MA243: Geometry

1. Lecture 1: Introduction, motivation, overview. Definition of a metric.

1.1. Practical matters.
• Lecturer: Helena Verrill, office: B1.35, email: [email protected]
• Lectures:
– Monday 11-12: week 1 and 3, will be WLT, weeks 2 and 4-10: L3
– Tuesday 11-12: L4
– Friday 11-12: WOODS-SCAWEN
• Lecture Notes: A pdf of the lecture notes will be posted at the beginning of each week. The
lecture capture will be available after the lectures.
• References: This course is based on the book:
Reid and Szendrői (chapters 1-6) Geometry and Topology.
For more details for the parts on hyperbolic geometry we also use:
Ratcliffe (parts of chapters 1-3) Hyperbolic Geometry.
• TAs: Rob Phillips and James Rawson
• Support Classes: Tuesday 4pm (Rob) and Thursday 2pm (James). Starting week 2, both in
MS.04.
• Assignmnents: There are 4 assignments, contributing 15% to your final mark. The best 3
assignments count.
• Quizzes: There will be optional quizzes posted on moodle for you to check understanding. These
will not contribute to your grade.
• Forum: Please post your questions to the moodle forum and join in dicusssions online.
• Goals: This course aims to introduce you to
– Euclidean space,
– spherical geometry,
– hyperbolic geometry,
∗ For these geometries, defining the
· respective metrics (ie. how distance is measured),
· proving they satisfy the triangle inequality
· describing the groups of isometries
· constrasting parallel lines and angles of triangles for these geometries.
– projective geometry.
• Prerequisites:
– Linear algebra: vector space, basis, linear map, matrix, rank-nullity theorem
– Group theory: group, identity, associtive, inverses
– Relations: transitive, reflective, equivalence
– Hyperbolic function: sinh, cosh
We will review these topics, but it will be useful if you can refresh your memories in advance.
• Exam: The exam counts for 85% of the mark. This will probably have a similar content to the
exams for the last three years, as well as to the homework assignments.
• Vevox: We will have quizzes on vevox, the vevox ID will always be: 151-597-945

Let me know if you find any typos in these note, or have any questions!

1
MA243 December 8, 2023 2

1.2. Course contents overview:

Cartesian geometry −  − Spherical geometry −  − Hyperbolic geometry −  − Projective geometry

Comparison of 2-dimensional metric geometries:

Euclidean Spherical Hyperbolic
Parallel lines through a point parallel to a line unique don’t exist not unique
Triangle angle sum π >π <π

Klein’s Erlangen program: describes geometry in terms of groups of transformations.

Geometry “=” Group theory
Geometry Group
Euclidean Rn O(n) n Rn
Spherical S n O(n + 1)
Hyperbolic Hn O+ (1, n)
Projective Pn P GL(n + 1)

1.3. Metric Spaces.

Definition 1. A metric d on a set X is a function d : X × X → [0, ∞) such that ∀x, y ∈ X, it is:
(1) Non-degenerate:
d(x, y) = 0 ⇐⇒ x = y
(2) Symmetric:
d(x, y) = d(y, x)
(3) Satisfies the triangle inequality:
d(x, y) ≤ d(x, z) + d(z, y)
The pair (X, d) is called a metric space. The function d may also be called a distance function.
1.4. The Euclidean metric. Definitions and notation. For x, y ∈ Rm , we write:
• x = (x1 , x2 , . . . , xm ),
• y = (y1 , y2 , . . . , ym ),
• Inner product: hx, yi = m i=1 xi yi = x1 y1 + x2 y2 + · · · + xm ym
P

This is the same as the p dot product, x · y.

• Euclidean Norm: ||x|| = hx, xi (Note, |λ| is the abolute value of a real number.)
• Euclidean metric: d(x, y) := ||x − y||
Note, the Pythagorean theorem is true by definition!
To prove the Euclidean metric is actually a metric we need some results about the inner product.
Lemma 1 (Bilinearity of the inner product). The inner product on Rm is bilinear and symmetric, that is
for a, b, c ∈ Rm and λ ∈ R, we have
ha + b, ci = ha, ci + hb, ci and hλa, bi = λ ha, bi and ha, bi = hb, ai
in particular ha−b, a−bi = ha, ai − 2 ha, bi + hb, bi.
Proof. Follows from the rules of vector and matrix multiplication. Details left as an exercise. 
MA243 December 8, 2023 3

2. Lecture 2: The Euclidean metric. The Cauchy-Schwartz inequality.

Lemma 2 (Cauchy-Schwartz inequality). For all x, y ∈ Rn , hx, yi2 ≤ ||x||2 · ||y||2 Or, equivalently,
| hx, yi | ≤ ||x|| · ||y|| x
When y 6= 0, there is equality if and only if x = λy for some λ ∈ R.

||z||
Proof. If y = 0, we get 0 ≤ 0, which is true. Now assume y 6= 0.

||
||x
Recall from A-level maths, if the angle between x and y is θ, then:
y
hx, yi = ||x|| · ||y|| cos(θ). λy
||y||
The result follows since cos θ ≤ 1.
0
For a proof from first principles: Let z := x − hx,yi
||y||2
y. Using bilinearity, y
x, ||y||
λ =
we have
hx, yi hx, yi hx, yi hx, yi hx, yi
     
2
0 ≤ ||z|| = x − y, x − y = hx, xi − 2 x, y + y, y
||y||2 ||y||2 ||y||2 ||y||2 ||y||2
2 2 2
hx, yi hx, yi hx, yi hx, yi
  
2 2
= ||x|| − 2 hx, yi + hy, yi = ||x|| − 2 + .
||y||2 ||y||2 ||y|| ||y||
 2
Rearranged, this is ||x||2 ≥ hx,yi
||y|| , which gives the required result.
Equality is acheived if and only if ||z|| = 0. Because the norm is non-degenerate ||z|| = 0 if and only if
z = 0, which is the case if and only if
hx, yi
x = λy where λ =
||y||2

To save on notation, we prove that the Euclidean metric is translation invariant.
Lemma 3 (Translation invariance of Euclidean metric). For x, y, z ∈ Rm we have d(x, y) = d(x−z, y−z).
Proof. d(x, y) = ||x − y|| = ||(x − z) − (y − z)|| = d(x − z, y − z). 
Lemma 4 (Triangle inequality for the Euclidean metric). For all x, y, x ∈ Rn we have d(x, y) ≤ d(x, z) +
d(z, y), with equality if and only if either z = x, or y − z is a positive scalar multiple of z − x.
We want to show that d(x, y) ≤ d(x, z)+d(z, y). By translation invariance
Triangle inequality in Rm
(Lemma 3), we can replace x, y, z by 0, y0 = y − x and z0 = z − x z
respectively. (because the distances between these points are the same as
the distances between x, y, z. The only reason to change is to cut down ||y
on notation.) So by definition of the metric, it is sufficient to prove
−
z|
||

|
||y0 || ≤ ||z0 || + ||y0 − z0 ||.
−z

This is equivalent to ||y0 ||2 ≤ (||z0 || + ||y0 − z0 ||)2 . Using the Cauchy
||x

y
Schwartz inequality, we have ||z0 ||·||y0 −z0 || ≥ | hz0 , y0 − z0 i | ≥ hz0 , y0 − z0 i.
So
(||z0 || + ||y0 − z0 ||)2 = ||z0 ||2 + 2||z0 || · ||y0 − z0 || + ||y0 − z0 ||2 y||
||x −
x
(1) ≥ ||z0 ||2 + 2| z0 , y0 − z0 | + ||y0 − z0 ||2
||x − y|| ≤ ||x − z|| + ||z − y||
(2) ≥ ||z0 ||2 + 2 z0 , y0 − z0 + ||y0 − z0 ||2
= z0 , z0 + 2 z0 , y0 − z0 + y0 − z0 , y0 − z0 = z0 − (y0 − z0 ), z0 − (y0 − z0 ) = y0 , y0 = ||y0 ||2 ,
as required. Looking at the lines (1) and (2), we see equality is obtained if and only if both: z0 = 0 or
y 0 − z0 is a multiple of z0 , and hz0 , y0 − z0 i is positive. In terms of the original x, y, z, this is if and only
if z − x = 0 or y − z is a positive multiple of z − x. 
MA243 December 8, 2023 4

Question: (Thinking about transformations) How are these triangles related?

(See www.mathamaze.co.uk/MA243/MA243week1.html, or video capture of discussion)

3. Lecture 3: Colinearity. Euclidean space. Isometries. Euclidean Isometries.

Theorem 1 (The Euclidean metric is a metric). d(x, y) = ||x − y|| on Rm is a metric on Rm .
Proof. We need to show this satisfies the three properties of metrics.
Non-degenerate: Since xi and yi are real, (xi − yi )2 = 0 ⇐⇒ xi = yi .
Thus d(x, y) = 0 ⇐⇒ xi = yi for i = 1, . . . m ⇐⇒ x = y.
Symmetric: Since (xi − yi )2 = (yi − xi )2 , we have d(x, y) = d(y, x).
Triangle inequality: was proved in Lemma 4. 
3.1. Isometries and Euclidean space.
Definition 2. If (X1 , d1 ) and (X2 , d2 ) are metric spaces, then a distance preserving map between (X1 , d1 )
and (X2 , d2 ) is a map
f : X1 → X2 f(b)
such that for any points P, Q in X1 , we have
f(c)
d2 (f (P ), f (Q)) = d1 (P, Q). T
y
An isometry is a bijective distance preserving map. m etr
iso f(a)
If f is an isometry, then (X1 , d1 ) and (X2 , d2 )
are said to be isometric.
b The triangle and its image under
Quiz: Does a distance preserving map have to be injective? a c an isometry are congruent
n
Definition 3. A Euclidean space is a metric space which is isometric to R , with the Euclidean metric,
for some integer n. We use the notation En to denote the metric space Rn together with the Euclidean
metric. If not specified otherwise, Rn is used to mean En .
WARNING: We are often confusing or conflating the concepts of vectors and points. We can add
vectors in Rn , but we can’t add points in En , unless we idenifity En with Rn , but there are infinitely
many ways to do this. Identifying a Eucidean space with Rn means choosing an origin and a basis.
Example 1.
y A Euclidean space
x+y =1
The line
L = {(x, y) ∈ R2 : x + y = 1}, 1
with metric given by the usual metric on R2 is a Euclidean space. We can
choose our identification with R by chosing 0 and a direction.
0 1 x

Note that adding points on L does not make sense: If the points are added together using addition in R2 ,
then the result is not on L. We can make addition on L make sense, but we would have to choose a zero.
Definition 4. A motion, also called a Euclidean isometry, is an isometry of a Euclidean space.
Note: A motion does not move.
MA243 December 8, 2023 5

3.2. Lines and Collinearity.

Definition 5 (Definition of a line). Let u, v ∈ Rn , v 6= 0. The line through u in direction v is
L := {u + λv|λ ∈ R} ⊂ Rn L

u v

A line in Rm is any subset of Rm which has this form. 0

Definition 6 (Collinearity). We say that x, y, z are collinear if there is a line L with x, y, z ∈ L.
Lemma 5 (Formula for collinearity). If x, y, z are distinct
points in Rn , then they are colinear if and only if for some
λ ∈ R we have Quiz: In this picture, λ = λ2 ∈ (0, 1),
and z is between x and y.
z = x + λ(y − x) Where is z on L if λ > 1 or < 0?
= (1 − λ)x + λy
= λ1 x + λ2 y
d1 = d2 =
where λ1 +λ2 = 1 and λ2 = λ. Further, in the case of collinear-
d(x, z) d(z, y)
ity |λ1 | : |λ2 | = d2 : d1 , where d1 = d(x, z) and d2 = d(z, y). L
Proof. (probably won’t use green proof. ) x z y
⇒: Suppose x, y, z are collinear, on a line L. By definition
of L, for some vectors u and v, all points on L have the form
u + λv. So for some λx , λy , λz ∈ R, we have
x = u + λx v, y = u + λy v, z = u + λz v
Solving for u and v, and setting λ = (1−λx )/(λy −λx ) gives
the result: 0
" # " #
λy x − λx y x−y λy − 1 1 − λx
z= + = x+ y
λy − λx λx − λy λy − λx λy − λx z = λ1 x + λ2 y
Now we have d(x, z) = ||x − z|| = ||x − λ1 x − λ2 y|| = λ1 + λ2 = 1
|λ2 |||x − y||, and similarly d(x, z) = |λ1 |||x − y||, which gives |λ1 | : |λ2 | = d2 : d1
|λ1 | : |λ2 | = d2 : d1 .
⇐: If z = x + λ(y − x), then z lies on the line through x in the direction of y − x. This line also contains
y, by taking λ = 1.

Alternative proof of ⇒ (same idea, different notation):
Proof 2. (of Lemma 5)
⇒ x, y, z being collinear means they lie on a line u + Rv, with v 6= 0. So, for some λx , λy , λz ∈ R,
        
u v 1 u v 1 u v 1
x= , y= , z=
λx λy λz
So,
    
x y z u v 1 1 1
(3) =
λx λy λz
we’re assuming λx 6= λy (since x 6= y), so the matrix
 
1 1 1
M :=
λx λy λz
has rank 2, so has kernel of rank 1, (rank nullity theorem) spanned by some vector (u1 , u2 , u3 )T ∈ R3 .
We can’t have u3 = 0, since this would imply that the first two columns are linearly dependent, implying
λx = λy , implying x = y, which we assume is not the case.
MA243 December 8, 2023 6

So we can scale by 1/u3 , and so replace (u1 , u2 , u3 ) by (−λ1 , −λ2 , 1) for some λ1 , λ2 ∈ R.
Now multiplying both sides of Equation 3 by (−λ1 , −λ2 , 1)T , we get:
−λ1
 
 
(4) x y z −λ2  = 0
1
Rearranged, z = λ1 x + λ2 y. Since (−λ1 , −λ2 , 1) is in the kernel of M , we must have λ1 + λ2 = 1. This
implies that we also have
−λ1
 
 
(5) z z z −λ2  = 0
1
Subtracting Equation 5 from Equation 4 gives
−λ1
 
 
x−z y−z 0 −λ2  = 0
1
I.e., λ1 (x − z) = −λ2 (y − z) So
|λ1 |||x − z|| = |λ2 |||y − z||.

Picture on desmos (https://www.desmos.com/calculator/lcgtswvxgm):
MA243 December 8, 2023 7

4. Lecture 4: Collinearity in terms of the metric

For this class, before the new material:
(1) Recap on triangle inequality:
For all x, y, x ∈ Rn we have d(x, y) ≤ d(x, z) + d(z, y), with equality if and only if either
z = x, or y − z is a positive scalar multiple of z − x.
(2) Finish proof of Lemma 5:
If x, y, z are distinct points in Rn , then they are colinear if and only if for some λ ∈ R we
have
z = x + λ(y − x) = (1 − λ)x + λy = λ1 x + λ2 y
where λ1 + λ2 = 1 and λ2 = λ. In the case of collinearity |λ1 | : |λ2 | = d(x, z) : d(z, y).
We need to finish the part about |λ1 | : |λ2 | = d2 : d1 , and ⇐. See notes for Lecture 3.

4.1. Note on collinearity and λ1 , λ2 . In Lemma 5, we have collinearity of x, y, z and, providing

x, y are distinct, we can write
z = λ1 x + λ2 y
with λ1 + λ2 = 1. The following picture shows possible signs of λ1 , λ2 :
z1 x z2 y z3  
λ1
(+, −) (+, 0) (+, +) (0, +) (−, +) z = λ1 x + λ2 y ⇐⇒ (x, y, z) λ2  = 0

−1
We can always scale the vector (λ1 , λ2 , −1) so that one term is −1 and the other two are non-negative;
the term which is −1 the corresponds to which of x, y, z is “between” the other two. E.g., if z = 3x − 2y,
then we can scale (3, −2, −1) to (−1, 2/3, 1/3), and write as x = 2/3y + 1/3z.
From now on, we identify Em with Rm , i.e., we choose and fix a zero. This is always possible, so the
results apply to any Euclidean space. We now give two lemmas which we use to express collinearity in
terms of the metric.
Now we can characterize collinearity in terms of the Euclidean metric:

Proposition 1. x, y, z are collinear if and only if If x, y, z are collinear, with z y

we have equality in the triangle inequality “in the middle”

z
(6) d(x, y) = d(x, z) + d(z, y) then d(x, y) = d(x, z) + d(z, y)
x

up to permuting x, y, z.

Proof. Easy case: If y = z, the last term is zero and the others are equal, so the result is true. Since
we are proving the result up to permutaion of the points, we now assume that all points are distinct.
⇐: In the proof of the triangle inequality, we saw that there is equality in (6) if and only if z = x (which
we are now assuming is not the case) or y − z is a positive multiple of z − x. I.e., for some λ ∈ R≥0 ,
y − z = λ(z − x). Rearranged,
(7) y = z − λ(x − z)
which by Lemma 5 implies x, y, z are collinear. (λ happens to be positive, so we don’t yet have ⇒.)
⇒: If x, y, z are collinear, and distinct, then by Lemma 5, and following discussion, we can permute
(rename) x, y, z so that z = λ1 x + λ2 y with λ1 , λ2 ≥ 0, λ1 + λ2 = 1. Then
1 λ1 λ1
y= z − x = z + (z − x)
λ2 λ2 λ2
So y − z is a positive multiple of z − x, and so by Lemma 4 (triangle inequality) Equation (6) holds. 
MA243 December 8, 2023 8

4.2. Isometries and Collinearity. We’re going to show that isometries map lines to lines, and use
this to write isometries in terms of matrices. Previously you have seen that vectors and matrices can be
used to define “motions”, but now we show that all isometries have this form.
Isometry in latex tikzpicture: T (c)
[shift = (7,-1), rotate=45]
N (b)
b
T (b)

c N (a)
a Apply an isometry N (c)
Not the result of an isometry
T (a)

Proposition 2. If T is an isometry of Em , then T maps lines to lines, i.e., x, y, z are collinear then
T (x), T (y), T (z) are also collinear. Moreover, if z = (1 − λ)x + λy, then T (z) = (1 − λ)T (x) + λT (y).
Proof. By Propsition 1, and since T is distance preserving,
x, y, z are collinear ⇒ d(x, y) = d(x, z) + d(z, y) after possible permutation of x, y, z
⇒ d(T (x), T (y)) = d(T (x), T (z)) + d(T (z), T (y))
⇒ T (x), T (y), T (z) are collinear by Proposition 1
If λ ∈ [0, 1], the final statement follows from Lemma 5, which implies (1 − λ) : λ = d(z, y) : d(x, z) =
d(T (z), T (y)) : d(T (x), T (z)). Cases λ 6∈ [0, 1] are obtained by reordering x, y, z appropriately. 
Definition 7. A linear map is a function L : V → W between two vector spaces over k, such that
L(λv + µw) = λL(v) + µL(w)
for all v, w ∈ V , and for all λ, µ ∈ k. For us, the field k is almost always R; we might extend to C
sometimes.
Definition 8 (Affine maps). A map T : Rn → Rk is affine if it is of the form T (x) = L(x) + b for all
x ∈ Rn for some linear map L : Rn → Rk linear and b ∈ Rk . Note that b = T (0).
From linear algebra, you know that a linear map L : Rn → Rk is described by a k by n matrix A.
Notation: For a k × n matrix A and b ∈ Rk , write T(A,b) for the map
T(A,b) (x) = Ax + b
for x ∈ Rn .
Theorem 2. The set of all invertible affine maps from Rn to Rn ,
n o
Aff(n) := T(A,b) : A is an n × n invertible matrix, b ∈ Rn
is a group. The composition law is
T(A1 ,b1 ) ◦ T(A2 ,b2 ) = T(A1 A2 ,A1 b2 +b1 )
Proof. Exercise. 
This group law is denoted
Aff(n) ∼= GL(n) n Rn .
The notation n is just a way of saying that this is a special kind of product, called the “semi direct
product”, not the direct product of two groups.
Exercise: What is the inverse of TA,b ?
Note that the GL(n) is a subgroup of Aff(n) – this is the subgroup of affine maps that fix the origin.
Rn is the subgroup of Aff(n) consisting of the translations.
Exercise: GL(n) and Stab(v) (maps fixing v) are normal subgroups of Aff(n). Show by example that the
subgroup of translations is not normal.
We are going to show that Euclidean isometries are affine. But not just any old affine map.
MA243 December 8, 2023 9

