GEOMETRIES AND TRANSFORMATION GROUPS

JIE DU

Contents

Introduction 2
1. Transformations and Groups 3
2. Subgroups and the Group of Isometries 5
3. Reflections and Isometries 7
4. Translations and rotations on R2 9
5. Congruence and the symmetry group of a geometric figure 11
6. Classification of Plane Isometries 13
7. Similarities 15
8. Dilatations 17
9. Classification of Plane Similarities 19
10. More on Groups 21
11. Collineations 23
12. Darboux’s Theorem 25
13. Affine Transformations 27
14. The Real Projective Line RP 1 , Plane RP 2 and Space RP n 29
15. The Principle of Duality in RP 2 31
16. Desargues’ Theorem and Pappus Theorem 33
17. Projective Transformations in RP n 35
18. Projective Plane Transformations 37
References 40

Date: September 14, 2023.

2000 Mathematics Subject Classification. Primary 17B37, 20G05, 20C08;
Secondary 16G20, 20C33.
Key words and phrases. affine Hecke algebra, affine quantum Schur algebra, cyclic quiver, Drinfeld
double, loop algebra, quantum group, Schur–Weyl duality, Ringel–Hall algebra, simple representation.
1
2 JIE DU

Introduction

Mathematics went through quite a revolution around the turn of the 20th century. In
particular, an axiomatic approach infiltrated the mathematical paradigm, both as a tool
to ensure mathematical rigour and to abstract common principles working in a variety of
different settings.
First year undergraduate mathematics emphasises computation over abstraction and
rigour. Later year courses (and Pure Mathematics in general) reverse this, so students
need to learn some new skills and some new ways of thinking about mathematical objects.
This course is designed to help you develop the ability to write rigorous mathematical
proofs in a setting where the level of abstraction is still quite modest. As such it will
serve as an excellent preparation for the third year Pure Mathematics courses.
We will investigate various transformations on the Euclidean and projective spaces. We
will first study several types of transformations such as translations, reflections, rotations
in the R2 , etc. in terms of groups. We will then look at symmetries, i.e. transformations
of geometric figures that preserve some property (such as distance or angles between
lines), and projective geometry. Projective transformations can change a conic section of
one type to another, e.g. an ellipse to a hyperbola.
What is geometry?
Felix Klein (1849–1925)’s geometry is a set S, called a space, on which a group G
operates. The study of properties invariant under the operations is called a geometry.

Space Transformation groups Geometry Invariants Linear groups

Rn isometries Euclidean geometry distance On (R)

Rn similarities Similarity geometry angle R>0 On (R)

Rn collineations Affine geometry parallelism GLn (R)

RP n projective transf. Projective geometry collinearity P GLn+1 (R)

R2n smplectic transf. Symplectic geometry symp. type Sp2n (R)

 GEOMETRIES AND TRANSFORMATION GROUPS 3

1. Transformations and Groups

Definition 1.1. A transformation on Rn is a bijection from Rn to Rn . We will denote

by B(Rn ) the set of all transformations on Rn .
In particular, a transformation on the Euclidean plane R2 is called a “plane transfor-
mations”.

Examples 1.2. (1) The identity map 1 : Rn → Rn is a transformation.

(2) Fix b ∈ Rn . The map Tb : Rn → Rn , where Tb (x) = x + b, is called a translation.
A translation is a transformation.
Another more geometric way to view translation is as follows: the map takes a point P
−→
to the point P 0 such that the vector from P to P 0 is equal to the vector b, i.e., P P 0 = b.
We denote a translation that takes P to P 0 by τP,P 0
(3) The map TA,b formed by a matrix multiplication (by invertible matrix A) followed
by a translation (by vector b) is a transformation. This is, again, easy to see. [It is
onto because: for v, then we see TA,b (A−1 (v − b)) = v. One-to-one is clear.] Note that
Tb = T0,b .
(4) A transformation π on Rn with n ≥ 2 that sends every line l to a line π(l) is called
a collineation. [The cubic function sends R to R.]

Definition 1.3 (Group). A group is a set G equipped with a map

∗ : G × G −→ G, (g, h) 7→ g ∗ h = gh,

called the multiplication map, satisfying the following axioms

(G1) Associativity, i.e. g, h, k ∈ G, then (gh)k = g(hk)

(G2) Existence of identity, i.e. there is an element denoted by e in G called the identity
of G such that eg = g = ge for any g ∈ G. (Will prove uniqueness, then notation
1G .)
(G3) Existence of inverse, i.e. for any g ∈ G, there is an element denoted by h ∈ G called
an inverse of g such that gh = hg = e. (Will prove uniqueness, then notation
g −1 .)

A group G is called commutative or abelian if gh = hg for all g, h ∈ G

Proposition 1.4. (1) The set B(Rn ) of all transformations on Rn together with the
operation of composition forms a group.
(2) The set T (Rn ) of all translations on Rn together with the operation of composition
forms a group.
4 JIE DU

(3) The set C (Rn ) of collineations on Rn together with the operation of composition
forms a group.

Proof. (1) Straightforward to check (G1)–(G3).

