Geometry III/IV

Anna Felikson
Durham University, 2020-2021

Dedicated to the memory of Ernest Borisovich Vinberg

Contents
0 Introduction and History 3
0.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
0.2 Axiomatic approach to geometry . . . . . . . . . . . . . . . . . . . . . . . . . 6
0.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1 Euclidean Geometry 10
1.1 Isometry group of Euclidean plane, I som(E2 ). . . . . . . . . . . . . . . . . . . 10
1.2 Isometries and orthogonal transformations . . . . . . . . . . . . . . . . . . . . 14
1.3 Discrete groups of isometries acting on E2 . . . . . . . . . . . . . . . . . . . . 17
1.4 3-dimensional Euclidean geometry . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Spherical geometry 24
2.1 Metric on S 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2 Geodesics on S 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Polar correspondence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Congruence of spherical triangles . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5 Sine and cosine rules for the sphere . . . . . . . . . . . . . . . . . . . . . . . . 31
2.6 More about triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.7 Area of a spherical triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.8 Isometries of the sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.9 Platonic solids and their symmetry groups (NE) . . . . . . . . . . . . . . . . . 39
2.10 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3 Affine geometry 42
3.1 Similarity group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Affine geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4 Projective geometry 49
4.1 Projective line, RP1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2 Projective plane, RP2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3 Some classical theorems on RP2 . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.4 Topology and metric on RP2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.5 Polarity on RP2 (NE) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.6 Hyperbolic geometry: Klein model . . . . . . . . . . . . . . . . . . . . . . . . 63
4.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

1
5 Möbius geometry 72
5.1 Group of Möbius transformations . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2 Types of Möbius transformations . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.3 Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.4 Möbius transformations and cross-ratios . . . . . . . . . . . . . . . . . . . . . 80
5.5 Inversion in space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.6 Stereographic projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

6 Hyperbolic geometry: conformal models 87

6.1 Poincaré disc model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2 Upper half-plane model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.3 Elementary hyperbolic geometry . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.4 Area of hyperbolic triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

7 Other models of hyperbolic geometry 108

7.1 Klein disc, revised . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7.2 The model in two-sheet hyperboloid . . . . . . . . . . . . . . . . . . . . . . . 110
7.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

8 Classification of isometries of H2 115

8.1 Reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.2 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
8.3 Horocycles and Equidistant curves. . . . . . . . . . . . . . . . . . . . . . . . . 119
8.4 Discrete reflection groups in S 2 , E2 and H2 (NE) . . . . . . . . . . . . . . . . 122
8.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

9 Geometry in modern maths - some topics (NE) 126

9.1 Taming infinity via horocycles . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
9.2 Three metric geometries: S 2 , E2 , H2 , unified . . . . . . . . . . . . . . . . . . . 128
9.3 Discrete groups of isometries of H2 : Examples . . . . . . . . . . . . . . . . . . 131
9.4 Hyperbolic surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
9.5 Review via 3D hyperbolic space . . . . . . . . . . . . . . . . . . . . . . . . . . 139
9.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

10 Bibliography and links 146

Here “(NE)” marks non-examinable sections.

2
0 Introduction and History
0.1 Introduction
What to expect or 8 reasons to expect difficulties.

Our brain has two halves: one is responsible for multiplication of polyno-
mials and languages, and the other half is responsible for orientation of
figures in space and the things important in real life.
Mathematics is geometry when you have to use both halves.
Vladimir Arnold

Geometry is an art of reasoning well from badly drawn diagrams.

Henri Poincaré

1. Structure of the course:

• It will be a zoo of different 2-dimensional geometries - including Euclidean, affine,

projective, spherical and Möbius geometries, all of which will appear as some
aspects of hyperbolic geometry.
Why to study all of them?
- They are beautiful!
- we will need all of them to study hyperbolic geometry.
Why to study hyperbolic geometry?
- Important in topology and physics, for example.
Example. When one looks at geometric structures on 2-dimensional closed
surfaces, one can find out that only the sphere and torus carry spherical and
Euclidean geometries on them, and infinitely many other surfaces (all other closed
surfaces) are hyperbolic (see Fig. 1).
(Given the time there will be more on that at the very end of the second term).

spherical Euclidean hyperbolic

Figure 1: Geometric structures on surfaces.

• There will be just a bit on each geometry, hence the material may seem too easy.

• But it will get too difficult if you will miss something (as we are going to use
extensively almost everything...)

3
2. Two ways of doing geometry: “synthetic” and “analytic”

• “Synthetic” way:
- List axioms and definitions.
- Then formally derive theorems.
Question: is there any object satisfying the axioms?
- Build a “model”: an object satisfying the axioms (and hence, theorems).

• “Analytic” way:
- Build a model
- Work in the model to prove theorems (using properties of the model).

The same object may have many different models.

Example 0.1. A group G2 = {e, r} = ⟨r|r2 = e⟩

(Group with 2 elements, e, r, with one generator r and relation r2 = e).
Model 1: Let r be a reflection on R2 (and e an identity map).
Model 2: {1, −1} ∈ Z with respect to multiplication.

We will sometimes use different models for the same geometry - to see different aspect
of that geometry.

3. “Geometric” way of thinking:

Example 0.2. Claim. Let AB C be a triangle, let M and N be the midpoints of AB

and B C . Then AC = 2M N .

We will prove the claim in two ways: geometrically and in coordinates. Geometric
proof will be based on Theorem0.3.
Notation: given lines l and m, we write l||m when l is parallel to m.
BA BC
Theorem 0.3. If AB C is a triangle, M ∈ AB , N ∈ B C , then M N ||AC ⇔ BM
= BN
.

Proof. (Geometric proof):

1) M N ||AC (by Theorem 0.3).
2) Draw N K ||AM , K ∈ AC (see Fig. 2, left).
3) Then AK = K C (by Theorem 0.3).
4) The quadrilateral AM N K is a parallelogram
(by definition of a parallelogram - as it has two pairs of parallel sides).
5) Hence, M N = AK (by a property of a parallelogram).
6) AC = 2AK = 2M N (by steps 3 and 5).

Proof. (Computation in coordinates):

We can assume that A = (0, 0), C = (x, 0), B = (z , t) (see Fig. 2, right).
Then M = ( z2 , 2t ), N = ( x+2 z , z2 ).
Therefore, M N 2 = ( x+2 z − z2 )2 + ( 2t − 2t )2 = ( x2 )2 . Hence, M N = x/2, while AC = x.

4
 B B = (z, t)

M N M = ( z2 , 2t ) N = ( x+z
2
, 2t )

A K C A = (0, 0) C = (x, 0)

Figure 2: Two proofs of the theorem about midline.

Note that even in the second proof we used geometry to simplify the computation:
we assumed that A = (0, 0), i.e. that all points of the plane are equally good, and
that after taking A to the origin we can rotate the whole picture so that C get to the
horizontal line.

4. We will use some results from Euclidean geometry without reproving.

- We need some basics.
- The complete way from axioms takes time.
- It is not difficult (was previously taught in schools).
- You can find proofs in books (will give some references).
- Hopefully, by now you have already mastered logical/mathematical thinking (and
don’t need the course on Euclidean geometry as a model for mastering them).

5. We will use many diagrams:

- They are useful
- but be careful: wrong diagrams may lead to mistakes.

Example 0.4. “Proof” that all triangles are isosceles (with demystification):
[Link]

6. Problem solving in Geometry

- Is not algorithmic (one needs practice!)
- Solution may be easy – but how to find it?
(additional constructions? which model to use? which coordinates to choose? ...
- all needs practice!)
For getting the practice we will have Problem Classes and Assignments:
- weekly sets of assignments;
- some questions will be starred - to submit for marking fortnightly (via Gradescope).
- other questions - to solve!
- There will be hints - use them if you absolutely don’t know how to start the question
without them (it is much better to attempt the questions with hints than just to read
the solutions).

5
7. “Examples” will be hard to tell from “Theory”:
“Problem”=”one more theorem”
“Proof of a Theorem”=”Example on problem solving”.
8. Group approach to geometry
Klein’s Erlangen Program: In 1872, Felix Klein proposed the following:
each geometry is a set with a transformation group acting on it.
To study geometry is the same as to study the properties preserved by the group.
Example 0.5. Isometries preserve distance;
Affine transformations preserve parallelism;
Projective transformations preserve collinearity;
Möbius transformations preserve property to lie on the same circle or line.

Why to speak about possible difficulties now?

- not with the aim to frighten you but
- to make sure you are aware of them;
- to inform you that they are in the nature of the subject;
- to inform you that I know about the difficulties- and will try my best to help;
- to motivate you to ask questions.
Remark. For seven top reasons to enjoy geometry check Chapter 0 here:
[Link] kisilv/courses/[Link]

0.2 Axiomatic approach to geometry

Ptolemy I: Is there any shorter way than one of Elements?
Euclid: There is no royal road to geometry.
Proclus
“One must be able to say at all times instead of points, straight lines and
planes - tables, chairs and beer mugs.”
David Hilbert
Geometry in Greek: γ ϵω µϵτ ρια, i.e. measure of land (“geo”=land, “metry”=measure).
Brief History:
• Origin: Ancient Egypt ≈ 3000 BC
(measuring land, building pyramids, astronomy).
• First records: Mesopotamia, Egypt ≈ 2000 BC.
Example: Babylonians did know Pythagorean theorem
at least 1000 years before Pythagoras.
• Greek philosophy brought people to the idea
that geometric statements should be deductively proved.
• Euclid (≈ 300 BC) realised that the chain of proofs cannot be endless:
A holds because of B (Why B holds?)
B holds because of C (Why C ?)
C ....
To break this infinite chain ... ⇒ C ⇒ B ⇒ A we need to
- Accept some statements as axioms without justification;
- Agree on the rules of logic.

6
Euclid’s Postulates:
1. For every point A and for every point B not equal to A there exists a unique line
that passes through A and B .
2. For every segment AB and for every segment C D there exists a unique point
E such that B is between A and E and such that segment C D is congruent to
segment B E .
3. For every point O and every point A not equal to O, there exists a circle with
centre O and radius OA.
4. All right angles are congruent to each other.
5. (Euclid’s Parallel Postulate) For every line l and for every point P that does not
lie on l, there exists a unique line m passing through P that is parallel to l.

In “Elements” Euclid derives all known by that time statements of geometry and number
theory from these five postulates.

Hilbert’s axioms
By XIXth century in is clear that Euclid’s axioms are not sufficient: Euclid still used
some implicit assumptions.
Example 0.6 (Euclid’s Theorem 1). : There exists an equilateral triangle with a given
side AB .
Euclid’s proof:
- Draw a circle CA centred at A of radius AB (see Fig. 5).
- Draw a circle CB centred at B of radius AB .
- Take their intersection C = CA ∩ CB and show that △AB C is equilateral.
What is wrong with the proof: Why do we know that the circles do intersect?

A B

Figure 3: Euclid’s proof of existence of equilateral triangle.

This shows that we need to have more axioms. Hilbert has developed such a system
of axioms, which contains 5 groups of axioms (roughly corresponding to Euclid’s pos-
tulates). See handout for the list.
You don’t need to memorise - neither Euclid’s nor Hilbert’s axioms!

7
Example 0.7. Given a triangle AB C and a line l crossing the segment AB , can we
state that l we cross the boundary of AB C again on it’s way “out of the triangle”? See
Fig. 4.

B B

A C A C

Figure 4: Pasch’s Theorem and Crossbar Theorem.

If we want to derive this obvious fact from the axioms, we need to work quite a lot,
in particularly, using Betweenness Axiom BA4. It will go as follows.

Definition 0.8. Given a line l and points A, B ∈ / l we say that A and B are on
the same side of l if A = B or the segment AB does not intersects l. Otherwise, A
and B are on the opposite sides of l. We will denote these situations A, B |∗ and A|B
respectively (when it is clear which line is considered).

Axiom 0.9 (BA4, Plane separation). (a) A, C |∗ and B , C |∗ imply A, B |∗.

(b) A|C and B |C imply A, B |∗.

Remark 0.10. The Axiom BA4 guarantees that the geometry we get is 2-dimensional.

Theorem 0.11 (Pasch’s Theorem). Given a triangle AB C , line l, and points A, B , C ∈

/
l. If l intersects AB then l intersects either AC or B C .

Proof. (1) By Definition 0.8, we have A|B .

(2) Since C ∈/ l, BA4(a) implies that either A, C |∗ or B , C |∗.
(3) Suppose that A, C |∗. By BA4(a), this implies that C |B (otherwise we have A, B |∗
in contradiction to (1)). Therefore, l ∩ B C ̸= ∅ (by Definition 0.8).
(4) The case if B , C |∗ is considered similarly.

Remark 0.12. In the case, when l enters the triangle AB C through a vertex C one
can show that l intersects AB (this statement is called Crossbar Theorem and its proof
is more than twice longer).

Remarks
1. We will not work with axioms (neither in Euclidean geometry no in any other).

2. We appreciate this magnificent building of knowledge and use theorems of Eu-

clidean geometry when we need them.

8
 3. Some basic theorems are listed in the handout out Euclidean geometry (with
brief ideas of proofs and references, where available).
4. More detailed treatment of basics can be found in
M. J. Greenberg, Euclidean and Non-Euclidean Geometries, San Francisco: W.
H. Freeman, 2008.
5. Sometimes one can find many proofs of the same theorem.
For example, see [Link] for 122 proofs of
Pythagorean theorem.

What to do with the list of Theorems?

1. You don’t need to memorise!
(This is just an index for the references later on )
2. Read, understand and illustrate the statements (to be aware of them).
3. Do HW Question 1.1 (using the first survey on DUO).
Remark 0.13. Axiomatic approach is designed to eliminate geometry from geometry.
Now belongs to the history of mathematics. However, some elements of it still could
be useful as parts of school education as
• an example of logical arguing;
• a demonstration that even “evident” statements should be justified.
Example: “My opinion is the right one”.
Remark 0.14. Hilbert’s axiom system is shown to be
1. Consistent (i.e. there exists a model for it).
2. Independent (i.e. when removing any axiom one gets another set of theorems).
3. Complete (for any statement A in this language either holds “ A” or its negation
“not A”.

0.3 References
- A further discussion of Klein’s Erlangen Program can be found in Section 5 of
Nigel Hitchin, Projective Geometry, Lecture notes. Chapters 1, 2, 3, 4.
(See also “Other Resources” on DUO if you want to have all chapters in one pdf).
- Elementary exposition of most basic facts of Euclidean geometry can be found in
A. D. Gardiner, C. J. Bradley, Plane Euclidean Geometry, UKMT, Leeds 2012.
(The book is available from the library).
- Elementary but detailed exposition of basic facts of Euclidean geometry (and of
many other topic of the current module):
A. Petrunin, Euclidean plane and its relatives. A minimalist introduction.
- For the detailed treatment of axiomatic fundations of Euclidean geometry see
M. J. Greenberg, Euclidean and Non-Euclidean Geometries, San Francisco:
W. H. Freeman, 2008.
(The book is available from the library).
- Euclid’s "Elements", complete text with all proofs, with illustration in Geometry
Java applet, website by David E. Joyce.

9
1 Euclidean Geometry
1.1 Isometry group of Euclidean plane, I som(E2 ).
From now all, we will forget about axiomatic and will use some facts of Euclidean
geometry as “preknown”.
By Euclidean plane E2 we will understand R2 together with a distance function
d(A, B ) on it satisfying the following axioms M1-M3 of a metric:
Definition 1.1. A distance on a space X is a function
d : X × X → R, (A, B ) 7→ d(A, B ) for A, B ∈ X satisfying
M1. d(A, B ) ≥ 0 (d(A, B ) = 0 ⇔ A = B );
M2. d(A, B ) = d(B , A);
M3. d(A, C ) ≤ d(A, B ) + d(B , C ) (triangle inequality).

Remark. Triangle inequality appears in the list of Euclidean facts as E25. It was
proved using Cauchy-Schwarz inequality in Linear Algebra I, see also Section 1 of
G. Jones, Algebra and Geometry , Lecture notes,
which you can find in “Other Resources” on DUO.

We will use the following two models of Euclidean plane:

a Cartesian plane: {(x, y ) | x, y ∈ R} with the distance d(A1 , A2 ) = (x1 − x2 )2 + (y1 − y2 )2 ;
p

a Gaussian plane: {z | z ∈ C}, with the distance d(u, v ) = |u − v |.

Definition 1.2. An isometry of Euclidean plane E 2 is a distance-preserving transfor-

mation of E2 , i.e. a map f : E2 → E2 satisfying d(f (A), f (B )) = d(A, B ) for every
A, B ∈ E2 .
We will show that isometries of E2 form a group, but first we recall the definition.

Definition. A set G with operation · is a group if the following for properties hold:
1. (Closedness) ∀g1 , g2 ∈ G have g1 · g2 ∈ G;

2. (Associativity) ∀g1 , g2 , g3 ∈ G have (g1 · g2 ) · g3 = g1 · (g2 · g3 );

3. (Identity) ∃e ∈ G such that e · g = g · e = g for every g ∈ G;

4. (Inverse) ∀g ∈ G ∃g −1 ∈ G s.t. g · g −1 = g −1 · g = e.

Theorem 1.3. (a) Every isometry of E2 is a one-to-one map.

(b) A composition of any two isometries is an isometry.

(c) Isometries of E2 form a group (denoted I som(E2 )) with composition as a group

operation.

Proof. (a) Let f be an isometry. By M1, if f (A) = f (B ) then d(f (A), f (B )) = 0.

So, by definition of isometry, d(A, B ) = 0, which by M1 implies that A = B .
Hence, f is injective.

10
 Sketch of proof of surjectivity:
– Suppose X ∈
/ f (E2 ). Let y = f (A).
– Consider a circle CA (r) centred at A of radius r = d(X, Y ). Notice that
f (CA (r)) ⊂ Cy (r).
– Take B ∈ CA (r), consider f (B ) ∈ Cy (r).
– There are two points on CA (r) on any given distance smaller than 2r from
B . Hence, CA (r) contains two points on distance d(f (B ), X ). Therefore,
X ∈ f (CA (r)). The contradiction proves surjectivity, and (a) is done.

X
f

B A f (B) Y

Figure 5: To the proof of surjectivity of isometry.

(b) Given two isometries f and g , we need to check that the composition g ◦ f is an
isometry. Indeed,
g f
d(g (f (A), g (f (B )) = d(f (A), f (B )) = d(A, B ),
where the first (resp. second) equality holds since g (resp. f ) is an isometry.
(c) We need to prove 4 properties (axioms of a group):
1. Closedness is proved in (b).
2. Associativity follows from associativity of composition of maps.
3. Identity e := idE2 is the map defined by f (A) = A ∀A ∈ E2 . It clearly belongs
to the set of isometries.
4. Inverse element g −1 does exist as g is one-to-one (and it is an isometry).

Example 1.4. Examples of isometries of E2 :

• Translation Tt : a 7→ a + t;
• Rotation Rα,A about centre A by angle α.
On complex plane, Rα,0 writes as z 7→ eiα z ;
• Reflection rl in a line. Example: if the line l is the real line on C, then rl : z → z¯.
For a general formula of reflection: see HW 2.7.
• Glide reflection: given a vector a and a line l parallel to a, consider ta ◦ rl = rl ◦ ta .
Definition 1.5. Let AB C be a triangle labelled clockwise.
An isometry f is orientation-preserving if the triangle f (A)f (B )f (C ) is also labelled
clockwise. Otherwise, f is orientation-reversing.

11
Proposition 1.6 (Correctness of Definition 1.5). Definition 1.5 does not depend on
the choice of the triangle AB C .
Proof. Suppose that △AB C has the same orientation as f (AB C ). Take a point D on
the same side of the line AB as C . Then △AB D has the same orientation as f (AB D).
Hence, given the points A, B , Definition 1.5 does not depend on the choice of C .
Now we change points one by one moving from any triangle to any other as follows:
AB C → A′ B C → A′ B ′ C → A′ B ′ C ′ . (One should be a bit more careful here if some
triples of points are collinear, but then we just insert an extra step and may be change
the order. We skip the details here).

Example 1.7. Translation and rotation are orientation-preserving,

reflection and glide reflection are orientation-reversing.
Remark 1.8. Composition of two orientation-preserving isometries is orientation-
preserving;
composition of an or.-preserving isometry and an or.-reversing one is or.-reversing;
composition of two orientation-reversing isometries is orientation-preserving.
Proposition 1.9. Orientation-preserving isometries form a subgroup (denoted I som+ (E2 ))
of I som(E2 ).
Proof. We need to check the set I som+ (E2 ) forms a group, i.e. satisfies the four
properties of a group:
1. Closedness follows from Remark 1.8;
2,3. Associativity and Identity follow in the same way as in the proof of Theorem 1.3.
4. Inverse element: consider g ∈ I som+ (E2 ) and let g −1 ∈ I som(E2 ) be the inverse
in the big group. Suppose that g −1 is orientation-reversing. Then by Remark 1.8
g ◦ g −1 is also orientation-reversing, which contradicts to the assumption that g ◦
g −1 = e when considered in the whole group I som(E2 ). The contradiction shows
that g −1 is orientation-preserving, and hence I som+ (E2 ) contains the inverse
element.

Definition. Triangles △AB C and △A′ B ′ C ′ are congruent (write △AB C ∼

= △A′ B ′ C ′ )
if AB = A B , AC = A C , B C = B C and ∠AB C = ∠A B C , ∠B AC = ∠B ′ A′ C ′ ,
′ ′ ′ ′ ′ ′ ′ ′ ′

∠AC B = ∠A′ C ′ B ′ .

Theorem 1.10. Let AB C and A′ B ′ C ′ be two congruent triangles. Then there exists
a unique isometry sending A to A′ , B to B ′ and C to C ′ .
Proof. Existence:
1. Let f1 be any reflection sending A → A′ , A′ → A (if A ̸= A′ , f1 is unique and
given by reflection with respect to perpendicular bisector to AA′ , see Fig. 6, left;
if A = A′ we can take f1 = id, identity map).
2. Let f2 be a reflection s.t. f2 (A′ ) = A′ , f2 (f1 (B )) = B ′ . This f2 does exist: it
is given by reflection with respect to perpendicular bisector to B B ′ , see Fig. 6,
middle (denote the perpendicular bisector by l2 ). Notice that A′ ∈ l2 .
Exercise: Show that A′ ∈ l2 by using E14.

12
 3. We have A′ = f2 (f1 (A)), B ′ = f2 (f1 (B )).
If f2 (f1 (C )) and C ′ lie in the same half-plane with respect to A′ B ′ , then the
congruence △AB C ∼ = △A′ B ′ C ′ implies C ′ = f2 (f1 (C )): (indeed, in this case
triangles △A C f2 (f1 (C )) and △B ′ C ′ f2 (f1 (C )) are isosceles, so the heights of
′ ′

these triangles dropped from the points A′ and B ′ respectively are two different
perpendicular bisectors for the segment C ′ f2 (f1 (C )), which contradicts to E9, see
Fig. 6, right). So, f2 ◦ f1 maps AB C to A′ B ′ C ′
If f2 (f1 (C )) and C ′ lie in different half-plane with respect to A′ B ′ , apply f3 =
rA′ B ′ (reflection with respect to A′ B ′ ), then use the above reasoning to see that
f3 ◦ f2 ◦ f1 maps AB C and A′ B ′ C ′ .

Uniqueness: Suppose the contrary, i.e. there exist f , g ∈ I som(E2 ), f ̸= g such that
f : △AB C → △A′ B ′ C ′ and g : △AB C → △A′ B ′ C ′ . Then φ := f −1 ◦ g ̸= id and
φ(△AB C ) = △AB C . Choose D ∈ E2 : φ(D) ̸= D (it exists as φ is non-trivial!).
Then d(A, D) = d(A, φ(D)), d(B , D) = d(B , φ(D)), d(C, D) = d(C, φ(D)), which by
E14 means that all three points A, B , C lie on the perpendicular bisector to Dφ(D).
This contradicts to the assumption that AB C is a triangle.

C0

l1 A0 B0 C0 f2 (f1 (C))
0
f1 (A) = A

l2

A f1 (C) f1 (B) A0 B0

Figure 6: Isometry as a composition of reflections.

Corollary 1.11. Every isometry of E2 is a composition of at most 3 reflections.

(In particular, the group I som(E2 ) is generated by reflections).

Remark 1.12. The way to write an isometry as a composition of reflections is not unique.

Example 1.13. We can write rotation and translation as compositions of two reflec-
tions (see (a) and (b) below; a glide deflection can be written as a composition of three
reflection (see (c)).

(a) Let l1 ||l2 be two parallel lines on distance d. Then rl2 ◦ rl1 is a translation by 2d
along a line l perpendicular to l1 and l2 .

(b) Let 0 = l1 ∩ l2 be two lines intersecting at O. Let φ be angle between l1 and l2 .

Then rl2 ◦ rl1 is a rotation about O through angle 2α.

(c) Let l be a line, and a a vector parallel to l. To write the glide reflection ta ◦ rl ,
use (a): consider two lies l1 ||l2 orthogonal to l lying on the distance a/2 from
each other. Then by (a) ta = rl1 ◦ rl2 , so that ta ◦ rl = rl1 ◦ rl2 ◦ rl .

13
Theorem 1.14 (Classification of isometries of E2 ). Every non-trivial isometry of E2
is of one of the following four types: reflection, rotation, translation, glide reflection.

Proof. We can see from the proof of Theorem1.10 that every isometry of E2 is a com-
position of at most 3 reflections. Consider possible compositions:

0. Composition of 0 reflections is an identity map id.

1. Composition of 1 reflection is the reflection.

2. Composition of 2 reflections is either translation or rotation (see Example 1.13).

3. Composition of 3 reflections: one can prove that is a glide reflection (this is not
done in Example 1.13!), for the proof see HW 2.3.

Definition 1.15. Let f I som(E2 ). Then the set of fixed points of f is

F ixf = {x ∈ E2 | f (x) = x}.

Example 1.16. Fixed points of id, reflection, rotation, translation and glide reflection
are E2 , the line, a point, ∅, ∅ respectively.

Remark 1.17. Fixed points together with the property of preserving/reversing the
orientation uniquely determine the type of the isometry.

Proposition 1.18. Let f , g ∈ I som(E2 ).

(a) F ixgf g−1 = g F ixf ;

(b) g f g −1 is an isometry of the same type as f .

Proof. (a) We need to proof that g (x) ∈ F ixgf g−1 ⇔ x ∈ F ixf .

See HW 3.2 for the proof.

(b) Applying (a) we see that fixed points of f and g f g −1 are of the same type, also
they either both preserve the orientation or both reverse it. Hence, the isometries
f and g f g −1 are of the same type by Remark 1.17

1.2 Isometries and orthogonal transformations

a. Isometries preserving the origin O = (0, 0)
• From HW 2.7 we see, that a reflection preserving O is a linear map:

x → Ax A ∈ GL2 (R).

More precisely, if l is a line through O and a a vector normal to l (i.e. the line l
is given by equation (a, x) = 0, where (∗, ∗) is the dot product), then

(a,x)
rl (x) = x − (x,x)
a.
a

(a,x) = 0

14
 • Every isometry preserving O is a composition of at most 2 reflections (this fol-
lows from the proof of Theorem 1.10, or, alternatively, from the classification of
isometries). Hence, it is either an identity map, or a reflection or a rotation.

• So, if f ∈ I som(E2 ) and f (O) = O, then f (x) = Ax for some A ∈ GL2 (R).

Proposition 1.19. A linear map f : x → Ax, A ∈ GL(2, R) is an isometry if and only if

A ∈ O(2), orthogonal subgroup of GL(2, R) (i.e. iff AT A = I = AAT , where AT is A
transposed).

Proof. See HW 3.3.

b. General case
Let (b1 , b2 ) = f (O), denote b = (b1 , b2 ). Then t−b ◦ f (O) preserves O. So, in view of
Proposition 1.19, t−b ◦ f (x) = Ax for some A ∈ O2 (R), which implies that

f (x) = tb ◦ (Ax) = Ax + b.

Proposition 1.20. (a) Every isometry f of E2 may be written as f (x) = Ax + t.

(b) The linear part A does not depend on the choice of the origin.

Proof. (a) is already shown. (b) Move the origin to arbitrary other point u = (u1 , u2 )
and denote by y = x − u the new coordinates (see Fig. 7). Then

f (y ) = f (x) − u = Ax + b − u = A(y + u) + b − u = Ay + (Au + b − u).

x y f (x) = Ax + b

f (y)
O

O0 = u

Figure 7: Linear part of isometry: independence of the origin.

Example 1.21. Let A ∈ O2 (R) then det A = ±1.

  
−1 0 x
• Consider the reflection rx=0 with respect to the line x = 0: rx=0 = .
0 1 y
Clearly, in this case det A = −1.
 
cos α sin α
• Consider a rotation by angle α, RO,α = . In this case det A = 1.
− sin α cos α

15
Proposition 1.22. Let f (x) = Ax + t be an isometry.
f is orientation-preserving if det A = 1 and orientation-reversing if det A = −1.
Proof. First, notice that translation does not affect the orientation, so. we can assume
that f preserve the origin. An origin-preserving isometry is either identity, or reflection,
or rotation, and for all of them the statement holds.

Remark. Let l be a line through O forming angle α with the horizontal line x = 0.
Then rl = g −1 rx=0 g , where g = RO,−α (check this!). So,
det rl = det g −1 det rx=0 det g = −1.
Exercise 1.23. (a) Show that any two reflections are conjugate in I som(E2 ). (i.e.
that given any two reflections r1 and r2 there exists an isometry g ∈ I som(E2 )
such that r1 = g −1 r2 g ).
Hint. If l is a line not through the origin, then there exists a translation t such
that l′ = t(l) is a line through the origin and rl = t−1 rl′ t.
(b) Not all rotations are conjugate (only rotations by the same angle),
not all translations are conjugate (only the ones by the same distance)
and not all glide reflections are conjugate (only the ones with translational part
by the same distance).

Proposition 1.24. Let A, C ∈ l ∈ E2 . Then the line l gives the shortest path from A
to C .
Proof. Idea: approximate the path from A to C by a broken line AA1 A2 A3 . . . An−1 An C
and apply triangle inequality |AC | ≤ |AB | + |B C | repeatedly:
|AC | ≤ |AA1 | + |A1 C | ≤ |AA1 | + |A1 A2 | + |A2 C | ≤ · · · ≤ |AA1 | + ·|An C |,
with at least one inequality being strict if AA1 A2 A3 . . . An−1 An C ̸= AC.

A2
An
A1
C

Figure 8: A broken line approximating a path.

Analytically: given a path γ : [0, 1] → E2 with γ (0) = A = (0, 0) and γ (1) = C = (c, 0),
write
s s
Z 1  2  2 Z 1  2
C dx dy dx
l (γ ) = + dt ≥ dt
A 0 dt dt 0 dt
Z 1 Z 1
dx dx 1
= dt ≥ dt = x(t) 0 = x(1) − x(0) = b − 0 = d(A, B ).
0 dt 0 dt

16
1.3 Discrete groups of isometries acting on E2
Definition 1.25. A group acts on the set X (denoted G : X ) if
∀g ∈ G ∃fg , a bijection X → X , s.t. fgh (x) = (fg ◦ fh )(x), ∀x ∈ X, ∀g , h ∈ G.

Example 1.26. Here are some examples of group actions:

(a) Let G = ⟨ta ⟩ be a group generated by a translation ta . Every element of G can

be written as tka for some k ∈ Z. Clearly G : E2 with all elements of G acting as
translations tka = tka .

(b) I som(E2 ) acts on the set of all regular pentagons.

(c) (Z, +) : E2 in the following way:

Take any vector a, then n ∈ Z will act on E2 as the translation tna .

Definition 1.27. An action G : X is transitive if ∀x1 , x2 ∈ X ∃g ∈ G : fg (x1 ) = x2 .

Example 1.28. (a) The action of I som(E2 ) on the set of regular pentagons is not
transitive (it cannot take a small pentagon to a bigger one).

(b) Theorem 1.10 shows that I som(E2 ) acts transitively on the set of all triangles
congruent to the given one.

(c) I som(E2 ) acts transitively on points of E2 (this directly follows from (b)).

(d) The action of I som(E2 ) on lines is transitive (as for any two lines l1 and l2 there
is an isometry taking l1 to l2 .

(e) Theorem 1.10 also implies that I som(E2 ) acts transitively on flags in E2 , where
a flag is a triple (p, r, H + ) such that p ∈ E2 is a point, r is a ray from p and H +
is a half-plane bounded by the line containing r.

(f) I som(E2 ) does not act transitively on the circles or triangles.

Definition 1.29. LetSG : X be an action. An orbit of x0 ∈ X under the action G : X

is the set orb(x0 ) := g x0 .
g ∈G

Example 1.30. (a) The group O2 of isometries preserving the origin O acts on E2 .
For this action orb(O) = O (i.e. orbit of the origin is one point) and all other
orbits are circles centred at O (see Fig. 9, left).

