Affine and Projective Geometry

Adrien Deloro
Şirince ’22 Summer School

These are lecture notes with exercises for a two week course of 12×2 = 24 hours given
in 2016, 2107, and 2022 at the Nesin Matematik Köyü in Şirince, Turkey. They should be
accessible to a 3rd year student in mathematics; syllabus and prerequisites are described
below. I wish to thank Joshua Wiscons for useful conversations, and Fırat Kıyak for his
many questions and suggestions.
If found with flaws or mistakes, this document compiled on 9th August 2022 should
be returned to [email protected]. Please mention the date.

Contents
Introduction and prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Affine planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Projective planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3 The Desargues property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4 A detour through space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5 The coordinatisation theorem: Hilbert’s version . . . . . . . . . . . . . . . . . . . 25
6 The Pappus property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7 Semi-linear automorphisms of vector spaces . . . . . . . . . . . . . . . . . . . . . 37
8 Automorphisms of projective planes . . . . . . . . . . . . . . . . . . . . . . . . . . 42
9 Group-theoretic analysis of Desargues and Pappus . . . . . . . . . . . . . . . . . . 47
10 Artin’s coordinatisation (1): dilations and translations . . . . . . . . . . . . . . . . 53
11 Artin’s coordinatisation (2): the skew-field . . . . . . . . . . . . . . . . . . . . . . 59
12 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Each section corresponds to a lecture of 2 hours, with the introduction fitting into § 1.
The class can be taught in a different order as there are independences between lectures.
§§ 1–6 form the initial socle. Then §§ 7–8 are one block; so are §§ 10–11. Apart from
these dependencies, lectures § 7–11 can be permuted freely.

Introduction
The real plane R2 may be studied with different notions and tools; these are called struc-
tural layers and are conveniently imagined as tracing sheets. The following list is not
comprehensive, and items may be combined or not:
• points and lines;
• coordinates and vectors;
• distances;

1
 • areas;
• the dot product;

• the complex multiplication;

• angles;
• angle measurements.

These lead to distinct mathematical theories, all developed in order to formalise our
intuition of what ‘the physical plane’ is.
In this course we restrict ourselves to the most basic layer where one has only points
and lines. The relevant notion is that of an affine plane. It is just a collection of points
and lines; a point can be on a line or not. Geometric intuition, or physical observation
in our small part of the universe (a sheet of paper can do) will suggest axioms describing
the behaviour. These axioms define abstract affine planes, and the usual affine plane R2
is only one example of such structures. Quickly one realises that projective geometry
behaves in a nicer way. Essentially it amounts to adding ‘points at infinity’ to an affine
plane. The resulting abstract projective planes are more homogeneous, more worth a
mathematician’s time. The course is dedicated to their study.
When compared to analytic number theory, differential geometry or algebraic topo-
logy, our topic is certainly elementary since there is no ‘higher’ structure (no topologies,
no categories, etc.). But the word elementary is no synonym of easy: it refers to a set of
methods, not to a level of difficulty. Incidence geometries were studied, among others,
by Hilbert, E. Artin, Tarski; and are therefore not an unworthy subject. Beyond the first
study of abstract projective planes lies combinatorial geometry, full of challenging open
problems. We will not go that far. Further reading is suggested at the end of the lecture
notes.
The course covers basic material on projective planes. It describes beautiful phe-
nomena. It returns to elementary aspects and therefore teaches one to separate struc-
tural layers. It exemplifies the power of group-theoretic language. And it is not part of
the standard curriculum. For all these reasons it is very suitable for a summer school.

Prerequisites
The class is supposedly accessible to a third year student.

Geometry: High-school level. The use of cartesian coordinates; line equations, the dif-
ference between points and vectors.

Linear algebra: Basic knowledge. Vector spaces and coordinates. We shall need the
(lesser known) basic linear algebra over skew-fields, which works exactly the same
however confusing it might sound at first.
Fields and skew-fields: The definitions. One should be aware of the existence of the
latter as they appear everywhere in the course.

Group theory: Groups will play a prominent role in advanced sections like §§ 7, 8, 9,
10, 11. To follow them one should be fully comfortable with subgroups, normal
subgroups, and group actions (stabilisers, transitivity).

Before we start, recall two definitions and related phenomena.

2
Fields and skew-fields; Wedderburn’s theorem
Definition (field). A field is a structure (F; 0, 1, +, ⋅) where:
• (F; 0, +) is an abelian group (called ‘the additive group’);
• (F ∖ {0}; 1, ⋅) is an abelian group (called ‘the multiplicative group’);

• ⋅ is distributive over +, viz. (a + b) ⋅ c = a ⋅ c + b ⋅ c and likewise on the other side.

Equipped with their natural algebraic operations, Q, R, C are fields. One should
know that there exist finite fields.
Theorem (Galois). For every prime power q = p k , there exists a (unique up to isomorph-
ism) field with q elements. Every finite field has order a prime power.

One should also know that commutativity may be dropped.

Definition (skew-field). A skew-field is a structure (F; 0, 1, +, ⋅) where all axioms of a
field hold, except that ⋅ is not required to be commutative.
An example of a non-commutative skew-field is the quaternion algebra H = R ⊕
Ri ⊕ R j ⊕ Rk, subject to the usual ‘Hamilton identities’. There one has i j = k ≠ − k = ji.
It should be checked or known that H is a skew-field, but not a field. For reference one
may see my 2021 course on quaternions.1
Theorem (Wedderburn). Any finite skew-field is commutative.

We shall not prove this classical number-theoretic result but it considerably clarifies
the picture. Though non-commutative skew-fields are therefore uncommon objects in
general mathematics, they play a central, unavoidable role in the study of affine and
projective planes. This involves doing some linar algebra over skew-fields.
Remark (linear algebra over a skew-field). One has to distinguish between left- and
right-vector spaces, not confusing them. (The risk is serious when V = Fn .) Basic linear
algebra such as elementary elimination theory and dimension theory still apply.
On the other hand, there is no fully satisfactory/fully accepted theory of determ-
inants over a skew-field. Therefore claiming invertibility of a matrix with entries in a
non-commutative skew-field because of its ‘determinant’ is simple nonsense.

1 Affine planes
Abstract. § 1.1 formalises the setting: incidence geometries. In common cases, the
abstract incidence relation can be taken to be set-theoretic membership. § 1.2 intro-
duces affine planes, and the notion of parallelism. § 1.3 associates to any skew-field
F a concrete affine plane A2 (F), which is the usual system of points and lines in F2 .
Not all affine planes are of this form.

1 https://webusers.imj-prg.fr/~adrien.deloro/teaching-archive/

Sirince-Quaternions.pdf

3
1.1 Incidence geometries
The purpose of this course is a description of certain point-line configurations. There
are two natural approaches here:
• work with points only and a ternary collinearity relation (with intended meaning:
‘the three points are on some common line’);
• work with points and lines as objects of distinct types, and a binary incidence
relation (with intended meaning: ‘the point is on the line’).
Though the former option has its virtues, we retain the latter for better readibility.
1.1.1. Definition (incidence geometry). An incidence geometry is a triple Γ = (P , L, I)
where:
• P is a set of objects called ‘points’ ;
• L is a set of objects called ‘lines’;
• I is a relation on P × L called ‘incidence’, which intuitively says when a point is
on a line or not.
Manipulating two distinct sets of objects requires using variables in a consistent
manner. From now on, a, b, . . . , a′ , b 1 , . . . , p, q, . . . will always denote points, while
ℓ, m, ℓ′ , m 1 , . . . will always denote lines. This is implicit in our notation. For instance,
(∀a) means: ‘for any point a. . . ’.
1.1.2. Definition (collinear). Given an incidence geometry, points a 1 , . . . , a n are collin-
ear if there is a line ℓ such that for each k one has a k Iℓ.
The incidence relation I is of an abstract nature; instead of ‘p is incident to ℓ’ one may
be more used to writing ‘p ∈ ℓ′ . Indeed replacing I by ∈ is allowed, up to isomorphism.
This requires a definition.
1.1.3. Definition (isomorphism). Two incidence geometries Γ1 = (P1 , L1 , I 1 ) and Γ2 =
(P2 , L2 , I 2 ) are isomorphic if there are bijections f ∶ P1 → P2 and g∶ L1 → L2 such that:
(∀a)(∀ℓ)(aI 1 ℓ ↔ f (a)I 2 g(ℓ)).
1.1.4. Proposition (reducing to membership). Let Γ = (P , L, I) be an incidence geo-
metry such that any two lines having the same points are actually equal. Then there is
an incidence geometry Γ′ = (P ′ , L′ , ∈) where incidence is set-theoretic membership, and
Γ ≃ Γ′ as geometries.

Proof. The idea is to keep the same points, and write new lines as sets of points. To
each ℓ ∈ L associate S ℓ = {p ∈ P ∶ pIℓ}, a subset of P. Now consider L′ = {S ℓ ∶
ℓ ∈ L}, a family of subsets of P. We claim that Γ = (P , L, I) and Γ′ = (P ′ , L′ , ∈) are
isomorphic.
Let f be the identity function on P and g∶ L → L′ map ℓ to S ℓ . Clearly f is a
bijection and g is surjective. Actually g is injective: if S ℓ = S m then lines ℓ and m have
the same points, so by assumption they are equal. Finally for any (p, ℓ) ∈ P × L, one
has pIℓ iff p ∈ S ℓ . Hence ( f , g) is an isomorphism Γ ≃ Γ′ .

So instead of always saying that ‘p is incident to ℓ’ in Γ one may implicitly work in

Γ′ and use the common phrases: ‘p is in ℓ’, ‘p belongs to ℓ’, ‘ℓ contains p’.

4
1.2 The definition
The definition of an affine plane (Definition 1.2.2 below) requires an auxiliary notion.
1.2.1. Definition (parallel). Two lines ℓ, m are parallel if either ℓ = m or there is no point
incident to both, viz.:
ℓ//m iff [ℓ = m] ∨ [(∀a)¬(aIℓ ∧ aIm)].
(Notice that it captures the intuition of R2 , not of R3 .) This notion plays a role only
when studying affine planes, and disappears when studying projective planes (§ 2).
1.2.2. Definition (affine plane). An affine plane A is an incidence geometry satisfying
the following axioms AP1 , AP2 , AP3 .
AP1 . (∀a)(∀b)[(a ≠ b) → (∃!ℓ)(aIℓ ∧ bIℓ)].
(Meaning: given any two distinct points, there is a unique line incident to both.)
AP2 . (∀a)(∀ℓ)(∃!m)[(aIm) ∧ (ℓ//m)].
(Meaning: through any point there is a unique line parallel to a given line.)
AP3 . There exist three non-collinear points.
1.2.3. Remarks.
• The axiomatisation is in the sole terms of points and lines; words such as angle,
distance, or even between are banned from the vocabulary. They would lead to
other (more powerful and less elementary, but equally interesting) mathematical
theories.
• AP3 removes degenerate configurations, the most extreme of which being P =
∅ = L. See exercice 1.4.3.
• By the axioms and proposition 1.1.4, one may suppose that the incidence relation
is ∈. See exercise 1.4.4.

1.3 The affine plane over a skew-field

We wrote the definition of an affine plane by ‘painting from nature’, viz. by observing
the real affine plane. Our definition captures more than the real case; skew-fields appear
naturally.
• If F is a skew-field, then elements of F2 will be written as rows, not as columns,
viz. as pairs (x, y).
• We use the same pair notation for points and vectors; this results in no confusion.
Recall the following basics of high-school affine geometry.
• Vectors may be added by putting (x 1 , y 1 ) + (x 2 , y 2 ) = (x 1 + x 2 , y 1 + y 2 ).
• Scalars act on vectors by letting λ ⋅ (x, y) = (λ ⋅ x, λ ⋅ y); this turns F2 into a
left-F-vector space.
• Last, vectors act on points: if p = (x 1 , y 1 ) and Ð→
v = (x 2 , y 2 ), let p + Ð
→
v = (x 1 +
x 2 , y 1 + y 2 ). (This may look like the sum of two vectors, but here p is a point, and
there is no addition of points. Our notation simply does not reflect the essential
difference in nature between points and vectors.)

5
1.3.1. Definition (affine plane over a skew-field A2 (F)). Let F be a skew-field. Let A2 (F)
be the following incidence geometry:

• points are pairs (x, y) ∈ F2 ;

• lines are sets of the form {p + t Ð
→
v ∶ t ∈ F} for p a point and Ð
→
v ≠Ð
→
0 a non-zero
vector;
• incidence is set-theoretic membership.

This generalises the familiar case of A2 (R) as it allows skew-fields. One should prac-
tice a little with skew-fields (using quaternions for instance).
1.3.2. Proposition. Let F be a skew-field. Then A2 (F) is an affine plane.
1.3.3. Remark (important). Not all affine planes are of the form A2 (F). The Desargues
property (§ 3) characterises those affine planes of the form A2 (F).

Proof. Points are pairs p = (x, y) ∈ F2 ; lines are sets of the form
Ð→
v = {p + t v ∶ t ∈ F}
ℓ p,Ð
→

for p ∈ F2 and Ð →
v ≠Ð →
0 . Notice that one always has p ∈ ℓ p,Ð →
v.
We must check axioms AP1 , AP2 , AP3 . Call two vectors proportional if both are
non-zero and one (equivalently, each) is a multiple of the other. (Since the field is non-
commutative, we do mean Ð v→2 = λ ⋅ Ð
v→1 with λ on the left. For a left-vector space we
never write ‘vector times scalar’.)
Step 1. Two lines ℓ 1 = ℓ p ,Ð → and ℓ 2 = ℓ p 2 ,Ð Ð→ Ð
→
1 v1 v→2 are equal iff [ v 1 and v 2 are proportional
vectors, and Ð
pÐ →
1 p 2 is a (possibly null) multiple of those].

Verification. First suppose ℓ 1 = ℓ 2 . Since p 2 ∈ ℓ 1 there is t ∈ F with p 2 = p 1 + t Ð v→1 . So

ÐÐ →
p 1 p 2 = t v 1 is a multiple of v 1 . But also p 1 ∈ ℓ 2 so there is s ∈ F with p 1 = p 2 + sÐ
Ð→ Ð
→ v→2 .
Now:
p1 = p2 + sÐ v→2 = p 1 + t Ð
v→1 + s Ð
v→2 ,
which simplifies into t Ð v→ + s Ð
1 v→ = Ð
2 0 . Since neither Ð
→ v→ nor Ð
1 v→ is Ð
2
→
0 , they are propor-
tional.
Conversely we shall prove one inclusion, and the other will follow by symmetry.
So let q ∈ ℓ 1 . By definition there is t ∈ F with q = p 1 + t Ð
v→1 . Now by hypothesis there
are λ, µ ∈ F with Ð pÐ→ = λÐ
p
1 2
→ and Ð
v 2
→ = µÐ
v 1 v→ (we even know that µ ≠ 0). Then:
2

q = p2 − Ð
pÐ → Ð → Ð→
1 p 2 + t v 1 = p 2 + (tµ − λ) v 2 ∈ ℓ 2 ,

proving ℓ 1 ⊆ ℓ 2 . ◇

Step 2. Two lines ℓ 1 = ℓ p ,Ð

→ and ℓ 2 = ℓ p 2 ,Ð Ð
→ Ð
→
1 v1 v→2 are parallel iff v 1 and v 2 are proportional
vectors.

Verification. Suppose ℓ 1//ℓ 2 ; we wish to show that Ð

v→1 and Ð
v→2 are proportional vectors.
If ℓ 1 = ℓ 2 then this follows from the previous claim. So suppose ℓ 1 ≠ ℓ 2 ; since they

6
 are parallel this means that there is no point on both lines. Hence the equation:

p1 + t1 Ð
v→1 = p 2 + t 2 Ð
v→2

has no solution for (t 1 , t 2 ) ∈ F2 .

Work in coordinates, say p i = (x i , y i ) and Ð
v→i = (a i , b i ). So the system of equa-
tions:
x + t a = x2 + t2 a2
(S) ∶ { 1 1 1
y1 + t1 b1 = y2 + t2 b2
has no solution for (t 1 , t 2 ) ∈ F2 . Notice that multiplicative coefficients are on the
right and variables t 1 , t 2 on the left; this is due to non-commutativity of F.
Remember that we want to prove that Ð v→1 and Ð
v→2 are proportional vectors, and
that the assumption is that (S) has no solution. There are three cases, two of which
are equivalent.
Case 1a. Suppose a 1 = 0. Then since Ð v→1 ≠ Ð→
0 one has b 1 ≠ 0. If a 2 ≠ 0 then there is a
unique solution t 2 = (x 1 − x 2 ) ⋅ a 2 to the first equation; now t 2 being known
−1

there is a unique solution in t 1 to the second. This is a contradiction to (S)

having no solution, so actually a 2 = 0. Then since Ð v→2 ≠ Ð
→
0 one has b 2 ≠ 0, and
Ð
→ Ð
→
clearly v 1 and v 2 are proportional vectors.
Case 1b. If b 1 = 0, the desired conclusion follows similarly.
Case 2. Suppose a 1 ≠ 0 and b 1 ≠ 0. Then multiply the first equation on the right by
λ = a −1
1 b 1 , and subtract from the second, getting:

y 1 − x 1 λ = y 2 − x 2 λ + t 2 (b 2 − a 2 λ).

If b 2 − a 2 λ ≠ 0 then there is a (unique) solution in t 2 , which then gives rise to

a (unique) solution in t 1 : so (S) has a solution, a contradiction.
Therefore b 2 = a 2 λ = a 2 a −1
1 b 1 . Finally:

Ð
v→2 = (a 2 , a 2 a −1 −1 Ð
→
1 b 1 ) = a 2 a 1 ⋅ (a 1 , b 1 ) = a 2 a 1 v 1 ,
−1

so Ð
v→1 and Ð
v→2 are proportional vectors. ◇

Step 3. A2 (R) satisfies axioms AP1 , AP2 , AP3 .

Verification. We check them in order.

AP1 . Let p 1 ≠ p 2 be points. Then Ðv =Ð

→ pÐ → Ð →
1 p 2 ≠ 0 and ℓ = ℓ p , Ð
→ clearly contains p 1
1 v
and p 2 . Uniqueness is clear as well.
AP2 . Let p be a point and ℓ = ℓ q,Ðv be a line. Then clearly m = ℓ p,Ð
→ →
v contains p and
is parallel to ℓ; but also uniqueness is clear.

AP3 . Simply take (0, 0), (0, 1), and (1, 0). ◇

This completes the proof.

1.3.4. Remark (the right-version (F)2 A).

7
 • Using the right action, define a geometry (F)2 A with ‘right-lines’ {p+ Ð
→
v t ∶ t ∈ F}.
• Points in A2 (F) and (F)2 A are the same, but lines may differ. Technically, sets
{p+t Ð →
v ∶ t ∈ F} and {p+ Ð →
v t ∶ t ∈ F} need not be equal. (This may seem confusing
at first, but the left- and right-linear structures are simply not the same.)
• (F)2 A is another affine plane: same proof as Proposition 1.3.2, exchanging sides.
• For a commutative field F, left- and right-lines are equal, so A2 (F) = (F)2 A. For
quaternions H, lines differ but A2 (H) ≃ (H)2 A (exercise 1.4.7). For arbitrary F,
planes A2 (F) and (F)2 A need not be isomorphic.

1.4 Exercises
1.4.1. Exercise.
1. Write AP3 in logical form.
2. (Tedious.) Using only points and a ternary relation for collinearity, define parallel-
ism, then axiomatise affine planes.

1.4.2. Exercise. In an affine plane, prove that//is an equivalence relation.

1.4.3. Exercise. List all incidence geometries satisfying AP1 and AP2 but not AP3 .
1.4.4. Exercise. Let A = (P , L, I) be an affine plane. Prove that any line has at least
two points, and that two lines with the same points are equal. Conclude that there A is
isomorphic to an affine plane of the form A′ = (P ′ , L′ , ∈) .
1.4.5. Exercise. Let P be the set of points on the Euclidean sphere S2 (R) = {x ∈ R3 ∶
∥x∥ = 1} and L be the set of ‘great circles’, viz. circles of maximum radius on the sphere.
Prove that (P , L, ∈) is a geometry satisyfing AP1 . What about AP2 ?
1.4.6. Exercise. Prove that the following is the smallest affine plane. Can you identify it?
a b

c d
One point is removed;
diagonals do not meet.