(2) For a, b ∈ Rn , Ta Tb = Ta+b , Tb−1 = T−b .
(3) [Ex,] We want to prove that if τ is a collineation, then so is τ −1 . Let l be a line and
suppose l = `(A, B). Consider the line m = `(τ −1 (A), τ −1 (B)). Since τ (m) pass through
A, B and τ (m) is a line, it follows that τ (m) = l. Hence, τ −1 (l) = m is a line. 

Note that T (Rn ) is commutative.

Definition 1.5 (Subgroup). Let (G, ∗) be a group. A nonempty subset H ⊆ G is said

to be a subgroup of G, denoted by H ≤ G, if (H, ∗) is a group.

Like the Subspace Theorem in Linear Algebra, we have

Proposition 1.6. A nonempty subset H of a group G is a subgroup if and only if the

following two closure conditions are satisfied:
(SG1) Closure under multiplication, i.e. if h, k ∈ H, then hk ∈ H;
(SG2) Closure under inverse, i.e. if h ∈ H, then h−1 ∈ H.
In particular, 1H = 1G .

Definition 1.7. For groups G, H, a map f : G → H is called a group homomorphism

if f (xy) = f (x)f (y) for all x, y ∈ G. A bijective group homomorphism is called an
isomorphism. In this case, we say that G is isomorphic to H. Notation: G ∼
= H.
 GEOMETRIES AND TRANSFORMATION GROUPS 5

2. Subgroups and the Group of Isometries

Examples 2.1 (of subgroups). (1) Let GLn (R) := {A ∈ Mn (R) | A is invertible}. This
is a group with matrix multiplecation. We now define a set

GL(Rn ) := {TA : Rn → Rn , x 7→ Ax | A ∈ GLn (R)}.

Then GL(Rn ) ≤ B(Rn ). The group GLn (R) and GL(Rn ) are isomorphic in the sense that
there is a bijective map φ : GLn (R) → GL(Rn ), A 7→ TA such that φ(AB) = φ(A)φ(B).
(2) On (R) = {Q ∈ GLn (R) | QT Q = I} is a subgroup.
(3) Let S be a subset of a group G. Define the following set:
hSi := {se11 ∗ se22 ... ∗ semm : si ∈ S, ei = ±1, ∀i, m ≥ 0}
= {t1 ∗ ... ∗ tn : ti ∈ S ∪ S −1 , i = 1, ..., n; n ≥ 0}
= set of all products of elements in S, and their inverses.
Then hSi ≤ G, S is a subgroup of G
Another way to define the group hSi is the following: hSi is the smallest subgroup of
G that contains all the elements of S, i.e., hSi = ∩H≤G H.

Definition 2.2. We call hSi the subgroup of G generated by S.

Exercise: (1) GLn (R) is generated by elementary matrices.

(2) Describe the subgroup generated by one element: hgi.

Notation:

• space: Rn ;
• points: A, B, C, P, Q, R, ... with position vectors a, b, c, p, q, r....;
• transformations: τ, π, σ, δ...;
• lines: l, m, n, ..; line equations in Rn : x = a + λv for all λ ∈ R;
• planes in Rn : x = a + λu + µv for all λ, µ ∈ R;
• Hyperplanes through a ∈ Rn with normal n ∈ Rn − 0:

Hn,a = {x ∈ Rn | (x − a) · n = 0} = hni⊥ + a.

For points P, Q in Rn , we may also define the perpendicular bisector of the

line segment P Q to be the hyperplane H that passes through the midpoint and
perpendicular to P Q. So H has the equation (x − m)  (p − q) = 0, where m =
1
p
2
(p + q). It is clear that, for all X ∈ H, d(X, P ) = kx − mk2 + kp − mk2 =
p
kx − mk2 + kq − mk2 = d(X, Q).
6 JIE DU

The Euclidean space Rn is equipped with a dot product

 : Rn × Rn −→ R, (a, b) 7→ a  b = a1 b1 + · · · + an bn .
√
• Length of a vector: kak = a  a;
• Distance between two points P, Q: kp − qk;
a·b
• Projection of a on b: projb (a) = b·b
b;
a·b
• Angle between a and b: cos(θ) = kakkbk ;
• Orthogonality: a ⊥ b ⇐⇒ a  b = 0;

Definition 2.3. An isometry on Rn is a map τ : Rn → Rn which preserves distance

between points: d(P, Q) = d(τ (P ), τ (Q)), ∀P, Q ∈ Rn .

Lemma 2.4. The set of isometries which fix the zero vector is equal to the set of (linear)
maps that represent multiplication by an orthogonal matrix. (This is proved in the first
assignment.)

Theorem 2.5. (1) Isometries can be decomposed into a translation multiplied by a linear
map, which can be represented by an orthogonal matrix. In other words, for every τ ∈
I (Rn ), τ = Tb ◦ TQ , for Q orthogonal. In particular, an isometry is a transformation.
(2) The set I (Rn ) of all isometries forms a subgroup of the group B(Rn ) of all trans-
formations. In particular, we have I = T O.

