.
π(zj )) = 2 . ∞) = 2 1 + z1 2 . κ(z1 . z2 ∈ C ∪ {∞} algebraically by using the formula for stereographic projection. Exercise: 7. Hence projection with centre N from the tangent plane to C preserves angles. 8. the straight line from N to z crosses the tangent plane at π(z) and the complex plane C at the same angle θ. z2 ) = d(π(z1 ). A similar argument to that used above to ﬁnd the chordal metric shows that stereographic projection is conformal: it preserves the angle between curves. Prove the formula for the chordal distance between two points z1 . z are antipodal if and only if z = −1/z . is ∞ then we interpret this as κ(z1 . Prove that the chordal metric is a metric on the Riemann sphere. Lecture 8 32 . Show that z. For. N z1 z2 and 1 + zj 2 and d(N. π(z2 )) = 2z1 − z2  1 + z1 2 1 + z2 2 . 1+z1 2 1+z2 2 N π(z2 )π(z1 ) are similar with scale factor √ Conse When one of the points. zj ) = So the triangles quently. 9. say z2 . Two points z. Consequently. z ∈ C ∪ {∞} are antipodal if their stereographic projections satisfy π(z ) = −π(z). 1 + zj 2 2 √ . so they are at the opposite ends of a diameter of the Riemann sphere.z1 π(z1 ) N π(z2 ) z2 We know that d(N. in the diagram below. π also preserves the angle between two curves that meet at z.
(1) 8. Then we can deﬁne a map T : C∪{∞} → C∪{∞} by T :z→ az + b . (2) 2z −1 + z2 . Proof: We can write any circle or straight line in C ∪ {∞} as ao z2 + az + az + a∞ = 0 where a0 . t) = So (1) is equivalent to (a0 + a∞ ) + aw + aw + (a0 − a∞ )t = 0 . A curve Γ in C ∪ {∞} is a circle or a straight line if. These maps are called M¨bius transformations and form a o group M¨b under composition of maps.2 M¨bius Transformations o Let a. The map o φ : GL(2. d be complex numbers with ad−bc = 0. Note that the plane intersects the sphere if. 2 1 + z 1 + z2 .N π(z) θ z 0 θ C Tangent plane at z It is convenient to deﬁne circles in C ∪ {∞} to mean both straight lines and circles. c. a∞ ∈ R and a ∈ C. This same condition ensures that (1) does describe a circle or straight line rather than the empty set. The following result explains why this is so. b. the stereographic projection π(Γ) is a circle on the Riemann sphere. and only if. and only if. C) → M¨b . o Lecture 8 a b c d →T 33 . a2 − a0 a∞ 0. cz + d Note that T (∞) = a/c and T (−d/c) = ∞. so it is a circle on the sphere. The stereographic projection π(z) is (w. This is the intersection with the Riemann sphere of a plane. Proposition 8.1 Stereographic projection preserves circles.
a1 . a∞ ) and (b0 . Hence. Proposition 8. b1 . o a b c d →T is a group homomorphism whose kernel consists of the two matrices I and −I. We denote this quotient by PSL(2. Sa (a1 ) = 1 and Sa (∞) = ∞. b∞ ) . C)/{I. Let T be a M¨bius transformation with o inverse S:z→ Then T (Γ) = {z : S(z) ∈ Γ} = z : p0 az + b2 + p(az + b)(cz + d) + p(az + b)(cz + d) + p∞ cz + d2 = 0 Expanding this gives an expression of the form: z : q0 z2 + qz + qz + q∞ = 0 which is clearly another circle. Recall that M¨bius transformations map circles to circles. then Sb ◦ T ◦ Sa ﬁxes 0. Γ2 be two disjoint circles on the Riemann sphere. o Proposition 8. so φ : SL(2. For we choose three points on the ﬁrst and ﬁnd a M¨bius transformation that maps them o to three points on the second. Let Γ1 . b. cz + d −1 has Sa (a0 ) = 0. T (a1 ) = b1 and T (a∞ ) = b∞ . there is a unique M¨bius transformation T with T (a0 ) = b0 . It follows from this that we can ﬁnd a M¨bius transformation that maps any circle in P onto any o other circle. the M¨bius o group is the quotient SL(2. −I}. c. Exercise: 10. Consequently. If T is −1 another M¨bius transformation with the same properties.3 For any triples of distinct points in the Riemann sphere. C) → M¨b . Then the matrix a b is in SL(2.2 M¨bius transformations map circles to circles o A M¨bius transformation maps any circle on the Riemann sphere to a circle on the Riemann sphere. C). (a0 . o Proof: The M¨bius transformation o Sa : z → a1 − a∞ a1 − a0 z − a0 z − a∞ . C) and call it the projective special linear group. 1 and ∞ so it must o be the identity. a b = λI for some scalar λ ∈ C \ {0}.is a group homomorphism. az + b . Show that there is a M¨bius transformao tion that maps them to two circles in C centred on 0. Usually we choose the scalar λ so that the determinant ad − bc is 1. This c d shows that a M¨bius transformation is unaltered when we multiply each of the coeﬃcients a. Now c d This occurs if and only if a = d and b = c = 0. A matrix M = a b c d is in the kernel of this homomorphism when az + b =z cz + d for all z ∈ C ∪ {∞} . d by a o nonzero scalar λ. Lecture 8 34 . T = Sb ◦ Sa has the required properties. o Proof: Let Γ be the circle {z : p0 z2 + pz + pz + p∞ = 0}.
Let T be represented by the matrix M = with det M = 1. z  . Then we have = . A M¨bius transformation is an isometry of the Riemann sphere (for the chordal metric) if. M (z)  2 w. Conversely. z) = 2Jw − z 1+ Jw2 1+ z2 = 2 − w2 z2 − w1 z1  w1 + w2 2 2 z1 + z2 2 2 = 2 w. T (z)) = κ(J(T w). then the two matrices This implies that −d b c −a =± and both represent T . −d b c −a Therefore. 2 M (w). equivalently. This means that M ∗ M = I or.Proposition 8.4 Isometries of the Riemann sphere. and only if. · on C2 and has determinant 1. M b = a. Then J(z) is the unique point of P that is at a distance 2 from z. Similarly. Write w∗ = ∗ w2 w2 w1 κ(Jw. −a This shows that J ◦ T ◦ J is itself a M¨bius transformation represented by the matrix o a b c d . c d M ∈ SU(2) means that the matrix M preserves the complex inner product ·. write w = w1 /w2 with Each z ∈ P can be written as z = z1 /z2 with z = z2 ∗ −w2 w1 w1 ∗ ∗ w= . which is impossible. if T is an isometry. Proof: Let J : P → P be the map z → −1/z that maps a point to the antipodal point. Hence any M¨bius isometry T must satisfy o a b J ◦ T ◦ J = T . bz − a −d b c . T (z)) = and so T is an isometry. that d = a. o z1 ∈ C2 . b we see that κ(T (Jw). Hence we must have the − sign and this shows that M is unitary. w z Since J ◦ T = T ◦ J and M a. z) M (w) M (z) w z Lecture 8 35 . so that Jw = w1 /w2 = −w2 /w1 . c = −b and a2 + b2 = 1. o it is represented by a matrix a b M= ∈ SU(2) . suppose that the M¨bius transformation T is represented by the matrix M ∈ SU(2). We then have c d J ◦T ◦J :z → −dz + c . a b c d The + sign requires that d = −a and c = b so 1 = ad − bc = −a2 − b2 . z  = = κ(Jw.
Lecture 9 36 . / Theorem 9. Choose a M¨bius o o transformation A with A(z0 ) = 0 and A(z∞ ) = ∞. Let T be the M¨bius transformation represented by the matrix M = o The vector z1 z2 is an eigenvector of M precisely when T (z1 /z2 ) = z1 /z2 . o o Proof: Suppose that the M¨bius transformation T has two ﬁxed points z0 and z∞ . hyperbolic if it is conjugate to Mk for k ∈ R (k = −1. that the matrix M is conjugate to some λ = 0. Choose A with A(z0 ) = ∞. Suppose that T has 1 ﬁxed point z0 only. 0. This means that z1 /z2 is λ 0 0 λ−1 for a ﬁxed point of T . Then A ◦ T ◦ A−1 ﬁxes ∞ alone. This means that it is z → z + b for some b = 0. 1.1 shows that every nonidentity transformation falls into one of these classes. When are two of the M¨bius transformations Mk and P conjugate? o 12.¨ 9 VISUALISING MOBIUS TRANSFORMATIONS 9. Then the conjugate A ◦ T ◦ A−1 ﬁxes 0 and ∞. Exercise: 11.1 Fixed points of M¨bius transformations o A nonidentity M¨bius transformation has either 1 or 2 ﬁxed points in P. loxodromic if it is conjugate to Mk for k ∈ C with k = 1 and k ∈ R. then it is conjugate in M¨b to Mk : z → kz for some k = 0. Find all of the M¨bius transformations that commute with Mk for a ﬁxed k.1 Fixed Points a b c d with ad − bc = 1. or to 1 1 . This gives corresponding results for the M¨bius transformations. +1. 0. Hence describe the o group Z(T ) = {A ∈ M¨b : A ◦ T = T ◦ A} o for an arbitrary M¨bius transformation T . +1). If it has 1. If it has 2. A nonidentity M¨bius transformation is said to be: o parabolic if it is conjugate to P . Consequently. This λ 0 implies that A ◦ T ◦ A−1 is represented by a matrix for some λ = −1. 0 λ−1 A ◦ T ◦ A−1 = Mλ2 . from Linear Algebra. By replacing A by b−1 A we get A ◦ T ◦ A−1 = P . We know. Theorem 9. Let us o 0 1 prove this directly using the transformations. It is simple to tell which by considering the trace. then it is conjugate o in M¨b to P : z → z + 1. elliptic if it is conjugate to Mk for k = 1 (k = 1). Describe the set {A(zo ) : A ∈ Z(T )} for zo a point in o P.
Finally. Elliptic Lecture 9 Hyperbolic 37 . o Show that T is parabolic if and only if (tr M )2 = 4 det M . This means that the matrix M is conjugate to 1 1 0 for λ a square root of k. tr M < −2 or tr M > 2. An o elliptic transformation is conjugate to z → eiθ z. Now we simply note that 0 1 λ−1 tr λ 0 0 λ−1 = λ + λ−1 and tr 1 0 1 1 =2. show that the characteristic equation for M is t2 − (tr M )t + 1 = 0 . −2 < tr M < 2. This moves points along logarithmic spirals away from one ﬁxed point and towards / the other. This moves points along arcs of circles from one ﬁxed point towards the other. or to ± . / Corollary 9. A hyperbolic transformation is conjugate to z → kz for k > 1.2 with determinant Proof: λ 0 We know that T is conjugate to Mk or to P . a parabolic transformation is conjugate to z → z + 1. 14. This rotates the sphere ﬁxing 0 and ∞. hyperbolic or loxodromic. The points are mapped away on one side and towards on the other. Establish similar conditions for T to be elliptic. Deduce that the trace determines the eigenvalues of M . This maps points along a circle through the single ﬁxed point.Trace determines conjugacy class of a M¨bius transformation o a b Let T be a nonidentity M¨bius transformation represented by a matrix M = o c d 1. Suppose that the M¨bius transformation T is represented by the matrix M but that det M = 1. tr M ∈ R. The pictures below illustrate this. Exercise: 13. A loxodromic transformation is conjugate to z → kz for k ∈ R. Then T is parabolic elliptic hyperbolic loxodromic ⇔ ⇔ ⇔ ⇔ tr M = ±2. It is now fairly simple to visualise how M¨bius transformations act on the Riemann sphere. Each point is moved along a circle. If M is a 2 × 2 matrix with determinant 1.
J(z) inverse points for Γ.Loxodromic Parabolic 9. z are inverse points for Γ if every circle orthogonal to Γ through z also passes through z . Any o circle through z orthogonal to Γ is therefore mapped to a circle through T (z) orthogonal to T (Γ).3 M¨bius transformations preserve inverse points o Let T be a M¨bius transformation and Γ a circle in P. Proof: We know that the M¨bius transformation T preserves angles and maps circles to circles. for every point z there is a unique point z so that z. z are inverse points for Γ. If z and z are inverse points for Γ then T (z). we say that z and z itself are inverse points. o T (z ) are inverse points for T (Γ). The original circle passes through z so the the image passes through T (z ) as required. We will prove later that.2 Inversion Let Γ be a circle on the Riemann sphere. Lecture 9 38 . Two distinct points z. Lemma 9. For example. Also. when z ∈ Γ. two points are inverse for the real axis R ∪ {∞} when they are complex conjugates of one another.4 Inversion For each circle Γ in the Riemann sphere and each point z ∈ P there is a unique point J(z) with z. Proposition 9.
Show that inversion preserves the magnitude of angles but reverses their orientation. o Proof: It is clear that C(z) = z is the unique point with z.5 The composition of two inversions is M¨bius. as is its inverse J ◦ C. Example: Inversion in the unit circle. If T (z) = Hence. J(z) inverse points for Γ. so that the M¨bius transformations form the orientation preserving isometry o group of B 3 for this metric. z More generally. we can ﬁnd a M¨bius transformation o T that maps R ∪ {∞} onto Γ. z are inverse points for Γ when they lie on the same halfline from c to ∞ and z − c z − c = r2 . Our ultimate aim is to ﬁnd a metric on the ball B 3 = {x ∈ R3 : x < 1. if J1 . o Lecture 9 39 . If Γ is a circle. z−c Exercise: 15. az + b az + b . Before achieving this goal we will study the simpler case of those M¨bius transformations that map a disc onto itself. inversion in T is given by J(z) = T −1 (T (z)) = 1 . C(z) inverse points for R ∪ {∞}. then C ◦ T ◦ C(z) = .The map J is called inversion in Γ. o cz + d cz + d C ◦ J = C ◦ T ◦ C ◦ T −1 = (C ◦ T ◦ C) ◦ T −1 is a M¨bius transformation. The inversions will then be reﬂections in hyperbolic planes. o Proof: Let C be the particular inversion z → z. then two points z. J ﬁxes every point of the circle Γ. Show that inversion maps any circle to another circle. o Now. o Proposition 9. This shows that C ◦T ◦C is a M¨bius transformation. Inversions in circles are analogous to reﬂections. The M¨bius transformation o T : z → −i z+i z−i maps the unit circle T onto the real axis. Hence J(z) = c + r2 . Now the lemma o shows that J(z) = T (C(T −1 (z))) is the unique point with z. Then inversion in Γ is given by J = T ◦ C ◦ T −1 . (It is a rotation of the Riemann sphere about an axis through 1 and −1. It is an involution: J 2 = I and reverses orientation. o The composition of an even number of inversions is a M¨bius transformation. For any circle Γ we can ﬁnd a M¨bius transformation T that maps R ∪ {∞} onto Γ. J2 are two inversions we can write J2 ◦ J1 = (J2 ◦ C) ◦ (C ◦ J1 ) to show that J2 ◦ J1 is a M¨bius transformation. when Γ = {z ∈ C : z − c = r}.) Hence. . so it is certainly not a M¨bius transformation.
so J2 ◦ J1 (z) = R2 z . then we can conjugate them to get two concentric circles {z : z = 1} and {z : z = R}. This transforms Γ1 and Γ2 to two straight lines through 0 (and ∞). w2 . Clearly we can obtain any elliptic transformation in this way.It is quite easy to identify the M¨bius transformations that we obtain by composing two inversions. parabolic or hyperbolic transformation in this way. There are no other ways that two circles can intersect. then we can conjugate by a M¨bius transformation that sends these points to 0 and ∞ respeco tively. Γ2 respectively. if Γ1 and Γ2 are disjoint. if Γ1 and Γ2 touch at a single point w. o Suppose that J1 . J2 (z) = . The inversions become reﬂections in these lines. so loxodromic transformations can not be the composition of two inversions. If Γ1 and Γ2 cross at two points w1 . Similarly. z z Hence J2 ◦ J1 is a hyperbolic transformation. Finally. So J2 ◦ J1 is the elliptic transformation rotating about an axis through 0 and ∞ through twice the angle between the lines. J2 are inversions in the circles Γ1 . Show that any loxodromic transformation is the composite of 4 inversions. Lecture 9 40 . It is clear that we can obtain any elliptic. then we can transform w to ∞ and see that J2 ◦ J1 is a parabolic transformation ﬁxing ∞. Exercise: 16. Then J1 (z) = R2 1 .