(b) The group Z × Z acts on E2 by integer translations (a, b) (where a ∈ Z and b ∈ Z

are the first and the second components respectively). Then the orbit of any
point is a shift of the set of all integer points (see Fig. 9, right).

Definition 1.31. Let X be a metric space. An action G : X is discrete if none of its

orbits possesses accumulation points, i.e. given an orbit orb(x0 ), for every x ∈ X there
exists a disc Dx centred at x s.t. the intersection orb(x0 ) ∩ Dx contains at most finitely
many points.

Example 1.32. (a) Consider the action Z : E1 defined by gn x = 2n x for n ∈ Z. The

action is not discrete as orb(1) = {2n } and the sequence 1/2n converge to 0 ∈ E1 ,
see Fig. 10, left.

17
 Figure 9: Orbits of O2 (left) and Z2 = Z × Z (right)

(b) The action Z × Z acts on E2 by translations: let G = ⟨t1 , t2 ⟩, where t1 , t2 are

translations in non-collinear directions. This action is discrete as every orbit con-
sists of isolated points, see Fig. 9, right.

(c) (Reflection group). Given an isosceles right-angled triangle, one can generate a
group G by reflections in its three sides, G = ⟨r1 , r2 , r3 ⟩. Then G : E2 is a discrete
action.
To show that the action is discrete, consider a tiling of E2 by isosceles right-
angled triangles such that any adjacent tiles are reflection images of each other,
see Fig. 10, right. Then
- each of the three generators r1 , r2 , r3 preserves the triangular tiling;
- there are finitely many isometries taking a tile to itself (2 isometries here);
- hence, every tile contains only finitely many points of any given orbit;
- every ball intersects only finitely many tiles;
- which implies that every ball contains finitely many points of each orbit, i.e.
the group acts discretely.

1 1
0 22 2 1 2

Figure 10: A non-discrete action (left) and a discrete action (right).

18
Definition 1.33. An open connected set F ⊂ X is a fundamental domain for an action
G : X if the sets g F, g ∈ G satisfy the following conditions:

1) X = g F (where U denotes the closure of U in X );

S
g ∈G

2) ∀g ∈ G, g ̸= e, F ∩ g F = ∅;

3) There are only finitely many g ∈ G s.t. F ∩ g F ̸= ∅.

Examples of fundamental domains: any of the triangles in the tiling shown in Fig. 10
is a fundamental domain for the action described in Example 1.32(c).

Definition 1.34. An orbit space X/G for the discrete action G : X is a set of orbits
with a distance function

dX/G = min {dx (ˆ

x, yˆ)}.
ˆ∈orb(x), yˆ∈orb(y )
x

Example 1.35. (a) Z : E1 acts by translations, then an interval is a fundamental

domain. Identifying its endpoints we see that the orbit space E1 /Z is a circle.

(b) Z2 : E2 (generated by two non-collinear translations), then a parallelogram is a

fundamental domain of the action and the orbit space E2 /Z2 is a torus.

Figure 11: Fundamental domain for Z2 : E2 and a torus as an orbit space.

19
1.4 3-dimensional Euclidean geometry

Model: Cartesian space (x1 , x2 , x3 ), xi ∈ R, with distance function

3
X p
d(x, y ) = ( (xi − yi )2 )1/2 = ⟨x − y , x − y ⟩.
i=1

We will not list all the axioms but will mention some essential properties.

Properties:

1. For every plane α there exists a point A ∈ α and a point B ∈

/ α;

2. If two distinct planes α and β have a common point A then they intersect by a
line containing A.

3. Given two distinct lines l1 and l2 having a common point, there exists a unique
plane containing both l1 and l2 .

Example. Three flies are flying randomly in one room. Find the probability that they
are all in one plane at some given moment of time.

Proposition 1.36. For every triple of non-collinear points there exists a unique plane
through these points.

Proof. Let A, B , C be the three non-collinear points. The lines AB and AC have a
common point A. Therefore, there exists a plane α containing the lines AB and AC ,
and hence, containing all three points A, B , C .

Definition 1.37. Given a metric space X , a distance between two sets A, B ∈ X is

d(A, B ) := inf (d(a, b)).
a∈A,b∈B
In particular, the distance between a point A and a plane α is d(A, α) := min(d(A, X )).
X ∈α

A A A

X0 l
α l X1 X0 X1 X0

Figure 12: Distance between a point and a plane (see Proposition 1.38).

Proposition 1.38. Given a plane α, a point A ∈ / α and a point X0 ∈ α, AX0 = d(A, α)

if and only if AX0 ⊥ l for every l ∈ α, X0 ∈ l.

20
Proof. “ ⇒”: First, we prove that AX0 = d(A, α) implies that AX0 ⊥ l for every l ∈ α,
X0 ∈ l. Suppose that l ∈ α, X0 ∈ l and l is not orthogonal to AX0 , see Fig. 12,
in the middle. Then there exists X1 ∈ l such that d(X1 , A) < d(X0 , A) (indeed,
this is the case when X1 is the point such that AX1 ⊥ l).

“ ⇐”: Suppose that AX0 ⊥ l, but d(A, X0 ) ̸= d(A, α) = d(A, X1 ), see see Fig. 12, right.
As it is shown above, AX1 ⊥ X1 X0 . Then there are two distinct lines through A
perpendicular to l, in contradiction with E9.

Corollary. Given a plane α and a point A ∈

/ α, the closest to A point X0 ∈ α is
unique.

A
β

β nα
α α
nβ
C B
α

Figure 13: Angle between a line and a plane (left) and between two planes (right).

Definition 1.39. (a) The point X0 ∈ α s.t. d(A, α) = AX0 is called an orthogonal projection
of A to α. Notation: X0 = projα (A).

(b) Let α be a plane, AB be a line, B ∈ α, and C = projα (A). The angle between
the line AB and the plane α is ∠(AB , α) = ∠AB C , where C = projα (A),
(see Fig. 13, left).
Equivalently, ∠(AB , α) = min(∠AB X ).
X ∈α

Exercise: Check the equivalence. Hint: use cosine rule.

Remark. Definition 1.37 implies that if AC ⊥ α then AC ⊥ l for all l ∈ α, C ∈ l.

Definition 1.40. The angle ∠(α, β ) between two intersecting planes α and β is the
angle between their normals (see Fig. 13 middle and right).
Equivalently, if B ∈ β , A = projα (B ), C = projl (A) where l = α ∩ β ,
then ∠(α, β ) = ∠B C A.

21
Exercise:

1. Check the equivalence.

2. Let γ be a plane through B C A. Check that γ ⊥ α, γ ⊥ β .

3. Let α be a plane, C ∈ α. Let B be a point s.t. B C ⊥ α. Let β be a plane

through C , β ⊥ α. Then B ∈ β .

a u

b v2 k v +k v
1 1 2 2
A
c v1
α

Figure 14: To Proposition 1.41.

Proposition 1.41. Given two intersecting lines b and c in a plane α, A = b ∩ c, and

a line a, A ∈ a, if a ⊥ b and a ⊥ c then a ⊥ α (i.e. a ⊥ l for every l ∈ α).

Proof. Given three vectors u, v 1 , v 2 in R3 such that (u, v 1 ) = 0 and (u, v 2 ) = 0 we

have (u, k1 v 1 + k2 v 2 ) = 0 for any k, 1, k2 ∈ R.

B B D

l
A C A C
α

Figure 15: Theorem of three perpendiculars (and it’s proof).

Theorem 1.42 (Theorem of three perpendiculars). Let α be a plane, l ∈ α be a line

and B ∈
/ α, A ∈ α and C ∈ l be three points. If B A ⊥ α and AC ⊥ l then B C ⊥ l.

Proof. 1. Let C D be a line through C parallel to AB , see Fig. 15. Then C D ⊥ α (as
AB ⊥ α).
2. Then C D ⊥ l (as C D ⊥ l′ ∀l ⊂ α. Also, l ⊥ AC (by assumption).
3. Hence, by Proposition 1.41 l ⊥ (plane AC D), i.e. l ⊥ B C (as B C ⊂ plane AC D).

22
1.5 References
- A nice discussion of the group of isometries of Euclidean plane can be found in
G. Jones, Algebra and Geometry, Lecture notes (Section 1).
(The notes are available on DUO, see “Other Resources” section).

- Discussion of the geometric constructions and constructibility of various geomet-

ric objects can be found in
G. Jones, Algebra and Geometry, Lecture notes (Section 8).
(The notes are available on DUO, see “Other Resources” section).

- More detailed discussion of Euclidean isometries can be found here:

N. Peyerimhoff, Geometry III/IV, Lecture notes (Section 1).

- To read more about the role of reflections for I som(E 2 ), look at

O. Viro, Defining relations for reflections I, arXiv:1405.1460v1.

- The following book (Section 1) provides an introduction to group actions:

T. K. Carne, Geometry and groups.
Also, one can find here a detailed discussion of the group of Euclidean isometries
(Sections 2-4) - as well as many other topics.

- Another source concerning groups actions:

A. B Sossinsky, Geometries, Providence, RI : American Mathematical Soc. 2012.
One can find the book in the library, see also Section 1.3 (pp.9–11) here.
Section 2.7 (pp.26-27) of the same source introduces group presentations and
gives many examples.

- Exposition of 3-dimensional Euclidean Geometry can be found in Chapter 1 of

Kiselev’s Geometry, Book II. Stereometry. (Adopted from Russian by Alexader
Givental).
(The book is not easily reachable at the moment. You can find a reference to
Amazon on Giventhal’s homepage. I should probably order the book for our
library... Please, tell me if you are interested in this book).

- Webpages, etc:

- Cut-the-knot portal by Alexander Bogomolny.

- Drawing a Circle with a Framing Square and 2 Nails.

23
2 Spherical geometry
In this section we will study geometry on the surface of the sphere.

Model of the sphere S 2 in R3 : (sphere of radius R = 1 centred at O = (0, 0, 0))

S 2 = { (x1 , x2 , x3 ) ∈ R3 | x21 + x22 + x23 = 1}

O
1

Figure 16: Sphere.

Sometimes we will consider sphere of radius R: { (x1 , x2 , x3 ) ∈ R3 | x21 + x22 + x23 = R}.

2.1 Metric on S 2
Definition 2.1. • Points A and A′ of S 2 will be called antipodal if O ∈ AA′ .

• A great circle on S 2 is the intersection of S 2 with a plane passing though O, see

Fig. 17, left.

Remark 2.2. Given two distinct non-antipodal points A, B ∈ S 2 , there exists a unique
great circle through A and B (as there is a unique 2-dimensional plane through 3 non-
collinear points A, B , O).

B
O

Figure 17: Great circles and distance on the sphere.

24
Definition 2.3. Given a sphere S 2 of radius R, a distance d(A, B ) between the points
A, B ∈ S 2 is π R, if A is diametrically opposed to B , and the length of the shorter arc
of the great circle through A and B , otherwise.
Equivalently, d(A, B ) := ∠AOB · R (with R = 1 for the case of unit sphere).
See Fig. 17, right.
Theorem 2.4. The distance d(A, B ) turns S 2 into a metric space, i.e. the following
three properties hold:
M1. d(A, B ) ≥ 0 (d(A, B ) = 0 ⇔ A = B );
M2. d(A, B ) = d(B , A);
M3. d(A, C ) ≤ d(A, B ) + d(B , C ) (triangle inequality).
Proof. M1 and M2 hold by definition. To prove M 3 we need to show

∠AOC ≤ ∠AOB + ∠B OC.

We will do it in the following 8 steps.

1. If B lies on a great circle CAC through A and C , then M3 holds (may turn into
equality). Assume B ∈ / CAC .

2. Suppose that ∠AOC > ∠AOB + ∠B OC , in particular, ∠AOC > ∠AOB .

3. Choose B1 inside AC so that ∠AOB1 = ∠AOB , see Fig. 18.

Choose B2 ∈ OB so that OB2 = OB1 .
Then AB1 = AB2 (since △AB1 O is congruent to △AB2 O by SAS).

B1

A O

B2
B

Figure 18: To the proof of triangle inequality for S 2 .

4. Since ∠AOC > ∠AOB + ∠B OC we have ∠AOC > ∠AOB2 + ∠B2 OC .

Also, ∠AOC = ∠AOB1 + ∠B1 OC .
Hence, ∠B2 OC < ∠B1 OC .

5. Recall the Cosine Rule in E2 : c2 = a2 + b2 − 2ab cos γ .

Note that given the sides a, b, for a larger angle γ between them we get a larger
side c.

6. Applying results of steps 4 and 5 to △OB1 C and OB2 C , we get B2 C < B1 C .

3,6
7. AB2 + B2 C < AB1 + B1 C = AC ≤ AB2 + B2 C
(here the last inequality is the triangle inequality on the plane).

25
 8. The contradiction obtained in 7 shows that ∠AOC ≤ ∠AOB + ∠B OC (where
equality only holds when B lies in the plane AC O).

2.2 Geodesics on S 2
Definition 2.5. A curve γ in a metric space X is a geodesic if γ is locally the shortest
path between its points.
More precisely, γ (t) : (0, 1) → X is geodesic if

∀t0 ∈ (0, 1) ∃ε : l(γ (t)|tt00 + ε

−ε ) = d(γ (t0 − ε), γ (t0 + ε)).

Proposition 2.6. Geodesics on S 2 are great circles.

Proof. Use the (spherical) triangle inequality and repeat the proof of Proposition 1.24.

Definition 2.7. Given a metric space X , a geodesic γ : (−∞, ∞) → X is called closed

if ∃T ∈ R, T ̸= 0 : γ (t) = γ (t + T ) ∀t ∈ (−∞, ∞), and open, otherwise.

Example. In E2 , all geodesics are open, each segment is the shortest path.
In S 2 , all geodesics are closed, one of the two segments of γ \ {A, B } is the shortest
path (another one is not shortest if A and B are not antipodal).
HW 4.1: describes a metric space containing both closed and open geodesics.

From now on: by lines in S 2 we mean great circles.

Proposition 2.8. Every line on S 2 intersects every other line in exactly two antipodal
points.
Proof. Let l1 = α1 ∩ S and l2 = α2 ∩ S be two lines on S 2 , see Fig. 19, left. Then

l1 ∩ l2 = (α1 ∩ α2 ) ∩ S 2 = ( line through origin ) ∩ S 2 ,

as O ∈ α1 ∩ α2 .

α2
α2

α1

α1

Figure 19: Intersection and angle between two lines on the sphere.

26
Definition 2.9. By the angle between two lines we mean the angle between the cor-
responding planes:
if li = αi ∩ S 2 , i = 1, 2 then ∠(l1 , l2 ) := ∠(α1 , α2 ), see Fig. 19, right.

Equivalently, ∠(l1 , l2 ) is the angle between the lines ˆl1 and ˆl2 , ˆli ∈ R3 ,
where ˆli is tangent to the great circle li at l1 ∩ l2 as to a circle in R3 .
Proposition 2.10. For every line l and a point A ∈ l in this line there exists a unique
line l′ orthogonal to l and passing through A.
Proof. Consider the plane α ∈ R3 such that l = α ∩ S 2 . We need to find another line
l′ = β ∩ S 2 , where β ∈ R3 is a plane orthogonal to α and such that O, A ∈ β . Let vα
be the normal vector at O to α, see Fig. 21, left. Since β ⊥ α, we see that vα ∈ β . So,
β is the plane spanned by the line OA and vα . This construction shows both existence
of l′ and uniqueness.

β
vα
A
α
O A O B

Figure 20: Existence and uniqueness of a perpendicular line on the sphere.

Proposition 2.11. For every line l and a point A ∈ / l in this line, s.t. d(A, l) ̸= π /2
there exists a unique line l′ orthogonal to l and passing through A.
Proof. Let B ∈ α be the orthogonal projection of A to the plane α, see Fig. 21, right.
Then l′ = β ∩ S 2 , where β = OAB .

Notice that given the points A, B in the line l, one of the two segments l \ {A, B }
is the shortest path between them.
Definition 2.12. A triangle on S 2 is a union of three non-collinear points and a triple
of the shortest paths between them.

Figure 21: Spherical triangles

27
2.3 Polar correspondence
Definition 2.13. Let l = S 2 ∩ Πl be a line on S 2 , where Πl is the corresponding plane
through O in R3 . The pole to the line l is the pair of endpoints of the diameter DD′
orthogonal to Πl , i.e. P ol(l) = {D, D′ }.
A polar to a pair of antipodal points D, D′ is the great circle l = S 2 ∩ Πl , s.t. the plane
Πl is orthogonal to DD′ , i.e. P ol(D) = P ol(D′ ) = l.

D0 D0

l l

D D

Figure 22: Polarity: P ol(l) = {D, D′ } (left) and P ol(D) = P ol(D′ ) = l (right).

Proposition 2.14. If a line l contains a point A then the line P ol(A) contains both
points of P ol(l).
Proof. 1. Let {D, D′ } := P ol(l), i.e. DD′ ⊥ αl , where l = αl ∩ S 2 . In particular,
OD ⊥ OA (see Fig. 23, left).

2. By definition, P ol(A) is the line l′ = S 2 ∩ αA , where αA ⊥ OA.

3. We conclude that OD ⊂ αA as AD ⊥ OA. Hence, D ⊂ P ol(A). Similarly,

D′ ⊂ P ol(A).

D0 A0

A C0

O A
l C
αl
B0 B
D
αA

Figure 23: Left: A ∈ l ⇒ P ol(l) ∈ P ol(A). Right: polar triangle.

Hence, polar correspondence transforms:

• points into lines;
• lines into points;
• the statement “A line l contains a point A” into
“The points P ol(l) lie on the line P ol(A)”.

28
Definition 2.15. A triangle A′ B ′ C ′ is polar to AB C (denoted A′ B ′ C ′ = P ol(AB C ))
if A′ = P ol(B C ) and ∠AOA′ ≤ π /2, and similar conditions hold for B ′ and C ′ , see
Fig. 23, right.

Remark. If A′ ∈ P ol(B C ), then to say “ ∠AOA′ ≤ π /2” is the same as to say that
A′ lies on the same side with respect to B C as A.

Exercise. Is there a self-polar triangle AB C on S 2 , i.e. a triangle AB C such that

P ol(AB C ) = AB C ?
Theorem 2.16 (Bipolar Theorem).

(a) If A′ B ′ C ′ = P ol(AB C ) then AB C = P ol(A′ B ′ C ′ ).

(b) If A′ B ′ C ′ = P ol(AB C ) and △AB C has angles α, β , γ and side lengths a, b, c,

then △A′ B ′ C ′ has angles π − a, π − b, π − c and side lengths π − α, π − β , π − γ .
Proof. (a) Since A′ ∈ P ol(B C ), we have OA′ ⊥ OC, OB . Since B ′ ∈ P ol(AC ), we
have OB ′ ⊥ OC, OA. From this we conclude that OC ⊥ OA′ , OB ′ , i.e. OC is
orthogonal to the plane OA′ B ′ , which implies that C ∈ P ol(A′ B ′ ). Also, we have
∠C OC ′ < π /2.
As similar conditions hold for A and B , we conclude that AB C = P ol(A′ B ′ C ′ ).

(b) - Angle β = ∠AB C between the spherical lines AB and B C is equal to the
angle between corresponding planes αAB and αB C in E3 .
- The length b′ in the spherical triangle A′ B ′ C ′ is given by definition by
b′ = ∠A′ OC ′ .
- As OA′ ⊥ αB C , OC ′ ⊥ αAB , we see ∠A′ OC ′ = π − β , see Fig. 24.
So, we get b′ = π − β .
- By symmetry, we get all other equations.

αBA
αBA
A0
αBC
b 0 = 6 A0 B 0 C 0

C0
αBC
β = 6 ABC

Figure 24: Proof of Bipolar Theorem.

29
2.4 Congruence of spherical triangles
Theorem 2.17. SAS, ASA, and SSS hold for spherical triangles.
Proof. The proofs are exactly the same as for similar statements in E2 .
SAS: This is an axiom (of congruence of trihedral angles in E3 ).

ASA: 1. Suppose that ∠B AC = ∠B ′ A′ C ′ , AC = A′ C ′ , ∠B C A = ∠B ′ C ′ A′ .

2. If AB = A′ B ′ , then △AB C ∼ = △A′ B ′ C ′ by SAS.
3. If AB ̸= A′ B ′ , consider B ′′ ⊂ A′ B ′ such that AB = AB ′′ .
4. Then △A′ B ′′ C ′ ∼ = △AB C by SAS, which implies that ∠B C A = ∠B ′′ C ′ A′ .
This means that the lines C B ′ and C B ′′ coincide, and hence B = B ′ (as a unique
intersection of two rays in the given half-space with respect to A′ C ′ ).

SSS: Assume that the corresponding sides of △AB C and △A′ B ′ C ′ are equal but the
triangles are not congruent, see Fig. 25. Consider a triangle AB C ′′ congruent to
A′ B ′ C ′ . Notice that C ′′ ̸= C , but AC = AC ′′ and B C = B C ′′ , which implies that
the segment C C ′′ has two distinct perpendicular bisectors (one constructed as the
altitude in the isosceles triangle AC C ′′ , and another as an altitude in isosceles
triangle B C C ′′ , see Remark 2.18 below). This contradicts to Proposition 2.10.

C 00 C

A B

Figure 25: Proof of SSS.

Notice that as soon as we have SAS property, we can immediately deduce the
following corollary:
Corollary 2.18. (a) In a triangle AB C , if AB = B C then ∠B AC = ∠B C A.

(b) If AB = B C and M is a midpoint of AC then B M ⊥ AC .

Proof. (a) Follows as △AB C ∼
= △C B A by SAS.
Then (b) follows as △B AM ∼= △B C M by SAS in view of (a).

In Euclidean plane, triangles with three equal angles are not necessarily congruent,
but only similar. This is not the case in S 2 :

30
Theorem 2.19. AAA holds for spherical triangles.

Proof. Consider the polar triangles P ol(AB C ) and P ol(A′ B ′ C ′ ). By Bipolar Theorem
(Theorem 2.16(b)) AAA for initial triangles turns into SSS for the polar triangles.
Hence, P ol(AB C ) is congruent to P ol(A′ B ′ C ′ ). Applying Theorem 2.16 again, we
conclude that AB C is congruent to A′ B ′ C ′ .

2.5 Sine and cosine rules for the sphere

a. Sine and cosine rules on the plane
Before discussing spherical sine and cosine rules, lets recall the statements for Euclidean
plane:
Consider a triangle on E2 with sides a, b, c and opposite angles α, β , γ , as in Fig. 26,
left. Then:

a b c
sine rule: sin α = sin β = sin γ

cosine rule: c2 = a2 + b2 − 2ab cos γ

Proof. Sine rule: Let A, B , C be the vertices of the triangle with the angles α, β , γ
respectively. Drop the perpendicular B H from B to AC , see Fig. 26, right. Then
B H = c sin α = a sin γ , which implies sinc γ = sina α . The other equality is obtained by
symmetry.
Cosine rule: With the same H as before, we have B H = a sin γ , C H = a cos γ , then

c2 = AH 2 + B H 2 = (b − C H )2 + B H 2
= (b2 − 2b · a cos γ + a2 cos2 γ ) + a2 sin2 γ = a2 + b2 − 2ab cos γ .

B B
β
c a c a

α γ α γ
A b C A H C

Figure 26: Triangle △AB C .

31
b. Sine and cosine rules on the sphere
Theorem 2.20 (Sine rule for S 2 ). sin a
sin α
= sin b
sin β
= sin c
sin γ
.

Proof. - Let H be the orthogonal projection of A to the plane OB C .

- Let Ab and Ac be orthogonal projections of H to the lines OB , OC respectively,

see Fig. 27, left.

- As AH ⊥ OH C and H Ac ⊥ OC ,
Theorem of three perpendiculars (Theorem 1.42) implies that AAc ⊥ OC .

- As OC ⊥ Ac H and OC ⊥ Ac A,
we see that ∠AAc H = ∠(OH C, OAc A) = ∠(OB C, OAC ) = γ
see Fig. 27, right.
△AH Ac △AOAc
- AH = AAc sin γ = AO sin(π − b) sin γ = R sin b sin γ .

- Similarly, AH = AAb sin β = · · · = R sin c sin β .

- We conclude that sin b

sin β
= sin c
sin γ
.

A A

Ac O Ac
O
H Ab H
C C
B B

Figure 27: Proof of the sine rule on the sphere.

Remark. If a, b, c are small then a ≈ sin a and the spherical sine rule transforms into
Euclidean one.

Corollary. (Thales Theorem) If a = b then ∠α = ∠β , i.e. the base angles in isosceles

triangles are equal.

Theorem 2.21 (Cosine rule for S2 ). cos c = cos a cos b + sin a sin b cos γ .

Proof. We skip the proof in the class, but one can find it in any of the following:
- Prasolov, Tikhomirov: Section 5.1, p.87;
- Prasolov: p.48.

32
 Remark. If a, b, c are small then cos a ≈ 1 − a2 /2 and the spherical cosine rule
transforms into Euclidean one.

Theorem 2.22 (Second cosine rule). cos γ = − cos α cos β + sin α sin β cos c.

Proof. Let A′ B ′ C ′ = P ol(AB C ) be the triangle polar to AB C . Then by Bipolar

Theorem (Theorem 2.16) a′ = π − α, cos a′ = − cos α, sin a′ = sin α. Applying the first
cosine rule (Theorem 2.21) to △A′ B ′ C ′ we get

cos c′ = cos a′ cos b′ + sin a′ sin b′ cos γ ′ ,

which implies
− cos γ = cos α cos β − sin α sin β cos c.

Remark.

(a) If a, b, c are small then cos a ≈ 1 and from the second cosine rule we have
cos γ = − cos α cos β +sin α sin β = cos(α + β ), which means that γ = π − (α + β ).
So, the second cosine rule transforms into α + β + γ = π .

(b) For a right-angled triangle with γ = π /2 we have sin γ = 1, cos γ = 0. So we

obtain:
sine rule: sin b = sin c · sin β ,
cosine rule: cos c = cos a cos b (Spherical Pythagorean Theorem).

(c) Is there a “second sine rule” on the sphere?

Writing the sine rule for the polar triangle only changes the places of numerators
and denominators in the sine rule and does not lead to anything new...

2.6 More about triangles

The following properties of spherical triangles are exactly the same as the corresponding
properties of Euclidean triangles:

Proposition 2.23. For any spherical triangle,

1: angle bisectors are concurrent;

2,3,4: perpendicular bisectors, medians, altitudes are concurrent.

5,6: There exist a unique inscribed and a unique circumscribed circles for the triangle.

Proof. - Parts 1,2 are discussed in HW 5.2 (and can be done as for E2 ).

- Parts 3,4 are discussed in HW 6.5 (here, one needs to use some projections to
reduce the statement to similar statements on E2 .

- Parts 5,6 follow directly from 1,2 respectively (as on E2 , one needs to think about
an angle bisector as a locus of points on the same distance from the sides of the
angle and a perpendicular bisector as a locus of points on the same distance from
the endpoints of the segment).

33
 However, not everything about spherical triangles works exactly the same way as
in Euclidean plane:

Example 2.24. Let M , N be the midpoints of AB and AC in a spherical triangle

AB C . Then M N > AC /2.
One can use cosine law to prove the statement, see HW 6.6.
Moreover, for some triangles in the sphere one can even have M N > AC , or even
M N > 100AC !
To see this take B to be the North Pole, and A and C to be the points on the same
parallel very close to the South Pole.

2.7 Area of a spherical triangle

We will denote area of X by S (X ) or by SX and will assume the following properties
of the area:

• S (X1 ⊓ X2 ) = S (X1 ) + S (X2 )

where ⊓ means a disjoint union, i.e. interior of X1 is disjoint from interior of X2 .

• S (S 2 ) = 4π R2 for a sphere of radius R.

Theorem 2.25. The area of a spherical triangle with angles α, β , γ equals

(α + β + γ − π )R2 ,

where R is the radius of the sphere.

Proof. 1. Consider a spherical digon, i.e. one of 4 figures obtained when S2 is cut
along two lines. See Fig. 28, left. Let S (α) be the area of the digon of angle α.

2. S (α) is proportional to α. Indeed we can divide the whole sphere into 2n con-
gruent digons, and obtain that S (π /n) = 4π R2 /2n. This will show the propor-
tionality for π -rational angles. For others we will apply continuity of the area.
As S (2π ) = S (sphere) = 4π R2 , we conclude that S (α) = 2αR2 .

3. - The pair of lines AB and AC meeting at angle α determines two α digons.

- Similarly, AB and B C gives two β -digons and AC, C B gives two γ -digons,
see Fig. 28, middle.
- The total area of all six digons is Sdigons = 2R2 (2α + 2β + 2γ ).
- Triangle AB C is covered by three digons, also triangle A′ B ′ C ′ antipodal to
AB C is covered by 3 digons.
- All other parts of S2 are covered only by one digon each, see Fig. 28, right.
- So,
3(SAB C + SA′ B ′ C ′ ) + SS2 \{△AB C ∪△A′ B ′ C ′ } = Sdigons .
Hence, 2(SAB C + SA′ B ′ C ′ ) + SS2 = Sdigons . Which implies

4SAB C + 4π R2 = 2R2 (2α + 2β + 2γ )

and we get SAB C = R2 (α + β + γ − π ).

34
 C0
A B

α
A0 B0
C

Figure 28: Computing the area of a triangle using digons

Corollary 2.26. π < α + β + γ < 3π .

Proof. The area of triangle is positive. Also, every angle is smaller than π .

Corollary 2.27. 0 < a + b + c < 2π .

Proof. Let A′ B ′ C ′ = P ol(AB C ) . Then α′ + β ′ + γ ′ > π , and by Bipolar Theorem

(Theorem 2.16) we have (π − a) + (π − b) + (π − c) > π , which implies a + b + c < 2π .

Theorem 2.28. No domain on S 2 is isometric to a domain on E2 .

Proof. One proof directly follows from sine or cosine rule, another from the sum of
angles of a triangle.
The third proof is by comparing the length of circles of radius r: a spherical circle
of radius r has length 2π sin r while in E2 such a circle would have length 2π r, see
Fig. 29 (we leave the computation as an excercise).

sin r

1
r

Figure 29: Computing length of spherical circle

35
2.8 Isometries of the sphere
Example 2.29. The following maps are isometries of S2 (as they are restrictions to
S 2 of isometries in E3 ):

- Rotation about a point A on the sphere may be understood as a restriction of

rotation of E3 about the corresponding diameter of the sphere.

- Reflection with respect a line l on S2 may be understood as a restriction of

reflection in E3 with respect to the plane α s.t. l = α ∩ S2 .

- Antipodal map is a restriction of the symmetry in E3 with respect to the point

O.

Figure 30: Examples of isometries on S2 : rotation, reflection and antipodal map.

Proposition 2.30. Every non-trivial isometry of S 2 preserving two non-antipodal

points A, B is a reflection (with respect to the line AB ).

Proof. - Suppose f ∈ I som(S2 ), such that f (A) = A, f (B ) = B , f (X ) = X ′ ̸= X .

- Since f is an isometry, we see that △AB X is congruent to △AB X ′ (by SSS),

see Fig. 31, left. Hence, ∠AB X = ∠AB X ′ .

- Since X ̸= X ′ , this implies that X and X ′ lie in different hemispheres with respect
to AB .

- Consider the point H ∈ AB such that ∠X H B = π /2. Then △H C B ∼ = △H X ′ B

by SAS. This implies that X ′ = rAB (X ) is a reflection image of X .

Proposition 2.31. Given points A, B , C , satisfying AB = AC , there exists a reflection

r such that r(A) = A, r(B ) = C , r(C ) = B .

Proof. Let M be the midpoint of B C , let r = rAM be the reflection with respect
to AM , see Fig. 31, right. Then △AM B ∼ = △AM C by SSS, which implies that
∠B M A = ∠AM C = π /2, and hence r swaps B and C .

Exercise. The line through B C in the proof above contains 2 segments with endpoints
B , C . Are there two distinct solutions for r?

36
 X0 C

A H B B A

Figure 31: To the proofs of Propositions 2.30 and 2.31

Example 2.32. A glide reflection is an isometry defined by f = rl ◦ RA,φ = RA,φ ◦ rl ,

where rl is a reflection with respect to a line l and RA,φ is a rotation about A = P ol(l),
see Fig. 32, left.

Theorem 2.33. 1. An isometry of S 2 is uniquely determined by the images of 3

non-collinear points.

2. Isometries act transitively on points of S 2 and on flags in S 2

(where a flag is a triples (A, l, h+ ), where A is a point, l is a line containing A,
and h+ is a choice of hemisphere bounded by l).

3. The group I som(S 2 ) is generated by reflections.

4. Every isometry of S 2 is a composition of at most 3 reflections.

5. Every orientation-preserving isometry is a rotation.

6. Every orientation-reversing isometry is either a reflection or a glide reflection.

Proof. 1-4 are proved similarly to their analogues in E2 .