Proof. Let τ (0) = b. Let σ = T−b ◦ τ , σ ∈ I (Rn ). Then σ(0) = 0. By the lemma
above, σ is an orthogonal linear isomorphism TQ for some orthogonal matrix Q. Thus,
multiplying our earlier expression by Tb gets us: τ = Tb ◦ σ. Hence, we see that, for any
τ ∈ I (Rn ), there exists an orthogonal matrix Q and a vector b such that τ (x) = Qx + b.
For (1) we check inverses:
Let τ be an isometry. Since τ is a transformation, there exists an inverse function
τ −1 ∈ B(Rn ). We seek to show that this inverse is also an isometry. Let A, B be two
points in Rn . Then, since τ is a transformation, there exists points A0 , B 0 such that
τ (A0 ) = A, τ (B 0 ) = B. So τ −1 (B) = B 0 , τ −1 (A) = A0 . Now, d(A0 , B 0 ) = d(A, B), since τ
an isometry. Thus d(τ −1 (A), τ −1 (B)) = d(A, B) Hence, τ −1 is an isometry. 

An isometry is not a linear map.

 GEOMETRIES AND TRANSFORMATION GROUPS 7

3. Reflections and Isometries

Definition 3.1. Let H be a hyperplane. The reflection σH in H is the mapping defined

by: 
P if P ∈ H;
σH (P ) =
P 0 if P is off H and H is the perpendicular bisector of P P 0 .
(in the sense that d(P, X) = d(P 0 , X) for all X ∈ H.)

Proposition 3.2. Let H be a hyperplane.

(1) A reflection σH is an isometry satisfying σH2 = 1.
(2) σH fixes a line m 6⊆ H if and only if m ⊥ H.
(3) σH fixes a line pointwise if and only if m ⊆ H.

Proof. Assume H = Hn,a . (1) By definition, σH2 = 1. Hence, it is a bijection. (For the
proof of isometry, see on-line notes.) (2) If m ⊥ H, it is clear σH (m) = m. Conversely,
let Q = σH (P ) for some P ∈ m, but off H. Then Q ∈ m and the line segment P Q is
perpendicular to H. Hence, m ⊥ H. (3) is clear. 

2 an
Theorem 3.3. If H = Hn,a , then there exist Q = I − nn
nnT ∈ On (R) and b = 2 nn n
such that σH (x) = Qx + b, .

Proof. By vector geometry,

xn an
σH (x) = a + [(x − a) − 2projn (x − a)] = x − 2projn (x − a) = x − 2 n+2 n.
nn nn
Now the assertion follows easily, noting that (x  n)n = (nT x)n = n(nT x) = (nnT )x. 

Corollary 3.4. If the hyperplane H has equation N1 X1 + · · · + Nn Xn + c = 0, then the

reflection σH in H has equation: for n = (N1 , · · · , Nn )T ,
2
σH (x) = x − (N1 x1 + · · · + Nn xn + c)n.
n·n
Proof. Note that (x − a) · n = N1 X1 + · · · + Nn Xn + c in the proof above. 

Definition 3.5. We say that m points P1 , P2 , . . . , Pm in Rn are in generic position if

the vectors pi − p1 , for i = 2, 3, . . . , m, are linearly independent. In particular, n + 1
points in Rn are in generic position if every hyperplane contains at most n of the n + 1
points.

Theorem 3.6. (1) An isometry on Rn that fixes n + 1 points in generic position is

the identity map.
(2) An isometry on Rn that fixes n points in generic position is a reflection or the
identity.
8 JIE DU

(3) An isometry that fixes n − 1 but not n points in generic position is a product of
two reflections.
(4) Every isometry (in Rn ) is a product of at most n + 1 reflections.

Proof. (1) If τ is an isometry, then by Theorem 2.5, there exists an orthogonal matrix
Q ∈ On (R), and a vector b such that τ (x) = Qx + b.
Now suppose points P1 , . . . , Pn+1 are in generic position with position vectors
p1 , . . . , pn+1 with τ (Pi ) = Pi , for every i. Then we have: Qpi + b = pi . This
Q(p1 − pi ) = p1 − pi for all i = 2, . . . , n + 1. Thus, Q fixes a basis for Rn and so
Q = In . Consequently, b = 0. Hence, τ = 1 proving (1).
(2) Suppose τ fixes points P1 , . . . , Pn which are in generic position. Then the hyperplane

H : x = p1 + λ1 (p2 − p1 ) + · · · + λn−1 (pn − p1 ), λi ∈ R,

contains all n points. If τ 6= 1, then there exists a point R off H such that τ (R) = R0 6= R.
So we have d(Pi , R) = d(τ (Pi ), τ (R)) = d(Pi , R0 ) for all i = 1, 2, . . . , n. Thus, for any
point A in H, since a = p1 + λ1 (p2 − p1 ) + · · · + λn−1 (pn − p1 ) for some λi , we see that

τ (a) = Q(p1 + λ1 (p2 − p1 ) + · · · + λn−1 (pn − p1 )) + b = a.

