NOTES ON HYPERBOLIC GEOMETRY

KYLE ORMSBY

Caveat emptor — These notes are in draft form and will evolve throughout the semester. Please
contact the author at [email protected] with any comments or corrections.
Much of these notes are highly parallel to Birger Iversen’s Hyperbolic geometry [Ive92] and they
should not be considered original work.

C ONTENTS
1. A surplus of triangles 2
2. What is hyperbolic geometry? 3
3. Quadratic forms 8
4. Real quadratic forms 11
5. The Lorentz group 16
6. Metric spaces and their isometries 21
7. Euclidean space 22
8. Spherical geometry 24
9. Hyperbolic space 27
10. The Klein and Poincaré disks 30
11. Möbius transformations 31
12. The Poincaré disk and half space 36
13. The Riemann sphere 39
14. The Poincaré half plane 42
15. The action of the special linear group on the upper half plane 43
16. Vector calculus on the trace zero model of the hyperbolic plane 45
17. Pencils of geodesics 47
18. Classification of isometries 49
19. The action of the special linear group in the trace zero model 51
20. Trigonometry 52
21. Angle of parallelism 54
22. Right-angled pentagons 55
23. Right-angled hexagons 56
24. Hyperbolic area 57
25. Fuchsian groups 58
26. Cusps and horocyclic compactification 62
27. The modular group and its fundamental domain 63
28. Locally finite and convex fundamental domains 65
29. Dirichlet domains 67
30. Compact polygons 68
31. Poincaré’s theorem 70
32. Triangle groups 72
33. The Klein quartic 74
References 75
1
 1. A SURPLUS OF TRIANGLES
This activity is inspired by Kathryn Mann’s notes DIY hyperbolic geometry [Man15]. Jay Ewing
kindly provided the class with approximately 900 equilateral triangles cut from 100# card stock
by the College’s laser cutter.
Take a stack of triangles and start taping them together along edges, with the restriction that
exactly n triangles meet at each vertex, where n is some fixed positive integer. You should find
some familiar shapes for n = 3, 4, 5: the tetrahedron, octahedron, and icosahedron.

F IGURE 1. The Platonic solids. You should have built the tetrahedron, octahedron,
and icosahedron. The cube and dodecahedron are built from squares and pen-
tagons, respectively. Image: Wikimedia Commons.

With n = 6, your job remains easy, but you now produce an infinite, flat triangular lattice.

F IGURE 2. The infinite triangular lattice produced by placing six triangles at each
vertex. Image: Wikimedia Commons.

When n = 7, life becomes more interesting. With 420◦ worth of angle at each vertex, the ar-
rangement cannot possibly fit in the plane. The shape you are creating is an approximation of the
hyperbolic plane.
2
 A geodesic in a metric space is a distance-minimizing curve. On the polygonal hyperbolic plane,
these are “straight lines,” where ‘straight’ has the obvious meaning on flat triangles, and lines con-
tinue straight over edges by temporarily flattening them. (Geodesics through vertices are prob-
lematic.)
Excercise 1.1. Draw several geodesics on your polygonal hyperbolic plane and try to describe
their features. More specifically:
(a) Draw two geodesics that start off parallel (inside of a triangle) but eventually diverge.
(b) Draw a geodesic that starts close to and parallel to an edge of a triangle. Observe that if follows
a path made up of triangle edges. Call such a path an edge path geodesic. Try to describe all
edge-path geodesics.
(c) Draw a large (not inside a paper triangle) triangle made up of geodesics.
(d) Is it possible to draw a large rectangle (quadrilateral where all sides meet at right angles)?
We will now investigate some properties of area in our polygonal hyperbolic plane.
Definition 1.2. A polygonal disk of radius r consists of all triangles with vertices at most r edges from
a central vertex. The combinatorial area of a region made up of triangles is the number of triangles
in the region. The combinatorial circumference of a region made up of triangles is the number of
triangle edges on the boundary of the region.
Excercise 1.3. Find formulæ for the combinatorial area and circumference of a polygonal disk of
radius r in the Euclidean plane (six triangles per vertex).
Excercise 1.4. Find formulæ or estimates for the combinatorial area and circumference of a polyg-
onal disk of radius r in the polygonal hyperbolic plane.
You may find Figure 3 useful in answering Excercise 1.4.

2. W HAT IS HYPERBOLIC GEOMETRY ?

Here are some slightly non-traditional axioms for Euclidean geometry. Throughout, we assume
that X is a nonempty metric space.1 A line (or geodesic line) in X is a distance-preserving function
γ : R → X, where R has the standard metric. Two lines are parallel if their images are disjoint.
An isometry is a function between metric spaces σ : X → Y which preserves distances and has a
distance-preserving inverse.

Incidence axiom: A unique line passes through any two distinct points of X.

Reflection axiom: If γ : R → X is a line in X, then X r γ(R) has two connected components.

There exists an isometry σ : X → X which fixes γ(R) pointwise but interchanges the connected
components of X r γ(R).

Parallel axiom: Given a line γ in X and a point p ∈ X r γ(R), there is a unique line through p
parallel to γ.

1Recall that a metric space is a set X equipped with a metric, that is, a function d : X × X → R such that for all x, y, z ∈
X, d(x, y) = 0 ⇐⇒ x = y (identity of indiscernibles), d(x, y) = d(y, x) (symmetry), and d(x, z) ≤ d(x, y) + d(y, z)
(triangle inequality).
3
F IGURE 3. A tessellation of the Poincaré disk model of the hyperbolic plane with
seven equilateral triangles around each vertex. Created with the utility Make hy-
perbolic tilings of images.

F IGURE 4. The author holding a paper model of a polygonal disk of radius 2 in

front of some preliminary data on Excercise 1.4.

4
 F IGURE 5. Henry Segerman has made some delightful 3D printed objects repre-
senting the polygonal hyperbolic plane. See this Numberphile video for a live
demonstration.

Theorem 2.1. A metric space which satisfies all three of these axioms is isometric to the Euclidean plane.
A space which satisfies the first two axioms but not the third is isometric to the hyperbolic plane H2 (after
rescaling).
We will not study the proof of this theorem, but rather state it to motivate the relative univer-
sality of the hyperbolic plane.
Remark 2.2. Spherical geometry is another interesting system, but it fails to satisfy the incidence
axiom: two antipodal points have infinitely many lines (great circles) passing through them.
The Poincaré half plane. The French polymath Henri Poincaré (1854–1912) is one of the key con-
tributors to the theoretical underpinnings of hyperbolic geometry. He introduced two models of
H2 , one based on the upper half plane in C, and the other based on the unit disk. He also supplied
one of the most memorable quotations regarding the process of mathematical discovery:
For fifteen days I strove to prove that there could not be any functions like those I
have since called Fuchsian functions.2 I was then very ignorant; every day I seated
myself at my work table, stayed an hour or two, tried a great number of combina-
tions and reached no results. One evening, contrary to my custom, I drank black
coffee and could not sleep. Ideas rose in crowds; I felt them collide until pairs in-
terlocked, so to speak, making a stable combination. By the next morning I had
established the existence of a class of Fuchsian functions, those which come from
the hypergeometric series; I had only to write out the results, which took but a few
hours.

The Poincaré upper half plane H is the set

H = {z ∈ C | Im z > 0}
equipped with the metric
|z − w|2
 
d(z, w) = arccosh 1 + .
2 Im(z) Im(w)
2These are analytic functions invariant under the action of a Fuchsian group. Fuchsian groups will be one of our
main topics of study and will be defined in due course.
5
 F IGURE 6. Henri Poincaré, a central figure in late nineteenth century geometry,
topology, and physics. Image: https://famous-mathematicians.com/
henri-poincare/

10

-3 -2 -1 1 2 3

F IGURE 7. The plot of cosh.

Recall that the hyperbolic cosine function is

ex + e−x
cosh(x) =
2
with plot shown in Figure 7.

The function arccosh is the inverse of cosh restricted to the nonnegative reals.
The lines in H are Euclidean circles with centers on the real axis or Euclidean lines perpendicu-
lar to the real axis. The reader is invited to think about how this model satisfies the incidence and
reflection axioms, but not the parallel axiom, perhaps by consulting Figure 8.

6
 F IGURE 8. Some not-so-random geodesics in H. Image: https://thatsmaths.
com/2013/10/11/poincares-half-plane-model/.

The group SL2 (R) of 2 × 2 real matrices with determinant one acts on H via isometries via the
assignment
 
az + b a b
z 7−→ for ∈ SL2 (R).
cz + d c d
In particular, SL2 (Z) acts on H via isometries, and this is crucial to interactions between hyperbolic
geometry and number theory.

The Poincaré disk. The open unit disk3 D = {z ∈ C | |z| < 1} can also be used as a model for
H2 . The metric again has an expansion in terms of an inverse hyperbolic trig function, and the
geodesics are equally easy to describe: they are Euclidean circles and lines orthogonal to the unit
circle S 1 = ∂D. See Figure 3 for an aesthetically pleasing arrangements of geodesics in D.

Discrete groups. Reflections in the lines of the tessellation in Figure 3 generate a discrete group of
isometries of H2 . Starting with a discrete group Γ of isometries of H2 , one may construct a polygon
∆ with a side pairing (called a fundamental domain) so that H2 /Γ is reconstructed by gluing together
the prescribed sides of ∆.
Poincaré was concerned with the converse problem of determining when a given hyperbolic
polygon with side pairing generates a discrete group with the given polygon as fundamental
domain. The necessary and sufficient conditions he deduced are called Poincaré’s Theorem, and the
elucidation of its statement and proof will be one of our primary concerns this semester.
When Γ is a discrete of Isom(H2 ), we call it a Fuchsian group. The quotient spaces H2 /Γ does not
satisfy the incidence and reflection axioms, but it does locally satisfy them; in other words, H2 /Γ is
locally isometric to H2 and as such is a geometry called a hyperbolic surface. It turns out that every
hyperbolic surface is isometric to some H2 /Γ, Γ a Fuchsian group, highlighting the importance of
these objects.

3Or is it disc? The spelling ‘disk’ seems to be preferred in American English, unless one is interacting with antiquated
compact disc (CD) technology.
7
 F IGURE 9. Poincaré named the discrete subgroups of Isom(H2 ) after Lazarus Fuchs
(1833–1902).

Sins of ommission. Following Iversen [Ive92], we will take a metric and linear algebraic ap-
proach to hyperbolic geometry, largely forgoing differential and Riemannian geometry (including
curvature).

3. Q UADRATIC FORMS
Students from last semester’s Math 412 must bite their tongues during the following two lec-
tures.
Orthogonality. Let k be a field of characteristic different from 2, and let E be a k-vector space of
finite dimension n. A quadratic form on E is a function Q : E → k satisfying Q(λv) = λ2 Q(v) for all
λ ∈ k, v ∈ E, and such that its polarization h , i = h , iQ : E × E → k given by
h , i : E × E −→ k
1
(v, w) 7−→ (Q(v + w) − Q(v) − Q(w))
2
is a symmetric bilinear form. Note that char k 6= 2 is essential for defining the polarization.
Excercise 3.1. If k = R, E = Rn , and Q : Rn → R is twice-differentiable and satisfies Q(λv) = λ2 v,
then the polarization of Q is necessarily a symmetric bilinear form.
Proposition 3.2. Polarization is a bijective correspondence between quadratic forms on E and
symmetic bilinear forms on E.
Proof. We claim that the assignment h , i 7−→ (v 7→ hv, vi) is a two-sided inverse to polarization.
Note that hλv, λvi = λ2 hv, vi, so this map is in fact quadratic. The computation
1 1
(Q(2v) − Q(v) − Q(v)) = (4Q(v) − 2Q(v)) = Q(v)
2 2
show that a quadratic form may be recovered from its polarization. We leave the other direction
as an exercise for the reader. 
8
 If a quadratic/symmetric bilinear form is fixed, we call Q(v) = hv, vi the norm of v (with re-
spect to the fixed quadratic/symmetric bilinear form). Two vectors v, w ∈ E are orthogonal when
hv, wi = 0. Two linear subspaces V, W ≤ E are orthogonal when hv, wi = 0 for all v ∈ V , w ∈ W .
Given a linear subspace V ≤ E, we write
V ⊥ = {w ∈ E | hv, wi = 0 for all v ∈ V }
and call this space the orthogonal complement of V .
Excercise 3.3. Check that V ⊥ is a linear subspace of E.
Definition 3.4. A quadratic form Q with polarization h , i on a finite-dimensional vector space E
is called nonsingular if hv, wi = 0 for all v ∈ E implies that w = 0.
Excercise 3.5. Given a symmetric bilinear form h , i on E, there is an associated linear transforma-
tion
q : E −→ E ∗
v 7−→ (w 7→ hv, wi)
where E ∗ is the k-linear dual 4 of E. Show that h , i is nonsingular if and only if the associated
map E → E ∗ is an isomorphism.
Theorem 3.7. Let E be a finite-dimensional k-vector space equipped with a nonsingular quadratic form
Q. Then for any linear subspace V ≤ E, we have
dim V + dim V ⊥ = dim E.
Proof. By Excercise 3.5, the map q : E → E ∗ is an isomorphism. The inclusion map i : V ,→ E has
dual
i∗ : E ∗ −→ V ∗
f 7−→ f ◦ i.
We claim that i∗ is surjective. To wit, if v1 , . . . , vk is a basis of V , we may extend it to a basis
v1 , . . . , vn of E. The reader may check that the matrix for i∗ with respect to bases v1∗ , . . . , vn∗ and
v1∗ , . . . , vk∗ is (Ik | 0), so i∗ is surjective.

4 Dual vector spaces play an important role in linear algebra and an outsized one in the theory of symmetric bilinear
and quadratic forms.
Definition 3.6. The k-linear dual of a k-vector space V is the Hom space
V ∗ = Homk (V, k).
Elements of V ∗ are called linear functionals or dual vectors.
There is a canonical map
V −→ (V ∗ )∗
v 7−→ (f 7→ f (v))
which is always injective and is an isomorphism when V is finite-dimensional. (We call the map canonical because it
does not depend on the choice of a basis or coordinates.) For V finite-dimensional, it is also the case that V ∼ = V ∗ , but
this isomorphism in non-canonical. Indeed, after choosing an ordered basis {v1 , . . . , vn } of V , we create a dual basis
{v1∗ , . . . , vn∗ } where vi∗ (vj ) is either 1 or 0 depending on whether j = i or j 6= i. It is straightforward to prove that
{v1∗ , . . . , vn∗ } is a basis of V ∗ and the linear map taking vi to vi∗ is an isomorphism.
Given a linear transformation f : V → W , we can form the dual transformation f ∗ : W ∗ → V ∗ which takes g : W → k
to the composite linear functional g ◦ f . This defines an injective linear transformation
Homk (V, W ) −→ Homk (W ∗ , V ∗ )
which is an isomorphism when both vector spaces are finite-dimensional. (Check this!)
9
 Now consider the the composite linear map i∗ ◦ q : E → V ∗ which is also surjective (since it is
the composite of an isomorphism and then a surjection). By rank-nullity,
dim E = dim V ∗ + dim ker(i∗ ◦ q).
Finally, one may check that ker(i∗ ◦ q) = V ⊥ , and we already know dim V ∗ = dim V , so we deduce
the desired equality. 
Given a quadratic form Q on a vector space E with h , i = h , iQ , one may form a Gram matrix
for Q by choosing a basis v1 , . . . , vn of E and then creating G ∈ Matn×n (k) with
Gij = hvi , vj i .
Note that G is a symmetric matrix. Also note that G is the matrix for q : E → E ∗ with respect to
the bases v1 , . . . , vn and v1∗ , . . . , vn∗ .
Excercise 3.8. Show that Q is nonsingular if and only if the Gram matrix G of Q has det G 6= 0.
Let (E, Q) be a vector space with quadratic form Q. An isometry of (E, Q) with another such
pair (E 0 , Q0 ) is a linear isomorphism σ : E → E 0 such that Q = Q0 ◦ σ. The automorphisms of (E, Q)
are the isometries (E, Q) → (E, Q). Under composition, automorphisms form a group O(Q) — or
O(E) when no confusion is possible — called the orthogonal group of (E, Q).
Proposition 3.9. Let Q be a nonsingular form on E. If σ ∈ O(Q), then det σ = ±1.
Proof. Pick a basis v1 , . . . , vn for E and let A be the matrix for σ with respect to this basis. Then
* +
X X
hσ(vi ), σ(vj )i = Aki vk , Ahj vh
k h
X
= A>
ik hvk , vh i Ahj
h,k

which show that A> GA is the Gram matrix for Q ◦ σ (with respect to v1 , . . . , vn ). In particular,
σ ∈ O(Q) if and only if A> GA = G. Applying the determinant, we get
det G = det A> det G det A = (det A)2 det G,
whence det A = ±1. 
P 2
When Q(x) = xi for x = (x1 , . . . , xn ) ∈ kn , we write On (k) for O(Q). Note that Q has Gram
matrix In , so A ∈ On (k) if and only if A> A = In .
Theorem 3.10. Let Q be a nonsingular quadratic form on a vector space E over C (or any algebraically
closed field). Then there exists a basis of E with respect to which Q has Gram matrix In .
Proof. See [Ive92, p.5]. 
Witt’s theorem. Fix a field k with char k 6= 2, let E be a finite-dimensional k-vector space, and let
Q be a quadratic form on E. Set h , i = h , iQ . A vector v ∈ E is isotropic if Q(v) = 0; otherwise, v
is nonisotropic. Given a nonisotropic vector v, define a linear transformation
τv : E −→ E
hx, vi
x 7−→ x − 2 v.
hv, vi
The reader may check that
hτv (x), τv (x)i = hx, xi ,
10
so τv ∈ O(Q). The orthogonal transformation τv is called reflection along v. Note that τv |v⊥ = id,
while τv (v) = −v. Hence τv is an involution: τv ◦ τv = id, but τv 6= id. Also note that det τv = −1.
We say that a group G acts on a set X when there is a function G × X → X, (g, x) 7→ g · x such
that e · x = x and g · (h · x) = (gh) · x. A group action is transitive when for all x, y ∈ X, there
exists g ∈ G such that g · x = y. The following proposition says that O(Q) acts transitively on any
“Q-sphere” in E.
Proposition 3.11. For any λ ∈ k× , O(Q) acts transitively on the set
SQ (λ) = {v ∈ E | Q(v) = λ}.
Proof. Given σ ∈ O(Q) and v ∈ SQ (λ), we define σ · v = σ(v). Since σ is an isometry, Q(σ(v)) =
Q(v) = λ, so σ · v ∈ SQ (λ). The other properties of a group action are obvious.
To prove transitivity, suppose that v, w ∈ SQ (λ). These vectors are nonisotropic since Q(v) =
Q(w) = λ 6= 0. Also observe that
hv − w, v + wi = hv, vi − hw, wi = λ − λ = 0,
that is, v − w and v + w are orthogonal. Note then that v = 12 ((v − w) + (v + w)), and thus
λ = Q(v) = 41 h(v − w) + (v + w), (v − w) + (v + w)i = Q(v − w) + Q(v + w). Since λ 6= 0, we
conclude that we cannot have both Q(v − w) and Q(v + w) equal to 0.
If v − w is nonisotropic, then
τv−w (v − w) = w − v and τv−w (v + w) = v + w.
Adding these formulæ gives τv−w (v) = w. A similar computation shows that if v + w is non-
isotropic then τv+w (v) = −w, whence τw ◦ τv+w takes v to w. We conclude that the subgroup of
O(Q) generated by reflections acts transitively on SQ (λ). 
Remark 3.12. Cartan’s theorem — which we will not prove — implies that O(Q) is generated by
reflections.
Theorem 3.13 (Witt). Let (E, Q) be a nonsingular quadratic form. Any isometry σ : (U, Q) ∼ = (V, Q),
where U, V are linear subspaces of E, can be extended to an orthogonal transformation of (E, Q).

Proof. First suppose that U is nonsingular and proceed by induction on dim U . If dim U = 1, pick
0 6= u ∈ U and set λ = Q(u). Since Q(σ(u)) = λ, we can apply Proposition 3.11 find σ 0 ∈ O(Q)
such that σ 0 (u) = σ(u). It is easy to check that σ 0 |U = σ, so σ 0 is our desired extension.
For the induction step, pick a nonisotropic vector u ∈ U . Use Proposition 3.11 to choose
τ ∈ O(Q) such that σ(e) = τ (e). It suffices to extend τ −1 ◦ σ : U → E to an orthogonal trans-
formation of E. This means that, without loss of generality, it suffices to extend σ : U → E under
the assumption that σ fixes a nonisotropic vector u ∈ U . Let W be the ortgonal space to u in U , and
let V be the orthogonal space to u in E. Then W and V are nonsingular, W ≤ V , and σ(W ) ≤ V .
By the induction hypothesis, σ|W : W → V extends to an orthogonal transformation of V , and
this extension may be extended to E via the identity, completing our proof in the case where U is
nonsingular.
We leave the case of U singular as a reading exercise; see [Ive92, pp.7–8]. 

4. R EAL QUADRATIC FORMS

We now specialize to k = R, the field of real numbers.
11
 F IGURE 10. Ernst Witt (1911–91) was a leading algebraist in the 20th century and
founded the theory of quadratic forms over an arbitrary field. Image: Oberwolfach
Photo Collection.

Sylvester types. The isometry type of a real quadratic form is determined by two numbers, called
the Sylvester type. In order to develop this theory, we first generalize the notion of orthonormal
basis.
Proposition 4.1. Let (E, Q) be a quadratic form on a real vector space E of dimension n. There
exists a basis e1 , . . . , en for E such that
(
0 6 j,
if i =
hei , ej i =
±1 or 0 if i = j.
Proof. We proceed by induction on n = dim E. If Q = 0, any basis for E satisfies these properties.
Thus we may assume there exists v ∈ E which is nonisotropic. Write Q(v) = ελ2 with ε = ±1
and λ ∈ R× . Set e1 = λ−1 v to get Q(e1 ) = ±1. This covers the n = 1 case. For n > 1, invoke the
induction hypothesis to find e2 , . . . , en a basis of the desired type for e⊥
1 . The basis e1 , . . . , en meets
our requirements. 
Definition 4.2. Call a basis of the type guaranteed by Proposition 4.1 an orthonormal basis.
The careful reader will note that this is a larger class of bases than those usually termed or-
thonormal (for which hei , ej i = δij ).
Theorem 4.3 (Sylvester). Let e1 , . . . , en be an orthonormal basis for (E, Q). The numbers
p = #{i | hei , ei i = −1} and q = #{i | hei , ei i = 1}
are independent of the orthonormal basis considered.