5: An orientation-preserving isometry of S2 is a composition of 2 reflections with
respect to some lines l1 , l2 . As any two lines intersect non-trivially on S2 , we conclude
that it is a rotation.
6: See Lemma 2.34 below.

Lemma 2.34. Let r1 , r2 , r3 be distinct reflections not preserving the same point of S 2 .
Then r3 ◦ r2 ◦ r1 is a glide reflection.

Proof. To show the lemma we will use non-uniqueness of presentation of an isometry

as a composition of reflections.
We will denote by rX
∗
a reflection with respect to the line lX ∗
. Also, denote g = r3 ◦ r2 ◦ r1 .
Notice, that the lines l1 , l2 , l3 are all distinct and not passing through the same
point.

37
 - Let A = l1 ∩ l2 . Let l2′ be the line through A orthogonal to l3 . There exists a line
l1′ through A such that r2 ◦ r1 = r2′ ◦ r1′ . Hence,

g = r3 ◦ r2 ◦ r1 = r3 ◦ (r2′ ◦ r1′ ) = (r3 ◦ r2′ ) ◦ r1′ ,

see Fig. 32 (the two diagrams in the middle).

- Similarly, let B = l3 ∩ l2′ . Let l3′′ ⊥ l1 be the line through B orthogonal to l1′ and
let l2′′ be the line such that r3 ◦ r2′ = r3′′ ◦ r2′′ (i.e. l3′′ ⊥ l2′′ ), see Fig. 32 (the two
diagrams on the right). Then we get

g = (r3′′ ◦ r2′′ ) ◦ r1′ = r3′′ ◦ (r2′′ ◦ r1′ ),

where r3′′ is the reflection in l3′′ and (r2′′ ◦ r1′ ) is the rotation about the point l2′′ ∩ l1
polar to l3′′ . Hence, g is a glide reflection.

A l2
ϕ l10
l10
l1 l20 l200
X
l l3 l3
f (X)
l300

Figure 32: Glide reflection and a composition of three reflections

Remark. We could try to prove the lemma shorter by saying that r3 ◦ r2 ◦ r1 =

(r3 ◦ r2 ) ◦ r1 is a composition of a rotation and reflection, as required. But we don’t
know (and it is not always true) that the centre of the rotation (r3 ◦ r2 ) is polar to the
line of reflection r3 .

Exercise. What is the type of the antipodal map?

Remark 2.35. Fixed points of isometries on S 2 distinguish the types of isometries.

Indeed, fixed points of identity map, reflection rl , rotation RA,α and a glide reflection
are the whole sphere, the line l, the pair of antipodal points A, A′ and the empty set
respectively.

Theorem 2.36. (a) Every two reflections are conjugate in I som(S 2 ).

(b) Rotations by the same angle are conjugate in I som(S 2 ).

Proof. Idea of proof:

(a) Let r1 and r2 be reflection with respect to the lines l1 and l2 . Let l be an angle
bisector for an angle formed by l1 and l2 . Then r2 = rl−1 ◦ r1 ◦ rl (indeed, rl takes
l2 to l1 , then r1 preserves l1 , then rl−1 takes l1 back to l2 , so, the composition
rl−1 ◦ r1 ◦ rl preserves l2 pointwise and changes the orientation, which means that
it coincides with r2 ).

38
 (b) Let A and B be the centres of the two rotations RA,φ ,RB ,φ , let l be the orthogonal
bisector of AB . Then RA,φ−1
= rl−1 ◦ RB ,φ ◦ rl . Also, RA,φ
−1
is conjugate to RA,φ
−1

since the rotation RA,φ = r2 ◦ r1 is a composition of some reflections r1 , r2 , and

the inverse is RA,φ
−1
= r1 ◦ r2 = r1−1 ◦ (r2 ◦ r1 ) ◦ r1 .

Remark 2.37. As S2 ⊂ E 3 , we have I som(S2 ) ⊂ I som(E3 ) (more precisely, isometries

of the sphere is the origin-preserving subgroup of isometries of E3 ). This is given by
orthogonal 3 × 3 matrices (i.e. matrices satisfying AT A = AAT = I .)
Orientation reversing isometries correspond to matrices with det = −1, while
orientation-preserving to ones with det = 1.
Orientation-preserving isometries form a subgroup given by

S O(3, R) = {A ∈ M3 |AT A = I , det A = 1}.

2.9 Platonic solids and their symmetry groups (NE)

(Non-examinable section)

We conclude our exposition of spherical geometry by a brief discussion of symmetry

groups of Platonic solids, i.e. regular polyhedra known since antiquity, namely

tetrahedron, cube, octahedron, dodecaghedron and icosohedron

(see Fig.33, left to right).

Definition 2.38. By a regular polyhedron we mean a polyhedron P with largest possi-

ble group of symmetries GP , i.e. the group GP should act on P by isometries mapping
its vertices to vertices, and the action GP : P should be transitive

- on vertices of P ;
- on edges of P ;
- on faces of P .

Moreover, GP should act transitively on flags in P , i.e. on triples (V , E , F ) where V is

a vertex, and E is an edge such that V ∈ E , and F is a face of P such that E ∈ F .

To find a fundamental domain of the action, one needs to choose a flag (V1 , E1 , F1 )
in P . Let A = V1 be a vertex, and B be a midpoint of the edge E1 and C be a centre
of the face F1 . Then one can check that the triangle AB C is a fundamental domain of
the action GP : P .
Projecting P from its center O to a sphere centred at O one can turn the triangle
AB C into a spherical triangle A′ B ′ C ′ . One can check that the angles of this spherical
triangle are

- ( π2 , π3 , π3 ) when P is a tetrahedron;

- ( π2 , π3 , π4 ) when P is a cube or an octahedron;

- ( π2 , π3 , π5 ) when P is a dodecahedron or an icosohedron.

39
Figure 33: Regular polyhedra (from left to right): tetrahedron, cube, octahedron,
dodecahedron and icosohedron.

One can also check that the group GP : S 2 is generated by reflections with respect to
the sides of the triangle A′ B ′ C ′ .

Remark 2.39. Let H : S 2 be an action. As the sphere is a compact set, H acts on S 2

discretely if and only if H is a finite group.

Remark 2.40. A group H generated by reflections on S 2 is finite if and only if

• H is generated by 1 or 2 reflections;

• H is generated by reflections with respect to the sides of one of the following

triangles with angles:

– ( π2 , π2 , πn ), n ∈ Z, n ≥ 2;
– ( π2 , π3 , π3 ), ( π2 , π3 , π4 ), ( π2 , π3 , π5 ).

Remark 2.41. Notice that the same group serves as the symmetry group for the cube
and the octohedron - this is because the cube is dual to the octahedron (if we take
a regular cube and mark the centeres of its faces, then the six marked points will be
vertices of a regular octahedron; also, we can obtain a cube if we highlight the centeres
of faces of the octahedron). Similarly, an icosohedron is dual to a dodecahedron, while
a tetrahedron is dual to itself.

40
2.10 References
- In this section, we have mostly followed the exposition in
V. V. Prasolov, Non-Euclidean Geometry (see Lecture I and pp. 48-49)
or you can find the same material in pp. 83-87 of
V. V. Prasolov, V. M. Tikhomirov Geometry.

- The spririt of our discussion of isometry group of the sphere follows the paper by
Oleg Viro: O. Viro, Defining relations for reflections. I, arXiv:1405.1460v1.

- For another exposition concerning the isometry group of the sphere see
G. Jones, Algebra and Geometry, Lecture notes (Section 2.2).

- More general notion of polarity comparing to the one considered in Section 2.3
is presented in Sections 16-17 of the following lecture notes:
A. Barvinok, Combinatorics of Polytopes.

- One can read about tilings by triangles in

V. V. Prasolov, Non-Euclidean Geometry , Lecture X, p. 34-36,
or in
V. V. Prasolov, V. M. Tikhomirov, “Geometry”, Section 5.5, p 185-187.

41
3 Affine geometry
An affine space is a vector space whose origin we try to forget about.
Marcel Berger

We consider the same space R2 as in Euclidean geometry but with larger group acting
on it.

3.1 Similarity group

Similarity group, S im(R2 ) is a group generated by all Euclidean isometries and scalar

multiplications, i.e. transformations given by (x1 , x2 ) 7→ (k x1 , k x2 ), k ∈ R.

Its elements may change size, but preserve the following properties:
angles, proportionality of all segments, parallelism, similarity of triangles.

This means that many problems in Euclidean geometry are actually problems about
“similarity geometry”.

Example 3.1. Consider the following theorem of Euclidean geometry:

A midline in a triangle is twice shorter than the corresponding side.

One can prove it as follows. Let M and N be the midpoints of AB and B C in the
triangle AB C , see Fig. 34. Let B = 0 be the origin, consider the map f : C → C
taking z → 2z , i.e. the map which doubles every distance. Then for every segment I
the length of f (I ) is twice the length of I . In particular, as f (M ) = A and f (N ) = C ,
we get |AC | = 2|M N |.

M N

A C

Figure 34: Length of midline using similarity

Remark. A map which may be written as a scalar multiplication in some coordinates

in R2 is called homothety (with positive or negative coefficient depending on the sign
of k ).

Here, one can find the picture of a pantograph and a Sylvester machine - two mecha-
nisms for implementing similarity (webpage by Rémi Coulon).

42
3.2 Affine geometry
Instead of scalar maps, as in “similarity geometry”, now we will consider all non-
degenerate linear maps.

Affine transformations are all transformations of the form f (x) = Ax + b where A ∈

GL(2, R).

Proposition 3.2. Affine transformations form a group.

Proof. We leave the proof as an exercise. You need to write f (x) = Ax + b and to
find the composition of two such maps, then to find f −1 and an identity map. The
associativity will follow from associativity of composition.

Example 3.3. (a) Consider the map f : C → C given by f (z ) = 2z + 2 + i. By

definition f ∈ Af f (R2 ), but also one can notice that f ∈ S im(R2 ).
     
x 2x + y + 1 2 1
(b) Now, consider f : → . As det = 3 ̸= 0, we conclude
y −x + y + 2 −1 1
that f ∈ Af f (R2 ). At the same time f ∈ / S im(R2 .

Figure 35: Examples of affine maps (see Example 3.3).

Affine transformations do not preserve length, angles, area.

Proposition 3.4. Affine transformations preserve

(1) collinearity of points;

(2) parallelism of lines;

(3) ratios of lengths on any line;

(4) concurrency of lines;

(5) ratio of areas of triangles (so ratios of all areas).

Proof. Linear maps preserve the properties (1)-(5), translations also preserve them.
So, affine maps, as their compositions, also preserve all these properties.

Proposition 3.5. (1) Affine transformations act transitively on triangles in R2 .

43
 (2) An affine transformation is uniquely determined by images of 3 non-collinear
points.
Proof. (1) Let AB C and A′ B ′ C ′ . We want to find a map f (x) = Ax + b such that
f (AB C ) = A′ B ′ C ′ . We will find it as a composition f = g ◦ h, where
     
g 0 1 0 h ′ ′ ′
A, B , C → , , → A ,B ,C .
0 0 1

The map h is easy to find, and so is the map g −1 . This implies that the compo-
sition f = g ◦ h exists.

(2) Suppose there are two different affine transformations f and g taking the non-
collinear points A, B , C to A′ , B ′ , C ′ . Then the transformation g −1 ◦ f ̸= id is a
non-trivial transformation preserving   all 
three
  points
 A, B , C . Let h be the affine
0 1 0
transformation taking the points , , to A, B , C . Then the affine
0 0 1      
0 1 0
transformation h ◦ (g ◦ f ) ◦ h preserves the points
−1 −1
, , (and it is
0 0 1
a non-trivial transformation, since it is conjugate to a non-trivial one).   Which
   is
0 1 0
a contradiction, as a transformation Ax + b taking the points , ,
    0 0 1
0 1 0
to themselves clearly has b = and A = .
0 0 1

Example 3.6. We will use the affine group to show the following statement of Eu-
clidean geometry:
The medians of a triangle in E2 are concurrent.
Proof.

- The statement is trivial for a regular triangle (as each of the three medians passes
through the centre of the triangle).

- Apply an affine transformation f which takes some regular triangle to the given
triangle AB C .

- f takes the medians of the regular triangle to the medians of AB C (as it maps
vertices to vertices and midpoints to midpoints).

- So, it takes the intersection of the three medians to the intersection of the three
medians of AB C .

Theorem 3.7. Every bijection f : R2 → R2 preserving collinearity of points, between-

ness and parallelism is an affine map.
Proof.

- Let g be anaffine
 map  which
 takes      
0 1 0 0 1 0
the points , , to f ( ), f ( ), f ( ).
0 0 1 0 0 1
(this map exists by Theorem 3.5).

44
 - We want to show f (x) = g (x) for all x ∈ R2 .

- We will denote the points by their complex coordinates, so by now we know the
desired property for 0, 1, i.

- As affine maps take parallel lines to parallel lines and f also preserves collinearity,
we conclude that f (x) = g (x) also for x = 1 + i (as 1 + i lies on the line though
1 parallel to the line through O and i and on also it lies on the line through i
parallel to the line through 0 and 1), see Fig. 36, left.

- Similarly, we use the points i, 1 + i, 1 to conclude the property for the point 2,
see Fig. 36 middle left.

- Applying this procedure, one can show the property for all integer points a + bi,
a, b, ∈ Z.

- Every half-integer point a + bi, a, b, ∈ 12 Z can be obtained as an intersection of

two segments with integer endpoints, so the property also holds for half-integer
points, see Fig. 36 middle right and right..

- Applying the previous step again, we obtain the property for 41 -integer points,
then for 81 -integer points, and so on... We will get smaller and smaller lattices.

- As f preserves betweenness and coincides with g on a dense set of points, we

conclude that f is continuous and f (x) = g (x) for all x ∈ R2 .

i 1+i
f (1+i)

0 1 2

Figure 36: To the proof of Theorem 3.7.

Remark. If f is a bijection R2 → R2 preserving collinearity,

then it preserves parallelism and betweenness.
Proof. Parallelism: of f takes parallel lines to the lines intersecting at the point A,
consider f −1 (A). It exists because f is a bijection, and it would lie on both of the
parallel lines as f preserves collinearity. The contradiction shows that f preserves
parallelism.
Betweenness: the argument here is much more involved, we will skip it. You can find
the argument on pp.40-41 in the book by Prasolov and Tikhomirov.
This allows as to reformulate Theorem 3.7 as follows.

Theorem 3.7’. (The fundamental theorem of affine geometry).

Every bijection f : R2 → R2 preserving collinearity of points is an affine map.

45
Corollary 3.8. If f : R2 → R2 is a bijection which takes circles to circles, then f is
an affine map.

Proof.

(1) The transformation f −1 maps three collinear points f (A), f (B ), f (C ) to 3 three

collinear points A, B , C .

- Indeed, if the points A, B , C are not collinear, then they are pairwise distinct
and there is a circle through A, B , C.
- Hence, f (A), f (B ), f (C ) are also pairwise distinct (as f is bijective) and lie
on a circle (since f maps circles to circles).
- Then f (A), f (B ), f (C ) cannot lie on one line.

(2) From (1) and Theorem 3.7’ we conclude that f −1 is affine, which implies that f
is also affine.

Remark. An affine transformation takes ellipses to ellipses. So, in Corollary 3.8 we

can change the circles to ellipses.

Example 3.9 (Parallel Projection). Consider two copies α and β of a two-dimensional

plane in R3 , let suppose that each of α and β are endowed with coordinates. Project
from α to β by parallel rays (the rays should not be parallel to any of α and β !).
Then we get a bijection between the two planes, and one can see that this bijection is
preserving parallelism (indeed, if two parallel lines l, m ∈ α are mapped to intersecting
lines l′ , m′ ∈ β , then what is the preimage of the intersection l′ ∩ m′ ∈ β ?). Applying
the fundamental theorem of affine geometry, we conclude that the parallel projection
is an affine map.

Proposition 3.10. Every parallel projection is an affine map, but not every affine map
is a parallel projection.

Proof. It is already shown in Example 3.9 that the parallel projections are affine maps.
To see the second statement, consider the affine map f : z → 2z :

- Suppose that f is a parallel projection f : α → β .

- The planes α and β are not parallel (otherwise, f would be an isometry, which
is not the case).

- Consider the line of intersection α ∩ β . Every point of this line is mapped by f

to itself, so the distance between two points on that line is preserved.

- At the same time z → 2z makes all distances twice longer. So, f : z → 2z cannot
be a parallel projection.

Exercise 3.11. Every affine map can be obtained as a composition of two parallel
projections. (See also p.18 in Geometry Lecture notes by Norbert Peyerimhoff).

46
3.3 References
- Most of the material above (and more information on affine geometry) may be
found in
G. Jones, Algebra and Geometry, Lecture notes (Section 3).

- For fundamental theorem of affine geometry and its corollaries see

V. V. Prasolov, V. M. Tikhomirov, Geometry , Section 2.1. pp.39-42.

- For another exposition of affine geometry, based on parallel projection, see

N. Peyerimhoff, Geometry, Lecture notes, (Section 2, Section 2.1 and 2.2.).

- Illustrating Mathematics by Rémi Coulon: a panthograph and a Sylvester ma-

chine.

47
 a

Albrecht Dürer, The Draughtsman of the Lute. Woodcut.

From Dürer’s “ Unterweysung der Messung mit dem Zyrkel und Richtscheyd”, 1525.
Image from [Link]
(OA public domain)
4 Projective geometry
Projective geometry is all geometry.
Arthur Caley
Motivation: We have considered larger and larger groups acting on the same space R2 ,
now we are going to consider even larger group P roj (a) of projective transformations:

I som(E2 ) ⊂ S im(2) ⊂ Af f (2) ⊂ P roj (2).

The bigger is the group acting, the smaller is the set of properties it preserves. Now,
we will extend the group so that is will only preserve collinearity (but not parallelism
or betweenness).
The group P roj (2) of projective transformations will act transitively on the pairs of
lines, in particular there will be transformations taking intersecting lines into parallel.
The intersection point of the lines in this case still needs to be mapped somewhere.
This motivates the idea of adding some points to the plane, namely “points at infinity”
(we will have infinitely many of them, more precisely, one point for each direction).

4.1 Projective line, RP1

Model:
- Points of the projective line are lines though the origin O in R2 .
On the plane with coordinates (x1 , x2 ) consider the line l0 given by the equation
x2 = 1. Then every line l through the origin O can be represented by the
coordinates of the intersection l ∩ l0 = (x, 1), except for the line Ox1 which does
not intersect l0 , see Fig. 37.
We will assign to Ox1 a special point, “point at infinity” and will denote it x∞ .

x2
l : x2 = 1

x1

Figure 37: Projective line: set of lines through O in R2 .

- Group action: GL(2, R) acts on

 R by mapping a line though O to another line
2

a b
through O: a matrix A = with ad − bc ̸= 0 maps the point (λx, λ) ∈ l to
c d
    
a b λx ax + b
=λ .
c d λ cx + d

If cx + d ̸= 0, we can write A : (x, 1) → ( ax+b

cx+d
, 1).
The point (−d/c, 1) is mapped to x∞ . So, GL(2, R) acts on RP1 .

49
 - Homogeneous coordinates: a line through O is determined by a pair of num-
bers (ξ1 , ξ2 ), where (ξ1 , ξ2 ) ̸= (0, 0).
The pairs (ξ1 , ξ2 ) and (λξ1 , λξ2 ) determine the same line, so are considered as
equivalent.
The ratio (ξ1 : ξ2 ) determines the line and is called homogeneous coordinates of
the corresponding point in RP1 .
The GL(2, R)-action in homogeneous coordinates writes as
 
a b
A : (ξ1 : ξ2 ) 7→ (aξ1 + bξ2 : cξ1 + dξ2 ), where A = ,
c d

and is called a projective transformation.

Remark. Projective transformations are called this way since they are compositions of
projections (of one line to another line from a point not lying on the union of that lines).
The following several statements will help us to prove that projective transformations
are exactly the set of all possible compositions of such projections.
Lemma 4.1. Let points A2 .B2 , C2 , D2 of a line l2 correspond to the points A1 , B1 , C1 , D1
of the line l1 under the projection from some point O ∈ / l1 ∪ l2 . Then
|C1 A1 | . |D1 A1 | |C2 A2 | . |D2 A2 |
= .
|C1 B1 | |D1 B1 | |C2 B2 | |D2 B2 |
Proof. For a triangle ∆ let S∆ denote the Euclidean area of ∆. Recall that given a
Euclidean triangle AB C with altitude B H one has
1 1
SAB C = |B H | · |AC | = |AB | · |AC | sin ∠B AC. (4.1)
2 2
In particular, SOC1 A1 = A1 C2 1 ·h , SOC1 B1 = A1 B2 1 ·h , where h is the distance from O to
the line l1 . Hence, we have
|C1 A1 | SOC1 A1 (4.1) |OC1 ||OA1 | sin ∠A1 OC1 |OA1 | sin ∠A1 OC1
= = = ,
|C1 B1 | SOC1 B1 |OC1 ||OB1 | sin ∠B1 OC1 |OB1 | sin ∠B1 OC1
which implies that

|C1 A1 | . |D1 A1 | |OA1 | sin ∠A1 OC1 . |OA1 | sin ∠A1 OD1
=
|C1 B1 | |D1 B1 | |OB1 | sin ∠B1 OC1 |OB1 | sin ∠B1 OD1
sin ∠A1 OC1 sin ∠B1 OD1 sin ∠A2 OC2 sin ∠B2 OD2
= · = · = R H S.
sin ∠A1 OD1 sin ∠B1 OC1 sin ∠A2 OD2 sin ∠B2 OC2

Definition 4.2. Let A, B , C, D be four points on a line l, and let a, b, c, d be their
coordinates on l. The value [A, B , C, D] := cc−a d−a
−b d−b
is called the cross-ratio of these
points.
So, we can reformulate Lemma 4.1 as follows.

Lemma 4.1’. Projections preserve cross-ratios of points.

50
 O l1
D1
C1
A1 B1

l2
A2 B2 C2 D2

Figure 38: Projection preserves cross-ratio.

Definition 4.3. The cross-ratio of four lines lying in one plane and passing through
one point is the cross-ratio of the four points at which these lines intersect an arbitrary
line l.

Remark. By Lemma 4.1’, Definition 4.3 does not depend on the choice of the line l.

Proposition 4.4. Any composition of projections is a liner-fractional map.

Proof. Let f be a composition of projections. Let a′ , b′ , c′ be images of points a, b, c

under a composition of projections. By Lemma 4.1’, [a, b, c, x] = [a′ , b′ , c′ , f (x)], i.e.

c − ax − a c′ − a′  f (x) − a′
= ′ .
c−b x−b c − b′ f ( x ) − b′

Expressing f (x) from this equation we get f (x) = αx+β

γ x+δ
for some α, β , γ , δ .

Proposition 4.5. A composition of projections preserving 3 points is an identity map.

Proof. We leave the proof as an exercise.

Hint: use f (x) = αx +β
γ x+δ
and show that if f fixes three points then either f (x) = x or
there is a quadratic equation with 3 roots.

Lemma 4.6. Given A, B , C ∈ l and A′ , B ′ , C ′ ∈ l′ , there exists a composition of

projections which takes A, B , C to A′ , B ′ , C ′ .

Proof.

• Consider any line l′′ such that A′ ∈ l′′ and l′′ ̸= l′ . Let O ∈ AA′ be any point, see
Fig. 39.

• Project B , C from O to l′′ . This will define points B ′′ and C ′′ respectively.

• Let P = B ′ B ′′ ∩ C ′ C ′′ . Project l′′ to l′ from P . The composition of the two

projections takes points A, B , C to A′ B ′ C ′ .

51
 C
B
A
l

B0 C0
A0 = A00
l0

B 00 C 00

l00
P
O

Figure 39: To the proof of Lemma 4.6.

Remark. If in the proof above B ′ B ′′ ||C ′ C ′′ we can chose another line l′′ so that the
lines will not be parallel (in particular, if we move l′′ so that it crosses B O and C O
closer to the point O, then the intersection P = B ′ B ′′ ∩ C ′ C ′′ moves also closer to O).

Theorem 4.7.

(a) The following two definitions of projective transformations of RP1 are equivalent:

(1) Projective transformations are compositions of projections;

(2) Projective transformations are linear-fractional transformations.

(b) A projective transformation of a line is determined by images of 3 points.

Proof. First, we will prove part (a) of the theorem.

(1)⇒(2) Compositions of projections are linear-fractional transformations by Proposi-

tion 4.4.

(1)⇐(2) We will prove this in three steps.

(i) We will now show that

linear-fractional transformations preserve cross-ratios.
Indeed, if yi = αxi +β
γ xi +δ
, then one can check that

(αγ − β δ )(xi − xj )
yi − yj = .
(γ xi + δ )(γ xj + δ )

Denote ui = 1
γ xi +δ
. Then

y3 − y1  y4 − y1 u3 · u1  u4 · u1
= [ x1 , x2 , x3 , x4 ] = [x1 , x2 , x3 , x4 ].
y3 − y2 y4 − y2 u3 · u2 u4 · u2

52
 (ii) Hence, a linear-fractional transformation is determined by the images of
3 points. Indeed, if there are two linear-fractional transformations f and
g which take A, B , C to A′ , B ′ , C ′ , then g −1 ◦ f is a non-triavial linear-
fractional transformaion preserving three points A, B , C , which is impossible
as would lead to a quadratic equation with 3 roots (compare to the proof of
Proposition 4.5).
(iii) Let f be a linear-fractional transformation. By Lemma 4.6, there exists a
composition of projections φ which takes A, B , C ∈ R to f (A), f (B ), f (C ).
In view of the part ((1)⇒(2)), the map φ is linear-fractional. Then Step
(ii) implies that φ = f (i.e. a linear-fractional map f is the composition of
projection φ).
This completes the proof of part (a) of the theorem. Part (b) follows now from
Step (ii).

4.2 Projective plane, RP2

Model:
- Points of RP2 are lines through the origin O in R3 .
Let x1 , x2 , x3 be coordinates in R3 and let α ∈ R3 be the plane x3 = 1.
For each line l ∈ / Ox1 x2 take a point l ∩ α, see Fig. 40.
For each line in the plane Ox1 x2 assign a “point at infinity” .

x3

α : x3 = 1

x2

x1

Figure 40: Projective plane: set of lines through O in R3 .

- Lines of RP2 are planes through O in R3 .

All points at infinity form a line at infinity (a copy of RP1 ).
- Group action: GL(3, R) (acts on R3 mapping a line though O to another line
through O).
- Homogeneous coordinates:
· A line though O is determined by a triple of numbers (ξ1 , ξ2 , ξ3 ), where
(ξ1 , ξ2 , ξ3 ) ̸= (0, 0, 0).

53
 · Triples (ξ1 , ξ2 , ξ3 ) and (λξ1 , λξ2 , λξ3 ) determine the same line, so are consid-
ered equivalent.
· So, lines are in bijection with ratios (ξ1 : ξ2 : ξ3 ) called homogeneous coordinates.
- Projective transformations in homogeneous coordinates:
A : (ξ1 : ξ2 , ξ3 ) 7→ (a11 ξ1 + a12 ξ2 + a13 : a21 ξ1 + a22 ξ2 + a23 ξ3 : a31 ξ1 + a32 ξ2 + a33 ξ3 ),
where A = (aij ) ∈ GL(3, R).
- Points and lines in RP2 :
· Points are lines through O in R3 ;
· Lines are 2-dimensional planes through O in R3 , see Fig. 41.
· A plane through O can be written as
a1 x1 + a2 x2 + a3 x3 = 0, (4.2)
where (a1 , a2 , a3 ) ̸= (0, 0, 0).
· If (a1 , a2 ) ̸= (0, 0) then the plane defined by Equation 4.2 makes a trace on
the plane x3 = 1; this trace if the line given by
(
a1
x + aa23 x2 = −1 for a3 ̸= 0
a3 1
a1 x 1 + a2 x 2 = 0 for a3 = 0

· The plane x3 = 0 gives a “line at infinity”.

x3

α : x3 = 1

x2

x1

Figure 41: Projective plane: lines are planes through O in R3 .

Remark.
(1) A unique line passes through any given two points in RP2 (as a unique plane
through the origin passes through any two lines intersecting at the origin).
(2) Any two lines in RP2 intersect at a unique point (as any two planes through O
in R3 intersect by a line through O).
(3) Relation 4.2 establishes duality between points and lines in RP2 :
(the point (a1 , a2 , a3 ) is dual to the plane a1 x1 + a2 x2 + a3 x3 = 0).
So, for any theorem about points in RP2 there should be a dual theorem about
lines.

54
Theorem 4.8. Projective transformations of RP2 preserve cross-ratio of 4 collinear
points.

Proof. - Let f be a projective transformation and let α ∈ R3 be the plane through

the origin containing the four collinear points whose cross-ratio we consider.

- Find an isometry i ∈ I som(R3 ) which takes α to the plane f (α).

- Let φ = f ◦ i−1 , i.e. f = φ ◦ i. Notice that φ is a projective transformation of

the projective line α (as φ is a composition of a projective transformation and
an isometry).

- i preserves cross-ratios (as it is an isometry), and φ preserves cross-ratios by

Theorem 4.7. This implies that f preserves cross-ratio of the considered points
(as a composition of cross-ratio preserving maps).

- As the quadruple of collinear points was chosen randomly, we conclude that f

preserves all cross-ratios.

Definition. A triangle in RP2 is a triple of non-collinear points.

Proposition 4.9. All triangles of RP2 are equivalent under projective transformations.

Proof. There exists an element of GL(3, R) which takes three given linearly indepen-
dent vectors to three other given linearly independent vectors.

Definition 4.10. A quadrilateral in RP2 is a set of four points, no three of which are
collinear.

Proposition 4.11. For any quadrilateral Q in RP2 there exists a unique projective
transformation which takes Q to a given quadrilateral Q′ .

Proof. - It is sufficient to prove the statement for the fixed quadrilateral

Q′ = Q0 = [(1, 0, 0), (0, 1, 0), (0, 0, 1), (1, 1, 1)].

Indeed, if we have projective transformations f : Q → Q0 and g : Q′ → Q0 , then

g −1 ◦ f is a projective transformation mapping Q → Q′ . Moreover, if φ ̸= g −1 ◦ f
is another projective transformation taking Q to Q′ then g ◦ φ ̸= f is another
projective transformation mapping Q to Q0 .

- By Proposition
 4.9 we
 may assume that Q = [(1, 0, 0), (0, 1, 0), (0, 0, 1), (a, b, c)].
a 0 0
Then f = 0 b 0 is the unique map taking Q0 to Q, which implies that f −1
0 0 c
is the unique map taking Q to Q0 .

Theorem 4.12. A bijective map from RP2 to RP2 preserving projective lines is a
projective map.

55
Proof. Consider a bijection f : RP2 → RP2 . Let l∞ be the line at infinity and f (l∞ )
be its image under f . Consider a projective map φ which maps f (l∞ ) to l∞ (it does
exists as there is a projective map taking any two points in RP2 to any other two
points in RP2 ). Then the map ψ = φ ◦ f takes l∞ to itself (so, one can restrict it two
ψ : R2 → R2 ). Also, ψ preserves collinearity (as a composition of the transformation
f preserving collinearity with a projective transformation).
Hence, by Fundamental Theorem of affine geometry the map ψ = φ ◦ f is affine.
This implies that the map f = φ−1 ◦ ψ is projective (as a composition of an affine and
projective transformations).

Corollary 4.13. A projection of a plane to another plane is a projective map.

Proof. As a projection preserves the lines, Theorem 4.12 implies that it is a projective
map.

Remark. A projection of a plane α to another plane β is not an affine map if α is not

parallel to β , as in this case some line from α will not be mapped to β .

4.3 Some classical theorems on RP2

Remark on projective duality:

point A = (a1 : a2 : a3 ) ←→ line lA : a1 x1 + a2 x2 + a3 x3 = 0

A ∈ lB ←→ B ∈ lA
line through A, B ←→ point of intersection: lA ∩ lB
3 collinear points ←→ 3 concurrent lines
... ←→ ...

Proposition 4.14 (On dual correspondence). The interchange of words “point” and
“line” in any statement about configuration of points and lines related by incidence does
not affect validity of the statement.
Proof. The relation a1 x1 + a2 x2 + a3 x3 = 0 is symmetric with respect to the coordi-
nates of the point X and the line lA , applying duality we only change the geometric
interpretation of the equations. Algebra remains the same.

Theorem 4.15 (Pappus’ theorem). Let a and b be lines, A1 , A2 , A3 ∈ a, B1 , B2 , B3 ∈

b. Let P3 = B1 A2 ∩ A1 B2 , P2 = B1 A3 ∩ A1 B3 , P1 = B3 A2 ∩ A3 B2 . Then the points
P1 , P2 , P3 are collinear.
Proof.
- Let P2′ = B1 A3 ∩ P1 P3 , let C = B1 A3 ∩ A1 B2 .
We need to show that P2 = P2′ , see Fig. 44, left.
- Consider a composition f of 3 projections:
A1 A
2 3 P
B1 A3 −→ b −→ B2 A3 −→ B1 A3 ,
A
where l1 −→ l2 denotes a projection of l1 to l2 from A, see Fig. 44, right.