So d(A, R) = d(τ (A), τ (R)) = d(A, R0 ). This shows H is the orthogonal bisector of RR0 .
Hence, τ = σH .
(3) Suppose τ is an isometry, and fixes the points P1 , . . . , Pn−1 in generic position.
Choose a point P = P0 so that P, P1 , . . . , Pn−1 are in generic position. Then P 0 = τ (P ) 6=
P . Let H be the perpendicular bisector of P P 0 . Then Pi ∈ H for all 1 ≤ i ≤ n − 1 and
σH (P 0 ) = P . So σH τ (Pi ) = Pi for all i = 0, 1, . . . , n − 1. Thus, by (2), σH τ is a reflection
or the identity. If the latter occurs, then τ fixed n points in generic position. Hence, τ is
a product of two reflections.
(4) Suppose τ is an isometry and let m be the maximal number of points in generic
positions which τ fixes. We apply a downward induction on m. If m = n + 1 or n, we are
done by (1) and (2). Suppose now m < n and the assertion is true for m + 1. Consider a
point P such that P 0 = τ (P ) 6= P and a hyperplane H containing the m points in generic
position and perpendicularly bisecting P P 0 . Then σH τ fixed m + 1 points in generic
position. By induction, σH τ is a product of n − m reflections. Hence, τ is a product of
n − m + 1 reflections. By induction, the assertion is true for all m = n + 1, n, . . . , 1, 0. In
particular, when m = 0, we see that τ is a product of n + 1 reflections. 

Corollary 3.7. The group I (Rn ) is generated by reflections Hn,a for all 0 6= n, a ∈ Rn .
 GEOMETRIES AND TRANSFORMATION GROUPS 9

4. Translations and rotations on R2

Theorem 4.1. An isometry τ in Rn is a translation if and only if τ is the product of

two reflections in parallel hyperplanes. If HkH0 kH00 , then σH σH0 σH00 is a reflection.

Proof. Let H = Hn,a and H0 = Hn,b . Then, for all x ∈ Rn , we have, by Theorem 3.3,
 
xn an
σH0 σH (x) = σH0 x − 2 n+2 n
nn nn
 
xn an 2 xn an bn
=x−2 n+2 n− x−2 n+2 n n+2 n
nn nn nn nn nn nn
 
bn an
=x+2 − n
nn nn
= T2progn (b−a) (x).
Conversely, suppose τP,Q be the translation sending P to Q. Let n be a vector parallel
−→
to P Q and M the midpoint of P Q. Then, for H = Hn,P and H0 = Hn,M , we have
σH0 σH = τP,Q . 

We now consider the R2 case.

Corollary 4.2. A plane isometry is a translation if and only if it is a product of two

reflections in parallel lines.

What is the product two reflections in intersecting lines?

Definition 4.3. A rotation on R2 about a point C, through angle θ, is the transforma-

tion that fixes C and otherwise sends a point P to a point P 0 , where d(C, P ) = d(C, P 0 ),
−→ −−→
and the angle from CP to CP 0 is θ (in anti-clockwise direction if θ > 0, and clockwise if
θ < 0). We denote this transformation by ρC,θ

Theorem 4.4. An plane isometry is a rotation if and only if it is the product of two
reflections in intersecting lines. Further, we have
θ
(1) if lines l, m intersect at C, and the directed angle from l to m is 2
∈ (− π2 , π2 ], then
ρC,θ = σm σl ;
(2) if lines p, q, r are concurrent, then there exists a line l such that σr σq σp = σl .

Proof. (2) follows easily from (1). Let L ∈ l, L 6= C, and let M ∈ m, M 6= C. By a

geometrical argument, one checks easily that
M
σm σl (C) = C = ρC,θ (C), m

σm σl (L) = σm (L) = ρC,θ (L), C

L l
σm σl (M ) = ρC,θ (M ).
10 JIE DU

Hence, ρC,θ = σm σl . Reversing the argument proves the converse. 

Corollary 4.5. (1) A non-identity rotation (on R2 ) fixes exactly one point.
(2) A rotation with centre C fixes every circle with centre C.
(3) The set of all rotations about a particular point (with centre at a particular point) is
a subgroup of the group of isometries; further still, it is a commutative subgroup.
In other words, if RC := {ρC,θ : θ ∈ R}, then RC ≤ I (R2 ).

Theorem 4.6. (1) The rotation ρ0,θ : R2 → R2 about the origin 0 and through angle θ
is the linear isomorphism TU,0 (x) = U x, where U is the following matrix:
 
cos(θ) − sin(θ)
U= .
sin(θ) cos(θ)
(2) If c is the position vector of C, then ρC,θ = Tc (ρ0,θ )T−c . Hence, ρC,θ has the
equation ρC,θ (x) = U x + b, where U defines ρ0,θ as in (1) and b = (I − U )c. Moreover,
at the group level, we have RC = Tc R0 T−c , or RC is conjugate to R0 .

Proof. (1) By Theorem 4.4, we may assume that ρ0,θ = σm σl , where l is the x-axis, and
m has equation:
sin(θ/2)X − cos(θ/2)Y = 0.
Hence σm , by Corollary 3.4, has the equations:
(
x0 = x − 2 sin( 2θ )(sin( 2θ )x − cos( 2θ )y) = (cos(θ))x + (sin(θ))y
y 0 = (sin(θ))x − (cos(θ))y.
Also, σl has (more obvious) equation: X 0 = X, Y 0 = −Y . Hence, by multiplying matrices,
we can see that:  
cos θ − sin θ
σm σl (x) = x = U x,
sin θ cos θ
proving (1). For (2), we have

Tc ρ0,θ T−c =Tc σm σl T−c = (Tc σm T−c )(Tc σl T−c ) = σm0 σl0 = ρC,θ ,

where m0 = Tc (m) k m, l0 = Tc (l) k l, and m0 , l0 intersect at C, and

ρC,θ (x) = U (x − c) + c = U (x) + b,

where b = (I − U )c. 