Proof. Fix an orthonormal basis e1 , . . . , en and let E− be the subspace spanned by those ei for
which hei , ei i = −1 or 0. Then Q(v) ≤ 0 for all v ∈ E− . If F is any positive definite5 subspace of E,
5A real quadratic form Q is positive definite if Q(v) > 0 for all nonzero v.
12
 F IGURE 11. James Joseph Sylvester (1814–97) invented the terms ‘matrix’ and ‘dis-
criminant.’ Image: MacTutor History of Mathematics Archive

then F ∩ E− = 0. According to the forthcoming dimension formula (Theorem 4.6),

dim F + dim E− = dim(F + E− ) + dim(F ∩ E− ).
Since dim E− = n − q, dim(F ∩ E− ) = 0, and dim(F + E− ) ≤ n, we get that
dim F + n − q ≤ n
that is, dim F ≤ q. Hence
sup{dim F | F ≤ E positive definite} = q,
proving that q is independent of the basis chosen. Applying this to the space (E, −Q) shows that
p is independent of basis as well. 
Definition 4.4. With the notation of Theorem 4.3, we say that (E, Q) has Sylvester type (−p, q).
Corollary 4.5. Two quadratic forms of the same dimension with equal Sylvester type are isometric.
We now state and prove the dimension formula which was crucial in the previous proof.
Theorem 4.6 (Dimension formula). For U and V subspaces of a finite-dimensional vector space E, we
ahve
dim(U ∩ V ) + dim(U + V ) = dim U + dim V.
Proof idea. Apply rank-nullity to the linear map
f : U ⊕ V −→ E
(u, v) 7−→ u − v,
observing that ker f ∼
= U ∩ V and im f = U + V 

See [Ive92, pp.10–11] for a couple more cute facts about Sylvester types.
13
Euclidean vector spaces. A Euclidean vector space is a finite-dimensional space E equipped with a
positive definite quadratic
p form (one for which Q(v) > 0 when v 6= 0). In such a space, a vector v
has length |v| = hv, vi. Throughtout this section, E stands for a fixed Euclidean vector space.
Theorem 4.7 (Cauchy–Schwarz). For all v, w ∈ E,
| hv, wi | ≤ |v||w|
and the inequality is strict when v and w are linearly independent.
Proof. If w = λv for λ ∈ R, then | hv, wi | = |λ hv, vi | = |λ||v| 2
 = |v||w|. If v and w are linearly
hv, vi hv, wi
indpendent, they span a Euclidean plane with Gram matrix . You can read the
hw, vi hw, wi
proof of [Ive92, I.3.5] to see that the sign of the determinant of this matrix is the same as that of
any Gram matrix for a form with the same Sylvester type. In particular,
 
hv, vi hv, wi
det >0
hw, vi hw, wi
so
|v|2 |w|2 − hv, wi2 > 0
and the Cauchy-Schwarz inequality follows. 
Theorem 4.8 (Triangle inequality). For all v, w ∈ E, |v + w| ≤ |v| + |w| and the inequality is strict
when v and w are linearly indpendent.
Proof. By direct computation,
|v + w|2 = |v|2 + |w|2 + 2|v||w| + 2(hv, wi − |v||w|).
By Theorem 4.7, the term in parentheses is nonnegative (and positive when v, w are linearly inde-
pendent), and the triangle inequality follows. 
Definition 4.9. The angle ∠(v, w) between two nonzero vectors v, w in a Euclidean vector space E
is
hv, wi
∠(v, w) = arccos
|v||w|
where arccos is the inverse to cos |[0,π] .
We now turn to the orthogonal group O(E). If n ∈ E is a unit normal vector for a hyperlane
H ≤ E (so |n| = 1 and H ⊥ n), then
τn (x) = x − 2 hx, ni n
is an orthogonal tranformation of E reflecting across the hyperplane H.
Theorem 4.10. Suppose dim E = n. Then all σ ∈ O(E) can be expressed as the product of at most n
reflections through hyperplanes.
Proof. We proceed by induction on n = dim E. The case n = 1 is trivial. For the induction step,
pick a vector v ∈ E of unit length. If σ(v) = v, then σ takes v ⊥ to itself and we may use the
inductive hypothesis to write σ as a product of at most n − 1 reflections. If σ(v) 6= v, observe that
reflection τ along the vector σ(v) − v interchanges v and σ(v). As such, τ ◦ σ fixes the vector v,
whence we can write τ ◦ σ as a product of at most n − 1 reflections. Multiplying on the left by τ
expresses σ as a product of at most n reflections. 
Recall from Proposition 3.9 that an orthogonal transformation has determinant ±1. As such, the
determinant is a surjective homomorphism det : O(E) → {±1}. The kernel SO(E) = {σ ∈ O(E) |
det E = 1} is thus a normal subgroup of O(E) of index two called the special orthogonal group.
14
Excercise 4.11. The group SO(E) acts transitively on S(E) = {v ∈ E | |v| = 1} when dim E ≥ 2.
Let’s now work in the Euclidean plan by assuming dim E = 2. The elements of SO(E) are then
called rotations. For e1 , e2 an orthonormal basis of E and σ ∈ SO(E) we have
σ(e1 ) = cos θe1 + sin θe2
for some θ ∈ R. Since σ(e2 ) is orthogonal to σ(e1 ), we have that the matrix for σ with respect to
e1 , e2 is
   
cos θ − sin θ cos θ sin θ
or .
sin θ cos θ sin θ − cos θ
Evaluating determinants, wesee that the first
 matrix gives a rotation while the second has deter-
cos θ − sin θ
minant −1, so σ has matrix . By the addition formulæ for trig functions, we get
sin θ cos θ
an isomorphism SO(E) ∼ = R/2πZ ∼ = S 1 = {z ∈ C | |z| = 1}. In other words, SO(E) is a circle.
Now assume that dim E = 3 so we are working in Euclidean 3-space. In this context, we define
a rotation to be an orthogonal transformation σ of E that fixes a line L and such that σ|L⊥ is a
(two-dimensional) rotation.
Proposition 4.12 (Euler). If dim E = 3, then every element of SO(E) is a rotation. If σ ∈ O(E) has
det σ = −1 then σ = ρτ where ρ is a rotation with axis L and τ is reflection in the plane L⊥ .
Proof. The characteristic polynomial for σ is
χ(t) = det(t · id −σ).
Note that if λ ∈ R is a real eigenvalue of σ with eigenvector v, then the equation σ(v) = λv (along
with σ preserving length) implies λ = ±1.
By Proposition 3.9, det σ = ±1. We also have χ(0) = det(−σ) = ∓1. Suppose det σ = 1 so
χ(0) = −1. Since χ(t) has leading term t3 , we know that limt→∞ χ(t) = ∞. By continuity of χ and
the intermediate value theorem, χ has a root on the positive real axis. By the previous paragraph,
this root must be 1.
Now suppose det σ = −1 so χ(0) = 1. Since limt→−∞ χ(t) = −∞, we learn that χ has root −1.
In either case, σ has an eigenline L with eigenvalue ±1. We learn that σ acts on L⊥ with eigen-
value 1 (so that det σ = ±1), and this is equivalent to Euler’s proposition. 

The following lemma will allow us to inductively describe the form that elements of O(E) take
according to dim E.
Lemma 4.13. Let σ : E → E be any linear endomorphism of a finite-dimensional nontrivial real
vector space. There exists a linear subspace V ≤ E with dim V = 1 or 2 which is stable under σ.
Read [Ive92, Lemma 4.11] for a cute proof using complexification. Recall that every complex
linear transformation has an eigenvalue and hence a stable linear subspace. The reader should
interpret Lemma 4.13 as a version of this result for real vector spaces (with lines replaced by lines
or planes).
Theorem 4.14. For σ ∈ O(E) there exists a decompositiion of E into an orthogonal sum of lines and
planes stable under σ.
Proof. By Lemma 4.13, there is a line or plane V ≤ E stable under σ. For v ∈ V ⊥ and w ∈ V ,
choose w0 ∈ V such that σ(w0 ) = w. Then hσ(v), wi = hσ(v), σ(w0 )i = hv, wi = 0, so V ⊥ is stable
under σ as well. We may now proceed by induction on dim E. 
15
Parabolic forms. We now drop the definite from our positive definite forms, looking at real qua-
dratic forms Q such that Q ≥ 0 (but which may have Q(v) = 0 for nonzero v). There are called
positive forms.
Theorem 4.15 (Cauchy-Schwarz). If Q is positive with polarization hi, then
hv, wi2 ≤ hv, vi hw, wi .
Proof. The same as that of Theorem 4.7. 
Corollary 4.16. Let (E, Q) be a positive quadratic form. A vector v is isotropic (hv, vi = 0) if and
only if v ∈ E ⊥ (that is, hv, wi = 0 for all w ∈ E).
We now look at the ortogonal group of such forms.
Theorem 4.17. Let (E, Q) be a positive quadratic form with dim E = m. If σ ∈ O(E) and σ|E ⊥ = id,
then σ can be written as the product of at most m reflections.
Proof. Full details are in [Ive92, Theorem 5.3]. This is essentially a more delicate version of Theo-
rem 4.10. 
Definition 4.18. A positive quadratic form (E, Q) is called parabolic if dim E ⊥ = 1. In this case, E ⊥
is called the isotropic line.
Observe that E ⊥ is stable under any σ ∈ O(Q), and thus when Q is parabolic σ has E ⊥ as an
eigenspace. Define the multiplier function
µ : O(Q) −→ R×
σ 7−→ eigenvalue of σ on E ⊥ .
This is a group homomorphism and thus
O∞ (Q) := {σ ∈ O(Q) | µ(σ) > 0}
is an index 2 subgroup of O(Q). (Indeed, O∞ (Q) is the kernel of the composition of µ with R× →
R× /R>0 .)
The antipodal map a : E → E, x 7→ −x commutes with all elements of O(E) and µ(a) = −1.
This gives us a decomposition
O(E) ∼= O∞ (E) × Z/2Z.
[Iversen goes on for another fives pages — how much of this do we need right now?]

5. T HE L ORENTZ GROUP
As hyperbolic geometers we will be unduly obsessed with (n + 1)-dimensional real quadratic
forms of Sylvester type (−n, 1). The archetype of such a form is
n
X
x 7−→ x2n+1 − x2i
i=1
on Rn+1 , but we will find coordinate free descriptions useful as well. The hyperboloid or pseudo-
sphere for such a form (E, Q) is
S(E) := {v ∈ E | Q(v) = 1}.
In coordinates, this is the locus of the equation
n
X
x2i = x2n+1 − 1.
i=1
The following Cauchy–Schwarz type inequality holds on S(E).
16
Lemma 5.1. If (E, Q) is an (n + 1)-dimensional real quadratic form of Sylvester type (−n, 1) and
x, y ∈ S(E), then
1 ≤ | hx, yi |
and the inequality is strict when x and y are linearly independent in E.
Proof. If x, y are linearly dependent and both in S(E), then x = ±y and the inequality is trivial.
When x, y are lineaerly independent they span a plane P of Sylvester type (−1, 1), (−1, 0), or
(−2, 0) (see Lemma 5.2 below). Since P contains x and y which have norm 1, it must be the case
that the Sylvester type is (−1, 1). The ensuing Discriminant Lemma 5.2 implies that
hx, xi hy, yi < hu, xi2
from which the Cauchy-Scwartz inequality follows. 
Lemma 5.2 (Discriminant lemma). Let P be a plane in E spanned by x, y, where E is an (n + 1)-
dimensional real quadratic form of Sylvester type (−n, 1). Define the discriminant of P to be
∆ := hx, xi hy, yi − hx, yi2 .
Then the Sylvester type of P is 
(−1, 1) if ∆ < 0,

(−1, 0) if ∆ = 0,

(−2, 0) if ∆ > 0.


Proof. We begin by proving that P contains a vector of norm −1. Choose an orthonormal basis
for E and use it to create a hyperplane V of Sylvester type (−n, 0). By the dimnension formula
Theorem 4.6,
dim P + dim E = dim(P ∩ E) + dim(P + E).
The left-hand side is n + 2 and dim(P + E) ≤ n + 1, so dim(P ∩ E) ≥ 1. It follows that P contains
vectors with negative norm, and thus its Sylvester type is (−1, 1), (−1, 0), or (−2, 0). Again by
[Ive92, I.3.5] the determinant of a (real) Gram matrix for a form of Sylvester type (−p, q) has sign
(−1)p . This implies our discriminant test. 
We can apply these lemmata to prove that the hyperboloid S(E) always has exactly two con-
nected components. In the following proof, refer to Figure 12 to track what is going on. Challenge:
Use a computer to draw the same picture one dimension up (and then send said picture to me so
I can include it in these notes). Better yet, build an interactive demo that allows the user to vary
parameters (such as the choices of x and y in the proof below).
Proposition 5.3. The hyperboloid S(E) has exactly two connected components and x, y ∈ S(E)
are in the same connected component if and only if hx, yi > 0.
Proof. Fix y ∈ S(E) and let V = y ⊥ be the hyperplane orthogonal to x. Then V has Sylvester type
(−n, 0) and separates S(E) into two parts:
H = {x ∈ S(E) | hx, yi > 0} and H − = {x ∈ S(E) | hx, yi < 0}.
It remains to show that H and H − are connected. They are clearly homeomorphic via the an-
tipodal map, so it suffices to prove that H is connected. We do so by exhibiting a homeomorphism
with the open unit disk
D := {x ∈ V | − hx, xi < 1}.
The affine line through x ∈ H and −y is parameterized by (x + y)t − y, t ∈ R, and its intersection
g(x) with V occurs when h(x + y)t − y, yi = 0, so
x − hx, yi y
g(x) = .
1 + hx, yi
17
 3

-1

-2

-3
-3 -2 -1 0 1 2 3
F IGURE 12. This picture gives the content of the proof of Proposition 5.3 when
E = R2 with form x22 − x21 . The blue curves represent S(E), the black line segment
is D, and y has been taken to be (0, 1), so H is the upper portion of S(E). The red
line segment joins −y to the black dot on H, and the red dot represents the value
that g takes.

We leave it to the reader to check (by applying Lemma 5.1) that g(x) ∈ D. One may furthermore
check that g : H → D is a homeomorphism by exhibiting its continuous inverse. 

Definition 5.4. A Lorentz transformation (or orthochroneous Lorentz transformation) of E is any σ ∈

O(E) which preserves the connected components of S(E). The group of Lorentz transformations
is called the Lorentz group and is denoted Lor(E). Lorentz transformations with determinant 1 are
called even or special Lorentz transformations and the group of such maps is denoted Lor+ (E).

By Proposition 5.3, we know that σ ∈ O(E) is a Lorentz transformation if and only if

hx, σ(x)i > 0 for allx ∈ S(E).
Since the antipodal map is in the center of O(E) but is not a Lorentz transformation, we get that
O(E) = Lor(E) × Z/2Z.
Suppose that c ∈ E with hc, ci = −1. Then τc (x) = x + 2 hx, ci c. We claim that τc ∈ Lor(E).
Indeed, for x ∈ S(E),
hτc (x), xi = hx, xi + 2 hx, ci · hc, xi = 1 + 2 hx, ci2 > 0
so this follows from Proposition 5.3.
18
 F IGURE 13. Hendrik Lorentz (1853–1928) is the Dutch physicist after whom
Lorentz transformations are named. Pictured on the right here in 1921 with Al-
bert Einstein. Image: Wikimedia Commons.

We now study Lorentz transformations in detail when dim E = 2. Pick an orthonormal basis
e, f for E with he, ei = 1 and hf, f i = −1. Then a point xe+tf lies in S(E) if and only if x2 −t2 = 1.
The component of the hyperbola containing (1, 0) can be parametrized by s ∈ R 7→ (cosh s, sinh s).
Given σ ∈ Lor(E) we have σ(e) = e cosh s + f sinh s for some s ∈ R. We know that hσ(f ), σ(f )i =
−1 and σ(f ) ⊥ σ(e), so
σ(f ) = ∓(e sinh s + f cosh s).
In the first case, the matrix for σ is
 
cosh s − sinh s
sinh s − cosh s

with determinant −1 and trace 0, so σ has eigenvalues 1 and −1. It is in fact the case that σ is
reflection along a vector of norm −1 (check this!).
In the second case, the matrix for σ is
 
cosh s sinh s
L(s) = .
sinh s cosh s

Observe that
L(s)L(t) = L(s + t)
by the following exercise.
19
 F IGURE 14. The isotropic cone C(E) for E = R3 with form z 2 − x2 − y 2 is in gray.
The blue surface is S(E). The red circle at z = 1 can be identified with P C(E) since
each line through 0 in C(E) contains a unique point on the circle. In the proof of
Proposition 5.7, this circle is instead taken as the boundary of the unit disk in the
z = 0 plane.

Excercise 5.5. The hyperbolic trigonometric functions satisfy the following angle addition for-
mulæ (which may be checked by direct manipulations of exponential functions):
sinh(s + t) = sinh s cosh t + cosh s sinh t,
cosh(s + t) = cosh s cosh t + sinh s sinh t.
Since L(s)L(t) = L(s + t), we learn that R → Lor+ (E), s 7→ L(s) is a group isomorphism when
dim E = 2.
Theorem 5.6. Let E be an (n+1)-dimensional real vector space equipped with a quadratic form of Sylvester
type (−n, 1). Any Lorentz transformation σ of E is the product of at most n + 1 reflections of the type τc
with c ∈ E and hc, ci = −1.
Proof. See [Ive92, Theorem I.6.11]. 
For the remainder of this section, we will study the action of Lor(E) on the isotropic cone
C(E) = {v ∈ E | hv, vi = 0}
and its projectivization P C(E): The multiplicative group R× acts on C(E) r {−} by scalar mul-
tiplication, and the orbit space P C(E) = C(E)/R× is the projective cone of E. We can think of
P C(E) as the set of isotropic lines in E. See Figure 14 for explicit representations of C(E) and
P C(E) when dim E = 3.

20
Proposition 5.7. Let E be an (n+1)-dimensional real vector space equipped with a quadratic form
of Sylvester type (−n, 1). Then P C(E) is homeomorphic to the sphere S n−1 = {x ∈ Rn | |x| = 1}.
Proof. Returning to the terminology of the proof of Proposition 5.3, let S n−1 = ∂D be the boundary
of the unit disk in the hyperplane V perpendicular to some fixed y ∈ S(E). Define a function
φ : V −→ E
z 7−→ 2z + y − hz, zi y.
By direct calculation using hy, zi = 0, we get
hφ(z), φ(z)i = (1 + hz, zi)2 .
If z ∈ ∂D, then hz, zi = −1, whence hφ(z), φ(z)i = 0. As such, φ induces a map ∂D → C(E) which
we may compose with the projection C(E) → P C(E) to get a map Φ : ∂D → P C(E). The rest of
[Ive92, Proposition I.6.12] justifies why Φ is a homeomorphism. 
For the statement of the following corollary, recall that the action of a group G on a set X is
faithful if g · x = x for all x ∈ X implies that g = e. (That is, only the identity element acts trivially
on X.)
Corollary 5.8. Suppose dim E ≥ 3. Then
(a) the Lorentz group acts faithfully on P C(E), and
(b) the truncated isotropic cone C × (E) = C(E) r {0} has two connected components and these
are preserved by the Lor(E).
Proof. See [Ive92, Corollary I.6.14 and Proposition I.6.15]. 
Remark 5.9. In the near future, we will define hyperbolic space Hn to be one of the two sheets
of the hyperboloid S(E). The projective space of E is P (E) = (E r {0})/R× , the set of lines
(through the origin) in E. The projective cone P C(E) is naturally a subspace of P (E). Let P S(E)
denote the subset of P (E) consisting of lines in E spanned by vectors of strictly positive norm.
The ∂P S(E) = P C(E).
There is a natural homeomorphism Hn → P S(E) assigning x ∈ Hn to the line spanned by x.
Under this identification, it makes sense to write
∂Hn = P C(E).
Remark 5.10. At this point, Iversen continues his first chapter covering Möbius transformations,
inversive products of spheres, and the Riemann sphere. We are going to jump ahead to Chapter II
next, looking at metric geometry. We will make it all the way to section II.5 on the Klein disk and
then circle back to I.7 before studying the Poincaré disk and upper half space.

6. M ETRIC SPACES AND THEIR ISOMETRIES

Recall that a metric space is a set X equipped with a function d : X × X → R satisfying the
following properties:
» identity of indiscernibles: d(x, y) = 0 ⇐⇒ x = y,
» symmetry: d(x, y) = d(y, x), and
» triangle inequality: d(x, z) ≤ d(x, y) + d(y, z).
An isometry between metric spaces X and Y is a surjective distance-preserving function σ : X → Y .
Excercise 6.1. Prove that every isometry is injective and hence a bijection. Show that the inverse
to an isometry is also an isometry. (It follows that the set of self-isometries of a metric space X
forms a group Isom(X) under composition.)
21
 The set of real numbers R with the metric d(x, y) = |y − x| is a metric space. The following
proposition describes the isometries of R.
Proposition 6.2. An isometry σ : R → R is either a translation x 7→ x + a or a reflection x 7→ b − x.
Proof. Since 1 = d(σ(0), σ(1)), we know that σ(1) = σ(0) ± 1. In the + case take τ to be translation
by σ(0), and in the − case take τ to be the reflection x 7→ σ(0) − x. We are going to prove that
σ = τ . Suppose for contradiction that there is some c ∈ R such that σ(c) 6= τ (c). Note that
d(σ(c), σ(0)) = d(c, 0) = d(τ (c), τ (0)) = d(τ (c), σ(0))
so σ(0) is the midpoint of σ(c) and τ (c). Similarly σ(1) = τ (1) is the midpoint of σ(c) and τ (c), but
this contradicts σ(0) 6= σ(1). 
Definition 6.3. Let J ⊆ R be an interval and X a metric space. A curve γ : J → X is a geodesic
curve if for all c ∈ J there is an open neighborhood U ⊆ J of c such that γ|U is distance preserving.
Lemma 6.4. Let u be a point of the open interval J ⊆ R and let γ : J → R be a geodesic curve with
γ(u) = 0. Then γ(t) = ε(t − u) for ε = ±1.
Proof. By Proposition 6.2, γ is locally affine on some open U ⊆ J and thus γ 0 |U = ±1. Since J is
connected, γ 0 = ±1 on all of J. The rest is calculus. 

7. E UCLIDEAN SPACE
Our current goal is to understand pthe isometries of a Euclidean space E of dimension n. Here
the length of a vector v ∈ E is |v| = hv, vi and the distance between two points v, w ∈ E is
d(v, w) = |v − w|.
The Cauchy-Schwarz Theorem 4.7 implies that E is a metric space.
Proposition 7.1. Every geodesic curve γ : R → E has the form
γ(t) = et + v
where v is some point in E and e ∈ E has length 1.
Proof. It is easy to check that t 7→ et+v is a geodesic. For the converse, suppose that γ is a geodesic.
For a ∈ R, choose an open interval J ⊆ R containing a on which γ preserves distance. By the sharp
triangle inequality, any three points a, b, c ∈ J have γ(a), γ(b), γ(c) on an affine line. It follows that
γ(J) is contained in an affine line. By Proposition 6.2, we can find e of length 1 such that
γ(t) = e(t − a) + γ(a)
on J. As such, γ is a differentiable curve with locally constant derivative. Since R is connected, γ 0
is constant on R. The rest is calculus. 