56
 - Notice that f takes C → B2 → B2 → C , so f (C ) = C .
Also it takes B1 → B1 → B1 A2 ∩ B2 A3 → B1 , so f (B1 ) = B1 .
One can check similarly that f (A3 ) = A3 and f (P2 ) = P2′ .

- So, this is a projective transformation of the line B1 A3 preserving the points

C, B1 , A3 . By Theorem 4.7 (b), f is identity map.

- Since f (P2 ) = (P2′ ) we conclude that P2 = P2′ .

A1 A2 A3 A1 A2 A3

P1 P3 P2 P1
P3 P2

B3 C B3
B2 B2

B1 B1

Figure 42: Pappus’ Theorem and its proof by composition of 3 projections.

Remark. Sketch of another proof of Pappus’ Theorem:

- By Proposition 4.11 there exists a projective map taking the points A1 A2 B2 B1
to vertices of a unit square.

- So, we may assume that the points A1 , A2 , A3 are (0, 1), (1, 1), (a, 1) and the
points B1 , B2 , B3 are (0, 0), (1, 0), (b, 0).

A1 A2 A3 = (a, 1)

B1 B2 B3 = (b, 0)

Figure 43: Another proof of Pappus’ Theorem.

- Then it is easy to compute the coordinates of the points P1 , P2 , P3 and check that
the points are collinear.

- To establish collinearity of the points, check that the vectors P1 P2 and P1 P3 are
proportional.

57
Remark 4.16 (Dual statement to Pappus’ theorem). Let A and B be points and
a1 , a2 , a3 be lines through A, and b1 , b2 , b3 be lines through B .
Let p1 be a line through b2 ∩ a3 and a2 ∩ b3 ,
p2 be a line through b1 ∩ a3 and a1 ∩ b3 ,
p3 be a line through b2 ∩ a1 and a2 ∩ b1 .
Then the lines p1 , p2 , p3 are concurrent.
(This is actually the same statement as Pappus’ theorem itself!)

A1 A2 A3 A p1
a1
a a2
a3

P1 p2
P3 b
P2
p3
B3 b3
b2
B2
b1 B

B1

Figure 44: Pappus’ Theorem and the dual statement.

Remark 4.17. Pappus’ theorem is a special case of Pascal’s Theorem (see Fig. 45):
If A, B , C, D, E , F lie on a conic then the points AB ∩ DE , B C ∩ E F , C D ∩ F A are
collinear.
We leave Pascal’s Theorem without proof, you can find the proof in
- V. V. Prasolov, V. M. Tikhomirov. Geometry, (2001). Section 4.2, p. 71.

C
B D
A
E

Figure 45: Pascal’s Theorem.

58
 P3
P2

P4

P1

P5
P6

Figure 46: Brianchon’s Theorem.

Remark 4.18. Dual to Pascal’s Theorem is Brianchon’s Theorem (see Fig. 46):
Let P1 P2 P3 P4 P5 P6 be a hexagon formed by 6 tangent lines to a conic. Then the lines
P1 P4 , P2 P5 , P3 P6 are concurrent.

S P2
A3
A2 P1
B3
A1

P3 B2

B1

Figure 47: Desargues’ Theorem.

Theorem 4.19 (Desargues’ theorem). Suppose that the lines joining the corresponding
vertices of triangles A1 A2 A3 and B1 B2 B3 intersect at one point S Then the intersection
points P1 = A2 A3 ∩ B2 B3 , P2 = A1 A3 ∩ B1 B3 , P3 = A1 A2 ∩ B1 B2 are collinear.
Proof. The idea of the proof is as follows. First, we will show a 3-dimensional analogue
of the statement (and this will be short and easy part (a)). Then, in part (b) of
the proof, we will get the 2-dimensional statement as a limit of deformation of the
3-dimensional configuration.

(a) Let α be a plane in R3 containing points A1 , A2 , A3 , and β be a plane containing

points B1 , B2 , B3 . Let l = α ∩ β be the intersection line. And suppose that the
lines joining the corresponding vertices of triangles A1 A2 A3 and B1 B2 B3 intersect
at one point S . Then all intersection points Pk = Ai Aj ∩ Bi Bj obviously lie in
the intersection line l = α ∩ β (as Ai ∈ α, Bi ∈ β for all i = 1, 2, 3), see Fig. 48,
left.

59
 (b) - Now, we consider the 2-dimensional configuration (we place it into a hori-
zontal plane γ in 3-dimensional space).
- Let O ∈
/ γ be any point such that the plane OA2 B2 ⊥ γ „ see Fig. 48, right.
- Choose a point A′2 ∈ OA2 , and consider a point B2′ = OB2 ∩ S A2 .
- Consider the triangle A1 A′2 A3 and B1 B2′ B3 , denote the planes containing
them by α and β respectively. By part (a) of the proof, the three intersection
points constructed for these triangles lie on the line l = α ∩ β .
- Now, we start to move the point A′2 towards A2 . The planes α and β
approach the initial horizontal plane γ . The intersection line l = α ∩ β
approaches some line in γ . This line at the limit will be the line containing
all three points P1 , P2 , P3 ∈ γ .

B20

A02 B3
S A3
A2 A S
3
A2 B2
B3 A1 B1
A1 B2

B1
O
l

Figure 48: Proof of Desargues’ Theorem.

4.4 Topology and metric on RP2

Remark 4.20 (Topology of RP2 ). RP2 is a set of lines through O in R3 , in other words
RP2 = S2 / ∼ , i.e. the sphere with antipodal points identified, which is equivalent to
a disc with the opposite points identified.
It includes a Möbius band, so, it is one-sided and non-orientable.

Figure 49: Topology of RP2 .

60
Remark 4.21 (Elliptic geometry).

- As RP2 = S2 / ∼, one can use the spherical metric to introduce the metric on the
set of points of RP2 . Then RP2 with this metric will be locally isometric to S2 ,
i.e. a small domain on RP2 is isometric to a small domain on S2 .

- However, most projective transformations to not preserve this metric. So, this
metric is not a notion of projective geometry.

- The geometry of RP2 with spherical metric (and a group of isometries acting on
it) is called elliptic geometry and has the following properties:

(1) For any two distinct points there exists a unique line through these points;
(2) Any two distinct lines intersect at a unique point;
(3) For any line l and point p (which is not a pole for l) there exists a unique
line l′ such that p ∈ l′ and l ⊥ l′ .
(4) The group of isometries acts transitively on the points (and lines) of this
geometry.

Remark 4.22 (Conic sections).

- Quadrics, i.e. the curves of second order on R2 (such as ellipse, parabola and
hyperbola) may be obtained as conic sections (sections of a round cone by a
plane, see Fig. 50).

- Ellipse, parabola and hyperbola are equivalent under projective transformations

(to see this, one can use projections of one plane to another from the tip of the
cone).

- To find out more about conic sections see

V. Prasolov, V. M. Tikhomirov. Geometry, (2001). Chapter 4. Conics and

[Link] 4.1. Plane curves of second order. pp.61-69.
This will constitute Additional 4H reading and will be examinable for students
enrolled to Geometry IV and V.

Figure 50: Conic sections: ellipse, parabola and hyperbola.

61
4.5 Polarity on RP2 (NE)
(Non-examinable section!)

Consider a trace of a cone C = {(x, y , z ) ∈ R3 | x2 + y 2 = z 2 } on the projective plane

RP2 - a conic.

Definition. Points A = (a1 , a2 , a3 ) and B = (b1 , b2 , b3 ) of RP2 are called polar with
respect to C if a1 b1 + a2 b2 = a3 b3 .

Example:

1. Points of C are self-polar.

2. Point (2 : 1 : 2) is polar to (1 : 2 : 2).

Definition. Given a point A ∈ RP2 , the set of all points X polar A is the line
a1 x1 + a2 x2 − a3 x3 = 0, it is called the polar line of A.

Example. Let A = (0, 0, 1) - the North Pole of the sphere, then its polar is the line
defined by x3 = 0, i.e. all points with coordinates (a1 , a2 , 0). So, the line a1 x1 + a2 x2 = 0
is the polar line for the point A = (0, 0, 1).

How to find the polar line:

Lemma 4.23. A tangent line to C at a point B = (b1 , b2 , b3 ) is x1 b1 + x2 b2 = x3 b3 .

We skip the proof of the lemma.

Proposition 4.24. Let A be a point “outside” C, let lP and lQ be tangents to C at P

and Q, where P, Q ∈ C, s.t. A = lP ∩ lQ . Then P Q is the line polar to A.

Proof. As A ∈ lP , we have a1 p1 + a2 p2 = a3 p3 , so P is polar to A.

As A ∈ lQ , we have a1 q1 + a2 q2 = a3 q3 , so Q is polar to A.
Therefore, P Q is the line polar to A, see Fig. 51, left.

Proposition 4.25. If A ∈ C then the tangent lA at A is the polar line to A.

Proof. The line x1 a1 + x2 a2 = x3 a3 is tangent at A by Lemma 4.23 and is polar to A

by definition of the polar line.

Proposition 4.26. Let A be a point “inside” of the conic C. Let p and q be two lines
through A. Let P and Q be the points polar to the lines p and q . Then P Q is the line
polar to A with respect to C.

Proof. P is polar to A, Q is polar to A, hence, P Q is polar to A, see Fig. 51, right.

Remark 4.27. 1. Polarity generalise the notion of orthogonality.

2. More generally, for a conic C = {x ∈ R3 | xT Ax = 0}, where A is a symmetric

3 × 3 matrix, the point a is polar to the point b if aT Ab = 0.

62
 P lA P

lA lA A
A A

Q Q

Figure 51: Polar line lA for a point A inside, on and outside of the conic.

3. We worked with a diagonal matrix A = dag {1, 1, −1}.

4. If we take an identity diagonal matrix A = dag {1, 1, 1} we get an empty conic

x2 + y 2 + z 2 = 0, which gives exactly the same notion of polarity as we had on
S 2.
(Indeed, the point (a1 : a2 : a3 ) is polar to a1 x1 + a1 x2 + a3 x3 = 0 which is the
orthogonal plane (x, a) = 0).

4.6 Hyperbolic geometry: Klein model

Historic remarks:

• Parallel postulate (or Euclid’s Vth postulate) claims that

/ l, there exists a unique line l′ such that l||l′

Given a line l and a point A ∈
′
and A ∈ l .

• For centuries, people tried to derive Euclid’s Vth postulate from other postulates.

• In 1870s it turned out that Euclid’s Vth postulate is independent of others, i.e.
there exists a geometry where

- all other postulates hold;

- parallel postulate is substituted by
“Given a line l and a point A ∈ / l, there exists more than one (infinitely
many) line l such that l ∩ l = ∅ and A ∈ l′ .
′ ′

• Names:

- Gauss, Lobachevsky, Bolyai - derived basic theorems of hyperbolic geometry;

- Beltrami, Cayley, Klein, Poincaré - constructed various models.

Klein Model: in interior of unit disc.

- Points of the model are interior points of the unit disc;

- Lines are chords.

63
 - Distance between two points is defined by:
1
d(A, B ) = ln[A, B , X, Y ] ,
2
where

· X, Y are the endpoints of the chord through AB , see Fig. 52, left;
.
· [A, B , X, Y ] = ||X
X A|
B|
|Y A |
|Y B |
is the cross-ratio;
· |P Q| denotes the Euclidean length of the segment P Q.

X
B
A
Y

Figure 52: Klein model.

Remark:

1. Axioms of Euclidean geometry are satisfied in the model

(except for Parallel Axiom!).

2. Parallel Axiom is obviously not satisfied (see Fig. 52, right):

/ l, there are infinitely many lines l′ s.t. A ∈ l and
Given a line l and a point A ∈
l ∩ l ′ = ∅.

Remark: We will spend a large part of the next term looking at hyperbolic geometry.
Our closest aims are to show that

(1) The distance introduced above satisfies axioms of metric;

(2) Isometries act transitively on the points in this model.

Theorem 4.28. The function d(A, B ) satisfies axioms of distance, i.e.

(1) d(A, B ) ≥ 0 and d(A, B ) = 0 ⇔ A = B ;

(2) d(A, B ) = d(B , A);

(3) d(A, B ) + d(B , C ) ≥ d(A, C ).

64
Proof. (1) d(A, B ) ≥ 0 by definition.
Let us show that d(A, B ) = 0 if and only if A = B . Indeed,

d(A, B ) = 0 ⇔ ln[A, B , X, Y ] = 0 ⇔ [A, B , X, Y ] = 1

x − a.y − a x−a y−b
⇔ =1 ⇔ · = 1,
x−b y−b x−b y−a
where a, b, x, y are coordinates of the points A, B , X, Y on the line AB .
−b
Notice that xx− a
−b
≥ 1 and yy− a
≥ 1, which implies that the product of these
numbers equal to 1 if and only if both of them are equal to 1, which is equivalent
to the condition a = b, i.e. A = B .

(2) d(A, B ) = d(B , A) since [A, B , X, Y ] = −[B , A, Y , X ] (which we know from

HW 7.8).

(3) We are left to show the triangle inequality d(A, B ) + d(B , C ) ≥ d(A, C ), this will
be done in Lemma 4.30 below.

Remark 4.29. On hyperbolic line:

- Let [y , x] ∈ R be an interval. For a, b ∈ [y , x] (as in Fig. 52, left) we define

1 1 x − ay − a
d(a, b) = ln[a, b, x, y ] = ln( .
2 2 x−b y−b
Notice that the logarithm makes sense as the argument is positive for all
a, b ∈ [y , x].

- d(a, a) = 0.

- d(a, b) → ∞ when b → x or a → y .

- Since ln[a, b, x, y ] = −ln[a, b, y , x] (as [a, b, x, y ] = 1/λ when [a, b, y , x] = λ), we

conclude that the endpoints X and Y are “equally good”, i.e. the line is not
oriented.

- For c ∈ [y , x] we have ±d(a, b) ± (b, c) ± d(c, a) = 0, since

x−a y−b x−b y−c x−c y−a

   
· · · = 1.
x−b y−a x−c y−b x−a y−c

- If c ∈ [a, b] then d(a, c) + d(c, b) = d(a, b).

Lemma 4.30 (Triangle inequality). Let A, B , C be three points in Klein model. Then
d(A, B ) + d(B , C ) ≥ d(A, C ).

Proof. (1) We start the proof with the following additional construction:

- Extend the sides of the triangle AB C till the boundary of the disc to obtain
the chords X Y , X1 Y2 and Y2 X1 respectively (see Fig. 53, left).
- Define P := X1 X2 ∩ Y1 Y2 .
- Define X ′ = Y X ∩ X1 X2 and Y ′ = Y X ∩ Y1 Y2 .

65
 P

Y1 X1

ε
Y0 X
X0 Y0 A C0 X0 X
Y 0 B
A C

Y2 X2

Figure 53: Proof of triangle inequality.

- Define C ′ = P S ∩ X Y ∈ [AB ].

(2) Consider the projection from P to the segment X Y . As it preserves cross-ratios,

we get [A, C, X1 , Y2 ] = [A, C ′ , X ′ , Y ′ ] and [C, B , X2 , Y1 ] = [C ′ , B , X ′ , Y ′ ].

(3) Claim: [A, C ′ , X ′ , Y ′ ] > [A, C ′ , X, Y ] and [C ′ , B , X ′ , Y ′ ] > [C ′ , B , X, Y ].

Proof of the claim. We need to move the endpoints of the segments to the outside
of the segment. We will show [A, C ′ , X ′ , Y ′ ] > [A, C ′ , X, Y ′ ] and then applying
similar movement (i.e. shifting Y to Y ′ ) we will get the statement.
Let a, c′ , x′ , y ′ , x denote the coordinates of the points A, C ′ , X ′ , Y , X and suppose
x − x′ = ε, see Fig. 53, right. Then

y ′ − c′ x′ − a x′ − a + ε
 
′ ′ ′ ′ ′
[a, c , x , y ] − [a, c , x, y ] = ′ −
y −a x ′ − c′ x ′ − c′ + ε
y ′ − c′ ε(c′ − a)
= ′ > 0,
y − a (x′ − c)(x′ − c′ + ε)

which proves the claim.

66
 (4) Finally, we compute:

def 1 1
d(A, C ) + d(C, B ) = ln[A, C, X1 , Y2 ] + ln[C, B , X2 , Y1 ]
2 2
1
= ln([A, C, X1 , Y2 ] · [C, B , X2 , Y1 ])
2
(2) 1
= ln([A, C ′ , X ′ , Y ′ ] · [C ′ , B , X ′ , Y ′ ])
2
(3) 1
> ln([A, C ′ , X, Y ] · [C ′ , B , X, Y ])
2
1 x − a y − c′ x − c′ y − b
= ln( · · · )
2 x − c′ y − a x − b y − c′
1
= ln[a, b, x, y ] = d(A, B ).
2

Isometries of Klein model

By isometries we mean transformations of the model preserving the distance, i.e. pre-
serving the disc and the cross-ratios.

β
E

α
D

Figure 54: To the proof of Theorem 4.31.

Theorem 4.31. There exists a projective transformation of the plane that

- maps a given disc to itself;

- preserves cross-ratios of collinear points;

- maps the centre of the disc to an arbitrary inner point of the disc.
Proof. We will give a sketch of a proof here.
1. Let C be the cone x2 + y 2 = z 2 , let the disc D = C ∩ α be the horizontal section
of the cone C by a plane α defined by z = const.

2. Let β be a plane s.t. β ∩ C is an ellipse E , see Fig. 54, left.

3. Let P be the projection of the disc D to the plane β from the apex S of the cone:
the projection takes the disc D to the ellipse E , this map is a projective trans-
formation (due to Corollary 4.13).

67
 4. Let i ∈ I som(E3 ) be an isometry such that i(β ) = α, suppose also that i takes
the centre of the ellipse to the centre of the disc D.

5. Consider an affine transformation A of the plane α which takes the ellipse i(E )
to the disc D.

6. Then the composition A ◦ i ◦ P takes D to D. The map A ◦ i takes the centre of

the ellipse to the centre of the disc, while P (0) lies as far from the centre as we
want depending on the choice of plane β , see Fig. 54, right.

7. The map A ◦ i ◦ P is a projective transformation, as it is the composition of

a projective transformation, isometry and affine transformation (i.e. of three
projective maps), see Fig. 55.

i
P isom A
proj aff

Figure 55: To the proof of Theorem 4.31.

Corollary 4.32.

- Isometries act transitively on the points of Klein model.

- Isometries act transitively on the flags in Klein model.

Proof. The theorem shows transitivity on points. To show transitivity on flags one
can:

- map a given point to the centre of the disc;

- then rotate the disc about the centre (it is an isometry in the sense of the model,
since it clearly preserves all cross-ratios, and hence preserves the distance).

- reflect the disc (in Euclidean sense) with respect to a line through O (again, it is
an isometry as cross-ratios are preserved).

Remark.

1. In general, angles in Klein model are not represented by Euclidean angles.

68
 2. Angles at the centre are Euclidean angles.
Indeed, two orthogonal (in Euclidean sense) chords make equal hyperbolic angles
(as one can take one of them to another by an isometry of the hyperbolic plane),
so, these angles are π /2. Similarly, all (Euclidean) angles of size π /n, n ∈ Z
represent hyperbolic angles of size π /n, and moreover, the angles coincide with
Euclidean ones for all π -rational angles. Finally, by continuity we conclude that
all angles at the centre of the disc coincide with Euclidean angles.

3. Right angles are shown nicely everywhere in the Klein model (see Proposition 4.33).

Proposition 4.33. Let l and l′ be two intersecting lines in the Klein model. Let t1
and t2 be tangent lines to the disc at the endpoints of l. Then l ⊥ l′ ⇔ t1 ∩ t2 ∈ ˜l′ ,
where ˜l′ is the Euclidean line containing the chord representing l′ .

Proof. - We know that at the centre of the disc right angles are shown by two
perpendicular diameters l0 and l0⊥ . Consider the lines p1 , p2 tangent to the disc
at the endpoints of l0 , see Fig. 56, left. Then l0⊥ is the line through O parallel
to the lines p1 , p2 . In other words, l⊥ is the line through O and the intersection
p1 ∩ p2 (which does not exist in E2 but is well-defined in RP2 .

- Let f be a projective transformation which maps the disc to itself, takes l0 to

l and O to l ∩ l′ (it does exist in view of Corollary 4.32). Notice that f is an
isometry of the model (as it preserves the disc and the cross-ratios). Hence, it
takes a pair of perpendicular lines to perpendicular (in the sense of hyperbolic
geometry) lines.

- Notice that the lines f (p1 ) = t1 and f (p2 ) = t2 are the tangent lines to the disc
at the endpoints of l (indeed, they should contain the endpoints of l but should
only have one intersection with the disc, being the images of the tangent lines
p1 and p2 ). So, f (l0⊥ ) is the line through f (O) and f (p1 ) ∩ f (p1 ), which exactly
means that l′ ⊥ l if and only if it passes through t1 ∩ t2 . See Fig. 56, right.

p1 l0⊥ p2 l0
t1
l0

l
t2

Figure 56: Right angles in the Klein model.

69
Pairs of lines in hyperbolic geometry: two lines in hyperbolic geometry are called

- intersecting if they have a common point inside hyperbolic plane;

- parallel if they have a common point on the boundary of hyperbolic plane;

- divergent or ultra-parallel otherwise.

Figure 57: Pairs of lines in the Klein model: intersecting, parallel and ultra-parallel.

Proposition 4.34. Any pair of divergent lines has a unique common perpendicular.

Proof. See Fig. 58.

l2
l1

Figure 58: Common perpendicular for any ultra-parallel lines l1 and l2 .

70
4.7 References
- Sections 4.1 and 4.2 (on projective line and projective plane) closely follow Lec-
ture II and Lecture III of
V. V. Prasolov, Non-Euclidean Geometry.
You can find the same material in Section 3.1 of
V. V. Prasolov, V. M. Tikhomirov, Geometry.

- Section 4.3 “Some classical theorems” follows the section on Pappus’ and Desar-
gues’ theorems in Chapter 3 of
V. V. Prasolov, V. M. Tikhomirov, Geometry.

- Most part of the material of Sections 4.4 and 4.5 (topology of projective plane
and polarity on projective plane) may be found in Part II of
E. Rees, Notes on Geometry, Universitext, Springer, 2004.
(the book is available on DUO in Other Resources).

- Section 4.6 follows Lecture IV of Prasolov’s book (or see pp.89-93 in Prasolov,
Tikhomirov).

- A very nice overview of projective geometry is provided by

R. Schwartz, S. Tabachnikov, Elementary Surprises in Projective Geometry

- Elliptic geometry is briefly described in

A. B Sossinsky, Geometries, Providence, RI : American Mathematical Soc. 2012.
One can find the book in the library, see also Section 6.7 (p.75) here.

- A very nice course on projective geometry is

N. Hitchin, Projective Geometry
(the notes are available on DUO in Other Resources).

- For an overview of history of non-Euclidean geometry see

A. B Sossinsky, Geometries, Providence, RI : American Mathematical Soc. 2012.
One can find the book in the library, see also Chapter 13 (p.119) here.

- Video:

- Why slicing cone gives an ellipse - video on Grant Sanderson’s YouTube

channel 3Blue1Brown.

71
5 Möbius geometry
Hierarchy of geometries
By now, we have considered a number of geometries - Euclidean, spherical, affine,
projective, even a bit of hyperbolic. But how are they related to each other?
One answer to this is given by Arthur Caley: “Projective geometry is all geometry.”
And indeed, as one can see from Fig. 59, Euclidean, affine spherical and the Klein
model of hyperbolic geometry are all subgeometries of projective geometry.

E3

S2 E2 H2

Sym(2) M öb

Af f (2)

RP(2)

Figure 59: Hierarchy of geometries

At the same time, when hyperbolic geometry is considered in the Klein model, it
allows to nicely see the lines, but is not very convenient for working with angles, which
are not represented well there. Our first aim now will be to consider Möbius geometry
- geometry of linear fractional maps on C which are angle-preserving. This geometry
will provide other models for hyperbolic geometry - the models where the lines look
more complicated, but the angles are just Euclidean angles.
(And Möbius geometry will not be a part of projective geometry - so, projective
geometry is not all geometry after all!).

5.1 Group of Möbius transformations

Definition 5.1. A map f : C ∪ {∞} → C ∪ {∞} given by f (z ) = az +b
cz +d
, a, b, c, d ∈ C,
ad − bc ̸= 0 is called a Möbius transformation or a linear-fractional transformation.

Remark. It is a bijection of the Riemann sphere C = C ∪ {∞} to itself.

Theorem 5.2. (a) Möbius transformations form a group (denoted Möb) with respect
to the composition, this group is isomorphic to

P GL(2, C) = GL(2, C)/{λI | λ ̸= 0}.

(b) This group is generated by z → αz , z → z + 1 and z → 1/z , where α, β ∈ C.

72
  
a b
Proof. (a) Given a matrix A = ∈ GL(2, C), let fA (z ) = az +b
cz +d
. In this way we
c d
can obtain any Möbius transformation. Moreover, since λaz +λb
λcz +λd
= az +b
cz +d
, we may
assume that ad − bc = ±1. Furthermore, we get a bijection between elements of
P GL(2, C) and linear-fractional maps. It is straight-forward to check that this
bijection respects the group structure, i.e.

fB ◦ fA = fB A .

(b) Consider any linear-fractional transformation f = az +b

cz +d
. We can write
a ad
az + b (cz + d) + b − a bc − ad
f (z ) = = c c
= − = f3 ◦ f2 ◦ f1 (z ),
cz + d cz + d c c(cz + d)

where
1 bc − ad a
f1 (z ) = cz + d, f2 (z ) = , f3 (z ) = z+ .
z c c
Clearly, each of f1 , f2 , f3 can be obtained as a composition of transformations
z → αz , z → z + β and z → 1/z . Furthermore, z → z + β = β ( βz + 1) is a
composition of z → αz and z → z + 1. So, we conclude that f (and, hence,
any linear-fractional transformation) is a composition of z → αz , z → z + 1 and
z → 1/z .

Example 5.3. The generators az , z + 1 and 1/z (a, b ∈ C) can be represented by

matrices      
a 0 1 1 0 1
, and
0 1 0 1 1 0
respectively.

Theorem 5.4. (a) Möbius transformations act on C ∪ {∞} triply-transitively.

(b) A Möbius transformation is uniquely determined by the images of 3 points.

Proof. We need to construct a map f ∈ M ob ¨ taking three given distinct points z1′ , z2′ , z3′
in C ∪ {∞} to any other three given distinct points z1 , z2 , z3 . We will construct a Möbius
transformation f0 : (0, 1, ∞) → (z1 , z2 , z3 ). Then f = f0 ◦ g0−1 , where g0 : (0, 1, ∞) →
(z1′ , z2′ , z3′ ).
Construction: we will construct f0 = az +b
cz +d
.

- We will assume that z1 , z2 , z3 ̸= ∞, otherwise, we will precompose with 1/(z + d).

- f0 (0) = b/d = z1 , which is equivalent to b = z1 d.

- f0 (∞) = a/c = z3 , which is equivalent to a = z3 c.

(z2 −z1 )d
- Hence, f0 (1) = z3 c+z1 d
c+d
= z2 , and we get c = (z3 −z2 )
.

- We have obtained a, b, c (all of them proportional to d), so we can cancel d (i.e.

assume d = 1) to get to get representative of f0 which takes (0, 1, ∞) to the
required points.

73
 - The constructed map is a Möbius transformation since
z2 − z1 2
ad − bc = z3 cd − z1 cd = (z3 − z1 ) d ̸= 0,
z3 − z2
(as zi ̸= zj by assumption). This proves part (a).

Uniqueness of the Möbius transformation f0 : (0, 1, ∞) → (z1 , z2 , z3 ) follows imme-

diately from the computation. If there are two maps f and h taking (z1′ , z2′ , z3′ ) →
(z1 , z2 , z3 ) then f ◦ g0 and h ◦ g0 are two maps taking (0, 1, ∞) → (z1 , z2 , z3 ), which is
impossible. This implies part (b).

Theorem 5.5. Möbius transformations

(a) take lines and circles to lines and circles;

(b) preserve angles between curves.

Proof. It is sufficient to check the statements for the generators:

• z → az : is a rotation about 0 by argument of a composed with a dilation by |a|;

• z → z + 1: translation by 1;

• z → 1/z : composition of a reflection z → z¯ and an inversion z → 1/z¯

(see Fig. 60, left for the action of z → 1/z¯).

All these transformations satisfy (a) and (b) (for z → 1/z recall the results from
Complex Analysis II - we will also show it independently below in Theorems 5.14
and 5.15).

Example.

1. See Fig. 60, left, for the action of z → 1/z¯.

2. Transformation z → 1/z takes the real line to itself and the circle (z − 21 )2 = ( 12 )2
to the line Re z = 1. The right angle between these two curves is preserved (see
Fig. 60, right).

z 1 x=1
1 z
z̄
0 1 0

z̄

Figure 60: Left: Transformation z → 1/z¯. Right: an angle preserved by z → 1/z .

74
5.2 Types of Möbius transformations
Consider the fixed points of the transformation f (z ) = az +b
cz +d
, i.e., the points satisfying

az + b
z= .
cz + d
This is a quadratic equation with respect to z , so it has exactly two complex roots
(these roots may coincide, in which case f has a unique fixed point).

Definition 5.6. A Möbius transformation with a unique fixed point is called parabolic.

Example: z → z + b, where b ∈ C∗ is a parabolic transformation (with a unique fixed

point ∞).

Proposition 5.7. Every parabolic Möbius transformation is conjugate in the group

¨ to z → z + 1.
M ob

Proof. Suppose that f is a parabolic transformation with z0 = f (z0 ).

Let g (z ) = z−1z0 , notice that g (z0 ) = ∞. Then the transformation

f1 (z ) := g ◦ f ◦ g −1 (z )

has a unique fixed point at ∞ (here we use the same reasoning as in Proposition 1.18(a)).
This implies that f1 (z ) = az +b
cz +d
with c = 0 (as f1 (∞) = ∞). By scaling a and b we may
assume f1 (z ) = az + b. Since f1 has a (double) root at infinity (and no other roots) we
see that the equation z = az + b, has the only solution z = − a−b 1 at infinity, which is
only possible when a = 1. We conclude that f1 = z + b, so f is conjugate to z → z + b.
Finally, let h(z ) = bz . Then
1
f2 (z ) := h−1 ◦ f1 (z ) ◦ h(z ) = (bz + b) = z + 1,
b
So, we conclude that f is conjugate to z → z + 1.

Proposition 5.8. Every non-parabolic Möbius transformation is conjugate in M ob

¨
to z → az , a ∈ C \ {0}.

Proof. Let z1 , z2 be the fixed points of a Möbius transformation f . The transformation

− z1
g (z ) = zz− z2
sends them to 0 and ∞. So, f1 (z ) = g f g −1 (z ) has fixed points at 0, ∞.
Hence, f1 (z ) = az , and we see that f is conjugate to z → az .

Definition 5.9. A non-parabolic Möbius transformation conjugate to z → az is called

(1) elliptic, if |a| = 1;

(2) hyperbolic, if |a| ̸= 1 and a ∈ R;

(3) loxodromic, otherwise.

75
Remark 5.10. Consider the dynamics of various types of elements when they are
iterated many times (see Fig. 61). We draw each type twice: in the first row all fixed
points a visible, while in the second row one fixed point is mapped to ∞ (but the
picture is more simple).
Parabolic elements are best understood when the fixed point is ∞ - then it is transla-
tion of all points by the same vector. Applying a Möbius transformation we see that
iterations of such a transformation move points a long circles through the fixed point.
All other elements are best viewed when the fixed points are 0 and ∞:
elliptic elements just rotate points around two equally good fixed points, while hyperbolic
and loxodromic elements have one one attracting fixpoint and one repelling.
Two fixpoints of a hyperbolic or a loxodromic transformation have different prop-
erties: one is attracting another is repelling.
Elliptic transformations have two similar fixpoints (neither attracting nor repelling).

Parabolic Elliptic Hyperbolic Loxodromic

Figure 61: Dynamics of parabolic, elliptic, hyperbolic and loxodromic elements.

Dynamics of Möbius transformations is nicely illustrated in the 2-minute video by

Douglas Arnold and Jonathan Rogness.

5.3 Inversion
Definition 5.11. Let γ ∈ C be a circle with centre O and radius r. An inversion Iγ
with respect to γ takes a point A to a point A′ lying on the ray OA s.t. |OA| · |OA′ | = r2 ,
see Fig. 62.

Proposition 5.12. (a) Iγ2 = id.

(b) Inversion in γ preserves γ pointwise (Iγ (A) = A for all A ∈ γ ).

Proof. This immediately follows from the definition.