1.4.7. Exercise. Let F be a skew-field.

1. Check that (F)2 A of Remark 1.3.4 is an affine plane.

2. Prove that (F)2 A = A2 (F) iff F is commutative.

3. Suppose that there is an anti-automorphism of F, viz. an additive bijection φ such
that (∀a, b)(φ(a ⋅ b) = φ(b) ⋅ φ(a)). Show that (F)2 A ≃ A2 (F). (The converse
can be proved in later exercises.)

8
2 Projective planes
Abstract. § 2.1 defines projective planes and gives one class of examples: the struc-
tures P2 (F) for F a skew-field. Not all projective planes are of this form. § 2.2
explores the relationships between the classes of abstract affine planes and of ab-
stract projective planes by so-called projectivisation and affinisation. § 2.3 applies
these operations to the concrete examples A2 (F) and P2 (F).

The need to ‘add points at infinity’ where parallel lines would meet was felt during
the Renaissance, and cannot be separated from the history of Painting. Early school
pencil-and-paper practice shapes our first intuition of geometry, and is convincingly
rendered by affine planes. However, trying to represent on a planar support the physical
3-dimensional space, begs for other means than linear projections. Indeed, our seeing
device (the eye) is a nearly punctual object. The theory of perspective thus naturally
takes place in projective geometry.
Mathematically, projective geometry is even more natural than its affine version.
Statements there are more concise; the algebraic behaviour is more homogeneous. This
course is really about projective planes.

2.1 The definition

2.1.1. Definition (projective plane). A projective plane is an incidence geometry (P , L, I)
satisfying the following axioms PP1 , PP2 , PP3 .
PP1 . Through any two distinct points there is a unique line.
PP2 . Any two distinct lines meet in a unique point.

PP3 . There are four points no three of which are collinear.

(The definition has a lovely symmetry called self-duality, to which we return in § 12.)
2.1.2. Remark (and notation). Both AP1 and PP1 say that any two distinct points lie on
a unique line. From now on, given two points a ≠ b of an affine or projective plane, (ab)
denotes the unique line through them.

We move to one source of examples. In a left-vector space, a (left-)vector line is a

1-dimensional left-vector subspace, which can be viewed as a line through the origin;
define a (left-)vector plane likewise.
2.1.3. Definition (projective plane over a skew-field P2 (F)). Let F be a skew-field. Let
P2 (F) be the following incidence geometry:

• as points, the left-vector lines of F3 ;

• as points, the left-vector planes of F3 ;
• as incidence relation, set-theoretic inclusion ⊆.

2.1.4. Remark. There also exists a right-version (F)2 P, which need not be equal nor
even isomorphic to P2 (F) (see Remark 1.3.4).
2.1.5. Proposition. Let F be a skew-field. Then P2 (F) is a projective plane.

9
2.1.6. Remark (important). Like in the affine case, not all projective planes are of the
form A2 (F). The Desargues property (§ 3) characterises those projective planes of the
form P2 (F).

2.1.7. Remark (continued). For instance, P2 (F9 ) has 92 + 9 + 1 = 91 points. But there
exist three more projective planes with 91 elements. To show that in total there are exactly
four, requires heavy computer assistance.2

Proof. We must check three things. This is easy as our construction relies on vectors
in dimension 3. Even though F need not be commutative, basic linear algebra remains
sound. We only use left-subspaces, and consequently omit ‘left’.

PP1 . Let α ≠ β be two projective points, i.e. two distinct vector lines L α , L β ⊆ F3 .
Then set H = ⟨L α , L β ⟩ ⊆ F3 , a vector plane of F3 . In projective terms, H is a line
λ. Now by definition, L α ⊆ H translates into αIλ, and βIλ likewise. Uniqueness
is clear as well, since there is no other vector plane containing both L α and L β .
PP2 . Start with two distinct projective lines λ ≠ µ. By construction they correspond
to distinct vector planes H λ ≠ H µ , and H λ ∩ H µ is a vector line L, which we see
as a projective point α. Obviously αIλ and αIµ, but uniqueness is clear as well.
PP3 . Vector lines spanned by vectors (1, 0, 0), (0, 1, 0), (0, 0, 1), and (1, 1, 1) define
suitable projective points.

The remarkable simplicity of the proof (cf. proof of Proposition 1.3.2) is already an
indication in favour of projective over affine geometry: even for ‘coordinatisable’ objects,
the projective version is easier to deal with.
2.1.8. Remark (projective coordinates). One can represent points of P2 (F) by equi-
valence classes of non-zero triples of coordinates, as follows. Consider the relation on
F3 ∖ {(0, 0, 0)}:

(a 1 , a 2 , a 3 ) ∼ (b 2 , b 2 , b 3 ) iff (∃λ)[(λa 1 = b 1 ) ∧ (λa 2 = b 2 ) ∧ (λa 3 = c 3 )].

This is the equivalence relation of proportionality of non-zero vectors. The set of equi-
valence classes is naturally the set of vector lines in F3 , i.e. points of P2 (F).
Write [a 1 , a 2 , a 3 ] for the class of (a 1 , a 2 , a 3 ); they are projective coordinates, defined
up to (left-)multiplication by a non-zero scalar. If [a 1 , a 2 , a 3 ] = [b 1 , b 2 , b 3 ] and a 1 =
b 1 ≠ 0, then a 2 = b 2 and a 3 = b 3 .

Projective coordinates are useful when bringing matrices into the picture. However,
since we treat F3 as a left-vector space, matrices will act from the right, viz. by row-matrix
product.

2.2 From affine to projective, and back

Another way to obtain projective planes is by ‘completing’ affine planes, a procedure we
now describe. In the affine world, some intersections are missing: if ℓ//m are distinct,
parallel affine lines, then we should add a ‘point at infinity’ where they meet at last. Now
2 C. Lam, G. Kolesova, L. Thiel. A computer search for finite projective planes of order 9, Discr. Math., 92,

pp.187–195, 1991.

10
if ℓ//m//n then the same point should be added for the missing intersection ℓ ∩ m and
for the missing intersection m ∩ n. So one is not working with pairs of parallel lines, but
with whole sets of pairwise parallel lines.

2.2.1. Lemma. Let A be any affine plane. Then//is an equivalence relation.

Proof. This holds of A2 (F) for F a skew-field, as one has a detailed description of
parallelism from the proof of Proposition 1.3.2. But we want a general proof, a proof
using only axioms AP1 , AP2 , AP3 . Remember that in this abstract setting ℓ and m are
parallel iff ℓ = m or ℓ and m do not meet. We freely replace A by an isomorphic plane
where incidence is given by membership.
There are three things to check:

Reflexivity. Every line ℓ satisfies ℓ = ℓ, so ℓ//ℓ.

Symmetry. If ℓ//m then either ℓ = m or ℓ ∩ m = ∅; in either case m//ℓ.
Transitivity. Suppose ℓ//m and m//n; we show that ℓ//n. If ℓ = n we are done. Otherwise
we must prove ℓ ∩ n = ∅. So suppose not; by AP1 there is a ∈ ℓ ∩ n. Now by
AP2 , there is a unique line parallel to m and containing a; however this applies
to both ℓ and n. So actually ℓ = n, a contradiction showing that ℓ ∩ n = ∅. Hence
ℓ//n, as desired.

2.2.2. Notation. For ℓ an affine line, let [ℓ] be its equivalence class and call it its direction.
Intuitively one should add a new point ‘at infinity’ for each direction; there will also
be a line ‘at infinity’. But ordinary, affine lines ℓ are now too short: we must force a
projective line to contain the direction. This explains why we go through ℓ̂ below.
2.2.3. Definition (projectivisation  of an affine plane). Let A = (P , L, ∈) be an affine
plane. The projectivisation of A is the incidence geometry  = (P̂ , L̂, ∈) defined as
follows:

• for each ℓ ∈ L, let ℓ̂ = ℓ ∪ {[ℓ]} be the ‘completion of line ℓ’;

• let ℓ∞ = {[ℓ] ∶ ℓ ∈ L} be the ‘line of directions’;
• P̂ = P ∪ ℓ∞ = P ∪ {[ℓ] ∶ ℓ ∈ L},
• L̂ = {ℓ̂ ∶ ℓ ∈ L} ∪ {ℓ∞ }.

ℓ′2 [ℓ 1 ]=[ℓ ′1 ]

ℓ2
ℓ′1 [ℓ 2 ]=[ℓ ′2 ]
ℓ1
ℓ∞
Complete each affine line ℓ into ℓ̂.
Do not forget line at infinity ℓ∞ .

2.2.4. Proposition. Let A be an affine plane. Then  is a projective plane.

11
 Proof. There are three axioms to check.
PP1 . Let α ≠ β be two points in P̂. We see several cases.
• If α = a ∈ P and β = b ∈ P, then by AP1 there is a unique line ℓ ∈ L
containing both. Clearly α, β ∈ ℓ̂. We also have uniqueness. If another line
λ ∈ L̂ contains α and β, then λ cannot be ℓ∞ , so λ = m̂ for some affine
line m ∈ L. Then a, b ∈ m, so by uniqueness in AP1 one has m = ℓ and
therefore λ = m̂ = ℓ̂. So ℓ̂ is the only line of L̂ containing α and β.
• Suppose α = a ∈ P but β ∈ P̂ ∖ P = ℓ∞ (the other case is similar). Then
by definition, there is ℓ ∈ L with β = [ℓ]. Now by AP2 there is a unique
m ∈ L with a ∈ m and m // ℓ. Then on the one hand α ∈ m̂, and on the
other hand [ℓ] = [m] ∈ m̂. Now to uniqueness. If a line λ contains a and
β, then it cannot be ℓ∞ . So it is of the form n̂ for some affine line n. Now
α ∈ n̂ implies a ∈ n, and β = [ℓ] ∈ n̂ implies [ℓ] = [n], that is, ℓ // n. By
uniqueness in AP2 one has n = m, and therefore λ = n̂ = m̂.
• Now suppose α, β ∈ ℓ∞ . Clearly ℓ∞ is the only line in L̂ incident to both
α and β.
PP2 . Exercise.
PP3 . Obvious from AP3 and the definition of the projectivisation.

The converse operation of ‘downgrading from projective to affine’ involves choosing

the line to be removed.
2.2.5. Definition (affinisation P̌ λ of a projective plane). Let P = (P , L, ∈) be a projective
plane. Fix one line λ ∈ L. The affinisation of P with respect to λ is the incidence geometry
P̌ λ = (P̌ , Ľ, ∈) defined as follows:

• for each µ ∈ L ∖ {λ}, set µ̌ = µ ∖ λ;

• P̌ = P ∖ λ;
• Ľ = { µ̌ ∶ µ ∈ L ∖ {λ}}.
Be careful that not all our notation reflects dependence on λ.

2.2.6. Remarks.
• The isomorphism type of P̌ λ depends on the line λ you chose to remove. Affinisa-
tion is not uniquely defined, one may not say ‘the affinisation’ without specifying
λ. (Cf. ‘’the projectivisation’, which is well-defined.)

• In particular, an arbitrary affinisation of the projectivisation of an affine plane A

may depend on the choice of line removed from Â, and need not be isomorphic
to A.
• But starting from projective P and letting A = P̌ λ , one has  ≃ P regardless of λ.

See exercise 2.4.4.

12
2.3 Projectivising A2 (F), affinising P2 (F)
Applying the projectivisation/affinisation procedures to A2 (F) and P2 (F) gives what
one expects.
2.3.1. Proposition. Let F be a skew-field. Then:
̂
(i) A 2 (F) ≃ P2 (F);

(ii) for any line, P̌2 (F) ≃ A2 (F).

Proof.
(i) We describe an isomorphism. Let H 0 ≤ F3 be a vector plane and x ∈ F3 be a
vector not in H 0 . Let H 1 = x + H 0 , an affine translate of H 0 . Clearly A2 (F), H 0 ,
and H 1 are isomorphic affine planes; in particular A ̂ 2 (F) and H ̂1 are isomorphic
projective planes. So it suffices to see that H ̂1 and P2 (F) are isomorphic, as
follows.
L
a x

H1
x + L′

O
H0
L′

L≤
/ H 0 is mapped to a, while L′ ≤ H 0 is mapped to [x + L′ ].

Let L ≤ F3 be a vector line. If L ≤/ H 0 , then L will intersect H 1 in a point say a;

map L to a. If L ≤ H 0 , then there is no intersection, so L should be mapped to a
point at infinity. In the previous construction, such points were the equivalence
classes (directions) of lines of H 1 ; so map L to [x + L], the direction of x + L
which is a line of the affine plane H 1 . This maps vector lines in F3 to points in
Ĥ 1 .
Now let H ≤ F3 be a vector plane. If H ≠ H 0 then H ∩ H 1 is a line ℓ of the affine
plane H 1 ; map H to ℓ̂ as in the projectivisation construction. If on the other
hand H = H 0 then map H to ℓ∞ . This maps vector planes in F3 to lines in Ĥ 1 .
It is an exercise to finish the proof:
• check that we have a bijection between points of P2 (F) and points of H ̂1 ;
̂1 ;
• check that we have a bijection between lines of P2 (F) and lines of H
• check that these bijections preserve incidence.
(ii) Exercise.

13
2.4 Exercises
2.4.1. Exercise. Show that the following is the smallest projective plane, and identify it.

The circle counts as a line.

2.4.2. Exercise. Let P be a projective plane with line λ. Show that A = P̌ λ is an affine
plane and  ≃ P.
2.4.3. Exercise. Let F be a skew-field and P = P2 (F). Prove that for any line one has
P̌ λ ≃ A2 (F).

2.4.4. Exercise (the isomorphism type of P̌ λ depends on the line you remove).
1. Let A be an affine plane and α ∈ Aut(A) be an automorphism. Prove that α extends
to a unique automorphism of Â.
2. Let P be a projective plane with two lines λ 1 , λ 2 . Let A i be the affinisation with
respect to λ i . Suppose that A1 ≃ A2 , and prove that there is α ∈ Aut(P) sending λ
to µ.
3. Deduce that the isomorphism type of P̌ λ may depend on the line you remove. Hint:
there is a projective plane with whose automorphism group is not transitive on lines
(see Exercise 3.4.6).

4. Also deduce that there exist two non-isomorphic affine planes with isomorphic pro-
jectivisations.
5. Prove however that if P ≃ P2 (F), then P̌ λ does not depend on the line you remove.
2.4.5. Exercise. Let P be a projective plane. Let ℓ 1 ≠ ℓ 2 be two lines and p ∉ ℓ 1 ∪ ℓ 2 . Show
that the following picture defines a bijection ℓ 1 ≃ ℓ 2 .

q1 q2
p
ℓ1 ℓ2

2.4.6. Exercise. Let P be a finite projective plane.

1. For ℓ a line and p ∉ ℓ, give a bijection between ℓ and L p = {m ∈ L ∶ p ∈ m}.
2. Deduce that there is an integer n such that: • every line has n + 1 points; • every
points belongs to n + 1 lines; • there are n 2 + n + 1 points and as many lines.

3. What is the value of n for P2 (Fq ), where Fq is the field of order q?

14
Remark. The integer n is called the order of n. It has been conjectured that for finite
P, the order is always a prime power; this is open. It is not even known whether there
is a projective plane of order 12; killing those of order 10 has been remarkably difficult.
Finitary aspects are fascinating but of astounding complexity.3
2.4.7. Exercise. Let A be a finite affine plane. Show that there is n such that: • every line
has n points; • every point is on n + 1 lines; • there are n 2 points in total; • there are n 2 + n
lines, falling in n + 1 parallel classes of n lines each. Hint: this exercise belongs to § 2, not
to § 1.
2.4.8. Exercise. Let G = SO3 (R). Let I = {i ∈ G ∶ i 2 = 1 ≠ i} be the set of involutions.
On I consider the relation iε j iff (i j = ji ≠ 1). Prove that (I, I, ε) ≃ P2 (R).

3 The Desargues property

Abstract. The Desargues property is not ‘yet another theorem’ in geometry. It is a
key combinatorial statement characterising those projective planes which are of the
form P2 (F). § 3.1 describes the configuration and its affine versions. § 3.2 constructs
projective planes which do not have the Desargues property. § 3.3 states the main
result, Hilbert’s Coordinatisation Theorem. (§§ 4 and 5 are devoted to proving it.)

3.1 Statement of the property

Recall that given two points a ≠ b of an affine or projective plane, (ab) denotes the
unique line through them. It also equals (ba).
3.1.1. Definition (projective desarguesian plane). A projective plane is desarguesian if it
has the projective Desargues property, which is the following axiom.
If abc, a′ b′ c ′ are two triangles such that (aa′ ), (bb ′ ), and (cc ′ ) meet, then
a ′′ = (bc) ∩ (b′ c ′ ), b ′′ = (ac) ∩ (a′ c ′ ), and c ′′ = (ab) ∩ (a ′ b ′ ) are collinear.
a′

a
c′
c
o

b
b′

a′′ c ′′ b′′
One should practice drawing similar pictures, obtained as follows. 1. Choose
a point o and draw three lines from it. 2. ‘Hang’ two triangles on them (each
line contains exactly one vertex of each triangle). 3. Match sides of the tri-
angles and draw intersections of matching sides. 4. The resulting intersec-
tions are collinear.
3 C. Lam, The search for a finite projective plane of order 10, Am. Math. Monthly, 98 no. 4, pp. 305–318,

April 1991.

15
3.1.2. Remark. No assumptions nor conclusions about o being on (a′′ )–(b ′′ )–(c ′′ ).
The following definition is not for learning. One should be aware that depending on
the position of the line at infinity, the projective Desargues property will have several,
different-looking affine avatars.
3.1.3. Definition (affine desarguesian plane). An affine plane is desarguesian if it has the
following affine Desargues property, which is the following axiom.
Let abc and a′ b′ c ′ be triangles (involving six distinct points). Suppose that
lines (aa′ ), (bb′ ), (cc ′ ) are pairwise distinct, and either that they are par-
allel or that they meet at the same point. Then:
• either (ab)//(a′ b′ ), (ac)//(a′ c ′ ), and (bc)//(b ′ c ′ );
• or (ab)//(a ′ b ′ ) but (ac) ∩ (a′ c ′ ) = {b′′ } and (bc) ∩ (b ′ c ′ ) = {a ′′ }
satisfy (a ′′ b′′ )//(ab);
• or two similar cases (involving the intersection c ′′ of (ab) and (a′ b ′ ));
• or a ′′ , b′′ , c ′′ are collinear.
The following is straightforward.
3.1.4. Lemma.
(i) Let A be an affine plane. Then A is desarguesian iff  is.
(ii) Let P be a projective plane with fixed line λ. Then P is desarguesian iff P̌ λ is.
We therefore focus on the projective version.
3.1.5. Remark. There is a form of ‘converse’. Let Desargues’ be the following property
(which a plane may have or not).
If two triangles abc and a ′ b ′ c ′ are such that the intersection points a′′ =
(bc) ∩ (b ′ c ′ ), b ′′ = (ac) ∩ (a ′ c ′ ) and c ′′ = (ab) ∩ (a′ b′ ) are collinear, then
lines (aa′ ), (bb′ ) and (cc ′ ) are concurrent.
It turns out that this condition is equivalent to the original Desargues property. This is
proved in § 12.
Some projective planes satisfy the Desargues property (P2 (F) does, which will be
proved); some do not (§ 3.2). Hilbert characterised which do.

3.2 Free planes and non-desarguesian planes

We shall construct non-desarguesian planes by use of a free construction. The general
idea behind such constructions is to add everything one needs, and nothing more. Here
it will mean adding lines to connect any two points, but also adding points to intersect
any two lines. Of course this creates new lines and points, so it must be done inductively.
3.2.1. Definition (free plane4 ). Let Γ0 = (P0 , L0 , ∈) be an incidence geometry.
• Construct a sequence of geometries Γn = (Pn , Ln ) by the following induction.
Let n be given.
4 Free projective planes where introduced in: M. Hall. Projective Planes. Trans. Amer. Math. Soc. 54

(1943), 229-277.