Now J(T (J(z))) = d b c a a b c d dz + c bz + a so the matrices and both represent the same transformation. Proposition 10. Any point z ∈ T satisﬁes J(T (z)) = T (J(z)) = T (z) so T maps the unit circle onto itself. Conversely. Consequently. T must map the unit disc onto either D or the complementary disc D . T (0) = b/a < 1 so it must map onto D. and only if.1 M¨bius Transformations of the unit disc o Inversion in the unit circle is J : z → 1/z. However. If T maps the unit disc o cz+d D onto itself. then the matrix is a b b a with a2 − b2 = 1 as required. Proof: Let T be the M¨bius transformation T : z → az+b with ad − bc = 1.1 M¨bius transformations of the unit disc o A M¨bius transformation T maps the unit disc onto itself if. suppose that T :z→ az + b bz + a with a2 − b2 = 1 . o o Lecture 10 41 . which is impossible if T maps D onto itself.10 THE HYPERBOLIC PLANE. If we have the − sign. This ﬁxes points of the unit circle T and interchanges the unit disc D and the complementary disc D = {z ∈ P : z > 1}. then it must also map the unit circle onto itself. Hence. I 10. then the matrix is a −b b −a with −a2 + b2 = 1. So T (0) =  − b/a > 1. This implies that J ◦ T ◦ J = T . If we have the + sign. This is the M¨bius group of the disc D. it is of the form: o z→ az + b bz + a with a2 − b2 = 1 . The M¨bius transformations that map D onto itself form a subgroup o M¨b(D) = o z→ az + b : a2 − b2 = 1 bz + a of the M¨bius group. This means that d b c a a b c d =± . it must map any pair of inverse points for T to another pair of inverse points. Then J ◦ T ◦ J = T .
Example: For each ω with ω = 1. o + = 10. o Exercise: 17. For any other disc ∆ in the Riemann sphere. c. This shows that every M¨bius transformation mapping the o upper halfplane onto itself is represented by a matrix in SL(2. Show conversely. More formally. + Inversion in the boundary of this is complex conjugation J : z → z. So a M¨bius transformation o T : z → az+b with ad − bc = 1 maps R2 onto itself when J ◦ T ◦ J = T and T (i) ∈ R2 . R). d ∈ R and ad − bc = 1. Show directly that the map z→ 1 1−zo 2 √ zo 2 1−zo  √ zo 1−zo 2 √ 1 2 1−zo  √ ωz + zo 1 + zo ωz maps the unit disc D onto itself when ω = 1 and zo ∈ D. b. For. and only if. o o We begin by deﬁning this metric for an inﬁnitesimal displacement dz. For z + zo T :z→ 1 + zo z Lecture 10 42 . we can ﬁnd a transformation T with T (D) = ∆. that every transformation in M¨b(D) is of this form. T ◦ A ◦ T −1 ∈ M¨b(∆) . These map the cz+d upper halfplane onto the lower halfplane. Now + + cz+d J ◦T ◦J :z → so we need a b c d =± a b c d .2 The Hyperbolic Metric on D We wish to deﬁne a new metric. If such a metric is to have each transformation in M¨b(D) as an isometry. az + b cz + d The + sign gives transformations z → az+b with a. −I}. These do map the upper cz+d halfplane onto itself. R)/{I. this means that a smooth curve γ : [a. for each zo ∈ D the o map z + zo z→ 1 + zo z is represented by the matrix and so is in M¨b(D). Then o A ∈ M¨b(D) if. This should have length ds = λ(z)dz for some density function λ : D → (0. So M¨b(R2 ) ∼ SL(2. c. ∞). o o A particularly important example is when ∆ is the upper halfplane: R2 = {x + iy ∈ C : y > 0}. then the function λ is o almost completely determined. This is a subgroup of the full M¨bius o o o group conjugate to M¨b(D). The − sign gives z → az+b with a. Also. the map z → ωz is clearly in M¨b(D). d ∈ iR and ad − bc = 1. the hyperbolic metric. b] → D should have length b L(γ) = a λ(γ(t)) γ (t) dt . o . b. we can do the same argument as above to ﬁnd the group M¨b(∆) of M¨bius transformations that map ∆ onto itself. on the unit disc D for which each of the M¨bius transformations in M¨b(D) will be an isometry.
So we must have λ(0)dz = λ(zo )T (0)dz . This means that a constant Euclidean displacement increases in hyperbolic length as we approach the boundary. This means that λ(0) = λ(zo )T (0) = λ(zo )(1 − zo 2 ) . K So the metric must be given by ds = 1−z2 dz at the point z ∈ D. Lecture 10 43 . Similarly. The hyperbolic density increases as we approach the boundary of the disc. It is usual to set the constant K to 2. An o inﬁnitesimal displacement dz from 0 is mapped by T to the T (0)dz at zo . So we deﬁne the hyperbolic density on D to be λ(z) = 2 1 − z2 for z ∈ D .is a M¨bius transformation mapping D onto D for each zo ∈ D and so must be an isometry. a displacement of constant hyperbolic length has decreasing Euclidean length as we approach the boundary.
M. so we may assume that z = R ∈ [0. Having deﬁned the hyperbolic density that gives the length of inﬁnitesimal displacements. Lecture 10 44 . to our Euclidean eyes. 1). z1 ) is the inﬁmum of the lengths b b L(γ) = a λ(γ(t)) γ (t) dt = a 2γ (t) dt 1 − γ(t)2 over all smooth curves γ : [a. Escher. This is the hyperbolic geodesic from z0 to z1 . all of the angels are of the same hyperbolic size even though. it is simple to deﬁne the hyperbolic metric. We will see that there is a path γ from z0 to z1 that has shortest hyperbolic length. The hyperbolic distance ρ(z0 .2 Hyperbolic geodesics from the origin. Circle Limit IV (1960) In Escher’s picture above. 1 − z The hyperbolic geodesic from 0 to z ∈ D is a radial line and it has hyperbolic length log Proof: It is clear that rotation about 0 preserves the hyperbolic length of any curve. b] → D that have γ(a) = z0 and γ(b) = z1 . Lemma 10.C. they appear to get smaller as we approach the boundary. 1 + z .
ρ(z0 .Use polar coordinates to write a curve γ : [a. z) > 0 whenever z = 0. 1 − z2 1 − T (z)2 bz + a2 − az + b2 So we see that λ(T (z))T (z) = 2 2 T (z) = = λ(z) . (bz + a)2 az + b bz + a This shows that ρ(T (z0 ). Since we can always ﬁnd a T ∈ M¨b(D) with T (z0 ) = 0. Then the hyperbolic length of γ is more than the hyperbolic length of its radial part. 1 − T (z)2 1 − z2 1 . This is the hyperbolic metric on D. The symmetry and triangle o inequality for ρ are obvious. b] → D as γ(t) = r(t)eiθ(t) .1 shows that any M¨bius transformation from D onto D must be of the form o T :z→ with a2 − b2 = 1. The length of this radial path is: R 0 2 dr = log 1 − r2 1+R 1−R so we see that ρ(0. Lecture 10 45 . z1 ). z1 ) = inf {L(γ) : γ is a smooth curve in D from z0 to z1 } gives a metric on D for which each M¨bius transformation that maps D onto itself is an isometry. The M¨bius transformation T maps o T onto itself and preserves angles. Theorem 10. for b L(γ) = a 2γ (t) dt = 1 − γ(t)2 b a 2r (t) + iθ (t)r(t) dt 1 − r(t)2 b a 2r (t) dt = L(r) . The previous lemma certainly shows that ρ(0. T (z1 )) = ρ(z0 . z1 ) > 0 whenever z0 = z1 . This is a segment of a straight line through 0 orthogonal to the unit circle. it follows that ρ(z0 . z) = log 1 + z 1 − z .3 The expression Hyperbolic metric on the unit disc. For this we have T (z) = A straightforward calculation gives: bz + a2 1 bz + a2 = = . so ρ is a metric. A o geodesic for this metric is the arc of a circle orthogonal to the unit circle. r (t) + iθ (t)r(t) = r (t)2 + r(t)2 θ (t)2 so we have equality only when θ (t) is identically 0. We will assume that γ(a) = 0 and γ(b) = z = R 0. The lemma shows that the geodesic from 0 = T (z0 ) to T (z1 ) is a radial line segment. This shows that the radial path from 0 to z is the shortest path from 0 to z. so the geodesic from z0 to z1 is an arc of the circle through z0 and z1 that is orthogonal to T. 1 − r(t)2 Moreover. Proof: Proposition 10.
J = C ◦ (C ◦ J) preserves hyperbolic lengths. An important example is when ∆ is the upper halfplane R2 . or meet at a point w ∈ T. parabolic or hyperbolic in these three cases. b The trace of the matrix is 2Re(a). we could work with any other disc ∆. Then γ is an arc of a circle Γ that is orthogonal to T. If T is hyperbolic. Then Proposition 9. For T to be an isometry we must have µ(z)dz = λ(T z)T (z)dz . We know that there is a M¨bius transformation T that maps ∆ onto D and we deﬁne the hyperbolic metric on ∆ so that o this is an isometry from ∆ with this metric to D with the hyperbolic metric. This ﬁxes the points az + b iIm(a) ± Re(a)2 − 1 . (This is rotation of the Riemann sphere about an axis through ±1. Therefore. if S is another M¨bius transformation that maps ∆ o onto D. Both C o and J maps the unit disc D onto itself. Im(z) The hyperbolic geodesics on the upper halfplane are half circles orthogonal to R. Suppose that J is inversion in a geodesic γ. Let T be a M¨bius transformation that maps D onto itself. If T is parabolic. then R = S ◦ T −1 will be a M¨bius transformation that maps D onto itself. one ﬁxed point w is in D and the other is 1/w. Then Proposition 10.) The hyperbolic metric on R2 will + then be ds = µ(z)dz for some density function µ. or not meet at all in the closure of D. This means that R o is an isometry for the hyperbolic metric on D. So µ(z) = −2 2 (z+i)2 2T (z) = 1 − T (z)2 1 − z−i z+i 2 = 1 . Lecture 11 46 . Hence T and S give the same metric on ∆. As at the end of §8 we see that the composition of these two inversions is elliptic. so C ◦ J will also do so. then Re(a) = ±1 and there is a single ﬁxed point on T. so Corollary 9.1 Inversions preserve the hyperbolic metric Inversion in a hyperbolic geodesic preserves the hyperbolic metric. then Re(a) > 1 or Re(a) < −1 and there are two ﬁxed points both on T. This deﬁnition does not depend on which transformation T we choose.5 shows that C ◦ J is a M¨bius transformation. The two geodesics may either cross at a point w ∈ D. Proof: It is clear that the inversion C : z → z preserves the hyperbolic metric.2 shows that T can not be loxodromic. For. II 11. Therefore inversion J in Γ maps D onto itself.1 shows that o az + b T :z→ for some a. together with halflines parallel to the imaginary axis. Proposition 11. We can now consider the M¨bius transformations that we get by composing two inversions in o hyperbolic geodesics. We will call this inversion in γ. Rather than working with the unit disc.11 THE HYPERBOLIC PLANE.1 The Hyperbolic Metric on a Half Plane. For this we can take T to be the map + T :z→i z−i z+i .2 Inversions Let γ be a hyperbolic geodesic for D. then −1 < Re(a) < 1. it must be a hyperbolic isometry. 11. If T is elliptic. b ∈ C with a2 − b2 = 1. Hence.
that T has either one or two ﬁxed points in P. Consequently. In this section we will look at some o examples of Fuchsian groups and see that there is a very great variety. By using the conjugates Mk for k = 1. (All other values for k give loxodromic transformations and these map no disc onto itself. Suppose that T has two ﬁxed points. A subgroup G of M¨b(∆) is a Fuchsian group if it is discrete. We will also try to ﬁnd fundamental sets for the group acting on ∆. the only discs Mk maps onto themselves are {z ∈ P : z < r} and {z ∈ P : z > r}. (b) Hyperbolic with two ﬁxed points. The orbits in this case are ﬁnite and a fundamental set is a sector of the unit disc. the orbits are doubly inﬁnite sequences (T n zo )n=−∞ that converge to the two ﬁxed points of T . Elliptic: By conjugating we get Mk : z → kz for some k with k = 1. (c) Parabolic with one ﬁxed point. 1. Hence we are in case (c). When k > 0 (k = 1). Hence we are in case (a). which lies on ∂∆. This generates an inﬁnite cyclic group which is always discrete Lecture 12 47 . When k = 1. one in ∆ and one in the complementary disc. so their limit points will lie on the unit circle. Hence we are in case (b). The only discs mapped onto themselves by P are the halfplanes {z ∈ P : Im(z) > c}. Then T is conjugate to Mk : z → kz for some k = 0. We will study Fuchsian groups by looking at the orbits of points in the hyperbolic plane. This is a rotation about 0 so the group it generates is discrete when it is of ﬁnite order. These orbits must be discrete. the only discs mapped onto themselves are the halfplanes {z ∈ P : Re(eiθ )z) > 0}. The tessellation by images of this fundamental set are shown in the ﬁrst diagram below.) Suppose that T has only one ﬁxed point. both on the boundary ∂∆ of ∆ in P. The images of the fundamental set under the group G cover all of the hyperbolic plane and form a tessellation of the plane.1 .1 Single generator Fuchsian groups Proposition 12. Hence the group generated n=∞ by the hyperbolic transformation T is discrete. The orientation preserving isometries for this metric are the M¨bius transformations that map ∆ onto o itself. and a fundamental set is the region between two suitable disjoint geodesics. 12. Hyperbolic: By conjugating we get Mk : z → kz for some k > 0 (k = 1). from Theorem 9. Then T is conjugate to P : z → z + 1.12 FUCHSIAN GROUPS We can give any disc ∆ ⊂ P the hyperbolic metric and take it as a model for the hyperbolic plane. Parabolic: By conjugating we get P : z → z + 1. the group generated by the elliptic transformation T is discrete when T is of ﬁnite order and a fundamental set is a sector from the ﬁxed point bounded by two half geodesics. This generates an inﬁnite cyclic group which is always discrete with the set {z : 1 z < k} as a fundamental set.1 A nonidentity M¨bius transformation T that maps a disc ∆ onto itself is one of the following: o (a) Elliptic with two ﬁxed points. Mk for k > 1 and P . Proof: We know. Often exhibiting this tessellation will be the simplest way to prove that the group is discrete. it is simple to see when the group G = T generated by a transformation T is discrete. These limit points form the limit set of the group.
with the set {z : 0 Re(z) < 1} as a fundamental set. Hence the group generated by the parabolic n=∞ transformation T is discrete; the orbits are doubly inﬁnite sequences (T n zo )n=−∞ that converge to the single ﬁxed point of T ; and a fundamental set is the region between two suitable geodesics that both end at the ﬁxed point.