A rotation of the form ρC := ρC,π is called a half-turn. A half-turn has equation:

x0 = −x + 2c, where c is the position vector of C.
Fact: If l is a line through the origin 0, then Tc σl T−c = σl0 , where l0 = Tc (l) is the
line parallel to l and through c.
Proof. Suppose l has equation x = λv for all λ ∈ R. Then l0 = Tc (l) has equation
x = c + λv. Clearly, Tc σl T−c fix l0 point wise. Now, the assertion follows from
Theorem 3.6(2) (for the n = 2 case). 
 GEOMETRIES AND TRANSFORMATION GROUPS 11

5. Congruence and the symmetry group of a geometric figure

Definition 5.1. (1) A figure F1 ⊆ Rn is congruent to a figure F2 ⊆ Rn if one can

be mapped onto the other by an isometry; i.e. if there exists an isometry τ such that
τ (F1 ) = F2 . Notation: F1 ∼
= F2 means F1 is congruent to F2 .

Theorem 5.2. If 4P QR ∼
= 4ABC in R2 (in the high school sense: same side lengths),
then there exists a unique plane isometry τ such that τ (P ) = A, τ (Q) = B, τ (R) = C.

Proof. The uniqueness is clear from Theorem 3.6(1). Let l be the perpendicular bisector
of P A. Then σl (P ) = A. Let Q0 = σl (Q) and let m be the perpendicular bisector of BQ0 .
Since d(A, B) = d(A, Q0 ), σm (Q0 ) = B and σm (A) = A. Finally, let n = `(A, B), the line
through the points A, B,
C
R

A B
P
Q

and let τ = σn σm σl . It is easy to check that τ (P ) = A, τ (Q) = B, τ (R) = C. 

Definition 5.3. Let Ω ⊆ Rn . A symmetry of Ω is an isometry τ such that τ (Ω) = Ω.

Proposition 5.4. (1) All the symmetries of a set Ω form a group sym(Ω), the “symmetry
group” of Ω.
(2) Let V = {v1 , v2 , . . . , vm } and τ : Rn −→ Rn be an isometry such that τ (V ) = V .
1
Then τ (cV ) = cV , where cV = m
(v1 + v2 + . . . + vm ) is the centre of mass of V .
(3) If G ≤ I (Rn ) be a finite subgroup, then there is some vector c ∈ Rn such that
gc = c for any g ∈ G. Moreover, G is isomorphic to a subgroup of the group On of
orthogonal matrices.

Proof. (1) Straightforward.

(2) By Theorem 2.5, we have τ (x) = U x + v for some B and v. Thus,
1 X 1 X 1 X
cV = τ (vi ) = U vi + v = U ( vi ) + v = U cV + v = τ (cV ).
m i m i m i
(3) Pick w ∈ Rn and let V = {g 0 w : g 0 ∈ G} ⊆ Rn be as in (2). Note V is finite as G
is finite. Note g(V ) = {gg 0 w : g 0 ∈ G} ⊆ V as gg 0 ∈ G for all g ∈ G. But that G is a
group implies g(V ) = V . Putting c = cV as in (2) gives the first assertion.
The group T−c GTc is isomorphic to G and fixes 0. By the Theorem above T−c GTc ≤
On , proving the last assertion. 
12 JIE DU

The symmetry group of a regular n-gon 4n is called a dihedral group. Notation:

Dn = sym(4n ).
Recall that any finite group G of isometries on Rn embeds in On . Let SOn = On ∩ SLn .
Clearly, [On : SOn ] = 2.

Lemma 5.5. Let K ≤ SO2 have order n, finite. Then K is cyclic group generated by
2π
the rotation σ about angle n
.

Proof. OK if n = 1, i.e. σ = id. Assume n > 1. Pick σ ∈ K, a rotation anti-clockwise

about angle θ where θ is the minimal positive amongst all such possibilities. We first
show hσi = K . We know hσi ⊆ K holds by the closure condition. Suppose h ∈ K is a
rotation anticlockwise about angle θ0 . Pick integer m so that mθ ≤ θ0 < (m + 1)θ. Note
σ −m h ∈ K is a rotation anticlockwise about angle 0 ≤ θ0 − mθ < θ. Minimality of θ
forces θ0 − mθ = 0 so θ0 = mθ and h = σ m ∈ hσi. Hence, K = hσi.
It remains to check θ = 2π
n
. Pick an integer l, so lθ ≤ 2π < (l + 1)θ. Since σ −l ∈ K is
a rotation anticlockwise about angle 0 ≤ 2π − lθ < θ, minimality of θ forces 2π = lθ. We
2π
must have l = n so θ = n
. This proves the lemma. 

Theorem 5.6. Any finite subgroup G of O2 is isomorphic to a cyclic or dihedral group.