We will now determine the group of isometries of a Euclidean space E, Isom(E). One example
is reflection τ in an affine hyperplane H ⊆ E. If n is a unit normal vector to H and u ∈ H, then
τ (x) = x − 2 hx − u, ni n.
Lemma 7.2. Suppose A1 , . . . , Ap and B1 , . . . , Bp are sequences of points of E such that
d(Ai , Aj ) = d(Bi , Bj ), 1 ≤ i, j ≤ p.
There there exists σ ∈ Isom(E) which is a product of at most p reflections such that σ(Ai ) = Bi .
22
Proof. We proceed by induction on p. If p = 1, reflect through the bisecting hyperplane
{x ∈ E | d(x, A1 ) = d(x, B1 )}.
For the induction step, let ρ be a product of at most p − 1 reflections taking A1 , . . . , Ap−1 to
B1 , . . . , Bp−1 . If ρ(Ap ) = Bp , we are done. If ρ(Ap ) 6= Bp , observe that
d(ρ(Ap ), Bi ) = d(ρ(Ap ), ρ(Ai )) = d(Ap , Ai ) = d(Bp , Bi ), 1 ≤ i ≤ p − 1.
This implies that B1 , . . . , Bp−1 lie on the hyperplane bisecting ρ(Ap ) and Bp . Take τ to be the
reflection through this hyperplane. Then σ = τ ρ does the job. 
Definition 7.3. A simplex in E ∼ = Rn is a sequence A0 , A1 , . . . , An ∈ E not contained in an affine
hyperplane.
Remark 7.4. It might be more standard to call the convex hull of such a collection of points an n-
simplex. In our terminology, 0 simplex is a point, the convex hull of a 1-simplex is a line segment,
the convex hull of a 2-simplex is a triangle, and the convex hull of a 3-simplex is a tetrahedron.
Lemma 7.5. Let A0 , . . . , An be a simplex of E. If α, β ∈ Isom(E) and
α(Ai ) = β(Ai ), 0 ≤ i ≤ n,
then α = β.
Proof. Set σ = β −1 α. It suffices to prove that σ = id, so assume for contradiction that there exists
P ∈ E such that σ(P ) 6= P . Since
d(σ(P ), Ai ) = d(σ(P ), σ(Ai )) = d(P, Ai ), 0 ≤ i ≤ n,
we learn that A0 , . . . , An all belong to the affine hyperplane bisecting P and σ(P ). This contradicts
the hypothesis that A0 , . . . , An is a simplex. 
Corollary 7.6. Let σ be an isometry of E that fixes an affine hyperplane H pointwise. Then either
σ is a reflection in H or σ = id.
Proof. Assume that σ 6= id and pick P ∈ E with σ(P ) 6= P . By the proof of the lemma, H is the
perpendicular bisector of σ(P ) and P . If τ is reflection in H, then τ σ fixes P and the points of H.
By the lemma, τ σ = id, whence σ = τ . 
Theorem 7.7. If E is a Euclidean space of dimension n and β ∈ Isom(E), then β can be expressed as the
product of at most n + 1 reflections.
Proof. Pick a simplex A0 , . . . , An in E. By Lemma 7.2, there exists an isometry σ which is the
composite of at most n + 1 reflections with σ(Ai ) = β(Ai ), 0 ≤ i ≤ n. By Lemma 7.5, σ = β. 
We are now ready to describe Isom(E) in terms of O(E) and translations. For a vector e ∈ E,
translation by e is the isometry te given by te (x) = e + x. The translations form a subgroup T (E) ≤
Isom(E) and T (E) ∼ = (E, +), the additive group underlying E.
Lemma 7.8. Any isometry ρ ∈ Isom(E) has a unique expression of the form
ρ = te σ, e ∈ E and σ ∈ O(E).
The orthogonal translation σ is called the linearization of ρ and is denoted ρ
~.
Proof. Let Isom0 (E) denote the isometries of E of the form te σ, e ∈ E, σ ∈ O(E). The reader may
check that
σte σ −1 = tσ(e) .
It follows that Isom0 (E) is closed under composition since
tf σte ρ = tf tσ(e) σρ = tv+σ(e) (σρ)
23
 π

1π
2

-1 1

F IGURE 15. The graph of arccos, the inverse of cos |[0,π] .

for f, e ∈ E and σ, ρ ∈ O(E). By looking at the formula for reflection in an affine hyperplane, we
see that all such transformations are in Isom0 (E). It now follows from Theorem 7.7 that Isom0 (E) =
Isom(E), as desired. 
Observe that linearization ρ 7→ ρ~ is a homomorphism Isom(E) → O(E), and ρ ~ = id if and only if
ρ is a translation. This implies that T (E) is the kernel of the linearization map. Since linearization
is also surjective, we learn that there is a short exact sequence
1 −→ T (E) −→ Isom(E) −→ O(E) −→ 1
and that O(E) ∼= Isom(E)/T (E).
Iversen has a bit more to say about isometries of Euclidean space on pp.63–64 which the inter-
ested student may pursue.

8. S PHERICAL GEOMETRY
We will briefly consider the geometry of spheres before moving on to hyperbolic space. Let E
be a Euclidean space of dimension n + 1 with unit sphere S n := S(E). By the Cauchy–Schwarz
Theorem 4.7,
hP, Qi ∈ [−1, 1]
for P, Q ∈ S n . As such, we can define the spherical distance between P, Q ∈ S n to be
d(P, Q) = arccos hP, Qi .
For the reader’s convenience, the graph of arccos is reproduced in Figure 15.
The function d is clearly symmetric. To check identity of indiscernibles, first suppose that
P, Q are linearly independent in E. Then the sharp version of Cauchy–Schwarz implies that
hP, Qi ∈ (−1, 1), so d(P, Q) 6= 0. If P, Q are linearly dependent but distinct, then P = −Q, so
hP, Qi = −1 and d(P, Q) = π. We will prove the triangle inequality after introducing some spher-
ical trigonometry.

24
Definition 8.1. A tangent vector to a point A ∈ S n is a vector T ∈ E with hT, Ai = 0. A tangent
vector of length 1 is called a unit tangent vector. The space of tangent vectors to A is
TA (S n ) := A⊥ .
Lemma 8.2. For A, B ∈ S n there exists a unit tangent vector U ∈ TA (S n ) such that
B = A cos d(A, B) + U sin d(A, B).
Proof. First suppose that A and B are linearly independent. Take U ∈ TA (S n ) ∩ span{A, B} of
length 1. Then B ∈ span{A, U }, so B = xA + yU for some x, y ∈ R. Since hA, U i = 0, we have
x2 + y 2 = 1, and hence there is an s ∈ [−π, π] such that
B = A cos s + U sin s.
Replacing (s, U ) with (−s, −U ) if necessary, we may assume that s ∈ [0, π]. Then
cos d(A, B) = hA, Bi = cos s,
so s = d(A, B), as desired.
If A and B are linearly dependent, then A = ±B, so sin d(A, B) = 0 and any unit tangent vector
U ∈ TA (S n ) will work. 

By a spherical triangle 4ABC we mean three points A, B, C ∈ S n which are linearly independent
in E. We define
a = d(B, C),
b = d(A, C),
c = d(A, B).
By Lemma 8.2, we may choose unit tangent vectors U, V ∈ TA (S n ) such that
B = A cos c + U sin c and C = A cos b + V sin b.
We then define ∠A = α = arccos hU, V i, the angle between U and V .
Proposition 8.3 (Spherical Law of Cosines). Using the above notation,
cos a = cos b cos c + sin b sin c cos α.
Proof. We present the proof when dim E = 3. This adapts to the general case by observing that
A, B, C sit on the sphere for their span, which is a 3-dimensional Euclidean space (at least when
A, B, C are linearly independent).
By definition, cos a = hB, Ci. Without loss of generality, we may take E = R3 and assume that
A = (0, 0, 1) and B is in the xz-plane making angle c with the z-axis, so B = (sin c, 0, cos c). If C
projects to N in the xy-plane, then the angle between N and the x-axis is α, and we conclude that
C = (sin b cos α, sin b sin α, cos b). We may then compute
hB, Ci = sin c sin b cos α + cos c cos b.
Equating this and the previous expression for hB, Ci gives the law of cosines. 

For hints on a coordinate-free proof, see [Ive92, II.3.5].

Theorem 8.4 (Spherical Triangle Inequality). Given A, B, C ∈ S n ,
d(A, B) ≤ d(A, C) + d(C, B)
and the inequality is sharp when A, B, C are linearly independent in E.
25
Proof. We prove the sharp inequality when A, B, C are linearly independent and leave the rest to
the reader. In this case, cos α < 1, Proposition 8.3 implies that
cos a < cos b cos c + sin b sin c = cos(c − b).
For c − b > 0, we deduce a > c − b, so c < a + b. If c − b < 0, then c < a + b is trivial since a > 0. 
The following description of geodesics in S n is what one would expect from the “great circle”
intuition for such objects. We send the reader to [Ive92, II.3.8] for a full proof.
Theorem 8.5. Any geodesic curve γ : R → S n is of the form
γ(t) = A cos t + T sin t
where A ∈ S n and T is a unit tangent vector to S n at A.
We will now describe Isom(S n ). It is of course that case that any σ ∈ O(E) induces such an
isometry, and we will show that all such isometries are of this form.
Lemma 8.6. Suppose A1 , . . . , Ap and B1 , . . . , Bp are sequences of points in S n such that d(Ai , Aj ) =
d(Bi , Bj ) for all 1 ≤ i, j ≤ p. Then there exists σ ∈ Isom(S n ) composed of at most p hyperplane
reflections with σ(Ai ) = Bi for 1 ≤ i ≤ p.
Proof. Given A, B ∈ S n , observe that the set of points equidistant from A and B is the inter-
section of S n with the linear hyperplane H orthogonal to A − B. An orthogonal reflection in
H will interchange A and B. The result follows from the argument given in the Euclidean case
(Lemma 7.2). 
Theorem 8.7. We have Isom(S n ) = O(E).
Proof. Suppose β ∈ Isom(E). Pick a spherical simplex in S n , that is, a sequence A0 , . . . , An of points
in S n linearly independent in E. By Lemma 8.6, there is an isometry σ ∈ O(E) which is the
product of at most n + 1 reflections in hyperplanes with σ(Ai ) = β(Ai ). Now proceed as in the
proof of Lemma 7.5. 
We conclude our diversions in spherical geometry by looking at triangles on S 2 . Throughout
this discussion, we will take it as given that the surface area of a sphere of radius R is 4πR2 , a
formula you should have derived in vector calculus. Consider a spherical triangle 4ABC with
interior angles α at A, β at B, and γ at C. We define the excess of this triangle to be E = α+β +γ −π.
Theorem 8.8 (Girard). The area of 4ABC is
area(4ABC) = R2 · E.
You can prove this theorem yourself via the following steps:
(1) Let PA be the lune6 built from 4ABC, the triangle across the BC line segment, and the lune
on the opposite side of the sphere. Similarly define PB and PC .
(2) Check that PA ∪PB ∪PC is the entire sphere and that the intersection of any two P ’s is 4ABC ∪
(a congruent triangle on the other side of the sphere).
(3) Via geometric inclusion-exclusion, deduce that
area(PA ) + area(PB ) + area(PC ) = 4πR2 + 4 area(4ABC).
(4) Argue that area(PA ) = 4αR2 , area(PB ) = 4βR2 , and area(PC ) = 4γR2 to deduce Girard’s
theorem.

6A lune is a region between two great circles. Note that every pair of great circles creates four lunes which come in
congruent pairs.
26
 F IGURE 16. Albert Girard (1596–1632) contemplating a sphere. Image: http://
mathshistory.st-andrews.ac.uk/Biographies/Girard_Albert.html.

We can use Girard’s theorem to effectively bound the sum of the interior angles of a spherical
triangle. Since the area is always positive, we learn that E > 0, and E approaches but never
achieves 0 as the triangle becomes very small. In order to place an upper bound on E, we need
to ask ourselves some existential questions about triangles: Take a tiny triangle 4ABC. Look at
the exterior of 4ABC on the sphere. Is this also a triangle with vertices A, B, C? If you think so,
then area is bounded by 4πR2 and we learn that E < 4π (again getting arbitrarily close but not
achieving this bound). It is perhaps more conventional to demand that our triangles are always
the smaller of the two regions defined by A, B, C. And if A, B, C all lie on a great circle (so that
4ABC is one of two halves of the sphere), we should probably view that case as degenerate. With
these conventions, we learn that E < 2π.
The following theorem summarizes the above discussion.
Theorem 8.9. Let 4ABC be a triangle on a sphere of radius R with interior angles α, β, γ. If 4ABC is
smaller than half the sphere, then
π < α + β + γ < 3π.
If 4ABC is larger than half the sphere, then
3π < α + β + γ < 5π.

9. H YPERBOLIC SPACE
Throughout this section, E is an (n+1)-dimensional real vector space equipped with a quadratic
form of Sylvester type (−n, 1). According to Proposition 5.3, the hyperboloid S(E) = {x ∈ E |
hx, xi = 1} has two “sheets,” that is, connected components homeomorphic to the unit disk in Rn .
Let Hn denote one of these sheets.
In order to put a metric on Hn , note that it follows from Lemma 5.1 and Proposition 5.3 that
hP, Qi ≥ 1 for all P, Q ∈ Hn . As such, we may define the hyperbolic distance between P and Q to be
d(P, Q) = arccosh hP, Qi .
(Recall that arccosh is the inverse to cosh |[0,∞) , and cosh is pictured in Figure 7.)
27
 Symmetry for d is obvious. To check identity of indiscernibles, suppose P 6= Q. Then P and Q
are linearly independent and thus hP, Qi > 1, so d(P, Q) > 0. We check the triangle inequality in
the next theorem.
Theorem 9.1 (Hyperbolic triangle inequality). For any A, B, C ∈ Hn ,
d(A, B) ≤ d(A, C) + d(C, B)
and the inequality is strict whenever A, B, C are linearly independent in E.
Proof. Set a = d(B, C), b = d(C, A), and c = d(A, B). Then the Gram matrix for A, B, C is
 
1 cosh a cosh b
∆ := det cosh a 1 cosh c .
cosh b cosh c 1
Expansion along the first row gives

∆ = (1 cosh2 c) − cosh a(cosh a − cosh b cosh c) + cosh b(cosh a cosh c − cosh b)

= 1 − cosh2 a − cosh2 b − cosh2 c + 2 cosh a cosh b cosh c
= (cosh2 b − 1)(cosh2 c − 1) − (cosh b cosh c − cosh a)2 − sinh2 b sinh2 c − (cosh b cosh c − cosh a)2
= (cosh a − cosh b cosh c + sinh b sinh c)(cosh b cosh c + sinh b sinh c − cosh a)
= (cosh a − cosh(b − c))(cosh(c + b) − cosh a)
1 1 1 1
= 4 sinh( (a + b + c)) sinh( (a + c − b)) sinh( (a + b − c)) sinh( (c + b − a)).
2 2 2 2
Setting p = 12 (a + B + c), we get
∆ = 4 sinh p sinh(p − a) sinh(p − b) sinh(p − c).

If A, B, C are linearly independent in E, then they span a subspace of type (−2, 1),7 so ∆ > 0.8
Without loss of generality, suppose c ≥ a, b. Then
1 1 1 1
p − a = (c − a) + b > 0 and p − b = (c − b) + a > 0.
2 2 2 2
Since ∆, p > 0, we conclude from the factorization of ∆ that p − c > 0, whence
a + b − c = 2(p − c) > 0,
which is the strict triangle indequality.
If A, B, C are linearly dependent, then ∆ = 0. We leave the rest to the reader. 
Definition 9.2. A tangent vector T to a point A ∈ Hn is a vector T ∈ E with hT, Ai = 0. If
hT, T i = −1, then we call T a unit tangent vector. The space of tangent vectors to Hn at A form a
hyperplane TA (Hn ) = A⊥ with Sylvester type (−n, 0).
Lemma 9.3. For all A, B ∈ Hn , there exists a unit tangent vector U ∈ TA (Hn ) such that
B = A cosh d(A, B) + U sinh d(A, B).

7Exercise: Check this!

8Indeed, the sign of the determinant of the Gram matrix of a real quadratic form of Sylvester type (−s, r) is (−1)s .
See [Ive92, I.3.5] for a proof.
28
Proof. If A = B then sinh d(A, B) = 0 and we can use any unit vector U ∈ TA (Hn ). If A 6= B, then
A and B are linearly independent and span a plane R of Sylvester type (−1, 1) by Lemma 5.2. (We
have ∆ < 0 because hA, Ai = hB, Bi = 1 and hA, Bi2 > 1 by Lemma 5.1.) As such, there is a unit
vector U ∈ R ∩ TA (H2 ) and we may write
B = Ax + U y
for some x, y ∈ R. Since hA, U i = 0, we get x2 − y 2 = 1, so there is some s such that
B = A cosh s + U sinh s.
Replacing (U, s) with (−U, −s) if necessary we may assume that s ≥ 0, whence
cosh d(A, B) = hA, Bi = cosh s
so s = d(A, B) as required. 
The lemma gives us a good guess as to what geodesics in Hn might be. Indeed, given A ∈ Hn
and a unit tangent vector U ∈ TA (Hn ), we can define a curve
γ : R −→ Hn
s 7−→ A cosh s + U sinh s.
We can then calculate the hyperbolic distance between γ(s) and γ(t) for s, t ∈ R: we have
cosh d(γ(t), γ(s)) = hγ(t), γ(s)i = cosh s cosh t − sinh s sinh t = cosh(t − s) = cosh |t − s|,
where the penultimate equality is a standard hyperbolic trig identity. It follow that d(γ(s), γ(t)) =
|t − s|, so γ is a geodesic curve.
Note that the image of γ is contained in the plane R spanned by A and U , and Lemma 9.3
implies that the image is exactly Hn ∩ R.
Proposition 9.4. Any geodesic curve γ : R → Hn is of the form
γ(t) = A cosh t + U sinh t
where A ∈ Hn and U is a unit tangent vector to Hn at A.
Proof. Suppose that γ is a geodesic and then fix a point u ∈ R and pick an open interval J ⊆ R con-
taining u and such that γ|J is distance preserving. By the sharp triangle inequality (Theorem 9.1),
for any three distinct points r, s, t ∈ J, the points γ(r), γ(s), γ(t) are linearly dependent. In partic-
ular, if we fix c, d ∈ J then γ(J) is contained in the plane R spanned by γ(c) and γ(d). Set A = γ(u)
and pick a unit vector U ∈ TA (Hn ) ∩ R. By Lemma 6.4 and the preceding discussion,
γ(t) = A cosh ε(t − u) + U sinh ε(t − u)
for ε = ±1 and t ∈ J. Thus γ is continuously differentiable on J with γ 0 (u) = εU . Since cosh is an
even function and sinh is an odd function, we immediately get
γ(t) = γ(u) cosh(t − u) + γ 0 (u) sinh(t − u)
for t ∈ J.
Now suppose σ is another geodesic agreeing with γ in a neighborhood of u. It suffices to show
that σ = γ on all of R. To this end, consider the set
S = {t ∈ R | σ(t) = γ(t), σ 0 (t) = γ 0 (t)}.
Continuity implies that S is closed. To see that S is open, suppose that t0 ∈ S. Then the above
discussion implies that both functions are equal to t 7→ γ(t0 ) cosh(t − t0 ) + γ 0 (t0 ) sinh(t − t0 ) in an
open interval containing t0 . Thus S is open, closed, and nonempty. Connectedness of R implies
that S = R, so σ = γ everywhere. 
29
 We now know the exact form of every geodesic curve γ : R → Hn , and we have in fact learned
that these geodesics are globally — not just locally — distance preserving. The interpretation of
the image of γ as Hn ∩ R for R = span{A, U } implies that there is a unique geodesic line through
any two distinct points of Hn .
We conclude this section by looking at the isometries of Hn . Recall that a Lorentz transformation
preserves the sheets of S(E) and thus induces an isometry of Hn . We will show that Isom(Hn ) ∼ =
Lor(E). To set terminology, we will call a reflection along a vector with negative norm a Lorentz
reflection.
Lemma 9.5. Let A1 , . . . , Ap , B1 , . . . , Bp ∈ Hn with
d(Ai , Aj ) = d(Bi , Bj ), 1 ≤ i, j ≤ p.
Then there exists σ ∈ Isom(Hn ) composed of at most p Lorentz reflections with σ(Ai ) = Bi .
Proof. Given A, B ∈ Hn let
{P ∈ Hn | d(A, P ) = d(B, P )}
denote their perpendicular bisector; this is necessarily the intersection of Hn and the linear hyper-
plane orthogonal to the vector N = A − B. By Lemma 5.1 and Proposition 5.3, we have
hN, N i = 2 − 2 hA, Bi < 0.
Reflection along N is a Lorentz reflection interchanging A and B. The proof now proceeds as that
for Lemma 7.2. 
Theorem 9.6. Every isometry of Hn is the restriction of a Lorentz transformation of the ambient space E,
so Isom(Hn ) ∼
= Lor(E).
Proof. Let A0 , . . . , An be a hyperbolic simplex, that is, n + 1 linearly independent points of E all in
Hn . By the previous lemma, there exists an isometry σ which is the composite of at most n + 1
Lorentz reflections such that σ(Ai ) = β(Ai ) for 0 ≤ i ≤ n. The proof now proceeds exactly as that
of Lemma 7.5. 

10. T HE K LEIN AND P OINCARÉ DISKS

We now begin the task of producing alternate models of hyperbolic space with desirable visual
and analytic properties. The first of these is the Klein disk.
Let Dn denote the open unit disk in an n-dimensional Euclidean space E. Then the space E ⊕ R
carries a natural quadratic form of Sylvester type (−n, 1), namely
h(A, a), (B, b)i = − hA, Bi + ab, A, B ∈ E, a, b ∈ R.
For Hn we take the “upper” sheet of the hyperboloid S(E ⊕ R) consisting of points (A, a) with
− hA, Ai + a2 = 1 and a > 0.
Consider the map p : Dn → Hn taking a point A to the point which is the intersection of Hn and
the line through (A, 1) and 0. This is pictured for n = 1 in Figure 17. The reader may check that a
formula for p is given by
(A, 1)
p(A) = p .
1 − hA, Ai

The Klein disk is Dn equipped with the metric given by transporting the metric for Hn along p,
that is, for A, B ∈ Dn
1 − hA, Bi
d(A, B) = arccosh p .
(1 − hA, Ai)(1 − hB, Bi)
30
 3

2.5

1.5

0.5

0
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

F IGURE 17. Left: A visualization of p : Dn → Hn when E = R. The disk Dn is

drawn at height 1, and the black dot maps to the red dot under p. Right: Felix
Klein (1849–1925), the namesake of the Klein disk. Responsible for the Erlangen
Program, which classifies geometries by their symmetry groups. Image: Emilio
Segrè Visual Archives / American Institute of Physics/Science Source.