5. Lecture 5: Isometries of Rm in terms of linear maps

First a more general result about affine maps between real vector spaces which may have different di-
mension. This result shows that affine maps send lines to lines.
Proposition 3. Given a map T : Rn → Rk , the following are equivalent
(1) T is affine.
(2) for all λ, µ ∈ R and x, y ∈ Rn ,
(†) T (λx + µy) − T (0) = λ(T (x) − T (0)) + µ(T (y) − T (0))
I.e., x → T (x) − b is a linear map, and T is linear if T (0) = 0.
(3) For all λ ∈ R,
(∗) T (λx + (1 − λ)y) = λT (x) + (1 − λ)T (y)
Proof. (1) ⇐⇒ (2): Define a map L by L(x) := T (x) − T (0).
T is affine ⇐⇒ L is linear, by definition of affine
⇐⇒ L(λx + µy) = λL(x) + µL(y), by defintion of linear
⇐⇒ T (λx + µy) − T (0) = λ(T (x) − T (0)) + µ(T (y) − T (0)) since L(x) = T (x) − T (0).
(2)⇒ (3): This is obtained by setting µ = 1 − λ in (Equation †); all terms involving T (0) cancel.
(3)⇒ (2): By Lemma 5, (3) is telling us that T maps lines to lines, since z = λx + (1 − λ)y is on the line
through x and y, and T (z) = λT (x) + (1 − λ)T (y) means that T (z) is on the line through T (x) and T (y).
This is stronger than just colinearity: it says that the relation between the points on a line is preserved.
Now we need to relate linearity to conservation of collinearity. We will use the relationship between the
red points on the blue lines in the figure.
Since it may be unclear whether av means the scalar a times the vector v, or the vector av, I have put
a coloured background under the vectors in the expressions of collinearity.
2λx, λx + µy and 2µy are collinear, since
 
1 1
λx + µy = 1 − 2 2λx + 2 2µy

So by Equation ∗, with λ = 12 ,
2λx (F)
1 1
T ( λx + µy ) = T ( 2λx ) + T ( 2µy ).
2 2
λx λx + µy 0, x and 2λx are collinear, with
x 2λx = 2λ x + (1 − 2λ) 0

0 So T ( 2λx ) = 2λT ( x ) + (1 − 2λ)T ( 0 ), so

y µy 2µy  
1 1
2 T (2λx) = λT (x) + 2 − λ T (0)
 
Similarly, 21 T (2µy) = µT (y) + 12 − µ T (0). Substituting in Equation F and rearranging gives the re-
quired expression (†) of linearity of T (x) − T (0). 
Corollary 1. Every Euclidean isometry T : Rm → Rn is affine, and n = m i.e.,T has the form
T (x) = L(x) + b for some linear map L : Rn → Rn and vector b ∈ Rn .
Proof. By Proposition 2 T preserves collinearity, so (3) of Proposition 3 is satisfied, so since (1) and
(3) are equivalent, T is affine. Since T is a bijection, L must also be a bijection. From linear algebra,
e.g., use the rank-nullity theorem, this implies n = m. 
Next we work out what kinds of linear map L can be. It turns out that being an isometry implies extra
conditions on L, so we want to understand what kind of L is possible.
MA243 December 8, 2023 10

5.1. Linear isometries have orthogonal matrices. We now find a characterisation of linear
isometries L in terms of the matrix A corresponding to L. We need a result, which expresses the inner
product in terms of the norm:

Lemma 6 (polarization identity for real inner prod-

ucts). For x, y in a vector space over R (or over a x+y
field with characteristic not 2), x y|
|
+

||x
||x

−
y|
1 1

|
hx, yi = ||x + y||2 − ||x − y||2
4 4

Proof. Exercise in manipulating the inner y

product, and recalling the definition of the 0
norm. 

Proposition 4. let L : Rn → Rn be a linear map with matrix A with respect to the standard basis. Then
the following are equivalent:
(1) L is an isometry
(2) ||L(x)|| = ||x|| for all x ∈ Rn , i.e., L is norm preserving
(3) hL(x), L(y)i = hx, yi ∀x, y ∈ Rn i.e., L preserves the inner product
(4) The matrix A is orthogonal, i.e., AT A = I
Proof. (1)⇒(2): Using definition of the metric, defining property of isometries, and linearity of L:
||x|| = d(x, 0) = d(L(x), L(0)) = d(L(x), 0) = ||L(x)||.
(2)⇒(3): This follows from the polarisation inequality in Lemma 6.
(3)⇒(4): Now compute the elements of AT A. By the formula for matrix multiplication
n
X n
X n
X
T T
(A A)ij = (A )ik Akj = Aki Akj = L(ei )k L(ej )k = hL(ei ), L(ej )i
k=1 k=1 k=1

So the entries of AT A are the inner products of the columns of A. Since the ei are the standard basis,
and by hypothesis (3), L preserves the inner product, we have
(
T 1 if i = j
(A A)ij = hL(ei ), L(ej )i = hei , ej i =
0 6 j
if i =
These are the elements of the identity matrix, so A is orthogonal.
(4)⇒ (1): Assuming (4), A is invertible with inverse AT , so L is a bijection. So we just need to check L
is distance preserving. First check that L is norm preserving:
||L(x)||2 = hL(x), L(x)i
= (L(x))T L(x) using matrix notation for dot product
= (Ax)T (Ax) writing L(x) = Ax
= xT AT Ax property of transpose
= xT x since A is orthogonal
= hx, xi , by definition of inner product
= ||x||2 , by definition of norm.

Now distance preservation follows from the linearity of L:

d(L(x), L(y)) = ||L(x) − L(y)|| = ||L(x − y)|| = ||x − y|| = d(x, y).
Since A has inverse AT , L is invertible, so L is an isometry. 
MA243 December 8, 2023 11

6. Lecture 6: Isometry group of Euclidean space

Now we can characterize isometries in terms of their matrices. Proposition 4 deals with isometries which
are linear maps. Corollary 1 showed how isometries are related to linear maps. So we have:
Corollary 2. A Euclidean isometry (motion) T : Rn → Rn is of the form T (x) = Ax + b for all x ∈ Rn
for some unique orthogonal n × n matrix A and some vector b ∈ Rn .
Conversely, if A is an orthogonal matrix, Ax + b is an isometry for all b ∈ Rm .
Proof.
T is a motion ⇐⇒ L(x) := T (x) − T (0) is a linear isometry
(distance is translation invariant by Lemma 3)
⇐⇒ L(x) = Ax for some orthogonal A by Proposition 4
⇐⇒ T (x) = Ax + b for some orthogonal A and b = T (0)
The matrix corresponding to a linear map with respect to a given basis is unique (result from linear
algebra). 
cos θ − sin θ
   
0 1
Example 2. The following matrices are orthogonal: I, , .
1 0 sin θ cos θ

6.1. Isometries of R2 . Recall the matrices for rotations and reflections are defined as follows:
π
2
−θ

π − 2θ
(0, 1) (0, 1)
(− sin θ), cos(θ)) (cos(2θ), sin(2θ))

(cos(θ), sin(θ))
(sin(2θ), − cos(2θ))
θ
θ (1, 0)
θ (1, 0)

Matrix of rotation through Matrix of reflection in line 2θ

θ anti-clockwise: at θ anti-clockwise to x-axis: note: in this picture
π > 2θ > π/2,
so cos(2θ) < 0
cos θ − sin θ
   
cos 2θ sin 2θ
Bθ := ρθ :=
sin θ cos θ sin 2θ − cos 2θ Note that ρα ◦ ρβ = B2(β−α)

Theorem 3. Let T : E2 → E2 be an isometry. Then T (x) = Ax + b where, with respect to the standard
basis, A = Bθ or ρθ and b ∈ R2 , so, T is a rotation or reflection, followed by a translation.
Proof. By Corollary 2, T (x) = Ax + b, with A orthogonal. Write A as a column matrix, A = [v, w].
Then AT A = I implies ||v|| = ||w|| = 1 and hv, wi = 0. Let v = (a, b). Since a2 + b2 = 1, and a, b ∈ R,
for some θ, (a, b) = (cos θ, sin θ). Since hv, wi = 0, and ||w|| = 1, w = ±(−b, a). These give the two
cases. 

We can classify isometries in E2 further.

Theorem 4. Every isometry of E2 has one of the following forms:
• Translation
• Rotation
• Reflection
• Glide reflection: reflection, followed by translation parallel to the line fixed by the reflection.
Proof. Exercise. See [RS] Section 1.14 for more details. 
MA243 December 8, 2023 12

6.2. Frieze groups: examples of isometries of R2 . The motives in the following frieze patterns
are related by translations, rotations, reflections or glide reflections - can you identify which of these
isometries are involved in the following frieze pattern?

6.3. Isometry Groups.

Definition 9. if (X, d) is a metric space, then Isom(X, d) is the group of all isometries from X to X,
with group law being composition of functions.
Exercise: Verify that Isom(X, d) is indeed a group.
6.4. The group of Euclidean isometries. We know from Corollary 2 that isometries of En have
the form
x → Ax + b
where A is an orthogonal matrix and b is a vector.
Definition 10 (The orthogonal group). The group of n × n real orthogonal matrices is defined by
O(n, R) = {A is an n × n real matrix : AT A = In }
We will usually just write O(n), since usually we will only consider matrices with real coefficients.
Lemma 7. O(n, R) is a group.
Proof. Exercise. 
As sets,
Isom(Rn , d) ∼
= O(n, R) × Rn
T ↔ (A, b) = (T − T (0), T (0))
As group,
Isom(Rn , d) ∼
= O(n, R) n Rn
If T (x) = Ax + b and T 0 (x) = A0 x + b0 , with A and A0 ∈ O(n), and b, b0 ∈ Rn , we have
(T ◦ T 0 )x = A(T 0 (x)) + b = AA0 x + Ab0 + b
6.5. Topology of Isom(E2 ) [non-examinable; from [RS chapter 8]]. Now we have seen that
all elements of Isom(E2 ) can be described by four parameters, θ ∈ [0, 2π), a ∈ R, b ∈ R, c ∈ {1, −1}:
cos(θ) −c sin(θ)
   
a
x 7→ x+
sin(θ) c cos(θ) b
Since a rotation of 2π is the same as a rotation through 0, we may think of the rotations as being
parameterised by points on the circle S 1 = {x ∈ R2 |||x|| = 1}. So, the elements of Isom(E2 ) correspond
to points in S 1 ×R2 ×{±1}. You can think of this as two disjoint copies of a kind of an infinite solid torus.
Now a continuous “motion” in R2 will correspond to a continuous path in the component corresponding
to rotations. (called SO(2).) The motion of an object in R2 from time 0 to 1 could be be parameterised
by a map from [0, 1] into S 1 n R2 . Our motions form a three dimensional topological space. For R3 , the
space of motions forms a 6 dimensional space. See [RS chapter 8] for more details.

[0, 1]2
sketch of S 1 × [0, 1]2 , representing motions
S1 mapping 0 to a point in [0, 1]2
MA243 December 8, 2023 13

7. Lecture 7: Complex Linear algbera

The goal of this section is to show that all isometries of Rn can be expressed as a composition of rotations
and reflections on 1 and 2 dimensional mutually orthogonal subspaces of Rn .

7.1. Extending a linear map from Rn to Cn . In order to prove results about orthogonal matrices,
it’s useful to work over C.
If L : Rn → Rm is a linear map, we can extend it to a map from L : Rn → Rm , by mapping u + iv to
L(u) + iL(v), for u, v ∈ Rn . The matrix describing the transformation is the same.

7.2. sesquilinear inner product. For z,w ∈ Cn we define the sesquilinear inner product by
m
X
hz, wi := z i · wi = z T · w
i=1

Where a is the complex conjugate of a. Note that ||v|| is a real number, and Cauchy-Schwartz still holds.
Lemma 8. If L is a linear isometry of Rn , with matrix A then the extension of L to Cn preserves the
sesquilinear inner product.
Proof. Take z, w ∈ Cn . Then by definition,
hL(z), L(w)i = hAz, Awi = (Az)T · (Aw) = (zT AT )(Aw) = zT AT Aw = zIn w = zT w = hz, wi


7.3. Direct sum. If W is a vector space containing subspaces U and V , then we say W is the direct
sum of U and and V , written W = U ⊕ V if U ∩ V = {0} and every element of W can be written uniquely
as a sum of an element of U and an element of V . E.g., R2 = (Re1 ) ⊕ (Re2 ), where e1 , e2 is a basis for
R2 . We can extend this concept, writing W = U1 ⊕ U2 ⊕ · · · ⊕ Un where Ui are subspaces of a vector
space V , Ui ∩ Uj = {0} for i 6= j, and any element of W can be written uniquely as w = u1 + · · · + un
with ui ∈ Ui .

7.4. Orthogonal complement. If V is a vector space, with a subspace W , then the orthogonal
complement of W in V is the vector subspace:
W ⊥ := {v ∈ V : hv, wi = 0 ∀w ∈ W }
If V is a complex vector space, the inner product here means the sesquilinear inner product. In MA251
you will see a proof that V = W ⊕ W ⊥ . This involves finding an orthonormal basis for W .
Lemma 9. If L is an isometry of Rn , and W is a subspace of Rn such that L(W ) = W , then also
L(W ⊥ ) = W ⊥ . This result also holds for L extended to Cn and W a subspace of Cn .
Proof. Note that since L is an isometry, it is a bijection, so has an inverse L−1 .

v ∈ L(W ⊥ ) ⇐⇒ L−1 (v) ∈ W ⊥

D E
⇐⇒ L−1 (v), w = 0 ∀w ∈ W
D E
⇐⇒ L(L−1 (v)), L(w) = 0 ∀w ∈ W, by Lemma 8
⇐⇒ hv, L(w)i = 0 ∀w ∈ W
⇐⇒ hv, wi = 0 ∀w ∈ W, since L(W ) = W
⇐⇒ v ∈ W ⊥

MA243 December 8, 2023 14

Corollary 3. Suppose L is an isometry of Rn , and W is a subspace with L(W ) = W . Now let

w1 , w2 , . . . , wp be a basis for W and wp+1 , wp+2 , . . . , wn be a basis for W ⊥ . Then with respect to this
basis, the matrix for L has the form:
 
matrix of L|W 0
0 matrix of L|W ⊥
 
A B
Proof. Let L have matrix with respect to this basis, so if w1 ∈ W , w2 ∈ W ⊥ , then
C D
L(w1 + w2 ) = Aw1 + Bw2 + Cw1 + Dw2 ,
with Aw1 , Bw2 ∈ W , Cw1 , Dw2 ∈ W ⊥ . Then L(W ) = W ⇒ C = 0 and L(W ⊥ ) = W ⊥ ⇒ B = 0.

MA243 December 8, 2023 15

8. Lecture 8: Normal form Theorem

Lemma 10. Let λ ∈ C be an eigen value of an isometry L of Rn , with eigen vector z in Cn . Then
(1) |λ| = 1
(2) The complex conjugate λ is also an eigenvalue, with eigenvector z.
(3) If λ 6= µ are eigenvalues with eigen vectors z and w respectively, then hz, wi = 0.
Proof. Since the norm is non degenerate, and eigen vectors are non-zero, (1) follows from:
hz, zi = hAz, Azi = hλz, λzi = λλ hz, zi
Since A is real, (2) follows from:
Az = Az = Az = λz
To prove (3), first compute:
hz, wi = hAz, Awi = hλz, µwi = λµ hz, wi
So, either hz, wi = 0 or 1 = λµ. If 1 = λµ, then µ = λµµ = λ|µ|2 = λ, since by (1) |µ| = 1. But this
contradicts λ 6= µ, and so hz, wi = 0. 
Theorem 5 (Normal form Theorem). Let L : Rn → Rn be a linear isometry, with matrix A ∈ O(n, R).
Then there is an orthonormal basis of Rn with respect to which L has a matrix of the form
 
Ik

 −Im 

B1
 
 
..
 
.
 
 
B`
where Ik is the k × k identity matrix, and Bi is the 2 × 2 matrix representing a rotation of θi degrees
counter clockwise about origin
cos θi − sin θi
 
Bi =
sin θi cos θi
Proof. We use induction on n. By Theorem 3, we already know the result for n = 2, since either
A = Bθ already has the required form, or A = ρθ has eigen values 1 and −1, and orthogonal eigen vectors
(cos θ, sin θ) and (sin θ, − cos θ). In this case, with respect to this basis L has matrix diag(1, −1).
Now suppose n > 2. The characteristic polynomial of A has at least one root, and so there is at least one
eigen vector v of L, with eigen value λ. By Lemma 10, either λ = ±1 or λ is not real.
Case 1: If λ = ±1, (fixed sign choice if both are possible) then rank(A ∓ I) < n, so W = ker(A ∓ I) ⊂ Rn
has dimension dim W = d ≥ 1, (rank-nullity theorem) then L|W = ±Id . By Corollary 3, since dim W ⊥ <
n, the result follows by induction.
Case 2: if λ is not real, then v = x + iy, with x, y both non-zero real vectors, since otherwise λ 6= λ
would both be eigen values of the same vector, by Lemma 10, which is impossible.
Let W = xR ⊕ yR. By Lemma 10, we have hv, vi = 0, so 0hx + iy, x − iyi = (x + iy)T (x + iy) =
||x||2 − ||y||2 + 2ihx, yi, and by comparing imaginary parts, we see that x, y are orthogonal; by comparing
real parts, they have the same length, and so by scaling, which will preserve the eigen value of v, we can
assmume x, y are orthonormal (length one and orthogonal). Since |λ| = 1, we have λ = cosθ + i sin θ for
some θ ∈ [0, 2π). By considering real and imaginary parts of L(x) + iL(y) = L(x + iy) = L(v) = λv =
(cos θ + i sin θ)(x + iy), we obtain that L is represented by Bθ with respect to the orthonormal basis x, y.
By Corollary 3, the result follows by induction. 
MA243 December 8, 2023 16

9. Lecture 9: Reflections
Definition 11. A hyperplane of Rn is an affine subspace of dimension n − 1, which has the form
Π = V + b = {b + v : v ∈ V } = {b + λ1 v1 + · · · + λm−1 vn−1 |λi ∈ R} = (Rvn )⊥ + b.
= {v ∈ Rn : hv, vn i = β}
where b ∈ Rn and V ⊆ Rn is a vector subspace of dimension n − 1, with basis v1 , . . . , vn−1 , and vn is
orthogonal to V , and β = hvn , bi.
For any c ∈ Rn , Π + c = V + (b + c). (Π is a coset of V as an additive subgroup of Rn .)
Definition 12. For a map T : Rn → Rn , the set of fixed points of T is
Fix(T ) := {x ∈ Rn : T (x) = x}
Definition 13. Let Π be a hyperplane. A reflection in Π is a Euclidean isometry ρΠ : Rn → Rn such
that Fix(ρΠ ) = Π.
P | hP

b
V +
−b

Π=
rp l ane
hype
, vi

re f
lect
|

b+v
ion

viv
b P −b, The reflection ρΠ in the hyperplane Π
b−h
P− exists and is unique, and is given by:
ρΠ : P 7→ P − 2 hP − b, vi v,
where Π = b + (Rv)⊥ and ||v|| = 1
ρΠ (P)

Theorem 6. The reflection ρΠ in the hyperplane Π exists and is unique.

Proof. Let Π = V + b. Pick v with V ⊥ = Rv. Take a basis of Rn consisting of v together with a
basis for V . With respect to this basis, we define a linear map ρ with matrix
−1
 
0
A= .
0 In−1
This fixes V , and is not the identity, so is a reflection. Exercise: the conjugate ρΠ := T −1 ρT of ρ by
T (x) = x − b, is a reflection in Π. Thus ρΠ exists. If R 6= id fixes Π, then S := T ◦ R ◦ T −1 fixes V . By
Lemma 9, S fixes V ⊥ = Rv, so S(v) = αv for some α ∈ R. By Proposition 4, |α| = 1, so since S 6= id,
α = −1, and S has matrix A, so S = ρ, so R = ρΠ , so the reflection in Π is unique. 
MA243 December 8, 2023 17

Stuff that I would have covered in week 3 if there was more time
Lemma 11. ρΠ (P) = P − 2 hP − b, vi v, where Π = b + (Rv)⊥ and ||v|| = 1.

Proof. Exercise. 

Lemma 12. Given distinct points P and Q ∈ Rn , there exists a reflection ρ with ρ(P) = Q.