Elliptic
Hyperbolic
Parabolic
In the case of a hyperbolic transformation T we call the hyperbolic geodesic joining the two ﬁxed points the axis of T . It is clear that T maps its axis onto itself. Exercise: 18. Show that a hyperbolic transformation T moves any point on its axis by a ﬁxed hyperbolic distance along the axis. This distance is called the translation length of T . Show how to calculate the translation length from the trace of a matrix that represents T . It is also useful to use Proposition 9.5 and consider the M¨bius transformation as a composition o of two inversions in hyperbolic geodesics.
12.2 Triangle Groups Let U be a “triangular” region bounded by three disjoint geodesics γ1 , γ2 , γ3 . Let Jk be inversion in the geodesic γk . Then the M¨bius transformations: A = J2 ◦ J1 , B = J3 ◦ J2 generate a group G. o Note that J1 ◦ J2 = A−1 ; J2 ◦ J3 = B −1 ; J3 ◦ J1 = B ◦ A ; J1 ◦ J3 = A−1 ◦ B −1 so any product of an even number of the inversions is an element of G. The set F that is the union of the closure of U in D and J2 (U ) is a fundamental set for G. (The closure of U is a fundamental set for the larger group generated by J1 , J2 and J3 .) The quotient D/G is homeomorphic to a sphere with three holes: a “pair of pants”. G is a free group generated by A and B. For suppose that we can write g ∈ G as a product g = Ak(1) B k(2) Ak(3) . . . B k(N ) with each k(j) = 0 .
If k(1) > 0 (so the product for g begins with an A) then the copy g(F ) of the fundamental set is separated from F by the geodesic γ1 . Similarly, if g begins with A−1 (or B or B −1 , then g(F ) is separated from F by J2 (γ1 ) (or γ3 or J2 (γ3 )). Hence we can only have g = I when the product for g is trivial and does not begin with any of A, A−1 , B, B −1 . Any orbit of G contains exactly one point in each copy g(F ) of the fundamental set. The ﬁxed points of each nonidentity element of G clearly give points in the limit set. The limit set is like a Cantor set: It is a subset of ∂D but omits points in the open intervals where F meets ∂D. Similarly, it omits points where A(F ), B(F ), AB(F ), etc meet ∂D. Lecture 12 48
B −1 A(F ) B −2 (F ) B −1 A−1 (F )
B −1 (F )
B 2 (F )
B
AB(F )
BA−1 (F )
B(F )
F
A(F )
AB −1 (F )
BA(F )
A
A2 (F )
A−1 (F )
A−1 B −1 (F ) A−1 B(F )
A−2 (F )
The group G is certainly discrete. For suppose that (gn ) is a sequence of nonidentity transformations in G that converge to I. Then gn (zo ) → zo as n → ∞. When we choose zo to be an interior point of the fundamental set F we see that this is impossible.
Lecture 12
49
We can also consider the groups generated by inversions in the sides of a triangle where the sides cross. Then the M¨bius transformations A, B are elliptic rather than hyperbolic. To get a Fuchsian o group these elliptic elements must be of ﬁnite order, so the angles of the triangle must be π/k for some integers k. The sum of the angles of a hyperbolic triangle is strictly less than π, so there are many triangles where these conditions are satisﬁed. For example we may construct a triangle with all angles π/4 and obtain a tessellation as shown below. The limit set in these cases is all of the unit circle.
Lecture 12
50
b with ad − bc = 1. Since z → z + 1 is in the modular group. c. Hence M is the quotient PSL(2. This is the same as wo + c c iy i S = {z : z − k 1 for all k ∈ Z} 2 1 0 1 2 x This means that. b. Consequently. d). There are always inﬁnitely many points in Ω where the imaginary part is greatest. −I}. d we can ﬁnd integers a. The Modular group M is the group of M¨bius transformations of the form o z→ az + b cz + d with a. there is a point wo ∈ Ω where Im(wo ) is maximal. for any point w ∈ R2 . Z) of the matrix group SL(2. This + group is certainly discrete because the integers are discrete. Hence there are only a ﬁnite number of pairs (c. In particular. d) with Im(T (w)) > Im(w). For a transformation T : z → az+b in the modular group + cz+d we have 1 (aw + b)(cw + d) = Im(w) . so the matrix a b is in the modular group. Consequently. wo + k is in Ω for each 0 1 k ∈ Z. Rather than studying it directly. d ∈ Z and ad − bc = 1. so only a ﬁnite number can lie within the unit disc. we will look at a closely related group ﬁrst: the Modular group. Note that the point wo must satisfy cwo + d 1 for each such pair c d d 1 so we see that wo must lie outside the shaded region below. The points cw + d all lie in the lattice Zw + Z. (c. we can choose wo to lie 1 in the region {z = x + iy ∈ R2 : − 2 < x 1 . we can ﬁnd a point wo in the orbit of w that lies in the + unshaded region or on its boundary. For each pair of coprime integers c. The three vertices all lie on the boundary ∂D and the compositions of two inversions is parabolic. Im(T (w)) > Im(w) ⇔ 1 > cw + d . For the matrix 1 1 is in SL(2. z 1} shaded below. The modular group acts as a group of hyperbolic isometries on the upper halfplane R2 . there are only ﬁnitely many values of Im(z) for z ∈ Ω which are greater than Im(w). + 2 Lecture 13 51 . Consider the orbit Ω of a point w ∈ R2 . We wish to ﬁnd a fundamental set for the modular group and the corresponding tessellation. Z) and so z → z + 1 is in the modular group. Im(T (w)) = Im cw + d2 cw + d2 Hence. Z) by {I. This group is important in may branches of Mathematics.13 THE MODULAR GROUP In many ways the most interesting triangle group is when the triangle has all of its angles 0.
two points in F are in the same orbit for the modular group if and only if they are either √ with y 1 3 or − 1 + iy and 1 + iy 2 2 2 ieiθ and ie−iθ with 0 θ 1 3π Proof: Let A : z → z + 1 and B : z → −1/z. we c d cz + d must have cz + d = 1. Since the imaginary parts of z and z are equal. Since they are in the same orbit. In the ﬁrst case. together with part of the boundary is a Proposition 13.1 Fundamental set for the modular group. z = z + k = Ak (z) for some integer k and we are in the ﬁrst case of the proposition. In the diagram below. 1 z = k − = Ak B(z) and we are in the second case. then we have shown that there is an element z in the orbit of z that lies within S. so their imaginary parts are equal. Z). If z lies in the set S = {w : w − k k 1 for all k ∈ Z} then there is an integer k with A (z) = z + k ∈ F . z 1} . If z lies outside this set. the set is divided into two triangles. z It is useful to divide the fundamental set for the modular group into two. there is a transformation T in the modular group with T (z) lying in the set + 1 F = {z = x + iy ∈ R2 : − 2 + x 1 2 . Both of these are in the modular group. Hence. For every point z in R2 . either c = 0 and d = ±1 or c = ±1 and d = 0. Suppose that z and z are in the same orbit and both lie within F . Moreover. we az + b a b must have z = for ∈ SL(2.iy i 2 1 0 1 2 x This set: {z = x + iy ∈ R2 : − 1 < x < + 2 fundamental set for the modular group. In the second case. z > 1}. The modular group permutes Lecture 13 52 . one shaded and the other unshaded. Then both have maximal imaginary part for that orbit. 1 2 .
iy 2 1 0 1 2 x Finally. iy 2 1 0 1 2 x Lecture 13 53 . the diagram below shows that there we can join together 6 of these triangles to form an ideal triangle which has all three vertices on the boundary of R2 . Inversions in the sides of this ideal + triangle give a subgroup of the group generated by inversions in sides of the smaller triangles.the shaded triangles and also permutes the unshaded ones. The inversions in the three sides of one of these triangles gives a larger group with the modular group as an index two subgroup.
2 Inversion We are used to thinking of extending the complex plane C by adding a point ∞ and identifying the resulting space with the Riemann sphere P.1 The Hyperbolic Metric Let B 3 be the unit ball {x ∈ R3 : x < 1} in Euclidean 3space. when x ∈ Σ. As in the case of the Riemann sphere we will deﬁne circles to include straight lines. Then the hyperbolic on B 3 restricts to give the plane hyperbolic metric on the D. We will see shortly that much more is true: the restriction of the hyperbolic metric to the intersection of B 3 with a sphere orthogonal to the unit sphere gives a hyperbolic plane metric. b] → B 3 is then b L(γ) = a λ(γ(t)) γ (t) dt = a 2γ (t) dt . Moreover. 14. then consider inversions. 1 − x2 b The hyperbolic length of a smooth curve γ : [a. 1 − γ(t)2 The hyperbolic metric ρ on B 3 is deﬁned by ρ(xo . while u is a unit normal to the plane Π(u. The N dimensional Euclidean space RN is extended by adjoining a point ∞ to obtain RN . So a sphere in R3 is either of the form ∞ ∞ S(c. The hyperbolic density on B 3 is λ(x) = 2 . and ﬁnally show that every M¨bius transformation is the composition o of inversions. 0) correspond to x1 + ix2 . The arguments used for the hyperbolic metric on the unit disc ( Lemma 10. x are inverse ∞ points for Σ if every circle orthogonal to Σ through x also passes through x . This is homeo∞ morphic to the unit sphere S N in RN +1 and we can use stereographic projection to identify RN with ∞ S N . We will only consider the 3dimensional case although the results apply in higher dimensions. Also. and spheres in R3 to include planes. t) = {x ∈ RN : x · u = t} ∪ {∞} for a unit vector u and t ∈ R.1 Hyperbolic metric on B 3 The hyperbolic metric ρ is a metric on the unit ball B 3 . Lecture 14 54 . To do this. x2 . we ﬁrst deﬁne the o metric. r). We call c and r the Euclidean centre and Euclidean radius of the sphere S(c. x1 ) = inf L(γ) : γ is a smooth curve in B 3 from xo to x1 .2 and Theorem 10.3 ) show that: Proposition 14. 1 − x The disc B 3 ∩ {x : x3 = 0} = {(x1 . 0) : x1 2 + x2 2 < 1} can be identiﬁed with the unit disc D in C by letting (x1 . We can also do this in higher dimensions. the hyperbolic geodesic from the 1 + x origin 0 to any point x ∈ B 3 is a radial path with hyperbolic length log . Two distinct points x. r) = {x ∈ R3 : x − c = r} for c ∈ R3 and r > 0. t). or else Π(u. x2 . we say that x and x itself are inverse points. We deﬁne inversion is a sphere Σ in R3 exactly as in §9. A curve that attains this inﬁmum is a hyperbolic geodesic from xo to x1 .2 . 14.14 HYPERBOLIC 3SPACE In this section we will deﬁne a hyperbolic metric on the unit ball in R3 so that the orientation preserving isometries are the full group of M¨bius transformations.
Any plane π through x and c cuts Σ in a circle σ and we know from Proposition 9. so we see that J(U ) cuts π in a circle. The map J is called inversion in Σ. Proof: We already know from Proposition 9. we know that x and J(x) are inverse points for the circle σ = Σ ∩ π in the plane π. Suppose ﬁrst that Σ = S(c. In this plane.) If γ is any circle through x that crosses Σ orthogonally. Then any plane π through cuts Σ is a circle σ and U in a circle u. Proposition 14. r) and U the sphere S(d. s). Let J be inversion in a sphere Σ. Lecture 14 55 . (Note that we should interpret this formula as saying that J(c) = ∞ and J(∞) = c.2 Inversion in spheres For each sphere Σ in R3 and each point x ∈ R3 there is a unique point J(x) with x and J(x) inverse ∞ ∞ points for Σ. so we only need to show that it does have the ∞ properties we require. Since inversion in higher dimensions is deﬁned in terms of inversions in 2dimensions.4 that there is a point x in π with x and x inverse points for σ. Consequently. then there is a plane π through γ. It is an involution that reverses orientation. This expression makes sense for any point x ∈ R3 . Let J be inversion in a sphere Σ. Inversion maps each of these to another sphere and the intersection of these two spheres is again a circle. We know that inversion in the circle σ sends u to another circle. γ must pass through J(x). Corollary 14.4 Inversion preserves circles.Proposition 14. We will use this to extend the result to higher dimensions. Then J maps any circle γ in RN onto another circle. A similar but simpler argument applies when Σ or U are planes. t). ∞ Proof: Let Σ be the sphere S(c. Let be the straight line through c and d. An entirely similar argument applies when Σ is a plane Π(u. then J(α) and J(β) ∞ also cross at an angle θ. so J(U ) must be a sphere. If two curves α and β in R3 cross at an angle θ. Then we have J(x) = x + 2(t − x · u)u . Indeed we know that the inverse point is given by J(x) = c + r2 x − c2 (x − c) . This is true for every plane π through .4 that inversion in a circle maps the plane to itself. r). We already know a lot about inversion in 2dimensions. Then J maps any sphere U in RN onto another sphere.5 Inversion preserves angles.3 Inversion preserves spheres. Let J be inversion in a sphere Σ. ∞ Proof: The circle γ is the intersection of two spheres. Proposition 14. we readily obtain a variety of results about inversion.
Note that this does not alter any angles. We wish to calculate the derivative J (x). To simplify the algebra. Lecture 14 56 . for inversion in the sphere S(c. 2 x x2 x2 J(x + h) = So we see that J is diﬀerentiable at x with J (x) : h → 1 x2 h− 2x · h x2 x .Proof: Let J be inversion in the sphere S(c. This is a scalar multiple of reﬂection in the plane orthogonal to x and so preserves angles (and reverses orientation). r). r) that inversion maps a sphere to another sphere. 1). Show from the formula J(x) = c + r2 x − c2 (x − c) . translate and enlarge the sphere so that it becomes the unit sphere S(0. Exercise: 19. Now J : x → 1 x2 x satisﬁes 1 (x + h) + 2x · h + h2 1 2x · h = J(x) + h− x + o(h) .
Proposition 15. Let J be inversion in a sphere Σ orthogonal to the unit sphere S 2 . Proof: Let x be a point of B 3 and choose any plane π through the origin and x. Proposition 14. Proposition 11. 2 Throughout this section we will only be concerned with spheres Σ orthogonal to the unit sphere. let Q(a) = a and set a2 1 − a2 a Σ= x : x − Q(a) = . Hence J must be an isometry at x. Lemma 15.1 Inversions are hyperbolic isometries.1 Inversions and the hyperbolic metric Let Σ be a sphere that is orthogonal to the unit sphere S 2 in R3 . any circle σ on the unit sphere S 2 is the intersection of S 2 with a sphere Σ orthogonal to S . Proposition 14. Let J be inversion in a sphere Σ and Q inversion in the unit sphere S 2 . Then π intersects Σ in a circle γ and the map J acts on π as inversion in γ.¨ 15 EXTENDING MOBIUS TRANSFORMATIONS TO HYPERBOLIC SPACE 15. This is a section of a circle orthogonal to the unit sphere S 2 .3 shows that inversion J in Σ ﬁxes σ and maps the unit sphere to another sphere. Hence J maps the unit sphere onto itself.1 shows that inversion in γ is an isometry for the hyperbolic metric on the disc π ∩ B 3 .2 For each point a ∈ B 3 there is an inversion J in a sphere orthogonal to S 2 that interchanges the origin and a. Conversely. Exercise: 20. Then it is simple to check that inversion in Σ maps 0 to a. Therefore. Show that Σ is orthogonal to S 2 if and only if J ◦ Q = Q ◦ J. We have already seen that the derivative of J is orientation reversing.5 shows that this image sphere is also orthogonal to Σ. Then J is an isometry for the hyperbolic metric but reverses orientation. Otherwise. and the circle σ = Σ ∩ S 2 where it meets the unit sphere.3 Hyperbolic geodesics Between any two distinct points of B 3 there is an unique path with shortest hyperbolic length. J maps the ball B 3 onto itself. Proposition 15. Proof: If a = 0. Lecture 15 57 . then inversion in any plane through the origin (that is reﬂection in such a plane) will do. Then Σ meets S 2 in a circle σ.