Since geodesic lines in Hn are given by intersections Hn ∩ R, R a linear plane in E ⊕ R, we see that
geodesics in the Klein disk are Euclidean lines in Dn .
For reasons that we will explore later, the Klein disk is not the most desirable model of Hn .
A much more attractive model based on the disk is due to Poincaré. Here we again work with
Dn ⊆ E and the same Hn ⊆ E ⊕ R, but we parametrize Dn via sterographic projection with center
(0, −1). Note that the inverse of this map was previously discussed (under the name g) around
Figure 12. A formula for f is given by
f : Dn −→ Hn
(2P, 1 + hP, P i)
P 7−→ .
1 − hP, P i
Transporting the hyperbolic metric along f results in the Poincaré disk with
|P − Q|2
 
d(P, Q) = arccosh 1 + 2
(1 − |P |2 )(1 − |Q|2 )
for P, Q ∈ Dn .
In order to say more about the Poincaré disk, we will need to take a diversion into Möbius
transformations and the inversive product of spehres, which we undertake in the next section.

11. M ÖBIUS TRANSFORMATIONS

Let E be a Euclidean space of dimension n and and let S be a sphere in E with center c and
radius r > 0. We can define an involution σ : E r {c} → E r {c} by the formula
x−c
σ(x) = c + r2 .
|x − c|2
31
 F IGURE 18. Cover art for the Italian prog rock band Möbius Project’s 2014 album
Ra Me Nivar. August Ferdinand Möbius (1790–1868) was a German mathematician
and astronomer. In addition to Möbius transformations, he is also the namesake of
the Möbius band and Möbius µ-function.

This has the effect of fixing S pointwise and swapping the interior and exterior of S , so we call
σ inversion in the sphere S . Note that σ(x) lies on the ray originating at c that goes through x.

We can imagine σ sending the center c of S to a “point at infinity” and also set σ(∞) = c. This
allows us to extend σ to Ê, the one-point compactification of E. Consider E as E ⊕ 0 ⊆ E ⊕ R.
Stereographic projection identies S(E ⊕ R) with Ê, and we give Ê the corresponding topology.
The extension σ̂ of σ to Ê is clearly continuous.
Now consider an affine hyperplane H ⊆ E with unit normal vector n and containing a point
u ∈ H. Reflection across H is is given by

λ : E −→ E
x 7−→ x − 2 hx − u, ni n.

This naturally extends to λ̂ : Ê → Ê with λ̂(∞) = ∞.

By a sphere in Ê, we will mean either a Euclidean sphere S in E or H ∪ {∞} for H an affine
hyperplane in E. We will consider both σ̂ and λ̂ to be inversions in spheres in Ê.

Definition 11.1. The Möbius group Möb(E) is the subgroup of homeomorphisms of Ê generated
by inversions in spheres.

Our present goal is to identify Möb(E) with the Lorentz group of a Minkowski space. To this
end, consider the space E ⊕ R2 with form

1
h(e, a, b), (f, c, d)i = − he, f i + (ad + bc), e, f ∈ E, a, b, c, d ∈ R.
2
32
 F IGURE 19. This picture shows C(E ⊕ R2 ) in blue and the image of ι in red when
E = R with the standard inner product. Note that in this case, the isotropic cone is
cut out by the equation x2 = yz and ι is the curve given by t 7→ (t, t2 , 1). The fact
that the red curve does not make a loop around C(E ⊕ R2 ) represents the failure of
ι to go through the line spanned by (0, 1, 0).



1 0 1/2
The form on R2 given by h(a, b), (c, d)i = + bc) has Gram matrix
2 (ad which has
1/2 0
Sylvester type (−1, 1). As such, E ⊕ R2 with the above form as Sylvester type (−n − 1, 1) and is
thus a Minkowski space of dimension n + 2.
Recall that C(E ⊕ R2 ) denotes the isotropic cone of vectors with norm 0. We define a map

ι : E −→ C(E ⊕ R2 )
e 7−→ (e, he, ei , 1).

Indeed,
1
h(e, he, ei , 1), (e, he, ei , 1)i = − he, ei + (he, ei + he, ei) = 0,
2

so f (e) is in the isotropic cone. Composing the with natural projection C(E ⊕ R2 ) → P C(E ⊕ R2 )
to the projective isotropic cone (the set of lines through 0 in C(E ⊕ R2 )) gives a map ι : E →
P C(E ⊕ R2 ). The reader may check that this is an injection identifying E with the complement
of the point in P C(E ⊕ R2 ) corresponding to the line through 0 and (0, 1, 0). We can extend ι to a
bijection ι̂ : Ê → P C(E ⊕ R2 ) by defining ι̂(∞) to be this line. This is the identification necessary
to link Möb(E) with a Lorentz group.

The group Lor(E ⊕ R2 ) acts on E ⊕ R2 in a manner which takes lines to lines and preserves h , i,
and thus induces an action on P C(E ⊕ R2 ) = ∂Hn+1 . Given τ ∈ Lor(E ⊕ R2 ), let P C(τ ) denote
the corresponding map on P C(E ⊕ R2 ). Then τ̂ = ι̂−1 P C(τ )ι̂ is a map Ê → Ê, as evidenced by
33
the following diagram:
ι̂
Ê P C(E ⊕ R2 )
τ̂ P C(τ )

Ê P C(E ⊕ R2 ).
ι̂

Theorem 11.2. The assignment τ 7→ τ̂ takes Lorentz transformations of E ⊕R2 to Möbius transformations
of Ê and is an isomorphism of groups Lor(E ⊕ R2 ) ∼
= Möb(E).
Proof. The definition of τ 7→ τ̂ immediately implies that this map preserves composition and take
the identity to the identity. Since the Lorentz group is generated by Lorentz reflections and the
Möbius group is generated by inversions in spheres, it suffices to prove the following three state-
ments: (1) every inversion in a sphere is τ̂ for some τ ∈ Lor(E ⊕ R2 ), (2) if τ is a Lorentz reflection,
then τ̂ is an inversion in a sphere, and (3) distinct Lorentz transformations induce distinct Möbius
transformations.
To prove (1), begin by considering a Euclidean sphere S ⊆ E with center c and radius r > 0.
Define the vector C = (c, hc, ci − r2 , 1) ∈ E ⊕ R2 . The reader may check that
hC, Ci = −r2 and 2 hC, ιxi = |x − c|2 − r2
for x ∈ E. With these observations in hand, one can further check that
r2
ισ(x) = τC ι(x)
|x − c|2
where σ is inversion in S , τC is reflection along C, and x ∈ E r {c}. This means precisely that
τ̂C = σ̂.
Now consider reflection λ in an affine hyperplane H ⊆ E as before. Define the vector N =
(n, 2 hn, ui , 0) for n a unit normal to H and some fixed u ∈ H. The reader may similarly check that
τ̂N = λ̂, concluding our proof of (1).
For (2), observe that any vector with negative norm of the form (e, a, 1) ∈ E ⊕ R2 is of the form
C above for a judicious choice of S . Similarly, any vector with negative norm of the form (e, a, 0)
is of the form N for some affine hyperplane H.9 This is enough for (2) since τc = τλc for all nonzero
scalars λ, and every vector in E ⊕ R2 may be scaled to take the form (e, a, 1) or (e, a, 0).
Finally, (3) follows from Corollary 5.8, which says that the action of the Lorentz group on the
projective cone is faithful (if τ acts as the identity, then it is the identity). 
Corollary 11.3. A Möbius transformation of a Euclidean space of dimension n can be written as
the product of at most n + 2 inversions.
Proof. This follows from the above theorem and Theorem 5.6. 
We will now analyze the effect of Möbius transformations on disks. Let N = N (E ⊕ R2 ) denote
the vectors in E ⊕ R2 with norm −1. Given U ∈ N , the set
S = {x ∈ Ê | hιx, U i = 0}
is a sphere in Ê with complement D q D − where
D = {x ∈ Ê | hιx, U i > 0} and D − = {x ∈ Ê | hιx, U i < 0}.
(The reader is encouraged to check that when U = (e, a, b), the sphere S has center 1b e and radius
|1/b|.) We call D and D − the disks bounding S , and D is called the disk with normal vector U .
9Consult [Ive92, p.34] for completely explicit derivations of S and H.
34
 4

-2

-4
F IGURE 20. This diagram illustrates inversion in the black circle. The red and blue
circles are interchanged.

-4 -2 0 2 4
This allows us to deduce the following famous and important corollary. An illustration may be
found in Figure 20.
Corollary 11.4. A Möbius transformation ν ∈ Möb(E) takes a sphere S in Ê to a sphere ν(S ).
Inversion in the sphere ν(S ) is given by νσν −1 where σ is inversion in S .

Proof. Write ν = ινι−1 for the image of ν ∈ Möb(E) under the inverse of the isomorphism Lor(E ⊕
R2 ) ∼
= Möb(E) given by Theorem 11.2. Let U be the normal vector for one of the disks D bounding
S . We claim that ν(D) is the disk with normal vector ν(U ) (which of course implies that ν(S ) is
a sphere). Indeed,
ν(D) = {ν(x) | hιx, U i > 0}
= {ν(x) | hνιx, νU i > 0} (since ν is an orthogonal transformation)
= {ν(x) | hινx, νU i > 0} (definition of ν)
= {y | hιy, νU i > 0},
and the final set is the disk with normal vector νU .
For the second part of the corollary, observe that ν = τU . Since
τν(U ) = ντU ν −1
(check this!) we learn that νσν −1 is reflection in the sphere ν(S ). 
Corollary 11.5. The Möbius group Möb(E) acts transitively on the set of disks in Ê.
Proof. By parametrizing the disks by their normal vectors in N = N (E ⊕ R2 ), this reduces to
checking that Lor(E ⊕ R2 ) acts transitively on N . To see this, note O(E ⊕ R2 ) acts transitively on
N = SQ (−1) by Proposition 3.11. The reader may now check that the stabilizer of any U ∈ N
intersects the Lorentz group trivially, so Lor(E ⊕ R2 ) acts transitively as well. 
Corollary 11.6. Let S be a sphere in Ê with dim E ≥ 2. A Möbius transformation which fixes S
pointwise is either the identity or inversion in S .
Proof. See [Ive92, I.7.12]. 
Definition 11.7. We say that two spheres S and T are orthogonal if the normal vectors for the
bounding disks of S and T are orthogonal.
35
Proposition 11.8. Let σ denote inversion in the sphere S and let τ denote inversion in the sphere
T . If S 6= T , then the following conditions are equivalent:
(a) S and T are orthogonal,
(b) στ = τ σ,
(c) τ (S ) = S ,
(d) σ(T ) = T .
Proof. It follows from Corollary 11.4 that the last three conditions are equivalent. Let H and K be
normal vectors for S and T , respectively. The isomorphism Möb(E) ∼ = Lor(E ⊕ R2 ) takes τ to
τK , Lorentz reflection along K. By the formula
τK (H) = H + 2 hH, Ki K
we learn that τK (H) = ±H if and only if hH, Ki = 0. This shows that (a) and (c) are equivalent. 
Definition 11.9. Say that points A and B of Ê are conjugated with respect to a sphere S if A 6= B
and inversion in S interchanges A and B.
Proposition 11.10. If A, B ∈ Ê are conjugated with respect to a sphere S , then any sphere con-
taining both A and B is orthogonal to S .
Proof. Let a = ιA and let n denote a normal vector for S . The vector b = a + 2 hn, ai n generates
/ S , hn, ai =
the isotropic line represented by B (that is, b ∈ ι̂B). Since A ∈ 6 0 and n ∈ span{a, b}.
The normal vector for any sphere through A is perpendicular to a and the normal vector for any
sphere through B is perpendicula to b. Thus the normal vector for any sphere through A and B is
perpendicular to a and b, and thus perpendicular to n (a linear combination of a and b). 
Proposition 11.11. Let D ⊆ Ê be an open disk. The group Möb(D) of Möbius transformations of
E which take D to D is generated by inversions in spheres orthogonal to S = ∂D.
Proof. Let N denote the normal vector for D. A Möbius transformation σ leaves D invariant if and
only if the corresponding Lorentz transformation σ fixes N . Thus we may identify Möb(D) with
Lor(N ⊥ ). This group is generated by reflections τK where hK, Ki = −1 and hK, N i = 0. 
The following corollary is immediate.
Corollary 11.12. Let H denote a linear hyperplane in E and let D be one of the half-spaces in E
bounded by H. Restriction from Ê to Ĥ defines an isomorphism Möb(D) ∼= Möb(H).
The composite of the inverse of this isomorphism with the inclusion Möb(D) ,→ Möb(E) is
known as Poincaré extension Möb(H) → Möb(E).

12. T HE P OINCARÉ DISK AND HALF SPACE

Recall that the Poincaré disk is the open unit disk Dn in an n-dimensional Euclidean space E
equipped with the metric
|P − Q|2
 
d(P, Q) = arccosh 1 + 2 .
(1 − |P |2 )(1 − |Q|2 )
This metric was transported to Dn from Hn ⊆ E ⊕ R along the map
f : Dn −→ Hn
(2P, 1 + hP, P i)
P 7−→
1 − hP, P i
which is stereographic projection from the point (0, −1).
36
 F IGURE 21. An icosahedron in the Poincaré disk D3 . Image: http://home.
sandiego.edu/~dhoffoss/rusmp/index.html.

Theorem 12.1. The group Möb(Dn ) of Möbius transformations of E that take Dn to Dn acts as the group
of isometries of the Poincaré disk Dn . In particular, any isometry of Dn extends to a Möbius transformation
of E.
Proof. First note that the inverse to f is f −1 : Hn → Dn given by the formula f −1 (A, a) = 1+a
A
. Let
τ denote reflection along a vector (N, n) ∈ E ⊕ R of norm −1. The reader may verify that the
corresponding transformation of Dn is f −1 τ f given by
P + (n + n hP, P i − 2 hP, N i)N
f −1 τ f (P ) = .
|N − nP |2
When n = 0, we have hN, N i = 1 and f −1 τ f is Euclidean reflection along N . When n 6= 0, we
have 1 + n2 = hN, N i and (by a harder check by the reader) f −1 τ f is inversion in the Euclidean
sphere N with center N/n and radius 1/n. Since (1/n)2 + 1 = hN/n, N/ni, N is orthogonal to
S n−1 = ∂Dn . Combining this observation with Theorem 9.6, Theorem 5.6, and Proposition 11.11,
we are done. 
Proposition 12.2. Every isometry of the Poincaré disk Dn ⊆ E can be written as ρµ where µ ∈
O(E) and ρ is a reflection in a sphere orthogonal to S n−1 = ∂Dn .
The proof given in [Ive92, II.6.5] depends crucially on inversive products of spheres, content we
have decided to skip but which may be read in [Ive92, I.8].
The geodesics in the Poincaré disk are arcs of circles which are perpendicular to S n−1 = ∂Dn at
both intersection points. (Here we include lines as degenerate circles.) This can proven by passing
a geodesic in Hn through f −1 : Hn → Dn .

The Poincaré half space is constructed as follows. Begin with a Euclidean vector space L of di-
mension n − 1 and form the Euclidean space E = L ⊕ R. The upper +
√ half space E is the set
of points (p, h) ∈ E with h > 0. Inversion σ in a sphere of radius 2 with center S = (0, −1)
maps E + onto the unit disk Dn . Furthermore, σ induces stereographic projection of ∂Dn onto
L̂ = ∂E + . (The reader should pause and convince themselves of the previous two statements.
37
 F IGURE 22. A screenshot from http://roice3.org/h3/isometries/ which
contains GIF animations of isometries of E + for dim E = 3.

Remember that inversions take spheres to spheres, and note that σ(S) = ∞ while σ pointwise
fixes the intersection of its sphere with E + .) Explicitly,
σ : E + −→ Dn
(p, 1 + h)
(p, h) 7−→ (0, −1) + 2 .
|p|2+ (1 + h)2
We can then use σ to transport the metric of Dn onto E + , resulting in
|P − Q|2
 
d(P, Q) = arccosh 1 + .
2hk
The upper half space E + with the metric d is called Poincaré half space.
Theorem 12.3. The group Möb(E + ) of Möbius transformations of E taking E + to E + acts as the group
of isometries of Poincaré half space. In particular, any isometry of E + can be extended to a Möbius trans-
formation of E.
Proof. Conjugation by σ on Möb(E) takes the subgroup Möb(Dn ) to Möb(E + ). This theorem then
follows from Theorem 12.1. 

Corollary 12.4. The group of isometries of the Poincaré half space E + is isomorphic to the full
Möbius group of L = ∂E + .
Proof. This follows from the previous theorem and Proposition 11.11. 

We note that the geodesics in E + are traced by circles in Ê orthogonal to ∂E + .

At this point, the reader might enjoy exploring the following virtual reality implementations of
hyperbolic 3-space:
http://h3.hypernom.com/
http://www.michaelwoodard.net/hypVR-Ray/
38
 F IGURE 23. A screenshot of h3.hypernom.com.

They are best explored with a VR headset, but should open in any web browser and interact
meaningfully with a smart phone (tip: lock screen rotation).

13. T HE R IEMANN SPHERE

We need one more auxiliary object before solely focusing on two-dimensional hyperbolic geom-
etry: the Riemann sphere. The Riemann sphere Ĉ is the one-point compactification of the complex
plane. One way to achieve this construction is via projective geometry, as the set of complex lines
in C2 . This in turn may be realized as the quotient of C2 r {0} by the action of C× by scalar mul-
tiplication. (The orbit of a point (z, w) is {(λz, λw) | λ ∈ C× }, the complex line spanned by(z, w)

z
minus (0, 0).) We will denote the equivalence class of a point (z, w) 6= (0, 0) by [z : w] or .
w
This model for Ĉ is sometimes denoted CP 1 , the complex projective line. A point z ∈ C = Ĉ r {∞}
corresponds to [z : 1], while ∞ corresponds to [1 : 0].
The group GL2 (C) acts on Ĉ via the formula
    
a b z az + bw
=
c d w cz + dw
where a, b, c, d ∈ C, ad − bc 6= 0. Note that
      
λ 0 z λz z
= = ,
0 λ w λw w
so scalar matrices in GL2 (C) acts as the identity. As such, we get an action of
PGL2 (C) = GL2 (C)/C×
on Ĉ. (Here we are identifying C× with the normal subgroup of nonzero scalar matrices in
GL2 (C).)
39
 Identifying C with points of the form [z : 1] and ∞ with [1 : 0], we see that action of PGL2 (C)
on Ĉ can also be written as
az + b a
σ(z) = , σ(−d/c) = ∞, σ(∞) =
cz + d c
 
a b
for −d/c 6= z ∈ C and σ represented by .
c d
Note that for A ∈ GL2 (C), the inverse to A is A−1 = det1 A A∨ where A∨ is the cofactor matrix
given by
 ∨  
a b d −b
= .
c d −c a
This permits the easy computation of the action of A−1 on Ĉ, especially since we may neglect the
determinant factor.
Recall that a group action of G on a set X is called transitive if for every x, y ∈ X there exists
g ∈ G such that g · x = y. A more stringent property is the notion of k-fold transitivity: for all k-
tuples of distinct points x1 , . . . , xk ∈ X and y1 , . . . , yk ∈ X, there exists g ∈ G such that g · xi = yi ,
1 ≤ i ≤ k. In the case of PGL2 (C) acting on Ĉ, we actually have a 3-fold (or triply) transitive action.
The following proposition says this and slightly more.
Proposition 13.1. For any triples A, B, C and P, Q, R of distinct points in Ĉ, there is a unique
σ ∈ PGL2 (C) transforming the first triple into the second triple.
Proof. Suppose that A, B, C ∈ Ĉ are represented by a, b, c ∈ C2 . Since a and b are linearly inde-
pendent, there exist λ, µ ∈ C such that c = λa + µb. Since A = [λa] and B = [µb], we may assume
without loss generality that c = a + b. Similarly, we may represent P, Q, R ∈ Ĉ by p, q, p + q ∈ C2 .
Observe that the linear transformation σ : C2 → C2 such that σ(a) = p and σ(b) = q satisfies
σ(a + b) = p + q. Descending to Ĉ, we see that the class represented by σ in PGL2 (C) has the
desired properties.
Suppose now that τ ∈ GL2 (C) acts as the identity on Ĉ. Then all vectors in C2 r {0} are
eigenvectors of τ , whence τ must be a scalar matrix. We conclude that PGL2 (C) acts faithfully on
Ĉ, and this implies the uniqueness of σ above. 

We have also intensively studied the group of Möbius transformations Möb(C) which acts on Ĉ.
Call a Möbius transformation even if it can be written as the product of an even number of sphere
inversions. We write Möb+ (E) for the subgroup of even Möbius transformations in Möb(E).
Theorem 13.2. The group GL2 (C) acts on Ĉ as even Möbius transformations, inducing an isomorphism
PGL2 (C) ∼ = Möb+ (C).
Proof. We leave this as a reading exercise. See [Ive92, I.9.5], which depends on [Ive92, I.7.7]. 

We now introduce the cross ratio, an important invariant of four points in Ĉ which in some
sense measures the failure of quadruple transitivity for the action of PGL2 (C) on Ĉ.
Definition 13.3. For points P, Q, R, S ∈ Ĉ, at least three of which are distinct, let p, q, r, s be 2 × 1
complex matrices representing these points. The cross ratio of P, Q, R, S is
[P, Q; R, S] = [det(p, r) det(q, s) : det(p, s) det(q, r)] ∈ Ĉ.
In a homework problem, you will verify that this value is well-defined and independent of the
choice of p, q, r, s.
40
 If σ ∈ GL2 (C), then
det(σ(p), σ(q)) = det(σ) det(p, q).
From this, it follows that
[σ(P ), σ(Q); σ(R), σ(S)] = [P, Q; R, S].
In other words, the cross ratio is constant on PGL2 (C)-orbits.
Since [P, Q, R, S] is well-defined when at least three of the inputs are distinct, we may fix distinct
P, Q, R ∈ Ĉ and then consider the function Ĉ → Ĉ, S 7→ [P, Q, R, S] in which S varies and we take
the cross ratio.
Proposition 13.4. Let P, Q, R ∈ Ĉ be distinct. Then the transformation
Ĉ −→ Ĉ
S 7−→ [P, Q; R, S]
is the action of an element of PGL2 (C) taking P, Q, R to ∞, 0, 1, respectively.
Proof. The reader may check that
  
−q2 det(p, r) q1 det(p, r) s1
[P, Q; R, S] =
−p2 det(q, r) p1 det(q, r) s2
where the 2 × 2 matrix has determinant det(p, r) det(q, r) det(p, q) 6= 0. The fact that the map takes
P, Q, R to ∞, 0, 1 is a direct calculation. 
Proposition 13.5. Two 4-tuples of distinct points P, Q, R, S ∈ Ĉ and X, Y, Z, W ∈ Ĉ are in the
same PGL2 (C)-orbit if and only if
[P, Q; R, S] = [X, Y ; Z, W ].
Proof. By Proposition 13.4, we know that the condition is necessary. Conversely, if the two cross
ratios are equal, take σ, τ ∈ PGL2 (C) such that σ(P, Q, R) = (∞, 0, 1) and τ (X, Y, Z) = (∞, 0, 1).
Then
[P, Q; R, S] = [∞, 0; 1, σ(S)] = σ(S)
where the final equality holds because [∞, 0; 1, −] acts as an element of PGL2 (C) and takes ∞, 0, 1
to ∞, 0, 1 (and is thus the identity by Proposition 13.1). Similarly, [X, Y ; Z, W ] = τ (W ). Thus τ −1 σ
takes S to W as required. 
Lemma 13.6. Whenever they are defined, the following two identities hold:
[P, Q; R, S] = [R, S; P, Q] and [P, Q; R, S] = [Q, P ; S, R].
Proof. Computation. 
Proposition 13.7. Every element of PGL2 (C) is conjugate10 to one of the form
   
a 0 1 1
or
0 1 0 1
for a ∈ C× .
Proof. Let σ ∈ PGL2 (C) be represented by a matrix S ∈ GL2 (C). A nonzero vector (v, w) ∈ C2 is
an eigenvector for S if and only if σ[v : w] = [v : w]. By Proposition 13.1, we conclude that for
σ 6= id, σ has either one or two fixed points in Ĉ.