16
 • Let X n be the set of unordered pairs {a, b} of distinct points of Pn which are not
connected by a line of Ln . Also let Yn be the set of unordered pairs {ℓ, m} of
distinct lines of Ln which do not meet in a point of Pn .
• For each {a, b} ∈ X n , add a new line ℓ{a,b} incident to a and b, but to no other
point. Also for each {ℓ, m} ∈ Yn , add a new point p{ℓ,m} on both ℓ and m, but on
no other line.
• Let Pn+1 = Pn ∪ {ℓ{a,b} ∶ {a, b} ∈ X n } and Ln+1 = Ln ∪ {p{ℓ,m} ∶ {ℓ, m} ∈ Yn }.
• Finally let P = ⋃N Pn and L = ⋃N Ln .
Then Γ = (P , L) is called the plane freely generated by Γ0 .
3.2.2. Remarks.
• In some sources one adds lines at odd stages and points at even stages, but this is
less symmetric (§ 12).
• Starting with no points and no lines, or just collinear points, or concurrent lines,
nothing will happen.
• Starting with Γ0 having bad incidence properties, the free geometry will retain
them.
3.2.3. Definition (admissible incidence geometry). An incidence geometry Γ = (P , L, I)
is admissible if there are no two distinct lines sharing two distinct points.

a b

The forbidden configuration.

3.2.4. Proposition. Let Γ0 = (P0 , L0 , ∈) be an admissible incidence geometry satisfying

PP3 . Then the plane freely generated by Γ0 is a projective plane.

Proof. As usual, three things must be checked.

PP1 . Let a ≠ b ∈ P be two distinct points. Say that a ∈ Pn and b ∈ Pm ; we may assume
n ≥ m.
First suppose that n = 0 and a, b are already connected by a line ℓ ∈ L0 . Then
ℓ is unique as such; moreover, we never introduce another line through a and b
since {a, b} ∉ X n . So ℓ remains unique even in L.
Now suppose that there is no ℓ ∈ L0 containing a and b. Regardless of the value
of n, at stage n + 1 we have introduced a line ℓ joining a and b. The argument
for uniqueness of ℓ is similar.

PP2 . Entirely similar.

PP3 . This is almost by assumption but not entirely. There are four points a, b, c, d ∈
P0 ⊆ P which are not collinear in L0 . We must see that they remain non-
collinear in the sense of L. But at stage 1, when we add line ℓ = (ab), it does
not contain c. And we never add another line through a and b. So there never
is a line containing a, b, c: they remain non-collinear also in L. Likewise for the
other triples.

17
3.2.5. Example. Start with the admissible incidence geometry consisting of four points
and no lines.
1

1
0
1
3 0
2
1
0 0 2
1

3 2
3
Numbers indicate the stages of appearance of points/lines. Imagine the fol-
lowing next steps.
This example is vacuously desarguesian: there are no Desargues configurations.
3.2.6. Remark. More generally, let Γ κ be the free plane generated by κ points and no
lines (κ may be infinite). Their class behaves roughly like the class of free groups. Here
are a few (non-trivial) properties.
• A subplane of a Γ κ is another Γ λ (λ need not be smaller than κ).
• As κ varies the Γ κ are pairwise non-isomorphic.
• However, they all have the same elementary theory.
The model theory of the free projective plane has been studied only recently.5
3.2.7. Theorem. There exist non-desarguesian projective and affine planes.

Proof. Start with the admissible incidence geometry Γ0 consisting of:

• ten points P0 = {o, a, b, c, a ′ , b ′ , c ′ , a ′′ , b ′′ , c ′′ },

• L0 = all lines involved in the Desargues configuration except line (a′′ b ′′ c ′′ ).

Let P be the plane freely generated by Γ0 . By Proposition 3.2.4, P is a projective
plane. In Γ0 there is no line joining a′′ , b ′′ , c ′′ , and in the construction we never add
a line simultaneously joining three existing points. So in P there is no line joining
a′′ , b′′ , c ′′ . Therefore P does not satisfy the Desargues property.

3.2.8. Remarks.
• Although the example of a non-desarguesian plane we gave is obviously infinite,
there exist finite, non-desarguesian projective planes.
• Hilbert’s Grundlagen der Geometrie, Theorem 33, contains a completely different
example of a non-desarguesian plane. His construction is rather involved.
5 T. Hyttinen and G. Paolini. First-order model theory of free projective planes. Ann. Pure Appl. Logic 172

(2) (2021), paper no. 102888.

18
3.3 Hilbert’s Coordinatisation Theorem
3.3.1. Theorem (Hilbert coordinatisation). Let P be a projective plane. Then the following
are equivalent:
(i) P is desarguesian;
(ii) there is a skew-field (unique up to isomorphism) F with P ≃ P2 (F);
(iii) P is isomorphic to a plane in some projective, 3-dimensional space (Definition 4.1.1
below).
3.3.2. Remark. The Desargues property is therefore a strong dividing line in the theory
of projective planes; the latter come in two sorts:
• desarguesian projective planes (of the form P2 (F) for some skew-field): our know-
ledge is satisfactory, viz. as satisfactory as our understanding of skew-fields;
• non-desarguesian projective planes (not of the form P2 (F) for any skew-field):
we know extremely little, and presumably there is little to say in general.
In particular, contemporary computers are not powerful enough to help us classify (?)
finite projective planes. One would need new ideas.
Notice that the free, non-desarguesian construction of § 3.2 shows that there exist
non-coordinatisable projective planes.

3.4 Exercises
3.4.1. Exercise. Return to § 3.1; we say that (o; a, b, c, ; a′ , b ′ , c ′ ; a ′′ , b′′ , c ′′ ) is a Desar-
gues configuration seen from o.
1. Check that (a ′′ ; b, c ′′ , b ′ ; c, b ′′ , b ′ ; a′ , o, a) is a Desargues configuration seen from
a ′′ .
2. From how many points is it a Desargues configuration?
3. Suppose P has the Desargues property. Prove that it has the ‘converse Desargues
property’ (hint: a′′ ). (Such tricks are explained in § 12.)
3.4.2. Exercise (harder). The ‘generic’ Desargues configuration is the one where o ∉ (a′′ b ′′ ):
there are 10 points, 10 lines, and every point is on exactly 3 lines.
Determine the automorphism group of the generic Desargues configuration, viz. the
group of bijections of the 10 points and 10 lines which preserve all incidence relations.
3.4.3. Exercise. Prove the Desargues property in P2 (F) using projective coordinates. (If
any simpler, prove all affine versions using affine coordinates, then conclude.)
3.4.4. Exercise (affine versions).
1. Removing various lines, draw all affine versions of the Desargues property. Below is
one.
a′
a
b b′
o
c
c′

19
 2. Prove that a projective plane P is desarguesian iff for any line λ, its affinisation P̌ λ
is.
3. Prove that an affine plane A is desarguesian iff its projectivisation is.
3.4.5. Exercise. Let Γ0 be an admissible incidence geometry satisfying PP3 . Show that if
the plane freely generated by Γ0 is some Γn from the construction, then Γ0 was a projective
plane already. Hint: exercise 2.4.5.
3.4.6. Exercise.
1. Let Γ0 be four points with no lines, and Γ be the free projective plane generated by
Γ0 . Prove that no substructure of Γ is isomorphic to the Fano plane of Exercise 2.4.1.
Hint: consider the least n such that Γn contains a copy of the Fano plane.
2. Let ∆ 0 be the Fano plane with an extra (not connected) point. Let ∆ be the free
projectiev plane generated by ∆ 0 . Show that ∆ contains a unique copy of the Fano
plane.
3. Deduce that there exist projective planes whose group of automorphisms does not
permute lines transitively.
3.4.7. Exercise (The Moulton plane6 ). Equip R2 with the following lines:
• all ordinary lines with non-positive slope (including horizontal and vertical lines);
• for each a > 0 and x 0 ∈ R2 , the set ℓ a,x 0 = {(x, 2a(x − x 0 )) ∶ x ∈ R≤0 } ∪ {(x, a(x −
x 0 )) ∶ x ∈ R≥0 }.
1. Draw a picture.
2. Check that this is an affine plane.
3. Prove that at least one instance of the affine Desargues properties fails. Hint: take
an ordinary Desargues configuration, translate and rotate so that nine points out of
ten are above zero, and two lines out of three passing through this point are ‘unbent’.
4. Conclude that there is a non-desarguesian projective plane.

4 A detour through space

Abstract. § 4.1 introduces projective 3-spaces. § 4.2 (optional) proves a non-planar
version of the Desargues property in such spaces. § 4.3 proves part of Hilbert’s
Coordinatisation Theorem. (The argument continues in § 5.)

We want to prove the following.

Theorem (Hilbert coordinatisation). Let P be a projective plane. Then the following are
equivalent:
(i) P is desarguesian;
(ii) there is a skew-field (unique up to isomorphism) F with P ≃ P2 (F);
(iii) P is isomorphic to a plane in some projective, 3-dimensional space.
We first explain the missing terminology.
6 F. Moulton, Simple non-desarguesian geometry, Trans. Amer. Math. Soc., April 1902

20
4.1 Projective 3-spaces
4.1.1. Definition (projective 3-space). A projective 3-dimensional space is an incidence
structure (P , L, Π) consisting of points, lines, and planes (and incidence relations), sat-
isfying the following axioms PS1 –PS6 .
PS1 . Any two points are uniquely collinear.
PS2 . Any three non-collinear points are uniquely coplanar.
PS3 . Any line and plane meet.
PS4 . Any two planes share (at least) a line.
PS5 . There are four non-coplanar points no three of which are collinear.
PS6 . Every line has at least three points.
4.1.2. Example. Let F be a skew-field. Then P3 (F), where points are (left-)vector lines
in F4 , lines are 2-dimensional (left-)vector subspaces of F4 , and planes are (left-)vector
hyperplanes of F4 , is a projective space (for set-theoretic incidence).
It will be a consequence of Hilbert Coordinatisation that every projective 3-space is
of this form. (Cf. dimension 2: there exist projective planes not of the form P2 (F).)
Like in Proposition 1.1.4, every projective 3-space is isomorphic to one where incid-
ence relations are of the form p ∈ ℓ ⊆ π. This simplifies notation and terminology.
4.1.3. Lemma. Let S = (P , L, Π) be a projective 3-space, and let π ∈ Π. Let Pπ = {p ∈
P ∶ p ∈ π} and Lπ = {ℓ ∈ L ∶ ℓ ∈ π}. Then (Pπ , Lπ , ∈) is a projective plane.

Proof. See exercise 4.4.2.

Therefore ‘isomorphic to a plane in some projective, 3-dimensional space’ means

‘isomorphic to Pπ for some π in some projective, 3-dimensional space’.

Proof of (ii)⇒(iii): P2 (F) embeds into a projective 3-space. We prove that for any
skew-field, P2 (F) embeds into P3 (F).
This is trivial in projective/homogeneous coordinates; one not at ease with them
would argue as follows. Using linear coordinates, embed F3 into F4 , say on the ‘ho-
rizontal’ hyperplane: map (x, y, z) to (x, y, z, 0). This is a linear embedding, so it
certainly preserves collinearity. Moreover, it preserves inclusion of vector subspaces.
So it induces well-defined, injective maps :

f ∶ {vector lines of F3 } → {vector lines of F4 },

g∶ {vector planes of F3 } → {vector planes of F4 },

which preserve ≤. These define an embedding of P2 (K) into P3 (K).

4.1.4. Remark (higher-dimensional projective geometries). One may wish to capture

n-dimensional projective spaces.
• When n is given, one can always introduce n distinct types of objects (points,
lines, and so on) and give a long list of axioms. This is clumsy.

21
 • One can consider only points P and ‘subspaces’ S equipped with a dimension
function S → N. But this is unsatisfactory, logically speaking, as the description
of the geometry now relies on the purely non-geometric notion of the integers.

• An elegant solution, ‘à la Veblen’, is the following. Call projective space any incid-
ence geometry satisfying PP1 , PP3 , and the Pasch axiom:
if a, b, c, d are points such that (ab)∩(cd) ≠ ∅, then (ac)∩(bd) ≠ ∅.
c

a b
The Pasch axiom describes coplanarity without mentioning planes.

4.2 Non-planar Desargues configurations

This subsection is not necessary but helps gain some intuition of the next.

4.2.1. Lemma. Work inside a projective 3-space S. Suppose (o; a, b, c; a′ , b ′ ; a′′ , b′′ , c ′′ )
is a non-planar Desargues configuration in S. Then a ′′ , b ′′ , c ′′ are collinear.

Proof. By assumption, abc and a ′ b ′ c ′ do not define the same plane: otherwise all
points and lines would be in that plane. Say (abc) is in some plane π 1 and (a′ b′ c ′ ) in
some other plane π 2 ≠ π 1 .

a′

a
c′
c
o
b
b′

a ′′ c ′′ b ′′

Then observe how a′′ ∈ (bc) ⊆ π 1 ; likewise, b ′′ , c ′′ ∈ π 1 . But similarly, a ′′ ∈

(b ′ c ′ ) ⊆ π 2 ; likewise, b′′ , c ′′ ∈ π 2 . So a ′′ , b′′ , c ′′ ∈ π 1 ∩ π 2 . One may prove that the

22
 intersection of two distinct planes is exactly a line (exercise 4.4.2): whence collinearity
of the three points.

4.3 ‘Embeddable implies desarguesian’

We prove that if a projective plane is isomorphic to a plane in a projective 3-space, then
it is desarguesian.

Proof of (iii)⇒(i). Let P be a projective plane isomorphic to some plane of a project-

ive 3-space S.
Consider a Desargues configuration (o; a, b, c; a ′ , b ′ , c; a ′′ , b ′′ , c ′′ ) taking place in
P. We want to show collinearity of a ′′ , b ′′ , c ′′ . The idea is to slightly ‘lift’ the picture,
work in S, and then project back.
a′
a ′1

a1
a c′
c
e
o
b
b′

a′′ c ′′ b ′′
c 1′′
b ′′1
Imagine ‘slightly rotating’ the planar triangles by ‘slightly lifting’ the vertices a
and a′ into a 1 and a ′1 , still preserving collinearity of o, a 1 , a ′1 . Mathematically
this is achieved through choosing e ∉ π then a1 ∈ (ae).

• Let e ∈ S ∖ P.

• Let a 1 ∈ (ae) be neither a nor e; there is such a point by PS6 .

• Also let a′1 = (oa 1 ) ∩ (ea ′ ). Having changed triangles, we must consider new
intersections.
• Finally let b ′′1 = (a 1 c) ∩ (a ′1 c ′ ) and c 1′′ = (a 1 b) ∩ (a′1 b ′ ); notice that a ′′ however
remains unchanged.

Step 1. Points b′′1 and c 1′′ are well-defined ; triangles a 1 bc and a ′1 b′ c ′ satisfy the assump-
tion of the Desargues property and are not coplanar.

Verification. First, the ‘Desargues intersections’ a′′ , b ′′1 , c 1′′ do exist; this is no longer
trivial as lines in S need not intersect. Point a ′′ = (bc) ∩ (b ′ c ′ ) has not changed. By
PS2 , let π ′ be the plane containing o, a 1 , c. It then contains a′1 ∈ (oa 1 ) and c ′ ∈ (oc),

23
 so π ′ also contains lines (a 1 c) and (a′1 c ′ ), which therefore meet: intersection b ′′1 is
well-defined. Similar argument for c 1′′ .
By construction, o, a 1 , a′1 are collinear. Since other vertices have not changed,
triangles a 1 bc and a ′ 1b′ c ′ meet the Desargues assumption. Moreover, these triangles
are not in the same plane. Otherwise that plane would be P; since a, a 1 ∈ P, one then
has e ∈ (aa 1 ) ⊆ P, a contradiction. ◇

Step 2. Points a′′ , b ′′1 , c 1′′ are collinear.

Verification. This is the argument given in Lemma 4.2.1. Using axiom PS2 , let π 1 be
the plane containing a 1 , b, c and π 2 be the plane containing a′1 , b ′ , c ′ . Clearly these
planes are distinct, so π 1 ∩ π 2 contains a line ℓ.
By Lemma 4.1.3, π 1 ∩ π 2 = ℓ is a line. Observe how: a ′′ ∈ (bc)∩(b ′ c ′ ) ⊆ π 1 ∩ π 2 =
ℓ, and likewise for b ′′ and c ′′ . Finally a ′′ , b ′′ , c ′′ are on ℓ, hence collinear. ◇

We now project back. Let p be the projection function from e onto π, defined as
follows. For any point x ≠ e, let p(x) be the meeting point of line (ex) and plane π,
which does not contain it. (The value of p at e remains undefined.) Notice that:
• p is the identity on π;
• p preserves incidence, i.e. if x ∈ ℓ, then p(x) ∈ p(ℓ).
Step 3. p(b′′1 ) = b ′′ and p(c 1′′ ) = c ′′ .

Verification. Since (ea 1 ) = (ea) and a ∈ π, one has p(a 1 ) = a. But also (ea ′1 ) = (ea ′ )
so p(a ′1 ) = a′ . In particular,

{p(b ′′1 )} = p ((a 1 c) ∩ (a′1 c ′ ))

= p((a 1 c)) ∩ p((a′1 c ′ ))
= (p(a 1 )p(c)) ∩ (p(a ′1 )p(c ′ ))
= (ac) ∩ (a ′ c ′ )
= {b ′′ }.

One obtains p(c 1′′ ) = c ′′ likewise. ◇

Since p preserves incidence (and therefore collinearity), the images of a ′′ , b′′1 , c 1′′
remain collinear. Therefore p(a′′ ) = a ′′ , p(b ′′1 ) = b′′ , and p(c 1′′ ) = c ′′ are collinear.

4.3.1. Remarks.
• Essentially the ideas here come from Desargues (who treated only F = R).
• Baldwin and Howard gave a direct proof (i)⇒(iii) without introducing coordin-
ates.7 It is quite involved.
It remains to prove that a desarguesian projective plane can be coordinatised. This
was first done by Hilbert and we sketch his method in § 5.
7 J. Baldwin. Formalization, primitive concepts, and purity. Rev. Symb. Log. 6 (1), 87–128. 2013.

24
4.4 Exercises
4.4.1. Exercise. Return to Remark 4.1.4. Prove that Pasch’s axiom is equivalent to:
if a line meets two sides of a triangle, it also meets the third side.
4.4.2. Exercise. Let S = (P , L, Π) be a projective 3-space. Prove the following.
1. If π 1 ≠ π 2 are distinct planes sharing line ℓ, then π 1 ∩ π 2 = ℓ.
2. If p, q are points then there are at least two planes containing them.
3. If p ≠ q are distinct points on plane π, then (pq) ⊆ π.
4. Lemma 4.1.3.

5 The coordinatisation theorem: Hilbert’s version

Abstract. A sketch of Hilbert’s proof that a desarguesian projective plane can be co-
ordinatised. The argument is long and pictures are increasingly involved, although
the conceptual level remains basic. We just give the first steps; some further steps
are left as exercises; some should just be read from Hilbert’s book, or imagined. In
more advanced sections (§§ 10–11), we shall give abstract arguments using group
theory.

Let P be a desarguesian projective plane. Here is the global strategy.

1. Fix some line λ ∈ L(P). Let A = P̌ λ ; then A is a desarguesian affine plane. Co-
ordinatisation actually takes place in A.
2. Fix one line ℓ of A and cleverly define two operations +, ⋅ on it. Then F = (ℓ; +, ⋅)
is a skew-field. (Checking it takes time. But arguments are repetitive and one
quickly gets the general ideal.)
3. Coordinatise A using F, so A ≃ A2 (F). (This is actually not as obvious as it seems,
and Desargues is used again.)
̂
4. Finally conclude P ≃ Â ≃ A 2 (F) ≃ P2 (F).

We do not cover (2) entirely, and we entirely omit (3). Details are to be found in
Hilbert’s Grundlagen der Geometrie, chapter V. The proof starts here.

5.1 The frame

m mx
π→m, //n (x)
ℓx
x

π→ℓ, //n (x)

0 ℓ
n nx

25
Notation for directions. Let ℓ, m, n be three pairwise non-parallel lines (later we shall
require them to be non-concurrent).
For x ∈ A we let:
• ℓ x be the line parallel to ℓ through x;
• m x be the line parallel to m through x;
• n x be the line parallel to n through x.