Let Jn be inversion in the sphere Σn . σN . In particular. o For every M¨bius transformation T : S 2 → S 2 the extension T : R3 → R3 maps the unit ball B 3 onto o ∞ ∞ itself and is an orientation preserving isometry for the hyperbolic metric. Consequently. This extension was introduced by Poincar´ o e and is often called the Poincar´ extension of the M¨bius transformation. we see that the shortest path from a to b must be the image of a radial path under J. However. . hence the composition T is also.Proof: Suppose that a and b are two distinct points in the ball B 3 .2 gives an inversion J in a sphere orthogonal to S 2 with J(A(0)) = 0. e o This extension is unique. Hence. This is true for every choice of γ so we see that the point T (x) is completely determined by the map T acting on the sphere. For suppose that x is any point of B 3 .5 M¨bius transformations as isometries of hyperbolic 3space. o Theorem 15. we must have A (tu) = tv Lecture 15 58 . For each of these circles. . A = J ◦ A is an isometry of B 3 that ﬁxes 0. This map agrees with T on the sphere S 2 and maps the ball B 3 onto itself. So we have ∞ a way to extend the M¨bius transformation T to the ball.4 Extensions of M¨bius transformations. Since it starts at the origin. Then each inversion J in a sphere orthogonal to S 2 maps γ to another circle orthogonal to S 2 . Proof: Proposition 15. A ◦ γu must also be a hyperbolic geodesic. Then A(0) ∈ B 3 so Lemma 15. ◦ J3 ◦ J2 ◦ J1 acts on the sphere S 2 as the M¨bius transformation T .1 . For any other value of a. 1) → tu is a hyperbolic geodesic with ρ(0. Now Corollary 14. this composition gives a map T from all o of R3 to itself. We o know from §9 that T can be written as the composition of inversions in an even number of circles. Choose a circle γ through x that crosses S 2 orthogonally at two points a and b. Since this inversion is a hyperbolic isometry. σ2 . Then the composition JN ◦ JN −1 ◦ . . 3 For each unit vector u ∈ S 2 we know that the path γu : [0.4 completes the proof. Suppose now that T is a M¨bius transformation. We already know that the radial path is the shortest path from a to b when a = 0 because of Proposition 14. Now we will prove that every orientation preserving isometry of B 3 for the hyperbolic metric is an extension of a M¨bius transformation. o Every orientation preserving isometry of hyperbolic 3space B 3 is T for some M¨bius transformation o T : S2 → S2. Proof: Suppose that A : B 3 → B 3 is an orientation preserving isometry for the hyperbolic metric on B . we see that T is the composition of an even number of the orientation reversing isometries Jn . T (x) lies on the circle orthogonal to S 2 that joins the two points T (a) and T (b) in S 2 . if we write T as a composition of inversions in two diﬀerent ways we must obtain the same extension T . the lemma shows that there is an inversion J in a sphere orthogonal to S 2 with J(a) = 0. there is a sphere Σn orthogonal to S 2 which intersects S 2 in the circle σn . .1 shows that each inversion Jn is an isometry for the hyperbolic metric. Proposition 15. Hence T is orientation preserving. γu (t)) = log(1 + t)/(1 − t). Hence. . This maps the unit sphere S 2 to itself. . say σ1 . Since T is the composition of an even number of inversions.
for some unit vector v. We can redo the arguments above for any ball in R3 and obtain a hyperbolic metric on the ball ∞ for which the orientation preserving isometries are the M¨bius transformations. These are inversions. 0. so we see that A . 1) maps the upper halfspace onto the ball and vice versa. ∞ ∞ Finally we know that every orthogonal linear map is the composition of reﬂections in plane through the origin. The most important o example is when the ball is the upper halfspace: R3 = {(x1 . This shows that A : R3 → R3 is orthogonal. is the composition of inversions in spheres orthogonal to S 2 . and hence A. Write v = α(u). Now observe that t→0+ lim ρ(tu1 . Since A is an isometry. x3 ) ∈ R3 : x3 > 0} . + The boundary of this is the extended complex plane C∞ = R2 . Lecture 15 59 . Usually we will not distinguish between T and its extension T . tu2 ) = 2u1 − u2  t for any unit vectors uk . We have now proved that the groups M¨b and Isom+ (B 3 ) are isomorphic with a M¨bius transforo o mation T corresponding to the extension T . We can show that any M¨bius o ∞ transformation acting on this boundary extends to an orientation preserving isometry of the upper halfspace for the hyperbolic metric with density: λ(x) = 1 . 15. It follows that α is an orthogonal linear map in O(3). This means that the M¨bius group M¨b acts on hyperbolic o o 3space B 3 . x2 . x3 We can also deduce the results for the upper halfspace directly from those for the ball B 3 for inversion in the sphere Σ = x : x + e3  = √ 2 where e3 = (0.2 The upper halfspace. this shows that α(u1 ) − α(u2 ) = u1 − u2  .
In this section we wish to study these isometries in more detail. A similar result holds for the intersection of any other sphere orthogonal to ∂H3 with H3 .1 shows that ρ(x. The orientation preserving isometries for hyperbolic 3space have been identiﬁed with the group of M¨bius transformations acting o on the boundary ∂H3 . 0) = log 1 + x 1 − x . Each M¨bius transformation is a composition of o 2 inversions and these map balls to balls. The plane {x : x3 = 0} meets the ball B 3 in the unit disc and the hyperbolic metric on B 3 restricts to the hyperbolic plane metric on this disc. In all of these we choose whichever model (B 3 or R3 ) is easiest and apply isometries to make + the calculations simple. Then the boundary of Σ is a circle σ ⊂ ∂B 3 . For suppose that Σ is the intersection with B 3 of a sphere orthogonal to ∂B 3 . + We have seen how to put a hyperbolic metric on the unit ball B 3 in R3 or the upper halfspace We will denote both of these by H3 and call them hyperbolic 3space. we will prove a series of simple results. A ball {x ∈ H3 : ρ(x. We know that there is a M¨bius transformation T o that maps this circle σ to the unit circle T. Note that the Euclidean centre will not normally be c nor the Euclidean radius ρo . γ c a Lecture 16 60 . so the result continues to hold for any c ∈ B 3 . + The hyperbolic geodesics are the arcs of circles orthogonal to the boundary ∂H3 . Then Proposition 14. Two hyperbolic planes that meet in H3 intersect in a hyperbolic geodesic.16 ISOMETRIES OF H3 R3 . Take H3 as B 3 and consider ﬁrst the case where c = 0. We think of geodesics as the straight lines for hyperbolic geometry. it must map Σ to the unit disc.1 Examples in Hyperbolic Geometry The hyperbolic metric is given by: ds = 2 dx 1 − x2 on B 3 and ds = 1 dx x3 on R3 . Hence Σ with the hyperbolic metric is isometric to the hyperbolic plane. Since T acts isometrically on B 3 . We call such an intersection of a sphere orthogonal to ∂H3 with H3 a hyperbolic plane in H3 . There is a unique point of a geodesic γ closest to a point c in H3 . Hence the ball is the set {x : x < tanh 1 ρo }. In order to develop our sense of what hyperbolic 3space is like. 16. c) < ρo } is a Euclidean ball which lies entirely within H3 .
So we see that the hyperbolic geodesic from a point c to the closest point of a hyperbolic geodesic γ is orthogonal to γ. The shortest path from x to α is the arc + {(sin φ. 0. The geodesic γ is then the arc of a circle orthogonal to the unit sphere ∂B . kx3 ) and this must be an isometry. cos θ) ∈ R3 . Note that the shortest hyperbolic path from 0 to a is a radial line. ρo = 0 1 dφ . x3 ) → (kx1 . + Consider the point x = (sin θ. α a x θ 0 Exercise: 21. T : z → eiψ z acts on R3 as + T : (x1 + ix2 . x3 ) : x3 > 0}. b) minimal. this is closest hyperbolically to 0. x2 . By applying an isometry we may assume that H3 = R3 and α is the geodesic {(0. So all of the points (k sin θ. This is a cone about the axis α. Draw the set of points that lie within a ﬁxed hyperbolic distance ρo of a geodesic α in the unit disc D and in the unit ball B 3 . 0. x3 ) and this must be an isometry. Hence all of the points {x ∈ R3 : x3 = x cos θ} + 1 are at a hyperbolic distance ρo = tanh−1 tan 2 θ from α. 1 − t2 Now the M¨bius transformation T : z → kz (k > 0) acts on R3 as o + T : (x1 . cos φ 1−t2 1+t2 Now make the change of variables t = tan 1 tan 2 θ 0 1 2 φ. 0 cos φ) : 0 This has hyperbolic length θ φ θ} . This gives: 2 ρo = 2 1 + t2 dt = 1 − t2 1 + t2 2 1 dt = tanh−1 (tan 2 θ) . Similarly.Take H3 to be B 3 and c = 0. x3 ) → (eiψ (x1 + ix2 ). Lecture 16 61 . kx2 . β in H3 which do not have a common endpoint on ∂H3 there are unique points a ∈ α and b ∈ β with ρ(a. 0. by the previous remark. so that cos φ = 1 tan 2 θ 0 and dt dφ = 1 (1 + t2 ). 3 For two geodesics α. This circle has a unique point a with smallest norm and. k cos θ) are at a hyperbolic distance ρo from α.
Now suppose that β is a geodesic in R3 joining points of ∂R3 neither of which is 0 or ∞. Then + + there will be a unique point b ∈ β with the angle sin−1 x3 x
minimal. This is the point closest to α. It is clear that the shortest path from α to β is a geodesic that crosses both α and β orthogonally.
γ
β α
If two geodesics have a common endpoint, then they contain points that are arbitrarily close. Let α and β be geodesics in R3 with ∞ as their common endpoint. Then + α = {(a1 , a2 , x3 ) : x3 > 0} ; Now it is clear that ρ((a1 , a2 , x3 ), (b1 , b2 , x3 )) β = {(b1 , b2 , x3 ) : x3 > 0} . (a1 , a2 ) − (b1 , b2 ) x3
so there are points of the two geodesics that are arbitrarily close together. Two geodesics α and β in H3 are either: (a) Parallel: when they have a common endpoint. (b) Cross: when they intersect at a point of H3 . (c) Skew: when they have no common endpoint and do not intersect. We have shown that two skew lines have a common normal joining the closest points of the two lines. In case (b) there is also a common normal through the point of intersection. However, there is no such normal for parallel lines unless the lines are identical.
16.2 Axes of Isometries Let T be a M¨bius transformation. When we think of T acting on the Riemann sphere, it has two o ﬁxed points, say a and b. When we think of T acting on the hyperbolic 3space H3 , then there is a geodesic α joining a and b. T maps this geodesic to itself since it ﬁxes the endpoints. We call α the axis of the M¨bius transformation. o The M¨bius transformation T acts isometrically on H3 , so T must move the points of the axis a o ﬁxed hyperbolic distance. This is called the translation length of T . It is 0 for elliptic transformations but nonzero for hyperbolic and loxodromic transformations. Lecture 16 62
Exercise: 22. Show that the translation length of the transformation Mk : z → kz is log k. Hence show how to ﬁnd the translation length of the M¨bius transformation o z→ 2z + 1 . 5z + 3
A parabolic transformation does not have an axis. If there were any geodesic that were mapped onto itself, preserving direction, then both endpoints would be ﬁxed. This can not occur if there is just one ﬁxed point on ∂H3 .
Lecture 16
63
17 INVOLUTIONS A M¨bius transformation R which has ﬁnite order must be elliptic. In particular, an involution o which has R2 = I must be elliptic. We can conjugate R so that its ﬁxed points are at 0, ∞ ∈ R3 . The + axis α of R is then the positive x3 axisjoining 0 to ∞. Then R : z → −z on P and R : (x1 , x2 , x3 ) → (−x1 , −x2 , x3 ) This ﬁxes every point of the axis. Note that, if π is a hyperbolic plane that contains the axis α of the involution R, then R maps this plane to itself but interchanges the halfspaces on either side of it. On the plane π, the involution acts as inversion in the geodesic α. In §11 we saw that every M¨bius transformation of the disc could be o written as the composite of two inversions. Here we will prove that each M¨bius transformation on P o can be written as the composite of two involutions. Proposition 17.1 Isometries of H3 are compositions of two involutions. Every M¨bius transformation can be expressed as R2 ◦ R1 for two elliptic involutions R1 , R2 . o Proof: The identity is R2 for every involution R. Suppose that P is parabolic. Then we may assume that it is P : z → z + 1. This acts as P : (x1 , x2 , x3 ) → (x1 + 1, x2 , x3 ) on R3 . For this, take + R1 : z → −z and R2 : z → 1 − z . Then R1 , R2 are involutions with P = R2 ◦ R1 . Note that the axes of R1 , R2 are {(0, 0, x3 ) : x3 > 0} and
1 {( 2 , 0, x3 ) : x3 > 0}
for x ∈ R3 . +
which have a common endpoint at the ﬁxed point of P . Suppose that T is a M¨bius transformation with 2 ﬁxed points. We may assume that these are 0 o and ∞. So T : z → λ2 z for some λ = 0. Then T acts on R3 as + T : (x1 + ix2 , x3 ) → (λ2 (x1 + ix2 ), λ2 x3 ) . For this, take 1 λ2 and R2 : z → . z z These are involutions with T = R2 ◦ R1 . Note that the axes of R1 , R2 are R1 : z → {(cos θ, 0, sin θ) : 0 < θ < π} and {(λ cos θ, 0, λ sin θ) : 0 < θ < π} .
These are identical when λ = ±1 and T = I. They cross at (0, 0, 1) when λ = 1 and T is elliptic. Otherwise, they are skew and T is loxodromic or hyperbolic with its axis normal to both.