In other words, a finite subgroup G of O2 is isomorphic to one of the groups on the
following list:
Cn , n = 1, 2, 3, . . . ,
(5.6.1)
Dn , n = 2, 3, 4, . . .
Proof. Let G ≤ O2 be a finite subgroup. If G is also in SO2 , we just apply Lemma 5.5.
Assume G  SO2 and let τ ∈ G − SO2 . τ is a reflection about say line l. Rotate to
change coordinates so that l is horizontal. This does not change the isomorphic class of
G. Then τ = 01 −10 . Let K = G ∩ SO2 ≤ SO2 . (For example, if l is parallel to cosθ



sinθ
,
cos2θ sin2θ −sinθ −1 −sinθ
and the reflection cosθ τ cosθ is 01 −10 .) By Lemma 5.5,





then τ = sin2θ −cos2θ sinθ cosθ sinθ cosθ
2π
K = hσi, where σ is the rotation anticlockwise about angle n
for some n. ∴ Dn ≤ G as
hτ, σi = Dn ⊆ G. It remains to prove the other way around inclusion.
Note that, by 3rdIT, G/K = G∩SO G ∼
= G·SO 2
⊆ O2 /SO2 ∼
= {SO2 , τ SO2 } ∼
= Z/2Z,
2 SO2
forcing |G/K| ≤ 2. But K 6= τ K. So we must have G/K = {K, τ K}. Hence every
element in G has form σ i or τ σ i , i.e. G ≤ Dn too. So G is dihedral. 

Theorem 5.7. Any finite subgroup of SO3 is either cyclic, dihedral or the rotational
symmetry group of a Platonic solid. In other words, a finite subgroup G of SO3 is
isomorphic to one of the groups on the list (5.6.1) together with:

RSym(regular polyhedra) : T , O, I.
 GEOMETRIES AND TRANSFORMATION GROUPS 13

6. Classification of Plane Isometries

The product of three reflections in concurrent of parallel lines is a reflection.

Definition 6.1. A plane isometry τ is called a glide reflection with axis c (a line) if there
exist distinct lines a, b which are perpendicular to c such that τ = σc σb σa (= σb σa σc ).

Note a, b will be parallel in R2 , so σa σb is a translation and σc σb , σa σc are halfturns.

Proposition 6.2. (1) A glide reflection is a composition of a reflection in line a and a

halfturn centred at a point off a.
(2) A glide reflection is a translation followed by a reflection
(3) A glide reflection fixes no points.
(4) A glide reflection fixes exactly one line, the axis, c.
(5) The midpoint of any point and its image under a glide reflection lies on its axis (c)

Theorem 6.3. Distinct lines p, q, r are neither concurrent, nor parallel, if and only if
σr σp σq is a glide reflection

Proof. The “if” is clear. That is, if lines p, q, r are concurrent or parallel, then σr σp σq is
a reflection. Conversely, suppose p, q, r are neither concurrent, nor parallel.

Case 1. p ∩ q 6= ∅, say p ∩ q = Q. Then

Q is off r (since p, q, r not concurrent). Let q
Q
P be the foot of the perpendicular from Q
to r, and m = `(P, Q). Then there ex- p
P
ists a line l (also through Q) such that: r
σq σp = σm σl . Thus σr σq σp = σr σm σl = m
(σr σm )σl = (halfturn) ◦ (reflection), which
is a glide reflection.

Case 2. Suppose p k q, and that r intersects p at P , q at Q. Now, let M be the foot of

the perpendicular from point Q to line p. Let m = `(M, Q). Let us consider the product
σr σq . It is a rotation (about point Q). There exists a line l such that σl σm = σr σq . Thus
we have: σr σq σp = σl σm σp = σl (σm σp ) = (reflection) ◦ (halfturn) = σl (σl0 σn ) (where l0 is
a line parallel to l through M , and n is a line perpendicular to l also through m. So we
have a glide reflection by defn at the end). 
14 JIE DU

Theorem 6.4 (Classification for Plane Isometries). Each non-identity plane isometry is
exactly one of the following:

translation rotation reflection glide reflection

Definition 6.5. An isometry that is the product of an even (resp., odd) number of
reflections is said to be even (resp., odd).

We would like to show that the intersection of the set of even isometries and the set of
odd isometries is the empty set; an isometry cannot be both “even” and “odd”.

Theorem 6.6. A product of 4 reflections on R2 is a product of 2 reflections.

Proof. Suppose σs σr σq σp is given with lines p, q, r, s. Pick a point P on p.

Then there exists lines q 0 , r0 with P on q 0

q
such that σr σq = σr0 σq0 . (If q, r intersect
at C, then so does q 0 , r0 .) Similarly, there C
P p
exists lines r00 , s0 with P on r00 such that
σs σr0 = σs0 σr00 . Since p, q 0 , r00 are concurrent
at P , then there exists a line l such that r s
σr00 σq0 σp = σl .

Hence σs σr σq σp = σs0 σr00 σq0 σp = σs0 σl . 

Corollary 6.7. No isometry is both even and odd. (May argue by using determinants.)

Proof. By Thm6.6, we may assume σr σq σp = σs σt . Then there exist lines l, m such that
σm σl = σs σr σq σp = σt , which is imposible. 

Corollary 6.8. The set E of all even isometries forms a subgroup.

So far, we have
 (
 translations
even − − rotations



nonidentity plane isometries − − − (
 reflections
odd − − glide reflection



 GEOMETRIES AND TRANSFORMATION GROUPS 15

7. Similarities

Definition 7.1. A transformation α : Rn → Rn is called a similarity of ratio r > 0 if

d(α(P ), α(Q)) = rd(P, Q), for all P, Q ∈ Rn .

The following properties of similarities are mostly straightforward.