Notation for projections. Also construct:

π→ℓ, //n ∶ A → ℓ
x ↦ ℓ ∩ nx .

Since n x // n is not parallel to ℓ, this makes sense. Define π→m, //n ∶ A → m similarly.
Notice that π→ℓ, //n is the identity on ℓ (and similarly for π→m, //n ).

The prime function. Now consider:

ℓ → m
a ↦ a′ ∶= n a ∩ m

clearly a bijection between ℓ and m. For any x ∈ A one has (π→ℓ, //n (x))′ = π→m, //n (x).
Finally let 0 = ℓ ∩ m and notice 0′ = 0.

5.2 Addition

m mb

a′ p a,b
ℓ a′
ℓ
0 a b a+b
n

5.2.1. Notation. For a, b in ℓ, define:

• p a,b = ℓ a ′ ∩ m b (p is for ‘plus’);
• a + b = π→ℓ, //n (p a,b ).
5.2.2. Proposition. (ℓ; +) is an abelian group with identity 0.

Proof. The proof is a series of lemmas.

5.2.3. Lemma. + is well-defined; for any a ∈ ℓ one has 0 + a = a + 0 = a.

Proof. Since ℓ and m are not parallel and // is an equivalence relation, p a,b is
uniquely defined. So is a + b ∈ ℓ. Neutrality is obvious as the construction then
degenerates: ℓ 0′ = ℓ so p 0,a = ℓ ∩ m a = a, and m 0 = m so p a,0 = ℓ a ′ ∩ m = a′ , whence

26
 a + 0 = π→ℓ, //n (a′ ) = a.

5.2.4. Lemma. Addition is commutative.

Proof. We may suppose that neither a nor b is 0. We may also suppose a ≠ b.

Consider the following picture.

m ma mb
b ′ ℓb′
p b,a qb

a′ ℓ a′
qa p a,b
ℓ
0 a b
n

Introduce:
• q a = m a ∩ ℓ a′ ;

• qb = mb ∩ ℓb′ .
Step 1. Points 0, q a , q b are collinear.

Verification. Let r = (0q a ) ∩ ℓ b ′ . Consider the configuration (0; a′ , q a , a; b ′ , r, b).

b′
a′
qa r
0
a
b

Then (a ′ a) = n a and (b ′ b) = n b are parallel; also, (a ′ q a ) = ℓ a ′ and (b ′ r) =

ℓ b ′ are parallel. So by the Desargues property, (aq a ) = m a and (br) are parallel.
Therefore also (br) // m a // m b ; now the only parallel to m b through b being m b ,
one has r ∈ (br) = m b . Since also r ∈ ℓ b ′ by construction, it must be r = q b . By
construction, 0, q a , q b = r are now collinear. ◇

Step 2. Lines (p b,a p a,b ) and n are parallel.

27
 Verification. Consider the Desargues configuration (q b ; p b,a , q a , p a,b ; b′ , o, b).

b′
p b,a
0 qa
qb
p a,b
b

Then (b ′ o) = m and (p b,a q a ) = m a are parallel, and also (ob) = ℓ and (q a p a,b ) =
ℓ a ′ are parallel. By the Desargues property, (b ′ b) = n b and (p b,a p a,b ) are parallel.
◇

Since (p b,a p a,b ) // n, one has n p b,a = n p a,b . Hence b + a = π→ℓ, //n (p b,a ) =
π→ℓ, //n (p a,b ) = a + b.

5.2.5. Lemma. + has opposites.

Proof. Let a ∈ ℓ; we may assume that a ≠ 0. We must find b ∈ ℓ such that a + b = 0;

by commutativity this will suffice. Let q = n o ∩ ℓ a ′ (which makes sense as n and ℓ are
not parallel), then b = ℓ o ∩ m q .
We claim that a + b = 0. Indeed, p a,b = ℓ a ′ ∩ m b = ℓ a ′ ∩ m q = q ∈ n o , so
a + b = π→ℓ, //n (p a,b ) = π→ℓ, //n (q) = 0.

5.2.6. Lemma. + is associative.

Proof. Let a, b, c ∈ ℓ; we may suppose that all three are distinct, and distinct from 0.
Consider the following picture.

(b + c)′ p b+c ,a

(a + b)′ p a+b,c
b ′ p b,c
p b,a

a a+b c b+c

• We know that (p b,a p b,c ) = ℓ b ′ is parallel to ((a + b)′ p(a+b),c) ) = ℓ(a+b)′ ,

which is parallel to ((b + c)′ p b+c,a ) = ℓ b+c .

28
 • We also know that ((a + b)′ (b + c)′ ) = m is parallel to (p b,a p b+c,a ) = m a ,
which is parallel to (p b,c p a+b,c ) = m c .

• Finally we know that a + b = b + a, so p b,a ∈ n(b+a) = n a+b . This means that

((a + b)′ p b,a ) = n a+b is parallel to n b+c = ((b + c) (b + c)′ ) = (p b,c (b + c)′ ).
By exercise 5.5.1, (p b+c,a p a+b,c ) is parallel to n. This means:

(b + c) + a = π→ℓ, //n (p b+c,a ) = π→ℓ, //n (p a+b,c ) = (a + b) + c.

Consequently a + (b + c) = (a + b) + c, and addition is associative.

This completes the study of addition.

5.3 Multiplication
Let 1 = ℓ ∩ n; since ℓ, m, n not to concur, one has 1 ≠ 0.

t a,b

αb
a′
1′ α
0 1 b a⋅b

5.3.1. Notation. For a, b ∈ ℓ, define:

• α = (1a ′ );
• t a,b = α b ∩ m (t is for ‘times’);
• a ⋅ b = π→ℓ, //n (t a,b ).
5.3.2. Proposition. (ℓ ∖ {0}; ⋅) is a group with identity 1.

Proof. Exercise 5.5.2.

5.4 Skew-field and coordinatisation

5.4.1. Proposition. (ℓ; +, ⋅) is a skew-field.
Both + and ⋅ are well-understood separately, so only distributive laws remain to be
checked; since ⋅ need not be commutative, there are two equations to check, namely
a ⋅ (b + c) = a ⋅ c + b ⋅ c and (a + b) ⋅ c = a ⋅ c + b ⋅ c. This is extraordinarily painful,
though always by repeated use of the Desargues property. One should have a (brief)
look at Hilbert’s Grundlagen der Geometrie.

29
5.4.2. Proposition. A ≃ A2 (F).

Proof. Not as trivial as it seems. One key step is to show that the isomorphism type of
F = (ℓ; +, ⋅) depends on none of the choices we made (we fixed ℓ, m, n, 0, 1, . . . ). This
uses the Desargues property. Then ‘the same coordinates can be used everywhere’, and
one retrieves A ≃ A2 (F).

5.4.3. Corollary. F is unique up to isomorphism.

Proof. Suppose P is desarguesian and gives rise to two fields F and F′ . Then P ≃
P2 (F). But when applied to P2 (F), one can check step by step that Hilbert’s method
yields F. It follows that F′ ≃ F: the coordinatising skew-field is unique up to isomorph-
ism.

5.4.4. Remark (coordinatising the uncoordinatisable). One can still attempt to coordin-
atise non-desarguesian affine planes using weaker algebraic structures than skew-fields.
So-called ternary rings serve this purpose. A ternary ring is a pair (R, T) where R is a set
and T∶ R 3 → R a ternary operation of some form, which aims at capturing the behaviour
of (a, x, b) ↦ ax + b. These algebraic structures are arguably not ‘core mathematics’. 8

5.5 Exercises
5.5.1. Exercise. In a desarguesian affine plane, let x 1 , x 2 , y 1 , y 2 , r, s be such that (x 1 x 2 )//
(y 1 r)//(y 2 s) and (y 1 y 2 )//(x 1 s)//(x 2 r). Prove that if (x 1 y 1 )//(x 2 y 2 ), then (rs)//(x 1 y 1 ).

y2 s

y1 r

x1 x2

5.5.2. Exercise (multiplication).

1. Show that multiplication is well-defined.
2. Prove that each a ∈ ℓ ∖ {0} has a left-inverse.

3. Let p = a ⋅ b; also let q 1 = (1p′ ) ∩ n b and q 2 = (1p′ )c ∩ n b⋅c . Draw and understand
the following picture.
8 One possible first reference for near-structures is C. Weibel, Survey of non-desarguesian planes, Not. ams,

54 no. 10, pp. 1294–1303, November 2007.

30
 (p ⋅ c)′

βc

(b ⋅ c)′
q2

β
p′

b′ q1
a ′

0 1 b c b⋅c
α αb

4. Prove that o, q 1 , q 2 are collinear. (Hint: introduce r = (oq 1 ) ∩ (1p′ )c .)

5. Prove that (bp′ ) and ((b ⋅ c) (p ⋅ c)′ ) are parallel.
6. Deduce that multiplication is associative.
7. Conclude that each a ∈ ∖{0} has a two-sided inverse.
5.5.3. Exercise. Read Chapter V of Hilbert’s Grundlagen der Geometrie (in any lan-
guage).
5.5.4. Exercise. Let J 9 = {0, ±1, ±i, ± j, ±k} with:
• addition: associative, with 3x = 0 and such that 1 + i = j and 1 − i = k;
• multiplication: like in quaternions.

1. Write the addition table. Prove that J 9 is not a skew-field.

2. (Harder.) Prove that P2 (J 9 ) (thought of as non-zero triples [x, y, z] modulo left-
action) is a projective plane.

6 The Pappus property

Abstract. § 6.1 defines the Pappus property. § 6.2 proves that ‘Pappus implies Desar-
gues’. § 6.3 returns to coordinatisation: a projective plane has the Pappus property
iff it is isomorphic to P2 (F) for some commutative field F.

31
6.1 Statement of the property
6.1.1. Definition (pappian projective plane). A projective plane is pappian if it has the
projective Pappus property, which is the following axiom.
If a, b, c are collinear points and a ′ , b ′ , c ′ are collinear points (being six non-
collinear points in total), then the intersection points a ′′ = (cb′ ) ∩ (bc ′ ),
b ′′ = (ac ′ ) ∩ (ca ′ ), and c ′′ = (ab ′ ) ∩ (ba′ ) are collinear.
o a b c

c′
b′
a′

b ′′ c ′′
a ′′

a1 a2 o a1 a5 a3

a6 a3 a6
a2
a4

a5 a4 b′′ c ′′
a ′′
How to remember Pappus: draw two pictures, erase one. 1. Draw an ordinary
hexagon. You should see ‘opposite sides’. Imagine that points a 1 , a 3 , a 5 are
on one line ℓ 1 , and points a 2 , a 4 , a 6 are on another line ℓ 2 . 2. Draw two
lines ℓ 1 and ℓ 2 . Hang your hexagon on ℓ 1 and ℓ 2 (do not care for convexity,
not an incidence-theoretic notion) [dashed]. 3. Rematch opposite sides of
the hexagon: (a 1 a 2 ) with (a 4 a 5 ), and so on (it is worth looking at the left
picture here). Get resulting intersections [dotted]. 4. Draw the ‘Pappus line’
(a′′ − b′′ − c ′′ ) and erase the left picture.

6.1.2. Definition (pappian affine plane). An affine plane A is pappian if it has the affine
Pappus property, which is the following axiom.
Let a, b, c be collinear points and c ′ , b ′ , a′ be collinear points (being six
non-collinear points in total). If (ab ′ )//(ba′ ) and (c ′ b)//(b ′ c), then (ac ′ )//
(ca ′ ).

o c′ b′ a′
This is just one affine Pappus configuration. Draw the other one.

32
 As one expects, pappianism is preserved under projectivisation/affinisation.

6.2 Pappus implies Desargues

6.2.1. Theorem (Hessenberg). Let P be a projective plane. If P is pappian, then it is also
desarguesian. The same holds of affine planes.

Proof. Let (o, a, b, c, a′ , b ′ , c ′ , a ′′ , b ′′ , c ′′ ) be a Desargues configuration; we aim at

showing that a′′ , b ′′ , c ′′ are collinear. We shall introduce four new points and apply
Pappus’ Theorem three times. Let:

• i = (ab) ∩ (a ′ c ′ );
• j = (oi) ∩ (bc);
• j′ = (oi) ∩ (b ′ c ′ );

• k = (oa) ∩ (bc ′ ).

a′

a k b′
b
o j
i
j′
c
c′

a ′′ b′′ c ′′
Hard to read, for reference only. The initial configuration,
with known collinearities; points o, i, j, j′ are collinear.

Step 1. b ′′ , j, k are collinear.

Verification. Consider the triples a, i, b and o, c, c ′ , in this order. Notice that they are
collinear by construction and the Desargues assumption.
b
i
a

o c c′
No notion of betweenness, but order matters.

33
 Moreover:
• (ac) = (ab ′′ ) and (ic ′ ) = (a′ c ′ ) = (a ′ b ′′ ) so (ac) ∩ (ic ′ ) = b′′ ;

• by definition, (oi) ∩ (cb) = j;

• by definition again, so (ao) ∩ (bc ′ ) = k.
By Pappus’ Theorem, we find the desired collinearity. ◇

Step 2. c ′′ , j′ , k are collinear.

Verification. Now consider the triples i, a ′ , c ′ and o, b, b ′ . They are also collinear by
construction and the Desargues assumption.

c′
a′
i

o b b′

Now:

• (ib) = (ab) = (ac ′′ ) and (a ′ b ′ ) = (a′ c ′′ ), so (ib) ∩ (a′ b′ ) = c ′′ ;

• by definition, (io) ∩ (c ′ b ′ ) = j′ ;
• (oa ′ ) = (oa) = (ok) and (bc ′ ) = (bk), so (oa′ ) ∩ (bc ′ ) = k.

Again by Pappus’ Theorem, the desired points are collinear. ◇

Step 3. a ′′ , b ′′ , c ′′ are collinear.

Verification. Finally consider the triples i, j, j′ and b, c ′ , k. By construction, they are

collinear.
j′
j
i

b c′ k

But:
• (ic ′ ) = (a′ c ′ ) = (a ′ b ′′ ) and ( jk) = ( jb′′ ) by Step 1, so (ic ′ ) ∩ ( jb′′ ) = b ′′ ;
• (b j) = (bc) = (ba′′ ) and (c ′ j′ ) = (b ′ c ′ ) = (b ′ a ′′ ), so (b j) ∩ (c ′ j′ ) = a ′′ ;
• (ib) = (ab) = (ac ′′ ) and ( j′ k) = ( j′ c ′′ ) by Step 2, so (ib) ∩ ( j′ k) = c ′′ .

Pappus’ Theorem gives the conclusion. ◇

34
 This is exactly the desired collinearity for the Desargues property. Of course we
treated only the case where all points where distinct. Rigorously speaking one ought
to handle the degenerate cases as well, but this is tedious.
As for the affine case, if A is pappian, then so is its projectivisation Â. By the
projective case, Â is therefore desarguesian. Now removing the line at infinity we just
added, A is desarguesian as well.

6.2.2. Remarks.
• Not treating annoying cases is always dangerous: Hessenberg’s proof was no more
fully rigorous than ours.9
• The converse fails: there exist desarguesian, non-pappian affine/projective planes.
• However, for finite planes, Desargues implies Pappus: Proposition 6.3.4 of the next
subsection.

6.3 Pappus’ affine and projective theorems

6.3.1. Theorem (Pappus’ theorem). Let F be a skew-field. Then A2 (F) is pappian iff
P2 (F) is pappian iff F is commutative.

Proof.
Step 1. A2 (F) is pappian iff P2 (F) is.

Verification. Add or remove lines, bearing in mind Proposition 2.3.1. ◇

The rest of the proof is easier in affine planes.

Step 2. If A2 (F) is pappian, then F is commutative.

Verification. Let λ, µ ∈ F× . Return to the description of affine planes A2 (F), notably

Step 2 of Proposition 1.3.2: we have a description of parallelism. Let a = (1, 0), b =
(λ, 0), c = (µ ⋅ λ, 0) and c ′ = (0, 1), b ′ = (0, µ), a ′ = (0, λ ⋅ µ).
Ð→ Ð→ Ð→ Ð→
Notice that ba ′ = λ ⋅ ab ′ and b ′ c = µ ⋅ c ′ b, so (ba′ )//(ab ′ ) and (b ′ c)//(c ′ b). By
Ð→ Ð→
the Pappus property, (ac ′ )//(ca ′ ). Therefore there is ν ∈ F× such that ca ′ = ν ⋅ ac ′ .
In coordinates, λ ⋅ µ = ν and −µ ⋅ λ = −ν. This implies λµ = µλ, viz. commutativity
of F. ◇

Step 3. If F is commutative, then A2 (F) is pappian.

Verification. There are two cases.

Case 1. Suppose lines (ac) and (c ′ a′ ) are parallel.

9 See A. Seidenberg. Pappus implies Desargues. Amer. Math. Monthly 83(3), 1976.

35
 a τ1 b τ2 c

c′ b′ a′

Let τ 1 be the translation mapping a to b. Then since (b ′ c ′ ) = (c ′ a ′ )//(ac) =

(ab) which is the axis of the translation, τ 1 ((b ′ c ′ )) = (c ′ a′ ).
Also, τ 1 ((ab ′ ))//(ab ′ )//(ba′ ), but τ 1 ((ab ′ )) contains τ 1 (a) = b. So by AP2 ,
one has τ 1 ((ab ′ )) = (ba ′ )
In particular, τ 1 (b′ ) ∈ (c ′ a′ ) ∩ (ba ′ ) and τ 1 (b ′ ) = a′ .
One can prove likewise that the translation τ 2 taking b to c also takes c ′ to b ′ .
So the translation τ = τ 2 τ 1 = τ 1 τ 2 takes a to c and c ′ to a ′ . As a conclusion
(ac ′ )//τ((ac ′ )) = (ca′ ).
We merely used commutativity of composition of translations; i.e. that
(F2 ; Ð
→
0 , +) is an abelian group. So far commutativity of F itself (i.e. of mul-
tiplication ⋅) was not required.
Case 2. This is more interesting. Suppose that (ac) and (c ′ a ′ ) meet at say o.
This is the picture we gave earlier, and we shall adapt the translation argument
using another tool.
Let h 1 be the homothety with centre o mapping a to b, and h 2 be the homothety
with centre o mapping b to c.
Notice that:
b = h 1 (a) ∈ h 1 ((ab ′ ))//(ab′ )//(ba′ ),
so h 1 ((ab′ )) = (ba′ ). Moreover, since h 1 is a homothety, h 1 ((ob′ )) = (ob′ ) =
(oa ′ ). Therefore h 1 (b ′ ) = a ′ . Likewise, one proves h 2 (c ′ ) = b ′ .
As a conclusion,

(ac ′ )//h 2 ○ h 1 ((ac ′ )) = (h 2 h 1 (a) h 2 h 1 (c ′ )) = (c h 2 h 1 (c ′ )).

By commutativity of F, two homotheties with same centre commute. So actu-

ally h 2 h 1 (c ′ ) = h 1 h 2 (c ′ ) = a ′ , and now (ac ′ )//(ca ′ ), as desired. ◇

This completes the proof.

6.3.2. Corollary (Pappus = coordinatisable in a commutative field). Let P be a projective

plane. Then P is pappian iff P ≃ P2 (F) for some commutative field F.

Proof. The converse implication is one direction of Pappus’ Theorem 6.3.1. For the
direct implication, let P be pappian. Then by Hessenberg’s Theorem 6.2.1, P is desar-
guesian. Therefore by Hilbert’s Theorem 3.3.1, there is a skew-field F with P ≃ P2 (F).
Now P2 (F) is pappian, and by Pappus’ Theorem 6.3.1, F is commutative.

6.3.3. Remarks.
• Our argument goes through Hessenberg’s and Hilbert’s theorems. I am not aware
of a quicker proof directly producing coordinates from the Pappus property.

36
 • As a consequence of Corollary 6.3.2, the theory of affine/projective planes is un-
decidable.10

Return to the implication ‘Pappus implies Desargues’. The general converse fails: let
F be any non-commutative skew-field, for instance the quaternions H. Then P2 (H) is
desarguesian but not pappian. However, in the finite case, the converse does hold.
6.3.4. Proposition. Let P be a finite, desarguesian projective plane. Then P is pappian.