Proposition 17.2 Let R1 , R2 be involutions with axes α1 , α2 in H3 . Then (a) If α1 = α2 then R2 ◦ R1 = I. (b) If α1 and α2 are parallel, then R2 ◦ R1 is parabolic with the common endpoint of α1 and α2 as its ﬁxed point on ∂H3 . (c) If α1 , α2 are skew, then R2 ◦ R1 is loxodromic or hyperbolic with axis normal to both α1 and α2 . Lecture 17 64
then we can assume that γ(0) is the point where α1 meets γ and γ(τ ) is the point where α2 meets γ. This means that R1 (γ(t)) = γ(−t) . it maps π(t) onto π(2τ + t). Exercise: 23. T maps the plane π(0) that bounds one side onto the plane π(2τ ) that bounds the other. So the translation length of T is 2τ : twice the distance between the axes α1 and α2 . Conjugate so this is the x3 axis in R3 . R2 be involutions with axes α1 . Consequently. α2 in H3 . The last proposition shows that we can write T as the composite R2 ◦ R1 of two involutions with axes α1 . Since T is an isometry. For (c). Then R1 : z → 2a1 − z and R2 : z → 2a2 − z and α1 = {(a2 . γ α2 α1 Lecture 17 65 . Hence the set F = {π(t) : 0 t < 2τ } is a fundamental set for the group G = T . Let R1 . x3 ) : x3 > 0} for some a1 . and α2 . There is then a common normal γ to α1 and α2 and this is the axis for T . 2 wj Rj : z → z 2 2 and R2 ◦ R1 : z → (w2 /w1 )z is either hyperbolic or loxodromic.Proof: (a) is obvious. x3 ) : x3 > 0} and R2 ◦ R1 : z → 2(a2 − a1 ) + z is parabolic with its single ﬁxed point at ∞. If we parametrise γ by hyperbolic length. We will concentrate on the case where T is loxodromic or hyperbolic. T (γ(t)) = R2 (R1 (γ(t))) = γ(2τ + t) . For (b) conjugate so that in R3 we have: + α1 = {(a1 . Finally. Show that R2 ◦ R1 is hyperbolic when both α1 and α2 lie in a hyperbolic plane. since the other cases are simpler. let us look at the group G = T generated by a single orientation preserving isometry T of H3 . The geodesic αj is then a halfcircle that crosses the x3 axis orthogonally. a2 ∈ C. Each point x ∈ H3 is closest to some point of γ and those points which are closest to γ(t) ﬁll a hyperbolic plane π(t). we know that there is a unique geodesic γ normal to both α1 and α2 . so it must + join points ±wj . R2 (γ(t)) = γ(2τ − t)) .
there is a fundamental set bounded by two hyperbolic halfplanes that meet on the axis of T . When T is parabolic. Exercise: 24. How are fundamental sets for G = T acting on D related to fundamental sets for G acting on B 3 ? t τ} Lecture 17 66 . When T is elliptic of ﬁnite order. R2 . there is a fundamental set bounded by two hyperbolic planes that touch at the ﬁxed point of T on ∂H3 .Note also that half of F : {π(t) : 0 is a fundamental set of the larger group R1 . Suppose that T is a M¨bius transformation that maps the unit disc D onto itself. Then T also acts o as an isometry of the hyperbolic 3space B 3 .
o Every Fuchsian group is certainly discrete when we think of it as a subgroup of M¨b rather than a o subgroup of M¨b(D). ρo ) and B(c− . octahedral and icosahedral groups. Then there is a unique closed hyperbolic ball B(c. This shows that there is at least one closed ball of smallest radius containing S. o 18. so it must lie in the shaded region. Now suppose that there are two diﬀerent closed balls with smallest radius ρ0 which contain S. A c− m c+ Proposition 18. Let m be the midpoint of the hyperbolic geodesic from c+ to c− . A Kleinian group is a discrete subgroup of M¨b. ρo ). A).1 Finite Kleinian Groups Any ﬁnite subgroup of M¨b is certainly a Kleinian group. This radius is less than ρo . ρ) of smallest hyperbolic radius that contains S. ρo ). which is a contradiction.2 Finite Kleinian groups are conjugate to subgroups of SO(3). However. Lemma 18. We can apply an isometry to H3 to move m to the origin. We will assume that cn → c as n → ∞. we will show that these ﬁnite o groups are all conjugate to ﬁnite subgroups of SO(3). The centres cn all lie within ρ1 of any chosen point of S. Then c+ = −c− . Then there is a sequence ρn that decreases to ρ0 and a sequence of centres cn with S ⊂ B(cn . o o Lecture 18 67 . Every ﬁnite subgroup of M¨b is conjugate in the M¨bius group to a subgroup of SO(3). We will think of these groups acting as isometries of the hyperbolic 3space H3 . Proof: Set ρo be the inﬁmum of the radii ρ for which there is a centre c ∈ H3 with S ⊂ B(c. We will see many more examples later.18 KLEINIAN GROUPS M¨bius transformations are represented by 2 × 2 complex matrices so a group of M¨bius transforo o mations is discrete if it is a discrete subset of the set of all 2 × 2 matrices. In the picture below. ρo ) and B(c− . ρ). so we can ﬁnd a convergent subsequence using the Bolzano – Weierstrass Theorem.1 Let S be a nonempty ﬁnite subset of H3 . So we already know which groups can arise: cyclic and dihedral groups together with the tetrahedral. ρn ). ρo ). Then S ⊂ B(c. say B(c+ . This is contained in the closed ball centred on m with radius ρ(m. S must lie in the intersection of the two balls B(c+ .
xo ) ρ(T (xo ). Lemma 18. ρo )) = B(T (c). This means it is often fairly straightforward to show that a group acts discontinuously by exhibiting a suitable fundamental set. so there is a ρo with K ⊂ B(xo . a) + ρ(T (a). If G acts discontinuously at xo . T (xo )) < δ} is ﬁnite. y) < 2 δ for n −1 elements Sn = TN ◦ Tn of G satisfy N . xo ) < δ which contradicts G acting discontinuously at xo . If a neighbourhood of xo meets only ﬁnitely many copies of F . then we can ﬁnd a ∈ K with T (a) ∈ K. T (a)) + ρ(T (a). Then. Each T ∈ G acts isometrically on B 3 and permutes the elements of the orbit Ω. It has a ﬁnite orbit Ω = G(x) for any point x ∈ B 3 . Now conjugate by a M¨bius transformation that maps c to the origin. xo ) < 2ρo . Lecture 18 68 .3 Let G act discontinuously at a point x0 ∈ H3 . This implies that T (c) = c. 18. there is a natural number N with ρ(Tn (xo ). for any compact set K ⊂ H3 . then the stabilizer: Stab(xo ) = {T ∈ G : T (xo ) = x0 } is a ﬁnite group and so conjugate to a ﬁnite subgroup of SO(3). say B(c. So ρ(T (xo ). Then the copies T (F ) for T ∈ G tessellate all of H3 . 1 In particular. Suppose that there were inﬁnitely many such elements T of G. Then G becomes a group of o hyperbolic isometries that ﬁx the origin. This proposition shows that we need to consider inﬁnite Kleinian groups in order to obtain new and interesting examples of such groups. Proof: Since K is compact. So Ω = T (Ω) ⊂ T (B(c.Proof: We will consider the ﬁnite group G acting on the unit ball B 3 . Consequently the ρ(Sn (xo ). Suppose that F is a fundamental set for G acting on H3 . ρo ). xo ) = ρ(xo . it is certainly bounded. then G acts discontinuously at xo . This means that there are inﬁnitely many Sn ∈ G with ρ(Sn (xo ). ρo ) . The lemma shows that this is contained within a unique smallest closed hyperbolic ball. Then the Bolzano – Weierstrass theorem shows that we can ﬁnd a sequence of distinct elements (Tn ) with Tn (xo ) converging to some point y ∈ H3 as n → ∞. the set {T ∈ G : T (K) ∩ K = ∅} is ﬁnite.2 Discontinuous Action Let G be a subgroup of M¨b = Isom+ (H3 ). To do this we need to think more carefully about the action of a Kleinian group on hyperbolic 3space. ρo ). TN (xo )) < 1 δ + 1 δ = δ 2 2 for n N . The group G acts discontinuously at xo ∈ H3 if there o is some δ > 0 for which {T ∈ G : ρ(xo . xo ) = ρ(Tn (xo ). so every element of G ﬁxes c. The group G acts discontinuously on H3 if it acts discontinuously at each point of H3 . These must be elements of SO(3). If K ∩T (K) = ∅.
) We will now show that acting discontinuously on H3 is equivalent to being a discrete group of M¨bius transformations.For example. a2 . Dodecahedral tessellation of H3 . This implies that Tn (xo ) → xo for any point xo ∈ H3 .uiuc.edu/graphics/pix/Special Topics/ . 1) ∈ R3 . Then there is a sequence (Tn ) of nonidentity transformations in G with Tn → I as n → ∞. So G does not act discontinuously at xo . Lecture 18 69 . o Proof: Suppose ﬁrst that G is not discrete. The group of M¨bius transformations o that are symmetries of this tessellation clearly acts discontinuously on H3 . Then Sa (k) = a. (This corresponds to the + M¨bius transformation on P given by z → a3 z + (a1 + ia2 ). o Theorem 18. Set −1 Sn = STn (k) and Rn = Sn ◦ Tn . It is more convenient to work on the upper halfspace. A group of M¨bius transformations is discrete if and only if it acts discontinuously on H3 . 0.) Use the Bolzano – Weierstrass theorem to o ﬁnd a sequence of distinct transformations Tn ∈ G with Tn (k) → y as n → ∞. we can construct a regular dodecahedron in H3 with each pentagonal face having each angle a rightangle and with the angles between faces being rightangles. (See http://www. T (xo )) < δ. The copies of this dodecahedron obtained by reﬂecting (inverting) in the faces tessellates all of H3 . Suppose now that G does not act discontinuously at some point xo ∈ H3 . so conjugate so that xo is mapped to k = (0. + For each a ∈ R3 deﬁne Sa : x → a3 x + (a1 .4 Discrete if and only if acts discontinuously. 0).geom. Then there is a δ > 0 with inﬁnitely many T ∈ G satisfying ρ(xo .
Lecture 18 70 .org/∼hatch/HyperbolicApplet/ .) Then we have Rn (k) = k so each Rn is in the group Stab(k). as n → ∞ . on the subsequence. so Sn → Sy as n → ∞. Now we have Tn (k) → y.(Note that Sn and Rn need not be in G. Hence. This stabilizer is conjugate to the group SO(3). which is compact. Hence we can ﬁnd a subsequence Rn which converges to a transformation R∞ as n → ∞. we have T n = Sn ◦ R n → Sy ◦ R ∞ This shows that G is not discrete. Look at: http://www.plunk. We can also consider groups acting discontinuously on other spaces and prove similar results.
For any Kleinian group acting on B 3 the limit set Λ(xo ) is a closed. Hence Tn (xo ) and Tn (x1 ) will converge to the same point of ∂B 3 . Since the orbit Ω = G(xo ) satisﬁes T (Ω) = Ω for each T ∈ G. and Tn (x1 ) stay the same hyperbolic distance apart they get closer together for the Euclidean metric as the points get closer to the boundary. we see that T (Λ(xo )) = Λ(xo ). Hence. Then we have shown that G acts discontinuously on H3 = B 3 . Let G be a Kleinian group with an invariant disc ∆ ⊂ P. Note that when we talk about limit points of the orbit we are using the Euclidean (or. We will now show that Λ(xo ) is independent of the point xo . Each T ∈ G is an isometry for the hyperbolic metric.1 The limit set is closed and Ginvariant. Ginvariant subset of ∂H3 = S 2 . Proof: The hyperbolic density at a point x is λ(x) = 2 1 = 2 cosh2 2 ρ(0. However the ﬁxed points of an elliptic transformation need not lie in the limit set. we can ﬁnd ρo with λ(x) > Lecture 19 1 ε for ρ(0.2 Let (xn ) and (yn ) be two sequences of points in B 3 with ρ(xn .19 LIMITS OF ORBITS Let G be a Kleinian group acting on the ball B 3 and choose a point xo ∈ B 3 . Similarly the ﬁxed point of a parabolic transformation is in the limit set. Lemma 19. For suppose that x1 is another point of B 3 . Then ρ(xo . and T is continuous on all of R3 . Suppose that T is a loxodromic or hyperbolic transformation in the group G. x1 ) is ﬁnite. so ρ(T (xo ). 71 . Give an example of a Kleinian group for which the limit set is empty. If the sequence of points (xn ) in B 3 converges in the Euclidean metric to a limit point u ∈ ∂B 3 . The hyperbolic metric is not deﬁned on the boundary so it does not make sense to ask for limit points in the hyperbolic metric. 26. both the ﬁxed points of T are in the limit set for G. Although these points Tn (xo ). We will see shortly that this limit set is independent of the point xo we choose. T (x1 )) = ρ(xo . x) > ρo − K . better still. inside R3 with ∞ the chordal metric) and consider the limit points of the orbit. Proof: Λ(xo ) is obviously closed since it is G(xo ) \ B 3 . Then T n (xo ) tends to one of the ﬁxed points of T as n → ∞ and T −n (xo ) tends to the other. chordal) metric. Give an example of an elliptic element of a Kleinian group with ﬁxed points that do not lie in the limit set. x1 ). So the limit set is ∞ Ginvariant. Recall that a limit point of a set Ω ⊂ R3 is a point u for which there is a sequence (wn ) of distinct points wn ∈ Ω which converge to u. This implies that T (xo ) tends to 1 as T runs through G. so wn − u → 0 as n → ∞. We can think of the points G(xo ) as lying inside the space R3 with the Euclidean metric (or. The set of limit points of the orbit G(xo ) is called the limit set of G and will be denoted by Λ(xo ). So only ﬁnitely many points of the orbit G(xo ) lie within any hyperbolic ball. Proposition 19. x) . yn ) K for each n ∈ N. then the sequence (yn ) will also converge to u for the Euclidean metric. Show that the limit set of G is a subset of ∂∆. for any given ε > 0. 1 − x2 So. Exercise: 25.
so one endpoint will lie within the sphere π. b) = L(γ) = γ λ(x) dx inf {λ(x) : x ∈ γ} b − a . For almost all Kleinian groups the ﬁxed points of loxodromic and hyperbolic transformations are dense in the limit set. sinh ρ(0. Tn (x1 )) = ρ(xo . b) K and ρ(0. We already know that every ﬁxed point of a hyperbolic or loxodromic transformation does lie in the limit set. The limit sets Λ(xo ) and Λ(x1 ) are equal for any Kleinian group G and any points xo . Show that the limit set is {∞} but that there are no hyperbolic or loxodromic transformations in G. The lemma shows that Tn (x1 ) − u → 0 as N → ∞. we must have k2 = (c + r)2 = 12 + r2 . Then ρ(a. x1 ) . is orthogonal to ∂B 3 and crosses the radius from 0 to c normally. This Euclidean sphere has radius r and centre k.4 Let α be a hyperbolic geodesic that passes through a point c ∈ B 3 . 2c Any hyperbolic geodesic through c goes inside π in one direction. so ρ(Tn (xo ). Let G be the group generated by the single parabolic transformation P : z → z + 1. b. Proof: A point u is in Λ(xo ) if there is a sequence (Tn ) in G with Tn (xo ) − u → 0 as N → ∞. They are called elementary groups and we will not be concerned with them. Since π cuts the unit sphere orthogonally. It meets B 3 in a hyperbolic plane. The exceptional groups are ones with very simple structure. yn gives the result. Exercise: 27. Lemma 19. This endpoint u satisﬁes √ u − c 2r . if ρ(a. c) 2c Proof: Draw the Euclidean sphere π that passes through c. Proposition 19. x1 ∈ B 3 . So u ∈ Λ(x1 ). Lecture 19 72 . then K ρ(a. b) 1 b − a .3 Limit set is independent of the base point.Suppose that γ is a hyperbolic geodesic joining the points a. Now each Tn acts isometrically on B 3 . a) > ρo . At least one of the endpoints u of α satisﬁes √ 2 1 − c2 √ u − c = . Hence. So r= 1 − c2 . ε Applying this to the pairs of points xn .