10Elements g and h of a group G are conjugate when there exists k ∈ G such that k −1 gk = h.
41
 First suppose that σ has two fixed points, A, B ∈ Ĉ. Let τ ∈ PGL  2 (C) satisfy τ (∞) = A,
a b
τ (0) = B. Then τ −1 στ has fixed points ∞ and 0. Recalling that z = az+b
cz+d , we see that any
c d
representing matrix for τ −1 στ is diagonal. Multiplying by a complex number, we get a represent-
ing matrix of the first type.
Now suppose that σ has only one fixed, A ∈ Ĉ. Choose an auxiliary point C 6= A and let τ be
the element of PGL2 (C) satisfying τ (∞) = A, τ (0) = C, τ (1) = σ(C). Then
 τ −1 στ fixes ∞ and
a 1
takes 0 to 1. As such, it can be represented by a matrix of the form for some a ∈ C× . This
0 1
matrix can only have one eigenvalue, so we must have a = 1. 

Given a matrix S ∈ GL2 (C), we define

(tr S)2
tr2 S = .
det S
For λ ∈ C× , we compute
(tr λS)2 λ2 (tr S)2
tr2 λS = = 2 = tr2 S
det λS λ det S
so tr2 descends to an invariant on PGL2 (C). It is similarly straightforward to check that tr2 is
constant on conjugacy classes in PGL2 (C).

Proposition 13.8. Let σ, τ ∈ PGL2 (C) r {id}. Then σ and τ are in the same conjugacy class in
PGL2 (C) if and only if tr2 σ = tr2 τ .

Proof. The preceding discussion exhibits the necessity of the condition on tr2 . For the converse,
note that
   
2 1 1 2 a 0
tr =4 and tr = 2 + a + a−1
0 1 0 1
for a ∈ C× . Since we are assuming a 6= 1, we see that these values
 are distinct,
 −1 and  2+a+a =
−1

a 0 a 0
2 + b + b−1 if and only if b = a±1 . It remains to show that and are conjugate,
0 1 0 1
which follows from
     
0 −1 a 0 0 1 1 0
=
1 0 0 1 −1 0 0 a
 −1   
a 0 1 0
and = a−1 . 
0 1 0 a

14. T HE P OINCARÉ HALF PLANE

For the rest of the course, we will focus on hyperbolic geometry in dimension 2. We will do
much of this work in the sl2 R model (introduced in the next section), but we pause here to develop
the upper half plane model. Let E denote the standard Euclidean space R2 and write H 2 = E +
for the half-space with second coordinate positive, and identify this space with complex numbers
with positive imaginary part. This carries the hyperbolic metric

|w − z|2
 
d(z, w) = arccosh 1 +
2 Im(z) Im(w)
42
for z, w ∈ C. This is equivalent to the following formulæ as the reader may verify:
1 |w − z|
sinh d(z, w) = p
2 2 Im(z) Im(w)
1 |w − z|
cosh d(z, w) = p
2 2 Im(z) Im(w)
1 |w − z|
tanh d(z, w) = .
2 |w − z|
Recall that GL2 (R) acts on C r R via
 
a b az + b
z=
c d cz + d
and this has imaginary part
 
Im z a b
det .
|cz + d|2 c d
As such we can define an action of GL2 (R) on H 2 via the formula
(
  az+b
a b cz+d if ad − bc > 0,
z = az+b
c d if ad − bc < 0.
cz+d

Proposition 14.1. The above action identifies PGL2 (R) with Isom(H 2 ).
Proof. By Möbius extension, this boils down to PGL2 (R) being the Möbius group of R ⊆ C. We
leave the details to the reader. 
Proposition 14.2. The geodesics in H 2 are traced by (generalized) circles in Ĉ perpendicular to the
real axis.
Proof. Note that
1 e2t + e2s (et − es )2
cosh(s − t) = (es−t + et−s ) = = 1 + = cosh d(ies , iet )
2 2et es 2et es
for s, t ∈ R. This implies that R → H 2 , s 7→ ies is a geodesic in H 2 , so the positive imaginary
axis is a geodesic line. The group PGL2 (R) acts transitively on circles in Ĉ perpendicular to R
(because PGL2 (R) acts transitvely on R̂). The group PGL2 (R) is transitive on geodesics in H 2 as
well, implying the proposition. 

15. T HE ACTION OF THE SPECIAL LINEAR GROUP ON THE UPPER HALF PLANE
In Proposition 14.1, we saw that PGL2 (R) was the group of isometries of the upper half-plane
 
2 × λ 0
H . Every element of GL2 (R) has determinant in R , and the action of a scalar matrix
0 λ
2
changes this determinant by λ . As such, every element of PGL2 (R) has a representing matrix with
determinant ±1. If the determinant is +1, the associated transformation is orientation-preserving,
and otherwise it is orientation-reversing. As such, the group PSL2 (R) = SL2 (R)/{±I} acts as the
group of orientation-preserving isometries of H 2 .
Our present goal is to understand some qualitative features of how elements of this group act
on H 2 . In order to ease notation and nomenclature, we will just discuss the action of SL2 (R), but
the careful reader should remember that −A acts in the same way as A on H 2 .
43
 We can understand how SL2 (R) acts on H 2 by focusing on the fixed points of each transforma-
tion. Indeed, for σ ∈ SL2 (R), σ(z) = z if and only if az + b = (cz + d)z, i.e.,
(15.1) cz 2 + (d − a)z − b = 0.
By the quadratic formula, this has roots
p
a−d± (d − a)2 + 4bc
.
2c
The discriminant (d − a)2 + 4bc controls the number of real roots and can be expanded as
d2 − 2da + a2 + 4(ad − 1) = (a + d)2 − 4 = tr2 σ − 4.
15.1. Hyperbolic transformations. When σ ∈ SL2 (R) has tr2 σ − 4 > 0, (15.1) has two real roots
and we call σ a hyperbolic transformation. This means that the action of σ on H 2 has no fixed points,
but when the action is extended to H 2 ∪ ∂H 2 there are two distinct fixed points A, B on R = ∂H 2 .
The geodesic ` in H 2 joining A and B is stable under σ, and in fact every (Euclidean) circle through
A and B itself by σ. These circles (called hypercycles for `) are the orbits of the subgroup of SL2 (R)
consisting of hyerbolic transformations with fixed points A and B.
15.2. Parabolic transformations. When σ ∈ SL2 (R) has tr2 σ − 4 = 0, (15.1) has a single real root
and we call σ a parabolic transformation. In this case, σ has a single fixed point P on R = ∂H 2 and
any circle in H 2 ∪ ∂H 2 tangent to R at the point P is stable under the action of σ. Such a circle is a
called a horocycle with center P , and they form the orbits of the subgroup of SL2 (R) consisting of
parabolic tranformations with fixed point P .
15.3. Elliptic transformations. When σ ∈ SL2 (R) has tr2 σ − 4 < 0, (15.1) has two conjugate
complex roots, only one of which is in H 2 . In this case, we call σ an elliptic transformation and it has
a unique fixed point A in H 2 (and none in ∂H 2 ). The (hyperbolic) circles centered at A are stable
under σ. Hyperbolic circles with center A are the orbits of the subgroup of SL2 (R) consisting of
elliptic transformations with fixed point A.
15.4. Nomenclature. Why are we calling these transformations hyerbolic, parabolic, and elliptic?
The terminology is based on the shape of a plane conic ax2 + bxy + cy 2 = 1, which depends on
its discriminant d = b2 − 4ac; it is a hyperbola when d > 0, a parabola when d = 0, and an ellipse
when d < 0.
15.5. Conjugacy classes. The conjugacy classes in SL2 (R) are also determined by the sign of
tr2 σ − 4. Indeed, we have the following theorem.
 
2 λ 0
Theorem 15.2. Let σ ∈ SL2 (R). If tr σ > 4, then σ is conjugate to a unique matrix of the form
0 1/λ
   
1 1 −1 1
with |λ| > 1. If tr2 σ = 4, then σ is conjugate to exactly one of ±I, ± , or ± . If
0 1 0 −1
 
cos θ − sin θ
tr2 σ < 4, then σ is conjugate to a unique matrix of the form other than ±I.
sin θ cos θ
A proof based on the action of SL2 (R) on R2 can be found in [Con, §3]. Note, though, that an
alternate proof can be established via the action of SL2 (R) on H 2 . Indeed, for σ, τ ∈ SL2 (R) where
σ has (potentially identical) fixed points A1 , A2 ∈ H 2 ∪∂H 2 , τ στ −1 has fixed points τ (Ai ). This and
our above analysis shows that the sign of tr2 σ − 4 is preserved by conjugation. To show that any
two transformations of the same type (hyperbolic, parabolic, or elliptic) are in the same conjugacy
class, one may use Proposition 13.1. The precise conjugacy class representatives may be deduced
by specifying “nice” fixed points. For example, the rotation matrices correspond to fixed point i.
44
15.6. Construction and visualization. There are some distinguished families of matrices of each
type. Indeed, for t ∈ R,
 t 
e 0
is hyperbolic with fixed points 0, ∞,
0 e−t
 
1 t
is parabolic with fixed point ∞, and
0 1
 
cos t − sin t
is elliptic with fixed point i.
sin t cos t
Given (potentially identical) points A1 , A2 ∈ H 2 ∪ ∂H 2 , we can conjugate the above matrices by
a matrix C ∈ SL2 (R) taking their fixed points to A1 , A2 . This results in a family of matrices in
SL2 (R) with the desired fixed points, and we can use this to produce some nice pictures of how
elements of SL2 (R) act on H 2 ∪ ∂H 2 . The reader can find animations produced in this fashion at
http://people.reed.edu/~ormsbyk/341/SL2R.html.
15.7. Iwasawa decomposition. We are actually tantalizingly close to deriving the Iwasawa de-
composition of SL2 (R). We send the reader to [Con, Appendix A] for details. (This might make a
good final project!)
Consider the following subgroups of SL2 (R):
  
cos t − sin t 
K= t∈R ,
sin t cos t 
  
r 0 
A= r > 0 , and
0 1/r 
  
1 x 
N= x∈R .
0 1 
Theorem 15.3. We have a decomposition of SL2 (R) as SL2 (R) = KAN , meaning that every g ∈ SL2 (R)
has a unique representation as g = kan with k ∈ K, a ∈ A, and n ∈ N .
As topological spaces, K ∼ = S1, A ∼
= R>0 ∼= R, and N ∼= R. The function f : K ×A×N → SL2 (R)
given by f (k, a, n) = kan is a homeomorphism identifying the homeomorphism type of SL2 (R) as
S 1 × R2 . Since R2 ∼= D2 , the open unit disk, we can (topologically) think of SL2 (R) as the interior
of a solid torus, S × D2 . Beware that SL2 (R) is not isomorphic to S 1 × R2 as a group!
1

16. V ECTOR CALCULUS ON THE TRACE ZERO MODEL OF THE HYPERBOLIC PLANE
Consider the real vector space sl2 (R) consisting of 2 × 2 real matrices with trace 0,
  
a b 
sl2 (R) = a, b, c ∈ R .
c −a 
(It turns out that this space is naturally identified with the tangent space at I to SL2 (R). This
makes it the Lie algebra of SL2 (R), whence the lowercase mathfrak font.) We compute
    2 
a b a b a + bc 0
=
c −a c −a 0 a2 + bc
so matrices R ∈ sl2 (R) satisfy R2 = (− det R)I.
We imbue sl2 (R) with the symmetric bilinear form
1
hR, Si = − tr(RS).
2
45
(This bears much similarity to the form you considered in Problem 5 of the first homework assign-
ment, but is slightly different.) We compute
1 1
hR, Ri = − tr(R2 ) = − tr((− det R)I) = det R
2 2
so we may also view h , i as the polarization of the determinant quadratic form.
We can check the Sylvester type of our inner product by producing an orthonormal basis for
sl2 (R). The reader is invited to check that
     
1 0 0 1 0 1
(16.1) , ,
0 −1 1 0 −1 0
works with determinants −1, −1, 1, respectively. It follows that sl2 (R) is a Minkowski space (with
Sylvester type (−2, 1)).
We now introduce the trilinear form vol on sl2 (R), defined by
1
vol(K, L, M ) = − tr(KLM )
2
for K, L, M ∈ sl2 (R). This form is alternating in the sense that it takes the value 0 when any two
of K, L, M are equal. For instance,
tr(KLK) = tr(LK 2 ) = − tr(L det K) = − det K tr L = 0.
The form vol is a volume form in the sense that it takes the value ±1 on any orthonormal basis of
sl2 (R). (This might be a good time to remember the universal property of the determinant of a
matrix.) Indeed, we can compute
       
1 0 0 1 0 1 1 −1 0
(16.2) vol , , = − tr = 1.
0 −1 1 0 −1 0 2 0 −1
In addition to h , i and vol, we need the operation ∧ on sl2 (R) defined by
1
K ∧ L = (KL − LK)
2
for K, L ∈ sl2 (R). We compute
1 1
hK ∧ L, M i = − tr(KLM − LKM ) = (vol(K, L, M ) − vol(L, K, M )) = vol(K, L, M )
4 2
where the last equality used that vol is alternating and hence − vol(L, K, M ) = vol(K, L, M ).
The following proposition summarizes some relations between these operations.
Proposition 16.3. For all A, B, C, D, Ai , Bi ∈ sl2 (R), i = 1, 2, 3, the following identities hold:
(a) A ∧ (B ∧ C) = hA, Ci B − hA, Bi C,
(b) hA ∧ B, C ∧ Di = hA, Ci hB, Di − hA, Di hB, Ci, and
(c) vol(A1 , A2 , A3 ) vol(B1 , B2 , B3 ) = det(hAi , Bj i)i,j .
Proof. Identities (a) and (b) are easy computations. For (c), begin by fixing A1 , A2 , A3 . Both sides
of the formula are alternating in B1 , B2 , B3 , so we may assume that B1 , B2 , B3 is a fixed positively
oriented orthonormal basis. Now altering A1 , A2 , A3 , we see that both sides are alternating in the
Ai , so we can in fact take A1 = B1 , A2 = B2 , and A3 = B3 to be the same orthonormal basis. We
have already computed in (16.2) that the left-hand side is 1 for the basis in (16.1). We also have
det diag(−1, −1, 1) = 1, so the desired equality holds. 
46
 17. P ENCILS OF GEODESICS
 
a b
The hyperboloid in sl2 (R) is the set of trace 0 matrices with determinant 1. If R = ,
c −a
then det R = −a2 − bc and if this quantity is 1, then we know that c 6= 0. Thus the sign of
c distinguishesbetween
 the sheets of the hyperboloid. We let h denote the sheet consisting of
a b
matrices R = with a2 + bc = −1 and c > 0. This is the sl2 (R) model of the hyperbolic
c −a
plane.

17.1. Normal vectors. Recall that geodesics γ : R → h can be parametrized as

γ(t) = A cosh t + T sinh t
where A = γ(0) and T = γ 0 (0) ∈ A⊥ with hT, T i = −1. With this notation, we define the normal
vector to γ to be
N = γ 0 (t) ∧ γ(t).
More explicitly, we may compute
N = (A sinh t + T cosh t) ∧ (A cosh t + T sinh t) = (T ∧ A)(cosh2 t − sinh2 t) = T ∧ A,
so it is in fact the case that N is independent of t. We have that N is orthogonal to T and A (since
hK ∧ L, M i = vol K, L, M and vol is alternating), and the norm of N is
hT ∧ A, T ∧ Ai = hT, T i hA, Ai − hA, T i hA, T i = −1.
We can recover γ 0 (t) from N by the formula
γ 0 (t) = γ(t) ∧ N.
(This follows from the definition of N and Proposition 16.3(a).) Note that N depends on the
orientation of γ, so the normal vector of an unoriented geodesic curve is only determined up to
sign.
We can now verify that h satisfies the reflection axiom!
Theorem 17.1 (Reflection Axiom). The complement of a geodesic ` in h has two connected components.
Reflection τ = τN along a normal vector N for ` will fix ` pointwise and interchange the two connected
components of h r `.
Proof. Let N be a normal vector for `. The function h → R, X 7→ hX, N i is zero for X ∈ ` and
nonzero on h r `. Thus the sets U and V where the function is positive and negative, respectively,
separate h r `. If τ = τN , then
τ (X) = X + 2 hX, N i N
from which it follows that hX, N i = − hτ (X), N i. This means that U and V are interchanged by τ .
It remains to show that U is connected. Consider two distinct points A, B ∈ U . Pick a tangent
vector T to h at A such that
B = A cosh d + T sinh d
where d = d(A, B). Let γ(t) = A cosh t + T sinh t, t ∈ R. Then
hγ(t), N i = cosh t(hA, N i + hT, N i tanh t).
If hT, N i ≥ 0, we get that γ(t) ∈ U for all t ∈ R. If hT, N i < 0, we can use properties of tanh to find
r ∈ R such that hγ(t), N i < 0 for all t > r and hγ(t), N i < 0 for all t < r. As such, γ(−∞, r) ⊆ U ,
γ(r, ∞) ⊆ V , and the geodesic arc [A, B] is a subset of U . This shows that U is path connected and
hence connected. 
47
 If ` is an oriented geodesic curve, then h r ` has two connected components and we can distin-
guish these sides of ` by its normal vector N : a geodesic curve through A ∈ ` with tangent vector
N at A passes from the negative side of ` to the positive side of `.

17.2. Angles. Recall that the tangent space of h at a point A ∈ h is TA (h) = A⊥ , the vectors in
sl2 (R) orthogonal to A. Call a pair of tangent vectors X, Y at A positively oriented if vol(A, X, Y ) >
0. Given a unit tangent vector S ∈ TA (h), one can check that S, S ∧ A forms a positively oriented
orthonormal basis for TA (h).
Given two unit tangent vectors S, T ∈ TA (h), the oriented angle ∠or (S, T ) ∈ R/2πZ between S
and T is defined by
T = S cos ∠or (S, T ) + S ∧ A∠or (S, T ).

17.3. Intersecting geodesics. Suppose that ` and m are oriented geodesics in h passing through
a common point A. Set U equal to the tangent vector to ` at A and set V equal to the tangent
vector to m at A. The directional angle from ` to m is defined to be α = ∠or (V, −U ). In this case,
−U = V cos α + V ∧ A sin α, and we can compute hU, V i = cos α and U ∧ V = A sin α. For the
normal vectors H = U ∧ A and K = V ∧ A, we compute
hH, Ki = cos α, H ∧ K = A sin α.
In particular, | hH, Ki | < 1. It turns out that the converse is true as well:
Proposition 17.2. Two distinct geodesics ` and m in h intersect if and only if their normal vectors
H and K satisfy | hH, Ki | < 1.
Proof. We have already observed one direction. For the other, suppose that | hH, Ki | < 1. The
discriminant for the plane E spanned by H and K is
∆ = hH, Hi hK, Ki − hH, Ki2 = 1 − hH, Ki2 > 0.
By the Discriminant Lemma 5.2, E has Sylvester type (−2, 0). Thus E ⊥ has type (0, 1) and inter-
sects h nontrivially. In fact, the point A which is the intersection of E and h is the intersection of `
and m. 
We say that geodesics ` and m are perpendicular in h if they meet at a point A with angle π/2.
Corollary 17.3. Geodesics ` and m in h are perpendicular if and only if their normal vectors are
orthogonal.
17.4. Geodesics with a common perpendicular geodesic. Suppose ` and m are geodesics both
perpendicular to another geodesic n. Let A be the point at which ` and n intersect, and let B
be the point at which m and n intersect. Let N be the normal vector for n corresponding to the
orientation of n from A towards B. Use the normal vector H = A ∧ N for ` and the normal vector
K = −B ∧ N for m. We find that
hH, Ki = hA, N i hN, Bi − hA, Bi hN, N i = hA, Bi .
As such,
hH, Ki = cosh d(A, B).
We also want to show that vol H, N, K = − sinh d(A, B). To prove this, observe that N ∧ K = B
since K = N ∧ B. Also vol(H, N, K) = hH, Bi, so
B = A cosh d(A, B) + H sinh d(A, B).
This gives hH, Bi = hH, Hi sinh d(A, B) = − sinh d(A, B) as required.
48
Theorem 17.4. Two distinct geodesics ` and m are perpendicular to the same geodesic if and only if their
normal vectors H and K satisfy | hH, Ki | > 1. The common perpendicular geodesic is unique whenever it
exists.
Proof. See [Ive92, III.2.9]. The unique geodesic has normal vector the unique vector of norm −1 in
span{H, K}⊥ . 
17.5. Lines with a common end. A geodesic ` in h spans a plane of type (−1, 1) in sl2 (R). The
two isotropic lines in this plane are called the ends of `. A normal vector H for ` is orthogonal to
the two ends of `.
Proposition 17.5. Two distinct geodesics ` and m in h have a common end if and only if their
normal vectors H and K satisfy | hH, Ki | = 1.
Proof. See [Ive92, III.2.10]. The common end is span{H, K}⊥ . 
17.6. Pencils of geodesics. A pencil is a set P of geodesics in h of the form
P = {` | ` a geodesic with normal vector in P }
where P is a fixed linear plane in sl2 (R). Note that P is the span of its vectors of norm −1, so P is
the span of the normal vectors of geodesics in the pencil it determines. The plane P has Sylvester
type (−2, 0), (−1, 1), or (−1, 0), and we will examine the nature of P according to this type.
If P has type (−2, 0), then the line P ⊥ has type (0, 1) and intersects h in a point A. The pencil P
is the set of geodesics through A.
If P has type (−1, 1, then P ⊥ has type (0, −1) and is the span of a normal vector of a geodesic `.
The pencil P is the set of all geodesics perpendicular to `.
If P has type (−1, 0), then P ⊥ is isotropic and P is the set of geodesics with end P ⊥ .
The following statements now have straightforward verifications. See [Ive92, p.98] for details.
Proposition 17.6. (a) Through two distinct points A, B ∈ h, there is a unique geodesic `.
(b) Given a point A ∈ h and a geodesic `, there is a unique geodesic passing through A which is
perpendicular to `.
(c) Given a point A ∈ h and an end S, there is a unique geodesic through A with end S.
(d) Given an end S and a geodesic ` with ends different from S, there is a unique geodesic per-
pendicular to ` with end S.
(e) There is a unique geodesic with given ends R and S.