Proof. We use Hilbert’s Coordinatisation Theorem 3.3.1. Let P be desarguesian. Then

there is a skew-field F with P ≃ P2 (F). Now F is finite, so by Wedderburn’s Theorem
from the Introduction, F is commutative. By Pappus’ Theorem 6.3.1, P is pappian.

6.3.5. Remark. This argument given here is quite algebraic as it relies on the introduc-
tion of a coordinate system, and its study using Wedderburn’s Theorem. The first essen-
tially geometric proof of Proposition 6.3.4 was found by Tecklenburg.11 But finitess is
important, so number-theoretic estimates must play a role. Tecklenburg’s proof relies,
in the very end, on cyclotomic polynomials—as a matter of fact, the very same kind of
argument that Witt so aptly introduced in order to prove. . . Wedderburn’s Theorem.

6.4 Exercises
6.4.1. Exercise. Draw the missing case of the affine Pappus configuration.

6.4.2. Exercise. Let P be a pappian projective plane. Show that P̌ λ is pappian (with respect
to any line). Conversely, show that the projectivisation of a pappian affine plane remains
pappian.

7 Semi-linear automorphisms of vector spaces

Abstract. § 7.1 returns to the general linear group over a skew-field. § 7.2 (op-
tional) shows how both linear automorphisms and field automorphisms induce
proportionality-perserving additive automorphisms of a vector space. § 7.3 intro-
duces semi-linear maps and studies the group ΓΛ(V ) of semi-linear automorph-
isms of a vector space.

Let F be a fixed skew-field. Eventually we want to describe automorphisms of P2 (F);

in the process we must understand certain transformations of F3 , and the general case
is worth exploring.
Let V be a left-vector over F. We reserve variables λ, µ for scalars, even implicitly,
and x, y for vectors, dropping arrows. Since we reserve the word ‘collinearity’ for points
in incidence geometries, we prefer to say that two non-zero vectors x, y are proportional
if there is λ ∈ F with y = λx. This is an equivalence relation. The quotient set is naturally
the set of vector lines in V .
10 J.
Makowsky. Can one design a geometry engine? On the (un)decidability of certain affine Euclidean
geometries. Ann. Math. Artif. Intell. 85 (2), pp. 259—291, 2019.
11 H. Tecklenburg. A proof of the theorem of Pappus in finite Desarguesian affine planes, J. Geom., 30(2),

pp. 172–181, 1987.

37
7.1 The general linear group
The definition should be known, but skew-commutativity has unexpected effects.
7.1.1. Definition (general linear group). Let GL(V ) be the group of linear automorph-
isms, viz. of additive automorphisms such that for all λ, v one has g(λv) = λg(v).
These objects are also called automorphisms of the vector space V .
7.1.2. Remarks.
• Scalar maps need not be linear. More precisely let λ ∈ F. If the space is non-trivial,
then (v ↦ λv) is linear iff λ ∈ Z(F), the centre of F.
• We do not claim that F× Id is a normal subgroup of GL(V ). As a matter of fact,
F× Id need not be contained in GL(V ). (This is exactly the first remark.)
• In finite dimension, the map (v ↦ λv) is not represented by the diagonal matrix:

⎛λ ⎞
D=⎜ ⋱ ⎟.
⎝ λ⎠

Indeed linear maps (in finite dimension) can be represented by matrices acting
from the right on row vectors, and the linear map R ↦ R ⋅ D is not the scalar
action λ (the latter being ‘multiplication from the left’). One may check this with
quaternions.

7.2 Two constructions

This subsection serves as an introduction to the next; it may be skipped at the expense
of increased conceptual difficulty. We describe two constructions of proportionality-
preserving, additive automorphisms of F3 . (This would work in any dimension but the
subsection is pedagogical in nature.) Even if F is commutative, non-linear phenomena
occur.
7.2.1. Lemma (linear automorphisms induce projective automorphisms). Any automorph-
ism of the vector space F3 induces an automorphism of the projective plane P2 (F).

Proof. Bear in mind that P2 (F) = P(F3 ) is the set of (left-)vector lines in F3 , with lines
the set of (left-)vector planes in F3 . Any g ∈ GL(F3 ) preserves (left-)proportionality,
so induces a map P(g)∶ P2 (V ) → P2 (V ) (in functorial notation), or [g]∶ P2 (V ) →
P2 (V ) (in equivalence class notation). Explicitly, define γ = P(g) = [g] by letting:

γ([v]) = [g(v)],

which is well-defined and bijective. When acting on vector subspaces, f preserves

maps vector planes to vector planes. Therefore γ maps projective lines to projective
lines, and preserves incidence (inclusion). So γ is an automorphism of P2 (F).

7.2.2. Lemma (skew-field automorphisms induce projective automorphisms). Any auto-

morphism σ ∈ Aut(F) of the skew-field F induces an automorphism σ of the projective
plane P2 (F).

38
 Proof. Let σ∶ F → F be a skew-field automorphism. This induces a map V (σ)∶ F3 →
F3 (in functorial notation). Explicitly, extend σ to a map f = σ̂ on F3 by letting, in
coordinates:
f ((x, y, z)) = (σ(x), σ(y), σ(z)).
The map f is an additive bijection, but not a linear one: f (λ ⋅ v) = σ(λ) ⋅ f (v) need
not equal λ ⋅ f (v). (Technically, f is called ‘σ-semi-linear’.)
Still, f preserves proportionality. Therefore (still in functorial notation), it induces
a map P2 (σ)∶ P2 (F) → P2 (F). Explicitly, define σ = P2 (σ) = P( f ) = [ f ] by letting:

σ([v]) = [ f (v)].

Now σ is clearly an automorphism of P2 (F).

7.2.3. Remarks.
• In projective coordinates, σ([λ, µ, ν]) = [σ(λ), σ(µ), σ(ν)].
• The proof decomposes into two steps: first show that σ induces f = σ̂ on F3 , then
quotient down to an automorphism σ = P( f ) = P2 (σ) of P2 (F) = {non-zero
vectors}/proportionality. This motivates the study of maps like f .

7.3 The group of semi-linear automorphisms

We study (additive) automorphisms of V which preserve proportionality. Even in the
commutative case, this is more general than GL(V ). Recall that GL(V ) is the group
of additive automorphisms which commute with the scalar action, viz. which preserve
proportions. We weaken this condition.
7.3.1. Definition (semi-linear map). Let σ ∈ Aut(F) be a field automorphism. An ad-
ditive morphism f ∶ V → V is σ-semi-linear if for all λ ∈ F and x ∈ V one has:

(∀λ)(∀v)[ f (λv) = σ(λ) f (v)].

(This is defined only for additive maps.) The sum of semi-linear maps attached to
distinct automorphisms is not semi-linear. This is a strong indication that we should
focus on the group structure and not look for ring structures. Indeed, the composition
of two semi-linear maps is a semi-linear map, for the composition of automorphisms;
and the inverse of a semi-linear additive automorphism is one, for the inverse field auto-
morphism.
7.3.2. Definition (group of semi-linear automorphisms). Let ΓΛ(V ) be the group of
semi-linear additive automorphisms of V .
7.3.3. Examples.
1. Every g ∈ GL(V ) is linear, viz. IdF -semi-linear, so GL(V ) ≤ ΓΛ(V ). (Actually
GL(V ) ⊴ ΓΛ(V ) as one sees in Step 3 below.)
2. F× Id ≤ GL(V ). For λ ∈ F× , let σ λ (µ) = λµλ−1 be conjugation by λ, a field
automorphism. Then (v ↦ λv) is σ λ -semi-linear.
3. More generally, if σ ∈ Aut(F), then σ̂ of Lemma 7.2.2 is σ-semi-linear.

39
 The following reduces semi-linear, additive automorphisms of a vector space into
two parts: linear and field-automorphic.

7.3.4. Proposition. Suppose V has dimension ≥ 2. Then:

(i) ΓΛ(V ), the group of semi-linear additive automorphisms, is exactly the group of ad-
ditive automorphisms preserving proportionality. (In symbols, ΓΛ(V ) = Aut(V ; +, ∼
) where ∼ is proportionality.)
(ii) ΓΛ(V ) ≃ GL(V ) ⋊ Aut(F).

(iii) Every f ∈ ΓΛ(V ) decomposes in a unique way as f = g ○ σ̂ where g ∈ GL(V ) and

σ̂ is induced by some σ Aut(F).
7.3.5. Remarks.
• In dimension 1, every additive automorphism preserves proportionality so the
proposition fails.
• One may not drop additivity: otherwise, stabilise each line and permute inside
each line, independently. This preserves proportionality.
• Even in the commutative case, GL(V ) may be proper in ΓΛ(V ): it only depends
on Aut(F). (The phenomenon went unnoticed because Aut(R) = {Id}.)

Proof.
Step 1. ΓΛ(V ) preserves proportionality.

Verification. If f ∈ ΓΛ(V ) is σ-semi-linear, then for given λ and v, one has f (λv) =
σ(λ) f (v). So proportionality is preserved. ◇

Step 2. An additive automorphism preserving proportionality is semi-linear.

Verification. Let f ∶ V → V be an additive automorphism preserving proportionality.

For λ ∈ F× and x ∈ V ∖ {0} there is σx (λ) ∈ F with f (λx) = σx (λ) f (x). Also define
σx (0) = 0.
If x and y are (left-)independent vectors, then so are f (x) and f (y). Now:

σx (λ) f (x) + σ y (λ) f (y) = f (λx) + f (λy)

= f (λ(x + y))
= σx+y (λ) f (x) + σx+y (λ) f (y).

Using independence, we see that σx (λ) = σx+y (λ) = σ y (λ) does not depend on x;
we write it σ. (The argument fails if dim V ≤ 1.) By construction, σ(1) = 1.
Now:

σ(λ + µ) f (x) = f ((λ + µ)x)

= f (λx) + f (µx)
= (σ(λ) + σ(µ)) f (x),

40
 and:

σ(λ ⋅ µ) f (x) = f ((λ ⋅ µ)x)

= f (λ ⋅ (µx))
= σ(λ) f (µ(x))
= σ(λ) ⋅ (σ(µ)x).

These imply additivity and multiplicativity, respectively. So σ∶ F → F is a field em-

bedding. Finally consider f −1 , which preserves proportionality: the same applies,
so there is a field embedding τ∶ F → F with f −1 (λx) = τ(λ) f −1 (x). Therefore for
arbitrary λ:
λx = f ( f −1 (λx)) = f (τ(λ) f −1 x) = σ(τ(λ))x,
which proves that σ is onto, hence a field automorphism. ◇

This proves (i). We now embed Aut(F) into ΓΛ(V ). For σ ∈ Aut(F) let σ̂ be
the automorphism defined as follows. Fix a (left-)basis B = {e i ∶ i ∈ I} of V . For
x = ∑ λ i e i let:
σ̂(x) = ∑ σ(λ i )e i .
This is a σ-semi-linear automorphism. Let Σ̂ be the set of operators σ̂; clearly Aut(F) ≃
Σ̂ ≤ ΓΛ(V ). (The construction depens on the choice of B.)
Step 3. ΓΛ(V ) = GL(V ) ⋊ Σ̂.

Verification. We prove GL(V ) ⊴ ΓΛ(V ). If g ∈ GL(V ) and f ∈ ΓΛ(V ) is σ-semi-

linear, then for arbitrary λ, x one has:

( f −1 g f )(λx) = f −1 ○ g(σ(λ) f (x)) by σ-semi-linearity of f

= f (σ(λ)g( f (x)))
−1
by linearity of g
= σ (σ(λ))( f
−1 −1
g f )(x) by σ -semi-linearity of f −1 ,
−1

so f −1 g f is linear.
We prove GL(V ) ∩ Σ̂ = {Id}. If f ∈ GL(V ) ∩ Σ̂, then f = σ̂ is linear, so σ = Id
and f = Id ˆ = Id.
It remains to show that GL(V ) and Σ̂ generate ΓΛ(V ). Let f ∈ ΓΛ(V ); say f is
σ-semi-linear. We give two arguments.
• First option. Let g = f ○ σ̂ −1 , an additive automorphism which is σ ○ σ −1 -semi-
linear. This means that g is linear and we are done.

• Second option. The image f (B) is a basis and GL(V ) acts transitively on bases,
so up to composing on the left with some g ∈ GL(V ) we may assume that
f ∈ ΓΛ(V ) fixes B pointwise. But f is σ-semi-linear, so it coincides with σ̂
everywhere; hence f ∈ Σ̂ and we are done.

So ΓΛ(V ) ≃ GL(V ) ⋊ Aut(F), proving (ii). And (iii) is a mere rephrasing.

We now have full understanding of semi-linear automorphisms of V .

41
7.4 Exercises
7.4.1. Exercise. Let F be a skew-field and V be a non-trivial left-vector space. Prove that
Z(GL(V )) = Z(F× ) Id.

8 Automorphisms of projective planes

Abstract. Let P be a projective plane and G = Aut(P). There are two cases:
• P is coordinatisable (= desarguesian): we know a lot about G;
• P is non-coordinatisable (= non-desarguesian): nothing general can be said
about G as an abstract group.

It must be borne in mind that the present section deals mostly with coordinatisable
(= desarguesian) projective planes.

8.1 A lemma on coordinates

The following technical lemma confirms that the coordinate system is ‘built in’ the in-
cidence geometry P2 (F). Though it naturally belongs to the flow of the argument of
Theorem 8.2.1 we present it here as its proof is long.12

8.1.1. Lemma. If α ∈ Aut(P2 (F)) fixes [1, 0, 0], [0, 1, 0], [0, 0, 1] and [1, 1, 1], then it is
of the form P2 (σ) = [σ̂] for some skew-field automorphism σ.
Proof not covered in class.

Proof.

Consider the following variable points for λ, µ ∈ F:

q λ = [λ, 0, 1],
r µ = [0, µ, 1],
s λ, µ = [λ, µ, 1].

Note that (λ, 0, 1) = λe 1 + e 3 ∈ ⟨e 1 , e 3 ⟩, so q λ ∈ (p 1 p 3 ); moreover q λ ≠ p 1 . Likewise

r µ ∈ (p 2 p 3 ) ∖ {p 1 }. Finally (p 1 r µ ) ∩ (p 2 q λ ) = {s λ, µ }.

12 I follow A. Keedwell, Self-collineations of desarguesian projective planes, Am. Math. Monthly, 82 no. 1,

pp. 59–63, January 1975.

42
Step 1. There is a function σ∶ F → F such that for all λ ∈ F:

φ([λ, 0, 1]) = [σ(λ), 0, 1].

Verification. Since φ fixes each p i , it stabilises each line through two of them. There-
fore there is a map σ∶ F → F with φ(q λ ) = q σ(λ) . (We cannot prove yet that it is a
field automorphism.) In coordinates:

φ([λ, 0, 1]) = [σ(λ), 0, 1],

as claimed. ◇

Step 2. One also has φ([λ, µ, 1]) = [σ(λ), σ(µ), 1]. Moreover, σ(1) = 1.

Verification. In a similar way there is τ∶ F → F with φ([0, µ, 1]) = [0, τ(µ), 1]. But
then:

φ(s λ, µ ) = φ((p 1 r µ ) ∩ (p 2 q λ ))
= (p 1 r τ(µ) ) ∩ (p 2 q σ(λ) )
= s σ(λ),τ(µ) ,

in coordinates: φ([λ, µ, 1]) = [σ(λ), τ(µ), 1].

But when λ = µ the point r λ,λ is on (p 1 p 4 ), which is stabilised by φ. So φ(r λ, λ )
is some r ν,ν . In coordinates:

[σ(λ), τ(λ), 1] = [ν, ν, 1].

This proves σ(λ) = τ(λ): functions are equal.

It follows φ([λ, µ, 1]) = [σ(λ), σ(µ), 1]. And clearly, σ(1) = 1. ◇

Step 3. One even has φ([1, λ, µ]) = [1, σ(λ), σ(µ)].

Verification. Likewise, there is σ ′ ∶ F → F such that:

φ([1, λ, µ]) = [1, σ ′ (λ), σ ′ (µ)],

and σ ′ (1) = 1. But then:

[1, σ(λ), 1] = φ([1, λ, 1]) = [1, σ ′ (λ), 1],

so σ ′ = σ. ◇

Step 4. σ is a field automorphism and φ([λ, µ, ν]) = [σ(λ), σ(µ), σ(ν)].

43
 Verification. By the above,

[1, σ(λ−1 ), σ(λ−1 )] = φ([1, λ−1 , λ−1 ])

= φ([λ, 1, 1])
= [σ(λ), 1, 1]
= [1, σ(λ)−1 , σ(λ)−1 ],

proving σ(λ−1 ) = σ(λ)−1 . Therefore:

[σ(λ), σ(λµ), 1] = φ([λ, λµ, 1])

= φ([1, µ, λ−1 ])
= [1, σ(µ), σ(λ−1 )]
= [σ(λ), σ(λ)σ(µ), 1],

proving multiplicativity.
Hence:

φ([λ, µ, ν]) = φ([1, λ−1 µ, λ−1 ν])

= [1, σ(λ−1 )σ(µ), σ(λ−1 )σ(ν)]
= [σ(λ), σ(µ), σ(ν)].

This and σ(1) = 1 implies that φ fixes [1, 0, 1] and [0, 1, 1]. It therefore stabil-
ises the line through them, which is exactly {[λ, µ, λ + µ] ∶ λ, µ ∈ F}. Therefore
φ([λ, µ, λ + µ]) has the same form, which proves additivity of σ. Surjectivity is no
issue; it is a field automorphism. ◇

It is now clear that φ is induced by σ ∈ Aut(F), in our notation φ = σ = P2 (σ).

8.2 The projective version

We now let V = F3 and let ΓΛ(F3 ) act on F3 ; since it preserves proportionality, it induces
an action on P2 (F). For f ∈ ΓΛ(F3 ) we let [ f ] be the induced map. If p ∈ P2 (F) has
projective coordinates p = [x, y, z], viz. if p = [v] with v = (x, y, z), then:

[ f ](p) = [ f (v)].

8.2.1. Theorem.
(i) The action of ΓΛ(F3 ) on P2 (F) induces a group homomorphism ΓΛ(F3 ) → Aut(P2 (F)).
(ii) The kernel of the action is F× Id, the group of scalar maps.

(iii) The action induces Aut(P2 (F)).

(iv) Every α ∈ Aut(P2 (F)) factors uniquely as α = γ ○ β with γ = [g] and β = [σ̂] for
some g ∈ GL(V ) and σ ∈ Aut(F).
8.2.2. Remarks.

• Have another look at Remark 7.1.2; F× Id is a normal subgroup of ΓΛ(F3 ) be need

not be contained in GL(F3 ).

44
 • In (iv), γ and β are well-defined but g is not; g is unique only up to Z(F× ) Id =
Z(GL(F3 )).

Proof. Let f ∈ ΓΛ(V ). Then f ∶ F3 → F3 maps vector planes to vector planes. There-
fore, the induced map P f ∶ P2 (F) → P2 (F) maps projective lines to projective lines. It
is obviously incidence preserving. Therefore P f ∈ Aut(P2 (F)).
Step 1. The kernel of the action is F× Id.

Verification. Suppose f ∈ ΓΛ(V ) acts trivially on P2 (F). Then for any x ∈ F3 there
is λ x ∈ F with f (x) = λ x x. Using the ordinary tricks, λ x does not depend on x. So
f = λ Id ∈ F× Id. (Recall that we do not assert linearity of these maps: neither did we
suppose linearity of f .) The converse is obvious. ◇

To prove surjectivity is not as easy. Let X be the set of 4-tuples of points in P2 (F)
no three of which are collinear.
8.2.3. Lemma. The group PGL3 (F) acts transitively on X, viz. if (p 1 , p 2 , p 3 , p 4 ) and
(q 1 , q 2 , q 3 , q 4 ) are in X, then there is φ ∈ GL3 (F) such that ψ = P(φ) maps each p i to
qi .