Lemma 5.13. If P ′ = Iγ (P ) and Q′ = Iγ (Q) then △OP Q is similar to △OQ′ P ′ .

76
 A

A0
r

Figure 62: Inversion: |OA| · |OA′ | = r2 .

Proof. Since |OP | · |OP ′ | = r2 = |OQ| · |OQ′ | we have

|OP | |OQ′ |
= ,
|OQ| |OP ′ |

see Fig. 63. As ∠P OQ = ∠P ′ OQ′ , we conclude that △P OQ ∼ △Q′ OP ′ (by sAs).

P0

O Q0 Q

Figure 63: Inversion: △OP Q ∼ △OQ′ P ′ .

Theorem 5.14. Inversion takes circles and lines to circles and lines. More precisely,

1. lines through O are mapped to lines through O;

2. lines not through O are mapped to circles through O

3. circles not through O are mapped to circles not through O.

Proof. Consider an inversion Iγ with respect to a circle γ .

1. This part is evident from the definition.

2. Let l be a line, O ∈
/ l. Let Q ∈ l be a point such that OQ ⊥ l, see Fig. 64. Let
P ∈ l be any point of l and let P ′ = I γ (P ), Q′ = Iγ (Q).
By Lemma 5.13, △P OQ ∼ △Q′ OP ′ , so ∠OP ′ Q′ = π /2. This implies that P ′
lies on the circle with diameter OQ′ (by converse of E26). This implies that Iγ (l)
is the circle with the diameter OQ′ .

77
 P

P0

O Q0 Q
l

Figure 64: Inversion takes lines not through origin to the circles through origin.

3. Let γ0 be a circle O ∈ / γ0 . Let l be a line through O and the centre of γ0 .

Let {P, Q} = l ∩ γ0 , R ∈ γ0 , and let Iγ takes the points P, Q, R to P ′ , Q′ , R′
respectively, see Fig. 65.
Be Lemma 5.13, we have ∠OP R = ∠OR′ P ′ which implies ∠RP Q = ∠P ′ R′ R.
Also, we have ∠OQR = ∠OR′ Q′ . Since P Q is the diameter of γ0 , we have
∠P RQ = π /2, which implies that ∠RP Q + ∠OQR = π /2. Therefore, ∠Q′ R′ P ′ =
π /2, and hence, R′ lies on the circle with diameter Q′ R′ .

R0
l
O Q0 P0 P Q

γ0

Figure 65: Inversion takes circles not through origin to the circles not through origin.

Theorem 5.15. Inversion preserves angles.

Proof. Let Iγ be the inversion with respect to the circle γ . Let l be a line such that
O∈ / l, see Fig. 66. Then Iγ (l) is a circle γ through O and the tangent line to γ line at
the point O is parallel to l (one can see it for example from the symmetry with respect
to the line orthogonal to l dropped from O). This implies that if l1 , l2 are two lines not
through the origin, then the angle between them is preserved by the inversion.
For two circles (or a line and a circle) we measure the angles between tangent lines
to them (and this angle is preserved as shown above).
If one or both of l1 , l2 pass through O then it the image of such line is still parallel
to initial line, so the angle is still preserved.

Remark. Inversion may be understood as “reflection with respect to a circle”.

78
 l0

γ1
l

Figure 66: Inversion preserves angles.

Example 5.16. Let I1 be inversion with respect to the unit circle √ centred at the
origin, and I 2 be the inversion with respect to the circle of radius 2 centred at −i,
√

see Fig. 67. Notice that I√2 takes the unit circle to the real line.

−1 0 1

−i

Figure 67: Reflection is a conjugated inversion.

Define r := I√2 I1 I√2 . Then r(x) = x for every x ∈ R, and it is easy to see that r
swaps the half-planes defined by the real line.
As r is a composition of inversions, it preserves the angles, which (together with
preserving all points of real line) implies that r is a reflection, see Fig. 68.
Theorem 5.17. Every inversion is conjugate to a reflection by another inversion.
Proof. As in Example 5.16, given an inversion Iγ with respect to a circle γ , consider
an inversion I ′ with respect to a circle forming angle π /4 with γ : then I ◦ Iγ ◦ I is a
reflection.

Theorem 5.18. Every Möbius transformation is a composition of even number of

inversions and reflections.
Proof. By Theorem 5.2 Every Möbius transformation is a composition of transforma-
tions az , z + 1, 1/z . We will check that each of these transformations is a composition
of inversions and reflections.

79
 r

Figure 68: r preserves all real points and preserves angles, hence r is a reflection.

- az = |a|ei Arg a z is a composition of a dilation and a rotation. A rotation by angle

α is a composition of reflections with respect to the lines meeting at angle α/2 (by
Example 1.13). Dilation D : z → r2 z (r ∈ R) is a composition of two inversions
I1 and Ir with respect to circles of radius 1 and r centred at the origin:
1
Ir ◦ I1 (z ) = Ir ( ) = r2 z = D(z ).
z¯

- Translation z + 1 is a composition of two reflections (again by Example 1.13).

- The map f : z → 1/z is a composition of an inversion 1/z¯ and a reflection z¯:

f = z¯ ◦ z1¯ .

Notice that we described each of the generators as a composition of even number

of reflections and inversions.

Remark 5.19. Inversion and inversion change orientation of the plane, but Theo-
rem 5.18 says that a Moöbius transformation is expressed through even number of
them. Hence, it shows that Möbius transformations preserve orientation.

See here for an animation demonstrating properties of inversion (by M. Christersson).

5.4 Möbius transformations and cross-ratios

Definition 5.20. For z1 , z2 , z3 , z4 ∈ C ∪ {∞}, the complex number
z3 − z1 z4 − z1
[z, z2 , z3 , z4 ] = / ∈ C ∪ {∞}
z3 − z2 z4 − z2
is called the cross-ratio.

Theorem 5.21. Möbius transformations preserve cross-ratios.

Proof. This is an easy computation for each of the generators az , z + 1, 1/z (check!).

Corollary 5.22. A Möbius transformation is determined by images of 3 points.

Proof. If f ∈ M ob
¨ , f : a, b, c → a′ , b′ , c′ and y = f (x), then y can be computed from
the linear equation [a, b, c, x] = [a′ , b′ , c′ , y ].

80
Remark 5.23. Points z1 , z2 , z3 ∈ C are collinear if and only if z1 −z2
z1 −z3
∈ R (i,e. when
vectors z1 − z2 and z1 − z3 are proportional over R).

Proposition 5.24. Points z1 , z2 , z3 , z4 ∈ C ∪ {∞} lie on one line or circle if and only
if [z1 , z2 , z3 , z4 ] ∈ R.

Proof. By Theorem 5.4 there exists a Möbius transformation f which takes z1 , z2 , z3

to 0, 1, ∞. Let x ∈ C and y = f (x). It is easy to see (using Remark 5.23) that y lies
on a real line if and only if [0, 1, ∞, y ] is real. Hence, x lies on the same line or circle
as z1 , z2 , z3 if and only if [z1 , z2 , z3 , z4 ] ∈ R.

Remark 5.25. Geometric proof of Proposition 5.24:

Consider 4 points on the same circle. By E28, ∠z1 z4 z2 = ∠z1 z4 z3 , which means
that Arg( z3 −z2 ) = Arg( z4 −z2 ). This implies that [z1 , z2 , z3 , z4 ] = z3 −z2 z3 −z2 ∈ R.
z3 −z1 z4 −z1 z3 −z1 z3 −z1


Conversely, if z4 does not lie on the same circle as z1 , z2 , z3 , then the angles at z3 and
z4 are different and the cross-ratio is not real.

z3

z1

z2 α
z4

Figure 69: Geometric meaning of real cross-ratio.

Proposition 5.26. Given four distinct points z1 , . . . , z4 ∈ C ∪ ∞, one has

[z1 , z2 , z3 , z4 ] ̸= 1.

Proof. Suppose that [z1 , z2 , z3 , z4 ] = 1. Then by Proposition 5.24 the points lie on one
line or circle, so we may assume that [z1 , z2 , z3 , z4 ] = [x,0, 1, ∞], where x ∈ R (here we
use triple transitivity of M ob
¨ ). So, [z1 , z2 , z3 , z4 ] = 11− x ∞−x
−0 ∞−0
= 1 − x, this only equals
to 1 when x = 0, which is impossible as the points z1 , z2 , z3 , z4 (and hence, the points
x, 0, 1, ∞) are distinct by assumption.

Example 5.27. (a) Two parallel lines are not M ob ¨ -equivalent to two concentric cir-
cles (as circles are disjoint while lines are tangent at ∞, i.e. sharing one point).

(b) Let lx be a line given by Re(z ) = x, x ∈ R. Is there a Möbius transformation

taking l0 , l1 , l2 to l0 , l1 , l3 ?

81
 To answer the question consider a line or circle γ orthogonal to all three of l0 , l1 , l2 .
It is easy to see that γ is a line orthogonal to li (justify this!). Let A, B , C, D be
the points where γ intersects respectively l0 , l1 , l2 (where D = ∞). Then
2 − 0∞ − 0
[A, B , C, D] = [0, 1, 2, ∞] = = 2/1 = 2.
2−1 ∞−1
Similarly, let A′ , B ′ , C ′ , D′ be the points where γ intersects respectively l0 , l1 , l3
(where D′ = ∞). Then
3 − 0∞ − 0 3
[A′ , B ′ , C ′ , D′ ] = [0, 1, 3, ∞] = = .
3−1 ∞−1 2
As for λ = 2 none of λ, 1 − λ, λ1 , 1−1 λ , 1−−λλ , 1−−λλ coincides with 32 , we conclude that
there is no Möbius transformation taking l0 , l1 , l2 to l0 , l1 , l3 .
Remark 5.28. Does reflection/inversion preserve cross-ratio?
Example: reflection z → z¯ takes the cross-ratio [z1 , z2 , z3 , z4 ] to [z1 , z2 , z3 , z4 ].
The same will happen for every inversion/reflection f : we can find a Möbius trans-
formation g which takes F ixf to R ∪ {∞}, then f = g ◦ z¯ ◦ g −1 , where g preserves
cross-ratios, and z¯ conjugates.
We conclude that inversions and reflections take cross-ratios to conjugate numbers.
Corollary 5.29. Cross-ratios of four points lying on a line or a circle are preserved
by inversions and reflections.

5.5 Inversion in space

Definition 5.30. Let S ∈ R3 be a sphere of radius r centred at O. The inversion IS
with respect to S takes a point A ∈ R3 to a point A′ on the ray OA s.t. |OA|·|OA′ | = r2 ,
see Fig.70.

A0
r
O

Figure 70: Inversion in space: |OA| · |OA′ = r2 .

Theorem 5.31 (Properties of inversion).

(1) Inversion takes spheres and planes to spheres and planes.

(2) Inversion takes lines and circles to lines and circles.

(3) Inversion preserves angles between curves.

|C A | | D A |
(4) Inversion preserves cross-ratio of four points [A, B , C, D] = /
|C B | |D B |
.

82
Proof. Let I be inversion in the sphere S centred at O of radius r.

(1) Let l be a line through the centre O of S orthogonal to a plane or sphere α,

see Fig. 71. Let Π be any plane containing l. Restricting I to Π we obtain an
inversion on the plane Π. So the circle or line Π ∩ α will be mapped to some
circle of line q . Now, notice that all data (including the definition of inversion,
the sphere S , the line l and the set α) are preserved by rotation around l. So, if
we rotate Π, we will see exactly the same image of Π ∩ α as in Π (but rotated).
Rotating a circle or a line around the line l through the centre of the circle (or a
point on the line) we get a sphere or a plane.

P
P0

Figure 71: Inversion in space: rotating the circle get a sphere.

(2) A line is an intersection of two planes, a circle is an intersection of a plane and a

circle. Applying the result of part (1) we see that the image of a line of a circle
is a line or a circle.

(3) Similarly to 2-dimensional case, a line l is mapped by an inversion to a circle

through O with a tangent line at O parallel to l.

(4) Let P ′ = I (P ) and Q′ = I (Q). Then in the same way as in 2-dimensional case
we see that △OP Q ∼ △OQ′ P ′ , one can see it by considering the restriction of
I to to the plane OP Q (or to any plane containing O, P, Q if they are collinear).
Therefore,
|P Q| |OP | |OP | |OQ| |OP | · |OQ|
= = · = .
|P ′ Q′ | |OQ′ | |OQ| |OQ′ | r2
This implies that
| C A| |D B | |OA|·|OB |·|OC |·|OD|
[A, B , C, D] |C ′ A ′ |
· |D ′ B ′ | r4
= |C B | |D A |
= |OA|·|OB |·|OC |·|OD|
= 1.
[A′ , B ′ , C ′ , D′ ] ·
|C ′ B ′ | | D ′ A′ | r4

5.6 Stereographic projection

Definition 5.32. Let S be a sphere centred at O, let α be a plane through O. Let
N ∈ S be a point with N O ⊥ α. The map π : S → α s.t. π (A) = α ∩ N A for all
A ∈ S is called a stereographic projection, see Fig. 72, left.

83
 N

A
N
α A
A0 A0 O

Figure 72: Stereographic projection: definition (left) and as a restriction of inversion

(right).

Proposition 5.33 (Properties of stereographic projection).

(a) Stereographic projection takes circles to circles and lines.

(b) Stereographic projection preserves angles.

(c) Stereographic projection preserves cross-ratio.

Proof. Properties (a)-(c) listed in the proposition are properties (2)-(4) of inversion
with respect to sphere. Hence, it is sufficient to show that the stereographic projection
is a restriction of some inversion to the sphere. √
For the unit sphere S centred at the origin consider the sphere S√2 of radius 2
centred a the point N , see Fig. 72, right. Notice that S√2 makes angle π /4 both with
the sphere S and the plane α. This means that the inversion IS√2 maps S to α, which
implies that IS√2 (A) = A′ = N A ∩ α.

Remark. Another argument to see that IS√2 maps A to A′ is based on similarity of

triangles △A′ ON ∼ △QAN , where Q = S ∩ ON (the triangles are similar by AAA).
Then
|A′ N | |ON |
= ,
2|ON | |N A|
√
which implies that |A′ N | · |N A| = 2|ON |2 = ( 2|ON |)2 .

Remark. We used stereographic projection when proved that the formula for the area
of spherical triangle S△ = α + β + γ − π , see Fig. 28.

Remark. Another way to define stereographic projection, is to project from N to the

plane α tangent to S at point opposite to N ; see Fig 74, left. This projection has the
same properties as the one in Definition 5.32. To see, that the properties of the new
projection are the same, just notice that the images on the two planes only differ by a
scaling (projection from N).

Remark. This version of stereographic projection is nicely illustrated in a 1-minute

video by Henry Segerman.

84
 N

α
0
A

Figure 73: Another way to define stereographic projection.

Remark. Why do we need all these properties of 3-dimensional inversions and stere-
ographic projection? We used 3-dimensional inversion to show the properties of the
stereographic projection, and the later will be used to show that different models of
H2 give rise to the same geometry.

Example 5.34 (Steiner Porism). A circle γ1 lies inside another circle γ2 . A circle C0
is tangent to both γ1 and γ2 . A circle Ci is tangent to three circles: γ1 , γ2 and Ci−1 , for
i = 1, 2, 3 . . . . It may happen that either all circles Ci , i ∈ N are different, or Cn = C1
for some n. Show that the outcome does not depend on the choice of the initial circle
C0 (but only depends on γ1 and γ2 ).

Proof. First, we need to show that every two disjoint circles are Möbius-equivalent to
two concentric circles. This can be done using an appropriate sequence of inversions.
We will skip the proof here, as we will see another, shorter explanation of this later
based on hyperbolic geometry (see Example 6.18 below).
Once the circles γ1 and γ2 are mapped to concentric circles, the statement follows
trivially (as any choice of initial circle C0 may be transformed to any other choice by a
rotation around the common centre of the concentric circles).

Figure 74: Steiner Porism.

85
5.7 References
- One can read about hierarchy of geometries (with more examples than we had
at the start of Section 5) in
A. B Sossinsky, Geometries, Providence, RI : American Mathematical Soc. 2012.
One can find the book in the library, see also Section 1.4 (pp.119–124) here.

- Our exposition of Möbius transformations, Inversion and Stereographic projec-

tion is expansion of Lecture V in
V. V. Prasolov, Non-Euclidean Geometry.
You can find the same material on pp.93-95 of
V. V. Prasolov, V. M. Tikhomirov, Geometry.

- Another exposition about Möbius transformations is contained it Section 1.1 of

Hyperbolic Geometry by Caroline Series.

- For introduction and discussion of inversion see the following sources

- Malin Christersson, Circle inversion (Illustrated introduction with proofs).

- Tom Davis, Inversion in a circle. (An article showing how to use inversions
(but not giving the proofs of basic properties of inversion).

- I have borrowed the “Proof without words” for Ptolemy theorem (you can find it
in the Problems Classes notes) from the cut-the-knot portal. Which in its turn
refers to the following paper:

- W. Derrick, J. Herstein, Proof Without Words: Ptolemy’s Theorem,

The College Mathematics Journal, v. 43, n. 5, November 2012, p 386.

- Animations:

- Loxodromic transformation in the page by Paul Nylander. Also, scroll down

to find the animation of the coloured version.

- Videos:

- Dynamics of Möbius transformations is illustrated in the 2-minute

video by D. Arnold and J. Rogness.
- Animation demonstrating Inversion in circles, by M. Christersson.
- 1-minute video illustrating stereographic projection by Henry Segerman.

86
6 Hyperbolic geometry: conformal models
6.1 Poincaré disc model
Model: H2 =unit disc D = {|z | < 1, z ∈ C};
∂ H2 = {|z | = 1}, boundary, called absolute;
- lines: parts of circles or lines orthogonal to ∂ H2 , see Fig. 75, left;
- distance: a function of cross-ratio;
- angles: same as Euclidean angles.
Group: Isometries (i.e. Möbius transformation, inversions, reflections - preserving the
disc).
Notice that these transformations indeed preserve distance (when distance is a function
of cross-ratio), angles, set of lines.

X
B
A

Figure 75: Poincaré disc model.

Proposition 6.1. For any two points A, B , ∈ H2 there exists a unique hyperbolic line
through A, B .

Proof. Let I be the inversion with respect the absolute. Consider I (A). Let γ be
the (Euclidean) circle/line through A, B , I (A) (it does exist and is unique as every
Euclidean triangle has a unique circumscribed circle), see Fig. 76, left. Let X = γ ∩ ∂ H2
(exists as A is inside absolute and I (A) is outside). Then I (γ ) = γ (as it swaps A with
I (A) and preserves X ). This implies tat γ ⊥ ∂ H, and hence, γ ∩ D is the hyperbolic
line through A, B .
Notice that any (Euclidean) line/circle containing a hyperbolic line through A, B
should be preserved by inversion I (as the Euclidean line/circle should be orthogonal
to ∂ H2 ), so, it should contain A, I (A), B , and hence, coincide with γ .

Remark 6.2. The same holds for A, B , ∈ H2 ∪ ∂ H2 . Indeed, if one of the points
A, B (say, A) is not lying on the absolute, we can use the same proof as before. If
A, B , ∈ ∂ H2 , then any (hyperbolic) line through A, B should be orthogonal to the
absolute at A and B , or, in other words, orthogonal to the tangent lines lA and lB to
the absolute at points A, B , so the point Q = lA ∩ lB is the (Euclidean) centre of the
corresponding (Euclidean) circle, see Fig. 76, right. If A, B ∈ ∂ H2 are diametrically
opposed points of the absolute, then the same reasoning shows that the diameter AB
is the unique hyperbolic line through A and B .

87
 A

A
X B
I(A) B

Figure 76: Construction of a line through given points A and B .

X A | |Y A |
Definition 6.3. d(A, B ) = ln|[A, B , X, Y ]| = ln ||X /
B | |Y B |
,
where X, Y are the points of the absolute contained in the (hyperbolic) line AB , see
Fig. 75, right.
Theorem 6.4. d(A, B ) satisfies axioms of the distance.
Proof. 1. d(A, B ) ≥ 0 and d(A, B ) = 0 if and only if A = B : this is evident (in the
same way as for Klein model). More precisely, logarithm is zero if and only if the
cross-ratio equals one, which in view of Proposition 5.26.
2. d(A, B ) = d(B , A) since [A, B , X, Y ] = [B , A, Y , X ].
3. Triangle inequality d(A, B ) + d(B , C ) ≥ d(A, C ) will by proved below in Corol-
lary 6.12.

Example 6.5. Examples of isometries in the Poincaré disc model:

1. Rotation about the centre of the model;
2. reflection with respect to a diameter;
3. inversion with respect to a circle representing a hyperbolic line.
(Notice that such an inversion preserves the boundary of the disc, as a circle
representing a hyperbolic line is orthogonal to the absolute; also, it takes inside
of the disc to inside as the hyperbolic line is preserved pointwise).
All of these transformations preserve the disc and preserve the cross-ratios, so are
isometries. Inversions and reflections described above play a role of hyperbolic reflec-
tions - they fix a hyperbolic line pointwise and swap the half-planes.
Proposition 6.6. Let l ∈ H2 be a (hyperbolic) line, A ∈ H2 or A ∈ ∂ H2 be a point,
A∈/ l. Then there exists a unique line l′ through A orthogonal to l.
Proof. Existence: Let Il be an inversion with respect the line l. Il swaps the points A
and A′ = I (A) (A′ ∈ H2 as Il takes the points of H2 to the points of H2 ), see Fig. 77,
left. This implies that Il preserves the hyperbolic line AA′ . So, AA′ is orthogonal to l
and we can take l′ = AA′ .
Uniqueness If l′ ⊥ l then Il (l′ ) = l′ , so l′ contains the point A′ = Il (A), which implies
l′ = AA′ .

88
 A0

A A
A0
l
l

Figure 77: Construction of a perpendicular line to l through A.

Proposition 6.7. Let l ∈ H2 be a (hyperbolic) line, A ∈ H2 be a point, A ∈ l. Then

there exists a unique line l′ through A orthogonal to l.

Proof. Let Ia be the inversion in the absolute. Then Ia (l) = l (as Euclidean circles/lines
representing hyperbolic lines are orthogonal to the absolute). Let A′ = Ia (A) (notice
that A′ ∈/ H2 ), see Fig. 77, right.
Let l′ be the line or circle through A and A′ such that l′ ⊥ l (it does exist in view of
Remark 6.2 applied to the shaded disc bounded by l). Then Ia (l′ ) = l′ (as Ia swaps A
and A′ and preserves the point l′ ∩ ∂ H2 ). From this we conclude that l′ ⊥ l and hence,
l′ represents a hyperbolic line orthogonal to l and containing A.

Proposition 6.8. Every hyperbolic segment has a midpoint.

Proof. Let X, Y ∈ ∂ H2 be the endpoints of the hyperbolic line AB . Let f ∈ M ob ¨ be a

map of the unit disc to itself such that f (X ) = X , f (Y ) = Y where X Y is a diameter
′ ′

of the disc, and f (Z ) = Z for some Z ∈ ∂ H2 (this map exists by triple transitivity of
Möbius transformations on the points). We will show that the segment f (A)f (B ) has
a midpoint, and this will imply the same for AB as f preserves the cross-ratio.
From now on we assume that A, B lie on a diameter. Consider a point B ′ on the
same diameter such that B lies between A and B ′ . Then
|X A|  |Y A| |X A| |Y B | |X A| |Y B ′ |
d(A, B ) = |ln | = |ln | < |ln | = d(A, B ′ ),
|X B | |Y B | |X B | |Y A| |X B ′ | |Y A|

which means that d(A, B ) is a strictly monotone function. It is also clearly continuous.
Now, consider a point T = T (t) moving from the point A = T (0) to the point
B so that d(A, T ) = t (i.e. B = T (d0 )) where d0 = (A, B ))). Then d(A, T ) grows
monotonically from 0 to d0 while d(T , B ) declines monotonically from d0 to 0, which
implies that in some intermediate point M these distances coincide.

Remark. When B = B (t) runs along a ray AX from A to X , the distance d(A, B (t))
grows monotonically from 0 to ∞.

89
Theorem 6.9. The isometry group group of H2 acts transitively

(1) on triples of points of the absolute;

(2) on points in H2 .

Proof. (1) There exists f ∈ M ob ¨ taking any three given points of the absolute to
any other three given points. Then f takes the absolute to itself. If it takes the
outside of the disc to the inside, consider f ′ = Ia ◦ f where Ia is the inversion with
respect to the absolute. Then f ′ preserves the disc and preserves the cross-ratio,
so it is an isometry.

(2) Let O be the centre of the disc and A be a point. It is enough to find an isometry
fA which takes A to O (and when we need to map A to B we will consider a
composition fB−1 ◦ fA ). Let M be the (hyperbolic) midpoint of the hyperbolic
segment OA (the midpoint exists in view of Proposition 6.8). Let l′ be the
(hyperbolic) line orthogonal to OA and containing M (exists by Proposition 6.6),
see Fig. 78. Then inversion Il′ with respect to l′ preserves the line OA (as l′ ⊥ OA)
and swaps the points of OA lying on the same hyperbolic distance from M , i.e.
Il′ (A) = O. So, Il′ is the required isometry (as Il′ (D) = D).

A M O

l0

Figure 78: Mapping the point A by isometry to the centre O of the disc.

Remark. Isometries act transitively on flags (one can map a point to the centre of
the disc, then rotate around the centre and reflect with respect to a line through the
centre).

Proposition 6.10. For C ∈ AB , d(A, C ) + d(C, B ) = d(A, B ).

Proof. Take A to O by an isometry. Then the segment AB is represented by a segment

of Euclidean line. For points on a line we have checked that in Remark 4.29 when
considered the Klein model, where d(A, B ) = 21 |ln[a, b, x, y ]|.

Lemma 6.11. In a right-angled △AB C with ∠C = π /2, holds d(B , C ) < d(B , A).

90
Proof. By transitivity of isometries on the points of H2 we may assume that B is the
centre of the disc. Let γ be a (Euclidean) circle centred at B passing through C , see
fig 79. Notice that it is also a hyperbolic circle (i.e. the set of points on the same
distance from the centre of the disc). As B C is the radius of γ , we see C B ⊥ γ (as the
Euclidean sets). Since the side B C of the hyperbolic triangle AB C is represented by
a line or circle orthogonal to the absolute and orthogonal to C B , we conclude that A
lies outside of γ . Hence, B A > B A′ = C B where A′ = γ ∩ B A.

A A0
C B
γ
γ1

Figure 79: In a right-angled △AB C with right ∠C , holds d(B , C ) < d(B , A).

Corollary 6.12. Triangle inequality: for C ∈

/ AB , d(A, C ) + d(C, B ) > d(A, B ).
Proof. Let H be orthogonal projection of B to AC (in hyperbolic sense, i.e. H = l ∩ AC
where l is the hyperbolic line orthogonal to AC through B ). Then

d(A, B ) + d(B , C ) > d(A, H ) + d(H, C ) ≥ (A, C ).

Remark. Triangle inequality implies that (a) distance is well-defined, and

(b) hyperbolic lines are geodesics in the model.

By a hyperbolic circle centred at a point A we mean a set of points on the same

(hyperbolic) distance from A.
Proposition 6.13. (a) Hyperbolic circles are represented by Euclidean circles in the
Poincaré disc model.
(b) Every Euclidean circle in the disc represents a hyperbolic circle.
Proof. (a) Let O be the centre of the disc. A hyperbolic circle centred at O is
represented by a Euclidean circle (with the same centre). To obtain a circle
centred at a point A ∈ H 2 , A ̸= O, we apply an inversion Il′ which swaps O and
A (as in the proof of Theorem 6.9). The inversion takes the circle centred at O
to a circle centred at A (in hyperbolic sense). In Euclidean sense the image of a
circle under an inversion is either a circle or a line; it is a circle as it is contained
inside the disc D.

91
 (b) Let C ⊂ D be a (Euclidean) circle with Euclidean centre Q. Let let A and B
be the points of intersection of the circle C with the Euclidean lien OQ. Let M
be the hyperbolic midpoint of the hyperbolic segment AB . We can map M to
O by an isometry f of H2 (in view of Theorem 6.9). Then f (C ) is orthogonal
to the (Euclidean) line AB (as OQ ⊥ C ), it is a circle, and passes through two
points f (A) and f (B ) on the same Euclidean distance from O = f (M ). We
conclude that as a Euclidean set f (C ) is the circle centred at the origin. So, it is
a hyperbolic circle. And hence C is also a hyperbolic circle.

Notice that the Euclidean centre of the circle with hyperbolic centre A ̸= O is different
from A.
Theorem 6.14. An isometry of H2 is uniquely determined by the image of a flag.
Proof. Let O ∈ H2 be a point, l be a (hyperbolic) ray from O and h+ be a choice of
a half-plane with respect to the (hyperbolic) line containing l, let F = (O, l, h+ ) be a
flag. It is sufficient to show that if f is an isometry of H2 and f (F ) = F then f = id.
First, notice that if f (F ) = F then f (l) = l pointwise, as f preserves the cross-ratio
[O, B , X, Y ] where B ∈ l and X, Y are the endpoints of the line containing l.
Now, Let C ∈ / l be any point. As d(O, C ) = d(O, f (C )) and d(B , C ) = d(B , f (C )),
we see that f (C ) lies on the intersection of two hyperbolic circles: one centred at O of
radius OC and another centred at B of (hyperbolic) radius B C . This implies that f (C )
lies on the intersection of two Euclidean circles (in view of Proposition 6.13). Since
f (h+ ) = h+ we conclude that f (C ) = C for any choice of the point C , i.e. f = id.

Theorem 6.15. Every isometry of the Poincaré disc model can be written as either
az +b
cz +d
(Möbius transformation) or acz¯z¯++db (anti-Möbius transformation).
Proof. Let g ∈ I som(H2 ).Let F be a flag. Proving transitivity of isometry on flags we
have constructed a Möbius or anti-Möbius transformation f : F → g (F ). Clearly, it is
an isometry. Uniqueness shown in Theorem 6.14 implies that g = f .

Corollary 6.16. An isometry of H2 is uniquely determined by the images of three

points of the absolute.
Proof. It follows as Möbius and anti-Möbius transformations are determined uniquely
by images of the points.

Here, you can find some Hyperbolic Geometry Artworks by Paul Nylander.
Corollary 6.17. “Isometries preserve the angles”, i.e. hyperbolic angles coincide with
Euclidean ones.
Proof. This follows since Möbius and anti-Möbius transformations preserve angles.

Exercise 6.18. We can use the results of hyperbolic geometry to show the following
statement:
Let C1 and C2 be two disjoint circles. Then there exists a Möbius transformation
which takes them to two concentric circles.

92
To show this, first use some Möbius transformation to map C1 inside of C2 . Then
imagine that C2 is the Poincare disc model and C1 is a circle inside of it...

Proposition 6.19. The sum of angles in a hyperbolic triangle is less than π .

Proof. Since the isometries act transitively on points of hyperbolic plane, it is sufficient
to show the statement for a triangle with one vertex at the centre of the disc. For such
a triangle we compare angles of Euclidean triangle with angles of the hyperbolic one:
the angle at the centre coincide, while the angles at other points a strictly smaller in
hyperbolic case see Fig. 80.

Figure 80: Sum of angles in hyperbolic triangle is smaller than in the Euclidean one.

Remark. On can show that if α + β + γ < π then there exists a triangle with angles
α, β , γ .

6.2 Upper half-plane model

Model: H2 = {z ∈ C, I mz > 0};
∂ H2 = {I mz = 0}, absolute;
- lines: rays and half-circles orthogonal to ∂ H2 ;
- distance: d(A, B ) = ln[A, B , X, Y ] ;
- angles: same as Euclidean angles.
Group: isometries, i.e. Möbius transformation, inversions, reflections
(preserving the half-plane).

Proposition 6.20. This defines the same geometry as Poincaré disc model.

Proof. Consider a Möbius transformation f = 1i zz− +1

1
(inverse to Cayley map) taking
the unit disc to the upper half-plane. f preserves cross-ratio, and hence, preserves the
distance.

93
 Figure 81: Upper half-plane model.

This implies that we can use in this model all results obtained in the Poincard́isc
model.

Proposition 6.21. In the upper half-plane, hyperbolic circles are represented by Eu-
clidean circles.

Proof. We know this for the Poincaré disc, so applying a Möbius transformation we
get the same for the upper half-plane model.

Remark. If the two models are so similar, why do we need both?

This is because different properties are better viewed in different models.
Example:

- In the disc model it is easier to see that (hyperbolic) circles are represented by
Euclidean circles.

- In the upper half-plane it is easier to see that there is a hyperbolic line through
every two points on the absolute (if two points A and B are given by a, b ∈ R
then the corresponding line is given by a (Euclidean) circle of radius |a − b|/2
centred at (a + b)/2).

The distance between two points is also easy to compute in the upper half-plane.