Proof. Let v = (P − Q)/||P − Q||. Let b = 12 P + 21 Q. By Lemma 5, b is a point on the line

⊥
D Q. Let Π = Rv +E b. Then ρΠ (P) = Q, because using the formula in Lemma 11,
between P and
(P−Q) (P−Q) (P−Q)
ρ(P) = P − 2 P − 12 P − 12 Q, ||P−Q|| ||P−Q|| = P − hP − Q, P − Qi ||P−Q|| 2 = Q. 

Lemma 13. If W = U ⊕ V for some vector spaces U, V and W , and R : U → U is a linear reflection,
then R ⊕ I : W → W is also a reflection.

Proof. The map R ⊕ I is defined by R ⊕ I(u + v) = R(u) + v for u ∈ U, v ∈ V . R has order 2

and fixes a codimension 1 subspace U0 of U . Hence R ⊕ I also has order 2, and fixes U0 ⊕ V which is a
hyperplane in W . 

Reflections Generate Isom(En )

Proposition 5. Any linear isometry T : Rn → Rn of Rn is the product of at most n reflections.

Proof. This result hold when n = 1, since by Lemma 10 in this case λ = ±1, corresponding to 0 or 1
reflections. Now suppose n > 1. If T 6= I, we’re done. If T 6= I, there is some P 6= 0 with T (P) = Q 6= P.
By Lemma 12, there is a reflection R1 with R1 (Q) = P, so R1 ◦ T fixes P.
Now note that R1 is linear. Not all reflections are linear (they are all affine). In general, a reflection might
be in a plane which is not through the origin. However, in this case, the hyperplane perpendicular to
the line from P to Q and midway between them does pass through the origin. From previous discussion,
the plane that you have to reflect in to get from P to Q is perpendicular to the vector P − Q. I.e., it
is a subspace V , plus some vector b. And the mid point between P and Q is b = (P + Q)/2. So the
mirror plane is V + b. If we can show that P + Q is perpendicular to P − Q, then P + Q lies in the
reflection hyperplane, i.e., b is in V , and so then V + b = V . And reflecting in a hyperplane through
the origin is linear, because the hyperplane is fixed, and so 0 is fixed. We know reflections are affine
(linear plus constant), so if 0 maps to 0, then the reflection is linear. So we just have to check that
hP + Q, P − Qi = 0. But hP + Q, P − Qi = hP, P i − hQ, Qi = ||P ||2 − ||Q||2 . But isometries preserve
length, so since Q = T (P ), we have ||Q|| = ||T (P )|| = ||P ||, so P + Q is perpendicular to P − Q, as
required, and R1 is linear.
Since R1 and T are both linear, R1 ◦ T is linear. And the space W = RP is also fixed by R1 ◦ T . By
Lemma 9 W ⊥ is fixed by R1 ◦ T . By induction (R1 ◦ T )|W ⊥ is a product of at most n − 1 reflections,
(R1 ◦ T )|W ⊥ = R2 ◦ · · · ◦ Rk with k ≤ n. By Lemma 13, these reflections can be considered as reflections
of Rn , which all fix W .
Now we have (Rk ◦ · · · ◦ R2 ◦ R1 ◦ T )|W ⊥ = I|W ⊥ , but since R1 ◦ T |W = I|W , and all Ri are trivial on
W , then also (Rk ◦ · · · ◦ R2 ◦ R1 ◦ T )|W = I|W , So since this map is trivial on both W and W ⊥ , it is the
identity, and so Rk ◦ · · · ◦ R2 ◦ R1 ◦ T = I from which we obtain the result

T = R1 ◦ R2 ◦ · · · ◦ Rk .

Theorem 7. Any isometry of En is the product of at most n + 1 reflections.

Proof. Suppose T (0) 6= 0. Then by Lemma 12, there is some reflection R1 with R1 (T (0)) = 0, so
R1 ◦ T fixes 0, and so is a linear isometry. By Proposition 7, R1 ◦ T is the product of at most n reflections,
and hence T is the product of at most n reflections. 
MA243 December 8, 2023 18

Example 3. Let T be a rotation through 90◦ about the origin of R2 . Then T has
0 −1 A2
matrix A = . Take any point which is not fixed, e.g., P = (1, 0). There is
1 0
a reflection mapping (1, 0) to T (1, 0) = (1, 0), which has matrix
 
0 1
A1 = .
1 0
Now A2 := A1 ◦ A must fix the line (1, 0)R pointwise. The orthogonal complement,
(0, 1)R must also be fixed by A1 ◦ A, so either is the identity, which would imply
A1 = A, which is not the case, or A1 ◦ A is a reflection; we can check A1
0 −1
    
0 1 1 0
A1 ◦ A = = .
1 0 1 0 0 −1
So A2 is a reflection in the x-axis, and A = A1 ◦ A2 . So T can be acheieved by a
reflection in y = 0 followed by a reflection in x = y. If we choose a different point
P, we will get a different decomposition.
9.1. Classification of isometries of R3 . Using the Normal form theorem, all isometries of R3 are
one of the following:
(1) Translation, by a vector v ∈ R3 .
(2) Rotation about some axis
(3) Twist: A rotation about some axis, followed by a translation in the direction of this axis
(4) Reflection in some plane
(5) Glide: Reflection in a plane, followed by a translation by a vector parallel to the plane.
(6) Rotary relection: Rotation followed by a reflection in a plane perpendicular to the rotation axis.
Proof is omitted. See course text book, [RS].
MA243 December 8, 2023 19

9.2. Euclid’s parallel postulate and angles in triangles.

9.3. Euclid’s fifth postulate.
Proposition 6. Given a line L and a point P ∈ R2 , with P 6∈ L, there is a unique line L0 with P ∈ L
and L ∩ L0 = ∅.
Proof. Given L = Rv + b, let L0 = Rv + P. This contains P and is a line by construction.
Suppose w ∈ L ∩ L0 . Then w = αv + b = βv + P, which implies P = (α − β)v + b ∈ L, a
contradiction. Hence L ∩ L0 = ∅, and the required L0 exists. L L0
Suppose there was another line L00 = Rv0 + P also through P with L ∩ L0 = ∅.
To find w ∈ L ∩ L00 , we want to solve w = b + αv = P + βv0 , which means solving

−α
  P
b − P = v v0 ,
β
which is possible if the matrix[vv0 ]
is invertible, which is the case if v and v0 are independent.
So, if L ∩ L00 = ∅, then v0 is a multiple of v, and so L0 = L00 . Hence L0 is unique. 
9.4. Angles and Triangles. We now define angles and recall that the sum of angles in a Euclidean
triangle is π. This is equivalent to the parallel postulate, given Euclid’s other axioms and postulates.
Definition 14. If A, B are points in Rn , the line segment between A and B is defined by
[A, B] := {(1 − λ)A + λB : λ ∈ [0, 1]}
If A, B, C are points in Rn , then their triangle is
B
∆(ABC) := [A, B] ∪ [B, C] ∪ [C, A].
We define the (measure of the) angle ∠(BAC) to be θ ∈ [0, π) such that
hB − A, C − Ai
cos(θ) := ∈ [−1, 1] C
||B − A||||C − A||
A
This is well defined by the Cauchy-Schwartz inequality, Lemma 2.
There are other possible ways to define angle, for example, as defined by Euclid.
Lemma 14. The angles in a triangle are preserved by an isometry.
Proof. This follows from Proposition 4, that an isometry is norm and inner product preserving. 
Proposition 7. The angles of a triangle ∆ABC sum to π.
Proof. See [RS] page 19. 
MA243 December 8, 2023 20

Week 4: Preamble to lecture proper

Note, I am currently skipping the rest of the results about reflections, because some questions about
spherical geometry are on the homework, so I want to make sure this is covered. We will try and fit in
some more about reflections if there is time at the end of this or some other lecture. The result about
reflections generating all isometries will not be used later in the course.
Also in the lecture, following the results of the beginning of course evaluation, I will give a very short
recap on the academic goals of the course.
Also, given that we are now discussing another metric, I will talk a little bit about metrics in general,
especially about when metric spaces can be isometric, because this is relevant to the homework this week.

10. Lecture 10: Spherical geometry: The sphere and the spherical metric
Definition 15. The n-dimensional sphere of radius r ≥ 0 is defined by
Srn := {(x1 , . . . , xn+1 ) ∈ Rn+1 : x21 + · · · + x2n+1 = r2 }
= {x ∈ Rn+1 : ||x|| = r}
S n := S1n
Example 4. Spheres S n for n = 0, 1, 2:

−1 1

S 0 is two points S 1 is a circle S 2 is a sphere

Definition 16. A spherical line or great circle is the intersection of Srn with a 2-dimensional vector
subspace of Rn+1 .
z

Definition 17. P, Q ∈ Srn are antipodal if Q = −P.

In the figure (right), P and −P are shown on two great P
circles, with 4 possible paths along great circles. These
lines all have the same length. Antipodal points have in- O
finitely many “shortest paths” between them, and so are
a special case.
z x −P y

P
Lemma 15. If P and Q ∈ Srn are not antipodal, then there is a
unique great circle containing both of them
O Proof. P and Q have length r and so are non-zero. If
they are not antipodal, then they are not multiples of each
other, since a line Rv only intersects Srn at the points λv
x Q y where ||λv|| = |λ|||v|| = R, which has only two solutions,
λ = ±R/||v||. Thus the only two dimensional vector subspace
containing P and Q is
RP ⊕ RQ = {αP + βQ : α, β ∈ R} 
MA243 December 8, 2023 21
z
Definition 18. The spherical distance between two points
P, Q ∈ Srn is the length of the shortest arc of a great circle join-
ing them. We will show that this distance is a metric, called P
the spherical metric, which we can also define by
D E
P, Q
dSrn (P, Q) := r cos−1 
  θ O
r2
 rθ

where we take cos−1 to be in [0, π]. x Q y

Notes: Spheres of different radii are not isometric.

A sphere can be projected to the plane; there are lots of ways to represent a sphere in the plane.
IMPORTANT CONVENTION: For simplicity, from now on, we take the radius r = 1.
In order to prove that dS n is a metric, we need to prove the triangle inequality.

Definition 19. A spherical triangle a in Srn consists of three

distinct vertices, P, Q, R ⊂ S n and three arcs of great circles P
joining them, which do not intersect except at the vertices, and
a specified area enclosed by these arcs. Note that there are 8 a
triangles with vertices P, Q, R.
The spherical angle at P between the lines PQ and PR is the
dihedral angle between the planes containing these lines. In the R
Q
figure, this is the angle between the red and blue planes. This
angle can also be computed as the angle between the tangent
lines to the sphere that lie in these two planes and touch the
sphere at P.

11. Lecture 11: Comparison of lines in R2 and S 2

Euclid’s fifth postulate: Given a line L and a point P ∈ R2 , with P 6∈ L, there is a unique line L0
with P ∈ L and L ∩ L0 = ∅.
There are no parallel lines in S 2 . Here, by lines we mean great circles, and by parallel, we mean lines
with no intersection. Thus Euclid’s parallel postulate does not hold for S n . More precisely:
Proposition 8. If C and D are two distinct great circles on S 2 , then the intersection C ∩ D is equal to
a pair of antipodal points.
Proof. By definition, there are 2-dimensional subspaces V and W in R3 such that C = S 2 ∩ V and
D = S 2 ∩ W . Since C =6 D, we must have V 6= W . This implies that W ∩ V is one dimensional, since if
it was zero dimensional, the space V ⊕ W would be four dimesional, but this is a subpace of R3 , which
is three dimensional. Since W 6= V , W ∩ V has dimension strictly less than 2. Hence dim(V ∩ W ) = 1
and for some v, with ||v|| = 1, we have W ∩ V = Rv, and so C ∩ D = S 2 ∩ V ∩ W = S 2 ∩ Rv = {v, −v},
which is a set of two antipodal points. 
Note: Great circles in S n for n > 2 do not always intersect.
MA243 December 8, 2023 22

12. Lecture 12: Main formula of spherical trigonometry

(This was covered partly in lecture 11 and partly in lecture 12)

Proposition 9 (Main formula of spherical trigonometry). Let P

α, β, γ be the side lengths of a spherical triangle with vertices
P, Q, R ∈ S 2 on the unit sphere. a γ
β
α := d(Q, R), β := d(P, Q), γ := d(P, R),
where d = dS 2 is the spherical metric. Let a be the spherical angle R
Q α
between arcs PQ and PR. Then
cos α = cos β · cos γ + sin β · sin γ · cos a

Proof. Note that the distance on the sphere, in Definition 18, is given in terms of the inner product. By
Proposition 4, any linear isometry of R3 preserves the inner product, so, if we apply a linear isometry of
R3 to the sphere and the points on it, the distances between the points will be unchanged. Therefore, we
may apply an isometry to map the vertices of the triangle to more convenient positions.

There is an isometry L of R3 which maps P to (0, 0, 1), P = (0, 0, 1)

since these points are equidistant from 0. This may be
followed by a rotation about the z axis to map Q to a β
point in the xz plane (the plane where y = 0). Thus, γ
Q
without loss of generality, we may assume P = (0, 0, 1),
and Q has y coordinate 0. We can use this to write the
Euclidean coordinates of Q and R in terms of a, β and γ, R
as in the figure: a
Q = (sin β, 0, cos β)
R = (sin γ cos a, sin γ sin a, cos γ) Q0 = (1, 0, 0) R0
By definition,
cos α = hQ, Ri = sin β sin γ cos a + cos β cos γ
which gives the required result. 
Note, since it may be hard to see why P, Q and R can be put in this form, there are futher pictures on
the next two pages.
MA243 December 8, 2023 23

P = (0, 0, 1)

γ
β)
Q = (sin(β), 0, cos(β))
sin(

R = (sin γ cos(a), sin γ sin(a), cos β)

sin(γ
)
β γ

sin(a
)

Q0 = (1, 0, 0) R0 = (cos(a), sin(a), 0)

MA243 December 8, 2023 24

12.1. “Cut out” proof of Main theorem for spherical trigonometry.

cos α = cos(a) sin(β) sin(γ) + cos(β) cos(γ)

R = sin(γ)R0 + (0, 0, cos(γ))

β
P = (0, 0, 1)
γ

z-axis
sin(β)
Q = (sin(β), 0, cos(β))

cos(β)
sin(γ)

cos(γ)
β
γ
0

R0 = (cos(a), sin(a), 0)
Q0 = (1, 0, 0)

x-axis line in xy-plane

P = (0, 0, 1)
Q0 = (1, 0, 0)
Q = ( sin β, 0, cos β)
R0 = ( sin a, cos a, 0)
R = (sin γ cos a, sin γ sin a, cos γ)

Cut out the circle, and fold along the vertical axis. As you “open or close” the circle, the angles a, which
is the dihedral angle between the two planes (the plane containing 0, P and R and the plane containing
0, P, Q), and the angle α, which is the angle between the two vectors Q and R will vary.
The minimum possible value of cos(a) is −1, which is obtained before the paper is folded, ie., lying flat,
unfolded. In this case, either α = β + γ, when β + γ ≤ π, or, if β + γ > π, as in this little picture, then
if a = π, we still have cos(α) = cos(β + γ), so β + γ = 2π − α.

β γ
MA243 December 8, 2023 25

Lemma 16. The spherical distance satisfies the triangle inequality. That is, with P, Q, R, α, β, γ and a
as in Proposition 9, we have α ≤ β + γ, with equality, or with α + β + γ = 2π ⇐⇒ a = π. Furthermore,
a = 0 ⇐⇒ α = |β − γ|.
Proof. Since P, Q, R are contained in a vector space V of dimension 3, and so lie on the unit sphere
S2 ∼
= V ∩ S n , it is enough to prove the result for n = 2. We may assume P, Q, R are distinct, since
otherwise the result is easy.
By definition, α, β, γ ∈ [0, π], so sin α, sin β ≥ 0. Since cos a ≥ −1, by
f (x)
Proposition 9 1
(††) cos α = cos β · cos γ + sin β · sin γ · cos a
cos α
≥ cos β · cos γ − sin β · sin γ = cos(β + γ)
Case β + γ ≤ π: By definition, α ∈ [0, π]. On [0, π], cos(x) is strictly x
decreasing, so cos α ≥ cos(β + γ) ⇒ α ≤ β + γ. There is equality in (††) α π β+γ π
2
if and only if cos(a) = −1, which is if and only if a = π.
Case β + γ > π: Since α ∈ [0, π], the inequality is immediate. Equality in
(††) occurs ⇐⇒ a = π, but now β + γ > π so cos(β + γ) = cos(α) ⇐⇒ cos(β+γ)
α = 2π − (β + γ) ⇐⇒ α + β + γ = 2π. The final statement follows −1 f (x) = cos x
similarly, since a = 0 ⇐⇒ cos(α) = cos(β − γ). 
Proposition 10. The spherical distance dS n is a metric on S n
Proof. Symmetry follows from symmetry of the inner product on Rn . Non-degeneracy follows since
cos−1 ∈ [0, π], so d(P, Q) = 0 ⇐⇒ hP, Qi = 1, and since ||P|| = ||Q|| = 1, this implies P = Q. The
triangle inequality is proved in Lemma 16. 
MA243 December 8, 2023 26

13. Lecture 13: Spherical isometries

In this section we write d := dS n . The goal is to show that Isom(S n , d) ∼
= O(n + 1).
Easy direction first:
Lemma 17. If T is a linear Euclidean isometry of Rn+1 , then the restriction T |S n to S n is a spherical
isometry.
Proof. Since T preserves the Euclidean distance, and fixes 0, it must map S n to S n , since this is
the set of points distance 1 from 0. Since T has an inverse which is also a Euclidean isometry, and so also
maps S n to S n , then T restricted to S n is a bijection. By Proposition 4 T preserves the inner product,
and thus preserves the spherical distance cos−1 (hx, yi) on S n . Hence TS n is a spherical isometry 

Isometries are given by linear maps

Lemma 18. A bijective map f : Rk → Rk , which preserves the standard inner product on Rk is a linear
isometry of Rk .
p
Proof. If the inner product is preserved, then since d(x, y) = hx − y, x − yi, then f must preserve
distance: ||f (x) − f (y)||2 = hf (x) − f (y), f (x) − f (y)i = hf (x), f (x)i − 2 hf (x), f (y)i + hf (y), f (y)i =
hx, xi − 2 hx, yi + hy, yi = ||x − y||. Given that f is bijective, f must be a Euclidean isometry. Since
the inner product is preserved, f (0) = 0. By Corollary 1 isometries of Rk fixing 0 are given by linear
maps. 
Remark: We can drop the assumption that f is a bijection in Lemma 18, as this follows since if
ai f (ei ) = 0, then by taking the inner product with f (ej ), we find aj = 0 for 1 ≤ j < k.
P

Proposition 11. If T is an isometry of S n , then T extends to a Euclidean isometry Tb of Rn+1 , that is,
there is a Euclidean isometry Tb of Rn+1 , and Tb|S n = T .
Proof. Given an isometry T of S n , define Tb by radial extension, i.e., by

0 if x = 0
(8) Tb(x) = 
x

T
||x|| ||x|| if x =
6 0
Tb(x)

x 
x

T ||x||
x Tb
||x||

We will show that Tb is an isometry of Rn+1 . First, Tb is a bijection, because Td−1 is the inverse of T
b:
First, note that since T and T −1 n
preserve the S , only 0 maps to 0, so we assume x 6= 0. Then, using
the fact that ||T (y)|| = 1 for all y ∈ S n , we have
x z x
         
−1 −1 −1
T
d T (x) = T
b d T ||x|| = T ||z||, where z = T ||x||
||x|| ||z|| ||x||
x
  
= T −1 T ||x||, since ||z|| = ||x||
||x||
x
 
= ||x|| = x.
||x||
Now we show Tb preserves the inner product on Rn+1 . Since T preserves the spherical metric, for x, y ∈ S n ,
hx, yi = cos (dS n (x, y)) = cos(dS n (T (x), T (y)) = hT (x), T (y)i, so T preserves the inner product restricted
MA243 December 8, 2023 27

to S n . Suppose x, y ∈ Rn . Assume x, y 6= 0, since otherwise the result is easy.

D E   x  
y
 
Tb(x), Tb(y), = T ||x||, T ||y||
||x|| ||y||
x y
    
= T ,T ||x||||y||, since the inner product is bilinear
||x|| ||y||
x y
 
= , ||x||||y||, since T preserves the inner product on S n
||x|| ||y||
= hx, yi by bilinearity
Hence by Lemma 18 Tb is linear Euclidean isometry. 
Corollary 4. There is a group isomorphism
Isom(S n , dS n ) ∼
= O(n + 1)
Proof. This follows since Lemma 17 tells us we have a map O(n + 1) → Isom(S n , dS n ). This map
must be injective, since S n contains a basis for Rn+1 , so for A ∈ O(n + 1), the mapping TA |S n determines
the map TA . Proposition 11 tell us that this map is surjective. 
MA243 December 8, 2023 28

14. Lecture 14: Spherical Triangles

In this section we restrict to S 2

In the Euclidean plane, the angles of a triangle sum to π.