18. C LASSIFICATION OF ISOMETRIES

By Theorem 5.6, an isometry of h is the product of at most three reflections in geodesics in h. In
particular, every even reflection is a product αβ of reflections in geodesics a and b. The relative
position of a and b (in terms of pencils) will give finer structure on Isom+ (h). We will see that
odd isometries are easier: they are all glide reflections, that is, the composite of a translation along
a geodesics and a reflection in the same geodesic.
Theorem 18.1. Let α, β, γ be reflections in three geodesics a, b, c, respectively, belonging to the same pencil
P. Then the product αβγ is a reflection in a geodesic from P.
Proof. See [Ive92, p.99]. The argument hinges on considering how α, β, γ act on a nonzero vector
in P ⊥ where P is the plane in sl2 (R) associated with P. 
For a triangle 4ABC in h, let ma , mb , and mc be the perpendicular bisectors of BC, AC, and
AB, respectively.
Corollary 18.2. The perpendicular bisectors ma , mb , mc for 4ABC lie in a pencil.
49
Proof. Let P be the pencil containing ma and mb and let ` ∈ P be the geodesic through A. Let µa ,
µb , and α denote reflection in ma , mb , and `, respectively. By Theorem 18.1, µa µb α is a reflection in
a geodesic m ∈ P. But µa µb α(A) = µa µb (A) = µa (C) = B, so m = mc , implying that mc ∈ P. 
18.1. Horolations. Horolations are a form of even isometry that are also referred to as limit rota-
tions or parabolic transformations.
Definition 18.3. Let S be an isotropic line in sl2 (R). A horolation with center S is an isometry of
the form αβ where α and β are reflections in geodesics a and b with end S.
Note that the set of horolations with center S forms a group. Indeed, the tricky part is closure
under multiplication, but this is a corollary of Theorem 18.1. More is true:
Corollary 18.4. The group of horolations with center S is abelian.
Proof. Suppose ρ, σ, τ are reflections in geodesics in h with end S. Then ρστ = (ρστ )−1 = τ −1 σ −1 ρ−1 =
τ σρ. For four such reflections, we get
(αβ)(γδ) = (αβγ)δ = (γβα)δ = γ(βαδ) = γ(δαβ) = (γδ)(αβ)
as desired. 
The reader should consult [Ive92, p.101] for further details on horolations. It is actually the case
that the horolations with center S are isomorphic (as a group) to (R, +). We also have complete
control over the orbits of this group acting on h.
18.2. Rotations. Rotations are another class of even isometries of h that we have previously called
elliptic transformations.
Definition 18.5. A rotation around a point A of h is an isometry of the form αβ where α and β are
reflections in geodesics a and b passing through A.
Rotations around A ∈ h formm a group isomorphic to R/2πZ. The orbits of this group are the
circles centered at A.
18.3. Translations. Translations are the class of even isometries of h that we previously called
hyperbolic transformations.
Definition 18.6. Let k denote a geodesic in h. A translation along k is an isometry of h of the form
αβ where α and β are reflections in geodesics a and b perpendicular to k.
Translations along k form a group isomorphic to R. The reader should take a moment to think
through the orbits of this group acting on h.
18.4. Odd isometries. The odd isometries of h are those which are the composition of an odd
number of reflections. They all have a fairly simple form:
Proposition 18.7. Every odd isometry of h is a glide reflection of the form τ κ where τ is a translation
along a geodesic k and κ is reflection in k.
Proof. Let φ be an odd isometry of h. Fix A ∈ h. If A = φ(A), then φ is a reflection in a geodesic
through A. If A 6= φ(A), let m denote the geodesic through A and φ(A) and let ` denote the
perpendicular bisector for A and φ(A). Let M denote the midpoint between A and φ(A), which is
also the intersection of ` and m. Let µ and λ denote reflection through m and `, respectively. Then
λφ is even and fixes the point A, so λφ is a rotation around A. By Theorem 18.1, λφ = µν where ν
is a reflection in a geodesic n through A. Let k be the geodesic through M perpendicular to n and
let κ denote reflection in k. Then
φ = λ(µν) = (λµ)(νκ)κ.
50
Since λµ and νκ are both half-turns around points of k, τ = (λµ)(νκ) is a translation along k,
proving the result. 

19. T HE ACTION OF THE SPECIAL LINEAR GROUP IN THE TRACE ZERO MODEL
We have seen that Isom(h) ∼
= Lor(sl2 (R)), and we will use this to identify Isom(h) with PGL2 (R) =
GL2 (R)/R× . (Here R× has been identified with the subgroup of nonzero scalar matrices.) To do
so, we begin by defining an action of GL2 (R) on sl2 (R). We set
σX = sign(σ)σXσ −1
where σ ∈ GL2 (R), X ∈ sl2 (R), and sign(σ) = ±1 is the sign of the determinant of σ. We leave
it to the reader to check that conjugation by σ is an orthogonal transformation of sl2 (R), and the
sign factor is there so that the action preserves sheets.
Theorem 19.1. The action of GL2 (R) on sl2 (R) induces an isomorphism
PGL2 (R) ∼ = Lor(sl2 (R)).
Proof. You have already checked that each σ ∈ GL2 (R) induces a Lorentz transformation of sl2 (R).
This means that we have a group homomorphism f : GL2 (R) → Lor(sl2 (R)). It remains to show
that f is surjective with kernel R× .
For surjectivity, recall that Lor(sl2 (R)) is generated by Lorentz reflections, so it suffices to show
that for every K ∈ N (sl2 (R)), the reflection τK is in the image of f . The reader may check by
calculation that for all K, X ∈ sl2 (R), we have
KX + XK = −2 hK, Xi I.
If K ∈ N (sl2 (R)), then K = K −1 . Multiplying on the right by this factor and rearranging gives
(19.2) − KXK −1 = X + 2 hK, Xi K.
On the left, we see the action of K on X, and on the right we see τK (X), so f is surjective.
The determination of the kernel of f is more straightforward and we ask the reader to think
through the details. Begin by assuming that σ acts trivially on sl2 (R) and use this to show that σ
commutes with every 2 × 2 real matrix. It follows that σ is a scalar matrix, as desired. See [Ive92,
p.105] for details. 
Corollary 19.3. The action of SL2 (R) on sl2 (R) identifies PSL2 (R) with the special Lorentz group
Lor+ (sl2 (R)).
Proof. It suffices to show that the action of σ ∈ GL2 (R) is an even Lorentz transformation if and
only if det σ > 0. To this end, pick a decomposition of the action of σ on sl2 (R) as a product
r1 , . . . , rs of reflections along vectors K1 , . . . , Ks ∈ N (sl2 (R)). It follows from (19.2) that σ and the
matrix K1 · · · Ks ∈ GL2 (R) have the same action on sl2 (R). As such, there is a scalar λ ∈ R× such
that
σ = λK1 . . . Ks .
Taking the determinant of both sides, we get det σ = λ2 (−1)s , so det σ > 0 if and only if s is
even. 
We can finally connect the geometric classification of the previous section to the square trace
analysis we performed in Section 15. If σ ∈ SL2 (R), then its action on h can be decomposed as αβ
where α, β are reflections in geodesics h, k, respectively. Let H and K denote normal vectors for h
and k, respectively.
Theorem 19.4. In the notation of the previous paragraph,
tr2 (σ) = 4 hH, Ki2 .
51
Proof. By the proof of the preceding corollary, we can write σ = εHK where ε = ±1. Thus
tr σ = ε tr(HK) = −2ε hH, Ki .
Squaring both sides proves the theorem. 
This allows us to complete our dictionary between geometric and algebraic descriptions of even
isometries of h. Geometrically, these tranformations are always the product of reflections in two
geodesics, h and k. The relative position of these geodesics determines (and is determined by) the
range in which the square trace tr2 σ of the associated σ ∈ PSL2 (R), as summarized here (compare
with [Ive92, Table III.4.8]):
» The geodesics h and k intersect in h if and only if 0 ≤ tr2 (σ) < 4. In this case, σ is elliptic /
a rotation.
» The geodesics h and k have a common perpendicular in h if and only if tr2 (σ) > 4. In this
case, σ is hyperbolic / a translation.
» The geodesics h and k have a common end in ∂h if and only if tr2 (σ) = 4. In this case, σ is
parabolic / a horolation / a limit rotation.
The conjugacy class in PGL2 (R) of a general matrix σ ∈ GL2 (R) is deteremined by tr2 (σ) and
sign det σ. See [Ive92, Proposition III.4.9] for details.

20. T RIGONOMETRY
Consider a hyperbolic triangle 4ABC, oriented so that its normal vectors point inwards. Let
the angle at A be α, the angle at B be β, and the angle at C be γ; let the length of the side opposite
A be a, opposite B be b, and opposite C be c.
Proposition 20.1 (Hyperbolic cosine and sine relations). With the conventions above,
cosh a = cosh b cosh c − sinh b sinh c cos α
and
sinh a sinh b sinh c
= = .
sin α sin β sin γ
Proof. Recall that
B = A cosh c + V sinh c and C = A cosh b − U sinh b
for V the unit tangent vector to AB at A and U the unit tangent to CA at A. We have hU, V i = cos α,
so
hB, Ci = cosh b cosh c − sinh b sinh c cos α.
But we also know that hB, Ci = cosh a, and the cosine relation follows.
The sine relation follows from evaluating vol(A, B, C) in a symmetrical fashion. See [Ive92,
II.5.1] for details. 
Proposition 20.2 (Alternative cosine relation). With the conventions above,
cos β cos γ + cos α
cosh a = .
sin β sin γ
Let H, K, L be the inward directed normal vectors for BC, CA, and AB, respectively. Then, by
Proposition 16.3(b) and the discussion around the definition of oriented angle,
hH ∧ K, L ∧ Hi = hK, Hi hH, Li − hH, Hi hK, Li = cos β cos γ + cos α.
We also have
H ∧ K = C sin γ and L ∧ H = B sin β.
52
Combined with the definition of hyperbolic distance, we get
hH ∧ K, L ∧ Hi = hB, Ci sin γ sin β = cosh a sin γ sin β.
Corollary 20.3. The sum of the angles in a hyperbolic triangle satisfies
α + β + γ < π.
Proof. Without loss of generality, assume α ≥ β, γ. By the alternative cosine relation and manipu-
lation of trig functions, cosh a > 1 becomes
cos(π − α) < cos(β + γ).
When β + γ ≤ π, we get π − α > β + γ, so α + β + γ < π as desired. If π ≥ β + γ, then the display
above implies cos(α + π) < cos(β + γ), whence α + π < β + γ, contradicting α ≥ β, γ. 
Theorem 20.4. Let α, β, γ ∈ (0, π) be real numbers with α + β + γ < π. Then there exists a triangle
4ABC in h with ∠A = α, ∠B = β, and ∠C = γ.
Proof sketch. Since β + γ < π − α, we know that
cos(β + γ) > cos(π − α).
By angle addition formulæ, this implies
sin β sin γ < cos β cos γ + cos α,
whence there exists a > 0 such that
sin β sin γ cosh a = cos β cos γ + cos α.
Now pick B, C ∈ h with d(B, C) = a and draw a geodesic ` through B forming angle β with BC;
also draw a geodesic k through C forming angle γ with BC. Use vector calculus on normal vectors
to check that ` and k intersect with angle α at a point A. Then 4ABC is the desired hyperbolic
triangle. See [Ive92, III.5.4] for details. 
We will now investigate the so-called Lambert quadrilaterals. These are quadrilaterals 2ABCD
with three right angles. In hyperbolic geometry, the fourth angle in such a quadrilateral is nec-
essarily acute, as follows from the second relation below. In particular, there are no hyperbolic
rectangles!

Proposition 20.5. In a Labert quadrilateral 2ABCD with right angles at A, B, and D and ∠C = γ,
cosh d(A, D) = cosh d(B, c) sin γ and sinh d(A, B) sinh d(D, A) = cos γ.
Proof sketch. Let H, K, L, M be inward directed normal vectors for BC, CD, DA, and AB, respec-
tively. First show that
cosh d(B, C) sin γ = hK, Li
by proceeding as in the proof of the alternative cosine relation. Since hK, Li = cosh d(A, D), this
gives the first equality.
For the second equality, use Proposition 16.3(c) to get
 
0 hK, Li hK, Hi
vol(K, L, M ) vol(M, L, H) = det −1 0 hM, Hi = hK, Hi .
0 −1 0
By the discussion at the start of §17.4 for hH, Ki = cos γ, we get the second identity. 
Of course, sinh is positive on (0, ∞) but cos t ≤ 0 for π/2 ≤ t ≤ 3π/2, so the fourth angle in a
Lambert quadrilateral must be acute.
53
 F IGURE 24. J.H. Lambert (1728–74) was a Swiss polymath who introduced hyper-
bolic trig functions, proved the irrationality of π, invented the hygrometer (which
measures humidity), and produced one of the first three-dimensional theories of
color.

F IGURE 25. A tiling of the Poincaré disk by Lambert quadrilaterals.

21. A NGLE OF PARALLELISM

Consider a hyperbolic triangle 4ABC in which A ∈ ∂h is an end. Since 4ABC is not contained
in h, the results of the previous section do not apply. Visually, though, we might suspect that
∠A = 0, and the following proposition justifies this convention.
54
Proposition 21.1. In the above setup,
cosh a sin β sin γ = cos β cos γ + 1.
Proof. Proceeding as we did for standard hyperbolic triangles, we still have
cosh a sin β sin γ = cos β cos γ + hK, Li .
Since AB and AC have a common end, hK, Li = ±1, and the value must be +1 since the left-hand
side above is positive. 
Note that when ∠C = π/2, the proposition implies that
cosh a sin β = 1,
sinh a tan β = 1,
tanh a sec β = 1
where the second two equalities follow from simple manipulation of the first.

22. R IGHT- ANGLED PENTAGONS

We have already seen that Lambert quadrilaterals always have fourth angle acute; i.e., there are
no rectangles in hyperbolic geometry. Bizarrely, there are pentagons with five right angles in h, as
we presently explore. (We prove existence in Proposition 22.3.)
Proposition 22.1. Suppose that DP1 P2 P3 P4 P5 is a right-angled pentagon with d(Pi , Pi+1 ) = ai for
1 ≤ i ≤ 5 (where we make the convention that P6 = P1 ). Then
cosh a5 = sinh a2 sinh a3 and cosh a3 = coth a2 coth a4
where coth = cosh / sinh is the hyperbolic cotangent function.
Proof. For 1 ≤ i ≤ 5 let Ni be the inward normal vector for Pi Pi+1 . Since all the angles are right,
we know hNi , Ni+1 i = 0 for all i. It follows from Proposition 16.3(c) that
vol(N1 , N2 , N3 ) vol(N2 , N3 , N4 ) = hN1 , N4 i .
Since P1 P2 and P4 P5 have the common perpendicular P5 P1 , the second display in §17.4 implies
that hN1 , N4 i = cosh a5 . Also by 17.4, the volume terms evaluate to − sinh a2 and − sinh a3 , respec-
tively, giving the first formula.
We defer the proof of the second identity to [Ive92, III.7]. The proof includes the more general
fact that
(22.2) sinh a2 sinh a4 cosh a3 = hN1 , N5 i + cosh a2 cosh a4
without the assumption that ∠P1 = π/2. When ∠P1 = π/2, hN1 , N5 i = 0 and the identity follows.

Proposition 22.3. Let a1 , a2 > 0 be real numbers. There exists a right-angled pentagon with two
consecutive sides of length a1 , a2 if and only if sinh a1 sinh a2 > 1.
Proof. Necessity of the condition follows from the first formula in Proposition 22.1 since cosh a5 >
1 for a5 > 0. The rest of the proof is a geometric construction. Begin by placing P1 , P2 , P3 such
that d(P1 , P2 ) = a1 , d(P2 , P3 ) = a2 , and ∠P2 = π/2. Draw the geodesic k3 through P3 perpendic-
ular to P2 P3 and the geodesic k5 through P1 perpendicular to P1 P2 . From our study of Lambert
quadrilaterals, we know that
sinh a1 sinh a2 = hN3 , N5 i > 1,
implying that k3 and k5 have a common perpendicular geodesic. 

55
 F IGURE 26. A right-angled pentagon constructed in NonEuclid via the method of
Proposition 22.3. Here A = P1 , B = P2 , C = P3 , k3 = CF , k5 = AE, and IH is the
common perpendicular to CF and AE.

23. R IGHT- ANGLED HEXAGONS

In future work, we will construct so-called hyperbolic pants by gluing together two congruent
right-angled hexagons.
Proposition 23.1. Suppose that 7P1 · · · P6 is a right-angled hexagon with d(Pi , Pi+1 ) = ai for
1 ≤ i ≤ 6 and P7 = P1 . Then
sinh a2 sinh a6 sinh a4
= =
sinh a5 sinh a3 sinh a1
and
sinh a2 sinh a4 cosh a3 = cosh a6 + cosh a2 cosh a4 .
The reader is encouraged to study the proof in [Ive92, III.8]. The following theorem provides
for the existence of right-angled hexagons and classifies them by the length of three alternating
sides.
Theorem 23.2. Let a2 , a4 , a6 > 0 be real numbers. Then there exists a right-angled hexagon in h such that
three alternating sides have lengths a2 , a4 , a6 . The hexagon is unique up to isometry.
Proof. First choose a3 to be the unique positive real number making the second formula in Propo-
sition 23.1 hold. Start the construction of the hexagon by placing points P2 , P3 , P4 , P5 in the plane
distances a2 , a3 , a4 apart and with ∠P3 = ∠P4 = π/2. Let h1 be the geodesic passing through P2
at a right angle to P2 P3 , and let h5 be the geodesic passing through P5 at a right angle to P5 P4 . Let
N1 and N5 denote the inward directed normal vectors for h1 and h5 . By (22.2) and the previous
56
proposition,
hN1 , N5 i = cosh a6 > 1.
As such, h1 and h5 have a common perpendicular geodesic h6 . Let P1 be the intersection of h1 and
h6 , and let P6 be the intersection of h5 and h6 . The above relation on normal vectors implies that
d(P1 , P6 ) = a6 . Uniqueness up to isometry is left to the reader. 

24. H YPERBOLIC AREA

We send the reader to [Ive92, III.9] to verify the following assertions.
Recall that the upper half-plane H 2 has metric
|z − w|2
 
d(z, w) = arccosh 1 + .
2 Im(z) Im(w)
Given z = x + iy ∈ H 2 with x, y ∈ R (and y > 0), we can build the matrix F (z) ∈ sl2 (R) given by
1 x −|z|2
 
F (z) = .
y 1 −x
A calculation reveals that F : H 2 → h is a PGL2 (R)-equivariant11 isometry.
One may also calculate that that for u, v ∈ R, v > 0, the derivative of F at z = x + iy is the linear
mapping given by
1 x −x2 + y 2
   
1 1 −2x
DFz (u + iv) = u− 2 v.
y 0 −1 y 1 −x
As such the norm of DFz (u + iv) is
1 2
(u + v 2 ).
det DFz (u + iv) = −
y2
Moral exericse: show that this implies that DFz preserves angles. As such, we are justified in
making the following statement.
Proposition 24.1. The transformation F : H 2 → h preserves oriented angles.
We can also make the calculation
1
vol(F (z), DFz (r + is), DFz (u + iv)) = (rv − su).
y2
In the language of differential forms, this justifies the formula
1
F ∗ vol = 2 dx ∧ dy,
y
whence we get the Riemannian metric
1
(dx2 + dy 2 )
y2
on the half-plane.12 This means that the area of a region ∆ in H 2 is
Z
1
area(∆) = 2
dx dy.
∆ y
We now check that the area of a hyperbolic triangle with angles α, β, γ is
(24.2) π − (α + β + γ).
11This means that F (σz) = σ · F (z) for all σ ∈ PGL (R), where σ · F (z) = sign(σ)σF (z)σ −1 .
2
12The reader unfamiliar with Riemannian metrics can safely ignore this claim.
57
In HW7, you will use this to verify that the area of a hyperbolic n-gon ∆ is
X
(24.3) area(∆) = (n − 2)π − ∠int P
P

where the sum is of interior angles at all vertices P of ∆.

Our verification of (24.2) follows [Thu97, §2.4]. First consider an ideal triangle with all vertices
on ∂H 2 . We may assume that the triangle
√ has vertices ∞, (−1, 0), and (1, 0).13 This is the region
2
{x + iy ∈ H | −1 ≤ x ≤ 1 and y ≥ 1 − x2 }. As such, its hyperbolic area is
Z 1Z ∞ Z 1 Z π/2
1 dx 1
√ 2
dy dx = √ = cos θ dθ = π.
−1 1−x 2 y −1 1−x 2
−π/2 cos θ
(Moral exercise: verify the above manipulations.)
Now suppose we have a “2/3-ideal triangle” with two vertices on ∂H 2 and one in H 2 . Let A(θ)
denote the area of such a triangle with angle π − θ at the finite vertex. (An isometry chase implies
that this value is well-defined.) We claim that A is an additive function (proven in the following
paragraph): A(θ1 + θ2 ) = A(θ1 ) + A(θ2 ) for θ1 , θ2 , θ1 + θ2 ∈ (0, π). If this is the case, then A is
Q-linear, and since it is continuous it is also R-linear. Since A(π) = π by the preceding paragraph,
we learn that A(θ) = θ and the area of a 2/3-ideal triangle is the complement of the angle at its
finite vertex.
We now verify that A is additive. (The reader is encouraged to draw a picture as they go
through this argument or to view the video lecture for visual supplements.) Let 4P QR be a 2/3-
ideal triangle with P, Q ∈ ∂H 2 and ∠R = π − θ1 . We can construct 4QRS with S ∈ ∂H 2 and
∠R = π − θ2 . Let O be the intersection of the geodesics P R and QS, and let T be the end of P O
which is not P . Then 4T RS has ∠R = π − (θ1 + θ2 ) and area A(θ1 + θ2 ). Notice, though, that a
rotation through O exhibits that 4P OQ ∼ = 4T OS, whence
A(θ1 + θ2 ) = A(θ1 ) + A(θ2 ).
Finally, we ask the reader to verify that an arbitrary hyperbolic triangle (with two or three
vertices in H 2 ) is the difference of an ideal triangle and two or three 2/3-ideal triangles. (Draw a
picture!) The area formula (24.2) follows.
You will also show in HW7 (via an integral computation) that the area of a hyperbolic disk D
of radius r is
(24.4) area(D) = 2π(cosh r − 1) = 4π sinh2 (r/2).
Note that this quantity is proportional to er as r → ∞, in stark contrast with the area of Euclidean
disks (proportional to r2 ).

25. F UCHSIAN GROUPS

A discrete subgroup of PSL2 (R) ∼= Isom+ (H2 ) is called a Fuchsian group. Presently, we study
slightly more general objects, namely discrete subgroups of PGL2 (R).