Lemma 19. Let α be its axis joining the two ﬁxed points a+ . Hence. Then the ﬁxed points of all the hyperbolic and loxodromic transformations in G are dense in the limit set for G. Then Proposition 19. except the endpoints of α. Since the limit set is inﬁnite.6 Limit sets are perfect. Corollary 19.3 shows that it suﬃces to prove that the ﬁxed points are dense in Λ(xo ). a− . Then An (v) → u± as n → ±∞. we know that An (v) tends to one endpoint u+ of the axis α as n → +∞ and to the other u− as n → −∞.π c 1 r 0 k Proposition 19. Hence both the ﬁxed points u± of A are not isolated in the limit set. Tn (xo )) → ∞. Lecture 19 73 . Tn (xo )) √ As n → ∞. We know that both these ﬁxed points lie in the limit set. Now the −1 conjugate Tn ◦ A ◦ Tn is hyperbolic or loxodromic and has axis Tn (α) which passes through Tn (xo ). Proof: Let A be the hyperbolic or loxodromic transformation in G. Then there is a sequence (Tn ) in G with Tn (xo ) − u → 0 as n → ∞. For each point v ∈ ∂B 3 = S 2 . Let u ∈ Λ(xo ). so ρ(0. The lemma shows that one of the endpoints Tn (a+ ) or Tn (a− ) satisﬁes Tn (a± ) − u 2 . Proof: Suppose that the limit set is not ﬁnite. sinh ρ(0. Choose a point xo ∈ B 3 that lies on the axis α. Let A be a hyperbolic or loxodromic transformation in G with axis α.5 Let G be a Kleinian group that contains a hyperbolic or loxodromic transformation. The limit set of a Kleinian group G that contains a hyperbolic or loxodromic transformation is either ﬁnite or perfect.4 shows that there is another ﬁxed point v of a hyperbolic or loxodromic transformation that is not ﬁxed by A. Recall that a subset Q of a metric space is perfect if no point of Q is isolated. some sequence of endpoints Tn (a± ) converges to u.
Set Kn+1 to be the closure of Kn ∩ B(y. Otherwise.Every point of the limit set is a limit of ﬁxed points of hyperbolic or loxodromic transformations in G. Set K0 = Λ. so it is compact. perfect. so no point in the limit set is isolated. Corollary 19. there must be another / point. If xn+1 ∈ Kn then take Kn+1 = Kn . The limit set of a Kleinian group G that contains a hyperbolic or loxodromic transformation is either ﬁnite or uncountable. We will construct a decreasing sequence of nonempty. Suppose that Λ were countable and enumerate its points as (xn )n∈N . However. / This completes the inductive construction. perfect. compact. Since Kn is perfect. closed set with xn+1 ∈ Kn+1 ⊂ Kn . 2 No point of this closure is isolated. Suppose that Kn has been deﬁned and is a nonempty. closed subsets Kn of Λ with xn ∈ Kn . Then the intersection Kn can / not be empty. so Kn+1 is a nonempty. closed subset of Λ. in Kn . since Λ is compact. 1 ρ(xn+1 . it does not contain any of the points xn so it must be empty. Now the result follows from Cantor’s theorem that a perfect. xn+1 ∈ Kn . Lecture 19 74 . perfect. Proof: For the limit set Λ is a closed subset of the sphere S 2 . y)) . metric space is uncountable. say y.7 Limit sets are ﬁnite or uncountable. The previous proposition shows that it is perfect.
)3 = 3 3 3 where the digits x1 . Cn consists of 2n disjoint intervals each of length 3−n . . 3 ] ∪ [ 3 . n∈N The Cantor set is clearly closed and bounded in R. Consequently. 1] 1 2 1 2 7 8 1 2 19 20 7 8 25 26 C3 = [0.. xk → 3k xk /2 2k is a bijection. 27 ] ∪ [ 27 . The Cantor set C is the intersection C= Cn . since the endpoints of each component interval of Cn lie in C. x3 . 2} . . It is nonempty. . are each 0 or 1 or 2. 1] . 3 3 3 with x1 .)3 and 2 3 = (0. 2 Exercise: 28. when n = min{k : yk = zk } 2yk 3k is a metric on ZN . . 9 ] ∪ [ 9 . so we can identify C with the inﬁnite group ZN . 9 ] ∪ [ 9 . Show that 2 d(y..200000 .022222 . . 3 ] ∪ [ 3 . This shows that the Cantor set has the same cardinality as the unit interval. . so it is compact. 9 ] ∪ [ 9 . for the map C → [0. . Then x ∈ Cn if and only if it can be written as x= x3 x2 x1 + 2 + 3 + .. 1] . 3 ] ∪ [ 3 . . 1] in base 3 as x3 x2 x1 + 2 + 3 + . z) = 0 2−n when y = z. Cn+1 is obtained from Cn by removing the open middle third from each of these intervals. . Hence the Cantor set is uncountable. . For example. 2} . .20 HAUSDORFF DIMENSION 20. 9 ] ∪ [ 9 . the Cantor set consists of all x ∈ [0.)3 . . Show that the map φ : ZN → C deﬁned above is a homeomorphism from ZN 2 2 2 with this metric. .1 Cantor Sets Deﬁne a sequence of sets Cn ⊂ [0.x1 x2 x3 . 27 ] ∪ [ 27 . 27 ] ∪ [ 27 . Lecture 20 75 . . (yk ) → 2 is a bijection. x2 . Note that this is true even for the endpoints of Cn .. It also shows that the map φ : ZN → C . . xn ∈ {0. 1] as follows: C0 = [0. we can write any x ∈ [0. Let y = (yk ) and z = (zk ) be sequences in ZN . More carefully. . . 1 3 = (0. 1] 1 2 1 2 7 8 C2 = [0. x = (0. . 1] 1 2 C1 = [0. 1] that can be written as ∞ x= k=1 xk 3k with each xk ∈ {0. 27 ] ∪ [ 27 . At each stage. x2 .
. . 3 ] and C ∩ [ 3 .} of subsets of M is a δcover for M when M = Un has diameter at most δ. Exercise: 29. A collection {U1 . The diameter of M is diam(M ) = sup{d(x. Then x − y > 3−n for some n ∈ N. Any nonempty. compact. So the ddimensional measure can only be ﬁnite and nonzero when 1=2 1 d 3 .63093 . homeomorphic to the Cantor set (but we will not prove this). . log 3 Hence the Cantor set has fractional dimension 0. V2 (X) the area of X. x → 3 x + 3 2 3 1 2 send C homeomorphically onto the subsets C ∩ [0. so the Cantor set is disconnected. Vd (X) = ∞ for k > d. Each of them is an contraction 1 with scale factor 3 . .. . U2 . in fact. x → 1 x and s1 : C → C . Hence. metric space which is both perfect and totally disconnected is called a Cantor set. The most important property of the Cantor set is its selfsimilarity. y) : x. so x and y are in diﬀerent components of Cn . For suppose that x. 1]. 2. y ∈ M } . We set d Hδ (M ) = inf n∈N n∈N Un and each set diam(Un )d : (Un )n∈N is a δcover for M . We will be particularly interested in subsets of RN with the Euclidean metric or RN with the chordal metric. . . that is d = log 2 = 0. Consider a subset X of RN and let Vd (X) denote the ddimensional volume of X. This gives another proof that the set C is uncountable. 3.63093 . the Cantor set is totally disconnected: the only connected subsets of C are the singletons. then we expect that 0 for k < d. we would expect Vd (C) = Vd (s0 (C)) + Vd (s1 (C)) = 1 d 3 Vd (C) + 1 d 3 Vd (C) = 2 1 d 3 Vd (C) . For the Cantor set C we know that C is the disjoint union of s0 (C) and s1 (C). . So V1 (X) is the length of X. All such sets are. If X is a set of dimension k = 1. The maps 1 s0 : C → C . If s is a contraction with scale factor λ then Vd (s(X)) = λd Vd (X) .The Cantor set is easily seen to be perfect. . V3 (X) the volume of X. .. The sets Cn for N 1 are disconnected. 76 Lecture 20 . . U3 . y were two diﬀerent points of C. For now we will do this informally.2 Hausdorﬀ Dimension Let M be a metric space with metric d. Later we will deﬁne the Hausdorﬀ dimension of a set and prove the results properly. for example the Cantor set or limit sets of Kleinian ∞ groups. Indeed. 20. How do these selfsimilarities act on ZN ? 2 We can use these selfsimilarities to ﬁnd the “dimension” of the Cantor set.
but the Hausdorﬀ measure at the critical value dimH M may be any number between 0 and ∞ including both endpoints. y) for all x. This shows that if Hs (M ) < ∞ if Hd (M ) > 0 then then Hd (M ) = 0 for s < d . HKδ (f (M )) Kd diam(Un )d . So there is a critical value at which the Hausdorﬀ dimension Hs (M ) jumps from ∞ to 0. then Hd (f (M )) K d Hd (M ) . o Proposition 20. It is usually very hard to calculate the Hausdorﬀ measure at this critical value and only a little easier to ﬁnd the Hausdorﬀ dimension. Furthermore. It is usually either 0 or ∞. A map f : M → N between two metric spaces is KLipschitz if d(f (x). Then each f (Un ) has diameter at most Kdiam(Un ). Any map f : RN → RN that is diﬀerentiable with bounded derivative is certainly Lipschitz by the mean value theorem. Indeed. Hs (M ) increases as δ we deﬁne the ddimensional Hausdorﬀ measure of M as: d d Hd (M ) = lim Hδ (M ) = sup Hδ (M ) : δ > 0 δ→0 0. then dimH f (M ) dimH M .1 If f : M → N is a KLipschitz map. for example. Now . Hs (M ) = ∞ for s < d . there is at most one dimension d for which Hd (M ) is ﬁnite and nonzero. d K d Hδ (M ). This measures the ddimensional size of the set M . Lecture 20 77 .2 Lipschitz maps preserve Hausdorﬀ dimension If f : M → N is a Lipschitz map. 0 for s > dimH M . So. for each set M . A map f : M → N is biLipschitz if it is Lipschitz and it has an inverse which is also Lipschitz. Proof: Suppose that (Un ) is a δcover for M . This value is called the Hausdorﬀ dimension dimH M of M . y ∈ M . Hence. For each 0 < δ < 1 and any δcover (Un ) of M we have diam(Un )d δ d−s diam(Un )s . A map f : M → N is Lipschitz if it is KLipschitz for some ﬁnite K. Suppose that M is a metric space and 0 s < d. Taking the limits as δ Corollary 20.d As we decrease δ. so (f (Un )) is a Kδcover for f (M ). Note that Hs (M ) = ∞ for s < dimH M . diam(f (Un ))d d So. we will show that. so the class of allowed δcovers is reduced. f (y)) Kd(x. every M¨bius transformation is biLipschitz. d So Hδ (M ) s δ d−s Hδ (M ). Such a map is certainly uniformly continuous. 0 gives the result.
K d Hd (M ) for every This Corollary shows that the Hausdorﬀ dimension of a set is unchanged by biLipschitz maps. α Cd(x. First observe that the set Ck consists of 2k intervals each of length 3−k . suppose that (Un ) is a δcover of C. the Hausdorﬀ dimension of the limit set Λ(G) of a Kleinian group G is unchanged by a M¨bius transformation. 1]. If M is the disjoint union of two subsets M1 . x → 3 x + 3 2 3 are each Lipschitz maps with Lipschitz constant 1 . Their inverses are also Lipschitz with constants 3. as required. for such a map. We are now in a position to prove properly that the Cantor set C has Hausdorﬀ dimension 1 log 2/ log 3. The contraction maps 3 1 s0 : C → C0 . y)α for all x. The Proposition shows that Hd (f (M )) value of the dimension d. For the converse. The Hausdorﬀ measure Hd is an (outer) measure. we need to show that for this dimension we do indeed have 0 < Hd (C) < ∞. we have 0 < Hd (C) < ∞ then we see that 2 so the Hausdorﬀ dimension d must satisfy 2 1 d 3 = 1 and =1 that is d= log 2 . for some value d. A map f : M → N is αH¨lder continuous if there is a constant C < ∞ with o d(f (x). So Λ(T GT −1 ) and Λ(G) have the same Hausdorﬀ dimension. Let C0 = C ∩ [0. f (y)) Show that. If. 78 . In particular. These intervals form a δcover of C for δ = 3−k . dimH f (M ) 1 dimH M . y ∈ M . o Exercise: 30.Proof: Suppose that f is KLipschitz. log 3 However. M2 . Hence dimH f (M ) dimH M . We will show that diam(Un )d This will prove that Hd (C) Lecture 20 1 2 1 2 . So d Hδ (C) 2k (3−k )d = (2 × 3−d )k = 1 because d = log 3/ log 2. x → 1 x and s1 : C → C1 . 3 ] and C1 = C ∩ [ 2 . This shows that Hd (C) = Hd (C0 ) + Hd (C1 ) = 2 1 d 3 Hd (C) 1 d 3 for every dimension d. Letting k ∞ this gives Hd (C) 1. 3 So 1 d Hd (C0 ) = Hd (C1 ) = 3 Hd (C) . then Hd (M ) = Hd (M1 ) + Hd (M2 ). We will not use measure theory but we will need the following very simple consequences: If M ⊂ N then Hd (M ) Hd (N ).
1 diam(Vn ) = (3 ) =2 and j=1 diam(Wj )d = 2K−k (3−K )d = 2−k (2 × 3−d )K = 2−k are equal. by expanding the sets Un by a small amount. So it suﬃces to prove that diam(Wj )d 1 when (Wj ) is a cover of C by the component intervals of CK . Now any two components of Ck are distance at least 3−k apart. So we can assume that (Un ) is a ﬁnite collection of open sets that cover C. all 2K intervals must appear in the sum and the result is clear. so Un can not meet more than one of them. Now consider one of the intervals Vn . The intersection CK ∩ Vn consists of 2K−k intervals each of K−k length 3−K . Choose an integer K so that each Un has diameter greater than 3−K . we can ensure that they are open. with 3−k−1 diam(Un ) < 3−k for some natural number k < K.First observe that. Let Vn be this component. Lecture 20 79 . Since C is compact. Then (Vn ) is a cover for C and diam(Un )d Hence it will suﬃce to prove that diam(Vn )d for a ﬁnite cover of C by the intervals Vn . so diam(Un ) 1 3 diam(Vn ). It has length 3−k . Since these are a cover. Suppose that Un is one of these sets. Then 2K−k d −k d −k d 1 3 diam(Vn ) = 1 3d diam(Vn )d = 1 2 diam(Vn )d . which we will denote by (Wj )2 j=1 . there is then a ﬁnite subcover.
compact subsets of M . D0 D1 D2 D3 D4 √ Each Dn consists of 4n squares with side length 4−k and hence diameter 4−k 2. y ∈ X there is a point c ∈ X.21 CALCULATING THE HAUSDORFF DIMENSION Proposition 21. so H1 (D) H1 ([0. with π1 (D) = [0. The Hausdorﬀ distance D on K(M ) is D(K. 1] × [0. 1]. Proof: Consider ﬁrst a subset X of R with dimH X < 1. 1]) = 1 . 1] ⊂ R2 . then Dn+1 is obtained by dividing each square into 16 smaller squares and keeping only the 4 shown in the diagram below. ∞) are then disjoint open sets containing x and y respectively with union M . Therefore we see that D has Hausdorﬀ dimension 1. L) = inf{δ > 0 : K ⊂ Lδ and L ⊂ K δ } . z → d(x. y are two distinct points of the metric space M .2 Hausdorﬀ distance The Hausdorﬀ distance D is a metric on K(M ). it is clear that D is totally disconnected. so X can not contain any such interval. Then f : M → R . c) and f −1 (c. If Dn has been deﬁned and consists of a ﬁnite number of disjoint closed squares. Lecture 21 80 .1 If a metric space M has dimH M < 1. However. It is called the δneighbourhood of K. z) is 1Lipschitz.1 Invariant Sets Let M be a metric space and K(M ) the collection of nonempty. The sets f −1 (−∞. 21. Any nonempty open interval U has H1 (U ) = 1. Covering D by these squares shows that √ H1 (D) 2. δ) : x ∈ K} is an open set containing K for each δ > 0. We construct Cantor dust as follows: D0 = [0. Cantor dust is the intersection Dn . / Now suppose that x. Lemma 21. then M is totally disconnected. Hence there is a c ∈ f (M ) with 0 = / f (x) < c < f (y). The projection π1 onto the ﬁrst coordinate is a 1Lipschitz map.2 shows that dimH f (M ) < 1. This means that between any two points x. If K is one of these compact sets then K δ = {B(x. so Corollary 20. The converse fails as the example of Cantor dust shows.