P
We now show that the angles of a (non-trivial) spherical triangle sum
to a number strictly greater than π.
First, note that the angle sum for a spherical triangle is not constant.
θ
For example, the triangle in the figure, where P = (0, 0, 1) and Q =
(1, 0, 0) and R = (cos θ, sin θ, 0) can have angle sum π + θ for any
θ ∈ (0, 2π). π π
2 2
We will show that the angle sum is determined by the area of the Q R
triangle.

We will make some assumptions about area, in particular, assume that you will learn about a rigourous
definition of area in terms of integration in analysis.

Assumptions:
(1) The total surface area of a unit sphere is 4π.
(2) The area of a region of S 2 is invariant under application of θ
any isometry of S 2 .
(3) If X and Y are two non intersecting regions on S 2 , then
area(X ∪ Y ) = area(X) + area(Y )
(4) The segment of a sphere, σθ that lies between two semi- Segment σθ
θ
circular segments of great circles, from P to −P, meeting area 2π 4π
at angle θ has area 2θ.
Let Σθ be the union of σθ and −σθ . So Σθ has area 4θ.

Theorem 8. (Girard’s theorem) Let ∆PQR be a spherical triangle on S 2 , with P, Q, R distinct points,
and such that the interior of ∆PQR does not intersect the great circles which contain the sides of ∆PQR,
and with non-zero area. Then
area (∆PQR) = a + b + c − π
Proof. Notice that area(∆PQR) = area(−∆PQR), since x 7→ −x
is an isometry, and area is preserved by isometries. Also notice
that since we assume that the triangle does not contain any of the
extended edges in its interior, these triangles are disjoint. Notice
P that S 2 is the union of Σa , Σb and Σc , and that
a σa ∩ σb = σb ∩ σc = σa ∩ σb = ∆PQR
and similarly for the images of these areas under x 7→ −x.
b c Let
Q R Sa = σa \ ∆PQR
Sb = σb \ ∆PQR
Sc = σc \ ∆PQR
Then we have a disjoint union:
S 2 = Sa ∪ Sb ∪ Sc ∪ (−Sa ) ∪ (−Sb ) ∪ (−Sc ) ∪ ∆PQR ∪ (−∆PQR).
MA243 December 8, 2023 29

Since we have a disjoint union σa = Sa ∪ ∆PQR), we have area(σθ ) = area(Sθ ) + area(∆PQR), for
θ = a, b, c, so
area(S 2 ) = area(Σa ) + area(Σa ) + area(Σa ) − 4 area(∆PQR)).
By assumption Σθ has area 2θ so
4π = 4(a + b + c) − 4 area(∆PQR),
and the result follows. 
Remark: This result also holds if we do not make the restriction that the triangle does not contain
the extensions of its edges in its interior, but then ∆PQR and (−∆PQR) can not be guaranteed to be
disjoint, and so we need to consider several different cases. To save time, I will not prove the general
case, which is not needed for the following result.
Corollary 5. The sum of the angles of a spherical triangle, satisfying the hypothesis of Theorem 8, is
strictly greater than π.
Proof. This follows from Theorem 8. 
Schwarz triangles: Suppose ∆ is a spherical triangle, and G
is the group generated by reflections in the sides of ∆. Suppose
these reflections result in a tiling of the sphere – this means that
S 2 = ∪T ∈G T (∆) and T1 (∆)◦ ∩ T2 (∆)◦ 6= ∅ ⇐⇒ T1 = T2 ,
where X ◦ means the interior of X. Then the angles must divide
π. This means that if ∆ tiles S 2 through reflections, then the
angles are πp , πq , πr for positive integers p, q, r such that
1 1 1
+ + >1
p q r
The only solutions are (p, 2, 2), (3, 3, 2), (4, 3, 2), (5, 3, 2). The
later is shown on the right.
(You will probably see more of this if you take the third year course on reflection groups.)
MA243 December 8, 2023 30

15. Lecture 15: Hyperbolic Geometry

Motivation. So far we have looked at Euclidean space, which is flat, zero curvature, and the sphere,
which is curved – positive curvature. Hyperbolic space has everywhere constant negative curvature.
Investigating these concepts in more detail is beyond the scope of this course.

zero curvature positive curvature negative curvature

Distortion. There is no way to draw the sphere on the plane without some distortion. This gives
rise to various different map projections. (Possibly topic for a second year essay).
Example, of a non isometric map from S 2 to R2 :
imagine a 3D sphere flat version of the sphere

distortion

Similarly, the hyperbolic plane can’t be mapped isometrically to R2 or R3 . Depicting it in R2 or R3 must

give distortions. There are several standard ways to put the hyperbolic plane in Euclidean space, two of
which are as follows, which either “squash” into a disk, or “stretch” onto a hyperboloid:
We will use the hyperboliod model, since it’s closest to the sphere, which we have just studied. The
disadvantage of this model is that it does not preserve angles (not conformal).
Approx isometric
view, but not
smooth

squashed into a disk stretched to a hyperboloid

top sheet of two

sheeted
p hyperboloid:
x2 + y 2 − z 2 = i

Poincaré disk model

MA243 December 8, 2023 31

Hyperbolic space
We will use the model of hyperbolic space defined as follows. An advantage of this definition is that
results are very similar to results in the case of the sphere, just with a “twist”, which is to replace the
usual Euclidean metric hx, yi = xT y with the the Lorentz inner product. We define
−1
 
0
Jn :=
0 In−1
Note that Jn has n rows and columns and acts on Rn . If x, y ∈ Rn , then xT Jy means xT Jn y.
Definition 20. The Lorentz inner product of two vectors x, y ∈ Rn is defined by:
hx, yiL := −x1 y1 + x2 y2 + x3 y3 + · · · xn yn
= xT Jy
Definition 21. The Lorentz norm of a vector x ∈ Rn is given by
q
||x||L := hx, xiL

We always take this root to be positive or positive imaginary, or zero. WARNING: hx, xi may be negative,
so ||x||L may be imaginary, so is not a norm.
Definition 22. The n-dimensional hyperbolic space is the metric space with underlying set
n o
Hn := (x1 , x2 , . . . , xn+1 ) ∈ Rn+1 : −x21 + x22 + x23 + · · · + x2n+1 = −1 and x1 > 0
n o
= x ∈ Rn+1 : ||x||L = i and x1 > 0

and with metric defined by

hx, yiL
 
−1
d
Hn (x, y) := cosh = cosh−1 (− hx, yiL )
||x||L ||y||L
where we have used that on Hn , ||x||L = i. We will review the hyperbolic cosine cosh shortly.

Examples: The x1 -axis is usually drawn pointing up.

H1 is the upper half of the hyperbola: H2 is the upper sheet of a hyperboloid:a
x1
4 H1 :√−y2 +x2 =i and y>0

parametrization:
(sinh(t), cosh(t)) 2
1
x2
−4 −2 2 4
−1
−2

−4
ahttps://www.geogebra.org/m/xqnc3sdm

The Goal of this chapter is to

(1) show that the hyperbolic metric is a metric
(2) determine the group of hyperbolic isometries
(3) investigate the failure of the parallel postulate
(4) show that triangles have angles sum less than π
First we need to understand hyperbolic sine and cosine, and check dH n is well defined.
MA243 December 8, 2023 32

Hyperbolic trigonometry [please review for homework]

Definition 23. The hyperbolic sine and cosine, sinh and cosh are defined by

exp(θ)−exp(−θ) exp(θ)+exp(−θ)
sinh(θ) = 2 = −i sin(iθ) cosh(θ) = 2 = cos(iθ)
2 y 2 y

1 1

x x
−2 −1 1 2 −2 −1 1 2

−1 −1
cosh(x)
sinh(x) cosh−1 (x)
−2 −2

sinh−1 : R → R cosh−1 : [1, ∞) → [0, ∞)

q q
y 7→ ln(y + y 2 + 1) y 7→ ln(y + y 2 − 1)
Geometric interpretation:
Formulae: These can be derived from properties of
the exponential, or the relationship with cos and sin, (cosh(θ), sinh(θ)
and formulae you already know
hyperbolic
θ
cosh(a + b) = cosh(a) cosh(b) + sinh(a) sinh(b) h
gt
sinh(a + b) = sinh(a) cosh(b) + cosh(a) sinh(b)
en

cosh0 (a) = sinh(a) sinh0 (a) = cosh(a) (cos(θ), sin(θ))

cl
oli

2 2
cosh (a) − sinh (a) = 1
erb

le
ng
hyp

comparison with sine and cosine:

th
circle
trigonometric area = θ

θ
θ
area = 2
cos(a + b) = cos(a) cos(b) − sin(a) sin(b) 2
sin(a + b) = sin(a) cos(b) + cos(a) sin(b)
1 1
cos0 (a) = − sin(a) sin0 (a) = cos(a)
2 2 hyperbola
cos (a) + sin (a) = 1

The definition of the metric for Hn is comparable to that for S n :

hx, yi hx, yiL
   
−1 −1
dS n (x, y) = cos dHn (x, y) = cosh
||x||||y|| ||x||L ||y||L
So results about the hyperbolic metric are analagous to results for the spherical metric.
hx,yiL
To show that dHn (x, y) is well defined, we need to prove that if x, y ∈ Hn , then ||x||L ||y|| L
∈ [1, ∞). We
will follow [R3.7–3.10], and find an isometry which makes these results easier to prove. First we need to
investigate the Lorentz inner product, and the maps of Rn+1 which preserve it. Eventually we will prove
that these are all linear. We follow [RS, Appendix B] for Lorentz linear algebra.
MA243 December 8, 2023 33

Lemma 19. There is a parameterization

f : R → H1
f : (t) 7→ (cosh(t), sinh(t)) t ∈ [0, ∞)
There is a parameterization
f : R2 → H 2
f : (t, θ) 7→ (cosh(t), cos(θ) sinh(t), sin(θ) sinh(t)), t ∈ [0, ∞), θ ∈ [0, 2π),

where R2 is given with polar coordinates, and H2 with the Cartesian coordinates of R3 .
Proof. For the first parameterisation, we use
the fact that cosh(t)2 −sinh(t)2 = 1, and that cosh(t)
is strictly increasing on [0, ∞), and symmetric, and
that sinh(−t) = − sinh(t).
For the parameterisation of H2 , this is the inverse
of the projection from R3 given by (x1 , x2 , x3 ) →
(x2 , x3 ). This is a bijection between H2 and R2 , since
for any (x2 , x3 ) there is a unique solution to
(?) − x21 + x22 + x23 = −1 and x1 > 0.
In polar coordinates,
(x2 , x3 ) = (r cos(θ), r sin(θ)),
so (?) becomes
−x21 + r2 = −1, x1 > 0,
the equation for H1 , parametrised by
(x1 , r) = (cosh(t), sinh(t)),
from which the result follows by substitution. 

15.1. The hyperbolic metric is well defined.

Lemma 20. For x, y ∈ Hn we have − hx, yiL ≥ 1, and hx, yi > 0.
Proof. Let x = (x1 , x2 , . . . , xn ) and y = (y1 , y2 , . . . , yn ). Let x
b = (1, x2 , . . . , xn ) and y
b = (1, y2 , . . . , yn ).
Then since x, y ∈ Hn , we have
(9) x21 = 1 + x22 + x23 + · · · + x2n = ||x
b ||2
y12 = 1 + y22 + y32 + · · · + yn2 = ||y
b ||2 ,

where the norm on the right is the Euclidean norm. So we have by the Cauchy-Schwartz inequality for
the Euclidean norm, Lemma 2,
hx b i2 ≤ ||x
b, y b ||2 ||y
b ||2 = (x1 y1 )2
i.e.,
(10) |1 + x2 y2 + x3 y3 + . . . xn yn | ≤ |x1 y1 |
so
1 + x2 y2 + x3 y3 + . . . xn yn ≤ |x1 y1 | = x1 y1
where we have used that x1 , y1 are positive, since x, y ∈ Hn . Rearranged, we get
hx, yiL = −x1 y1 + x2 y2 + x3 y3 + · · · + xn yn ≤ −1,
which multiplying through by −1 gives the first inequality.
MA243 December 8, 2023 34

By the Cauchy-Schwartz inequality, we have

|x2 y2 + x3 y3 + · · · + xn yn |2 ≤ (x22 + · · · x2n )(y22 + · · · yn2 ) = (x21 − 1)(y12 − 1)
= x21 y12 − y12 − x21 + 1
< x21 y12 ,
so we have
x1 y1 > |x2 y2 + x3 y3 + · · · + xn yn |
where we have used that x1 ≥ 1 and y1 ≥ 1, which follows from Equation 9. So
hx, yi = x1 y1 + (x2 y2 + x3 y3 + . . . xn yn )
≥ x1 y1 − |x2 y2 + x3 y3 + . . . xn yn | > 0,
which gives the second inequality. 
Corollary 6. dHn is well defined.
Proof. This is immediate from Lemma 20 and the definition of dHn . 
15.2. Terminology inspired by special relativity: Time, space, and light like vectors.
Definition 24.



 space-like if ||x||L ∈ (0, ∞), i.e., if hx, xiL > 0

 light-like if ||x||L = 0, i.e., if hx, xiL = 0



A point x ∈ Rn is time-like if ||x||L ∈ i(0, ∞), i.e., if hx, xiL < 0

positive if x1 > 0






 negative if x1 < 0
A time-like or space-like vector x ∈ Rn is positive if x1 > 0 and negative if x1 < 0.
Note that vectors in Hn are all time-like.
x1 (“time”)

on the cone
light like;
||x||L = 0 inside cone:
positive time like H2 = surface of hyperboloid
outside cone: = {x : ||x||L = i, z > 0}
space like region
x2 (“space”)

x3 (“space”)

inside cone:
negative time like Lorentz transoformations fix H2
MA243 December 8, 2023 35

16. Lecture 16: The Lorentz group

Definition 25. A map T : Rn → Rn is a Lorentz transformation if it preserves the Lorentz inner product:
hT (x), T (y)iL = hx, yiL ∀x, y ∈ Rn
We say that T is positive if x is positive time-like if and only if T (x) is positive time-like.

Lemma 21. If T : Rn+1 → Rn+1 is a positive, bijective Lorentz transformation, then T |Hn ∈ Isom(Hn , dHn ).
Proof. Since Hn = {x ∈ Rn : hx, xiL = −1 and x is positive}, T preserves Hn . Since T is bijective,
it has a bijective inverse, which is also a positive Lorentz transformation, and so T |Hn is a bijection.
Distance on Hn is given in terms of the Lorentz inner product, so T is distance preserving. Hence T |Hn
is an isometry. 

Euclidean isometries correspond to orthogonal matrices. The Lorentzian analogy is given by:
Definition 26. An n × n real matrix A is Lorentz orthogonal if

−1
 
T 0
A JA = J :=
0 In−1
Definition 27. The Lorentz group is the group of Lorentz orthogonal n + 1 × n + 1 matrices, denoted

O(1, n) = {A ∈ Mn+1×n+1 |AT JA = J}

The positive Lorentz group is the subgroup of O(1, n) which maps positive time like vectors bijectively to
positive time like vectors. This is denoted O+ (1, n).
Lemma 22. O(1, n) and O+ (1, n) are groups.
Proof. Exercise. 
new Lorentz
orthonormal
√ basis:
√
(2, 3) and ( 3, 2)
Example: √ !
√2 3
∈ O(1, 1) H1
3 2
old Lorentz
orthonormal basis:
standard basis

We check this by showing that:

√ ! √ ! √ ! √ ! 
−1 −2 − −1 0
 
√2 3 0 √2 3 √2 3 √ 3
= = .
3 2 0 1 3 2 3 2 3 2 0 1

By Lemma 24 this means that the columns of this matrix form a Lorentz orthonormal basis for R2 .

Lorentz translation
 
cosh γ sinh γ
More generally for any γ ∈ R, Aγ = defines an element of O+ (1, 1), which we can think
sinh γ cosh γ
of as inducing a translation on H1 , since we have
      
cosh γ sinh γ cosh x sinh γ sinh x + cosh γ cosh x cosh(x + γ)
= =
sinh γ cosh γ sinh x cosh γ sinh x + sinh γ cosh x sinh(x + γ)

The spaces (E1 , dE1 ) and (H1 , dH1 ) are isometric, via the paramterisation x 7→ (sinh x, cosh x).
MA243 December 8, 2023 36

WARNING: first component, x1 is the vertical axis

x1
 
cosh(2) sinh(2)
A2 =
pi 7→ qi = pi + 2 sinh(2) cosh(2)
Q0
H1
Pi 7→ Qi = A2 Pi
R1 f : x 7→ (sinh x, cosh x)
P3 Q1
p3 p2 p1 p0 q3 q2 q1 q0
p0 p1 p2 p3 P2 Q2
0 −0.5 −1 −1.5 P1 P0 Q3
Pi = f (pi ); Qi = f (qi ) x2
P0 = (1, 0); Q0 = (cosh(2), sinh(2))
Since cosh(x) ∈ [1, ∞), the image of Aγ applied to any element of H1 is in H1 , and by linearity, Aγ
preserves all positive elements of R2 , and so Aγ ∈ O+ (1, n).
Quiz: What do you think a “reflection” matrix for H1 looks like?
Note that the parameterisation is not a projection to the x2 axis:
x1
illustration of the
isometry from
(R1 , dR1 ) to (H1 , dH1 ).
The red dots are equally H1
spaced, distance 0.5
apart in the hyperbolic
metric, and the blue
dots are equally spaced
distance 0.5 apart in
the Euclidean metric.
R1 x2

To Do:
Important results we hope to have time to prove later:
• The hyperbolic metric is a metric
• All isometries of Hn are given by elements of O(1, n).
• For a hyperbolic triangle with angles a, b, c and area A, we have A = π − a − b − c.
Things we will actually do:
• Think about lines in the hyperbolic plane, by projection to the plane x1 = 0.
• Think about what are all the isometries of the hyperbolic plane, e.g., translation, rotation, etc
MA243 December 8, 2023 37

Some different views of the hyperbolic line. The hyperbolic line is isometric to R with the
Euclidean metric, under the map t 7→ (cosh(t), sinh(t)). So the simplest way to think about H1 would be
to imagine the usual Euclidean line.
However, H2 can not be embedded smoothly and isometrically in R3 , so is harder to imagine. There are
several common models to work with. Since these models are easier to understand in the analogous one
dimensional cases, the following picture shows 4 different ways to view H1 .

Lower, cyan line t∈R

parameter, mapping to red upper hyperbola
Red hyperbola (cosh(t), sinh(t))
vertical component first.
(point, mapping to either orange or green lines)
Green line sinh(t)
vertical projection from hyperbola
Orange segment (−1, 1) sinh(t)/(1 + cosh(t)), Poincare line.
Projection of the hyperbola from the point (0, −1).

17. Lecture 17: Hyperbolic lines

We need to prove the triangle inequality, and discuss lines and triangles. It will be useful to introduce
some terminology about hyperbolic lines.
Definition 28. A 2-dimensional sub-vector space V of Rn+1 is called a Lorentz plane if it contains a
timelike vector. A hyperbolic line is the intersection of Hn with any Lorentz plane.
Hyperbolic line segments are the shortest distances between two points in the hyperbolic metric. (We do
not have time to prove this.)
Examples are shown in the following Geogebra figures: line; triangle; parallel lines. 1

Lemma 23. For P 6= Q ∈ Hn , there is a unique hyperbolic line L containing P and Q.