Proof. As always in group theory, we may choose p 1 , p 2 , p 3 , p 4 . Let e 1 = (1, 0, 0) ∈

F3 and p 1 = [e 1 ]; define e 2 , e 3 , p 2 , p 3 likewise. Also let e 4 = (1, 1, 1); notice that any
three of the e i ’s generate F3 , so any three of the p i ’s are non-collinear in P(F3 ) =
P2 (F). It is therefore enough to prove the lemma with p i = [e i ].
Write each q i in projective coordinates, say q i = [ f i ] with f i = (a i ,1 , a i ,2 , a i ,3 ) ∈
F3 . Let:
⎛ a 1,1 a 1,2 a 1,3 ⎞
A = ⎜a 2,1 a 2,2 a 2,3 ⎟ .
⎝ a 3,1 a 3,2 a 3,3 ⎠
Introduce unknown scalars λ 1 , λ 2 , λ 3 and consider the matrix:

⎛ λ 1 a 1,1 λ 1 a 1,2 λ 1 a 1,3 ⎞

M λ = ⎜ λ 2 a 2,1 λ 2 a 2,2 λ 2 a 2,3 ⎟ .
⎝ λ 3 a 3,1 λ 3 a 3,2 λ 3 a 3,3 ⎠

Regarldess of the λ i ’s, one has (1, 0, 0) ⋅ M λ = λ 1 ⋅ (a 1,1 , a 1,2 , a 1,3 ) = λ 1 ⋅ f 1 , so p 1 ⋅

[M λ ] = q 1 and likewise for indices 2 and 3. So it suffices to find λ 1 , λ 2 , λ 3 such that
p 4 ⋅ [M λ ] = q 4 .
Now observe that:

(1, 1, 1) ⋅ M λ = (λ 1 , λ 2 , λ 3 ) ⋅ A.

Therefore, p 4 ⋅ [M λ ] = [λ 1 , λ 2 , λ 3 ] ⋅ [A]. But A maps the independent triple e 1 , e 2 , e 3

to the independent triple f 1 , f 2 , f 3 . So it is left-invertible: for any row R there is a
row S such that S ⋅ A = R. We apply this with R = f 4 ; take suitable S and write it as
S = (λ 1 , λ 2 , λ 3 ). Then p 4 ⋅ [M λ ] = [S ⋅ A] = [R] = [ f 4 ] = q 4 , we are done.

Step 1. ΓΛ(F3 ) covers Aut(P2 (F)).

45
 Verification. Let α ∈ Aut(P2 (F)). Consider the four points p 1 = [1, 0, 0], p 2 =
[0, 1, 0], p 3 = [0, 0, 1], and p 4 = [1, 1, 1]. No three of them are collinear, meaning
(p 1 , p 2 , p 3 , p 4 ) ∈ X. Let q i = α(p i ); no three of these are collinear so (q 1 , q 2 , q 3 , q 4 ) ∈
X. By Lemma 8.2.3 there is γ = P(g) doing γ(p i ) = q i . Up to considering γ−1 ○ α, we
may therefore suppose that α fixes each p i . Then by Lemma 8.1.1, there is a skew-field
automorphism σ such that α = [σ̂]. So α is induced by ΓΛ(V ). ◇

Step 2. Proof of (iv).

Verification. We use Proposition 7.3.4. Let α ∈ Aut(P2 (F)). Then there is f ∈ ΓΛ(F3 )
with α = [ f ]. Now f = g ○ σ̂ with g ∈ GL(F3 ) and σ ∈ Aut(F). Then α = [ f ] =
[g] ○ [σ̂] has the desired form.
It remains to check uniqueness. If also α = γ 2 ○ β 2 with γ 2 = [g 2 ] and β 2 = [σ̂2 ],
then [g 2−1 g] = [σ̂2 σ̂ −1 ], so letting h = g 2−1 g ∈ GL(F3 ) and τ = σ2 σ −1 ∈ Aut(F), we
have [h] = [τ̂]. For e i in the chosen basis, there is λ i such that h(e i ) = λ i e i ; by the
usual tricks λ does not depend on i. Now h = λ Id is linear, so λ ∈ Z(F× ). So g = λg 2
and therefore γ 2 = γ; we are done. ◇

This gives a full description of Aut(P2 (F)).

8.2.4. Remarks (the ‘fundamental theorem’).

• Some sources tend to forget automorphisms of type P2 (σ) because their reference
field, the real field R, has no non-trivial automorphisms.
• More in general: if V is a vector space of dimension at least 3, let P(V ) be the set of
its vector lines. Any automorphism of P(V ) preserving collinearity of projective
points is of the form P( f ) for some semi-linear automorphism f of V . There are
elementary proofs avoiding coordinates.13
• By calling this (or similar) result ‘the fundamental theorem of projective geo-
metry’, one focuses only on coordinatisable structures. This terminology hides
the theory of non-desarguesian projective planes and should be avoided.

8.3 The general case

Nothing general can be said about abstract projective planes. As a matter of fact, Mendel-
sohn proved the following difficult result.14
8.3.1. Theorem (Mendelsohn). Let G be a group, and let κ be any infinite cardinal. Then
there exists a projective plane P = (P , L, I) such that Aut(P) ≃ G and card P = max(κ, card G).

(Notice that nothing is said about finite projective with given finite automorphism
group. Determining for which finite groups G there is finite P with Aut(P) ≃ G is likely
to be extremely difficult if not impossible.)
13 C.-A. Faure, An elementary proof of the fundamental theorem of projective geometry, Geometriae Ded-

icata 90, pp. 145–150, 2002.

14 E. Mendelsohn. Every group is the collineation group of some projective plane. J. Geom., 2, pp. 97–106,

1972.

46
 However, there is another difficult result, known only for finite projective planes.15
8.3.2. Theorem (Ostrom-Wagner). Let P be a finite projective plane. If Aut(P) is doubly
transitive, then P is coordinatisable.
8.3.3. Remark. The suitable model-theoretic analogue is a research question.
Experts in projective planes tend to use the word ‘collineation’ for an incidence-
preserving (viz. collinearity-preserving) morphism. We do not see the point of introdu-
cing specific terminology where general algebraic language is sufficiently illuminating.

8.4 Exercises
8.4.1. Exercise. Let F be a skew-field.
1. Show that:
a b
[x ∶ y] ⋅ [ ] = [xa + yc ∶ xb + yd]
c d
defines an action of PGL2 (F) on F ∪ {∞}.
2. Prove that the action is always 3-transitive.
3. Prove that it is sharply 3-transitive iff F a commutative field.
4. Let F be a skew-field such that any two non-zero elements are conjugate in F× (these
do exist, but it is hard16 ). Prove that in this case, the action is even 4-transitive.

9 Group-theoretic analysis of Desargues and Pappus

Abstract. Interpretations of the Desargues and Pappus properties in group-
theoretic terms. § 9.1 interprets the Desargues propery as transitivity of a certain
family of group actions. § 9.2 introduces partial maps called projectivities; the group
of self-projectivities of a line is 3-transitive. § 9.3 characterises the Pappus property
as sharp 3-transitivity of the latter.

9.1 Group-theoretic interpretation of the Desargues property

9.1.1. Definition. Let P be a projective plane; let p be a point and ℓ be a line. Let G p,ℓ
be the group of automorphisms of P which:

• stabilise (setwise) each line through p,

• and fix (pointwise) ℓ.
The definition supposes neither p ∈ ℓ nor p ∈ ℓ.

9.1.2. Proposition. Let P be a projective plane. Then P is desarguesian iff for all choices
of p and ℓ one has:
15 T.
Ostrom, A. Wagner, On projective and affine planes with transitive collineation groups, Math. Z., 71,
pp. 186–199, 1959.
16 Exercise 3 p. 238 in: P. Cohn, P., Skew fields, theory of general division rings, Encyclopedia of Mathematics

and its Applications, vol. 57, Cambridge University Press, 1977.

47
 for each line m through p, the group G p,ℓ is transitive on m ∖ ({p} ∪ ℓ).

Proof.
Step 1. Direct implication.

Verification. Suppose transitivity. Let (o; a, b, c; a′ , b ′ , c; a′′ , b ′′ , c ′′ ) be a (non-

degenerate) Desargues configuration. We want to prove that a ′′ , b′′ , c ′′ are collinear.
Let p = o and ℓ = (a′′ b ′′ ); we have to show that c ′′ ∈ ℓ.
By assumption there is g ∈ G p,ℓ taking a to a′ . This g fixes (a ′′ b′′ ) pointwise,
so it fixes b′′ . Therefore it takes (ab′′ ) to (a ′ b ′′ ). It also fixes (cc ′ ) setwise, as a line
through o. Now (cc ′ ) ∩ (ab ′′ ) = c, while (cc ′ ) ∩ (a ′ b ′′ ) = c ′ , so g(c) = c ′ . The
same method proves g(b) = b′ . Let x = (ab) ∩ ℓ. By definition, g(x) = x. Therefore
x = g(x) ∈ g((ab)) = (a ′ b ′ ). So (ab) and (a′ b′ ) meet on ℓ. But the intersection
point is c ′′ , by definition. ◇

Step 2. Converse implication.

Verification. We only sketch the idea as it plays a key role in Artin’s proof of Hilbert’s
Coordinatisation Theorem.
Suppose desarguesianity. Take p, ℓ, then m through p and a, a ′ ∈ m ∖ ({p} ∪ ℓ).
If a = a then g = Id does; so suppose not, so that m = (aa′ ). We define a partial
′

map ǧ a,a ′ ∶ P ∖ (ℓ ∪ m ∪ {p}) → P as follows.

Let x ∈ P∖(ℓ∪m∪{p}), compute t x = (ax)∩ℓ then y x = (a ′ t x )∩(px),

and let ĝ a,a ′ (x) = y x .

We extend this ‘generically given function’ to all of P by trading the base pair
(a, a ′ ) for another one. Let b be any point in P ∖ (ℓ ∪ m ∪ {p}) and b ′ = ǧ a,a ′ (b).
Since a ≠ a ′ , one has b ≠ b ′ ; so let m′ = (bb′ ). Now for x ∈ P ∖ (ℓ ∪ m′ ∪ {p}), let
g(x) = ĝ b,b ′ (x).
This is well-defined. Indeed, if x ∉ (ℓ ∪ m ∪ m′ ∪ {p}), then ĝ a,a ′ (x) = ĝ b,b ′ (x)
reduces to a Desargues property, which holds.

48
 a′
m

a
b ′ = ǧ a,a ′ (b)
b
p

x
g(x) = ǧ a,a ′ (x) = ǧ b,b ′ (x)

ℓ
tx tx tb
(using b) (using a)

There are two ways to construct g(x); they agree.

(Whether p ∈ ℓ or p ∉ ℓ remains irrelevant.)

Notice that g is now also defined on m, and maps a to a ′ . Complete it by letting

g fix p and ℓ pointwise. By construction, g ∈ G p,ℓ sends a to a ′ . It is a collinearity-
preserving bijection. ◇

The equivalence is proved.

9.1.3. Remarks.
• The action is then sharply transitive as there is a unique g ∈ G p,ℓ taking a to a′ .
This can be seen during the proof (at each step, there is no choice), or after the
proof. Indeed, once P is known to be desarguesian, it follows P ≃ P2 (F) for some
skew-field F. Removing ℓ has the effect of taking us to the affine plane A2 (b F).
Then the action of (the restriction of) G p,ℓ is either that of translations parallel to ℓ
(p ∈ ℓ) or of homotheties with centre p. In either case, there is only one possibility
for g: this is sharp transitivity.

• There is more group theory to do here, notably in the finite case. For example,
Gleason17 proved the following.
Let P be a finite projective plane. Suppose that for all p, ℓ with p ∈ ℓ, one has
G p,ℓ ≠ {1}. Then P is desarguesian.
André18 proved a similar result, quantifying over pairs (p, ℓ) with p ∉ ℓ.

9.2 Projectivities
This subsection and the next deal with local functions, viz. maps not defined on the
whole of a projective plane. We introduce a relevant group for the local analysis of
projective planes. Its elements are called self-projectivities. Defining it requires some
terminology.
17 A. Gleason, Finite Fano Planes, Amer. Jour. Math, 78 no. 4, pp. 797–807, October 1956
18 J. André, Über Perspektivitäten in endlichen projektiven Ebenen, Arch. Math. 6, pp. 29–32, 1954.

49
9.2.1. Definition (perspectivity). Let ℓ ≠ m be two lines and p ∉ ℓ ∪ m. The perspectivity
ℓ Ð→ m is the map taking a ∈ ℓ to a ′ = (ap) ∩ m.
p

a′
a p

ℓ m

9.2.2. Definition (projectivity). A projectivity from ℓ to m is any composition of per-

spectivities ℓ Ð→ ℓ 1 . . . Ð→ ℓ n = m.
a1 an
If m = ℓ the map is called a self-projectivity of ℓ.
9.2.3. Notation (group of self-projectivities of a line). Let Proj(ℓ) be the group of self-
projectivities of ℓ.

Let ℓ, m be two lines. Since there is a perspectivity between, Proj(ℓ) ≃ Proj(m). So

the object does not depend on the line; it captures the local behaviour ‘anywhere’.
9.2.4. Remark. I do not know whether self-projectivities always extend to automorph-
isms of P.

9.2.5. Remarks (the projective line). In this remark we assume the presence of a co-
ordinate system, viz. P ≃ P2 (F) for a skew-field F.
• Since Proj(ℓ) does not depend on ℓ, we may focus on the line with equation z = 0,
viz. ℓ = {[x, y, 0] ∶ x, y ∈ F}. Hence ℓ ∈ P2 (F) can be represented as P1 (F) =
{[x ∶ y]; (x, y) ∈ F2 ∖ {(0, 0)}} = F ∪ {∞}.

• Be careful that as an incidence structure, P1 (F) = F ∪ {∞} consists of only one

line. So as an incidence structure, Aut(P1 (F)) = Sym(P1 (F)) = Sym(F ∪ {∞}).
(Geometers usually consider more structure on P1 (F) such as the cross-ratio,
which makes sense over commutative fields.)
We return to P1 (F), equipped with the notion of self-projectivites coming from
its embedding as a line in the plane.
• Every element of PGL2 (F) can be written as a self-projectivity. Now, every self-
projectivity of ℓ ∈ P2 (F) can (in theory) be computed in coordinates, but it is
unclear to me whether one remains in PGL2 (F). (There could be ‘inner field
automorphisms’, viz. automorphisms of F of the form x ↦ λ−1 x λ.)

• However, if F is commutative, then every self-projectivity comes from an element

of PGL2 (F). Therefore Proj(ℓ) ≃ PGL2 (F). The action of PGL2 (F) on P1 (F) is
known: see exercise 8.4.1.
One tends to avoid speaking of ‘the projective line’ when there is no coordinate system
behind.

9.2.6. Proposition. Let P be any projective plane with line ℓ. Then Proj(ℓ) acts 3-transitively
on ℓ.

50
 Proof. Let f 1 , f 2 be two distinct points on ℓ: we show that the stabiliser of f 1 and f 2 in
Proj(ℓ) acts transitively on ℓ ∖ { f 1 , f 2 }. Let x and y be points there.
Let m 1 ≠ ℓ be a line through f 1 , and m 2 ≠ ℓ be a line through f 2 . Choose any
p ∈ m 1 ∖ ℓ. Now let q x = (px) ∩ m 2 and q y = (py) ∩ m 2 . Finally, let r = m 1 ∩ m 2 .

m2 m1

f1 f2 x y
ℓ
qx

qy

Consider the two perspectivities ℓ Ð→ m 1 and m 1 Ð→ ℓ. One may follow on the

qx qy
picture that they have the following effect:

ℓ Ð→ m1 Ð→ ℓ
qx qy
f1 ↦ f1 ↦ f1
f2 ↦ r ↦ f2
x ↦ p ↦ y.

So the resulting self-projectivity of ℓ fixes f 1 and f 2 , and takes x to y.

9.3 Group-theoretic interpretation of the Pappus property

By Proposition 9.2.6, Proj(ℓ) is always 3-transitive on ℓ. Sharpness of the action is equi-
valent to the Pappus property.
9.3.1. Proposition. Let P be a projective plane. Then Proj(ℓ) is sharply 3-transitive on ℓ
iff P is pappian.

Proof. Suppose P is pappian. By Corollary 6.3.2, there is a commutative field F with

P ≃ P2 (F). Now the action of Proj(ℓ) on ℓ is equivalent to that of PGL2 (F) on F∪{∞},
which is known to be sharply 3-transitive (exercise 8.4.1).
For the converse we cannot use exercise 8.4.1, because we do not have a coordinate
system yet; one must give a purely geometric argument. Let (o, a, b, c, a′ , b ′ , c ′ ) be a
Pappus configuration. Let a ′′ = (bc ′ ) ∩ (cb′ ), b′′ = (ac ′ ) ∩ (ca ′ ), and c ′′ = (ab ′ ) ∩
(ba′ ). We want to show that a′′ , b′′ , c ′′ are collinear. Set ℓ′′ = (b′′ c ′′ ); we shall prove
a′′ ∈ ℓ′′ .
We introduce two auxiliary points:
• q = (aa′ ) ∩ ℓ′′ ;

51
 • z = ℓ′ ∩ ℓ′′ .
o a b c
ℓ

c′
b′
a′ ℓ′

q ℓ′′ z
b ′′ c ′′
a ′′
We must prove a ′′ ∈ (b′′ c ′′ ).

First consider the following projectivity and its effect:

ℓ Ð→ ℓ′′ Ð→ ℓ′
′a a
a ↦ q ↦ a′
b ↦ c ′′ ↦ b′
c ↦ b′′ ↦ c′
o ↦ z ↦ z

We need two more auxiliary points, which we do not show on picture:

• α ′′ = (b′ c) ∩ ℓ′′ ;
• γ′ = (bα ′′ ) ∩ ℓ′ .

Next consider this other projectivity (undefined point x plays no real role, since (xb) =
(bb ′ )):
ℓ Ð→ ℓ′′ Ð→ ℓ′
′ b b
a ↦ c ′′ ↦ a′
b ↦ x ↦ b′
c ↦ α ′′ ↦ γ′
o ↦ z ↦ z
Now the assumption that Proj(ℓ) is sharply 3-transitive on ℓ immediately implies
that for given lines m, m′ , there is a unique projectivity m → m′ taking a distinct triple
of m to one of m′ . This means that γ′ = c ′ .
Hence α ′′ ∈ (bγ ′ ) = (bc ′ ) and α ′′ ∈ (b ′ c) so α ′′ = a ′′ lies on ℓ′′ = (b′′ c ′′ ), as
desired.

9.4 Exercises
9.4.1. Exercise. Return to the proof of Step 2 in Proposition 9.1.2. Which Desargues prop-
erty is used?
9.4.2. Exercise (cross-ratio). In this exercise, F is a commutative field. We consider the
usual action of PGL2 (F) on ℓ = P1 (F). We extend field arithmetic to include ∞ with
1/0 = ∞, and so on. The cross-ratio of x 1 , x 2 , x 3 , x 4 is:
(x 1 − x 3 )(x 2 − x 4 )
γ(x 1 , x 2 , x 3 , x 4 ) = .
(x 1 − x 4 )(x 2 − x 3 )

52
Prove that the group of bijections of ℓ preserving γ is exactly PGL2 (F).
9.4.3. Exercise. Give a geometric proof that if P is pappian, then Proj(ℓ) is sharply 3-
transitive. (First prove that any product of self-projectivities is a product of at most two
perspectivities.)

10 Artin’s coordinatisation (1): dilations and translations

Abstract. This section and the next give a more conceptual proof of Hilbert co-
ordinatisation using group theory. § 10.1 introduces certain automorphisms called
dilations. § 10.2 studies those without a fixed point, called translations. § 10.3 in-
vestigates them further, assuming a weak form of the Desargues property.

Throughout A is an affine plane.

• We do not always assume desarguesianity. Instead, the results will describe the
progressive effects of increasingly stronger forms of the Desargues property.
• It will be convenient, when working with automorphisms f ∶ A → A, to use clas-
sical ‘high-school’ notation x ′ = f (x).