The compact set K is bounded. K) . . A subset F of M is an invariant set for C1 . K) D(M. c2 . provided that M is complete. so K = L. . Hence we obtain the triangle inequality: D(M. y ∈ M . L) = D(L. where c = max{c1 .L)+D(L. If D(K. . so Cn (K) is compact and hence C(K) ∈ K(M ). It is clear that D(K. . y) for all x. compact. Proposition 21. Cn (0))} + cR of the origin. So we see that M ⊂ K D(M. L ∈ K(M ). . Hence. .3 Invariant sets A ﬁnite set C1 . Let C1 . . . L). . . Choose a point x ∈ L. C(L)) Hence C is a contraction. Hence. c2 . . L ∈ K(M ) with K ⊂ Lδ we have Cn (K) ⊂ Cn (Lδ ) ⊂ Cn (L)cn δ . cD(K. cN respectively. compact. L) = 0. invariant set. Proof: For each compact subset K ∈ K(M ) deﬁne C(K) = n=1 N Cn (K). L) + D(L. so there is some δ > 0 with K ⊂ B(x. then K ⊂ {Lδ : δ > 0} = L and L ⊂ K. if we choose R large enough we will have C(Q) ⊂ Q. . L) . C2 . For a third set M ∈ K(M ) suppose that M ⊂ Lε and L ⊂ K δ . . CN when N F = n=1 Cn (F ) . . For two sets K. 0 A contraction on the metric space M is a map C : M → M for which there is a constant c with c < 1 and d(C(x). . CN of contractions on the Euclidean space EM have a nonempty. Its image under C lies within a distance max{d(0. .K) . . δ) ⊂ Lδ . This implies that Q ⊃ C(Q) ⊃ C 2 (Q) ⊃ C 3 (Q) ⊃ C 4 (Q) ⊃ . C2 . There is only one such nonempty. L) is ﬁnite. Lecture 21 81 . Each map Cn is continuous. K). C(K) ⊂ C(L)cδ This implies that D(C(K). so c D(K. . Consider the closed ball Q = B(0. C(y)) c d(x. then D(C(K). . . R). Then M ⊂ Lε ⊂ (K δ )ε ⊂ K δ+ε . cN } . If K and L were both nonempty. invariant set. . It remains to show that there is at least one invariant set. . We are looking for a set F with C(F ) = F . compact. A similar argument with K and L interchanged shows that the Hausdorﬀ distance D(K. .Proof: Let K. CN be a ﬁnite collection of contractions with constants c1 . invariant sets. C2 . C(L)) K = L. The contraction mapping theorem shows that such a map has a unique ﬁxed point in M .
n n=1 Proof: First note that the function f : s → N number d with n=1 cd = 1. We can also use the argument in the last proposition to give an upper bound on the Hausdorﬀ dimension of the invariant set. Show that there are other nonempty invariant sets. cn(K) diam(F ). C 2 (K). . Hence the proof above show that the sets K. C 3 (x). . C2 . n(K)) of K integers with 1 n(k) N . invariant set. C 2 (x). C(x). C 3 (K). and the intersection Exercise: 31. If c = max{cn : n = 1. Let cn < 1 be constants with d(Cn (x). compact. y) for all x. . . It is clear that C(F ) = F . CN be contractions for the Euclidean metric on EM with F as the nonempty. What is the unique nonempty. C n (Q) is therefore a nonempty compact set. 2. invariant set for C0 . . n cs is strictly decreasing. . . . . converge to the invariant set in the Hausdorﬀ metric for any starting set K ∈ K(M ). C(K). . so F is For a contraction mapping C : M → M we know that the sequence x. y ∈ EM . . Proposition 21. then K ⊃ C(K) ⊃ C 2 (K) ⊃ C 3 (K) ⊃ . . . . ◦ Cn(K) (F ) . compact. Since F is invariant. . n(2). . This has diameter at most cn(1) cn(2) . Cn (y)) Then dimH F cn d(x. N } < 1 then this diameter is at most cK diam(F ). C1 be the contractions on R given by 1 C0 : x → 3 x C n (K) is the invariant set.The intersection F = invariant. For any such sequence n deﬁne a set Cn (F ) as Cn (F ) := Cn(1) ◦ Cn(2) ◦ . Fix a natural number K and consider sequences n = (n(1). . Find the unique nonempty. . d where d is the unique solution to the equation N cd = 1 . . . If we choose K as a compact set with C(K) ⊂ K. C1 . . compact.4 Let C1 . converges to the ﬁxed point. . invariant set for a single contraction C : M → M ? Let C0 . . so there is a unique positive n The set F is bounded and so of ﬁnite diameter. and 1 C1 : x → 3 x + 2 3 . we have F = n Cn (F ) 82 Lecture 21 . .
. . V C1 (V ) C2 (V ) Lecture 21 83 . .where the union is over all sequences of K integers. H (F ) diam(F ) and the Hausdorﬀ dimension of F is at most d. 119120. 1) if d(C(x). Let d be the unique number with cd = 1. The proof above should be compared with the proof that the Hausdorﬀ dmeasure of the Cantor set is ﬁnite when d = log 2/ log 3. . C2 . compact invariant set F . Falconer. So Hδ (F ) diam(F )d for every n d d δ > 0. by K. Consequently. . C2 (V ). CN (V ) disjoint and n Cn (V ) ⊂ V . . These similarities have a unique nonempty. . cn(K) diam(F ))d = diam(F )d . see “Fractal Geometry”. For this cover we have diam(Cn (F ))d n n (cn(1) cn(2) . Suppose that there is a bounded open set V with C1 (V ). Combining this with the previous Proposition shows that the Hausdorﬀ dimension is exactly d and that 0 < Hd (F ) < ∞. . c2 . Theorem 21. For a proof of this result. . CN be similarities on the Euclidean space EM with scale factors c1 . . Then the Hausdorﬀ dmeasure Hd (F ) is greater than 0 and so the Hausdorﬀ dimension of F is at least d. although we need stronger restrictions on the contractions and the metric space M . The other part of the proof. can also be generalised. cN . for all x.5 Let C1 . showing that the Hausdorﬀ dmeasure is larger than 0. . We will not prove it since it would involve a little probability. C(y)) = c d(x. pp. . which is not examinable. . A map C : M → M is a similarity with scale factor c ∈ [0. y ∈ M . y) Such a similarity is certainly a contraction map. The Theorem below gives a useful result in this direction. Hence the sets Cn (F ) form a δcover for F provided that cK diam(F ) δ. d The last equality follows from the choice of d to satisfy cd = 1. .
1 The Cantor Set The two similarities C0 : R → R . This is n 1 d 2 × ( 3 ) = 1. 1] 3 3 C 2 ([0. Hence the dimension d of the Cantor set is the unique solution of cd = 1. 1) as the set V in Theorem 21. 1]) = C3 etc. 1]) = C2 C 3 ([0.22 EXAMPLES OF HAUSDORFF DIMENSION We can now determine the Hausdorﬀ dimension of many selfsimilar sets. We can take the open interval (0. 22. 1 ] ∪ [ 2 . For we see that C([0. so d = log 2/ log 3. 2 3 and the invariant set is the limit of this sequence.5 . 1]) = C1 = [0. Lecture 22 84 . x → 1 x and C1 : R → R . x → 1 x + 3 3 have the Cantor set as an invariant set.
3 1 For this we have 4 similarities each with scale factor 3 . Then 4( 1 )d = 1.2 The von Koch snowﬂake This is the curve constructed by an iterative process starting from the unit interval [0. At each stage. as in the diagram. 1) and height 2 3 as the set V . so the Hausdorﬀ dimension of the von Koch 3 snowﬂake is log 4/ log 3. Lecture 22 85 . 1] ⊂ R2 .22. each straight line segment is replaced by 4 line segments each 1 as long. V The von Koch snowﬂake Images of V . We can take the open isosceles triangle with √ 1 base (0.
Lecture 22 86 .Three copies of the von Koch curve making a snowﬂake.
22. We begin with a ﬁlled in equilateral triangle. so the Hausdorﬀ dimension of Sierpi´ski’s gasket is log 3/ log 2. We can take the interior of the initial 2 1 triangle as the set V . So we have 3 similarities 3 each with scale factor 1 . At each stage we replace each equilateral triangle by 3 triangles each 1 the size. Then 3( 2 )d = 1.3 The Sierpi´ ski Gasket n This is the set constructed as follows. n Lecture 22 87 .
There is a M¨bius transformation T that maps γ − onto γ + and o δ − onto D \ δ + . The group G acts discontinuously on all of the remainder of P and the set Φ = P \ (∆− ∪ ∆+ ) is a fundamental set. δ + . Let G be the group generated by T . So we may take T = A−1 ◦ U ◦ A. Proof: There is a hyperbolic geodesic ν normal to both γ − and γ + . We will sometimes abuse the notation by writing (P \ Λ(G))/G as P/G. w− to 0 and w+ to ∞. Lecture 23 88 . ∞ Now the map R+ z U :z→ R− is a hyperbolic M¨bius transformation that maps A(γ − ) onto A(γ + ). The quotient (P \ Λ(G))/G is obtained from Φ by identifying the two circles Γ− and Γ+ to obtain a torus. These bound two disjoint halfplanes δ − .1 Fuchsian Groups We begin with the following simple exercise. w+ where we may assume that w− is in the boundary of δ − and w+ in the boundary of δ + . So the geodesics A(γ ± ) must be halfcircles perpendicular to this: A(γ ± ) = {z ∈ R2 : z = R± }. The quotient D/G is obtained by identifying the two sides γ − o and γ + of f to get an annulus (ring) as shown above. γ + be two hyperbolic geodesics in D that do not meet either in D or on its boundary. Proposition 23. This has endpoints w− .1 Let γ − . o φ δ− T δ+ γ− φ γ+ γ− = γ+ D/G The region f = D \ (δ − ∪ δ + ) between γ − and γ + is a fundamental set for the group G generated by the single M¨bius transformation T .23 SCHOTTKY GROUPS 23. Conjugate by a M¨bius o transformation A chosen to map D onto R2 . The M¨bius o transformation T acts on all of the Riemann sphere and maps ∆− onto P \ ∆+ . We can also think of D as a subset of the Riemann sphere P. Then ν is mapped to the imaginary + axis. Then the limit set Λ(G) consists of just the two ﬁxed points of T . These circles enclose two disjoint discs ∆± . The geodesics γ ± are parts of circles Γ± orthogonal to the unit circle ∂D.
when we think of G acting on H3 we get a quotient H3 /G which is a 3dimensional solid that has (P \ Λ(G))/G as its boundary. The group G of M¨bius transformations acts on D and we obtain a surface D/G for the quotient. These bound disjoint halfspaces H± . The circles Γ± o are the boundaries of hyperbolic planes G ± in B 3 = H3 . This surface is o orientable since each M¨bius transformation is orientation preserving. We can also think of the M¨bius transformation acting on hyperbolic 3space H3 . Similarly. We can do the same for Fuchsian groups G generated by more than one element. the two parts D/G and J(D)/G are separated by the quotient of the limit set. The quotient H3 /G is obtained from F by identifying the two planes G − and G + to get a solid torus.) Lecture 23 89 . G acts on J(D) = P\D. Then (P \ Λ(G))/G is the boundary of H3 /G. The torus (P \ Λ(G))/G is obtained by taking the two annuli D/G and J(D)/G and joining them along their boundaries. The inversion J in ∂D induces an orientation reversing bijection from D/G to J(D)/G. When we think of G acting on all of the Riemann sphere we get a quotient (P\Λ(G))/G that consists of D/G and J(D)/G joined together along their boundaries. Similarly. The M¨bius transformation T maps D− onto H3 \ D+ . (Recall that we have seen examples of Fuchsian groups which have all of the unit circle in the limit set Λ(G). o The quotient J(D)/G is another surface. The set F = H3 \ (D− ∪ D+ ) is a fundamental set o for G. In this case. called the dual of D/G.∆− T ∆+ Γ− Φ Γ+ (P \ Λ(G))/G Γ− = Γ+ Note that inversion J in the unit circle ∂D maps D to the complementary disc J(D) and sends Φ to itself with the part φ inside D going to the part J(φ) outside D.
Then P/G is the boundary of H3 /G. The circles Γ± o are the boundaries of hyperbolic planes G ± in B 3 = H3 . let Γ− and Γ+ be disjoint circles o k k bounding 2K disjoint discs ∆− and ∆+ . We can do the same when we have more than one pair of circles. ∆− . ∆+ in P.2 Schottky Groups In the previous section all of our circles were orthogonal to the unit circle. . o Proposition 23. The set Φ = P \ (∆− ∪ ∆+ ) is a fundamental set for the group G acting on P \ Λ(G). In this case we get a Schottky group generated by several M¨bius transformations. Then J + (J − (zo )) = S(zo ) = zo so J − (zo ) = J + (zo ). Then S = J + ◦ J − is a nonidentity M¨bius transformation. The proposition shows that there are M¨bius transformations o k k Tk that map Γ− onto Γ+ and ∆− onto P \ ∆+ . more accurately. The special case when the circles Γn are orthogonal to the unit circle was the Fuchsian case dealt with in the previous section. For k = 1. ∆+ and T be as in the proposition. Then there is a M¨bius o transformation T that maps Γ− onto Γ+ and ∆− onto P \ ∆+ . Γ+ . Let Φ be the set Φ=P\ k=1 K ∆− ∪ ∆+ k k . Let J ± be inversion in the circle Γ± . These are the Schottky groups. T2 . Proof: We could prove this by adapting the proof of Proposition 23. Conjugate by a M¨bius transformation A that sends zo to 0 and J − (zo ) to ∞. However. we will give a diﬀerent argument that works entirely on the Riemann sphere. The M¨bius transformation T maps D− onto H3 \ D+ . The set F = H3 \ (D− ∪ D+ ) is a fundamental set o for G. . We have already considered the Schottky groups for K = 1 but Schottky groups with more than one generator are more interesting.) The map T = A◦U ◦A−1 now has the required properties. S(J − (zo )) = J + ◦ J − ◦ J − (zo ) = J + (zo ) = J − (zo ) so J − (zo ) is also ﬁxed by S. So A(Γ± ) must be a circle {z ∈ P : z = r± } for some 0 < r± < ∞. Let Γ− . 90 Lecture 23 . 2. (It is hyperbolic if eiθ = 1 and loxodromic otherwise. the points zo and J − (zo ) must be distinct.23. The group G generated by T1 . So S has 2 ﬁxed points. Also. So the quotient is a torus (the surface of a ring doughnut).1 for hyperbolic 3space. Γ+ be two disjoint circles bounding two disjoint discs ∆− . Consequently the map U : z → eiθ r+ r− z maps A(Γ− ) onto A(Γ+ ). . K. we will begin by considering groups generated by a single M¨bius transformation. We can also think of the M¨bius transformation acting on hyperbolic 3space H3 . Note that J − and J + both interchange these two ﬁxed points. (P \ Λ(G))/G) is obtained from this fundamental set by identifying the two circles Γ− and Γ+ . TK is called the k k k k (classical) Schottky group for the discs.2 Let Γ− . for variety. o Let zo be one ﬁxed point. . . As in the last section. . . . In this section we do not insist on this and produce Kleinian groups rather than Fuchsian groups. Since Γ− and Γ+ are disjoint. The quotient H3 /G is obtained from F by identifying the two planes G − and G + to get a solid torus (the body of a ring doughnut). Then A ◦ J ± ◦ A−1 is o inversion in the circle A(Γ± ) and must interchange 0 and ∞. The group G generated by T has a limit set Λ(G) consisting of the two ﬁxed points of T . The quotient P/G (or. These bound disjoint halfspaces D± .
say Tj . Γ+ k T1 −1 T1 (Φ) T1 (Φ) Φ T2 −1 T2 (Φ) T2 (Φ) The Tessellation for the Schottky Group. One of these is Tk (Γ− ) = j k k but all of the others are strictly inside ∆+ . Now apply another one of the generators. We will see shortly that the intersection of any chain of nested discs is a point and the closure of these points is the limit set Λ(G).This is the region outside all of the discs ∆± except that we have included half of the bounding circles. The k image Tj (Tk (Φ)) is a subset of Tj (∆+ ). is tessellated by the images g(Φ) for g ∈ G. The image Tk (Φ) lies inside ∆+ and is bounded by the 2K circles Tk (Γ± ). Hence we get a pattern of j k nested discs as shown below. The remainder of the Riemann sphere: P \ Λ(G). k We will show that Φ is a fundamental set for the Schottky group G acting on P \ Λ(G). which is itself a disc inside ∆+ . These are clearly locally ﬁnite so we see that G acts discontinuously on P \ Λ(G). Lecture 23 91 .