Proof. Let V be the subvector space of Rn+1 spanned by P and Q, and set L = Hn ∩ V . 
We will show that Euclid’s parallel postulate does not hold in H2 .
Definition 29. A one dimensional subvector space V = Rv of Rn is called time-like, space-like or
light-like depending on whether v is time-like, space-like or light-like.
Definition 30. Let L1 = H2 ∩ Π1 and L2 = H2 ∩ Π2 be two distinct lines in H2 , for two dimensional
subvector spaces Π1 and Π2 . Let v ∈ Π1 ∩ Π2 \ {0}, and V = Rv = Π1 ∩ Π2 .
(1) V is time-like: L1 and L2 intersect at a point x, with {x} = V ∩ H2 .
(2) V is space-like: L1 and L2 are parallel and diverge.
(3) V is light-like: L1 and L2 are ultraparallel.
1www.geogebra.org/3d/xk4cauzn ; www.geogebra.org/3d/ktnpt5ae ; www.geogebra.org/3d/bdtqpp8v
MA243 December 8, 2023 38

If a plane does not contain a time like vector, it can’t intersect H2 , as for example in this case:

Any plane has the form (Rv)⊥ for some vector v. As long as v is space line, we’re guaranteed that
H2 ∩ (Rv)⊥ is non empty.
You can move the point P in the following geogebra graph to see this for yourself:
https://www.geogebra.org/3d/tu9fgsgj

WARNING: We are using variables x1 , x2 , x3 , whereas in geogebra these are called z, y, x.

Theorem 9. Let x ∈ H2 , and let L be a line in H2 which

does not contain x. Then there are infinitely many lines
in H2 which pass through x and do not intersect L.

Proof. Let Π be a 2 dimensional vector subspace such that L = Π ∩ H.

We will find infinitely many lines Lt , containing x and disjoint from Π, where t may take infinitely many
different values, and Lt 6= L0t for t 6= t0 .
Each Lt is of the form Πt ∩ H2 for some subspace Πt .
We are going to find a way to choose Πt so that Πt ∩ Π is space like. This means that Πt ∩ Π can not
contain any time like vector, and so in particular can’t contain any point in H2 , and so Lt ∩ L = ∅. We
will give the sequence of steps of the argument together with illustrations from geogebra.
MA243 December 8, 2023 39

https://www.geogebra.org/3d/jz6j8cc4

step 1: Pick a point u in L

https://www.geogebra.org/3d/jz6j8cc4

step 2:
• Pick a point v in Π that is space like.
• This can be any point, but to be spe-
cific, let’s take a non zero point on
the line where Π intersects the plane
{x1 = 0}.
• Since the two planes Π and {x1 = 0}
are two different two dimensional sub-
spaces of R3 , they must intersect in a
line, which is spanned by some vector
v.
• Since we’ve chosen v = (0, v2 , v3 ) 6=
(0, 0, 0) we must have hv, viL = v22 +
v32 > 0, so v is a space like point on Π

https://www.geogebra.org/3d/faf89g7m

step 3: Consider the line segment S from v to

u. From our description of lines in Euclidean
space, we know this is the set of points of the
form
Pt = (1 − t)v + tu
My goal is to find infinitely many space like
points on this line segment S. (This is a Eu-
clidean line, not a hyperbolic line). I could find
other space like points, but this is one possible
method.
MA243 December 8, 2023 40

step 4
Note that
• v is space like, and
• u is time like. https://www.geogebra.org/3d/xfsqjeq5
• And there is a continuous function g(t)
from [0, 1] to R mapping t to g(t) :=
hPt , Pt iL
• Since g(0) = hv, viL = α > 0 and
g(0) = hu, uiL = −1, for any value of
a ∈ (0, α), by the intermediate value
theorem, there is some point Pta ∈
[0, 1) with g(ta ) = a, i.e., Pta is a space
like point on Π, and since these points
all have different values of their Lorentz
norm, they are different points. So we
have infinitely many space like points
in Π on this line segment.
We can add the light cone to emphasise that
points on S which are not in the light cone are
space like points on Π.
It’s not that easy to see, but the green area is
the light cone.
https://www.geogebra.org/3d/bxhgbkf2

step 5
Now consider one of the space like points Pt , and
the point x on H.
MA243 December 8, 2023 41

https://www.geogebra.org/3d/uakw5h7p

step 6
• Now consider one of the space like
points Pt , and the point x on H.
• We let the plane Πt be the span of the
vectors x and Pt .
• This plane intersects Π in the subspace
RPt spanned by the vector Pt .
• This is because Π contains Pt , since Pt
was chosen to be on Π and Πt contains
Pt since Πt was defined to be the span
of Pt and x.
• Since RPt ⊂ Π ∩ Πt , we must have
RPt = Π ∩ Πt , since otherwise they
would be equal, but x is in Πt by con-
struction, but is not in L, and thus not
in Π by assumption.

Let Lt = H ∩ Πt . Since this contains the time like point x, it is a line. (not empty).
We’re now going to double check that the Lt lines are the infinite collection of lines we’re looking for. A
projected to R2 picture looks like:
• The lines Lt and L are disjoint, since the intersection is Lt ∩ L = Πt ∩ H ∩ Π ∩ H = Πt ∩ Π ∩ H =
RPt ∩ H = ∅, where the final step is because all points on RPt are space like (or light like at
the origin), but all points on H are time like. So the intersection is empty.
• Now we just have to check that the lines Lt are all different. I.e., if t1 6= t1 then Lt1 6= Lt2 . This
is because the planes Πt1 and Πt2 are different, which is because the vectors Pt1 and Pt2 are not
colinear. If they were colinear, then 0 would be a point on the line through Pt1 and Pt2 which
is the same as the line through u and v by construction. But the origin can’t be on the line
between these points, or else we would be able to write 0 = (1 − t)u + tv for some real number
t, but then u and v would be linearly dependent, but this contradicts one being time like and
the other being space like. since all non-zero multiples of a space like vector are space like, and
similarly for time like vectors.
So, Πt1 and Πt2 are different planes. This implies their intersections with H2 are different,
since if not, they would have two points in common, but any two points will span a plane, unless
they are colinear. Two points on H2 can not be colinear, since for any timelike y, there is a
unique α with αy on H.
So, Lt are different for different t.
Hence there are infinitely many lines Lt which contain x but do not intersect L, as required.

MA243 December 8, 2023 42

18. Lecture 18: Sketching hyperbolic lines

(We first will spend some time going over the proof covered at the end of last lecture)
We can’t easily draw on a hyperboloid when we only have flat paper.
To represent hyperbolic lines in H2 , we can project from H2 to the x1 = 0 plane.
The projection map is given by
π : H 2 → R2
(x1 , x2 , x3 ) 7→ (x2 , x3 )
In this section, we describe the images of lines in H2 under this projection map.
Suppose Π is a plane given by ax1 + bx2 + cx3 = 0.
For there to be a time like point on Π, so that L is non-empty, we must have −a2 + b2 + c2 > 0.
Then on L = Π ∩ H2 , we have ax1 = −bx2 − cx3 , and also x21 = 1 + x22 + x23 , and x1 > 0, so we obtain

(bx2 + cx3 )2 = a2 (1 + x22 + x23 )

which is a quadratic equation, which defines a conic section.
(b2 − a2 )x22 + 2bcx2 x3 + (c2 − a2 )x23 = a2
We can factor the left hand side to get
(b2 − a2 )(x2 − α1 x3 )(x2 − α2 x3 ) = a2
where
! √
1 q −2bc ± 2a b2 + c2 − a2
α1 , α2 = −2bc ± 4b2 c2 − 4(b2 − a2 )(c2 − a2 ) = ,
2(b2 − a2 ) 2(b2 − a2 )
And since b2 + c2 − a2 > 0, there are two distinct asymptotes, given by x2 = α1 x3 and x2 = α2 x3 , so this
conic is a hyperbola.
Example lines:
case 1: If a = 0, then the line is 0 = bx2 + cx3 , which is a straight line through the origin.
E.g., taking c = 0 we have a hyperbolic line which projects to x2 = 0.
case 2: Now consider the case a 6= 0. Since we have x1 > 0, then −(bx2 + cx3 )/a = x1 > 0, so we can
use this to determine which half of the hyperbola
√ is obtained.
case 3: Now consider the point at P = ( 2, 1, 0)√on H2 . This point projects to (1, 0). For √ the plane
ax1 + bx2 + cx3 = 0 to pass through P , we need a 2 + b = 0, so we take a = 1, and b = − 2. c can be
anything. The projection of the line on R2 is
√
(− 2x2 + cx3 )2 = (1 + x22 + x23 )
So √
2x22 + 2 2cx2 x3 + c2 x23 = 1 + x22 + x23
So √
x22 + 2 2cx2 x3 + (c2 − 1)x23 = 1
√
If we replace c by 2c, we have
x22 + 4cx2 x3 + (2c2 − 1)x23 = 1
We will sketch some of these lines in class.
MA243 December 8, 2023 43

Desmos example https://www.desmos.com/calculator/kts5zeg1cb

(Desmos coordintes are x, y,

but we use x2 , x3 for these)
The line L is given by x = 0,
the point P = (1, 0). The hy-
perbolic lines are given by
x22 + 4cx2 x3 + (2c2 − 1)x23 = 1

https://www.geogebra.org/3d/kkhgqxbg

In Geogebra, define a hyper-

boliod, and a Lorentz plane.
These intersect in a hyperbolic
line L
If you pick 5 points on L, and
project to the plane x1 = 0,
then you can use the geogebra
tool to draw a conic through
these points, which is the or-
ange line in this picture.
MA243 December 8, 2023 44

19. Lecture 19: Different Kinds of Intersections of planes with H2

A plane through the origin in R3 which contains a time like vector intersects H2 in a hyperbolic line.
There are some other kinds of planes, with different curves obtained in H2 .
plane in R3 curve on H2
planes through the origin
Lorentz plane hyperbolic line
Plane with no time like vector empty set
planes not through the origin containing time like vector
normal is light like circles
normal is space like hypercycles
normal is time like horocycles
Hypercycles and horocycles both come in sets which are obtained from sets of parallel planes.
Each hypercycles is equidistant to some hyperbolic line, which you might consider them to be parallel to,
in some sense, and each hypercycle generally has different amounts of curvature.
Horocycles are all congruent to each other, but not to any hyperbolic line. They are also somewhat
similar to sets of parallel lines, but they are not straight lines, and not equidistant to any hyperbolic line.

A geogebra example: https://www.geogebra.org/3d/bmcussgf

This image shows a hyperbolic circle

Hyperbolic isometries of H2
Recall that if A is in O+ (1, n), then A defines an Isometry of Hn . This means that AT JA = J and TA
maps H2 to H2 .
We have not yet proved that all isometies of Hn have this form, but this does turn out to be true.
Almost every Lorentz transformation of H2 is given by one of the following matrices, up to a Lorentz
orthonormal change of basis, i.e., from the standard basis to another Lorentz orthormal basis.
MA243 December 8, 2023 45

matrix description fixed points eigen values picture – projection to x2 , x3 plane

Lorentz x3
translation
(by β
on the x2
axis); ori- x2

cosh(β) sinh(β) 0
 entation
 sinh(β) cosh(β) 0 preserving none (in H2 ) 1, exp(±β)
0 0 1 (direct);
hyperbolic

x3

Lorentz
glide
x2
  (in x2 direc-
cosh(β) sinh(β) 0
 sinh(β) cosh(β) 0  tion); ori- none (in H2 ) −1, exp(±β)
entation re-
0 0 −1
versing (in-
direct / op-
posite)

x3
rotation
(about the
point
(1, 0, 0));
  orientation x2
1 0 0 preserving
0 cos(θ) − sin(θ) one point 1, exp(±iθ)
(direct);
0 sin(θ) cos(θ) elliptic

x3
reflection
(in the line
x2 = 0);
orientation
x2
 
1 0 0 reversing
0 −1 0 a line 1, 1, −1
(indirect)
0 0 1

Note, each of the above matrices has an eigen value ±1 with an eigen vector which is space-like or
time-like. There are elements of O+ (1, 2) with light-like eigen vectors with eigen value 1, such as
MA243 December 8, 2023 46

 3
− 12 −1
 These correspond to “parabolic” transformations of H2 , which preserve the orien-
2
 1 1
−1. tation, and don’t fix any points of H2 , but whereas a Lorentz translation fixes a
2 2
−1 1 1 space-like vector when extended to R3 , parabolic transformations fix a light-like
vector.
Proof: See [RS, 3.11(page 47) and 3.16(page 58) and B.3 page 189], and the RS erratum
You are expected to be able to
• Check all the above are elements of O+ (1, 2).
• given an element of O+ (1, 2), determine which kind of transformation it corresponds to, e.g., by
considering fixed points and eigen values
• sketch images of simple shapes e.g., lines, points, triangle, under the action of transformations
• compute the composition of hyperbolic isometries
• find the matrix for a hyperbolic isometry given a description

Lorentz orthogonal matrices

Definition 31. A set of vectors v1 , . . . , vn ∈ Rn is Lorentz orthogonal if hvi , vj iL = 0 for i 6= j.
It is Lorentz orthonormal if 
0

 if i 6= j
hvi , vj iL = 1 if 2 ≤ i = j ≤ n

−1

if i = j = 1
Lemma 24. (1) The canonical basis e1 , . . . , en is Lorentz orthonormal.
(2) If v1 , . . . , vn are Lorentz orthonormal, then they form a basis of Rn .
(3) v1 , . . . , vn is a Lorentz orthonormal basis if and only if hvi , vj i = Ji,j , the element in the ith
row, j th column of J.
Proof. Exercise. 
For the Euclidean norm, the orthonormal vectors were on a sphere. For the Lorentz metric, we have
the first element of a Lorentz orthonormal basis is in Hn or in −Hn . The other elements of a Lorentz
orthonormal basis satisfy
hx, xiL = 1 = −x21 + x22 + · · · + x2n ,
which means they lie on another hyperoloid, as in the figure:

Suppose x1 , x2 is a Lorentz orthnormal basis for R2

x1 surfaces where elements of a
1 Lorentz orthonomal basis of
H
R3 live:
x1 lies on H1 or
−H1

x1

x2
x2

x2 lies on another
hyperbola
MA243 December 8, 2023 47

20. Lecture 20: Sketch proof that Isom(H2 , dH2 ) ∼

= O+ (1, n)
We won’t have time to prove this result. We already know that O+ (1, n) ⊂ Isom(H2 , dH2 ), so we just
need to prove the reverse. An outline is as follows:
(1) All linear Lorentz transformations correspond to elements of O(1, n).
(2) All Lorentz transformations are linear
(3) Any Isometry of Hn extends to a Lorentz transformation of Rn+1 .
(1) is almost by definition.
(3) can be acheived by using the fact that Hn does contain a basis for Rn+1 , so we can extend linearly,
and just have to check it actually works.
We don’t have time to prove all these steps in detail, but we can do (2). See course text book [R] or last
year’s notes for (1) and (3).
Proposition 12. If T : Rn → Rn is a Lorentz transformation then
(1) the image of the canonical basis T (e1 ), . . . , T (en ) is Lorentz orthonormal.
(2) T is linear.
Proof. ⇒ Suppose T is a Lorentz transformation.
hT (ei ), T (ej )iL = hei , ej iL since T is a Lorentz transformation
= Ji,j by Lemma 24, which implies T (e1 ), . . . , T (en ) is Lorentz orthonormal.
Pn Pn
Suppose x = i=1 xi ei ∈ Rn , and T (x) = i=1 bi T (ei ). By orthnormality of the ei and the T (ei ),
−b1 = hT (x), T (e1 )iL = hx, e1 iL = −x1
bi = hT (x), T (ei )iL = hx, ei iL = xi for 2 ≤ i ≤ n
Thus bi = xi for i = 1, . . . , n so T is linear.

Corollary 7. A Lorentz transformation is bijective.
Proof. This is because a Lorentz transformation is linear and maps a basis to a basis. 
Assuming that we now know that Lorentz transformations correspond to elements of O(1, n), to finish
off our proof that Isom(H2 , dH2 ) ∼
= O+ (1, n) we need to know which elements of O(1, n) preserve the sign
of x1 .
Lemma 25. If A ∈ O(1, n) then AT ∈ O(1, n).
Proof. A ∈ O(1, n) ⇒ AT JA = J ⇒ JAT JA = I ⇒ (JAT J)−1 = A ⇒ A(JAT J) = I ⇒ AJAT J =
I ⇒ AJAT = J ⇒ AT ∈ O(1, n). 
Proposition 13. For A = (ai,j ) ∈ O(1, n), we have A ∈ O+ (1, n) ⇐⇒ a1,1 > 0.
Proof. Let A have columns a1 , . . . , an .
⇒: If A ∈ O+ (1, n), then T (e1 ) = a1 = (a1,1 , a2,1 , . . . , an,1 ) ∈ Hn , so a1,1 > 0.
⇐: Denote the first column of AT by a10 . By Lemma 25, AT ∈ O(1, n), and so by definition of O(1, n),
a10 is in Hn or in −Hn . Since a1,1 > 0, we have a10 ∈ Hn . The first element of the vector Ax is
ha10 , xi = β ha10 , x/βi, where this is the usual inner product, and β = −i||x||L > 0. Since x/β ∈ Hn , by
Lemma 20 ha10 , x/βi is positive, so Ax is also positive, so we are done. 
MA243 December 8, 2023 48

Loose end: In our description of hyperbolic lines, we used that Lorentz planes are orthogonal to space
like vectors. We will now prove this.
Lemma 26. If w is Lorentz orthongonal to a time-like vector, then w is space like. In other words: If
x ∈ Rn with hx, xiL < 0, and w ∈ Rn \ {0} with hx, wiL = 0, then hw, wiL > 0.
Proof. Let x = (x1 , x2 , . . . , xn ) and w = (w1 , w2 , . . . , wn ). Let x0 = x − x1 e1 and w0 = w − w1 e1 .
Since hx0 , x0 iL ≥ 0 and hx, xiL = −x21 + hx0 , x0 iL , we must have x1 6= 0. If w1 = 0, then since w 6= 0, we
have hw, wiL = hw0 , w0 iL > 0, and we are done. So we may assume w1 6= 0. The vector x1Dw − w1 xE has
zero e1 component. This vector can not be zero, or x1 w = w1 x, but then 0 > hx, xiL = x, wx11 w =
L
x1
w1 hx, wiL = 0, a contradiction. Now

0 ≤ ||x1 w − w1 x||2L = hx1 w − w1 x, x1 w − w1 xiL = x21 hw, wiL + w12 hx, xiL
so
w12
hw, wiL ≥ − hx, xiL > 0.
x21 
Examples. In H1 ,
if x = (x1 , x2 ) and w = (w1 , w2 ), then hx, w, iL = 0 means x1 w2 = x2 w2 . For fixed
x this gives an equation for a line through the origin on which w must lie.
x
time-like region

light-like vectors
vectors Lorentz orthogonal
to x lie on this line and are
space-like when x is time-
like. They are orthogonal
space-like region
to Jx.

Jx
MA243 December 8, 2023 49

21. Lecture 21: Triangle Inequality (for H2 ) R

21.1. Hyperbolic triangles.
Definition 32. A hyperbolic triangle, denoted ∆PQR, consists of three distinct,
P
non-collinear points, P, Q, R on Hn , and the finite hyperbolic line segments joining
each pair of points, and the finite area enclosed by these lines.
Note, collinear is used to mean points are contained in the same hyperbolic line. Q
21.2. Hyperbolic angles: Motivation from spherical geometry. In order to prove the triangle
inequality, we will use angles in triangles. We are not going to give a proper definition of angle in this
course. (See course text book [R] (Ratcliffe) for comprehensive definition.)
Problem: Our model of hyprbolic space does not define a conformal embedding of H2 in R3 . Conformal
means angle preserving.
We haven’t even defined what angles are, but you will have to trust that whatever they are, they are not
preserved by stretching. Example:
 
1 0
The matrix does not preserve angles
0 2

So we can’t just measure hyperbolic angles by taking the dihedral angle between corresponding planes.
What do we do?
Recall
hP × R, P × Qi
For a spherical triangle: cos(a) = .
||P × R||||P × Q||
By analogy,
hP ×L R, P ×L QiL
(11) For a hyperbolic triangle: cos(a) = .
||P ×L R||L ||P ×L Q||L
It turns out the is invariant under Lorentz transformation. In this expression,
Definition 33. Let J = J3 For x, y ∈ R3 , the Lorentz cross-product of x, y is
−x2 y3 + x3 y2 −e1 e2 e3
   

x ×L y := J · (x × y) = x3 y1 − x1 y3
  = (Jy) × (Jx) =  x1 x2 x3 
x1 y2 − x2 y1 y1 y2 y3
Note about invariance:
Lemma 27. (Binet–Cauchy identity.) For x, y, z ∈ R3 ,
hx × y, z × wi = hx, zi · hy, wi − hx, wi · hy, zi
The following is a version of this identity for the Lorentz inner product, where the difference with the
usual formula is highlighted in red.
Lemma 28. (Binet–Cauchy identity for Lorentz inner product.) For x, y, z ∈ R3 , we have
hx ×L y, z ×L wiL = − hx, ziL · hy, wiL + hx, wiL · hy, ziL
This result tells us that the expression on the right hand side of Equation 11 is invariant under Lorentz
transformations, since the Lorentz inner product is.
Suppose x = (1, 0, 0), then we have
−e1 e2 e3 −e1 e2 e3
     
e1 e2 e3
x ×L y =  x1 x2 x3  =  1 0 0 = 1 0 0 =x×y
y1 y2 y3 y1 y2 y3 y1 y2 y3
So the angles are the same as in Euclidean space when x = 0.
So, we’re going to translate P to (1, 0, 0) by a Lorentz transformation so we can measure the triangle
angle at P.
MA243 December 8, 2023 50

Lemma 29. For points x, y, z in H2 , there is some Lorentz transformation T such that for some θ, β and
γ we have
Tx = ( 1, 0, 0)
T y = ( cosh(β), sinh(β), 0)
T z = ( cosh(γ), cos(θ) sinh(γ), sin(θ) sinh(γ))

Proof. Recall from our parameterisation of H2 , we know that we can write

x = (cosh(δ), cos(c) sinh(δ), sin(c) sinh(δ))

Let
   
1 0 0 cosh(w) sinh(w) 0
Aφ =  0 cos(φ) − sin(φ)  and Bw = sinh(w) cosh(w) 0

0 sin(φ) cos(φ) 0 0 1
These are hyprbolic rotation and translation matrices.
We have A−c x = (cosh(δ), sinh(δ), 0).
And B−δ A−c x = (cosh(0), sinh(0), 0) = (1, 0, 0).
Now we have that B−δ A−c y = (cosh(β), cos(b) sinh(β), sin(b) sinh(β)) for some β and b.
Now apply A−b , which fixes (1, 0, 0), and all the images of x, y, z under T = A−b B−β A−c have the required
form.