25.1. Discrete subgroups of locally compact groups. In order to initiate the study of Fuchsian
groups, we start with a few words on locally compact topological groups and their discrete sub-
groups. Students who have not studied point-set topology previously can focus on the theorem
statements below. The important topological notions are the following:

13Moral exercises: (1) Find isometries taking an arbitrary ideal triangle to this one. (2) Prove that isometries preserve
area.
58
 » topological space: a set equipped with a class of open subsets satisfying some axioms (empty
and total subsets are open, opens are closed under arbitrary unions, and opens are closed
under finite intersections); closed sets are (by definition) the complements of open sets;
» continuous function: a function f : X → Y between topological spaces such that the preim-
age of every open set in Y is open in X;
» [adjective] neighborhood of a point: a subset of the ambient space of [adjective] type containing
the point;
» Hausdorff space: a topological space for which every two points have disjoint open neigh-
borhoods;
» locally compact space: a Hausdorff space in which every point has a compact neighborhood;
» isolated point x of S ⊆ X: the point x has a neighborhood V in X such that V ∩ S = {x};
» discrete subset S ⊆ X: every point of S is isolated;
» accumulation point: a point w ∈ X is an accumulation point for S ⊆ X if every open
neighborhood of w contains infinitely many points of S.
We state without proof the following theorem on closed discrete subsets. (Note that {1/n | n ∈
Z+ } is a discrete subset of R which is not closed.)
Theorem 25.1. The following statements about a subset S of a topological space X are equivalent:
(1) S is closed and discrete,
(2) S has no accumulation points.
If, additionally, X is locally compact, then it is also equivalent that
(3) S intersects every compact subset of X in a finite set.
Henceforth, G will denote a locally compact topological group with identity element e. This
means that G is a locally compact space and that the product G × G → G and inverse G → G
functions are continuous. A discrete subgroup Γ ≤ G is a subgroup of G which is a discrete subset
of G.
Proposition 25.2. Let Γ ≤ G be a subgroup of G. If there exists an open neighborhood V of e ∈ G
such that V ∩ Γ = {e}, then Γ is a discrete subgroup of G.
Proof. Let V be such a neighborhood of e. Given σ ∈ G, we aim to construct an open neighborhood
U of σ in G such that U ∩ Γ = {σ} or ∅. We either have
σ(V −1 ) ∩ Γ = ∅ or σ(V −1 ) ∩ Γ 6= ∅
where V −1 = {v −1 | v ∈ V }. In the first case, σ(V −1 ) is a neighborhood of σ disjoint from Γ. In
the second case, we may take γ ∈ σ(V −1 ) ∩ Γ. Then σ ∈ γV , γ ∈ Γ, and
γ −1 (γV ∩ Γ) = V ∩ γ −1 Γ = V ∩ Γ = {e}.
We conclude that γV is an open neighborhood of σ meeting Γ in the unique point γ, as desired. 
We say that a group action of G on a topological space X is continuous if the associated function
G × X → X is continuous.
Corollary 25.3. Suppose G acts continuously on X and let Γ be a subgroup of G. If there exists a
point x ∈ X which is isolated in the orbit Γx and such that the stabilizer Γx = {γ ∈ Γ | γx = x} is
discrete in G, then Γ is a discrete subgroup of G.
Proof. Let U be an open neigborhood of x in X such that U ∩ Γx = {x}. Then the inverse image of
U along the map τx : G → X, g 7→ gx satisfies
τx−1 (U ) ∩ Γ = Γx .
Pick an open neighborhood V of e in G such that V ∩ Γx = {e}. Then V ∩ τx−1 (U ) isolates e in Γ,
and Proposition 25.2 applies. 
59
25.2. Discrete subgroups of the projective general linear group. In order to study Fuchsian
groups, we will need to specify a topology on PGL2 (R). To do so, first note that M2×2 (R) ∼
= R4
has the standard Euclidean topology, and GL2 (R) — the complement of the closed set defined by
ad − bc = 0 — is an open subset of M2×2 (R). A set U ⊆ GL2 (R) is open if it is open in M2×2 (R).
(Equivalently, open sets in GL2 (R) are of the form V ∩ GL2 (R) where V ⊆ M2×2 (R) is open.)
The quotient group PGL2 (R) = GL2 (R)/R× carries the natural quotient topology. Its open sets
correspond to open subsets of GL2 (R) that are stable under the action of R× .
We will use the action of PGL2 (R) on H 2 to study its discrete subgroups, but first it will be
convenient to produce some compact neighborhood of e ∈ PGL2 (R).
Theorem 25.4. Given z ∈ H 2 and ε > 0, let
R(z; ε) = {σ ∈ PGL2 (R) | d(z, σz) ≤ ε}.
Then R(z; ε) is a compact neighborhood of e.
Proof. Define the norm kσk of a matrix σ ∈ GL2 (R) via
  s 2 2 2 2
= a +b +c +d .
a b

c d | det σ|
Since kλσk = kσk for λ ∈ R× , the norm descends to a real-valued function on PGL2 (R), and it is
not hard to check that it is continuous. As such, the set
(25.5) {σ ∈ PGL2 (R) | kσk ≤ r}
√
is a neighborhood of e in PGL2 (R) for all r > 2. This neighborhood is compact because it is the
image of a closed and bounded set in GL2 (R) ⊆ M2×2 (R) ∼ = R4 .  
0 −1
We now show that R(i; ε) is a compact neighborhood of e, where i = ∈ h. A cal-
1 0
culation (see [Ive92, p.126]) reveals that 2 cosh d(i, σi) = kσk for σ ∈ PGL2 (R). Since sets of the
form (25.5) are compact neighborhoods of e in PGL2 (R), we learn that the set R(i; ε) is a compact
neighborhood of e as well.
Given an arbitrary point z ∈ h, transitivity of PGL2 (R) acting on h implies that there exists
τ ∈ PGL2 (R) such that τ z = i, so
R(z; ε) = {σ | d(τ −1 i, στ −1 i) ≤ ε}
= {σ | d(i, τ στ −1 i ≤ ε}
= τ −1 R(i; ε)τ.
We conclude that R(z; ε) is a compact neighborhood of e, as desired. 
Corollary 25.6. Let Γ ≤ PGL2 (R). If Γ is discrete, then for all z ∈ h and ε > 0 the set
{σ ∈ Γ | d(z, σ(z)) ≤ ε}
is finite. Conversely, if for one ε > 0 and one point z ∈ h the above set is finite, then Γ is a discrete
subgroup of PGL2 (R).
Proof. The set in question is R(z; ε)∩Γ. The result follows from Theorem 25.4 and Proposition 25.2.

Example 25.7. Let GL2 (Z) denote the group of 2 × 2 integer matrices with determinant ±1. This
is a discrete subgroup of GL2 (R) since any of the subsets
{σ ∈ GL2 (Z) | kσk ≤ n}
60
 √
is finite (and this is the intersection of GL2 (Z) with a compact neighborhood of e for n > 2). It is
a general fact that any discrete subgroup Γ ≤ GL2 (R) consisting of matrices with determinant ±1
maps onto a discrete subgroup of PGL2 (R), so PGL2 (Z) = GL2 (Z)/{±I} is discrete subgroup of
PGL2 (R).
Proposition 25.8. A discrete subgroup Γ ≤ PGL2 (R) acts on h with discrete orbits.
Proof. Let Γz be an orbit. By Corollary 25.6, a disk D with center z will contain finitely many
points from Γz. Thus there exists ε > 0 such that D(z; ε) ∩ Γz = {z}. We now show that in fact
d(u, v) ≥ ε for all u 6= v ∈ Γz, implying that Γz is discrete. To see this, choose σ ∈ Γ with σz = u.
Then σv 6∈ D(z; ε) so d(u, v) = d(σu, σv) ≥ ε. 
Remark 25.9. You will show in HW7 that there is a natural action of R on the circle R/2πZ for which
Z-orbits are not necessarily discrete, so the above proposition is special to the action of discrete
subgroups of PGL2 (R) on h.
Theorem 25.10. Let Γ ≤ PGL2 (R) be discrete. For compact subsets K, L ⊆ h, the sets σ(K) and L are
disjoint for all but finitely many σ ∈ Γ.
Proof. Given a compact set K ⊆ h, a point z ∈ h, and a number r > 0, it suffices to show that
D(z; r) and σ(K) are disjoint for all but finitely many σ ∈ Γ. (Moral exercise: why does this
suffice?) Choose s > 0 with K ⊆ D(z; r + s) and conclude from Corollary 25.6 that S = {γ ∈ Γ |
γ(z) ∈ D(z; r+s)} is finite. Observe that γD(z; r) = D(γ(z); r) and conclude that K ∩γD(z; r) = ∅
for all γ ∈ Γ r S. 
Remark 25.11. The property described in the theorem has a name: the action of PGL2 (R) on h
is properly discontinuous. This puts us in the unfortunate terminological bind of working with a
continuous properly discontinuous action, but such is life. Some authors use the term properly
discrete instead, but this is far less common. When a locally compact group G acts continuously
on a space X in a properly discontinuous fashion, the quotient map X → X/G is a covering map.
This leads other authors (for example, Hatcher) to call such actions covering actions.
Corollary 25.12. Let Γ be a discrete subgroup of PGL2 (R) and K a compact subset of h. Only
finitely many elements of Γ have fixed points in K.
Proof. For all but finitely many σ ∈ Γ, the sets K and σ(K) are disjoint. 
An elliptic fixed point for Γ on h is a point z ∈ h which is fixed by some proper elliptic element
σ ∈ Γ. (Recall that elliptic means σ ∈ PSL2 (R) with tr2 σ < 4; these correspond to hyperbolic
rotations.)
Corollary 25.13. Let Γ ≤ PGL2 (R) be discrete. The set of elliptic fixed points for Γ on h is a discrete
subset of h.
Proof. This follows immediately from the previous corollary and Theorem 25.1 as soon as the
reader has checked that every Fuchsian group is closed. 
In fact, a converse to Corollary 25.13 is true as well. Call Γ ≤ PGL2 (R) elementary if it fixes a
point of h, fixes a point of ∂h, or stabilizes a geodesic in h. It is a fact (proved in [Ive92, §IV.2]) that
subgroups of PGL2 (R) are elementary if and only if they are solvable.
Theorem 25.14 (Jakob Nielsen’s theorem). A non-elementary subgroup Γ of PGL2 (R) is discrete if and
only if the set of elliptic fixed points in h is discrete.

The proof of this theorem is difficult and occupies §§IV.2-4 of [Ive92]. Due to time constraints,
we will not present it here.
61
 F IGURE 27. Jakob Nielsen (1890–1959) was a Danish mathematician who pio-
neered geometric group theory. From 1952 to 1958, he was on the executive board
of UNESCO.
26. C USPS AND HOROCYCLIC COMPACTIFICATION
Note to the reader: Because of time constraints, these notes are going to switch to a more descriptive
style. See the corresponding portions of [Ive92] for full details and proofs. If you would like to see
some elaboration on any portion of these notes or the textbook, please say so on the Slack channel.
We now study how a discrete group Γ ≤ PGL2 (R) acts on the boundary of the hyperbolic plane.
A point S ∈ ∂h is a cusp for Γ if it is fixed by a proper horolation (or parabolic transformation) β ∈
Γ. (By proper, we mean not the identity element.) The stabilizer ΓS of a cusp consists of horolations
with center S and reflections in geodesics through S. A horodisk is a region in h bounded by a
horocycle. (Recall that horocycles are the orbits of the group of horolations with center S. In
the upper half-plane model, they are generalized circles in H 2 ∪ ∂H 2 parallel to ∂H 2 at S. When
S = ∞, this is just a horizontal line in h bounding the horodisk consisting of points with imaginary
part greater than some constant.)
Proposition 26.1. Given a cusp S ∈ ∂h for a discrete group Γ ≤ PGL2 (R), there exists a horodisk
D such that γ(D) ∩ D = ∅ for γ ∈ Γ r ΓS .
Similarly, we can choose a horodisk at a cusp so that its Γ-orbits avoid any compact subset of h:
Proposition 26.2. Given a cusp S ∈ ∂h for a discrete group Γ ≤ PGL2 (R) and a compact set K ⊆ h,
there exists a horodisk D with center S such that γ(D) ∩ K = ∅ for all γ ∈ Γ.
Corollary 26.3. Given cusps S, T ∈ ∂h for a discrete group Γ ≤ PGL2 (R) which are in different
Γ-orbits, for any horodisk E with center T there exists a horodisk D with center S such that γ(D)∩
E = ∅ for all γ ∈ Γ.
We will be very interested in the quotient spaces h/Γ for Γ ≤ PGL2 (R) discrete. By definition,
h/Γ is the set of Γ-orbits of points in h with the topology defined by the condition that U ⊆ h/Γ is
open if and only if q −1 U ⊆ h is open for q : h → h/Γ the quotient map. Observe that q −1 (q(U )) =
S 14
γ∈Γ γ(U ), so q takes open sets to open sets.

14This makes q an open map. Beware that not all cotinuous maps are open!
62
Corollary 26.4. If Γ ≤ PGL2 (R) is discrete and h/Γ is compact, then the group Γ contains no
proper horolations.
Proof. Cover h with open disks. The image of these disks is an open cover of h/Γ. By compactness,
we can take a finite subcover. Lifting back to h, we see that there is a finite set of disks in h meeting
all Γ-orbits. As such, we can find a compact set K ⊆ h meeting all Γ-orbits (the union of the disks).
Corollary 26.3 implies that Γ contains no proper horolations. 
For the rest of this section, fix a discrete group Γ ≤ PGL2 (R) and let Y = Y (Γ) be the subset of
h = h ∪ ∂h consisting of h and the cusps for Γ. We give Y the following topology: W ⊆ Y is open
if for all S ∈ W there exists a disk or horodisk with center S entirely contained in W . This is called
the horocyclic topology.
Since Γ acts continuously on Y , the orbit space X = Y /Γ is a topological space which is in fact
a topological surface (two-dimensional manifold) with boundary:
Theorem 26.5. The space X = Y /Γ is Hausdorff and every point has a neighborhood homeomorphic to an
open neighborhood of 0 in R2 or an open neighborhood of 0 in R × [0, ∞).
Proof. That X is Hausdorff follows from Proposition 26.2 and Corollary 26.3. We separately con-
struct neighborhoods of cusp and regular points. To construct a neighborhood of a cusp point, it
suffices to consider the point ∞ ∈ H 2 and Γ = Γ∞ (by Proposition 26.1). If Γ = Γ+ is generated
by z 7→ z + k for some k > 0, then Y /Γ is homeomorphic to the open unit disk D in the complex
plane via
Y /Γ −→ D
z 7−→ exp(2πiz/k).
In the general case, we can take Γ to be generated by z 7→ −zz and z 7→ z + k, whence X is the
orbit space for the action of complex conjugation on the complex unit disk in C.
For a regular point, it suffices to consider the origin 0 in the Poincaré disk and assume Γ = Γ0 . If
Γ = Γ+ is generated by z 7→ θz for θ a primitive n-th root of unity, then the fact that z 7→ z n is open
implies that H 2 /Γ ∼
= D. In the case that Γ/Γ+ is nontrivial, a nonidentity element corresponds to
a reflection of D in a line through 0, so we again get a half-space neighborhood. 
Corollary 26.6. The space X = Y /Γ is compact if and only if there exists a compact subset K ⊆ h
and a finite number of horodisks D1 , . . . , Dr with centers S1 , . . . , Sr cusps for Γ such that any
Γ-orbit in h meets K ∪ D1 ∪ · · · Dr .
When Γ is a Fuchsian group (a discrete subgroup of PSL2 (R)), the proof of Theorem 26.5 implies
that X = Y /Γ is a Riemann surface, that is, a one-dimensional complex-analytic manifold. When X
is compact, it is called the horocyclic compactification of h/Γ.

27. T HE MODULAR GROUP AND ITS FUNDAMENTAL DOMAIN

Consider the action of (R, +) on itself by translation. Let Γ ≤ R be the discrete group Z. To
understand R/Z, it is convenient to select a “nice” set of orbit representatives. Let U = (0, 1).
Then U and n + U are disjoint for all n ∈ Z r {0}, and each Z-orbit meets U = [0, 1]. Thus R/Z has
the same topology as U /(0 ∼ 1), which is clearly a model for the circle. The set U is a fundamental
domain for the action of Z on R.
Definition 27.1. A fundamental domain U for a group Γ ≤ PGL2 (R) is an open subset of H2 such
that U and γU are disjoint for γ ∈ Γ r {e} and such that each Γ-orbit meets the closure U of U .
We now construct a fundamental domain for the modular group PSL2 (Z).
63
Proposition 27.2. The three geodesics in the upper half-plane H 2 defined by
1 1
Re(z) = , Re(z) = − , and |z| = 1
2 2
bound an open set M which is a fundamental domain for the group PSL2 (Z).
Proof. Let σ, τ ∈ Isom+ (H 2 ) be the z 7→ −z −1 and z 7→ z + 1, respectively. The
 transformations
  
0 −1 1 1
corresponding matrices are S = and T = . Let ω = τ −1 : z 7→ z − 1 with matrix
1 0 0 1
 
−1 1 −1
T = . Fix w = yi for some y > 1 and note that the geodesics of the proposition are the
0 1
perpendicular bisectors for [w, τ (w)], [w, ω(w)], and [w, σ(w)], respecitvely. As such, we have
M = {z ∈ H 2 | d(w, z) ≤ d(w, γ(z)) for γ = σ, τ, ω}
We currently aim to find a point of M in the orbit of a given point z ∈ H 2 . Choose ν ∈ PSL2 (Z)
such that d(w, ν(z)) ≤ d(w, µ(z)) for all µ ∈ PSL2 (Z). (Important moral exercise: How do you
guarantee the existence of such ν?) In particular, for any of γ = σ, τ , or ω, we get
d(w, ν(z)) ≤ d(w, γ(z))
so ν(z) ∈ M .
It remains to show that any orbit O of PSL2 (Z) meets M in at most one point. We claim that
the function O → R, z 7→ Im(z) is bounded
 and attains its upper bound when z ∈ M . Indeed, the

a b
reader may check that for z ∈ M and ∈ SL2 (Z),
c d
 
az + b Im(z)
Im =
cz + d |cz + d|2
and |cz + d| > 1. Thus Im(µ(z)) ≤ Im(z) for all z ∈ M and µ ∈ PSL2 (Z). Furthermore, the
inequality is strict when c 6= 0, and when c = 0 the transformation is of the form z 7→ z + b, b ∈ Z.
Suppose now that z, w ∈ O ∩ M . By the above argument (and the fact that z and w are both
PSL2 (Z)-translates of each other), we learn that Im(z) = Im(w). But then w = z + b for some b ∈ Z
(and still z, w ∈ M ) so b = 0 and z = w. This verifies that |O ∩ M | ≤ 1. 

It is in fact the case that M ∪ {z ∈ ∂M | Re(z) ≥ 0} is a set of orbit representatives of PSL2 (Z)
acting on H 2 . We can think of H 2 / PSL2 (Z) as M with the left vertical glued to the right vertical
(by τ ) and the left half of the arc glued to the right half (via σ).15
Later, we will see that PSL2 (Z) is generated by σ and τ . The transformation ρ = τ σ is a rotation
by 2π/3, so we have the relations
σ 2 = e = ρ3 .
It is a consequence of Poincaré’s theorem (one of our end goals for the course) that this set of
relations is complete, that is
PSL2 (Z) ∼
= S, T | S 2 , (T S)3



in the language of group presentations.

Now consider the extended modular group PGL2 (Z). The reflection β in the imaginary axis di-
vides M into two halves, and it is in fact the case that the hyperbolic triangle ∆ with vertices

15To make this completely rigorous, we will need to observe that M is a locally finite fundamental domain. This is
covered in the next section.
64
i, ζ = e2πi/3 , ∞ is a fundamental domain for PGL2 (Z). Let α denote reflection in the left verti-
cal side [ζ, ∞] of ∆ and let γ denote reflection in [ζ, i]. One can then geometrically deduce the
identities
α2 = β 2 = γ 2 = e, αγα = γαγ, and βγ = γβ
in PGL2 (Z)! (Again, Poincaré’s theorem will tell us that these relations are complete.)
Lemma 27.3. Let Γ denote a discrete group of isometries of H 2 and suppose Π ≤ Γ is a subgroup
of finite index. If D is a fundamental domain for Γ and S ⊆ Γ is a full set of representatives for
Π\Γ, then the interior U of
[
F = αD
α∈S
is a fundamental domain for Π.
Proof. See [Ive92, Lemma V.1.4]. I’ll draw a picture in lecture. 
We will use this lemma to describe a fundamental domain for the level 2 modular group Γ(2).
We begin with the group G(2) which is the kernel of the surjective homomorphism PGL2 (Z) →
PGL2 (F2 ) which reduces the entries in an integer matrix mod 2. Since 1 = −1 in F2 = Z/2Z,
we have PGL2 (F2 ) = GL2 (F2 ). The action of GL2 (F2 ) on the three lines in F22 reveals that it is
isomorphic to the permutation group Σ3 . We can lift this group back to PGL2 (Z) as the dihedral
group D6 generated by α and γ, which have matrices
   
−1 −1 0 1
and ,
0 1 1 0
respectively. By the lemma, the union of the six translates of ∆ by D6 is a fundamental domain for
G(2). As such, the ideal hyperbolic triangle with vertices 0, −1, ∞ is a fundamental domain for
G(2).
The group Γ(2) is G(2)+ , the even isometries in G(2). Applying the lemma to the reflection β,
we see that the ideal quadrilateral with vertices 1, 0, −1, ∞ is a fundamental domain for Γ(2).

28. L OCALLY FINITE AND CONVEX FUNDAMENTAL DOMAINS

Throughout this section, let Γ be a discrete subgroup of PGL2 (R). In order for a fundamental
domain to provide good control over the geometry of H2 /Γ, it must additionally be locally finite
and convex. We will state the results here, but not pursue the proofs. See [Ive92, §§V.2–3] for
details.
A fundamental domain P ⊆ H2 for Γ is locally finite if every z ∈ H2 has a neighborhood which
meets σ(P ) for most finitely many σ ∈ Γ. Constructing a non-locally finite fundamental domains
require a certain perversity of thought and we will not pursue an example here. Beware, though,
that they exist!
When P is locally finite, we have the following two crucial results.
Proposition 28.1. If P ⊆ H2 is a locally finite fundamental domain for Γ, then the canonical pro-
jection map p : P /Γ → H2 /Γ is a homeomorphism.
Corollary 28.2. If P ⊆ H2 is a locally finite fundamental domain for Γ, then H2 /Γ is compact if
and only if P is bounded.
A region R ⊆ H2 is convex if for all A, B ∈ R, the geodesic arc [A, B] is contained in R. It turns
out that a convex locally finite fundamental domain P is a hyperbolic polygon: its sides are of the
form P ∩ σP , σ ∈ Γ r {e}; a vertex of P is of the form P ∩ σP ∩ τ P , σ 6= τ ∈ Γ r {e}. The boundary
of P is the union of these sides, and the vertices are necessarily endpoints of exactly two sides.
65
 For the remainder of this section, fix P a convex locally finite fundamental domain for Γ. A side
of P has a presentation as s = P ∩ σs P for a unique σs ∈ Γ r {e}. We define the ∗-operator on
sides, called the side pairing, by
∗s = P ∩ σs−1 P .
Then σs (∗s) = s, and s 7→ ∗s is an involution on the set of sides of P .
We will call an oriented side of P an edge. We can extend ∗ in a natural way to the set E of edges
of P . Let E+ be the set of edges with initial vertex in H2 (as opposed to ∂H2 ). Given an edge s with
initial vertex A, let ↓ s denote the second edge with initial vertex A. We define a new operator
Ψ : E+ −→ E+
s 7−→↓ ∗s.
Since Ψ is the composition of two involutions, it is a bijection.
Example 28.3. Let s be the Re(z) = −1/2 side of the modular fundamental
  domain M for PSL
 2 (Z),

2πi/3 1 −1 −1 1 1
oriented so that ζ = e is the initial vertex. Then σs = ω = with σs = τ = .
0 1 0 1
Thus ∗s = M ∩ τ M = t where t is the Re(z) = 1/2 side of M , oriented so that −ζ −1 is its initial
vertex. Thus Ψs =↓ ∗s is the oriented geodesic arc u = [−ζ −1 , i].
Lemma 28.4. The operator Ψ has finite orbits.
Proof. For s ∈ E+ with initial vertex A, the sequence of edges s, Ψs, Ψ2 s, Ψ3 s has initial vertices in
P ∩ ΓA, a finite set. 
Example 28.5. Continuing Example 28.3, one see that ∗u = v = [ζ, i] so Ψu =↓ v = s. Thus the
Ψ-orbit of s is just {s, u}. Similarly, one has Ψv = t and Ψt = v.
Lemma 28.6. Let s0 be an edge of P with initial vertex A and let si = Ψi s0 . Set σi = σsi . Then
there exists n ≥ 1 with σ0 σ1 · · · σn = e and such that
[
σ0 σ1 · · · σi P
0≤i≤n
is a neighborhood of A and [
σ0 σ1 · · · σi P
0≤i≤n
is a disjoint union.
Before discussing the proof, let’s see how this lemma plays out for P = M .
 