10.1 Dilations
The group of automorphisms of A2 (F) is F+2 ⋊ ΓΛ(F2 ). To ‘reconstruct F from A2 (F)’,
one may prefer a soluble subgroup. On the other hand, behaviour of Aut(A) in an ab-
stract affine plane A is quite unpredictable. Artin’s method is to focus on a special sub-
group of Aut(A).
10.1.1. Definition (the group of dilations). A dilation of A is an automorphism which
takes any line to a parallel line, viz. satisfying:

(∀a, b)[a ≠ b → (a ′ b ′ )//(ab)].

Let Dil(A) be the group they form under composition.

10.1.2. Remarks.
• In projective terms, a dilation is an automorphism of  fixing the line at infinity
pointwise (more precisely: the restriction to A of such an automorphism):

Dil(A) = StabAut(Â) (ℓ∞ ).

• In A2 (F), one has Dil(A) ≃ F+2 ⋊ F× .

10.1.3. Proposition.
(i) If a dilation f has two distinct fixed points, then f = IdA .

(ii) Any dilation is determined by the image of two distinct points.

53
 Proof.
(i) Suppose f fixes a and b. Using our convention to write x ′ for f (x), this means
a ′ = a and b′ = b. The proof will introduce two cases, points not on (ab) being
easier to understand. This kind of case division will be common.

a′ = a d b′ = b

• Let c ∉ (ab). Then a = a′ ∈ (a′ c ′ ) // (ac) since f is a dilation; so by

uniqueness in AP2 one has (ac ′ ) = (ac) and c ′ ∈ (ac). Likewise c ′ ∈ (bc);
these lines meet at c, so c ′ = c is fixed. This proves that any point not on
(ab) is fixed.
• Now let d ∈ (ab). Using any c ∉ (ab) (there exists such a point by AP3 ),
one has c ′ = c and d ∉ (ac). By the previous argument, d is fixed too.
Thus all points are fixed and f = IdA .
(ii) If f 1 , f 2 ∈ Dil(A) agree on two different points a ≠ b, then f 1−1 f 2 ∈ Dil(A) fixes
a and b, so by (i), one has f 1 = f 2 .

In A2 (F), dilations are either translations or homotheties. The difference is in the

number of fixed points. The next subsections build on this observation, in an abstract
affine plane.

10.2 Translations
10.2.1. Definition (translation). A translation of A is either the identity map IdA , or a
fixed point-free dilation. Let Trans(A) be their set (one must prove that it is a group).
10.2.2. Remark. The terminology is consistent with ordinary practice: if A = A2 (F),
then a translation in this sense is one in the usual sense.
Remember that given a point a and line ℓ, we use ℓ a to denote the unique line parallel
to ℓ through a. Moreover, for t a translation ≠ IdA and a a point, one has t(a) ≠ a so
line (at(a)) is well-defined.
10.2.3. Proposition.
(i) If t is a translation ≠ IdA then for any a, b ∈ A, one has (at(a))//(bt(b)).
(ii) If t ≠ IdA is a translation and b ∉ (aa′ ) then b ′ = (ab) a ′ ∩ (aa ′ )b .

a a′

b b′

54
(iii) Trans(A) ⊴ Dil(A) is a normal subgroup.
(iv) If two translations coincide at one point, they are equal.

No claims on abelianity so far.

Proof. Whenever working with only one dilation (eg. a translation), we implicitly use
notation x ′ = f (x).
(i) Suppose there is a counterexample. Then lines (aa′ ) and (bb ′ ); let c = (aa′ ) ∩
(bb′ ).
a a′

b b′

Now c ′ ∈ (a′ c ′ )//(ac) since t is a dilation. But a, a′ , c are collinear so c ′ ∈ (aa′ ).

Likewise, c ′ ∈ (bb′ ) and c ′ = c. Thus t has a fixed point: a contradiction.
(ii) Since t is a dilation, (a ′ b ′ )//(ab). But also, since t is a translation, (i) implies that
(bb′ )//(aa′ ). Now b ∉ (aa ′ ), so lines (aa′ ) and (ab) are not parallel. Therefore
lines (a ′ b ′ ) and (bb′ ) are not parallel, and their intersection is the point b ′ .
(iii) By definition, IdA is a translation; clearly the inverse of a translation is a transla-
tion; moreover if (t, d) ∈ Trans(A) × Dil(A) then dtd −1 has no fixed point. So
Trans(A) is a normal subset of Dil(A). It only remains to prove that Trans(A)
is stable under composition.
Let t 1 , t 2 be translations; we show t 2 t 1 ∈ Trans(A). We may supppose that
neither of t 1 , t 2 is IdA . To prove that the dilation t 2 t 1 is a translation, we show
that it is either fixed point-free or IdA . So suppose t 2 t 1 has a fixed point a.

t2 t 1 (a)
t 1 (b)
a
b

Pick any b ∉ (a t 1 (a)). We localise t 2 t 1 (b) on two non-parallel lines.

• On the one hand:

((t 2 (b) t 2 t 1 (b))//(b t 1 (b)) since t 2 ∈ Dil(A)

//(a t 1 (a)) since t 1 ∈ Trans(A), by (i)
//(t 2 (a) t 2 t 1 (a)) since t 2 ∈ Dil(A)
= (a t 2 (a)) since t 2 t 1 fixes a
//(b t 2 (b)) since t 2 ∈ Trans(A), by (i).

It follows that t 2 t 1 (b) ∈ (b t 2 (b)).

55
 • On the other hand, because t 2 t 1 is a dilation fixing a:

(a t 2 t 1 (b)) = (t 2 t 1 (a) t 2 t 1 (b))//(ab),

so t 2 t 1 (b) ∈ (ab).
If lines (b t 2 (b)) and (ab) are parallel, then reading the above computations we
see that (ab) = (a t 1 (a)), against the choice of b. So lines (b t 2 (b)) and (ab)
are not parallel, and their intersection is exactly b. Therefore t 2 t 1 (b) = b. But a
dilation fixing two points is the identity by Proposition 10.1.3. We are done.
(iv) If t 1 (a) = t 2 (a), then t ′ = t 2−1 t 1 is a translation by (iii). It has a fixed point, so
t 2−1 t 1 = IdA .

10.3 The (p,p)-Desargues property

The continued study of the group of translations needs a (weak) form of desarguesianity
of A.
10.3.1. Definition ((p,p)-Desargues). The (parallel, parallel)-form of the Desargues prop-
erty (for short: (p, p)-Desargues) is the following axiom.
Let (aa′ ), (bb′ ), (cc ′ ) be three distinct parallel lines. Suppose (ab)//(a ′ b ′ )
and (ac)//(a ′ c ′ ). Then (bc)//(b′ c ′ ).

a a′
b b′

c c′
Two triangles on three parallel lines. If two pairs of sides are parallel, so is
the third.
The property is also called the small Desargues axiom.
10.3.2. Proposition.
(i) A has (p,p)-Desargues iff Trans(A) is transitive on A.
(ii) If this holds, then Trans(A) is abelian.
10.3.3. Remarks.
• Since distinct translations never coincide anywhere, there can be at most one
translation t mapping given a to given b. So in (i), transitivity is equivalent to
sharp transitivity.
• The converse of (ii) need not hold: one can construct affine planes with Trans(A) =
{Id} (which is abelian, but not transitive).

Proof.
(i) First suppose that Trans(A) is transitive on A; we prove (p,p)-Desargues. Take

56
a (p,p)-Desargues configuration (a, b, c; a ′ , b ′ , c ′ ); suppose (ab) // (a ′ b ′ ) and
(ac)//(a ′ c ′ ). We want to show (bc)//(b ′ c ′ ). By transitivity, let t be the translation
mapping a to a ′ . Then by Proposition (ii), one has t(b) = (ab) a ′ ∩ (aa′ )b = b′
and t(c) = c ′ likewise. Now since t is a dilation, (b ′ c ′ )//(bc), as desired.
The converse is more interesting, and requires partial maps. Suppose (p,p)-
Desargues holds. For any pair of distinct points (a, a′ ), we define a partial map
ť a,a ′ ∶ A ∖ (aa′ ) → A by:

ť a,a ′ (b) = (ab) a ′ ∩ (aa ′ )b .

This is well-defined, but a partial map.

Step 1. If b ∉ (aa′ ) and b′ = ť a,a ′ (b), then ť a,a ′ and ť b,b ′ agree wherever both
are defined.

Verification. This is exactly the (p,p)-Desargues assumption. ◇

This defines a global map t a,a ′ ∶ A → A, and t a,a ′ (a) = a′ . We must prove that
t a,a ′ is a translation. Notice that if t a,a ′ (b) = b′ , then t a,a ′ = t b,b ′ .
Step 2. t a,a ′ is a dilation.

Verification. Bijectivity is obvious from the following picture.

a a′

x′

We claim that t a,a ′ is an automorphism. Suppose x, y, z are collinear (and

we may suppose they are distinct); we show that so are the images x ′ , y ′ , z ′ .

x x′

y y′

z z′

But we know that t a,a ′ = t x ,x ′ , so actually y′ = (xx ′ ) y ∩ (x y)x ′ ∈ (x y)x ′

and z ′ ∈ (xz)x ′ = (x y)x ′ by collinearity of x, y, z. This proves collinearity of
x ′ , y′ , z′ .
We claim that t is a dilation. But we already know that (x ′ y′ )//(x y), so
this is clear. ◇

57
 Finally t is a translation. Otherwise it has a fixed point b.

a a′

b′ = b

Up to changing base point we may suppose b ∉ (aa ′ ). Now (ab)//(a ′ b ′ ) = (a′ b)

so b ∈ (aa′ ), a contradiction.
(ii) Let t 1 , t 2 be translations. We want to show t 2 t 1 = t 1 t 2 ; we may assume that
neither is IdA , so neither has a fixed point. Let a ∈ A.

Case 1. Suppose a, t 1 (a), t 2 (a) are not collinear.

a t 2 (a)

t 1 (a)

Then by Proposition (ii):

t 2 (t 1 (a)) = (a t 1 (a)) t 2 (a) ∩ (a t 2 (a)) t 1 (a) = t 1 t 2 (a),

as desired.
Case 2. Suppose a, t 1 (a), t 2 (a) are collinear, say on ℓ. The above no longer
applies. By AP3 take b ∉ ℓ. By transitivity, there is a translation t 0 taking a
to b. If t 1 t 0 (a) = t 1 (b) ∈ ℓ, then ℓ = (t 1 (a)t 1 (b))//(ab) and b ∈ (ab) = ℓ, a
contradiction. So t 1 t 0 (a) ∉ ℓ. Therefore, by case 1 applied to pairs (t 1 t 0 , t 2 )
and (t 0 , t 2 ):

t 2 t 1 = (t 2 t 1 )(t 0 t 0−1 ) = t 2 (t 1 t 0 )t 0−1

= (t 1 t 0 )(t 2 t 0−1 ) = t 1 (t 0 t 2 )t 0−1
= t 1 t 2 t 0 t 0−1 = t 1 t 2 ,

and we are done again.

10.3.4. Remark. Abelianity can be proved assuming only the existence of translations
having different ‘directions’ (see Definition 11.2.2 below). The latter is proved to be a
consequence of (p, p)-Desargues, but is weaker.

10.4 Exercises
10.4.1. Exercise. Let f ≠ IdA be a dilation. Suppose that for any a, b ∈ A one has (aa′ )//
(bb ′ ). Prove that f is a translation.

58
11 Artin’s coordinatisation (2): the skew-field
Abstract. A sequel to § 10. We recover a vector geometry from an affine plane,
paying attention to assumptions. § 11.1 introduces homotheties and another form
of the Desargues property. § 11.2 constructs the ring of direction-preserving endo-
morphisms of Trans(A), provided the latter is abelian. § 11.3 shows it is a skew-field,
provided A has (p,p)-Desargues. § 11.4 finishes coordinatisation, provided A has
(c, p)-Desargues.

11.1 Homotheties and the (c,p)-Desargues property

We construct and interpret homotheties by arguments similar to translations.
11.1.1. Definition (homothety). A homothety of A is a dilation having at least one fixed
point (this includes IdA ).
For a ∈ A, let Hom a (A) be the group of dilations fixing a.
There is no equivalent of Proposition 10.2.3. Two homotheties with the same fixed
point which coincide at one other point, are equal; but this follows from Proposition 10.1.3.
11.1.2. Corollary. If A has (p, p)-Desargues then Dil(A) = Trans(A) ⋊ Hom a (A) for any
a ∈ A.

Proof. By Proposition (iii), Trans(A) ⊴ Dil(A); by definition, Trans(A) ∩

Hom a (A) = {IdA }. Now let d ∈ Dil(A) be any dilation. By (p, p)-Desargues there
is a translation t with t(a) = d(a). Hence t −1 d(a) = a so that t −1 d ∈ Hom a (A).
Therefore d ∈ Trans(A) ⋊ Hom a (A).

We give the relevant form of the Desargues property.

11.1.3. Definition ((c,p)-Desargues). The (concurrent, parallel)-form of the Desargues
property (for short: (c, p)-Desargues) is the following axiom.
Let (aa′ ), (bb′ ), (cc ′ ) be three distinct concurrent lines. Suppose (ab) //
(a ′ b ′ ) and (ac)//(a ′ c ′ ). Then (bc)//(b ′ c ′ ).

a′
a
b b′
o
c
c′
Two triangles on three concurrent lines. If two pairs of sides are parallel, so
is the third.
The property is also called the big Desargues axiom.
11.1.4. Proposition. A has (c,p)-Desargues iff for any line ℓ and point o ∈ ℓ, Homo (A)
acts transitively on ℓ ∖ {o}.
11.1.5. Remark. Here again, since an element of Homo (A) already fixes o and a non-
identity dilation has at most one fixed point, sharpness is for free.

59
 Proof. Exercise 11.5.1.

11.1.6. Corollary. If A has (c, p)-Desargues, then it has (p, p)-Desargues.

11.1.7. Remark. This corollary explains the classical (rather obscure) terminology ‘big
Desargues axiom’ versus ‘small Desargues axiom’: one is stronger than the other.

Proof. The statement is not obvious geometrically. We give an algebraic proof, using
the equivalences of Propositions 10.3.2 and 11.1.4. It suffices to show that Trans(A) acts
transitively. Let a, a ′ be distinct points of A.
Let o ∈ (aa′ ) ∖ {a, a ′ } (here we must assume that lines have more than three
points). By assumption, there is g 1 ∈ Homo (A) such that g 1 (a) = a′ . Also let b ∉ (ao),
and let b′ = (ab) a ′ ∩ (aa′ )b , then b ′′ = g 1 (b). By assumption, there is g 2 ∈ Hom a ′ (A)
such that g 2 (b ′′ ) = b′ .

o a a′

b b′
b ′′

We claim that t = g 2 g 1 is a translation taking a to a’. Of course t(a) = g 2 (a′ ) = a ′ ,

but also t(b) = g 2 (b ′′ ) = b ′ . So if t has a fixed point c, then one must have (ac) //
(a ′ c ′ ) = (a′ c), and therefore c ∈ (aa′ ). But c ∈ (bb ′ ) likewise, and these lines do not
meet. Hence t has no fixed point; it is a translation.

11.2 The ring of direction-preserving translations

Here is another idea. In coordinatised affine planes, one has Trans(A2 (F)) ≃ F2 , which
is a vector space over F. But a priori, Trans(A2 (F)) is merely a group. One may re-
trieve F as a special ring of endomorphisms of Trans(A2 (F)); the abstract formulation
requires two definitions.
11.2.1. Definition (ring of endomorphisms of an abelian group). Let (M; +) be an abelian
group. Then the set End(M) = { f ∶ (M; +) → (M; +)} of endomorphisms of the group
(M; +) forms a ring under:
• pointwise sum, viz. ( f + g)(x) = f (x) + g(x);
• composition, viz. (g ⋅ f )(x) = g( f (x)).
The respective identity elements are the zero (constant) map 0 M and the identity map
Id M .
In general, the ring is non-commutative.
11.2.2. Definition (direction of a translation).
• The direction of a translation t ≠ IdA is the equivalence class of (a t(a)) for any
a ∈ A. (This is well-defined by Proposition (i).) We denote it by d t .

60
 • An endomorphism λ∶ Trans(A) → Trans(A) is direction-preserving if: for any
t ∈ Trans(A) ∖ {IdA }, one has λ(t) ≠ IdA and d λ(t) = d t .
Let F = {λ ∈ End(Trans(A)) ∶ λ is direction-preserving} ∪ {0}.
In particular a direction-preserving endomorphism of Trans(A) is injective by con-
struction.
11.2.3. Proposition. Let A be an affine plane such that Trans(A) is abelian. Then F is a
subring of End(Trans(A)).

Proof. Clearly F contains 0 and 1; so it suffices to check stability under +, −, ⋅. Oppos-

ition will be an exercise.
Step 1. F is stable under +.

Verification. Let λ, µ ∈ F; we show λ + µ ∈ F. We may assume that neither λ, µ, nor

λ + µ is 0. Let t ∈ Trans(A) ∖ {IdA }. By definition, λ is injective, so λ(t) ≠ IdA and
µ(t) ≠ IdA likewise. Also λ(t) + µ(t) ≠ IdA as otherwise λ + µ = 0.
Since λ et µ are direction-preserving, d λ(t) = d t = d µ(t) . Now this is also the dir-
ection of λ(t) + µ(t), so d t = d λ(t)+µ(t) = d(λ+µ)(t) , so λ + µ is direction-preserving.
◇

Step 2. F is stable under ⋅.

Verification. Let λ, µ ∈ F; we may assume that neither λ nor µ is 0. Then both are
injective, and therefore λ ⋅ µ ≠ 0. Now for any t ∈ Trans(A) ∖ {IdA }, d(λ⋅µ(t) =
d λ(µ(t)) = d µ(t) = d t , so by definition, λ ⋅ µ ∈ F. ◇

Notice that since End(Trans(A)) is a ring, F inherits associativity and distributiv-

ity without a need to check them.

11.3 A skew-field assuming (p,p)-Desargues

Proposition 11.2.3 does not use any form of Desargues property, only abelianity of Trans(A).
11.3.1. Definition. For h ∈ Dil(A) any dilation, consider the transformation:

λh ∶ T → T
t ↦ hth−1 .

Notice that λ h is always a group automorphism of Trans(A), and direction-preserving

(by definition of a dilation): hence λ h ∈ F.
Recall from Proposition 10.3.2 that (p, p)-Desargues implies abelianity of Trans(A).
11.3.2. Proposition. If A has (p, p)-Desargues, then F is a skew-field. Moreover, for any
point a one has Hom a (A) ≃ F× .

Proof. The proof entirely relies on a geometric lemma, proved after its consequences.
11.3.3. Lemma. Suppose that A has (p,p)-Desargues. Let a ∈ A be fixed. Then for all

61
λ ∈ F ∖ {0} there is a unique h ∈ Hom a (A) with λ h = λ.
We first show how it implies the Proposition.
• Let λ ∈ F ∖ {0}. By the Lemma, λ = λ h for some h. Let µ = λ h −1 . Then µ ∈ F
and λµ = µλ = 1 in F, so F is a skew-field.

• There is a group homomorphism Hom a (A) ↠ F× ; this is actually an isomorph-

ism. For if h ∈ Hom a (A) lies in the kernel, then it means that λ h = 1, or equi-
valently, that for any t ∈ Trans(A), one has ht = th. However as soon as t ≠ IdA ,
one gets ht(a) = th(a) = t(a) so h fixes not only a but also t(a) ≠ a; being a
dilation it implies h = IdA , as desired.

Proof of the Lemma.

t λ(t)
a

t a,xx

λ(t a,x )

h(x)

By (p,p)-Desargues, for any x ∈ A there is a translation t a,x taking a to x. Define:

h(x) = (λ(t a,x )) (a)

We claim that h ∈ Hom a (A) and λ h = λ.

Step 1. h ∈ Hom a (A).

Verification. First notice that h is well-defined since t a,x exists and is unique. Also
h(a) = λ(IdA )(a) = IdA (a) = a since λ is a group endomorphism. But we have a
number of things to check that it is a dilation.
Let y ∈ A be another point. Then t a, y = t x , y t a,x since these two translations
coincide at a. Therefore, using that λ is a group endomorphism:

h(y) = (λ(t x , y )λ(t a,x )) (a) = λ(t x , y )(h(x))

In particular, if x ≠ y, then t x , y ≠ IdA and since λ ∈ F ∖ {0} is direction-preserving,

λ(t x , y ) ≠ IdA , proving h(y) ≠ h(x). So h is injective. But also, always since λ is
direction-preserving,
(h(x) h(y))//d t x , y //(x y)
An injective map with this property is always a dilation (preservation of collinearity
and surjectivity being consequences left as an exercise). So h ∈ Dil(A) and since
we already know h(a) = a, we have h ∈ Hom a (A). ◇

Step 2. λ h = λ.