Proof of the triangle inequality

Theorem 10. (Main theorem of hyperbolic trigonometry) For distinct points x, y, z in Hn , let

α = dHn (z, y)
β = dHn (x, z) x
γ = dHn (x, y)
a = the hyperbolic angle of ∆xyz at x. a
Then β γ
z
cosh α = cosh β · cosh γ − sinh β · sinh γ · cos a α
y
x1

sinh(γ)
 
cosh(γ),
z = cos(a) sinh(γ),
sin(a) sinh(γ)) sinh(β)
 
cosh(β),
γ y =  sinh(β) 
β 0
x = (1, 0, 0)
x2
cut out and fold along x1 axis; a is the angle between the two sides of the paper.
See how the distance between z and y varies as this angle varies.
MA243 December 8, 2023 51

22. Lecture 22: dHn is a metric, and intro to projective space

Triangle inequality in hyperbolic space
Proof of Main theorem of hyperbolic trigonometry. Since angles and lengths are invari-
ant under Lorentz transformations we may assume x, y, z have the form as in Lemma 29.
x =( 1, 0, 0)
y = ( cosh(β), sinh(β), 0)
z = ( cosh(γ), cos(θ) sinh(γ), sin(θ) sinh(γ))
Since we are in the special case where x = (1, 0, 0), the angle between the sides xz and xy is the angle
between the corresponding planes, which is a = θ.
Now we can compute α, β, γ in terms of the coordinates of x, y, z:

− hx, yiL = cosh(β)

− hx, ziL = cosh(γ)
− hz, yiL = cosh(β) cosh(γ) − cos(a) sinh(β) sinh(γ)
Combining these gives the required result:
cosh(α) = cosh(γ) cosh(β) − sinh β · sinh γ · cos a

Corollary 8. (The triangle inequality for the hyperbolic metric.) For x, y, z in Hn , we have 

dHn (y, z) ≤ dHn (y, x) + dHn (x, z)

Proof. Use notation of Theorem 10. Since cos(a) ≥ −1
cosh(α) ≤ cosh(γ) cosh(β) + sinh β · sinh γ = cosh(β + γ).
Since cosh is increasing on [0, ∞), this implies that α ≤ β + γ, the required result. 
Corollary 9. The hyperbolic metric is a metric.
Proof. Recall that
dHn (x, y) = cosh−1 (− hx, yiL ),
which we showed was well defined in Lemma 20.
Non degeneracy: If dHn (x, y) = 0, then − hx, yiL = 1. In Lemma 20, we showed that − hx, yiL ≥ 1. Since
Lorentz transformations preserve the Lorentz inner product, and we saw there is a Lorentz transformation
mapping x and y to the form in Lemma 29, where x = (1, 0, 0) and − hx, yiL = cosh(β) which is only 1
if β = 0 and x = y.
Symmetry: This is immediate from the fact that the Lorentz inner product is symmetric on Rn+1 .
Triangle inequality: This is Corollary 8.
Hence dHn satisfies the requirments to be a metric. 

Next: projective geometry!

(this is a wikipeadia picture from a page about emission theory, so not really about projective
geometry, but it reminded me of projective geometry).
MA243 December 8, 2023 52

Projective geometry
We take 3 views of this subject: Algebraic; Classical perspectivities; Combinatorial/Axiomatic.

22.1. Algebraic Definition of Projective Space.

Definition 34. Let V be an n+1 dimensional vector space over a field k, so V ∼ = k n+1 . Let V ∗ = V \{0}.
(This definition also works with V = A n+1 for any division ring A. A division ring is a ring where ax = b
has a solution x ∈ A for any non-zero a, b ∈ A. Most of the time we take k = R.)
Define an equivalence relation on V ∗ , by defining for x, y ∈ V ,
x ∼ y ⇐⇒ kx = ky.
Exercise: check that this is an equivalence relation, namely reflexive, transitive and symmetric.
The projective space of V is the set of equivalence classes in V ∗ under this relation:
Pn (k) := P(V ) := V ∗ / ∼
For x = (x1 , x2 , . . . , xn+1 ) ∈ V ∗ , we denote the equivalence class of x by
[x] = (x1 : x2 : · · · : xn+1 ) = {v ∈ V ∗ : v ∼ x}
When k = R, we write Pn = P(Rn+1 ), and call this n-dimensional real projective space.
Lemma 30. There is a bijection between P(V ) and the set of lines through the origin in V .
For x, y ∈ V ∗ , x ∼ y ⇐⇒ x = λy for some λ ∈ k \ {0}.
Proof. Exercise. 

Each line is an equivalence class of

points in P2 . The thick black line is
the class
picture of R2 [(1, 2)] = (1 : 2) = {(λ, 2λ) : λ ∈ R}
imagined as a
square
0 removed

The projective line P1 (R)

22.2. Equivalence class representatives for the projective line. Consider the red line in this
figure, given by y = 1. Then (nearly) every equivalence class of lines in P2 intersects this line exactly
once. E.g., the line [(1, 2)] intersects at ( 12 , 1). All points except (1 : 0) can be written in the form [(x, 1)]
for some x ∈ R. So P1 = {[(x : 1)] : x ∈ R} ∪ {(1 : 0)}, and we can think of (1 : 0) as a “point at infinity”.

We can take the point

( 21 , 1) to represent the
line y = 2x
y=1
These points, and all
points on the line
y = 2x project to ( 12 , 1) the line y = 0 does not
on the line y = 1. intersect y = 1
all other lines through
the origin do
MA243 December 8, 2023 53

22.3. Semi-circle identified with P1 .

Points on the semi-circle corre-
spond to lines through 0. E.g.,
the point at (cos(θ), sin(θ)) cor-
responds to the line y = tan(θ)x

In the above picture, the red and white dots at the end of the semi circle correspond to the same line,
y = 0, so these points are identified. Topologically (in topology, you can bend and stretch but not tear
a geometric object), P1 is a circle, which you can see after identifying these points, i.e., stretching round
and gluing together as in the next picture. We can also think of the {x = 1} and {y = 1} projections of
P1 as real lines that together cover the whole of of P1 , as in the following figure:

The red line represents all of P1 .

The blue line represents the part of P1 that projects to {y = 1}
The green line represents the part of P1 that projects to {x = 1}
This is a schematic diagram, only topologically equivalent to P1 .
MA243 December 8, 2023 54

23. Lecture 23: The projective plane; real and Fano.

The projective plane P2 (R)
Every point in P2 can be represented by a point on S 2 , but this includes each point twice. So we only
need the upper hemisphere Z ≥ 0. And we only need one half of the S 1 = S 2 ∩ {z = 0}.
Here is a Geogebra picture https://www.geogebra.org/3d/katc2xrq

Every point (a : b : c) in P2 is
represented by unique point
on either

 the plane Z = 1, if c 6= 0


the line Z = 0, Y = 1 if c = 0, b 6= 0

 the point (1, 0, 0) if b = c = 0


line at ∞

R2

R2 R point
z }| { z }| { z }| {
P2 ∼ {Z = 1} ∪ {Z = 0, Y = 1} ∪ {(1, 0, 0)}
= point at ∞
on line at ∞
You can think of Pn as Rn with a point added at ininfity for every radial direction in Rn .

Topology of the projective plane P2 (R)

If you take the hemisphere, and distort it to “glue together” points on opposite
side of the circular boundary on Z = 0, which lie on the same line in R3 , you
get a closed surface. This surface only can be viewed in R3 if it crosses itself.
One model is called “Boy’s surface”.
(Picture from Wikipedia, Ggleizer https://commons.wikimedia.org/wiki/
File:BoysSurfaceKusnerBryant.svg)

Lines and linear subspaces in projective geometry

For a vector space V , over a field k, there is a projection map from V ∗ = V \ {0} to P(V ) given by
π : V ∗ → P(V )

v 7→ kv
Definition 35. If V is a vector space V , with a k + 1 dimensional vector subspace W , then the
k-dimensional projective linear subspace P(W ) of P(V ) is the image under π of W in P(V ), that is
P(W ) := {x ∈ P(V ) : x ⊂ W } = π(W )
Warning: the dimension of P(W ) is one less than the dimension of W .
Definition 36. Let W be a subvector space of V .

 point if dim(W ) = 1
P(W ) is a line if dim(W ) = 2
plane if dim(W ) = 3

MA243 December 8, 2023 55

Lecture 23/24: The Fano plane and Projectivities

23.1. Fano Plane. In this section, we discuss the smallest possible projective plane.
Let V be the vector space (F2 )3 over the finite field of two elements, F2 = {0, 1}. What are the points
and lines in P(V )?
The elements of V are (0, 0, 0), (0, 0, 1), (0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1), (1, 1, 0), (1, 1, 1), which can be
pictured as points at the vertices of a cube:
(0, 1, 1) (1, 1, 1)

(0, 0, 1) (1, 0, 1)

(0, 1, 0) (1, 1, 0)

(0, 0, 0) (1, 0, 0)

The points of P(V ) are given by the one dimensional subspaces of V , which have the form F2 v = {0, v}
for non zero v. So there are 7 of these.
Lines are given by the two dimensional subspaces. Each of these has the form {0, v, w, v + w}
for some
v, w ∈ V . For example, the plane P = {(0, 0, 0), (0, 1, 0), (1, 0, 1), (1, 1, 1)}. So, there are 72 ways to
choose two points w and v, but any three non-zero points in P span P , so this would count the planes 3
1 7


times, so the number of planes is 3 2 = 7.
It’s not a coincidence that this is the same as the number of points, since each plane can also be defined
by the vector it is orthogonal to, e.g., P above is orthogonal to (1, 0, 1).
We can draw a diagram showing the points and lines as follows. Right shows labels for the points, with
a vector in V which generates each point in P(V ). This is called the Fano plane.
(0:1:1)

(1:0:1) (0:1:1)
(1:1:1)

(1:0:1) (0:0:1) (1:0:0)

23.2. Working in other finite fields. If we work in the field Fp = Z/pZ, which has p elements,
each line in Fnp has elements {0, v, 2v, 3v, . . . , (p − 1)v}, i.e., There are p − 1 non-zero points on any line.
Since Fnp \ {0} has pn − 1 points, this means it contains (pn − 1)/(p − 1) distinct lines, which are points
of Pn−1 (Fp ).
To determine the number of lines in Pn−1 (Fp ), we have to count the number of planes in Fnp . Each plane
is spanned by 2 a basis of non-zero vectors, u, v, which would be (pn − 1)(pn − p) sets of ordered basis,
because once the first, u is choosen, that rules out p choices for v, since v can not be a multiple of u.
but each plane would be counted multiple times. The number of times a plane is counted is equal to the
number of different choices of (ordered) basis. Once one non-zero basis vector is chosen, there are are
p2 − p choices for the remainder, so that’s a total of (p2 − 1)(p2 − p) possible basis. So this means there
n −1)(pn −p)
are (p
(p2 −1)(p2 −p)
lines.
Definition 37. For a vector space V and a subset Σ ⊂ P(V ), the projective cone of Σ is given by
[
Σ
e := L = {v ∈ V : v ∈ L ⊂ V for some line L ∈ Σ},
L∈Σ
i.e., elements of Σ may be considered to be lines, and we take their union.
The linear span or span of Σ is is the smallest projective linear subspace of P(V ) containing Σ,
hΣi := P(span(Σ)),
e
MA243 December 8, 2023 56

24. Lecture 24: Projective transformations and projective frames of reference

Perspectivities
Definition 38. A perspectivity f is a map between two distinct hyperplanes (linear subspaces of dimension
n−1), Π1 and Π2 in Pn , given by projection from a point O 6∈ Π1 ∪Π2 . That is, if P ∈ Π1 , then f (P ) ∈ Π2
is the point such that 0, P, f (P ) all lie on the same line.
f (P )
L2

f (Q) R

f (R)

Q
P

L1 O

(Next week we will prove this is well defined, i.e., there is a unique intersection point.)

Projective linear maps

Definition 39. The real projective general linear group P GL(n) = P GL(Rn ) is the group of invertible
real matrices up to scalar muliplication, that is the equivalence classes of the set of matrices in GL(n)
with the relation
A ∼ B ⇐⇒ A = λB for some λ ∈ R∗
Example, in P GL(2), we have    
1 2 2 4
∼
3 4 6 8
Definition 40. For A ∈ P GL(n + 1) the map
TA : Pn → Pn
[v] 7→ [Av]
is called a projective transformation, or a projectivity or a projective linear map.
TA is well defined, since for x, y ∈ Rn+1 , if x ∼ y, that is, if [x] = [y], then x = λy for some λ ∈ R, so
Ax = Aλy = λAy, so [Ax] = [Ay].
Also, if A ∼ B, then TA = TB , since if A = λB, then [Ax] = [λAx] = [Bx]. So, projective linear maps of
Pn correspond to elements of P GL(n + 1).

Big result
Perspectivities are the same thing as projective linear maps.
MA243 December 8, 2023 57

Projective basis frame of reference

Pn is not a vector space, so it doesn’t have a basis, but it’s very close to a vector space. Instead of a
basis, we have a projective frame of reference.
Definition 41. For a point P ∈ P(V ), the “lift” of P , is a non-unique vector Pe is given by:
Pe = any choice of non zero v ∈ P
For s + 1 points P0 , P1 , . . . , Ps in P(V ), the dimension of h{P0 , P1 , . . . , Ps }i = π(Span(P
f0 , . . . , P
f0 ) is at
most s, (by definition of projective dimension).
We say P0 , P1 , . . . , Ps are linearly independent if dim h{P0 , P1 , . . . , Ps }i = s.
Definition 42. A projective frame of reference for Pn is an ordered set of n+2 points, P0 , P1 , . . . , Pn+1 ∈
Pn , any n + 1 of which are linearly independent (and so span Pn )
Definition 43. The standard frame of reference for Pn is given by [e1 ], . . . , [en+1 ] together with [e1 +
· · · + en+1 ], where ei are the standard basis for Rn+1 .

semi-circle of representatives of P1

[e2 ]
Standard frame of reference points for P1
[e1 + e2 ]

[e1 ]

Example: Any three distinct points P1 , P2 , P3 in P1 is a frame of reference for P1 , since being distinct
means that P f1 , P
f2 , P
f3 are pairwise linearly independent.
Motivation: Why do we need n + 2 points? This is because n + 2 points is enough to determine a
projecive linear map of Pn , but n + 1 is not. This might be suprising, since the images of n points is
enough to determine a linear map on Rn .
Example: There are infinitely many projective linear maps P1 → P1 with
(1 : 0) 7→ (1 : 0)
(0 : 1) 7→ (0 : 1)

namely, any diagonal map, of the form Example: The radial projec-
4 4

α 0
 tion, with centre 0, from the
A= line x = 1 to the line x = 3
0 β 3 3 takes a point with y = 1 to
However, these maps are not all the same a point with y = 3. We can
map, since we have 2 2 paramaterise points on these
A : (1 : 0) 7→ (1 : 0) lines by their y-coordinates.
A : (0 : 1) 7→ (0 : 1) 1 1 If we complete these lines to
form copies of P1 , with point
A : (1 : 1) 7→ (α : β) with parameter y maping to
0 0
x=1

x=3

If we fix the image of (1 : 1), then we (y :1), then


the map is given
uniquely determine the projective linear 3 0
−1 −1 by : P1 → P1
map. 0 1
1
Example: The map t 7→ t from R to R can be extended to a map on P1 , given by
       
t 0 1 t 1
7→ = ,
1 1 0 1 t
MA243 December 8, 2023 58

extended to a map on P1 by allowing the map to apply to (1 : 0). This map has the effect of switching 0
and ∞, thought of as the points (0 : 1) and (1 : 0).
4
 
0 α
Any map : P1 → P1 , for α, β 6= 0 also −1 0 1 2 3 4
β 0 3
switches (0 : 1) and (1 : 0), and corresponds to y=3
α
t 7→ βt . red line, → blue line,
2
By specifying the image of (1 : 1), the map is {x = 1} → {y = 3}
uniquely determined. 0 7→ ∞
1
The diagram on the right shows a projection from ∞ 7→ 0
{x = 1} to {y = 3}, which corresponds to the map 1 7→ 3
0
7→ 3t
 
0 3 t

x=1
: P1 → P1 sending (1 : 1) 7→ (3 : 1)
1 0
−1
4 Example: In the figure on the right, we
−1 project from the line {x = 1} to the line
y= {y = − 12 x + 3}.
−1 0 3
2x+ 1 We parameterise {x = 1} by the y coor-
red 3
→ blue 2 2 dinate, and {y = − 21 x + 3} by the x co-
line line ordinate, though we could choose different
0 7→ 6 3
∞ 7→ 0 parameterisations.
1 4
1 7→ 2 With respect to this parameterisation, as
5
a map from P1 to P1 , this map is given
− 12 7→ ∞ 6
6 0 0 6
x=1

t 7→ 2t+1 by the matrix , and corresponds to

2 1
6
−1 t 7→ 2t+1 as a map on R

The maps in these examples have been written as maps from P1 to P1 , but drawn as maps from R → R.
To extend from R to P1 , we should embed each copy of R in a copy of P1 , which can be acheived by
replacing R2 in these figures by P2 .
Example
In all the above examples, the coordinates could be translated, and we would still have a projection. That
is, you don’t have to project from zero, e.g.:
4 Example: In the figure on the right, we
−1 project from the point (1, 1) from the line
y=
−1 0 3
{x = 2} to the line {y = − 12 x + 3.5}.
x
2 + We parameterise {x = 1} by the y
red → blue 3.5 1
2 coordinate−1, and {y = − 12 x+3.5} by the
line line 2
0 7→ 6 3
x coordinate−1, though we could choose
∞ 7→ 0 different parameterisations.
1 4
1 7→ 2 With respect to this parameterisation, as
O = (1, 1) 5
a map from P1 to P1 , this map is given
− 12 7→ ∞ 6
6 0 0 6
x=1

t 7→ 2t+1 by the matrix , and corresponds to

2 1
6
−1 t 7→ 2t+1 as a map on R
Note, if we were just told that in the above example, 1 (on the red line) maps to 2 (on the blue line) and
0 (on the red line) maps to 6 (on the blue line), by considering the intersection of the lines between these
points, we see the projection must be from (1, 1).
If we consider R2 extended to P2 , by assuming this plane is the Z = 1 plane in R3 , and embedding as
(x, y) 7→ (x : y : 1), then we have O = (1 : 1 : 1).
Quiz: Given this embedding, the blue line in the above extends to the line L2 given by Y = −X/2 + 3.5Z
in P. which point is in L2 , but not in the set of points where Z = 0?
MA243 December 8, 2023 59