Theorem 6.22. In the upper half-plane model,

|z − w|2
cosh d(z , w)) = 1 + .
2I m(z )I m(w)

Proof. We start by checking the formula for the points z = i, w = k i, k > 1:

∞−i  0−i
d(z , w) = | ln[i, k i, ∞, 0]| = | ln | = ln k .
∞ − ki 0 − ki
So, we obtain that
1 1 1
cosh(d(z , w)) = cosh(ln k ) = (eln k + e− ln k ) = (k + ).
2 2 k
At the same time,

|z − w|2 (k − 1)2 k2 2k 1 1 1
1+ =1+ =1+ − + = (k + ).
2I m(z )I m(w) 2·1·k 2k 2k 2k 2 k

94
 To check the formula for general z , w, apply a Möbius transformation taking the
points z , w to i, λi for some λ ∈ R+ (such a transformation exists in view of transitivity
of isometries on flags). Then the left hand side is preserved as it is a function of a
cross-ratio, while the right hand side is preserved since it is preserved by each of the
generators of Möbius transformations, i.e. by z → az , z → z + 1, z → 1/z .

Theorem 6.23. Every isometry of the upper half-plane model can be written as either
z 7→ az +b
cz +d
or z 7→ ac((−−z¯z¯)+
)+b
d
with a, b, c, d ∈ R, ad − bc > 0.

Proof. Isometries of the upper half-plane are isometries of the Poincaré disc conjugated
by Möbius transformations. So, from Theorem 6.15 we conclude that all isometries of
the upper half-plane are either Möbius or anti-Möbius transformations.
To conclude about the coefficients a, b, c, d, we will first consider orientation-preserving
isometries. Notice that is f is an isometry of the upper half-plane, then f takes the
real line to itself. This implies that one can choose the coefficients a, b, c, d to be real:
indeed, we have f (∞) = a/c ∈ R, f (0) = b/d ∈ R, since also f (1) ∈ R one can
conclude that a/b ∈ R (check!). Finally,

ai + b (ai + b)(ci − d) −bd − ac + i(bc − ad)

f (i) = = 2 2
= .
ci + d −c − d − c2 − d 2

As f maps the upper half-plane to the upper half-plane, I m(f (i)) = − cbc2− ad
+d2
> 0 which
is equivalent to ad−bc > 0. This finishes the proof for orientation-preserving isometries.
An orientation reversing isometry g may be considered as a composition of the map
r : z → −z¯ (which is a reflection with respect to the imaginary axis, see Fig. 82)
with an orientation- preserving isometry f = g ◦ r. Then g = f ◦ r−1 = f ◦ r. As
f (z ) = az +b
cz +d
, a, b, c, d ∈ R, ad − bc > 0 by above, we conclude that g (z ) = acz¯z¯++db with
the same restrictions on the coefficients.

−z̄ z

z̄

Figure 82: Reflection z → −z¯.

Equivalently, orientation-preserving isometries can be written as z 7→ az +b

cz +d
where
a, b, c, d ∈ R, ad − bc = 1. Hence, for the group of or.-preserving isometries we have

I som+ (H2 ) = P S L(2, R) = S L(2, R)/±I .

Orientation-reversing isometries can be written as z 7→ az¯+b

cz¯+d
with a, b, c, d ∈ R and
ad − bc = −1.

95
Remark. One can see that the distance is “larger” near the absolute. To see this
consider a line l and two points on it A0 A1 . We can map A0 to A1 so that the line l
will be mapped to itself and the half-planes with respect to l will not swap. The point
A1 will map to some point A2 on the same distance from A1 as A0 . Iterating the same
map we will get infinitely many points A3 , A4 , . . . on the same line with the condition
d(Ai , Ai+1 ) = d(A0 , A1 ).

Remark. Pairs of lines in the conformal models: see Fig. 83 for intersecting, parallel
and divergent (or ultra-parallel) lines in the Poincaré disc and in the upper half-plane.

Figure 83: Pairs of line in H2 : intersecting, parallel, divergent (ultra-parallel).

Example. Let l, l′ be parallel lines. Then d(l, l′ ) = 0 (where by distance between the
sets α and β we mean d(α, β ) = inf (A, B )).
A∈α,B ∈β
To show this, we consider the lines in the upper half-plane model, and we map
the point l ∩ l′ ∈ ∂ H2 to ∞. Then l and l′ are represented by vertical half-lines. By
applying an isometry z → az + b, a, b ∈ R we may also assume that l lies on the
imaginary axis and l′ on the line given by Re(z ) = 1, see Fig. 84 Consider the points
k i ∈ l and k i + 1 ∈ l′ . Then
1
cosh d(k i, k i + 1) = 1 + →1
2k 2
as k → ∞. This implies that d(k , k i) → 0 as k → ∞.

ki ki + 1

0 1

Figure 84: Parallel lines are on distance 0.

96
6.3 Elementary hyperbolic geometry

Remark. 1. Triangle inequality implies that d(A, B ) satisfies axioms of distance

and that hyperbolic lines are shortest paths.
2. In hyperbolic geometry, all Euclid’s Axioms, except Parallel Axiom hold, while the
later clearly does not hold (see Fig. 85).
3. Recall that Parallel Axiom for hyperbolic geometry says:
There are infinitely many lines l′ disjoint from a given line l and passing through a
given point A ∈/ l.

Figure 85: Parallel axiom does not hold in H2 : given a line l and a point A there are
infinitely many lines through A ultra-parallel with l, and there are two lines through
A parallel to l (labelled red).

Definition 6.24. For a line l and a point A ∈/ l, an angle of parallelism φ = φ(A, l) is

the half-angle between the rays emanating from A and parallel to l, see Fig. 86, left.
Equivalently: drop a perpendicular AH to l, then φ = ∠H AQ, Q ∈ l ∩ ∂ H2 ;
Equivalently: a ray AX from A intersects l iff ∠H AY ≤ φ, see Fig. 86, right.

A A

2ϕ ϕ

l H
Q X

Figure 86: Definition of angle of parallelism.

97
Remark. The angle of parallelism φ only depends on distance d(A, l) = min(A, B ) =
B ∈l
d(A, H ). To see this, map A to the centre of the disc by isometry, so that AH will be
mapped to a vertical ray. Then d(A, H ) completely determines l as AG ⊥ l.
Proposition 6.25. For a line l and a point A ∈
/ l, let a = d(A, l) and φ be the angle
1
of parallelism. Then cosh a = sin φ .
Proof. We will compute in the upper half-plane model. We may assume that H is the
point i, l be a vertical line through i, and Q := l ∩ ∂ H2 is the point ∞, see Fig. 87
The right angle at the point H will be formed by the imaginary axis and the circle
|z | = 1. So, the points of triangle △H AQ are given by H = i, Q = ∞, A = eiψ for
some ψ ∈ [0, π /2]. Notice that if O is the origin then OA is orthogonal to the circle
|z | = 1. So, ψ is the angle between the radius OA and the horizontal line, while φ is
the angle between the tangent to the circle at A and the vertical line - which means
that ψ = φ and A = eiφ .
Now, we can compute the distance d(A, H ) by

|i − eiφ |2 |i − eiφ |2
cosh d(A, H ) = 1 + = 1 +
2I m(i)I m(eiφ ) 2 sin φ
1 + 1 − 2 cos( π2 − φ) 1 − sin φ) 1
=1+ =1+ = .
2 sin φ 2 sin φ sin φ

H
ϕ
O
A

Figure 87: To the proof of Proposition 6.25.

Remark. When a → 0 we have cosh a → 1 which implies that sin φ → 1, i.e.

φ → π /2, as in Euclidean geometry.
Theorem 6.26 (Hyperbolic Pythagorean theorem). In a triangle with a right angle
γ , cosh c = cosh a cosh b.

98
Proof. Without loss of generality we may assume that C = i, A = k i, B = eiφ , see
Fig.88. Then using Proposition 6.25 we get
1
cosh a = .
sin φ
We also compute using the distance formula:
(k − 1)2 1 + k2
cosh b = 1 + =
2k 2k
and
cos2 φ + (k − sin φ)2 1 + k2
cosh c = 1 + = ,
2k sin φ 2k sin φ
which implies the theorem.

A = ki
α
b
c
C=i
a
B = eiϕ

α ϕ
X0 O

Figure 88: To the proof of Pythagorean Theorem and Lemma 6.6.

Remark. Here you can find an applet to make hyperbolic tessellations of images, by
Malin Christersson.

Remark. For small values of x we have

1 x2 x2 x2
cosh x = (1 + x + + ··· + 1 − x + + ...) = 1 + + ...,
2 2 2 2
so the Pythagorean theorem can be written as
c2 a2 b2
(1 + + . . . ) ≈ (1 + + . . . )(1 + + . . . ),
2 2 2
which is equivalent to c2 ≈ a2 + b2 , approaching the Euclidean version of Pythagorean
theorem.
Lemma 6.27. In a triangle with a right angle γ holds:

sinh a = sinh c sin α.

99
Proof. We compute using the same Fig.88 as before:
1 cos2 φ
sinh2 a = cosh2 a − 1 = − 1 = .
sin2 φ sin2 φ
Also,
1 + k2 2 (k 2 + 1)2 − 4k 2 sin2 φ
sinh2 c = cosh2 c − 1 = ( ) −1= .
2k sin φ 4k 2 sin2 φ
Next, we need to compute ∠α = ∠B AC . To do this, let x be the number such that
X = (x, 0) is the centre of the Euclidean circle containing the (hyperbolic) segment
AB . Notice that ∠AX O = α (as the X A is the radius of the circle representing
the hyperbolic segment AB and hence is perpendicular to it at A). Notice that as
X A = X B , we get x2 + k 2 = (cos φ − x)2 + sin2 φ, from where we get k 2 = 1 − 2x cos φ,
−k 2
and hence x = 21cos φ
.
Next, from the right-angled triangle △AX O we have
k2 k2 4k 2 cos2 φ
sin2 α = = 2 −1 = =
k 2 + x2 k 2 + ( 2kcos φ
)2 (k 2 − 1)2 + 4k 2 cos2 φ
4k 2 cos2 φ 4k 2 cos2 φ
= .
(k 2 + 1)2 − 4k 2 + 4k 2 cos2 φ (k 2 + 1)2 − 4k 2 sin2 φ
This implies that
cos φ
sinh c sin α = = sinh a.
sin φ

Theorem 6.28 (Sine rule). sinh a

sin α
= sinh b
sin β
= sinh c
sin γ
.
Proof. Let C H be the altitude in △AB C , see Fig. 89. Denote h = |C H | (where by
|C H | we mean the hyperbolic length of the segment C H ). Then from the right-angled
triangles AH C and B H C by Lemma 6.6 we get
sinh h = sinh b · sin α = sinh a · sin β ,
which implies sinh b
sin β
= sinh a
sin α
.

b h a

α β
A x H c−x B

Figure 89: To the proof of Sine and Cosine Rules.

Notice that in the proof of the sine rule we did not use any model!

Exercise:
cosh(a − b) = cosh a cosh b − sinh a sinh b (6.1)
Hint: one can prove it from the definition cosh x = 21 (ex + e−x ).

100
Exercise 6.29. In a triangle with a right angle γ holds:

tanh b = tanh c cos α.

The proof is a part of Assignment 15-16, Question 15.2.

Theorem 6.30 (Cosine rule). cosh a = cosh b cosh c − sinh b sinh c cos α.

Proof. Let C H be the altitude dropped from C , let h be its length, let x be the length
of AH , see Fig. 89. Then from Pythagorean Theorem we have

cosh b = cosh h cosh x, (6.2)

similarly

(6.1)
cosh a = cosh h cosh(c − x) = cosh h(cosh c cosh x − sinh c sinh x)
(6.2) cosh b
= cosh b cosh c − sinh c sinh x = cosh b cosh c − cosh b sinh c tanh x
cosh x
(6.5)
= cosh b cosh c − cosh b tanh b sinh c cos α
= cosh b cosh c − sinh b sinh c cos α.

Remark. For small values of a, b, c we get Euclidean sine and cosine laws.

Theorem 6.31 (Second cosine rule). cos α = − cosh β cos γ + sin β sin γ cosh a.

We omit the proof (one can find it in the book by Prasolov and Tikhomirov).

Exercise 6.32 (Congruence of triangles). a

Prove SSS, SAS, ASA and AAA rules of congruence of triangles in hyperbolic plane.
(One can do it either as a corollary of sine/cosine laws or directly, see also HW 13.3).
Hint: to prove AAA one can use the diagrams in Fig. 90.

β β
β β
α γ γ α γ γ

Figure 90: Proving AAA congruence.

Example: One can use the sine law to compute length of circle of radius r:

l(r) = 2π sinh r. (6.3)

101
Proof. To show it, we inscribe a regular n-gon Pn into the circle (we can draw it in the
disc model with the centre of Pn at the centre of disc, then the vertices of Pn will be
represented by vertices of regular Euclidean n-gon, see Fig. 91). Then we subdivide Pn
into 2n right-angled triangles, and compute perimeter of Pn as

P (Pn ) = 2nd,

where d is the half of the side of Pn . From the sine rule we have
sinh d sinh r
2π = ,
sin 2n sin π2

which implies that sinh d = sinh r sin πn .

Now, we compute the length of the circle as a limit of the perimeter of Pn when
n → ∞:
π x∼sinh x
l(r) = lim P (Pn ) = lim 2n · arcsinh(sinh r sin ) =
n→∞ n→∞ n
π π
lim 2n(sinh r sin ) = lim 2n sinh r · = 2π sinh r.
n→∞ n n→∞ n

Remark. Strictly speaking, using the sequence of polygons Pn , n → ∞ we only show

that the length of circle is larger or equal to 2π sinh r (as each arc is larger than the
corresponding geodesic side of the polygon). To get also the upper bound one needs to
consider regular polygons Qn for which the circle is inscribed (and it is a bit harder to
show that the perimeter of each polygon Qn is indeed larger than l(r)).

r
d

Figure 91: Computing the length of a circle of radius r.

So, the length l(r) of the circle of radius r in spherical, Euclidean and hyperbolic
geometry can be expressed by a similar formulae:

S2 E2 H2
l (r ) 2π sin r 2π r 2π sinh r

102
 l(a) l(b) l(c)
Corollary. Uniform statement for sine law in S 2 , E2 or H2 : sin α
= sin β
= sin γ
,
where l(r) is the length of circle of radius r in the corresponding geometry.

Remark. As we can see from the formula 6.3, in the hyperbolic geometry the circle
length l(r) grows exponentially when r → ∞. One can find examples of such structures
in nature (salad leafs, sea weeds, etc.).

6.4 Area of hyperbolic triangle

We will show that area of a hyperbolic triangle with angles α, β , γ is given by

S△AB C = π − (α + β + γ ). (6.4)

Notice that the formula makes sense by two reasons:

• As we know, α + β + γ < π , so the area computed by (6.4) is positive;

• the angles α, β , γ completely determine hyperbolic triangle up to isometry (AAA

congruence of triangles), so they should determine the area of the triangle.

Definition 6.33. A hyperbolic polygon with all vertices on the absolute is called
ideal polygon.

Notice, that in view of Theorem 6.9, all ideal triangles are congruent. In particular,
they have the same area.

Theorem 6.34. S△AB C = π − (α + β + γ ).

Proof. First, we introduce some notation:

• Let λ the area of an ideal triangle.

• Let f (α) be area of a triangle with angles (α, 0, 0) (i.e. with a triangle with one
vertex in H2 and two vertices at the absolute, see Fig. 92, left).

α π−α α α
β

Figure 92: Properties of area f (α) of triangle with angles (α, 0, 0).

We will show several properties of the function f (α):

(1) f (α)+ f (π − α) = λ (an ideal triangle can be assembled from two smaller triangles
as in Fig. 92, middle).

(2) f (α) + f (β ) = f (α + β ) + λ (a quadrilateral in Fig. 92, right can be splitted in

two triangles in two ways).

103
 (3) By Property (1), f ( π2 + π2 ) = λ, which implies f ( π2 ) = λ2 .
Next, by property (2), f ( π4 ) + f ( π4 ) = π2 + λ which implies f ( π4 ) = 34 λ.
By (1) again, f ( 34π ) = λ − 34 λ = λ4 .
And by (2) again: f ( π8 ) = 12 f ( π4 ) + λ2 = 78 λ.
.....................................................................
In the same way we obtain that f ( 2kmπ ) = λ(1 − 2km ).

(4) f (α) > f (α′ ) if α < α′ < π . Indeed, let 0 < β = α′ − α, then
(2) (1)
f (α) − f (α + β ) = λ − f (β ) = f (π − β ) > 0.

(5) From (3) and (4) we conclude that f (α) = πλ (π − α).

Using Property (5) in the same way as we used areas of spherical digons for proving
the formula of area for the spherical case, we will show that S△AB C = πλ (π − (α + β + γ ))
(see Lemma 6.35 below). Then, in Lemma 6.36 we will use small triangles to find out
that λ = π , which will finish the proof.

Lemma 6.35. S△AB C = πλ (π − (α + β + γ ))

Proof. An intersection of two lines at angle α produces two triangles with angles
(α, 0, 0) (“hyperbolic α-digons”) (if α = π /2 there are four such triangles but we will
be interested in one pair of non-adjacent ones). A triangle with angles α, β , γ produces
three pairs of hyperbolic digons, which all together cover an ideal hyperbolic hexagon,
see Fig. 93, left. Notice that the triangle △AB C itself is covered three times, while all
other parts of the hexagon only covered once. So,

2f (α) + 2f (β ) + 2f (γ ) − 2S△AB C = Sideal hexag on = 4λ,

where the last equality holds since an ideal hexagon can be composed of 4 ideal triangles
(see Fig. 93, right). From this we get f (α) + f (β ) + f (γ ) − S△ AB C = 2λ, i.e.

α β γ α β γ
S△AB C = λ((1 − ) + (1 − ) + (1 − ) − 2λ = λ(1 − − − ).
π π π π π π

Lemma 6.36. λ = π .

Proof. Consider a small right-angled triangle with sides a, b, c → 0, γ = π /2. Then the
triangle is almost Euclidean, i.e. α + β + γ → π , π − (α + β + γ ) → 0. This implies
that
γ= π π
π −(α+β +γ ) ≈ sin(π −(α+β +γ ) =2 sin( −(α+β )) = cos(α+β ) = cos α cos β −sin α sin β .
2
Next, we use two formulae concerning right-angled triangles and obtained in HW 15.2:

tanh b = tanh c cos α (6.5)

and
sinh a = sinh c sin α (6.6)

104
 Figure 93: Area of hyperbolic triangle.

and continue the computation as follows:

tanh b tanh a sinh a sinh b
π − (α + β + γ ) ≈ · − · =
tanh c tanh c sinh c sinh c
sinh a sinh b cosh2 c cosh c=cosh a cosh b
2 ( − 1) =
sinh c cosh a cosh b
sinh a sinh b a · b c2 ab
2 (cosh c − 1) ≈ 2
· = = SEuclidean triangle .
sinh c c 2 2
As S△AB C = πλ (π − (α + β + γ )) and π − (α + β + γ ) tend to the area of Euclidean
triangle when a, b, c decrease, we conclude that λ = π .

Remark. The value δ = π − (α + β + γ ) is called a defect of the triangle.

Corollary 6.37. Area of an n-gon: Sn = (n − 2)π − ni=1 αi .
P

Remark. (Refraction). Return for a minute to a real life. Recall, that the speed
of light depends on properties of the material. At the same time, light choose the
quickest path, which implies that a light ray bends when hitting the boundary of two
transparent materials.
More precisely, consider the edge of two transparent materials (we will call them
“air” and “water”). We assume that the speed of light in the air if v1 (m/sec) and in
the water is v2 < v1 (m/sec). We want to connect two given points (with coordinates
(0, h1 ) and (1, h1 ), see Fig. 94), one in the air and one in the water, by the “shortest
path” in sense that it is fastest for the light ray. It will be a broken line with two
segment (as inside each material the speed of light is constant), but we do not know in
advance at which point X = (x, 0) will the ray hit the boundary of water. Depending
on coordinate x, the time t(x) needed for the ray to travel between the two points is
p p
h21 + x2 h22 + (1 − x)2
t(x) = + ,
v1 v2
and we need to find minx t(x). We find where the derivative of t vanishes:
1 2x 1 2(1 − x)
t′ (x) = p − p = 0.
2v1 h21 + x2 2v2 h22 + (1 − x)2

105
From this equation (and using notation as on Fig. 94, left) we obtain

sin θ1 sin θ2
= .
v1 v2
This implies that if h1 = h2 then the ray will travel more in the air than in the water.

θ1 v1 h1

0 x 1
v2
θ2
h2

Figure 94: Refraction.

If the density of the material changes several times we get a broken line as the
shortest path, like the one in Fig. 95. And when the density grows continuously when
approaching the real line, so that the speed at the point (x, y ) is equal to 1/y , one can
get hyperbolic lines as geodesics (i.e. as trajectory of the light rays).

Figure 95: Broken lines as shortest paths.

106
6.5 References
- Section 6 is based on Lectures VI – VIII in Prasolov’s book.
Alternatively, see pp.95–104 in Section 5.2 of Prasolov, Tikhomirov.

- See also

- Hyperbolic Geometry. Lecture notes by Caroline Series.

- Hyperbolic Geometry. Lecture notes by Charles Walkden.

- A short paper to stimulate your non-Euclidean intuition:

Non-Euclidean billiards in VR by Jeff Weeks.

- Webpages, pictures, videos:

- Hyperbolic Geometry Artworks by Paul Nylander.

- Webpage on hyperbolic geometry by the Institute for Figuring (with hyper-
bolic soccer ball and crocheted hyperbolic planes).
- Tilings by (and after) Escher. Webpage on Mathematical Imagery by Jos
Leys.
- How to create repeating hyperbolic patterns, by Douglas Dunham (based
on Escher’s patterns). See also here.
- Playing Sports in Hyperbolic Space - Dick Canary in a Numberphile video
(by Brady Haran).

- Software:

- Applet for creating hyperbolic drawings in Poincaré disc.

- Applet to make hyperbolic tessellations of images, by Malin Christersson.

107
7 Other models of hyperbolic geometry
7.1 Klein disc, revised

Reminder:
- the model is inside the unit disc, lines are represented by chords.
- distance in Klein disc d(A, B ) = 12 ln[A, B , X, Y ] , where X, Y are the endpoints
of the chord through A, B , see Fig. 52;
- isometries are projective maps preserving the disc.
Theorem 7.1. Geometry of the Klein disc (DK ) coincides with geometry of the Poincaré
disc (DP ).
Proof. Idea of the proof: we will build a map f : DK → DP which takes the Klein
disc to the Poincaré disc and will show that the distance between points in DK will
coincide with the distance between their images in DP .
We will construct the map f as a composition of two projection. First, consider a
unit sphere S 2 ∈ R3 , and let DK = {(x, y , z ) ∈ R3 | z = 0, x2 + y 2 + z 2 < 1} be the
horizontal unit disc inside the sphere. Consider the orthogonal (vertical) projection
p : Dk → S 2 of the disc to the lower hemisphere:
p
p : (x, y , 0) 7→ (x, y , − 1 − x2 − y 2 ).

p s
Dk A2 A2 DP
A
A1 A1

Figure 96: Projecting the Klein disc to the hemisphere and then to the Poincaré disc.

Next, we apply a stereographic projection s : S 2 → DP from the North Pole (1, 0, 0)

which will take the lower hemisphere to the disc
DP = {(x, y , z ) ∈ R3 | z = 0, x2 + y 2 + z 2 < 1}
(see Fig. 96). The composition
f := s ◦ p
has the following properties:
- f maps Dk to DP bijectively (as a map of sets).
- An image of a line in Dk is a line in DP :
Indeed, a chord is mapped by p to a vertical semicircle, i.e. to a semicircle lying
on the sphere and orthogonal to the boundary of the horizontal disc. Then the
stereographic projection maps it to a part of a circle orthogonal to the boundary
of the disc (as s preserves the angles).

108
 - Let A, B ∈ Dk , let A1 = p(A), B1 = p(B ) and let A2 = f (A), B2 = f (B ).
Let X, Y be the endpoints of the chord AB . Then

[A, B , X, Y ] = [A2 , B2 , X, Y ]2 . (7.1)

Proof. As the stereographic projection s preserves cross-ratios, it is enough to

prove that
[A, B , X, Y ] = [A1 , B1 , X, Y ]2 .
Note that since p maps X Y to a semicircle, we have ∠X A1 Y = π /2, and more-
over, triangles △AA1 X , △AY A1 and △A1 Y X are all similar. So, we get
2
|X A| |X A| |AA1 | |X A1 |

= · = .
|Y A| |AA1 | |Y A| |Y A1 |
So, we compute
 2
XA  Y A XAXB X A1  X B1
[A, B , X, Y ] = = = = [A1 , B1 , X, Y ]2 .
XB Y B YA YB Y A1 Y B1
End of proof of (7.1).

- Denote by dK (A, B ) and dP (A2 , B2 ) the distance measured between A and B in

the Klein model and the distance between their images in the Poincaré model,
as in Fig. 96. Then
1 (7.1)
dK (A, B ) = |ln[A, B , X, Y ]| =
2
1
|ln[A2 , B2 , X, Y ]2 | = |ln[A2 , B2 , X, Y ]| = dp (A2 , B2 ).
2

X
X
A2 B2
A B
A2
Y A1 A

Figure 97: Comparing the cross-ratios in Klein and Poincaré disc models.

Remark: One can use light to project the hemisphere model to Klein disc, Poincare
disc and upper half-plane. See the following video by Hynry Segerman and Saul
Schleimer.

109
Remark: When to use the Klein disc model? It is useful in many cases when we need
to work with lines and right angles.

Examples:

- Right angles are displayed nicely in the Klein model (see Proposition 4.33, see
also Fig. 98, left).

- One can construct the common perpendicular to any two divergent lines (see
Proposition 4.34, see Fig. 98, middle left).

- One can construct a midpoint for any segment (see Fig. 98, middle right).

- and an angle bisector (see also Fig. 98, right).

- This implies that one can construct centres of the inscribed and circumscribed
circles (when exist).

Figure 98: Constructions in Klein model: right angles, common perpendicular, mid-
point, angle bisector.

Remark: circles in the Klein model are represented by ellipses. Indeed, a circle
centred at the centre of the disc is clearly represented by a Euclidean circle. Projective
transformations take a circle to an ellipse, or hyperbola, or parabola - however, out of
them only ellipses fit inside the unit disc. So we conclude that all circles are represented
by ellipses.

7.2 The model in two-sheet hyperboloid

Consider the hyperboloid H ∈ R3 given by the equation

x21 + x22 − x23 = −1,

where x1 , x2 , x3 ∈ R. This is a two-sheet hyperboloid, it has two connected components

(one with z > 0 and another with z < 0). We will projectivise it, i.e. identify the points
(x1 , x2 , x3 ) ∼ (−x1 , −x2 , −x3 ).
Model:

- H2 = {points of the upper sheet}. Can be also understood as the set of lines
through O intersecting H , see Fig. 99.

- the absolute ∂ H2 { are (projectivised) points of the cone x21 + x22 − x23 = 0 }, i.e.
lines spanning the cone.

110
 - Lines in H 2 : intersections of planes through O with the hyperboloid.

- Distance: d(A, B ) = 12 ln[A, B , X, Y ]

(cross-ratio of the four corresponding lines in R3 , see Fig. 99, right).

Group: - Isometries: projective transformations preserving the cone.

X
B
A

Figure 99: The model on two-sheet hyperboloid.

Theorem 7.2. This determines the same hyperbolic geometry as the Klein model.
Proof. We construct a bijective map pr : H → DK from the hyperboloid model H to
the Klein disc DK and will show that this map is distance-preserving.
Let DK = {x3 = 1} ∩ {x21 + x22 − x23 < 0} by the intersection of the horizontal plane
x3 = 1 with the inside of the cone. Let pr : H → DK be the projection from the origin
of the hyperboloid H to the disc DK , see Fig. 100. Then
• Points of H are mapped bijectively to points of DK .

• Lines in H (i.e. intersections of the hyperboloid with planes through the origin)
are mapped to lines in DK (i.e. intersection of the plane x3 = 1 with the planes).

• Cross-ratio of points in H (i.e. cross-ratio of the corresponding lines in R3 )

coincides with the cross-ratio of the points of their intersection with the plane
x3 = 1.
From this we conclude, that pr is a distance-preserving bijection.

For x = (x1 , x2 , x3 ), y = (y1 , y2 , y3 ) define a pseudo-scalar product by

⟨x, y ⟩ = x1 y1 + x2 y2 − x3 y3 .

Then
- points of the H2 : ⟨x, x⟩ = −1;

- points of the ∂ H2 : ⟨x, x⟩ = 0;

- hyperbolic line la : a1 x1 + a2 x2 − a3 x3 = 0, i.e. ⟨a, x⟩ = 0.

111
 DK

Figure 100: Projection from hyperboloid the the Klein disc. (Here the disc is drawn in
the plane z = 1/2 for the clarity of the diagram).

Remark. Geometric meaning of ⟨a, a⟩:

- if ⟨a, a⟩ > 0 then la intersects the cone producing a hyperbolic line;

- if ⟨a, a⟩ = 0 then la is tangent to cone producing the point a on the absolute;

- if ⟨a, a⟩ < 0 then la does not intersect the cone and gives no line
(but a corresponds to a point of H2 ).
To explain this, notice that the pseudo-scalar product ⟨a, a⟩ does not change when a
rotates around the axis Ox3 , in particular, without loss of generality we may assume
that a lies in the plane x2 = 0. So, we assume a2 = 0. We also assume that a1 > 0.
We can also assume that a3 > 0 as ⟨a, a⟩ = ⟨−a, −a⟩.
So, in the assumption that a2 = 0, a1 , a3 > 0 we have:
- If ⟨a, a⟩ > 0, then a21 − a23 > 0 and a1 > a3 which implies that the set la intersects
the cone at a1 x1 − a3 x3 = 0 i.e. where |x1 | < |x3 |, which is impossible inside the
light cone.

- If ⟨a, a⟩ = 0, then we get points where |x1 | = |x3 |, i.e. the point on the cone.

- If ⟨a, a⟩ < 0, we get |x1 | > |x3 | which gives a point inside the cone, so an inner
point of the model.

a a la
la
la
ha, ai > 0 ha, ai = 0 ha, ai < 0

Figure 101: Geometric meaning of pseudo-scalar square.

112
 ⟨u,v ⟩2
Theorem 7.3. cosh2 d(u, v ) = ⟨u,u⟩⟨v ,v ⟩
for u, v ∈ H2 , i.e. for u, v satisfying ⟨u, u⟩ < 0,
⟨v , v ⟩ < 0.
Proof. It is sufficient to prove the theorem for u = (0, 0, 1) and v = (x, 0, z ) (to see
this we first apply an isometry inside the Klein disc which takes any given point to
the centre of the disc, and then we project it to the hyperboloid; we can also apply a
rotation about the centre to ensure that one coordinate is zero). See Fig. 102. Then
the right-hand side is as follows:
⟨u, v ⟩2 (−z )2 x2 −z 2 =−1
= = z2.
⟨u, u⟩⟨v , v ⟩ −1 · (x2 − z 2 )
To compute the left-hand side, we first compute the distance d(u, v ) by the definition:

1 1 − 0  −1 − 0 1 z+x
d(u, v ) = ln x x = ln =
2 1 − z −1 − z 2 z−x
1 (z + x)2 z 2 −x2 =1 1
ln 2 = ln(x + z )2 = ln(x + z ).
2 z − x2 2
From this we conclude ed(u,v) = x + z , which implies
1 x2 +xz +1
2 ed(u,v) + e−d(u,v) 2 x+z+ x+z 2
z+ x+ z 2
cosh d(u, v ) = ( ) =( ) =( )
2 2 2
x2 +1=z 2 1 z (x + z ) 2 1
= (z + ) = (2z )2 = z 2 .
4 x+z 4

Remark. 7.3 12 . In the proof of Theorem 7.3, we start by applying isometrie to simplify
the computation, but why would the isometries preserve both parts of the equation? It
is clear from the definition of isometry that the LHS is preserved. To see that the RHS
is also preserved, we can write a required isometry as a composition of reflections and
check that every reflection r preserves the pseudo-scalar product ⟨u, v ⟩ for any vectors
u, v (see Theorem 8.1 below for the formula of the reflection and Remark 8.1 21 ). Notice
also, that the proof of Theorem 8.1 does not use the result of Theorem 7.3.

v
u
(−1, 0, 0) (1, 0, 0)

Figure 102: To the proof of Theorem 7.3 .

By a similar computation to the one in Theorem 7.3 one can prove the following
theorem.

113
 ⟨u,v ⟩ 2
Theorem 7.4. Denote by Q = Q(u, v ) := | ⟨u,u⟩⟨v ,v ⟩
|. Then

(1) if ⟨u, u⟩ < 0, ⟨v , v ⟩ > 0, then u gives a point and v give a line lv on H2 , and
sinh2 d(u, lv ) = Q;

(2) if ⟨u, u⟩ > 0, ⟨v , v ⟩ > 0 then u and v define two lines lu and lv on H2 and

– if Q < 1, then lu intersects lv forming angle φ satisfying Q = cos2 φ;

– if Q = 1, then lu is parallel to lv ;
– if Q > 1, then lu and lv are ultra-parallel lines satisfying Q = cosh2 d(lu , lv ).