Proposition 7.2. (1) An isometry is a similarity of ratio 1.

(2) A similarity fixing two points is an isometry.
(3) A similarity fixing n + 1 points in generic position is the identity.
(4) The set of all similarities in Rn forms a group, denote this set by S or S (Rn ).

Proof. (4) follows from the fact that R>0 is a group: If τ, σ ∈ B(Rn ) and
d(τ (P ), τ (Q)) = rd(P, Q), d(σ(P ), σ(Q)) = sd(P, Q), then d(στ (P ), στ (Q)) = rsd(P, Q)
and d(τ −1 (P ), τ −1 (Q)) = r−1 d(P, Q) for all points P, Q. 

Here is a special type of similarities.

Definition 7.3. A stretch of ratio r > 0 about point C is a transformation δC,r that fixes
C and otherwise sends a point P to a point P 0 , where P 0 is the unique point on the ray
from C through P such that d(C, P 0 ) = rd(C, P ). The equation of δC,r is

δC,r (x) = rx + (1 − r)c(= r(x − c) + c).

Theorem 7.4. If α is a similarity of ratio r > 0, and P is any point, then α = τ ◦ δP,r =
δP,r ◦ τ 0 , for some isometries τ, τ 0 . In particular, we have S = r>0 I δP,r = r>0 δP,r I .
S S

−1
Proof. First, observe that δP,r is a similarity itself, and δP,r = δP,r−1 is a similarity. Hence
−1 1 −1
α ◦ δP,r is a similarity, of ratio r × r
= 1. This means that α ◦ δP,r is an isometry, say τ .
Thus, α = τ ◦ δP,r , as required. 

Corollary 7.5. A similarity is a collineation that preserves betweenness, midpoints,

angles, perpendicularity, etc.

Definition 7.6. (1) A point reflection (or central inversion) about C(c) is the transfor-
mation ρC : Rn → Rn defined by

ρC (x) = −(x − c) + c = −x + 2c.

If n = 2, then ρC is a halfturn.
(2) A dilation about a point C is a stretch about C, or it is a stretch followed by a
point reflection about C.
16 JIE DU

Proposition 7.7. The dilation τ = ρC δC,r has the following equation:

τ (x) = (−r)x + (1 + r)c.

Hence, the notation δC,r represents a stretch if r > 0 or a stretch followed by a point
reflection if r < 0.

Proof. It is clear as ρC δC,r (x) = −[r(x − c) + c] + 2c = (−r)(x − c) + c = RHS. 

Lemma 7.8. For any r 6= 0, and any point P , we have

(1) δP,−r = ρP δP,r ;
(2) δP,1 = 1, δP,−1 = ρP ;
(3) δP,r δP,s = δP,rs ;
−1
(4) δP,r = δP,r−1 .

Proposition 7.9. (1) {δC,r | r ∈ R − 0} forms a group isomorphic to R× .

(2) All point reflections generate a group H that is a union of T and the set of all
point reflections.

Proof. (1) Exercise.

(2) A point reflection has order 2. Every element of H is a word in point reflections.
Since the product of two point reflections is a translation, while the product of a point
reflection with a translation is a point reflection. The assertion follows. 
 GEOMETRIES AND TRANSFORMATION GROUPS 17

8. Dilatations

Definition 8.1. A collineation τ on Rn is called a dilatation if, for every line l, l k τ (l)
(i.e., every line is parallel to the line it’s mapped to).

Example 8.2. Translations and dilations are dilatations.

For a line l with equation l : x = a + λv, we say that the direction of l is v. In other
→
− v
words, we define the direction of line l, l := kvk .
If points A, B ∈ l, then we define directed distance by setting
 −→ →
−
AB = d(A, B) if AB has same direction as l
−d(A, B) if we are in the other case
In other words, if A, B have position vectors a, b, respectively, then
−→ →
−
AB = b − a = (AB) l .
CB
In particular, for distinct collinear points A, B, C, if CA
= r 6= 0, then b − c = r(a − c).
CB
Lemma 8.3. (1) If A, B, C are collinear, distinct, with CA
= r 6= 0, then δC,r (A) = B.
AP 0
(2) For collinear points A, B, P, P 0 , if AP
PB
= P 0B
, then P = P 0 .
(3) Let τ be a dilatation and let τ (P ) = P 0 for every point P . If there exist points A, B
−→ −−→
such that AB and A0 B 0 have the same (resp., opposite) direction, then, for any points
−−→ −−→
C, D, CD and C 0 D0 have the same (resp., opposite) direction.
CB
Proof. (1) Since CA
= r implies b − c = r(a − c), δC,r (a) = r(a − c) + c = (b − c) + c = b.
AP 0
(2) Suppose PB
= PAP0 B = r. Then r 6= −1 (r = −1 =⇒ A = B). Thus, we have
p − a = r(b − p) and p − a = r(b − p0 ). Hence, (1 + r)p = rb + a = (1 + r)p0 and so
0

p = p0 .
−→ −−→ −→ −−→
(3) If AB and A0 B 0 are in the same direction, then so do AC and A0 C 0 . This is because
−→ −−→ −−→ −−→
if AC and A0 C 0 are in opposite direction, then BC and B 0 C 0 are not parallel. (Prove it!
If b − a = λ(b0 − a0 ), a − c = µ(c0 − a0 ) and c − b = ν(c0 − b0 ), where λ, µ > 0 and ν 6= 0,
then λ(b0 − a0 ) + µ(c0 − a0 ) = ν(b0 − c0 ). This is impossible since the LHS is in the 1st
quadrant of the plane R(b0 − a0 ) + R(c0 − a0 ) with ‘origin’ a0 , but the RHS is in the 2nd
or 4th quadrant.) In other words, 4ABC and 4A0 B 0 C 0 must have the same orientation.