0 −1
Example 28.7. If s0 = s, then s1 = u, σ0 = ω, and σ1 = σ = . The reader may check
1 0
that ωσ is a rotation by 2π/3 about ζ, and σ0 σ1 · · · σ5 = (ωσ)3 = e. I will draw the corresponding
neighborhood of ζ in the lecture.
Proof idea for Lemma 28.6. One checks that each σ0 · · · σi P has A as a vertex and angle ∠int Ai at A
where Ai is the initial vertex of si . When these angles sum to 2π, we get the result. 
Theorem 28.8. If P is a locally finite fundamental domain for Γ, then Γ is generated by
{γ ∈ Γ | γ(P ) ∩ P 6= ∅}.
If P is also convex, then Γ is generated by the side transformations σs (as s runs through the sides of P ).
We defer the proof to [Ive92, V.2.3 and V.3.14].
Example 28.9. The modular group is generated by the side transformations {σ, τ, ω} of M . Since
ω = τ −1 , we can further cut the set of generators down to {σ, τ }.
66
 F IGURE 28. Johann Peter Gustav Lejeune Dirichlet (1805–59) was a German num-
ber theorist and analyst. In 1837, he proved that every arithmetic progression of
integers contains infinitely many primes. He provided fundamental contributions
to the theory of Fourier series, and is credited with the modern definition of a func-
tion (“to any x there corresponds a single y”).

29. D IRICHLET DOMAINS

We now present Dirichlet’s method for constructing fundamental domains. Again, Γ ≤ PGL2 (R)
is discrete throughout. Recall that if K ⊆ H2 is compact, then σ(K) ∩ K = ∅ for only finitely many
σ ∈ Γ. This allows us to pick w ∈ H2 such that Γw = {e}, that is, if σ ∈ Γ satisfies σ(w) = w, then
σ = e. Indeed, one can use proper discontinuity of a continuous group action on a locally com-
pact set to prove that the collection of points with nontrivial stabilizer is discrete (moral exercise).
Regardless, fix such a w.

Given σ ∈ Γ, let Lσ (w) denote the perpendicular bisector of [w, σ(w)], and let Hσ (W ) denote
the component of the complement of H2 r Lσ (w) containing w. The Dirichlet domain with center w
is \
P (w) = Hσ (w).
σ∈Γr{e}

We may think of P (w) as the collection of points in H2 that are closer to w than they are to any
Γ-translate of w. In other words, P (w) is the Voronoi cell of w in the orbit Γw.
Theorem 29.1. For w ∈ H2 such that Γw = {e}, the Dirichlet domain P (w) is a convex locally finite
fundamental domain for Γ in H2 .
Proof. We first check that P (w) is open. While each Hσ (w) is open, it is not immediate that P (w)
is open since arbitrary intersections of open sets need not be open. Observe, though, that for any
r > 0 and D(w; r) the open disk of radius r centered at w,
\
P (w) ∩ D(w; r) = Hσ (w)
σ∈S

for S = {σ ∈ Γ r {e} | σ(w) ∈ D(w; 2r)}. Since S is finite, each P (w) ∩ D(w; r) is open, and it
follows that P )(w) is open.
67
 We now check that P (w) is a fundamental domain. For γ ∈ Γ, the reader may verify that
γP (w) = P (γw), so P (w) ∩ γP (w) = ∅ for γ ∈ Γ r {e}. To show that the closure P (w) of P (w)
meets every Γ-orbit, first note that
\
(29.2) P (w) = H σ (w).16
σ∈Γr{e}

Given a Γ-orbit O, pick z ∈ O with shortest possible distance from w. Then

d(z, w) ≤ d(σ −1 (z), w) = d(z, σ(w))
for σ ∈ Γ r {e}. Thus z ∈ H σ (w) for all such σ, so z ∈ P (w) by (29.2).
Convexity of P (w) is clear since it is the intersection of half-spaces. For local finitude, see [Ive92,
Proposition V.4.4]. 

We have already seen a Dirichlet domain, namely the modular domain M for Γ = PSL2 (Z) and
w = yi, y > 1. This follows from [Ive92, Lemma V.4.5], which says that if S ⊆ Γ r {e} and
\
PS (w) = Hσ (w)
σ∈S

meets each Γ-orbit

 in atmost
 one point,
 then
PS (w) = P (w). The set M was constructed in this
0 1 1 1 1 −1
fashion for S = , , .
1 0 0 1 0 1

30. C OMPACT POLYGONS

We will skip Chapter VI of [Ive92], moving straight to the discussion of Poincaré’s theorem in
Chapter VII. The missing content (such as the monodromy theorem) will be mentioned en passant
as we develop this final theorem.
Via the Dirichlet domain construction, we have seen that every discrete subgroup Γ ≤ PGL2 (R)
has a convex locally finite fundamental domain P in H2 . When Γ is finitely generated, P has
finitely many sides and is a polygon in H2 equipped with a side pairing s 7→ ∗s. Our present
goal is to understand which hyperbolic polygons and side pairings give rise to discrete groups of
isometries.
Let ∆ be a hyperbolic polygon in H2 with side pairing s 7→ ∗s on the set of edges E of ∆. We
additionally require that s and ∗s have the same length and that ∗(s−1 ) = (∗s)−1 where s−1 is s
with the opposite orientation. For s ∈ E , there is a unique isometry σs mapping ∗s to s in such a
way that the half-plane bounded by ∗s containing ∆ is mapped to the half-plane bounded by s but
opposite ∆. The isometry σs is called the side transformation determined by s. It is straightforward
to observe that
(30.1) σs−1 = σs and σ∗s = σs−1
for s ∈ E .
Given an edge s of ∆, let ↓ s denote the edge with the same initial vertex as s but different
terminal vertex. Define the edge operator
Ψ : E −→ E
s 7−→↓ ∗s.

16The inclusion ⊆ is immediate since intersections of closed sets are closed. For the opposite inclusion, consider
z ∈ H2 belonging to H σ (w) for all σ ∈ Γ r {e}. Then [w, z] ∈ H σ (w) for all σ, whence [w, z) is in P (w), so z ∈ P (w).
68
The edge operator is a bijection (both ∗ and ↓ are involutions) and E is finite, so Ψ has finite order.17
The cycles for Ψ on E are called edge cycles. The sequence of initial points of an edge cycle is called
a vertex cycle. (Note that vertices may repeat in a vertex cycle.)
Lemma 30.2. Let s1 , . . . , sr be an edge cycle with vertex cycle P1 , . . . , Pr and side transformations
σi = σsi for 1 ≤ i ≤ r. The cycle map σ = σ1 · · · σr is a rotation around P1 of angle congruent
modulo 2π to the sum of the interior angles
∠int P1 + ∠int P2 + · · · + ∠int Pr .
Proof. Choose an orientation of H2 such that ∆ is on the positive side of s1 near P1 . Then
∠or (si , ↓ si ) ≡ sign(si )∠int Pi (mod 2π)
for 1 ≤ i ≤ r where sign(si ) is +1 (resp. −1) if ∆ lies on the positive (resp. negative) side of si .
By the construction of the side transformations, sign(sm ) = sign(sm+1 ) det(σm ) for 1 ≤ m ≤ r.
Multiplying these equations together (and canceling sign(s2 ) · · · sign(sr )) we get
sign(s1 ) = det(σ1 · · · σr ) sign(sr+1 ).
Since sr+1 = s1 , we conclude that the cycle map σ1 · · · σr is an even isometry with fixed point P1 .
We now claim that
Xr
∠int Pi ≡ sign(sm )∠or (sm , σm · · · σr (s1 ))
i=m
for 1 ≤ m ≤ r. To prove as much, observe that si = σi (∗si ), si+1 =↓ ∗si , and ↓ si = σi (si+1 ). We
now proceed by downward induction on m. When m = r, we have
∠int Pr = sign(sr )∠or (sr , ↓ sr ) = sign(sr )∠or (sr , σr (s1 ))
as desired. To descend from the (m + 1)-th case to the m-th, observe that
r
X
∠int Pi ≡ sign(sm )∠or (sm , ↓ sm ) + sign(sm+1 )∠or(sm+1 , σm+1 · · · σr (s1 ))
i=m
≡ sign(sm )∠or (sm , ↓ sm ) + sign(sm+1 ) det(σm )∠or (↓ sm , σm · · · σr (s1 ))
≡ sign(sm )∠or (sm , ↓ sm ) + sign(sm )∠or (↓ sm , σm · · · σr (s1 ))
≡ sign(sm )∠or (sm , σm · · · σr (s1 )).
The m = 1 case along with sign(s1 ) = 1 gives the desired formula for the rotation angle. 
Example 30.3. Consider the hyperbolic octagon with side pairings pictured in Figure 29. Write
α = σa , β = σb , γ = σc , and δ = σd . The edge cycle starting with a is
a(∗b−1 )(∗a−1 )bc(∗d−1 )(∗c−1 )d
with corresponding vertex cycle
P1 P4 P3 P2 P5 P8 P7 P6
and cycle map
αβ −1 α−1 βγδ −1 γ −1 δ.

17This means there is a positive integer n such that the n-fold composition Ψn = id; the smallest such n is the order
of Ψ.
69
 F IGURE 29. A hyperbolic octagon with indicated side pairings results in a genus 2
surface. Sources: [Ive92] and http://web1.kcn.jp/hp28ah77/us33_revo.
htm.

31. P OINCARÉ ’ S THEOREM

We now introduce the cycle condition and prove Poincaré’s theorem on which convex polygons
and side pairings generate discrete subgroups of PGL2 (R).
Definition 31.1. Fix a compact convex hyperbolic polygon ∆ with side pairing s 7→ ∗s. We say
that an edge cycle s1 , . . . , sr for ∆ with vertex cycle P1 , . . . , Pr satisfies the cycle condition if the
interior angle sum at the Pi takes the form
r
X 2π
∠int Pi =
nc
i=1
for some positive integer nc . If all edge cycles of ∆ satisfy the cycle condition, then we say that its
side pairing satisfies the cycle condition.
By Lemma 30.2, the cycle map σ = σ1 · · · σr of an edge cycle satisfying the cycle condition is a
rotation about P1 of order nc .
Theorem 31.2 (Poincaré’s theorem). Let ∆ be a compact convex hyperbolic polygon with a side pairing
satisfying the cycle condition. Then the group Γ generated by the side transformations S = {σs | s ∈ E } is
a discrete subgroup of PGL2 (R) with ∆ as a fundamental domain. A presentation for Γ is given by hS | Ri
where R consists of the side relations (30.1) and the cycle relations
(31.3) σcnc = e
for each edge cycle c.
Example 31.4. Before diving into the proof of Theorem 31.2, let’s return to Example 30.3 and
consider the consequences. Suppose that the octagon presented there is regular with interior angle
π/4.18 Then there is a unique edge cycle with cycle map which is a rotation by angle 8 · π/4 = 2π,
so nc = 1 and σc = e. By Poincaré’s theorem, the group generated by the side transformations of
such an octagon is a discrete subgroup of PGL2 (R) with presentation
α, β, γ, δ | αβ −1 α−1 βγδ −1 γ −1 δ .



18Moral exercise: Construct such an octagon.

70
If the octagon is regular with interior angle π/(4n) for some positive integer n, then the group
generated by the side transformations is again discrete but now with presentation
α, β, γ, δ | (αβ −1 α−1 βγδ −1 γ −1 δ)n .



Proof sketch for Theorem 31.2. Let G = hS | Ri. By Equation 30.1 and Lemma 30.2 we get a canonical
homomorphism G → Γ denoted g 7→ g̃. Give G the discrete topology and define a topological
space
X = G × ∆/ ∼
where ∼ is the equivalence relation generated by
(gσs , p) ∼ (g, σs (p))

S g ∈ G, s ∈ E , and p ∈ ∗s. Here G × ∆ has the product topology (open sets are of the form
for
g∈G {g} × Ug where Ug is an open subset of ∆) and X has the quotient topology (open subsets
correspond to open, ∼-stable subset of G × ∆). Write g · p ∈ X for the equivalence class of
(g, p) ∈ G × ∆.
The evaluation map G × ∆ → H2 , (g, p) 7→ g̃(p) is continuous and compatible with ∼, so it
induces a continuous map f : X → H2 where f (g · p) = g̃(p). Note that the action of G on the
left-hand factor of G × ∆ induces an action of G on X (where g(h · p) = (gh) · p for g, h ∈ G and
p ∈ X), and the map f : X → H2 is equivariant in the sense that:
f (gx) = g̃f (x)
for g ∈ G and x ∈ X.
At this point, we see that we have a “nice” map f : X → H2 . The proof proceeds by showing
that f is in fact “very nice”: X is a complete hyperbolic surface, and f is a local isometry. One then
applies the monodromy theorem (our sin of omission, stated below as Theorem 31.7) to conclude
that f is a bijection. Since f is surjective (and clearly G → Γ is surjective), we know that
[
H2 = γ(∆).
γ∈Γ

By the definition of ∼, we have g · ∆ ∩ e · ∆◦

= ∅ for g ∈ G r {e}. Since f is injective, it transforms
disjoint sets into disjoint sets, and we learn that g̃(∆) ∩ ∆◦ = ∅ for g ∈ G r {e}. This implies that
ker(G → Γ) = {e}. (Indeed, if g̃ = e for some g ∈ G r {e}, then we would not have g̃(∆) and ∆◦
disjoint!) Thus G → Γ is an isomorphism and ∆ is a fundamental domain for Γ. 
In order to state the monodromy theorem, we need two definitions.
Definition 31.5. A hyperbolic surface is a metric space X with the shortest length property,19 for
which every point has an open neighborhood isometric to an open disk in H2 . A hyperbolic
surface is complete if every Cauchy sequence in the surface converges (in the surface).
Definition 31.6. A local isometry f : X → Y between metric spaces is a function such that every
point in x ∈ X has an open neighborhood which is mapped isometrically onto an open neighbor-
hood of f (x).
Theorem 31.7 (Monodromy theorem, see [Ive92, VI.4.8]). Let X be a complete hyperbolic surface. Any
local isometry X → H2 is a bijection.
The monodromy theorem has the following important corollary, which classifies complete hy-
perbolic surfaces.

19This property says that any two points x, y are joined by a geodesic of length d(x, y). See [Ive92, Definition VI.1.2]
71
Corollary 31.8 ([Ive92, Theorem VI.6.1]). Any complete hyperbolic surface X is isometric to a sur-
face of the form H2 /Γ where Γ is a torsion free discrete subgroup of PGL2 (R). Two such subgroups
Γ and Σ define isometric surfaces H2 /Γ and H2 /Σ if and only if Γ and Σ are conjugate subgroups
of PGL2 (R).
Example 31.9. We conclude with one more example of Poincaré’s theorem in action. Let 2ABCD
be a convex quadrilateral in H2 where opposite sides have equal length. In your homework, you
proved that the diagonals AC and BD have a common midpoint M and a half-turn with respect
to M take the parallelogram to itself. We also know that the interior angle sum of 2ABCD is less
than 2π. Let us assume that this sum takes the form
2π
∠A + ∠B + ∠C + ∠D =
n
for some integer n ≥ 2. Applying Theorem 31.2 to the side pairing ∗AD = BC and ∗AB = DC
gives edge cycle
ab ∗ a−1 ∗ b−1
and cycle map
αβα−1 β −1
for a = AD, b = BA, α = σa , and β = σb . Thus the parallelogram group Γ is isomorphic to
α, β | (αβα−1 β −1 )n .



32. T RIANGLE GROUPS

32.1. Triangles. We can use Poincaré’s Theorem 31.2 to identify the discrete groups of isometries
generated by hyperbolic triangles; these are the so-called triangle groups. Let 4ABC be a triangle
in H2 . Let α denote reflection in BC, β reflection in CA, and γ reflection in AB. We assume that
the interior angle measures of the triangle are
π π π
∠A = , ∠B = , ∠C =
p q r
for p, q, r positive integers such that
1 1 1
+ + < 1.
p q r
(The inequality guarantees that the total angle sum is less then π.) It turns out that for any three
such positive integers, such a triangle exists and and is unique up to isometry. The subgroup of
Isom(H2 ) generated by α, β, γ is denoted ∆(p, q, r) and is called the (p, q, r) triangle group.
The isometry γβ is a rotation around A by angle 2π/p, and similar remarks apply to vertices B
and C. We thus have reflection and cycle relations amongst α, β, γ given by
e = α2 = β 2 = γ 2
and
e = (γβ)p = (αγ)q = (αβ)r .
We can apply Poincaré’s theorem to the trivial pairing ∗a = a, ∗b = b, ∗c = c with edge cycles
cb−1 , ac−1 , a−1 b. This realizes the above relations, so
∆(p, q, r) ∼
= α, β, γ | α2 , β 2 , γ 2 , (γβ)p , (αγ)q , (αβ)r



Now consider the even part of ∆(p, q, r),

D(p, q, r) := ∆(p, q, r) ∩ Isom+ (H2 ).
72
 F IGURE 30. Walther von Dyck (1856–1934) was a German mathematician and stu-
dent of Felix Klein. He is credited with the modern definition of a mathematical
group.

This is called the (p, q, r) von Dyck group. The reader may check that D(p, q, r) is generated by
ρ2A = βγ, ρ2B = γα, and ρ2C = αβ and we may derive the relations
ρ2A ρ2B ρ2C = e = ρp2A = ρq2B = ρr2C .
One can show that these are a complete list of relations for D(p, q, r) by considering the quadrilat-
eral 2ABCD where D is the reflection of B across the geodesic AC. See [Ive92, p.210] for details.

32.2. Hyperbolic pants. Fix an orientation of the hyperbolic plane. Given a hyperbolic n-gon
∆, equip it with an oriented boundary cycle of edges a1 , . . . , an (an+1 = a1 ) where the terminal
vertex of ai matches the initial vertex of ai+1 and locally ∆ is on the positive side of the edges of
the boundary.
Let R be a rectangular hexagon in H2 with boundary cycle a1 , . . . , a6 and side pairing ∗ai = ai .
By Poincaré’s theorem, the reflections ρ1 , . . . , ρ6 in the sides of R generate a discrete group Π with
fundamental domain R. The complete set of relations amongst the ρi is
ρ2i = e andρi ρi+1 = ρi+1 ρi
for i = 1, . . . , 6 and ρ7 = ρ1 .
Now consider a second copy of R and sew it to the first copy along a2 , a4 , a6 as in Figure 31.
This results in a pair of hyperbolic pants which are the orbit space of the rectangular octagon ∆ =
R ∪ ρ4 (R) with side pairing indicated in the picture. The boundary cycle is abc ∗ b−1 def ∗ e−1 and
the side transformations are σb = ρ2 ρ4 , σe = ρ4 ρ6 . Applying Poincaré’s theorem, we get that the
group Γ generated by
ρ1 , ρ3 , ρ5 , ρ64 , ρ42 , ρ26
(where ρij = ρi ρj ) is a discrete subgroup of index 2 in Π with fundamental domain ∆. The cycle
and reflection relations may be written in the form
ρ21 = ρ23 = ρ25 = ρ64 ρ42 ρ26 = e
ρ1 ρ26 = ρ26 ρ1 , ρ3 ρ42 = ρ42 ρ3 , ρ5 ρ64 = ρ64 ρ5 .

73
 F IGURE 31. Hyperbolic pants assembled from two rectangular hexagons. Source:
[Ive92, p.211].

F IGURE 32. A fundamental domain of Γ(7) in the Poincaré disk. Source: Tony
Smith (via John Baez).

While we will not rehearse the details, our work with rectangular hexagons implies that hyper-
bolic surfaces homeomorphic to a pair of pants are uniquely determined (up to isometry) by the
lengths `1 , `2 , `3 ∈ R>0 of the the boundary components (cuffs and waist). This is important in the
Fenchel–Nielsen coordinatization of Teichmüller space, but we will not pursue this further.

33. T HE K LEIN QUARTIC

The Klein quartic is the surface H2 /Γ(7) where Γ(7) is the kernel of the reduction map PSL2 (Z) →
PSL2 (F7 ). The fundamental domain of Γ(7) is tiled by 24 heptagons, each of which is barycen-
trically subdivided into (2, 3, 7) triangles. These regions and the side pairing induced by Γ(7)
are indicated in Figure 32. The resulting surface H2 /Γ(7) has genus 3 and automorphism group
PSL2 (F7 ), which has order 168. It turns out that this is the maximal symmetry group of a genus 3
surface, making the Klein quartic a Hurwitz surface.

74
 F IGURE 33. A fundamental domain of Γ(7) in the upper half-plane produced by Sage.

John Baez maintains a delightful website devoted to the Klein quartic:

http://math.ucr.edu/home/baez/klein.html.
Please read it in preparation for our final class discussion.

R EFERENCES
[Con] K. Conrad. Decomposing SL2 (R). Accessed March 2020 at https://kconrad.math.uconn.edu/
blurbs/grouptheory/SL(2,R).pdf.
[Ive92] B. Iversen. Hyperbolic geometry, volume 25 of London Mathematical Society Student Texts. Cambridge University
Press, Cambridge, 1992.
[Man15] K. Mann. DIY hyperbolic geometry. Accessed January 2020 at https://math.berkeley.edu/~kpmann/
DIYhyp.pdf, 2015.
[Thu97] W.P. Thurston. Three-dimensional geometry and topology. Vol. 1, volume 35 of Princeton Mathematical Series.
Princeton University Press, Princeton, NJ, 1997. Edited by Silvio Levy.

75