7.3 References
- The exposition of Section 7 is partially based on parts of Lectures VI and XIII
of Prasolov’s book.
Alternatively, see Section 5.2 of Prasolov, Tikhomirov.

- The relation between the four models (even five, including the hemisphere!) of
hyperbolic geometry is described in Section 7 of
J. W. Cannon, W. J. Floyd, R. Kenyon, W. R. Parry, Hyperbolic Geometry.

- There are also other models of hyperbolic plane, which we do not consider in this
course. For some of them see Hyperbolic Spaces by John R. Parker.

- Webpages, videos:

- Stereographic projection and models for hyperbolic geometry, 3D toys to il-

lustrate by Henry Segerman.
- Illuminating hyperbolic geometry, Short video (4:25 min) by Henry Segerman
and Saul Schleimer on projecting the hemisphere model to Klein disc, Poincare
disc and upper half-plane.
- Even games on hyperbolic field on Zeno Rogue webpage
(this will be more related to the later parts of the course) .

114
8 Classification of isometries of H2
8.1 Reflections

Definition. A reflection rl with respect to a hyperbolic line l is an isometry preserving

the line l pointwise and swapping the half-planes.
Notice that as in Euclidean and spherical case, if A′ = rl (A) then the line AA′
is perpendicular to the line l (we can see this from the pair of congruent triangles
△AM B ∼ = △A′ M B where M = AA′ ∩ l and B ∈ l any point B ̸= M ).

Example. In the Poincaré disc and upper half-plane models: reflections are repre-
sented by Euclidean reflections and inversions.
More precisely,

- In the Poincaré disc, the (Euclidean) reflection with respect to any diameter and
an inversion with respect to any circle orthogonal to the absolute is hyperbolic
reflection (as it is an isometry which preserves the corresponding hyperbolic line
pointwise and swaps the half-planes).

- Similarly, in the upper half-plane, (Euclidean) reflections with respect to vertical

lines and inversions with respect to the circles orthogonal to the absolute are
hyperbolic reflections.

Example-exercise. In the Klein disc model: given A and l, one can construct rl (A).
Hint: we know how to construct a midpoint of a segment and a line orthogonal to the
given line and crossing it in a given point, see also Fig. 103.

Figure 103: Reflecting a point with respect to a line in the Klein model.

Next, we consider a reflection in the hyperboloid model. Let l = la = {x | ⟨x, a⟩ = 0},

where ⟨a, a⟩ > 0.

115
Theorem 8.1. In hyperboloid model: given a s.t. ⟨a, a⟩ > 0 (i.e. ⟨x, a⟩ = 0 defines a
line la ), the map ra : x 7→ x − 2 ⟨⟨x,a
a,a⟩
⟩
a is the reflection with respect to the line la .
Proof. First, notice that the map ra preserves the pseudo-scalar product ⟨x, y ⟩ (check
this by a direct computation similar to one in HW 2.7). This implies that the hy-
perboloid is mapped by ra to itself. Also, ra is a linear transformation: indeed, if
a = (a1 , a2 , a3 ) and x = (x1 , x2 , x3 ) then ra (x) = Ax where A = I − ⟨a,a
2
⟩
B for the
following matrix  2 
a1 a1 a2 −a1 a3
B = a1 a2 a22 −a2 a3  .
a1 a3 a2 a3 −a23
Hence, ra preserves the cross-ratio, and therefore preserves the distance. So, ra is an
isometry of the hyperboloid model. Furthermore, if ⟨a, x⟩ = 0 then ra (x) = x, which
means that ra preserves the line la pointwise. As ra (x) ̸= x for x ∈/ la , we conclude
that ra ̸= id. Hence, ra is the reflection with respect to la .

8.2 Classification
Theorem 8.2. Any isometry of H2 is a composition of at most 3 reflections.
Proof. The proof of the theorem is very similar to the proof of its Euclidean analogue,
Theorem 1.10.
Let f ∈ I som(H2 ) be an isometry. In view of Theorem 6.14 it is determined by an
image of a flag. Consider a flag F = (A, l, h+ ) where A ∈ H2 is a point, l is a ray from
A and h+ is a choice of half-plane with respect to l. Let f (F ) = (A′ , l′ , (h′ )+ ). Then f
can be obtained as a composition of the following three reflections:
- Let r1 be the hyperbolic reflection which takes A to A′ (i.e. the reflection with
respect to the perpendicular bisector of AA′ ).
- Let r2 be the reflection which preserves A′ and takes r1 (l) to l′ (i.e. the reflection
with respect to the angle bisector of the angle at A′ formed by r1 (l) and l′ ).
- If r2 ◦ r1 ̸= f , let r3 = rl′ be the reflection with respect to l′ .
Then the composition r3 ◦ r2 ◦ r1 (or just r2 ◦ r1 ) takes the flag F to f (F ) and hence,
coincides with f .

Example 8.3. Let l1 and l2 be two lines in hyperbolic plane and r1 and r2 be the
reflections with respect to them. What can we say about the composition r2 ◦ r1 ?
- If l1 ∩ l2 = A ∈ H2 , then we can take the intersection point A to the centre of the
Poincaré disc model, so that the hyperbolic reflections with respect to l1 and l2
will be represented by Euclidean reflections in the model, see Fig. 104, left. We
conclude that r2 ◦ r1 is a rotation about the intersection point A by the angle 2θ
where θ is the angle between the lines.
- If l1 ∩ l2 = A ∈ ∂ H2 , we can map the point A to the point ∞ of the upper half-
plane model, see Fig. 104, middle. Then the reflections r1 and r2 are represented
by Euclidean reflections with respect to vertical lines. Applying the a suitable
isometry az + b, a, b ∈ R we can assume that the vertical lines are Re(x) = 0 and
Re(x) = 1, and r2 ◦ r1 is a translation z → z + 2.

116
 - If l1 ∩ l2 = ∅, i,e, the lines are divergent, then they have a common perpendicular.
Let h be the common perpendicular, we can map it to the imaginary axis in the
upper half-plane model. Then the lines l1 and l2 are represented by half-circles
of centred at the origin, and after applying isometry z → az we may assume
that these are half-circles given by |z | = 1 and |z | = k (see Fig. 104, right).
The reflections r1 , r2 in this case are inversions with respect to the circles, i.e.
are given by z → 1/z¯ and z → k 2 /z¯. So, the composition can be written as
r2 ◦ r1 : z → k 2 /(1/z¯) = k 2 z .

l1
θ l2
ki
l1 l2
i l2
l1

0 1 0

Figure 104: Compositions of two reflections.

Corollary 8.4. A non-trivial orientation-preserving isometry of H2 has either 1 fixed

point in H2 , or 1 fixed point on the absolute, or two fixed points on the absolute.

Proof. By Theorem 8.2 any hyperbolic isometry is a composition of at most 3 reflec-

tions. Hence, an orientation-preserving isometry is either identity or a composition of
2 reflections. Compositions of two reflections are considered in Example 8.3, so we see
that the result depends on mutual position of the corresponding lines l1 , l2 and

- if the lines intersect, r2 ◦ r1 is a rotation and has a unique fixed point inside H2 ;

- if the lines are parallel, then r2 ◦ r1 has a unique fixed point at the boundary (and
no fix points inside H2 );

- if the lines are divergent, then r2 ◦ r1 has two fixed points at the boundary (and
no fixed points inside H2 ).

Definition 8.5. A non-trivial orientation-preserving isometry of H2 is called

- elliptic if it has 1 fixed point in H2 ,

- parabolic if it has 1 fixed point at ∂ H2 ,

- hyperbolic if it has 2 fixed points at ∂ H2 .

Exercise. An orientation-reversing isometry of H2 is either a reflection or a glide

reflection (see also HW 17.3).

117
Example 8.6. In the upper half-plane model, an orientation-preserving isometry is
represented by the transformation z 7→ az +b
cz +d
with a, b, c, d ∈ R, ad − bc = 1. When is
it elliptic? parabolic? hyperbolic?
We consider the fixed points, i.e. the solutions for
az + b
z= ,
cz + d
this is equivalent to cz 2 + z (d − a) − b = 0, which has the solutions
p
a − d ± (d − a)2 + 4bc
z1,2 = .
2c
Taking in account that ad − bc = 1, we can rewrite the discriminant D = (d − a)2 + 4bc
as follows:

D = (d − a)2 + 4bc = (d − a)2 + 4(ad − 1) = (d + a)2 − 4.

So, we conclude that

- when |d + a| < 2, we have D < 0 and the equation has two complex conjugate
roots (exactly one in the upper half-plane);

- when |d + a| = 2, we have D = 0 and the equation has a unique (double) real

root;

- when |d + a| > 2, we have D > 0 and the equation has two distinct real roots.

See Fig. 105. We conclude that the type of the transformation z → az +b

cz +d
depends on
 
a b
the trace a + d of the matrix : it is elliptic if |d + a| < 2, parabolic if |d + a| = 2
c d
and hyperbolic if |d + a| > 2.

Figure 105: Fixed points of elliptic, parabolic and hyperbolic transformations.

Warning: To apply the trace criterion one first need to check that ad − bc = 1!

Remark 8.7 (Invariant sets for isometries). The following sets preserved by elliptic,
parabolic and hyperbolic isometries respectively (but not pointwise), see Fig. 106:

• elliptic isometries preserve concentric circles;

• parabolic isometries preserve curves called horocycles;

• hyperbolic isometries preserve curves called equidistant curves.

118
 Type elliptic parabolic hyperbolic

Poincaré
disc

UHP

invariant circles horocycles equidistant curves

sets

Figure 106: Elliptic, parabolic and hyperbolic transformations in Poincare disc and
upper half-plane. Invariant sets (in blue) and families of lines orthogonal to them
(magenta).

8.3 Horocycles and Equidistant curves.

Motivation. A circle is a set of points on the same distance from a given point.

Properties:

1. All lines through the centre are orthogonal to the circle.

2. The distance between two concentric circles γ and γ ′ is constant

(i.e. given a point A ∈ γ and a closest to A point A′ ∈ γ ′ , the distance d(A, A′ )
does not depend on the choice of A).

Definition 8.8. A horocycle h is the following limit of circles:

- let P ∈ H2 be a point, and l be a ray from P ;

- for t > 0 let Ot ∈ l be a point s.t. d(P, Ot ) = t;

- let γ (t) be a circle centred at Ot of radius t;

- a horocycle is h = lim γ (t).

t→∞

- The point X = lim O(t) ∈ ∂ H2 is called the centre of the horocycle h.

t→∞

See Fig. 107.

119
 P X

Figure 107: Horocycle as a limit of circles.

Remark 8.9 (Horocycles in the models).

- In the Poincaré disc, every circle tangent to the absolute represents some horo-
cycle (and every horocycle is a circle tangent to the absolute).
To see that a given circle C tangent to the absolute is a horocycle, consider a
point P ∈ C . We can map P to the centre of the disc. Then C is mapped to the
circle in Definition 107

- In the upper half-plane, a horocycle can be represented by a circle tangent to

the absolute and by a line parallel to the absolute (the later is tangent to the
absolute at ∞).

- Isometries act transitively on the horocycles (in the upper half-plane, take the
centre of the horocycle to ∞, then scale by z → az if needed).

Properties of horocycles:

1. All lines through the centre of the horocycle are orthogonal to the horocycle.

2. The distance between two concentric horocycles h and h′ is constant.

(i.e. given a point A ∈ h and a closest to A point A′ ∈ h′ , the distance d(A, A′ )
does not depend on the choice of A).

Both of these properties are clear for the horocycle centred at ∞ in the upper half-plane,
and hence, hold for any other horocycle.

Definition 8.10. An equidistant curve e to a line l is a locus of points on a given

distance from l.

How to find the equidistant curve?

- Let l be the line 0∞ through 0 and ∞ in the upper half-plane, see Fig. 108, left;

- take a point A on the distance d = d(A, l) from l;

- then the map f : z → k z , k ∈ R+ is an isometry preserving the line 0∞, hence

d(l, f (A)) = d(l, A) = d;

120
 - by varying k get a (Euclidean) ray OA which represents a part of the horocycle;

- reflection rl with respect to l is also an isometry, so we get another ray;

- points inside the cone are closer to l than A, and points outside are further away.

l
d
A

Figure 108: Examples of equidistant curves.

Examples. In UHP, if l is a vertical ray 0∞, then e is a union of to (Euclidean) rays

from 0 making the same angle with l. If l is represented a half of Euclidean circle, then
e is a “banana”. In the Poincaré disc: also get banana, see Fig. 108.

Properties of equidistant curves:

1. All lines orthogonal to l are orthogonal to all equidistant curves for l.

2. Two equidistant curves to the same line stay on the same distance.

3. Distance between two equidistant curves e, e′ is achieved along an orthogonal line.

To see these properties, we draw l as a vertical half-line through 0 in the upper half-
plane, see Fig. 109. Then any equidistant curve to l is a union of two (Euclidean) rays
from the origin O; any line orthogonal to l is represented by a semicircle centred at O,
so it orthogonal to the equidistant curve.
Furthermore, if e and e′ are two distinct equidistant curves, A ∈ e is a point, and
A′ ∈ e′ is a point on the same line m ⊥ l, then d(A, A′ ) does not depend on the choice
of A ∈ e: any other choice A1 ∈ e may be obtained from the initial by the isometry
z → k z for some k ∈ R+ , this isometry will take l to itself, so d(A, l) = d(A′ , l).
Now, let e and e′ be two equidistant curves for l lying on distances d > d′ respec-
tively. Let A ∈ e and let m be a line through A orthogonal to l, let M = l ∩ m. Suppose
that d(A, e′ ) = d(A, B ) for some B ∈ e′ , B ∈
/ m. Let n be a line through B orthogonal
to l and N = n ∩ l. Then

d(A, l) = d(A, B ) + d(B , l) = d(A, B ) + d(B , N ) > d(A, N ),

where the last inequality holds in view of triangle inequality. We obtain d(A, l) >
d(A, N ) which is impossible as N ∈ l.

Remark 8.11. 1. Given an elliptic, parabolic or hyperbolic isometry f of H2 , for

every point A ∈ H2 there exists a unique invariant curve of f passing through A
(circle, horocycle or equidistant curve respectively). There is also a unique line
orthogonal to all invariant curves of f and passing through A.

121
 l N B0
e0
A0 M A0
e
A A

Figure 109: Properties of equidistant curves.

2. Representation of elliptic, parabolic and hyperbolic isometries as r2 ◦ r1 is not

unique: r1 is a reflection with respect to any line from the orthogonal family,
then there is a unique choice for r2 .

Remark 8.12. How do the horocycles and equidistant curves look like in the Klein
model? A horocycle is by definition a limit of circles, i.e. sets represented by ellipses
in the Klein model. So a horocycle is an ellipse tangent to the absolute. One can
show (using a projection) that equidistant curves are pieces of ellipses intersecting the
absolute.

8.4 Discrete reflection groups in S 2 , E2 and H2 (NE)

(Non-examinable section!)

Recall from Definition 1.31 that an action G : X of a group G on a metric space X

is discrete if none of its orbits has accumulation points.
Example. A group generated by two reflections with respect to two intersecting lines
(in S 2 , E2 or H2 ) is discrete if and only if the angle α between the lines satisfies
α = k π /m, where k , m ∈ Z+ (see also Problems Class 2, Question 2.1). The rays lying
on these lines bound a fundamental domain if and only if α = π /m for some m ∈ Z≥2 ,
integer greater than 1 (see also Problems Class 2, Question 2.2).
In general, discrete group actions in hyperbolic case are not so easy to construct,
and the aim of the current section is to describe one construction which produces
infinite series of examples.
Definition. A reflection group is a group generated by (finitely many) reflections. In
other words, we chose (finitely many) lines l1 , . . . , ln in X = S 2 , E2 or H2 and consider
a minimal group G = ⟨r1 , . . . , rn ⟩ containing the reflections r1 , . . . , rn with respect to
these lines.
A reflection group G = ⟨r1 , . . . , rn ⟩ may contain infinitely many reflections: for any
generating reflection rj , j ∈ 1, . . . , n and any element g ∈ G (i.e. g = rik ◦ · · · ◦ ri2 ◦ ri1 )
the element h = g rj g −1 is a reflection with respect to the line g (lj ). We will call all
these lines mirrors of G.
Question. When a reflection group G is discrete?
Suppose G : X is a discrete action. Then there is a point x ∈ X not lying on any
mirror of G (otherwise, all orbits have accumulation points). Moreover, since the orbit
of x should not have accumulation points, there is a disc centred at x containing no

122
other elements from the same orbit. Hence, there is a (possibly smaller) disc around x
not intersected by any mirror of G.
Let P be the largest connected set containing x and not intersected by any mirror
of G. It is clear, that P is bounded by mirrors of G (otherwise, it could be larger). If
P is also a bounded set in the metric space X , then P is bounded by finitely many
mirrors (otherwise one can find an accumulation point of these mirrors and the action
of G will not be discrete). So, if P is a bounded set, it is a polygon with finitely many
sides. In view of the example above, we conclude that all angles of P are of the form
π /ki , ki ∈ Z+ (otherwise the group is not discrete or P is crossed by some mirror).
Definition. Polygons with angles π /ki , ki ∈ Z+ are called Coxeter polygons. (And
more generally, in higher dimensions, a Coxeter polyhedron is a polyhedron whose
hyperfaces meet each other at angles π /ki , ki ∈ Z+ ).
Theorem. Every discrete reflection group in X = S 2 , E2 or H2 has a fundamental
domain, and if the fundamental domain is a bounded set then it is a Coxeter polygon.
Moreover, every Coxeter polygon is a fundamental domain for some discrete reflection
group.
. We have (partially) justified the first part of the theorem. The second part will
follow from a more general construction and Poincare’s theorem (which we will discuss
but will not prove later).
The theorem implies that studying Coxeter polygons is (more or less) equivalent to
studying reflection groups. Below we list what is known about them in S 2 , E2 or H2 .

S 2 : A spherical Coxeter digon exists for any angle π /k , k ∈ Z≥2 .

A spherical Coxeter triangle should have angles (π /k , π /l, π /m), k , l, m ∈ Z≥2 ,
integers larger than 1. As the sum of angles of a spherical triangle is larger than
π , we conclude that k1 + 1l + m1 > 1. Assuming without loss of generality that
k ≤ l ≤ m we obtain k = 2. Then, if l = 2 we may have any value of m ≥ 2. If
l = 3 then m = 3, or 4, or 5. And if l ≥ 4 there are no solutions.
We conclude that a spherical Coxeter triangle should have angles (π /k , π /l, π /m)
with (k , l, m) one of the following triples: (2, 2, m), (2, 3, 3), (2, 3, 4), (2, 3, 5).
(Notice that the corresponding reflection groups are symmetry groups of dihe-
dron, tetrahedron, cube and octahedron, icosahedron and dodecahedron respec-
tively. See also Section 2.9).
A spherical Coxeter quadrilateral or any other n-gon with n ≥ 3 does not exist
as the condition on the angles of a Coxeter polygon is not compatible with the
angle sum inequality for spherical n-gons, i.e. αi ≥ (n − 2)π .
P

E2 : A Euclidean Coxeter triangle should have angles (π /k , π /l, π /m) satisfying

1
k
+ 1l + m1 = 1. As before, assuming k ≤ l ≤ m one can easily check that this is
only possible for the following values of (k , l, m): (2, 3, 6), (2, 4, 4), (3, 3, 3).
A Euclidean Coxeter quadrilateral only can satisfy the angle sum condition when
all its angles are right angles (i.e. it is a rectangle).
From the same angle sum condition we conclude that no Euclidean Coxeter poly-
gon has more than 4 sides.
Remark: we can also have some unbounded domains serving as fundamental
domains of discrete reflection groups, i.e. any domain bounded by one or two
parallel lines, or two lines intersecting at an angle π /k , k ∈ Z≥2 , or a domain
bounded by two parallel lines and one line perpendicular to them.

123
 H2 : In hyperbolic plane, we have infinitely many Coxeter polygons.
For every triple (k , l, m) with k , l, m ∈ Z≥2 and k1 + 1l + m1 < 1 (i.e. any triple not
mentioned above) there exists a hyperbolic Coxeter triangle.
For four or more integer numbers (different from (2, 2, 2, 2)) there exists a hy-
perbolic Coxeter polygon with the corresponding angles (and more over, one can
show that there is an (n − 3)-parametric family of n-gons with given n angles!)
And each of these Coxeter polygons is a fundamental domain for the correspond-
ing discrete reflection group acting on H2 !

So, there are many more Coxeter polygons (and hence discrete reflection groups)
on hyperbolic plane than on the sphere or Euclidean plane.
Remark. The situation in higher dimension is even more different for the three types
of spaces.

• S d : Spherical Coxeter polytopes were classified by H. S. M Coxeter in 1934. There

are finitely many of them in each dimension d ≥ 2 (non-empty set of them); the
corresponding reflection groups are symmetry groups of regular polytopes.

• Ed : Euclidean were also classified by H. S. M. Coxeter. There are finitely many

of them in each dimension (up to affine transformations, like the ones changing
the proportions of sides in a rectangle).

• Hd : The question of classification of hyperbolic Coxeter polytopes is still a widely

open question, probably very far from a solution. The main things known to the
moment are:

- There are infinitely many examples of hyperbolic Coxeter polytopes in di-

mensions d = 2, 3, 4, 5, 6.
- Theorem (Vinberg, 1984).
Bounded hyperbolic Coxeter polytopes do not exist in Hd for any d ≥ 29.
- In dimensions d = 7 and d = 8, only 2 and 1 bounded examples are known
respectively.
- In dimensions d = 9, ..., 29 – nothing known at all!

124
8.5 References
- Material on types of isometries in hyperbolic geometry, and on horocycles and
equidistant curves is based on Lecture IX of Prasolov’s book.
Alternatively, see Section 5.3 in Prasolov and Tikhomirov (p. 113–116).

- Tilings of S 2 , E2 and H2 by triangles are described in Lecture X of Prasolov’s

book.
Alternatively, see Section 5 of the Addendum in Prasolov and Tikhomirov (p. 185–
187).

- One can find a detailed exposition concerning discrete reflection groups in

E. B. Vinberg (Ed.), Geometry II , Encyclopaedia of Mathematical Sciences, Vol.
29, Springer-Verlag.

- Webpages:

- Hyperbolic tessellations, including some printable hyperbolic tilings, web-

page by David E. Joyce.
- Dancing circles. Hyperbolic tilings dansing to music from Tchaikovsky’s
ballet “The Nutcracker”. On Mathematics Imagery by Jos Leys.

125
9 Geometry in modern maths - some topics (NE)
(Non-examinable section!!!)

The aim of this section is provide an overview of the zoo of different geometries we
have seen before, to unify them and to put in the context of modern mathematics.
Note, that the aim as stated is very broad and any sense of completeness here would
not be achievable within several lectures. So, it is just a brief discussion of a selection
of topics, which depends on and reflects my personal preferences.

9.1 Taming infinity via horocycles

Martin Hairer (2014 Fields medallist) gave a Collingwood Lecture 2015 in Durham and
spoke about “Renormalisation”; in particular, he said something close to: ”If you have
a diverging integral, subtract infinity (in a coherent way) and work then with finite
values”.

We illustrate this idea with horocycles:

- any point of a horocycle h is on infinite distance from the centre X of the horo-
cycle;
- two concentric horocycles are on a finite distance from each other;
- choose “level zero” horocycle, and measure the (signed) distance to it, see Fig. 110.

−d

X 0

Figure 110: Distance between horocycles is finite.

Lambda-length of an infinite segment:

- Given X, Y ∈ ∂ H 2 , choose horocycles hX and hY centred at these points.
- Let dX Y be the finite portion of the line X Y lying outside of hX and hY , see
Fig. 111, left. It is a signed length, may be zero or negative if hx intersects hy ,
see Fig. 111, middle.
- Define λX Y = exp(dX Y /2), the lambda-length of X Y .
Recall from Euclidean geometry Ptolemy Theorem (see Problems Class 5 for the
proof):

Ptolemy Theorem. In E2 , a cyclic quadrilateral AB C D satisfies

|AC | · |B D| = |AB | · |C D| + |AD| · |B C |.

126
 Y Y B

dXY A C
X X

D
λXY = exp(dXY /2) λAC · λBD =
here, dXY < 0 λAB · λCD + λAD · λBC

Figure 111: Lambda-length and hyperbolic Ptolemy Theorem.

There is a hyperbolic analogue of this theorem (see Fig. 111, right):

Theorem 9.1 (Hyperbolic Ptolemy Theorem). For a hyperbolic ideal quadrilateral
AB C D, choose any horocycles centred at A, B , C, D. Then
λAC · λB D = λAB · λC D + λAD · λB C .

Remark.
1. The proof of the hyperbolic Ptolemy Theorem can be done by a computation in
the upper half-plane (omitted). See also [33].
2. The identity does not depend on the choice of the horocycles: if we change one
horocycle taking another horocycles on distance d, then all lengths of arcs with
endpoint at the centre of that horocycle will change by the same value d, and
hence all lambda-length of the corresponding arcs will multiply by exp(d/2).
So, in the Ptolemy relation all summands of the identity will be multiplied by
exp(d/2), and the relation will be preserved.
3. This (together with applying an isometry to the quadrilateral) implies that it is
sufficient to check the identity for the configuration shown in Fig. 112, where the
identity rewrites as λAC = 1 + λB D .
4. Why do we care?
- Given an ideal triangle A1 , A2 , A3 and c12 , c23 , c31 ∈ R+ there exists a unique
choice of horocycles such that λAi Aj = cij (we leave this as an exercise).
- So, given a triangulated polygon A1 A2 . . . An and numbers λij > 0 associated
to the diagonals and sides of the polygon in the triangulation, one can
find a unique hyperbolic metric on the corresponding surface (and a unique
choice of horocycles) so that lambda lengths of the arcs of triangulation
coincide with the given numbers λij . In other words, the set of numbers
{λij } associated to the arcs of the triangulation provides the coordinates on
the space of all (decorated) hyperbolic structures on this polygon.

127
 A=0 B=1 C=x

Figure 112: Diagram for the proof of hyperbolic Ptolemy Theorem.

- One can retell the same story about any surface which one can triangulate
into ideal polygons.
- Using Ptolemy relation (several times if needed) we can compute a lambda
length of any diagonal in the polygon as a function of initial values λij
associated to the arcs of the triangulation.
- This describes one of the connections of hyperbolic geometry to a recent the-
ory of Cluster Algebras, which was introduced by Fomin and Zelevinsky in
2002 and turned out to be connected to numerous fields in mathematics and
mathematical physics (including combinatorics of polytopes, representation
theory, integrable systems).

9.2 Three metric geometries: S 2 , E2 , H2 , unified

One can consider a sphere of growing radius r, which eventually, when it is very big,
is approaching a plane. Notice that the distance on the sphere of radius r is given by
d(A, B ) = r · ∠AOB .
Similarly, we can introduce a parameter into the distance function on the hyperbolic
plane (say in the Klein disc model):
1
d(A, B ) = R · |ln[A, B , X, Y ]|,
2
so that when R is big the geometry is very similar locally to Euclidean one, while when
R is smaller it gets “more and more hyperbolic”, see Fig. 113.
Remark 9.2 (About curvature). More precisely, this is quantified by a notion of
curvature - which is very large and positive for a sphere of small radius, then decreases
to 0 when the radius of the sphere grows, then is zero for a plane, and small negative
for hyperbolic plane defined with large parameter R, and very large negative for a
hyperbolic plane defined with small “radius” R. The standard hyperbolic plane, which
we considered in the course, has constant curvature -1. We will not focus on the notion
of curvature in this course, but one can find more about it in any course of Differential
Geometry, see for example [7].
Below, we will show that spherical geometry (of any radius) can be unified with
hyperbolic geometry (again of any radius). This will explain why we have obtained
similar formulae in different geometries.

128
 d(A, B) = R · 12 [A, B, X, Y ] d(A, B) = r · ϕ

Figure 113: Spheres of growing radius r approach a plane – and similarly, hyperbolic
planes of “growing radius R” approach the same plane.

We will use complex projective geometry to show that the distance on S 2 can be
written as
r
d(A, B ) = ± |ln[A, B , X, Y ]|.
2i
We will take two points A and B on a sphere of radius r, assuming that ∠AOB = φ.

- The sphere of radius r is given by an equation x21 + x22 + x23 = r2 . For this
computation, we will assume that xi ∈ C, so x = (x1 , x2 , x3 ) ∈ C3 , or more
precisely x ∈ CP 2 , i.e. a triple (x1 , x2 , x3 ) is considered up to multiplication by
a (non-zero) complex number.

- In case of H2 we consider the hyperboloid as a sphere x21 + x22 + x23 = −R2 of

imaginary radius iR, rewriting this for x′3 = ix3 we get exactly the hyperboloid
model considered before.)

- In the hyperboloid model of H2 , the distance is expressed through the cross-ratio

[A, B , X, Y ] (where X and Y correspond to the endpoints of the lines through
the points A and B ).

- For the sphere, it is clear what are A and B (the points), but what would be the
objects corresponding to X and Y ?

X
B
A

Figure 114: {X, Y } = ΠAB ∩ {⟨x, x⟩ = 0}.

129
 - To find the points X, Y we use the same rule as in the hyperboloid model:

{X, Y } = ΠAB ∩ {⟨x, x⟩ = 0},

i.e. the intersection of the plane spanned by AB with the cone, see Fig. 114.

- Both in spherical and hyperbolic case, the plane through the points a = (a1 , a2 , a3 )
and b = (b1 , b2 , b3 ) is given by

(a1 + λb1 , a2 + λb2 , a3 + λb3 ),

where λ ∈ C∪∞ (the points a and b correspond to λ = 0 and λ = ∞ respectively).

- Intersection of this plane with the cone ⟨x, x⟩ = 0 gives

(a1 + λb1 )2 + (a2 + λb2 )2 + (a3 + λb3 )2 = 0.

- Taking in account ⟨a, a⟩ = r2 = ⟨b, b⟩ and ⟨a, b⟩ = r2 cos φ this gives

1 + 2λcosφ + λ2 = 0.

- Solving the equation for λ we get x = λ1 and y = λ2 :

λ1,2 = cosφ ± i sin φ,

which allows to compute the points x and y .

- [a, b, x, y ] = [0, ∞, λ1 , λ2 ] = λλ11−∞

−0
= λλ12 = exp(±2iφ), from this we con-
 λ2 − 0
λ2 −∞
clude that ln[a, b, x, y ] = ±2iφ and
r
φ=± |ln[a, b, x, y ]|.
2i

- Finally, we get for the spherical case:

r
d(A, B ) = ± |ln[A, B , X, Y ]|.
2i
Notice that by considering a sphere of imaginary radius r = iR we recover
the distance formula from the Klein model of hyperbolic geometry, d(A, B ) =
± R2 |ln[A, B , X, Y ]|.

Remark 9.3. This explains appearance of similar formulae in spherical and hyperbolic
geometries, in particular, this gives a proof of the second cosine law in the hyperbolic
case.
Different geometries of constant curvature can be explicitly compared as in the
following theorem:
Theorem 9.4 (Comparison Theorem, Aleksandrov-Toponogov). Given a, b, c ∈ R≥0
such that a + b < c, a + c < b and b + c < a, consider triangles in H2 , E2 and S 2 with
sides a, b, c. Let mH2 , mE2 and mS 2 be the medians connecting C with the midpoint of
AB in each of the three triangles (see Fig. 115). Then mH2 < mE2 < mS 2 .
The proof uses technique of Jacobi fields (which you can learn in Riemannian Ge-
ometry module). See Toponogov’s Theorem and Applications by Wolfgang Meyer for
the proof and applications.

130
 a m 2 b a b a b
H mE2 mS 2
c
c
c

Figure 115: Comparison Theorem: mH2 < mE2 < mS 2

9.3 Discrete groups of isometries of H2 : Examples

Hyperbolic plane is rich with examples of discrete groups of isometries. At the same
time it is not obvious even how to start to construct examples. We will describe one
construction with the rough idea like:
- take a good polygon;
- find a good pairing of its sides;
- this will provide a tessellation of the hyperbolic plane, and hence a discrete group
acting on it.

Assumption: Let G ∈ I som+ (H2 ) be a discrete group. We will assume that it has a
fundamental domain F s.t. F is a polygon with finitely many sides.
Then we will get a tiling of H2 (in a similar manner to how copies of squares can
tile a Euclidean plane).
By this we have described what do we mean by a good polygon, now we need to
formulate what is a good side pairing. For this we will need several steps:

ai

a4
F a1 gi−1
a3 g1 F aj
a2

g2 F

Figure 116: Side pairing.