62
 Verification. Take any t ∈ Trans(A); say t = t a,b for some b. Then λ h (t) = hth−1 is
the translation taking a to hth −1 (a) = ht(a) = h(b) = λ(t a,b )(a) = λ(t)(a). So
λ h (t) = λ(t), for any translation; hence λ h = λ in F. ◇

So h ∈ Hom a (A) and λ h = λ.

This proves the proposition.

11.4 Coordinatisation assuming (c, p)-Desargues

One can thus retrieve a skew-field under as weak a property as (p,p)-Desargues. Now
the F-module Trans(A) is actually a vector space. One may think this is close to co-
ordinatising, yet the dimension theory of Trans(A) is not clear at all.
Recall that the direction (Definition 11.2.2) of a translation t is denoted by d t .
11.4.1. Proposition. Suppose A has (c, p)-Desargues. Then:

(i) for t, t ′ ∈ Trans(A) ∖ {Id}, one has d t = d t ′ iff there is λ ∈ F with t ′ = λt;
(ii) Trans(A) is 2-dimensional as a left-vector space over F;
(iii) A ≃ A2 (F).

Proof. Recall from Corollary 11.1.6 that (c, p)-Desargues implies (p, p)-Desargues,
which in turn implies (sharp) transitivity of Trans(A) on A by Proposition 10.3.2.
Whevener a ≠ b ∈ A, we let t a,b be the unique translation taking a to b. Its direc-
tion is then the parallelism class of (ab).
(i) Let t, t ′ ∈ Trans(A) ∖ {IdA }. If there is λ ∈ F with t ′ = λt then λ ≠ 0, so by
definition, λ is direction-preserving; hence d t ′ = d t .
Now suppose d t ′ = d t , and say t = t o,a , t ′ = t o,a ′ . By (c, p)-Desargues, there is
h ∈ Homo (A) doing h(a) = a ′ . Then let λ = λ h . We know that:

λt = hth−1
= t o,hth −1 (o)
= t o,ht(o)
= t o,h(a)
= t o,a ′
= t′ .

(ii) It suffices to show that two translations with different directions span Trans(A).
These exist under (p, p)-Desargues. Let t 1 , t 2 have directions d 1 ≠ d 2 . Let t be
another translation. By (i) we may suppose that the direction of t is neither d 1
nor d 2 . Fix a ∈ A and let b = t(a).

63
 d2 d1

a b

Lines (d 1 )b and (d 2 ) a are not parallel, so they meet at say c. Clearly c is neither
a nor b. So t c,b has direction d 1 , which is the same as t 1 . By (i), there is λ ∈ F with
t c,b = λt 1 . Likewise, the direction of t a,c is d 2 , so there is µ ∈ F with t a,c = µt 2 .
Now:
t = t a,b = t c,b ○ t a,c = (λ(t 1 )) ○ (µ(t 2 )),
as transformations of A, which in vector space notation rewrites: t = λ ⋅ t 1 + µ ⋅ t 2 .
(iii) Fix any o ∈ A. Then by sharp transitivity, A = Trans(A) ⋅ o is parametrised by
Trans(A) ≃ F2 as a left-vector space over F. So we may represent points in A by
pairs in F2 .
Let p 1 ≠ p 2 be two points in A. Then for any q ∈ A one has equivalences:

q ∈ (p 1 p 2 ) iff q = p 1 or (p 1 q)//(p 1 p 2 )
iff q = p 1 or d t p 1 q = d t p 1 p 2
iff (∃λ ∈ F)(t p 1 ,q = λt p 1 , p 2 ).

So (p 1 p 2 ) = {p 1 + λ ⋅ t p 1 , p 2 ∶ λ ∈ F}. Thus lines in A correspond to parametric

affine lines in F2 . This gives an isomorphism A ≃ A2 (F).

11.4.2. Remarks.
• I do not know which finite values of dimF Trans(A) are possible in abstract affine
planes.
• Objects Trans(A) and F are intrinsic, viz. do not depend on anything. On the
other hand coordinatisation does: it depends on the choice of o. Notice that we
do not introduce other extrinsic data; Artin’ method requires no axes.
• Hilbert’s and Artin’s methods have in common that one has to go to the affine part
of a desarguesian projective plane. I am not aware of proofs entirely taking place
in the original projective plane.
• I am not aware either of quicker arguments directly coordinatising ‘projective 3-
spaces’ (Definition 4.1.1).19

11.5 Exercises
11.5.1. Exercise. Prove Proposition 11.1.4, viz.: A has (c,p)-Desargues iff for any line ℓ and
point o ∈ ℓ, Homo (A) acts transitively on ℓ ∖ {o}.
For the direct implication, first construct a partial map ȟ o,a→a ′ defined on A ∖ ℓ.
19 To some extent, the following sketches an answer, but not a very satisfactory one as it resembles an in-
termediate stage between Hilbert and Artin. M. K. Bennett, Coordinatization of affine and projective space,
Discrete Mathematics 4, pp. 219–231, 1973.

64
11.5.2. Exercise. There is a (c, c)-Desargues property; draw its picture. Now give a geo-
metric proof that (c, p)-Desargues implies (c, c)-Desargues.20
(It is a consequence of coordinatisation that (c, p) implies all other affine forms.)

12 Duality
Abstract. § 12.1 dualises geometric statements; if a geometric statement φ holds in
all projective planes, then so does its dual statement φ∗ . § 12.2 dualises projective
planes, viz. attaches to each projective plane a dual plane, with dual properties.
§ 12.3 studies the behaviour of the Desargues and Pappus properties under duality.

We use the following notation from mathematical logical. For Γ an incidence geo-
metry and φ a geometric statement (one about points, lines, incidence), write Γ ⊧ φ if
Γ satisfies φ.

12.1 Duality
Return to the definition of a projective plane (Definition 2.1.1). There is a noticeable
symmetry in the axioms when one exchanges points and lines.
12.1.1. Definition (dual statement). Let φ be a statement about projective planes. Its
dual statement is the statement φ∗ obtained by exchanging the words ‘point’ and ‘line’,
and exchanging the symbols ∈ and ∋.
12.1.2. Example. The dual statement of PP1 is PP1 ∗ = PP2 . Likewise, PP2 ∗ = PP1 . This
is not surprising since φ∗∗ = φ.
12.1.3. Theorem (duality theorem). Let φ be a statement about projective planes. Then φ
holds in every projective plane iff φ∗ does.
Be careful that ‘being desarguesian’ does not hold in every projective plane, so it is
not a suitable φ. But ‘Pappus implies Desargues’ is an example of such a statement.

Proof using mathematical logic. Since φ∗∗ is always the formula φ, one implication
is enough. We want to argue by symmetry of the axioms for projective planes. We
know that PP1 ∗ = PP2 , but we should also do something about PP3 .
Step 1. Let Γ be an incidence geometry satisfying PP1 and PP2 . Then in Γ, PP3 is
equivalent to:
PP3 ∗ . There are four distinct lines, no three of which are concurrent.

Verification. Suppose PP3 . There are four points a, b, c, d no three of which are col-
linear. So lines (ab), (ac), (bc), (bd) are four in number and no three of them con-
cur. Hence PP3 ∗ holds. The converse can be obtained similarly. ◇

Suppose that all projective planes satisfy φ. Then by ‘Skolem-Gödel’s completeness

theorem’ there is a proof of φ using only axioms PP1 , PP2 , PP3 . Now exchange the
20 A solution may be found in: M. Prażmowska, A proof of the projective Desargues axiom in the desar-
guesian affine plan, Demostratio Mathematica, xxxvii(4), pp. 921–924, 2004.

65
 words ‘points’ and ‘lines’ in the proof (and revert incidence accordingly). The resulting
text is a proof of φ∗ using only axioms PP1 ∗ = PP2 , PP2 ∗ = PP1 , and PP3 ∗ . Since the
latter is a consequence of PP1 , PP2 , and PP3 , it holds in any projective plane. Therefore
so does φ∗ .

12.1.4. Remark. The proof is not very satisfactory as it relies on proof theory. § 12.2 gives
a better one.

12.2 The dual plane

Let us now dualise planes.
12.2.1. Definition (dual incidence geometry). Let Γ = (P , L, ∈) be an incidence geo-
metry. Set P ∗ = L, L∗ = P, and take I ∗ to be ∋. Let Γ∗ = (P ∗ , L∗ , I ∗ ) be the dual
incidence geometry of Γ.
12.2.2. Remark (dual incidence as membership). In order to have I ∗ be ∈, proceed as
follows. For p ∈ P let L p = {ℓ ∈ L ∶ p ∈ ℓ}. Now take P ∗ = L, L∗ = {L p ∶ p ∈ P}, and
I ∗ =∈. We claim that this construction is isomorphic to the previous one.
Introduce better notation: let Γ1∗ be the original construction with I ∗ and Γ2∗ be the
one with ∈. Let f ∶ P1∗ = L → L = P2∗ be the identity map. Now let g∶ L∗1 = P → {L p ∶
p ∈ P} = L∗2 take p to L p . Then letting p∗1 = ℓ ∈ P1∗ and m∗1 = q ∈ L∗1 , one has:

p∗1 I 1∗ m∗1 ⇔ ℓ∋q ⇔ ℓ ∈ Lq ⇔ f (p∗1 ) ∈ g(m∗1 ).

12.2.3. Proposition. Let Γ be an incidence geometry and φ be a sentence. Then Γ ⊧ φ iff

Γ∗ ⊧ φ∗ . In particular, Γ is a projective plane iff Γ∗ is one.

Proof. Short, not rigorous: by induction on what a statement is.

Long, rigorous: define elementary formulas with parameters φ(a). Now a point
p ∈ P maps to the line p∗ ∈ L∗ and a line ℓ ∈ L to the point ℓ∗ ∈ P ∗ ; so parameters
a∗ are well-defined. Then prove by induction on formula with parameters: Γ ⊧ φ(a)
iff Γ∗ ⊧ φ∗ (a∗ ). (This is an easy induction.) In particular, a sentence is a formula with
no parameters, giving the result.

Since axioms of a projective plane are self-dual, but axioms of an affine plane are
not, we focus on projective planes. (The dual geometry of an affine plane is not an affine
plane.)
12.2.4. Remark. In general, P ≃/ P∗ . However P ≃ P∗∗ always and canonically.

Better proof of the duality Theorem 12.1.3. Suppose all projective planes satisfy φ.
In particular, for all projective planes P, one has P∗ ⊧ φ; hence P ⊧ φ∗ . So all projective
planes satisfy φ∗ . Applying to φ∗ , we get the converse.

Let F be a skew-field. Remember that the opposite skew-field Fop has same underly-
ing set, same addition, but multiplication:

a ⋅op b = b ⋅ a.

It is a skew-field, in general not isomorphic to F.

66
12.2.5. Remarks.
• There is a general notion of the opposite group G op , and one always has G op ≃ G
(using inversion). But knowing F× ≃ (F× )op is not enough as addition is not
necessarily preserved.
• F = Fop iff F is commutative.
• H ≃ Hop thanks to quaternion conjugation. In general, F ≃ Fop iff F admits an
‘anti-automorphism’.
• So in general, F ≃/ Fop .
12.2.6. Proposition. Let F be a skew-field. Then (P2 (F))∗ ≃ P2 (Fop ).
12.2.7. Remarks.
• In particular, if F is a commutative field, then (P2 (F))∗ ≃ P2 (F).
• The converse may fail. For instance, although H is not commutative, i.e. H ≠ Hop ,
one has H ≃ Hop . So (P2 (H))∗ ≃ P2 (Hop ) ≃ P2 (H).

Proof of the Proposition. One must pay attention to sides. There are three important
ideas.

1. The dual of a left-vector space is a right-vector space.

This is achieved by letting (φ ⋅ λ)(v) = φ(v) ⋅ λ. (Also notice that (λ ⋅ φ)(v) =
λ ⋅ φ(v) is not a linear form.)
More generally the usual duality pairing W ↝ W ⊥ maps left-F-vector subspaces
of V to right-F-vector subspaces of V ∗ . (If things are too abstract, one is al-
lowed to think in terms of equations, but then remember that in our conven-
tion matrices act from the right. Say elements of V are rows, elements of V ⊥ are
columns, and one computes ⟨v, φ⟩ by row-column multiplication, which returns
a scalar.)
With this in mind, duality theory works like in the commutative case.

2. A right-vector space over F is a left-vector space over Fop .

This is achieved by letting λ ∗ v = v ⋅ λ. Indeed:

(λ ⋅op µ) ∗ v = (µ ⋅ λ) ∗ v
= v ⋅ (µ ⋅ λ)
= (v ⋅ µ) ⋅ λ
= λ ∗ (µ ∗ v).

3. P2 (F) is obtained from F3 as a left-vector space. But F3 also bears a right-vector

space structure by letting (x, y, z) ⋅ λ = (x λ, yλ, zλ). These two actions differ,
but commute. Hence F3 is both a left-vector space and a right-vector space over
F, alternatively: both a F and Fop left-vector space. But these structures differ
as soon as F is non-commutative.
Let K = Fop for clarity. It will save notation to write PF = P2 (F) and PK = P2 (K).

67
 If L ≤ K3 is a (left-K-)vector line, then L⊥ ≤ K3 is a (right-K-)vector plane, so
L ≤ F3 is a (left-F-)vector plane. Likewise, if H ≤ K3 is a (left-K-)vector plane, then
⊥

H ⊥ ≤ F3 is a (left-F-)vector line. Now define:

f ∶ P(PK ) → L(PF ),
L ↦ L⊥

and:
g ∶ L(PK ) → P(PF ).
H ↦ H⊥
These are clearly bijections. Now let (p, ℓ) ∈ P(PK ) × L(PK ), say p = L ≤ K3 and
ℓ = H ≤ K3 . Since duality reverses inclusion:

p∈ℓ ⇔ L≤H ⇔ H ⊥ ≥ L⊥ ⇔ g(ℓ) ∋ f (p).

Therefore PFop = PK ≃ (L(PF ), P(PF ), ∋) ≃ P∗F , as desired.

12.3 Dualising Desargues and Pappus

The Desargues and Pappus properties may be called ‘self-dual’, but that would be in two
distinct senses:
• the Desargues property is equivalent to its dual, meaning that a plane has one iff
it has the other;
• the (much stronger) Pappus property implies the (much stronger too) self-duality
of a plane, meaning that the pappian plane is isomorphic to its dual.

12.3.1. Proposition (Desargues implies Desargues∗ ). Let P be a projective plane. Then

P is desarguesian iff P∗ is.

First proof. One implication suffices since P∗∗ ≃ P. Suppose P is desarguesian. Then
using Hilbert coordinatisation, there is a skew-field F with P ≃ P2 (F). Then by Pro-
position 12.2.6, P∗ ≃ (P2 (F))∗ ≃ P2 (Fop ) is desarguesian as well.

Second proof. To prove that P∗ satisfies Desargues, we check that P satisfies

Desargues∗ . One should first write Desargues∗ ; it so happens that it is the ‘converse
Desargues property’ (Remark 3.1.5).
So let ω, α, α ′ , β, β ′ , γ, γ′ be lines such that:
• ω, α, α ′ are concurrent;

• ω, β, β′ are concurrent;
• ω, γ, γ′ are concurrent.
To prove Desargues∗ is to prove that lines α ′′ = (β ∩ γ β′ ∩ γ′ ), β ′′ = (α ∩ γ α ′ ∩ γ′ ),
and γ′′ = (α ∩ β α ′ ∩ β ′ ) are concurrent.

68
 a
b
c

a′
b′
c′ ω
a0 b0 c0
α′ γ′
α β β ′ γ

Will dotted lines concur?

We name these points and a few more. Let:
• a 0 = α ∩ α ′ , b 0 = β ∩ β ′ , and c 0 = γ ∩ γ′ ;
• a = β ∩ γ and a ′ = β ′ ∩ γ′ ;

• b = α ∩ γ and b ′ = α ′ ∩ γ′ ;
• c = α ∩ β and c ′ = α ′ ∩ β′ .
With this notation we now have α ′′ = (aa′ ), and so on.
Consider the following Desargues configuration (mind the swap on line ω):

b0
c0
b c
a0
b′
c′

The various collinearities occur on lines ω, α, α ′ . Now compute the ‘Desargues points’:
• (c 0 b) = γ and (b 0 c) = β meet at a;

• (c 0 b ′ ) = γ′ and (b 0 c ′ ) = β ′ meet at a ′ ;
• (b′ b) = β ′′ and (c ′ c) = γ′′ meet at say x.
Since P is desarguesian, x ∈ (aa′ ) = α ′′ . So α ′′ , β′′ , γ′′ concur at x, as desired.

Assuming the Pappus property, the conclusion can be made even stronger.
12.3.2. Theorem (Pappus implies self-duality). Let P be a projective plane. If P is pappian,
then P∗ ≃ P. (In particular P∗ is pappian.)

Proof. Suppose P is pappian; we use coordinatisation. By the Pappus property, P ≃

P2 (F) for some commutative field F (Corollary 6.3.2). In particular Fop = F. Therefore

69
 taking duals with Proposition 12.2.6:

P∗ ≃ P2 (F)∗ ≃ P2 (Fop ) = P2 (F) ≃ P,

which is pappian by assumption.

12.3.3. Remarks.
• There is a geometric proof (without coordinatisation) that if P is pappian then so
is P∗ . This is much weaker than Theorem 12.3.2.

• I am not aware of a geometric proof that if P is pappian, then P ≃ P∗ (one would

have to go to self-dualities).
• Be careful that (P ≃ P∗ and desarguesian) does not imply commutativity of the
underlying skew-field: one only has Fop ≃ F, as in the quaternions.

12.4 Exercises
12.4.1. Exercise. Prove that in an incidence structure satisfying PP1 and PP2 , the following
are equivalent:

• every line has at least three points;

• every point is on at least three lines.
12.4.2. Exercise. Let F be a skew-field.

1. Prove that (F)2 P ≃ P2 (Fop ) (see Remark 2.1.4).

2. Prove that F ≃ Fop iff P2 (F) ≃ (P2 (F))∗ iff P2 (F) ≃ (F)2 P.
12.4.3. Exercise. Give a geometric proof (in the spirit of Proposition 12.3.1) that Desargues∗
implies Desargues.

12.4.4. Exercise. Give a geometric proof (in the spirit of Proposition 12.3.1) that a projective
plane P is pappian iff P∗ is.

Further reading
• E. Artin, Geometric algebra. Interscience Publishers, New York/London. 1957.
The coordinatisation theorem is proved in chapter 2. The whole book is a classic.
• P. Cameron, Projective and polar spaces. Queen Mary and Westfield Maths Notes,
13, London. 1992.
Available on the author’s webpage.21
• R. Hartshorne, Foundations of Projective Geometry. Harvard University Lecture
Notes, W. A. Benjamin, Inc., New York. 1967.
Technically out of print but on the internet one finds a modern typeset.
21 https://webspace.maths.qmul.ac.uk/p.j.cameron/pps/

70
 • D. Hilbert, Grundlagen der Geometrie. B. G. Teubner, Leipzig. 1899.
One may find a translation of this historical text online.22 Chapter 5 (on Desar-
gues’ Theorem) is the origin of the topic.

• D. Hughes and F. Piper, Projective planes. Graduate texts in mathematics, 6. Springer-

Verlag, New-York/Berlin. 1973.
Recommended reference on projective planes.
• H. Salzmann, D. Betten, T. Grundhöfer, H. Hähl, R. Löwen, and M. Stroppel,
Compact projective planes, with an introduction to octonion geometry. de Gruyter
Expositions in Mathematics, 21. 1995.
Adding back topological ingredients, a complete study of the projective plane over
the octonions.

22 https://math.berkeley.edu/~wodzicki/160/Hilbert.pdf

71