An examples of a perspectivity
We will show that any three distinct points P1 , P2 , P3 in P1 can be projected onto any other three distinct
points Q1 , Q2 , Q3 . This is called three transitivity of the action of P GL(2) on P1 . This result will
follow from Theorem 13 which shows that there is some A maping the standard frame of reference to
P1 , P2 , P3 , and B maps the standard frame of reference to Q1 , Q2 , Q3 , then BA−1 maps P1 , P2 , P3 to
Q1 , Q2 , Q3 . −1 0 1 2 3 4 5
Example:
How do I project be-
tween these set of
points???
−1 0 1 2 3 4 5

Algebraic description of problem: Let’s just write the algebra first. I.e., suppose we just want a
projecive linear transformation given by a matrix A ∈ P GL(2), and with

TA (0 : 1) = (1 : 1), TA (1 : 1) = (4 : 1), TA (2 : 1) = (5 : 1)
This means we want A so that
           
0 1 1 4 2 5
A ∼ , A ∼ , A ∼
1 1 1 1 1 1
where ∼ means equal up to non-zero scalar multiplication. i.e., we want some A and some constants
λ1 , λ2 , λ3 with            
0 1 1 4 2 5
A = λ1 , A = λ2 , A = λ3
1 1 1 1 1 1
You can either solve this directly, or by working via the standard frame of reference. It’s probably easier
to work via the standard frame of reference.
This means finding A as a composition of two maps, say B −1 and C, with
B −1 C
{0, 1, 2} {∞, 0, 1} {1, 4, 5}
Or, written as projective points, with the inclusion R → P1 given by t 7→ (t : 1).
B −1 C
{(0 : 1), (1 : 1), (2 : 1)} {(1 : 0), (0 : 1), (1 : 1)} {(1 : 1), (4 : 1), (5 : 1)}
(Note, we could also choose to include R1 in P1 as (1 : t), or many other ways; eventually this will lead
to the same formula for the map on R1 .)
Step 1: Find the matrix B:
Since B maps (1 : 0) to (0 : 1) and (0 : 1) to (1 : 1), it has the form:
 
0 β
B=
α β
Since B maps (1 : 1) to (2 : 1), we require
3
x=0

β : α + β = 2 : 1, i.e., 2(α + β) = β, so
red → blue
β = −2α, and the matrix is
2 line line
0 −2 ∞ 7→ 0
 
B=
1 −2 y=1 0 7→ 1
1 1 7→ 2
You don’t have to represent this geomet- −3 −2 −1 0 1 2 3
−2
rically, but one interpretation is as on the t 7→ t−2
right, which is a projection from the point 0
O (t : 1) 7→ (−2 : t − 2)
O = (2, 0) between the lines {y = 1} and (t : s) 7→ (−2s : t − 2s)
{x = 0}.
MA243 December 8, 2023 60

Step 2: Find the matrix C:

To project [e1 ], [e2 ], [e1 + e2 ] to (1 : 1), (4 : 1), (5 : 1),
red we need
 a matrix ofthe form
→ blue

5 1 4 α 0
line line A= , since then TA ([e1 ]) = (1 : 1),
∞ 7→ 1 1 1 0 β
−1 0 7→ 4 and TA ([e2 ]) =(4 : 1). To obtain T ((1 : 1)) =(5:
 A 
40 1 2 3 4 
1 7→ 5 1 5 1 4 α 5
1), we need A = , i.e., = ,
t 7→ t+4 1 1 1 1 β 1
3 t+1
i.e.,
(t : 1) 7→ (t + 4 : t + 1)
−1  
1 −4
       
α 1 4 5 5 1
2 = = = ,
β 1 1 1 −1 1 1 −4
meets red line at ∞ 1 −16
 
1 so the required element of P GL(2) is Note
O 1 −4
0
that we have ignored the determinant
  factor −3,
1 4
projection is from O = (4, 1) when computing the inverse of , since ele-
1 1
−1 ments of P GL(2) are only defined up to multiplca-
tion by non-zero scalar.
Step 3: Compute A = CB −1
Now to map (0 : 1), (1 : 1), (2 : 1) to (1 : 1), (4 : 1), (5 : 1), we take
−1
1 −16 0 −2 1 −16 −2 2
        
−1 14 2 7 1
A = CB = = = =
1 −4 1 −2 1 −4 −1 0 2 2 1 1
So, the final solution to the problem of finding the matrix is:
 
7 1
A=
1 1
The final solution to finding a perspectivity mapping 0, 1, 2 → 1, 4, 5 is the map
1 + 7t
t 7→
1+t
Note, the matrix computation is in P GL(2), so any 7 6
scale factors of the whole matrix are ignored.
We can check the result, e.g., 65
x=1

      0 7→ 1
7 1 2 15 4 5 1 7 4
→
x

TA ((2 : 1)) = = = (5 : 1)
+

1 1 1 3 2 7→ 5
y

3 4
=

Note, although the element of P GL(2) is unique, the (1 : t) 7→ (1 + t : 1 + 7t)

6

2 3
picture is not unique, because we could choose dif- t 7→ 1+7t
1+t
ferent parameterisations, or project from different 1 2
points, to different lines. One example projection
with this algberaic description is to the right. 0 1

In this examples, all we need to know is that a perspectivity is given by a projective linear map. Then
this map can be determined by knowing the images of three points.
You don’t have to draw the pictures. However, if you already know the lines and the point from which
you are projecting, then a diagram may help you find the images of ∞, 0 and 1 (projectively (1 : 0), (0 :
1), (1 : 1)), and so then you only need to compute one matrix, and get A without having to compute two
matrices and multiply.
MA243 December 8, 2023 61

25. Lecture 25: Projective dimension formula

Recall: For a projective linear subspace E ⊂ Pn , with cone Ee ⊂ Rn+1 , we have E = π(E), e and
dim E = dim E − 1.
e
Theorem 11. For projective linear subspaces E and F of P(V ),
dim(E ∩ F ) = dim E + dim F − dim hE, F i
with the convention dim ∅ = −1, and hE, F i is the linear subspace of
{P
g}
P(V ) spanned by E and F .
P Pe F
e
or
F Proof. Let n = dim E, m = dim F and k = dim E ∩ F . Then
plane representative of
part of P(V ) when dim Ee = n + 1 and dim Fe = m + 1. We also have E ^ ∩F = E e ∩ Fe
P(V ) = P2
(proof: exercise). Let v1 , v2 , . . . , vk+1 be a basis for E ∩ F . Then we
e e
can extend this to a basis BE = {v1 , v2 , . . . , vk+1 , vk+2 , . . . , vn+1 } of E, e
and a basis BF = {v1 , v2 , . . . , vk+1 , wk+2 , . . . , wm+1 } of Fe . The vectors
E in B := {v1 , . . . , vn+1 , wk+2 , . . . , wm+1 } clearly span hE, e Fe i. Suppose
Pn+1 Pm+1 Pn+1 Pm+1
i=1 ai vi + i=k+1 bi wi = 0, then i=1 ai vi = − bi wi , so since
Pn+1 Pm+1 Pn+1 i=k+1
a
i=1 i iv ∈ E
e and b
i=k+1 i i w ∈ Fe , we have i=k+1 i i ∈ E ∩ F , and
b w e e
E
e so can be written in terms of the basis v1 , . . . , vk , but since this gives a
relationship between elemtents of the basis BF , this must be the trivial
relationship, since BF is a basis, so all the coefficients must be zero. So B
must be a basis for span(E, e Fe ), and so
v1 ,...,vn+1 wk+2 ,...,wm+1
z }| { z }| {
e Fe )) = (n + 1) + (m + 1) − (k + 1)
dim(span(E,
⇒ dim hE, F i + 1 = (dim E + 1) + (dim F + 1) − (dim E ∩ F + 1)
and so the result follows. 
Remarks: (probably for after lecture reading)
Quiz: What is the corresponding result if E and F are replaced by vector subspaces U and V of Rn ?
Note: We have not defined the concept of dimension for subspaces which are not projective linear
subspaces, and this concept will not be covered in the course. So, while there is a comparable statement
about linear subspaces of Rn , with an inequality instead of equality, we are not going to talk about that.
Example: If E = L1 and F = L2 are lines in P2 , then they have projective dimension 1. In general,
their intersection is a point P = L1 ∩ L2 , which has projective dimension 0. The union spans all of R3 ,
which has dimension 2, so, provided E and F are distinct lines, we get the expected result, as shown in
the above figure.
0 1 1 2
z }| { z }| { z }| { z }| {
dim(L1 ∩ L2 ) = dim L1 + dim L2 − dim hL1 , L2 i
In the figure,(https://www.geogebra.org/3d/vry2xuh5) E is a line in P2 , so it’s drawn as a line in
a plane representing P2 , for psychological reasons. This is similar to how when we work with numbers
modulo 5, we write things like 3, when really we mean {5λ + 3 : λ ∈ Z}
Warning: The union of two vector subspaces is not a subspace. The span hE, F i is not the same as the
union E ∪ F . Because of this, we can’t use an inclusion exculsion formula for the dimension of D ∩ E ∩ F
if D, E, F are projective linear subspaces. There is a formula, but you can’t just automatically assume
it’s the same as the form you may have seen in probability or combinatorics. We can instead write, e.g.,
dim(E ∩ F ∩ D) = dim E + dim F + dim D − dimhF, Di − dimhE, F ∩ Di, and more symmetric, but more
complicated formulas are also possible.
MA243 December 8, 2023 62

Theorem 12. Any two distinct lines L1 and L2 in P2 intersect in a point.

Proof. The lines L1 and L2 are projections of planes through the origin, L f1 and L f2 in R3 , to
P2 .Since L1 and L2 have dimension 2, and are distinct, their intersection must have dimension 1, by
f f
Theorem 11, since the dimension of the span of L f2 is 3, since it can’t be greater, since it’s in R3 ,
f1 and L
and can’t be less, since L1 6= L2 . So the dimension of L1 ∩ L2 is zero, and so L1 ∩ L2 is a point. 

Remarks: (Mostly for after lecture reading, but hopefully there is time in class to write the equations
for the above pictured example, which shows the lines X = Z and X = 2Z, which meet at the point
(0 : 1 : 0).)
Theorem 12 shows that lines always meet in P2 . This can also be seen as a conseqence of the theorem we
proved that spherical lines always intersect, since P2 can be considered to be the sphere S 2 with antipodal
points identified. We can also consider parallel lines in R2 considered as a subset of P2 to meet in the
“line at infinity”, which can be seen by projecting P2 to a different plane: (https://www.geogebra.org/
3d/gnkzhywe)
In this figure, the lines x = 1 and x = 2 are parallel, drawn on a copy of R2 which is the plane Z = 1 in
R3 in this figure. x, y are the coordinates on this plane. These lines are (radial) projections of the planes
X = Z and X = 2Z in R3 , where R3 has coordinates X, Y, Z. These planes project to x = z and x = 2z
on the plane Y = 1, where they meet at (x, y) = (0, 0).
Similarly, lines y = mx + c for fixed m, and varying c, are all parallel in the plane Z = 1. These lines are
projections of Y = mX + cZ in R3 . If we project to X = 1, we get lines y = m + cz, which all meet at
(y, z) = (m, 0). You may also wish to consider these lines as projections to the sphere.
Note: Lines in P3 can be skew; skew lines do not intersect. A corresponding result to Theorem 12 for
P3 would be e.g. that a plane and a line, not contained in that plane, in P3 always intersect in a point,
since their span must be 3 dimensional, so their intersecion has dimension 2 + 1 − 3 = 0, which means
we have a point.
MA243 December 8, 2023 63

26. Lecture 26: Projective frames of reference and projective linear maps
Theorem 13. There is a bijection between projective transformations of Pn and projective frames of
reference of Pn , defined as follows:
φ : P GL(n + 1) → {projective frames of reference}
T 7→ {T ([e1 ]), T ([e2 ]), . . . , T ([en+1 ]), T ([e1 + · · · + en+1 ])}
Proof. We need to show that this map is well defined, injective and surjective.
Well defined: The image of the standard projective frame of reference under T must also be a projective
frame of reference, because T is invertible, so maps any basis of Rn+1 to another basis.
Surjectivity: Suppose we are given a projective frame of reference, P0 , P1 , . . . , Pn+1 . By definition, any
choice of n + 1 of these these are linearly independent in Pn , which means that a choice of lifts, a0 =
P
f0 , a1 = P fn is linearly independent in Rn+1 , and so for any nonzero α0 , . . . , αn , the matrix
f1 , . . . an = P
A with columns αi ai , is invertible.
| | |

...


A = α0 a0 α1 a1 . . . αn an 
| | ... |
We know that Aei is the ith column of A, and so for i = 0, . . . , n,
[TA (ei )] = [Aei ] = [αi ai ] = [ai ] = Pi
We want
n
" !#
X
TA ei = Pn+1 .
i=1
Pn Pn Pn
Let an+1P=P ]n+1 , so Pn+1 = [an+1 ]. We have [TA ( i=1 ei )] = [A( i=1 ei )] = [ i=1 αi ai ], so it’s sufficient
to solve ni=1 αi ai = an+1 , i.e., A(α0 , α1 , . . . , αn )T = an+1 . Since A is invertible, set
α0
 
 α1 
 .  = A−1 an+1 .
 
 .. 
αn
Since a0 , . . . , an+1 is a projective frame of reference, all αi in this solution must be non-zero, since
otherwise there would be a set of n vectors from a0 , . . . , an which was not linearly independent. Hence
A ∈ P GL(n + 1). So φ(TA ) = {P0 , . . . , Pn+1 }. Thus φ is surjective.
Injectivity:
Suppose that φ(TA ) = φ(TB ) for matrices A, B ∈ P GL(n). Then for i = 0, . . . , n, we have
TA (ei ) = TB (ei ),
so the columns of B are multiples of the columns of A, so for some diagonal matrix Diag(β0 , . . . , βn ),
B = A · Diag(β0 , . . . , βn ).
P P
Since φ(TA ) = φ(TB ), we have [TA ( ei )] = [TB ( ei )] so
h X i h X i h X i
A ei = B ei = A βi ei .
ei = A−1 0 = 0,
P P
We know that A isPinvertible, so we can’t have A ei = 0, since then we would have
but we know that ei 6= 0. So for some scalar λ,
X X
λA ei = A βi ei
rearranging, A (λ − βi )ei = 0 and since A is invertible, (λ − βi )ei = 0 which is only possible if λ = βi
P P

for each i = 0, . . . , n, which implies B = λA, and so TA and TB are equal as elements of P GL(n). 
MA243 December 8, 2023 64

27. Lecture 27: Cross ratio

Any three points on P1 can be mapped by a projective linear transformation to any other three points,
so the distance between points under projection is not preserved. But the maps are not totally random.
What characterises a projection, in terms of what is preserved? The answer is the cross ratio.
We will show that the cross ratio of four points has to be the same for it to be possible to map a set of
four points on a line to another set of four points on a line. So the cross ratio is an invariant of projective
linear maps.
R0
Definition 44. Let P, Q, R, S be four distinct points
on a line in Pn . Choose an approrpiate basis such
that P = (1 : 0) and Q = (0 : 1) (followed by n − 1
zeros, which we can ignore). Since Pe and Q
e span the Q=(0:1) R=(1:1) S0
line that R and S lie on, and since R, S 6= P , there
are λ, µ ∈ R such that we can write with respect to
this basis: S=(1: 12 )

R = (1 : λ), S = (1 : µ) P =(1:0) P0
Then the cross ratio is
λ in this example {P, Q; R, S} = 2,
{P, Q; R, S} = which is also the case for the pro-
µ jection to {P 0 , Q; R0 , S 0 }.

Warning: The cross ratio will change when the points are permuted, so it depends on the order P, Q, R, S
are given in.
Note: We could also map P, Q, R to the standard projective frame of reference for P1 , then the cross
ratio is the image of S.
It is convenient not to have to make a change of basis in order to compute the cross ratio, so the following
result is useful:

Proposition 14. The cross ratio is well defined. Moreover, if P, Q, R, S are points in P1 , then they can
be considered as lines in R2 . Let L be any line in R2 not through the origin. Let p, q, r, s be the vector
coordinates of the intersections of P, Q, R, S with L. Then
p−r q−s
  
{P, Q; R, S} = ,
p−s q−r
λv
where the ratios of vectors make sense, because they all lie in the direction of L, and we define v = λ.

P
p
−
r

R
q
−
p
−

r
s

q
− Q
s
L
MA243 December 8, 2023 65

Proof. In order to compute the cross ratio, we apply a change of basis transformation TA , so that
TA (P ) = (1 : 0), TA (Q) = (0 : 1), TA (R) = (1 : λ).
This means that
TA (p) = α(1, 0), TA (q) = β(0, 1), TA (r) = γ(1, λ), TA (s) = δ(1, µ),
and then by definition,
λ
{P, Q; R, S} =
.
µ
Note that any

other change of basis matrix B mapping P and Q to (1 : 0) and (0 : 1) must differ by
a 0
B= A for some a, b. So TB (r) = γ(a, bλ) = γa(1, b/aλ) and TB (s) = γ(a, bµ) = γa(1, b/aµ), so
0 b
the ratio of the second coordinates of r and s is still λ/µ, and so the cross ratio is well defined.
Since p, q, r, s are all chosen to be on a common line L, we must have that for some constants λ1 , λ2 , by
collinearity property of points in R2 , we have
r = λ1 p + (1 − λ1 )q
s = λ2 p + (1 − λ2 )q
Linear maps preserve the relationship between points on lines, so we also have
γ(1, λ) = λ1 α(1, 0) + (1 − λ1 )β(0, 1) = (αλ1 , β(1 − λ1 )) ⇒ λ = (1 − λ1 )/λ1
δ(1, µ) = λ2 α(1, 0) + (1 − λ2 )β(0, 1) = (αλ2 , β(1 − λ2 )) ⇒ µ = (1 − λ2 )/λ2
p−r q−s (1 − λ1 ) λ2 (1 − λ1 )/λ1 λ
  
= = =
p−s q−r (1 − λ2 ) λ1 (1 − λ2 )/λ2 µ


Example: Does there exist a projective linear map taking points P1 , Q1 , R1 , S1 = 0, 1, 2, 3 to points
P2 , Q2 , R2 , S2 = 0, 1, 7, 8, where the point t ∈ R is interpreted as a point (t : 1) in P1 ?
Solution: The cross ratios are
0−2 1−3 4 0−7 1−8 7×7
     
{P1 , Q1 ; R1 , S1 } = = 6= {P2 , Q2 ; R2 , S2 } = =
0−3 1−2 3 0−8 1−7 8×6
Since these are not equal, there is no such projective linear map.
Example: Does there exist a projective linear map taking points P1 , Q1 , R1 , S1 = 0, 1, 2, 5 to points
P2 , Q2 , R2 , S2 = 1, 4, 5, 6, where the point t ∈ R is interpreted as a point (t : 1) in P1 ?
Solution: The cross ratios are
0−2 1−5 8 1−5 4−6 8
     
{P1 , Q1 ; R1 , S1 } = = = {P2 , Q2 ; R2 , S2 } = =
0−5 1−2 5 1−6 4−5 5
So, there is such a map. Note that to compute the map, we only need to consider the images of the first
three points. Given that the cross ratios are equal, the fourth point of one set will automatically get
mapped to the fourth point of the other set.
Example: Determine whether or not there is a projective linear transformation (perspectivity) which
takes the points (1 : 0), (0 : 1), (1 : 1) and (1 : 2) to (1 : 1), (2 : 3), (4 : 5), (1 : 2), preserving the order of
the points. Either prove there is no such map, or find such a map by giving it’s matrix.
Solution: Since the first set of points includes a standard frame of reference, it may be easier just to
compute the matrix which maps (1 : 0), (0 : 1), (1 : 1) to (1 : 1), (2 : 3), (4: 5), and 
see where (1 : 2) is
α 2β
mapped to. We can see that the required matrix must have the form A = , and for TA (1 : 1) =
α 3β
 
2 2
(4 : 5), we must have A = . Now we can check that TA (1 : 2) = (6 : 8) = (3 : 4) 6= (1 : 2), so
2 3
there is no projective linear transformation which takes the points (1 : 0), (0 : 1), (1 : 1) and (1 : 2) to
(1 : 1), (2 : 3), (4 : 5), (1 : 2).
MA243 December 8, 2023 66

We could also have shown this using the cross ratio. By definition, the cross ratio of the first set of
points is {(1 : 0), (0 : 1); (1 : 1), (1 : 2)} = 12 For the second set of points, either we can map the first
3 −2
   
−1 1
three to (1 : 0), (0 : 1); (1 : 1), which is acheived by A , and we have TA−1 (1 : 2) = =
−2 2 2
−1
 
= (1 : −2), so {(1 : 1), (2 : 3); (4 : 5), (1 : 2)} = − 21 .
2
We could also compute the cross ratio geometrically, using Proposition 14, as in the following diagram.
Q−S
Q−R
P −S
P −R

S Q R P

R = (4 : 5)

Q = (2 : 3)

S = (1 : 2)

P = (1 : 1)

(0, 0)
−1 23
!
p−r q−s 2 2 −5 1
       
{P, Q; R, S} = = = =−
p−s q−r 5 1 13 5 4 2
Note, in this computation, I used the line y = 10, but any line would give the same result. I scaled all
points to have the same y = 10 coordinate, e.g., Q = (2 : 3) = (6 23 : 10).