K with the closure of all these 2K discs disjoint. Join the vertices labelled g and g ◦ Tk for k = 1. k=1 Proof: To see this we need to look more closely at the tessellation shown above. 2. 2. The Schottky group generated by the M¨bius transformations pairing discs ∆− and ∆+ for k = o k k 1. Two vertices g and h −1 in this graph are adjacent when g = hA for A one of the generators Tk or their inverses Tk . It is simplest to see the structure of this tessellation if we simplify the diagram. 3 . 1 Theorem 23. We have done this below for the case of two pairs of discs and coloured the edges from g to g ◦ T1 in blue and the edges from g to g ◦ T2 in red.T1 T2 (Φ) −1 T1 T2 (Φ) T1 (Φ) 2 T1 (Φ) Enlargement of the region ∆+ . is a free group on the generators (Tk )K .3 Schottky groups are free groups. . So draw a graph — the Cayley graph for G — by putting a vertex for each image g(Φ) and labelling it with the element g of G. Lecture 23 92 . . . . . . . K.
A1 ◦ A2 . So g = I unless the product is trivial. . Consider a product A1 ◦ A2 ◦ A3 ◦ . Hence G is a free group. .−1 −1 T1 T2 −2 T1 2 T1 T1 T2 −1 T1 T2 −1 T1 T1 −1 T1 T2 I −1 −1 T2 T1 −1 T2 T2 T2 T1 −1 T2 T1 −2 T2 2 T2 −1 T2 T1 The Cayley graph for a Schottky Group. ◦ AN = g follows edges from I to g and never turns back on itself. Repeat this until there are no such pairs. ◦ AN −1 where N 0 and each An is one of Tk or Tk . . A1 ◦ A2 ◦ A3 . Lecture 23 93 . then we can cancel them and reduce the length N by 2. Since there are no loops in the graph. In the graph. We need to show that no such product is the identity I (except for the trivial product of no elements). . . A1 . the path I. . . A1 ◦ A2 ◦ A3 ◦ . An+1 have An ◦ An+1 = I. . If two successive terms An . this implies that the path can not return to its starting point.
The set F = H \ (D ∪ D − + The quotient H3 /G is obtained from F by identifying the two planes Gk and Gk for k = 1. 2. . K. 3 This gives a solid ball with K handles. − + It is convenient to work in H3 . Then (P \ Λ(G))/G is the boundary of H /G. The M¨bius o − + 3 3 − + ) is a fundamental set for G. k k Γ− = Γ+ 1 1 Γ− = Γ+ 2 2 We can also think of the Schottky group acting on the hyperbolic 3space H3 . The circles Γ± are k ± ± the boundaries of hyperbolic planes Gk in B 3 = H3 . . transformations Tk maps Dk onto H \ Dk . . . − D1 + D1 F − D2 + D2 23.The quotient (P\Λ(G))/G is obtained from the fundamental set Φ by identifying the pairs of circles ∆− and ∆+ for k = 1. . 2. We will prove k this and also obtain a bound on the Hausdorﬀ dimension of the limit set. . K. . . These bound disjoint halfspaces Dk . This means that any curve in the fundamental set F that joins one of these planes to another must have length at least t. This gives a sphere with K handles. Lecture 23 94 .2 The Limit Set for a Schottky Group The pictures above suggest that nested sequences of the discs obtained as images of the basic discs ∆± under the group G are single points and that these points are dense in the limit set. so they are all at least a nonzero hyperbolic distance t apart. The hyperbolic planes Dk and Dk do not meet either in H3 nor on its boundary.
It follows from this lemma that an element g of length N in G has g(F) at a hyperbolic distance ρ N t from 0 and hence the Euclidean diameter of g(F) is at most 2/sinh N t. .4 Let D be a hyperbolic plane at a hyperbolic distance ρ from the origin in B 3 = H3 . For large ρ the observation that diam(D) 2 is better. The inequality is only useful when ρ is small. There is 1 element of length 0. Lemma 23. Let γ be the shortest hyperbolic path from this base point to the set g(F). 2K of length 1. Then the Euclidean diameter of D is at most 2/ sinh ρ.We can write any g ∈ G as a product g = A1 ◦ A2 ◦ . . and 2K(2K − 1)N −1 of length N . ◦ AN (F) from one of the bounding planes to another. This is essentially the same result as Lemma 19. This means that it must cross each of the regions A1 (F). .4 . So r= 1 − tanh2 1 ρ 1 − c2 1 2 = = . Since G is a free group. Since D is orthogonal to the unit sphere. . A1 ◦ A2 (F). 1 2c sinh ρ 2 tanh 2 ρ 1 0 r c k D The Euclidean diameter of D is at most 2r. Pythagoras’ Theorem shows that 1 + r2 = k2 = (c + r)2 . . . ◦ AN −1 where N 0. Proof: The hyperbolic plane D is part of a Euclidean sphere with centre k and Euclidean radius r. N − 1. Choose a base point in F. Lecture 23 95 . A1 ◦ A2 ◦ . Hence the hyperbolic length of γ must be at least N t. each An is one of Tk or Tk and An ◦ An+1 = I for n = 1. We will call it the length of g. When we look at this on the Riemann sphere it shows that diam(g(Φ)) 2/ sinh N t. The shortest path from the origin to D is a radial line from 0 to k that crosses the plane D at a point 1 c with c = tanh 2 ρ. ◦ AN −1 (F) and ﬁnally meets g(F). . then A1 ◦ A2 (F) and so on until it crosses into A1 ◦ A2 ◦ . 2. say the origin 0. the length N of such a product is determined by g. . . . This path begins in F. . . . then crosses into A1 (F). .
Thus we have gn = A1 ◦ A2 ◦ . Tk : k = 1. (Look at the diagram of the Tessellation for the Schottky Group. So we see that the ddimensional Hausdorﬀ measure is 0 for d> log(2K − 1) . Then Cn surrounds a disc containing all of the images gm (Φ) for m > n. So the Hausdorﬀ ddimensional measure of Λ(G) is at most d 2 d Hδ (Λ(G)) 2K(2K − 1)N −1 sinh N t for δ > 2 sinh N t . Note that we can not have An ◦ An+1 = I or else gn−1 = gn+1 and the two images gn−1 (Φ) and gn+1 (Φ) are the same. 2. . ◦ An . Hence this disc must contain a limit point of G. g(0))) converges. For any natural number N . Therefore the limit set is the complement of all these images g(Φ). sinh nt This certainly shows that their Euclidean diameter tends to 0 as n → ∞. . Our argument shows that the complement consists of 2K(2K − 1)N −1 2 discs each with Euclidean diameter at most sinh N t . . t (Much more delicate arguments show that the Hausdorﬀ dimension of the limit set is actually equal to the inﬁmum of those d for which the series exp(−dρ(0. The images gn (Φ) touch each other along a sequence of circles that are nested inside one another. no point of any of the images g(Φ) can be in the limit set since there are only ﬁnitely many copies h(Φ) for h ∈ G that border it. Suppose that Cn is the circle separating gn (Φ) from gn+1 (Φ). t This proves that the Hausdorﬀ dimension of the limit set is at most log(2K − 1) . K}. Now −1 the fundamental domain Φ only meets the images A(Φ) where A is one of Tk or Tk . . . We can also use this idea to ﬁnd an upper estimate for the Hausdorﬀ dimension of the limit set Λ(G). However.) g∈G Lecture 23 96 . When N is large this is approximately 2K(2K − 1)N −1 4e−N t d = 2K × 4d (2K − 1)e−td 2K − 1 N .Consider the tessellation given by the images g(Φ) for g in the Schottky group G.) We can use the previous lemma to estimate the size of these circles as diam(gn (Φ)) 2 . We wish to study a chain of these images gn (Φ) where every successive pair gn (Φ) and gn+1 (Φ) meet along a circle. . Hence gn (Φ) and −1 gn+1 (Φ) only meet when gn+1 = gn ◦ An where An ∈ {Tk . the complement of the union: {g(Φ) : g ∈ G has length at most N } certainly contains the limit set. So every sequence (gn ) as above gives us a point of the limit set.
This means that the generator T1 also varies. Ultimately. the point P corresponds to a singular point of the surface. It is interesting to k ask what happens as they move closer to one another and.1 How Schottky Groups Degenerate So far we have insisted that the discs ∆± for our Schottky groups are disjoint. As the discs move closer together. the green curve reduce to a point and the two discs ∆− 1 and ∆+ touch at a point P . However. Consider the Schottky group for the disjoint discs (∆± )K . then G could only be discrete if Tk were of ﬁnite order. if the circles crossed and the map Tk were elliptic. For example. and 1 1 k k=1 allow them to move closer. In the picture below the shortest (chordal) path between the two discs is marked in green. say ∆− and ∆+ . the group generated by them need not be o k discrete. Take one pair. so this green curve becomes shorter.24 DEGENERATE SCHOTTKY GROUPS 24. There are still M¨bius transformations that pair the circles Γ± . eventually touch or cross. 1 Γ− = Γ+ 1 1 ∆− 1 ∆+ 1 ∆− 2 ∆+ 2 Γ− = Γ+ 2 2 Γ− = Γ+ 1 1 ∆− 1 P ∆+ 1 ∆+ 2 ∆− 2 Γ− = Γ+ 2 2 Lecture 24 97 . In the quotient.
The transformation T1 also changes as we vary the discs. Lecture 24 98 . in the limit. T1 becomes a parabolic transformation that ﬁxes P . Then we get a tessellation as shown below. The most interesting case is when.
the group is a Fuchsian group. each of these quadrilaterals becomes a sphere with 3 punctures. A B D C The limit set in this case is a closed Jordan curve contained in the chains of circles shown below. in the special case where the 4 points A. k Lecture 24 99 . The fundamental set then breaks into two parts. both quadrilaterals bounded by arcs of each of the four circles. Let us consider the case where we have 2 pairs of circles Γ± and Γ± as shown below. C. Show that. T2 are parabolic with 1 2 T1 : A → A . Exercise: 32. T2 : B → D . T2 : C → C . B. We will also assume that the transformations T1 . When we identify the edges using T1 and T2 . T1 : B → D . D lie on a circle that crosses the 4 circles Γ± orthogonally.We can also let other pairs of circles touch.
z This is then a Fuchsian group preserving the upper and lower halfplanes.A special case of this is when our discs are: ∆− = {z ∈ C∞ : Re(z) < −1} 1 ∆− = {z ∈ C∞ : z + 1  < 1 } 2 2 2 The transformations are ∆+ = {z ∈ C∞ : Re(z) > 1} 1 1 ∆+ = {z ∈ C∞ : z − 2  < 1 } 2 2 1 . T1 : z → z + 2 and T2 : z → − ∆− 2 ∆+ 2 2 1 ∆− 2 0 ∆+ 2 1 2 ∆− 1 ∆+ 1 ∆− 1 ∆+ 1 Lecture 24 100 . Its limit set is the circle R ∪ {∞}. It is a subgroup of the modular group we looked at earlier.
The group is generated by two parabolic transformations ﬁxing points where two circles touch. Show that the limit set. The limit set is the Apollonian gasket. Four circles with each pair tangent. Exercise: 33.An even more degenerate case is when each disc touches the 3 others. This is the pattern obtained as follows: Given any 3 mutually tangent circles. the Apollonian gasket also has tetrahedral symmetry. there are 2 further circles each of which touches each of the original 3. Show that we can choose 4 circles each pair of which touch so that they have tetrahedral symmetry. This is illustrated below on the left. Limit set — The Apollonian Gasket Lecture 24 101 . Start with 3 mutually tangent circles and use this process to add further circles recursively.
By allowing these sorts of degenration for Schottky groups we can create very complicated. Lecture 24 102 .We ought really to draw this picture on the Riemann sphere. as illustrated below. An example is given below. fractal limit sets.
when G = Zω1 + Zω2 (with ω1 and ω2 linearly independent over R). the quotient C/G = C/(Zω1 + Zω2 ) is a torus. consider the group of translations G = {z → z + λ : λ ∈ Λ}. Cambridge University Press. that is. a cylinder or a torus. This has much more information. We can study them by studying the discrete groups of M¨bius o transformations that maps the unit disc to itself. It can be shown that these are homeomorphic to spheres with K handles for some K 0. including the complex structure on the surface. We call such a surface which looks locally like a piece of the complex plane a Riemann surface. C. we can obtain every compact Riemann surface k from a Schottky group. When Λ = {0}. u For further study I recommend the book Indra’s Pearls: The Vision of Felix Klein by David Mumford. by choosing the discs ∆± appropriately. Caroline Series. Hence every such compact orientable surface is the quotient of the Riemann sphere by a Schottky group. when Λ = Zω1 . For example. The Riemann Mapping Theorem (and the Uniformization Theorem) states that every Riemann surface is either the quotient of the unit disc D by a discrete group or else on of the special surfaces: P. Hence we can do complex analysis on the quotient D/G. Similar arguments apply to groups acting discontinuously on the complex plane. Each of the M¨bius transformations in G is o analytic and has an analytic inverse. 2002 (ISBN 0521352533). the quotient C/G = C/Zω1 is a cylinder. o Then the quotient D/G is an orientable surface.2 Riemann Surfaces * Let G be a group of M¨bius transformations that acts discontinuously on the unit disc D. Compact Riemann surfaces are compact orientable surfaces. This means that almost all Riemann surfaces are quotients of the unit disc. Lecture 24 103 . This leads us to think about classifying all compact Riemann surfaces by using the Schottky groups. Indeed. more is true. the quotient C/G is just C. and David Wright. This is a very active area of research known as the study of Teichm¨ller space. by studying Fuchsian groups. beautifully presented.*24.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.