Constructing a side pairing:

- Denote a1 , . . . , an the sides of the fundamental domain F (there are finitely many
by assumption).
- We have a tiling of H2 by copies of the fundamental domain F . For every side ai
of F consider the adjacent fundamental domain gi F , gi ∈ G, gi ̸= id such that
ai = F ∩ gi F
- Note that gi−1 (ai ) ∈ F , i.e. gi−1 (ai ) is another side of F , some aj , so that we get
a map gi−1 : ai → aj . As gj−1 = gi : aj → ai , we get a pairing for all sides of F .

131
Example: Let G : E2 be the group generated by two shifts g1 : z → z + 1 and
g2 : z → z + i. Then the fundamental domain for this group is a square, and the side
pairing it realised by g1 and g2 (g1 is pairing vertical sides of the square while g2 does
it for horizontal ones).

Constructing a graph Γ:
- Let A1 , . . . , An be the vertices of F . Every gi takes two adjacent vertices Ai Ai+1
to other two vertices.

- Consider an oriented graph Γ:

vertices of Γ are vertices Ai of F ,
edges of Γ are side pairings: Ai → Aj if gi (Ai ) = Aj

Example: For the group G : E2 generated by g1 : z → z + 1 and g2 : z → z + i the

graph Γ has 4 vertices, connected as in Fig. 117, right.

g2 F
A2 A3 A2 g1 A3

g1−1 F F g1 F g2 g2

A1 A4 A1 g1 A4
g2−1 F

Figure 117: Side pairing for the group G : E2 together with the graph Γ.

Properties of Γ and equivalent vertices:

- Notice that every vertex Ai is incident to exactly two edges of Γ (one coming
from the side pairing of Ai Ai+1 and another from Ai−1 Ai ).

- We conclude that Γ consists of finitely many cycles.

- Vertices in one cycles are called equivalent.

(These vertices lie in one orbit of G = ⟨g1 , . . . gn ⟩).

- Consider one cycle. Relabel the vertices and the maps so that A1 , . . . , AK be
consecutive vertices in one cycle and gi (Ai ) = Ai+1 .

Lemma. Let A1 , . . . , Ak make one cycle, so that gi (Ai ) = Ai+1 , gk (Ak ) = A1 , where
gi are side pairings of F and A1 , . . . , Ak ∈ H2 (but not ∂ H2 ). Then g = gk gk1 . . . g1 is
a rotation about A1 by the angle α1 + · · · + αk , where αi is the angle of F at Ai .
Proof. We will prove the Lemma in additional assumption that g1 , . . . , gk are orientation-
preserving.

132
 - Since g (A1 ) = gk gk1 . . . g1 (A1 ) = A1 and each of g ∈ G ⊂ I som+ (H2 ) is
orientation-preserving, g is a rotation about A1 .

- Consider
− α1 −α3 − α2
f = RA 1
gk . . . RA 3
g2 RA 2
g1 ,
where RA
αi
i
is a rotation around Ai by αi .

- Notice that f (A1 ) = A1 (g1 takes A1 to A2 , then RA

α2
2
preserves it, then g2 takes
to A3 , etc. till gk takes it back to A1 ).

- Denote e′i (resp. ei ) the edge of F preceding (resp. succeeding) the vertex Ai in
the clockwise order. We will assume that e′1 is paired to e2 by g1 , see Fig. 118,
left.

- Then, f (e1 ) = e1 . Indeed, g1 (e1 ) makes angle α1 with e2 , so, RA

α2
g (e1 ) makes
2 1
angle α1 with e2 . Then g2 RA2 g1 (e1 ) makes angle α1 with e3 and RA3 g2 RA
′ α2 α3 α2
g (e1 )
2 1
makes angle α1 with e3 . Continuing the same way till the end we get that f (e1 )
′

makes angle α1 with e′1 , so f (e1 ) = e1 .

- Since f (A) = A and f (e1 ) = e1 we conclude that f = id (by uniqueness of

isometry taking a flag to itself).

- Exercise: for every h ∈ I som+ (H2 ) holds hRA α

= Rhα(A) h.
(Hint: recall and use properties of fixed points, see Proposition 1.18).

- From the exercise above we have

− α1 −α3 − α2 − α1 −α3 −α2

id = f = RA 1
gk . . . RA 3
g2 RA 2
g1 = RA 1
gk . . . RA 3
g2 g1 RA 1
= ...
− α1 − αk − α3 − α2 −(α1 +α2 +···+αk )
= gk . . . g1 RA 1
◦ RA 1
◦ · · · ◦ RA 3
◦ RA 1
= gk . . . g1 RA1 ,
α1 +α2 +···+αk
which implies that gk . . . g1 = RA 1
.

A1 A1
e01 e1 e02
A2 A2
R2α2 g1 (e1 ) g2
g1 (e1 ) g1
g1 e2 F g1 F

A3
g2 F
g2 g1 F

Figure 118: To the proof of the lemma (left) and the claim (right).

Claim. Polygons gk F , gk gk−1 F ,. . . ,gk gk−1 . . . g1 F have a common vertex A1 , with

angles αk , αk−1 , . . . , α1 at A1 .

133
 The proof can be done inductively by considering the sets gi F , gi gi−1 F ,. . . ,gi gi−1 . . . g1 F
which share the vertex Ai+1 . We skip the proof but illustrate the first steps in Fig. 118,
right.

Corollary. Elements of the group ⟨g1 , . . . , gn ⟩ generated by side pairings tile the
neighbourhood of A1 if and only if α1 + . . . αk = 2π /m for m ∈ N.

The corollary provides us with a necessary condition, saying what we need to require
from a side pairing if we want it to define a discrete action. The following theorem
shows that this necessary condition is also sufficient:

Theorem 9.5 (Poincaré’s Theorem.). Let F ⊂ H2 be a convex polygon, with finitely

many sides, without vertices at the absolute, s.t.

a) its sides are paired by isometries {g1 , . . . , gn };

b) angle sum in equivalent vertices is 2π /mi for mi ∈ N.

Then

1) the group G = ⟨g1 , . . . , gn ⟩ is discrete;

2) F is its fundamental domain.

3) if all gi ∈ I som+ (H2 ) then defining relations in G are vertex relations like
(gk gk−1 . . . g1 )mi = e.

We omit the proof of the theorem, see the lecture notes Hyperbolic geometry by
Caroline Series for the details.

Remark.

1. Similar statement holds in E2 and S 2 .

2. In H2 we can allow ideal vertices (with an extra condition that for the corre-
sponding cycle the isometry gk gk−1 . . . g1 should be parabolic).

Examples: the following groups are discrete (more examples than we had in the
lecture):

1. P = regular hexagon in E2 , the group G generated by translations pairing the

opposite sides of P , see Fig. 119, left.
There are two classes of equivalent vertices, each containing 3 vertices. As αi =
2π /3 we have that the total angle for every cycle is 2π , as required.

2. P = regular hexagon in E2 , G generated by rotations by 2π /3 about three non-

adjacent vertices, see Fig. 119, middle.
(This time, there is one classes of 3 vertices, with total angle 3 · 2π /3 = 2π and
3 classes of single vertex, with angles 2π /3).

3. Regular hyperbolic octagon with angles π /4m, G generated by hyperbolic trans-

lations pairing the opposite sides of P , see Fig. 119, right.
There is one cycle of vertices, with total angle 8 · π /4 = 2π .

134
 4. Let P be a polygon all whose angles are integer submultiples of π , i.e. the angle
at the vertex Ai is π /mi , mi ∈ N (called Coxeter polygon), G generated by re-
flections with respect to the sides of P .
In this case, every side is paired with itself, every vertex forms a separate class
of equivalent vertices, and the cycle is given by gi+1 gi where gi and gi+1 are
reflections with respect to sides incident to the vertex Ai . Then gi+1 gi is a
rotation around Ai by angle 2αi = 2π /mi . By Poincaré Theorem, the group
G = ⟨g1 , . . . , gn ⟩ is discrete andP is its fundamental domain.

Figure 119: Examples of discrete groups defined by side parings of regular hexagon in
E2 and regular octagon with angles π /4 in H2 .

9.4 Hyperbolic surfaces

Definition 9.6. A surface S is called hyperbolic if every point p ∈ S has a neighbour-
hood isometric to a disc on H2 , see Fig. 120.

Figure 120: Hyperbolic surface.

How to construct such a surface? We will describe several ideas for that.

1. Glue from hyperbolic polygons.

Example: Euclidean torus can be glued from a square with identified opposite sides
(see Example 1.35).

Example: Hyperbolic surface of genus 2 (“two holed torus”) glued of a regular octagon
with angles π /4 (opposite sides identified by hyperbolic translations, as in Fig. 119,
right).
First of all, such a regular octagon exists: it is best viewed as a regular octagon
with the centre at the centre of the Poincaré disc. Then a very small regular octagon is

135
obtuse-angled, as a Euclidean regular octagon, then we increase the size of the octagon
(distance from the centre), and can eventually turn it into ideal octagon, with zero
angles. By continuity, somewhere in between there is a position where the regular
octagon has angles π /4.
One can check, that when identifying the sides as in Fig. 119, right, we identify all 8
vertices, so all 8 angles of size π /4 each will be glued together to form a neighbourhood
with a total angle 2π , as required. It is also easy to see, that every point on a side of
the octagon will have a complete hyperbolic neighbourhood (again of angle 2π ).
The sequence of identifications as in Fig. 121 shows that after gluing we obtain a
sphere with two handles.

Figure 121: Gluing a hyperbolic surface with two handles from a hyperbolic octagon.

Remark. Alternatively, we can glue the sides of the octagon as shown in Fig. 122.
This will result in the same topological surface (sphere with two handles), but the
hyperbolic structure obtained on it will be different.

Figure 122: Another side pairing for the octagon.

2. Pants decompositions.

A pair of pants is a sphere with three holes, see Fig. 123, left. A hyperbolic pair of
pants may be glued from two right-angled hyperbolic hexagons, as in Fig. 123, right.
Gluing several pairs of pants by the boundaries, one can get (almost) every compact
topological surface (see Fig. 124 ) Exceptions are a sphere and a torus, which naturally
carry spherical and Euclidean geometry, but not hyperbolic.

136
 Figure 123: Hyperbolic pair of pants from two right-angled hexagons.

Figure 124: Pants decomposition of a surface.

3. Quotient of H2 by a discrete group.

Let G : H2 be a discrete action. Consider an orbit space H2 /G.

Sometimes we get a hyperbolic surface, but not always (in particular, one needs to
require that G contains no elliptic elements and no reflections).

Example. Consider a regular octagon with angles π /4 and one of two side pairing
described above. Let the sides be paired by isometries g1 , g2 , g3 , g4 . Consider the
group G = ⟨g1 , g2 , g3 , g4 ⟩ generated by g1 , g2 , g3 , g4 . Then H2 /G is a hyperbolic surface
(topologically, sphere with two handles).

What will happen if G contains elliptic elements? One will obtain a surface with cone
singularities.

Example. Consider a regular hyperbolic quadrilateral with angles π /4 and opposite

sides identified. All four vertices will be identified. but the total angle at the resulting
point will be 4 · π /4 = π rather than 2π . Let G = ⟨g1 , g2 ⟩ be the group generated by
the side pairing isometries. Then, H2 /G will be a torus with a cone point (of angle π )
(this structure is called an orbifold rather than manifold).

4. Developing map.

Chose a point P on a hyperbolic surface. For each loop based at P (i.e. a path starting
from P and ending at P ) we construct an isometry on H2 in the following way:

- We can cover every point of the loop by a neighbourhood isometric to a disc in

H2 .

137
 - Since the loop is a compact set, we can choose a finite subcover in any open cover,
i.e. finitely many such neighbourhoods U1 , . . . , Un covering the whole loop.

- Chose a disc Ū1 ∈ H2 isometric to U1 .

- Then attach to Ū1 another disc Ū2 ∈ H2 in the same way as U2 is attached to U1 .

- Continuing in the same way we attach discs Ū3 , . . . Ūn - and finally, since on the
surface we started with a closed loop we attach one more disc Ūn+1 which is
obtained from attaching U1 to Un . The map from the surface to H2 constructed
in this way is called a developing map (or more precisely, the map is from the
paths on the surface starting from the point P ).

- Notice that in general Ūn+1 does not coincide with Ū1 , but they are always con-
gruent.

- Consider the isometry which takes Ū1 to Ūn+1 (“the isometry” since from the
chain of gluing we know which point of the the boundary of Ū1 is mapped to
which point of the boundary of Ūn+1 ).

Figure 125: Developing map.

So, each loop on S gives rise to an isometry of H2 . Consider a group G generated

by all these isometries for 2g generating loops. Then one can show that G acts on H2
discretely , and S = H2 /G is its orbit space.
One can also show the following properties:

- ∀g ∈ G, x ∈ H2 if g x = x then g = id (such an action G : X is called free).

- Every point x ∈ H2 has a neighbourhood containing no points of the orbit Gx.

So, we started with a hyperbolic surface and constructed a discrete group acting
on H2 .

5. Uniformisation theorem.

Theorem 9.7. Any closed oriented hyperbolic (or Euclidean,or spherical) surface is a
quotient of H2 (or E2 , or S 2 ) by a free action of a discrete group.

The proof can be derived from the Poincaré Theorem, see the lecture notes Hyper-
bolic geometry by Caroline Series.

138
9.5 Review via 3D hyperbolic space
I. Four models of H3

Ia. Upper half-space.

- Space: H3 = {(x, y , t) ∈ R3 | t > 0}.

- Absolute: ∂ H3 = {(x, y , t) ∈ R3 | t = 0}.

- Hyperbolic lines: vertical rays and half-circles orthogonal to the absolute.

- Hyperbolic planes: vertical (Euclidean) half-planes and half-spheres centred at

the absolute, see Fig. 126.

- Distance: d(A, B ) = |ln[A, B , X, Y ]| (where X, Y are the ends of the line, and
cross-ratio is computed in a vertical plane through A and B ).
2
- Distance formula: cosh d(u, v ) = 1 + |2uu−3vv|3 ,
(here |u − v |2 = (u1 − v1 )2 + (u2 − v2 )2 + (u3 − v3 )2 ).

A B

Y X

Figure 126: Upper half-space model of H3 .

Isometries.

- Example: Hyperbolic reflections = (Euclidean) reflections with respect to the

vertical planes and inversions with respect to the spheres centred at the absolute.

- f ∈ I som(H3 ) is determined by its restriction to the absolute:

Through every point in H3 we can draw two distinct lines intersecting in that
point, , see Fig. 126, right. So, if we know the images of the endpoints of the
lines then we know the images of the lines and their intersection.

- I som(H3 ) is generated by reflections (every isometry is a composition of at most

4 reflections). Restrictions to ∂ H3 are compositions of (Euclidean) reflections and
inversions.

- I som+ H3 ∼
= Möb.

Spheres, horospheres, equidistant surfaces.

- Spheres: Euclidean spheres (with another centre than in E2 ).

139
 - Horospheres (limits of spheres): horizontal planes and spheres tangent to the
absolute, see Fig. 127, middle.
Horospheres are submanifolds in H3 isometric to E2 (with the same isometry
group, same geodesics i.e. intersections of the horosphere h with the planes
through its centre at the absolute).

- Equidistant surface (to a line): vertical cone (or banana for “half-circle” lines, see
Fig. 127, right).

- Equidistant (to a plane Π represented by vertical half-plane): two (Euclidean)

planes at the same angle to a vertical plane (at Π ∩ ∂ H3 )

- Equidistant (to a plane Π represented by a hemisphere) or two pieces of spheres

at the same angle to the sphere representing Π.

Figure 127: Spheres, horospheres and equidistant surfaces in the upper half-space
model of H3 .

Ib. Poincaré ball.

- Obtained by inversion from the upper half-space model.

- Space: H3 = {(x, y , t) ∈ R3 | x2 + y 2 + t2 < 1}.

- Absolute: ∂ H3 = {(x, y , t) ∈ R3 | x2 + y 2 + t2 = 1}.

- Hyperbolic lines: parts of lines and circles orthogonal to ∂ H3 .

- Hyperbolic planes: parts of planes and spheres orthogonal to ∂ H3 .

- d(A, B ) = |ln[A, B , X, Y ]|
(X, Y the ends of the line, cross-ratio computed in a plane).

Both Poincaré models are conformal: hyperbolic angles are represented by Euclidean
angles of the same size.

Ic. Klein model.

- Space: H3 = {(x, y , t) ∈ R3 | x2 + y 2 + t2 < 1}.

- Absolute: ∂ H3 = {(x, y , t) ∈ R3 | x2 + y 2 + t2 = 1}.

- Hyperbolic lines: chords. See Fig. 128, left.

140
 - Hyperbolic planes: intersections with Euclidean planes.

- d(A, B ) = 12 |ln[A, B , X, Y ]| (X, Y the ends of the line).

- Angles are distorted (except ones at the centre).

- Right angles are easy to control, see Fig. 128, middle and right.

Figure 128: Klein model of H3 : lines, planes and orthogonality.

Id. Hyperboloid model.

- Hyperboloid: x21 + x22 + x23 − x24 = −1, x ∈ R4 .

- Pseudo-scalar product: ⟨x, y ⟩ = x1 y1 + x2 y2 + x3 y3 − x4 y4 .

- Space: ⟨x, x⟩ = −1.

- Absolute: ⟨x, x⟩ = 0.

- Hyperbolic planes: ⟨x, a⟩ = 0 for a s.t. ⟨a, a⟩ > 0.

- d(A, B ) = 21 |ln[A, B , X, Y ]| (cross-ratio of four lines).

⟨u,v ⟩2
- Formula: cosh2 (d(pt1 , pt2 )) = Q(pt1 , pt2 ) where Q⟨u, v ⟩ = ⟨u,u⟩⟨v ,v ⟩
.

II. Orientation-preserving isometries of H3

- In the upper half-space, orientation-preserving isometries take planes to planes,

i.e. circles lying on the absolute to circles on the absolute or lines on the absolute.
Hence, they correspond to Möbius transformation of the absolute ∂ H3 :
az + b
with z ∈ ∂ H3 , a, b, c, d ∈ C, ad − bc ̸= 0.
cz + d
There are the following types of them:

– Parabolic: 1 fixed point on ∂ H3 , the isometry is conjugate to z 7→ z + a.

– Non-parabolic: 2 fixed points on ∂ H3 , isometry is conjugate to z 7→ az .
- elliptic, |a| = 1, rotation about a vertical line.
- hyperbolic, a ∈ R, (Euclidean) dilation.
- loxodromic, (otherwise), “spiral trajectory” =composition of rotation
and dilation.

141
 Figure 129: Isometries of H3 : parabolic, elliptic, hyperbolic and loxodromic.

III. Some polytopes in H3

- Ideal tetrahedron. It is not unique up to an isometry! See Fig. 130, left.

Exercise: There are 3 pairs of dihedral angles α, β , γ satisfying α + β + γ = π ).

- Regular right-angled dodecahedron.

Indeed a regular dodecahedron of a very small size is almost Euclidean, so has
obtuse angles. A very large dodecahedron (i.e. an ideal regular dodecahedron
with all vertices on the absolute) has angles π /3. So, somewhere in between there
is a right-angled dodecahedron.

- Right-angled ideal octahedron.

Placing a vertex of an ideal octahedron to the infinity of the upper half-space
model, one can see that it can only have equal angles when all angles are right,
see Fig. 130, right.

Figure 130: Hyperbolic polytopes: a tetrahedron and a right-angled ideal octahedron.

IV. Geometric structures on 3-manifolds

- We can glue 3-manifolds from polytopes...

- But we need to check that we obtain complete neighbourhoods around around

edges and vertices.

- There is also a version of Poincaré theorem (which allows to construct discrete

group actions in H3 given a suitable side pairing of a hyperbolic polyhedron).

142
V. Geometrisation conjecture

• William Thurston (1982): all topological 3-manifolds are geometric manifolds,

i.e. every oriented compact 3-manifold without boundary can be cut into pieces
having one of the following 8 geometries:
S 3 , E3 , H3 , S 2 × R, H2 × R, N il, S ol and universal cover of S L(2, R).
(where N il, S ol and S L(2, R) are some exotic structures).

• (1982) William Thurston, (Fields medal, 1982)

proved geometrisation conjecture for some manifolds called “Haken manifolds”.
In particular, all closed atoroidal Haken manifolds are hyperbolic (this statement
is also called a hyperbolisation conjecture).

• (2003) Grigori Perelman, (Fields medal, 2006): general proof of the geometrisa-
tion conjecture.
This also proves Poincaré conjecture:
Every simply-connected closed 3-manifold is a 3-sphere. (Clay Millennium Prize).

Remark. In a “Not knot” 16-minute film (produced in 1991 by mathematicians at

the Geometry Center at the University of Minnesota, directed by Charlie Gunn and
Delle Maxwell) one can experience, how the hyperbolic 3-dimensional space feels from
inside and how 3-dimensional hyperbolic manifolds arise as knot compliments.

143
9.6 References
Section 9 touches several vast fields in mathematics. The aim of this list of references
is to give some possibility to start in each of these directions.

- Concerning lambda-lengths and hyperbolic version of Ptolemy theorem see

- Boris Springborn, Ideal hyperbolic polyhedra and discrete uniformization,

Discrete Comput. Geom., 64(1):63–108, 2020.
[Link]
- A similar idea based on horocycles (together with interplay between Eu-
clidean geometry and the two disc models of hyperbolic geometry) is beau-
tifully used in the following short paper:
K. Drach, R. E. Schwartz, A Hyperbolic View of the Seven Circles Theorem.

- Concerning Aleksandrov-Toponogov comparison theorems, see

- Wolfgang Meyer, Toponogov’s Theorem and Applications.

- Mikhael Gromov, Sign and Geometric meaning of curvature.
- Jeff Cheeger, David G. Ebin, Comparison theorems in Riemannian geome-
try, North-Holland Mathematical Library, vol. 9, North-Holland, Amster-
dam; American Elsevier, New York, 1975.

- To find more about fundamental domains of group actions, Poincaré theorem,

and hyperbolic structure on surfaces, start with
Caroline Series, Hyperbolic geometry.

- William Thurston, The geometry and topology of three-manifolds, Princeton Uni-

versity Mathematics Department (1979), lecture notes. here you will find separate
pdf files of chapters.

- Roice Nelson, Henry Segerman Visualising Hyperbolic Honeycombs. Describes

the way to visualise 3-dimensional hyperbolic tilings. Includes many beautiful
pictures (of tilings, indexed by their Schläfli symbols).

- Concerning Thurston’s hyperbolization theorem, see the following books:

– L Mark Lackenby, Hyperbolic Manifolds. Lecture notes (2000). Continues

here.
- Michael Kapovich, Hyperbolic Manifolds and Discrete Groups. Lectures on
Thurston’s hyperbolization.
- Danny Calegary, Chapter 2: Hyperbolic Geometry, of a forthcoming book
on 3-manifolds.

- And some general phylosophy:

- William Thurston, On Proof and Progress in Mathematics, Bulletin of the

American Math. Soc., Vol. 30, Number 2, April 1994, pp. 161–177.

144
- Videos, webpages, etc...

- Isometry Classes of hyperbolic 3-space, webpage by Roice Nelson, includes

animations of elliptic, hyperbolic, loxodromic and parabolic isometries.
- And other hyperbolic things by Roice Nelson (including hyperbolic Rubik’s
cube, honeycombs, etc...)
- Video comparing spherical, Euclidean and hyperbolic geometry (I don’t
know the author). Also, contains a link to videogames in hyperbolic plane.
- 3-dimensional space webpage with describtion, illustration and animation of
Thurston’s eight geometries, project by Rémi Coulon, Sabetta Matsumoto,
Henry Segerman, Steve Trettel.
- “Not knot”, a short film produced by the Geometry Centre in U. of Min-
nesota, popularising Thurston’s Geometrization Conjecture. (Original tapes
are difficult to find now).

- Further reading on the topics touched in this section (several books and papers
not necessarily accessible online...)

- A. F. Beardon, The Geometry of Discrete Groups, Graduate Texts in Math-

ematics, 91. Springer Verlag (1983).
- F. Bonahon, Low-Dimensional Geometry: From Euclidean Surfaces to Hy-
perbolic Knots. Student Mathematical Library, Volume: 49 (2009).
- [Link], Geometric structures on Manifolds, The Mathematical In-
telligencer 14, Number1, (1982), p. 5–16.
- [Link], Fuchsian groups, University of Chicago Press (1992).
- W. P. Thurston, edited by Silvio Levy, Three-dimensional geometry and
topology, Princeton Univ. Press (1997).

145
10 Bibliography and links

References
[1] J.W. Anderson, Hyperbolic Geometry, Springer (1999).

[2] A. Barvinok, Combinatorics of Polytopes. Lecture notes, University of Michigan

(2010).

[3] A. F. Beardon, The Geometry of Discrete Groups, Graduate Texts in Mathe-

matics,91. Springer Verlag (1983).

[4] F. Bonahon, Low-Dimensional Geometry: From Euclidean Surfaces to Hyper-

bolic Knots. Student Mathematical Library, Volume: 49 (2009).

[5] D. Calegary, Chapter 2: Hyperbolic Geometry, of a forthcoming book on 3-

manifolds. University of Chicago (2015).

[6] J. W. Cannon, W. J. Floyd, R. Kenyon, W. R. Parry, Hyperbolic Geometry.

Flavors of Geometry, MSRI Publications, Volume 31 (1997).

[7] M. do Carmo, Differential geometry of curves and surfaces, Imprint Englewood

Cliffs: Prentice-Hall (1976).

[8] T. K. Carne, Geometry and groups. Lecture notes, Cambridge University.

[9] J. Cheeger, D. G. Ebin, Comparison theorems in Riemannian geometry, North-

Holland Mathematical Library, vol. 9, North-Holland, Amsterdam; American
Elsevier, New York (1975).

[10] H. S. M. Coxeter, Introduction to Geometry, Wiley (1963).

[11] T. Davis, Inversion in a circle.

[12] K. Drach, R. E. Schwartz, A Hyperbolic View of the Seven Circles Theorem.

[13] [Link], Geometric structures on Manifolds, The Mathematical Intelli-

gencer 14, Number 1 (1982), p 5–16.

[14] A. D. Gardiner, C. J. Bradley, Plane Euclidean Geometry, UKMT, Leeds (2012).

[15] M. J. Greenberg, Euclidean and Non-Euclidean Geometries, San Francisco:

W. H. Freeman (2008).

[16] M. Gromov, Sign and Geometric meaning of curvature, IHES (1990).

[17] N. Hitchin, Projective Geometry, Lecture notes. Chapters 1, 2, 3, 4. University

of Oxford.

[18] G. Jones, Algebra and Geometry, Lecture notes, University of Southampton.

[19] M. Kapovich, Hyperbolic Manifolds and Discrete Groups. Lectures on

Thurston’s hyperbolization. University of Utah (1994).

[20] S. Katok, Fuchsian groups, University of Chicago Press (1992).

146
[21] Kiselev’s Geometry, Book I. Planimetry and Book II. Stereometry. (Adopted
from Russian by Alexander Givental). Publisher’s page of the book.

[22] Mark Lackenby, Hyperbolic Manifolds. Lecture notes (2000). Continues here.

[23] W. Meyer, Toponogov’s Theorem and Applications. Lecture Notes, College on

Differential Geometry, Trieste (1989).

[24] J. R. Parker, Hyperbolic Spaces. The Jyväskylä Notes, University of Durham

(2007).

[25] A. Petrunin, Euclidean plane and its relatives. A minimalist introduction. Lec-
ture notes, Pennsylvania State University (2020).

[26] N. Peyerimhoff, Geometry III/IV, Lecture notes (Section 1).

[27] V. V. Prasolov, Non-Euclidean Geometry. Lecture notes, Math in Moscow,

Publishing Company of Moscow Centre for Continuous Mathematical Educa-
tion, Moscow (2006).

[28] V. V. Prasolov, V. M. Tikhomirov, Geometry. American Mathematical Society,

Providence, Rhode Island, 2001.

[29] E. Rees, Notes on Geometry, Universitext, Springer (2004).

[30] R. Schwartz, S. Tabachnikov, Elementary Surprises in Projective Geometry,

arXiv:0910.1952v2.

[31] C. Series, Hyperbolic geometry, Lecture notes, University of Warwick, 2008.

[32] A. B. Sossinsky, Geometries, Providence, RI : American Mathematical Soc.

(2012).

[33] B. Springborn, Ideal hyperbolic polyhedra and discrete uniformization, Discrete

Comput. Geom., 64(1):63–108, (2020). [Link]

[34] W. P. Thurston, The geometry and topology of three-manifolds, Princeton Uni-

versity Mathematics Department (1979), lecture notes.

[35] W. P. Thurston, edited by Silvio Levy, Three-dimensional geometry and topol-

ogy, Princeton Univ. Press (1997).

[36] W. P. Thurston, On Proof and Progress in Mathematics, Bulletin of the Amer-

ican Math. Soc., Vol. 30, Number 2, April 1994, pp. 161–177.

[37] E. B. Vinberg (Ed.), Geometry II , Encyclopaedia of Mathematical Sciences,

Vol. 29, Springer-Verlag.

[38] O. Viro, Defining relations for reflections. I, arXiv:1405.1460v1

[39] O. Viro, Biflippers and head to tail composition rules arXiv:1412.2397v2.

[40] C. P. Walkden, Hyperbolic Geometry. Lecture notes, The University of Manch-

ester (2019).

[41] J. Weeks, Non-Euclidean billiards in VR, Bridges 2020 Conference Proceedings.

147
Webpages, videos, software, artworks, etc. :
- Webpages, websites, portals:

- Cut-the-knot portal by Alexander Bogomolny.

- Euclid’s "Elements", website by David E. Joyce.
- Drawing a Circle with a Framing Square and 2 Nails.
- Circle inversion, webpage by Malin Christersson (Illustrated introduction
with proofs).
- Geometry, webpage by Vladimir V. Kisil (Supporting materials to a course
in Euclidean, projective and inversive geometries).
- Hyperbolic tessellations, including some printable hyperbolic tilings, web-
page by David E. Joyce.
- Isometry Classes of hyperbolic 3-space, webpage by Roice Nelson, includes
animations of elliptic, hyperbolic, loxodromic and parabolic isometries.
- More hyperbolic things by Roice Nelson (including hyperbolic Rubik’s cube,
honeycombs, etc...)
- Illustrating Mathematics by Rémi Coulon (including Thurston’s eight ge-
ometries, fractals, Penrose tilings, pantograph, etc...)
- Hyperbolic tessellations by Don Hantch.
- Tiling page and Hyperbolic geometry page on The Geometry Junkyard
by David Eppstein.

- Videos, animations, etc:

- Why slicing cone gives an ellipse - video on Grant Sanderson’s YouTube

channel 3Blue1Brown.
- Loxodromic transformation in the page by Paul Nylander.
- Dynamics of Möbius transformations is illustrated in the 2-minute
video by D. Arnold and J. Rogness.
- Animation demonstrating Inversion in circles, by M. Christersson.
- 1-minute video illustrating stereographic projection by Henry Segerman.
- Playing Sports in Hyperbolic Space - Dick Canary in a Numberphile video
(by Brady Haran).
- Illuminating hyperbolic geometry, Short video (4:25 min) by Henry Segerman
and Saul Schleimer on projecting the hemisphere model to Klein disc, Poincare
disc and upper half-plane.
- Video comparing spherical, Euclidean and hyperbolic geometry (I don’t
know the author). Also, contains a link to videogames in hyperbolic plane.

- Films:

- “Not knot”, a short film produced by the Geometry Centre in U. of Min-

nesota, popularising Thurston’s Geometrization Conjecture.
- Dimensions. 9 films for wide audience, 13 min each, produced by: Jos Leys,
Étienne Ghys, Aurélien Alvarez.

148
- Software:

- Applet for creating hyperbolic drawings in Poincaré disc.

- Applet to make hyperbolic tessellations of images, by Malin Christersson.

- Artwork:

- Tilings by (and after) Escher. Webpage on Mathematical Imagery by Jos

Leys.
- Dancing circles. Hyperbolic tilings dancing to music from Tchaikovsky’s
ballet “The Nutcracker”. On Mathematics Imagery by Jos Leys.
- Polyhedra and Art webpage by George W. Hart.
- Polyhedral sculptures by George W. Hart.
- Hyperbolic Geometry Artworks by Paul Nylander.
- Webpage on hyperbolic geometry by the Institute for Figuring (with hyper-
bolic soccer ball and crocheted hyperbolic planes).
- How to create repeating hyperbolic patterns, by Douglas Dunham (based
on Escher’s patterns). See also here.
- Stereographic projection and models for hyperbolic geometry, 3D toys to il-
lustrate by Henry Segerman.
- D. Taimina, “Crocheting Adventures with Hyperbolic Planes”. Published by
A K Peters (2009).
- R. Nelson, H. Segerman, Visualising Hyperbolic Honeycombs, arXiv:1511.02851.

- Games:

- Games on hyperbolic field on Zeno Rogue webpage.

- Hyperbolica, Non-Euclidean adventure games.

149