For beginning of lecture 28

Lemma 31. Projective linear maps preserve collinearity. I.e., if P, Q, R ∈ Pn and A ∈ P GL(n + 1), with
corresponding map TA , then P, Q, R are collinear implies TA (P ), TA (Q), TA (R) are also collinear.
Proof. P, Q, R are collinear if and only if the lifts, vectors P̃ , Q̃, R̃ are coplanar, in the sense that
they span a (at most) two dimensional subspace of Rn+1 .
I.e., assuming they are all distinct, for some µ, λ, R̃ = µP̃ + λQ̃. (The case of not being distinct is easy,
since 2 points are always on a line.)
Since A defines an invertible linear map on Rn+1 , we have that TA (R̃) = µTA (P̃ ) + λTA (Q̃).
This implies that T̃A (P ), T̃A (Q), T̃A (R) span a (at most) two dimensional subspace of Rn+1 , and so
TA (P ), TA (Q), TA (R) are collinear.

MA243 December 8, 2023 67

28. Lecture 28: Pappus’ Hexagon Theorem

Pappus’ Theorem has been known since ancient times. It has been important in the development of alge-
braic geometry. Pascal’s theorem (hexagrammum mysticum) is a generalization. For history, variations
and applications see e.g., chapter 1 in “Perspectives on Projective Geometry” by Richter-Gebert, and
https://maa.org/sites/default/files/pdf/upload_library/2/Traves-Monthly-2014.pdf
Theorem 14 (Pappus’ Theorem). Let L, L0 ⊂ P2 be distinct projective lines.
Let P, Q, R ∈ L \ L0 , P 0 , Q0 , R0 ∈ L0 \ L be distinct points.
Then the intersection points
A := hP, Q0 i ∩ hP 0 , Qi, B := hP, R0 i ∩ hP 0 , Ri, C := hQ, R0 i ∩ hQ0 , Ri
R e
are collinear.
n lin
Q mo
om
n ac
o
lie
C
B,
P A,
C
L
B Desmos version:
A https: // www. desmos. com/ calculator/ df9debxsti

L0
P0 Q0 R0

Proof. Since P, Q, P 0 , Q0 are distinct and on two distinct lines, these points must be a projective
frame of reference, and so by Theorem 13 there is a projective transformation which takes these points
to the standard frame of reference for P2 .
Since a projective linear map preserves lines and intersection points, we can now assume
P = (1 : 0 : 0), Q = (0 : 1 : 0), P 0 = (0 : 0 : 1), Q0 = (1 : 1 : 1).
The picture, from https://www.geogebra.org/3d/zhwkackg shows these points on the projective plane
projected to a sphere. The line through P, Q, R is the “line at infinity”, where the sphere intersects the
z = 0 plane.
Since R ∈ hP, Qi, R, being a linear combination of
P and Q, has the form R = (1 : α : 0), where we
may assume the first coordinate is 1, since it can
not be zero, since R 6= Q. If the first coordinate is
non-zero, by scaling we can set it equal to 1. Since
R0 is on hP 0 , Q0 i, R0 has the form R0 = (1 : 1 : β),
where again, the first coordinates can not be zero,
since R0 6= P 0 .
Now compute the lines and intersection points:
hP, Q0 i = {(λ + µ : µ : µ)}

A = (0 : 1 : 1)
hP 0 , Qi = {(0 : λ2 : µ2 )}
hQ, R0 i = {(µ2 : µ2 + λ2 : µ2 β)}

B = (1 : α + β − αβ : β)
hR, Q0 i = {(µ + λ : µα + λ : λ)}
hP, R0 i = {(µ2 + λ2 : µ2 : µ2 β)}

C = (1 : α : αβ)
hR, P 0 i = {(µ : µα : λ)}
Now note that A, B, C all lie on the line α(1 − β)X −
Y + Z = 0. (Also, (β − αβ)Ae + Ce = B e for obvious
lifts.) 
MA243 December 8, 2023 68

Lecture 29: finish off proof of Pappus’ theorem

We didn’t quite complete the proof of Pappus’ theorem.
We found that A = (0 : 1 : 1), B = (1 : α, αβ).
The remaining details: prove that the intersection point of hQ, R0 i and hR, Q0 i is (1 : α + β − αβ : β),
and check A, B, C are on a line.
Simplified diagram:
Z
α)
Z −
= 0 (1
βX = +
X αX
= X
Y α
Y =

P = (1 : 0 : 0) Q = (0 : 1 : 0) R = (1 : α : 0) Z=0

A = (0 : 1 : 1) C = (1 : α + β − αβ : β)
B = (1 : α : αβ)

X=Y
0 0 0
P = (0 : 0 : 1) Q = (1 : 1 : 1) R = (1 : 1 : β)

Z=
βY
Y
=
Z

hQ, R0 i = π(Span((0, 1, 0), (1, 1, β)))

= π(Span((0, 1, 0), (1, 0, β)))
n o 
=π (λ, µ, βλ) ∈ R3 : λ, µ ∈ R )
= {(X : Y : Z) ∈ P2 : Z = βX}

hQ0 , Ri = π(Span((1, 1, 1), (1, α, 0)))

= π(Span((0, 1 − α, 1), (1, α, 0)))
n o 
=π (λ, λα + (1 − α)µ, µ) ∈ R3 : λ, µ ∈ R )
= {(X : Y : Z) ∈ P2 : Y = αX + (1 − α)Z}
The point on hQ, R0 i ∩ hQ0 , Ri must satisty Z = βX and Y = αX + (1 − α)Z = αX + (1 − α)βX, so is
given by C = (X : α + β(1 − α)X : X) = (1 : α + β − αβ : β).
Finally, A, B, C are all on the same line, since it can be verified that they are on the line
α(1 − β)X − Y + Z = 0

MA243 December 8, 2023 69

Lecture 29 continued: Axiomatic projective geometry

Definition 45. An axiomatic projective plane (P, L, I) consists of a set P , the set of points, and a set
L, the set of lines, and a relation I ⊂ P × L, the incidence relation, i.e., how lines intersect.
For p ∈ P and ` ∈ L, we write
p ∈ ` ⇐⇒ (p, `) ∈ I,
and say that p is contained in `.
Notation: for `1 , `2 ∈ L,
`1 ∩ `2 = {p ∈ P : p ∈ `1 and p ∈ `2 }
These sets must satisfy the following axioms:
(1) Every line contains at least distinct three points:
∀` ∈ L∃ distinct x, y, z ∈ P, such that (x, `), (y, `), (z, `) ∈ I
i.e., x, y, z ∈ `.
(2) Every point is contained in at least three distinct lines:

∀x ∈ P ∃ distinct l, m, n ∈ L, such that (x, l), (x, m), (x, n) ∈ I

i.e., x ∈ ` ∩ m ∩ n
(3) Any two points span a unique line:

∀x 6= y ∈ P ∃!` ∈ L such that (x, `), (y, `) ∈ I

(4) Any two distinct lines intersect at a unique point:

∀l 6= m ∈ L∃!x ∈ P such that (x, l), (x, m) ∈ I

Example P2 (F2 ) is the smallest axiomatic projective plane, and has 7 points and 7 lines.

Quiz: Which of the following is an axiomatic projective plane (with usual concepts of points and lines,
and standard definitions of intersections etc):

P2 (R), P3 (R), R2 , R3 , S 2 , S 1 , R, H2 ?

28.1. Projective planes over finite fields. The simplest way to obtain a finite axiomatic pro-
jective plane is to take P2 (Fq ). Here, Fq is a finite field with q elements, which exists for all q which is
a power of a prime, q = pn , for a prime p. In the case q = p, then Fp = Z/pZ, integers mod p. For
n > 1, Fq = Fp [x]/(f (x)) where f (x) is an irreducible polynomial of degree n. I.e., the quotient of the
polynomial ring Fp [x] by a prime ideal generated by f (x). If you’ve not seen this in algebra, then just
work with the case Fp .
Proposition 15. The projective plane P2 (Fq ) for a field Fq with q elements
• has q 2 + q + 1 points.
• There are q 2 + q + 1 lines.
• Each line contains q + 1 points.
• Each point is on q + 1 lines.
• P2 (Fq ) satisfies the axioms of an axiomatic projective plane.
MA243 December 8, 2023 70

Proof. • There are a total of q 3 − 1 points in (F3q )∗ . Each line in (F3q )∗ contains q − 1 points
(all the points on the line in F3q except 0). Lines in F3q only intersect at 0, so there are
q3 − 1
= q2 + q + 1
q−1
lines in F3q , and so this is the number of points in P2 (Fq ).
• The number of lines is equal to the number of points because any line ` in P2 (Fq ) corresponds
to a plane V through the origin in F3q , that is, ` = π(V ), and V = `.
e
Any plane through the origin, V in F3q corresponds to the line which is orthogonal to the
plane. That is, for some w, we have V = (wFq )⊥ , so there is a correspondance, [w] ↔ π((wFq )⊥ ),
and so the number of lines in P2 (Fq ) equals the number of points. This is part of a more general
duality between points and lines in P2 .
• Each plane V through the origin in (Fq )3 contains q 2 − 1 non-zero points, since if v, u is a basis
for the plane, the points are
V = {au + bv : a, b ∈ Fq }.
Within this plane, any line wFq = {aw : a ∈ Fq } through the origin contains q − 1 non-zero
points. Any two lines in (Fq )3 only intersect at 0. So there are q + 1 = (q 2 − 1)/(q − 1) lines
through the origin in each plane V through the originin F3q , which after projectivising become
q + 1 points (of the form [w] for w ∈ V ) on each line ` = π(V ).
• Each point [w] is contained in q + 1 lines. This is because the plane V = (wFq )⊥ in F3q contains
q + 1 lines through the origin. For each of these lines, vFq , we get a plane V = Span(v, w). Note
that since by definition w and v are orthogonal, they must not lie on the same line. So V is a
plane. Since w ∈ V , we get [w] = π(w) ∈ ` := π(V ). So [w] lies on q + 1 points.
• Since q ≥ 1, we have that axioms (1) and (2) hold for P2 (Fq ). So we just need to check axioms
(3) and (4).
• Axiom (3): If [u] and [v] are two distinct points in P2 (Fq ), then the line containing them both
is given by π(V ) where
V = {au + bv : a, b ∈ Fq }.
V is a 2-dimensional subspace, since u and v are linearly independent, (otherwise [u] = [v]), so
π(V ) is a line. Any line must have the form π(V ) for some 2-dimensional subspace V , and any
such space V must contain u and v if π(V ) contains [u] and [v]. So we have Span(u, v) ⊂ V ,
but since u and v span a 2-dimensional space, and V is 2-dimensional, we know from linear
algebra that Span(u, v) = V . Hence the line containing u and v is unique.
• Axiom (4): If `1 and `2 are two distinct lines, then these must have the form `1 = π((v1 Fq )⊥ )
and `2 = π((v2 Fq )⊥ ) for some vectors v1 , v2 . Then the point given by [v1 × vv2 ] is contained
in both these lines. Any such point [w] in `1 and `2 must be determined by the equations
w · v1 = w · v2 = 0. Since `1 and `2 are distinct, v1 and v2 must be linearly independent,
so the two equations defined by w · v1 = w · v2 = 0 are linearly independent. Since F3q is
3 dimensional, by the rank nullity theorem, the set of solutions to w · v1 = w · v2 = 0 is 1
dimensional, spanned by some vector w, which defines the unique point [w] on `1 and `2 , unique
by the 1-dimensionality of this set of solutions.

Note, in the class we only proved the first point; we may skip the rest due to lack of time. However, you
are expected to know this material.
MA243 December 8, 2023 71

Dobble
The game Dobble was invented based on a finite projective plane. Junior Dobble is isomorphic to P2 (F5 ),
and corresponds to a finite projective plane of order 6. Each line contains 6 points, each point is on 6
lines. In the game as sold, one card is missing.

The missing card is:

{dog, cat, camel, fish, ladybird, horse}.

Each line corresponds

to a different animal.

There are 31 different animal symbols, one for each line in the projective space P2 (F5 ).
I have made the following choice of animal to point of P2 (F5 ). A different choice would correspond to an
element of P GL(3, F5 ), which has order 372000.
animal orthogonal vector animal orthogonal vector
dog (0:0:1) duck (1:1:1)
elephant (1:0:0) parrot (1:1:2)
lion (4:0:1) crocodile (1:1:3)
rabbit (2:0:1) octopus (1:1:4)
tiger (3:0:1) shark (1:1:0)
whale (1:0:1) penguin (3:1:1)
kangaroo (0:1:1) zebra (3:1:2)
rooster (0:3:1) crab (3:1:3)
hippo (0:2:1) sheep (3:1:4)
tortoise (0:4:1) bear (3:1:0)
cow (0:0:1) horse (2:1:1)
dolphin (4:1:0) fish (2:1:2)
gorilla (4:1:1) cat (2:1:3)
snake (4:1:2) camel (2:1:4)
frog (4:1:3) ladybird (2:1:0)
owl (4:1:1)
E.g. the “cat” line, consists of points (x : y : z) with y = 2z − 2x, i.e., (x, y, z) · (2, 1, −2) = 0 meaning
“cat” is the line orthogonal to (2 : 1 : −2) ≡ (2 : 1 : 3) ≡ (4 : 2 : 1).
You can work out the point on P2 (F5 ) corresponding to your card by just considering two of the animals
on it. E.g., the card containing gorilla (4 : 1 : 1) and cat (2 : 1 : 3) must be a point (x : y : z) with
4x + y + z ≡ 2x + y + 3z ≡ 0 mod 5. From this we get x ≡ y + z (from first condition) and then
2y + 2z + y + 3z ≡ 3y ≡ 0 (sub in second) so y ≡ 0, and x ≡ z, so (up to scaling) the point is (1 : 0 : 1).
MA243 December 8, 2023 72

missing card:
dog,cat, camel, fish, ladybird, horse

dog, cat, camel, dog, cat, camel, dog, cat, camel,

fish, horse, fish, horse, fish, horse,
ladybird ladybird ladybird

kangaroo

rooster

elephant

rabbit

whale
tiger
lion
hippo

tortiose
cow

‘‘ d
o g ’’ ∞
li n e a t lines of the form {(αs : t : s) : (t : s) ∈ P1 } lines of the form {(t : αs : s) : (t : s) ∈ P1 }

u in
ng

b
pe

cra
dog, cat, camel, dog, cat, camel, dog, cat, camel,
fish, horse, fish, horse, fish, horse,
ladybird ladybird ladybird

ep
be she
ra

ar
go hin

zeb
s a
lp

og ke
ll

pa ro oc

du
do

ri

fr na

rr

ck
ot
l

c
ow

c od op
i le s
t
sh

u
ar
k

lines of the form {(t : t + αs : s) : (t : s) ∈ P1 } lines of the form {(t : αs − t : s) : (t : s) ∈ P1 } lines of the form {(t : 2t + αs : s) : (t : s) ∈ P1 }

(0 : 1 : 0)

(1 : −2 : 0) (0,4) (1,4) (2,4) (3,4) (4,4)

dog, cat, camel,
fish, horse,
ladybird

(0,3) (1,3) (2,3) (3,3) (4,3)

ho fis
rse h

(1 : −1 : 0)
(0,2) (1,2) (2,2) (3,2) (4,2) (1 : 0 : 0)

(0,1) (1,1) (2,1) (3,1) (4,1)

camlady
el bird

(1 : 2 : 0) (0,0) (1,0) (2,0) (3,0) (4,0)

cat

(1 : 1 : 0)

lines of the form {(t : αs − 2t : s) : (t : s) ∈ P1 } interpret (x, y) as (x : y : 1); all computations mod 5

This shows the 6 families of “parallel” lines. Note that “parallel” here means not meeting in the copy of
F25 ⊂ P2 (F5 ) given by (x, y) 7→ (x : y : 1).
MA243 December 8, 2023 73

29. Lecture 30: Finite Axiomatic Projective Geometries

Proposition 16 (Proposition/Definition). A finite axiomatic projective plane has n2 + n + 1 points for
some integer n, and has n + 1 points on each line and n + 1 lines through each point. Such a projective
plane is called a projective plane of order n.
Proof. Step 1: Set up a bijection, (by a “perspectivity”) between points on any two lines:
p
bijection between points
on `1 and points on `2
qi ↔ qi0
q4 `1
q2 q3
q1

q10 q20 q30

q40 `2

If `1 and `2 are two lines, then pick a point p not on `1 or `2 .

Construct p as follows:
• By axiom (4), `1 and `2 intersect at a unique point q. `1
• By axiom (3), `1 and `2 both contain at least three
distinct points, so `1 contains another point, q1 6= q, q1
0
`2 contains a point, q1 6= q. Since q is the unique point
on `1 and `2 , q1 is not in `2 and q10 is not in `1 , so
q
q1 6= q10 .
• By axiom (2) there is a unique line `3 through q1 and q10
q10 .
• By axiom (1) there are at least 3 points on `3 , so there
`2
must be another point p apart from q1 and q10 . Since `3
`3 meets `1 and `2 in unique points q1 , q10 , the point p
is not on `1 or `2 .
Now the bijection is defined as follows: Let q1 , . . . , qN be the points on `1 . Then by axiom (3) there is a
unique line `i joining qi to p. By axiom (4) there is a unique point qi0 on the line `i and `2 , so we map qi
to qi0 . This map must be injective by the uniqueness property, and surjective, since for any point on `2
we can obtain the required point on `2 by the same procedure.
This process works for any pair of lines, so all lines must have the same number of points, N .
step 2: Set up a bijection between points on a line and lines through a point:
For any line `, take a point p not on `, which is possible by the argument in step 1, since there must be
at least two different lines, by axiom (2).
Now any line `i through p intersects ` in a unique point qi , by axiom (4), so we have a map `i ↔ qi ,
which is injective by uniqueness, and surjective by axiom (3).
This means that all points are on N lines.
MA243 December 8, 2023 74

step 3: Compute the total number of points:

p picture indicates lines through

p, and points on them, but not
other lines

Let n = N − 1. So every line contains n + 1 points, and every point is on n + 1 lines.

Now take any point p. Every other point lies on exactly one of the n + 1 lines through p, and each of
these lines contains n points not counting p, so this gives (n + 1)n points not equal to p, so together with
p there are n2 + n + 1 points. 
Example: The Fano plane is a finite projective plane of order 2. P2 (F3 ) is a finite projective plane of
order 3.
Definition 46. We say two axiomatic projective planes (P, L, I) and (P 0 , L0 , I 0 ) are isomorphic if there
are bijective maps
fP : P → P 0 , fL : L → L0 , fI : I → I 0
such that for p ∈ P, ` ∈ L
p ∈ ` ⇐⇒ fL (p) ∈ fL (`)
Conjecture: Every finite projective plane has order q, where q is a power of a prime.
There are finite projective fields of order q for all powers of prime q, namely P2 (Fq ).
However, it’s not true that every finite projective plane is isomorphic to P2 (Fq ) for some finite field Fq
with q = pn elements for some prime p. There are 4 different finite projective planes of order 9. Just one
of these is P2 (F9 ).
Not very much is known about this problem.
See “A001231: Number of non-isomorphic projective planes of order n.” at https://oeis.org/A001231,
for the list that is known; the number of projective planes of order n = 2, . . . , 10:
n 2 3 4 5 6 7 8 9 10
n2 + n + 1 7 13 21 31 (43) 57 73 91 (111)
number of non-isomorphic planes 1 1 1 1 0 1 1 4 0
I have put 43 and 111 in brackets, because there are no projective planes with this number of points. It
is not currently known whether or not there is a projective plane of order 12.
The Bruck–Ryser Theorem says that for a finite projective plane of order q, with q ≡ 1 or 2 mod 4, q
is a sum of two squares.