A B C0

A0 B0
18 JIE DU

−−→ −−→
Similar argument for 4ACD shows that that CD and C 0 D0 are in the same direction.


Corollary 8.4. If points A, B, C are sent to A0 , B 0 , C 0 under a dilatation, then AB

A0 B 0
=
BC CA
B0C 0
= C 0 A0
.

Theorem 8.5. A dilatation is either a translation or a dilation. Hence, every dilatation

is a similarity.

Proof. Let τ be a dilatation. Consider points P1 , . . . , Pn , Pn+1 ∈ Rn in generic position

and the images Pi0 = τ (Pi ). Assume that P1 , . . . , Pn ∈ Rn lie in the hyperplane H. Then
τ (H) = H0 kH and also the translation τP1 ,P10 sends H to H0 . Let Pi00 = τP1 ,P10 (Pi ) for
i = 1, . . . , n, n + 1. Since `(P1 , Pj ) is parallel to `(P10 , Pj0 ) and also to `(P100 , Pj00 ), it follows
that `(P10 , Pj0 )k`(P100 , Pj00 ). Thus, `(P10 , Pj0 ) = `(P100 , Pj00 ) as P10 = P100 . Hence, 4P10 Pi0 Pj0 ∼
P10 Pi0 P10 P20
4P10 Pi00 Pj00 for all 2 ≤ i 6= j ≤ n + 1. By Corollary 8.4, we have P10 Pi00
= P10 P200
=: r for
all i = 2, 3, . . . , n + 1. (The n + 1 case is seen by applying the argument above to the
hyperplane through P1 , . . . , Pn−1 , Pn+1 .)
If r = 1, then Pi0 = Pi00 and τ = τP1 ,P10 is a translation. If r 6= 1, let C = P10 . Then,
by Lemma 8.3, δC,r sends every Pi00 to Pi0 . Hence, δC,r τP1 ,P10 sends every Pi to Pi0 for all i.
Therefore, τ = δC,r τP1 ,P10 . This is a dilation. 

Corollary 8.6. A dilatation in Rn fixes n point in generic position is the identity map.

Proof. Let δ be a dilatation that fixes P1 , . . . , Pn (n ≥ 2). Then δ is an isometry. By

Theorem 3.6, δ is either a reflection or identity. But reflection is NOT a dilatation.
Hence, δ is the identity. 
 GEOMETRIES AND TRANSFORMATION GROUPS 19

9. Classification of Plane Similarities

Theorem 9.1. A similarity without a fixed point is an isometry.

Proof for the n = 2 case. Suppose α is a similarity that is not an isometry. We would
like to prove that α has a fixed point. Since, by Theorem 8.5, a dilatation is either a
translation (which is an isometry), or a dilation (that has a fixed point), we may assume
α is not a dilatation (as otherwise we would be “done”). So there exists a line l such that
α(l) ∦ l. Let l intersect l0 := α(l) at A and let A0 = α(A). (The intersection exists for
non-parallel lines in R2 !)
If A = A0 , we are done (fixed point). Assume A 6= A0 . Let m be the line through A0
parallel to l and let m0 = α(m) (the image of the line m under α). Since l k m, we have
l0 k m0 (Similarities take parallel lines to parallel lines).

l m
B
B0
m0

P0
A0
l0 A

Let B = m0 ∩m, and let B 0 = α(B). Then B 0 is on m0 , and distinct from A0 (since m0 , l0
parallel). If B = B 0 , we are done (fixed point). We may assume B 6= B 0 , so l0 = `(A, A0 ),
m0 = `(B, B 0 ).
If `(A, B) k `(A0 , B 0 ), then d(A, B) = d(A0 , B 0 ) = d(α(A), α(B)) (as we would have a
parallelogram AA0 B 0 B). But then α is an isometry, contrary to our assumption
So we may assume that `(A, B) ∦ `(A0 , B 0 ). Let P = `(A, B) ∩ `(A0 , B 0 ). Then P is
A0 P
off both l0 , m0 . Now, l0 k m0 ⇒ AP
PB
= P B0
(by similar triangles). Suppose α has ratio
A0 P 0 A0 P A0 P 0
r, P 0 = α(P ). Then AP
PB
= rAP
rP B
= P 0B0
and so P B0
= P 0B0
. Since P 0 ∈ α(`(A, B)) =
`(A0 , B 0 ), Lemma 8.3(2) forces P = P = α(P ). Hence, α fixes a point.
0

Proof (by Robert Tan, a 2017 student). By Theorem 7.4, any similarity α = τ δ0,r for
some r > 0. So α has equation α(x) = rQx + b for all x ∈ Rn , where Q is an orthogonal
matrix and b ∈ Rn . Suppose now that α has no fixed point. In other words, the
linear system (rQ − I)x = −b has no a solution. This implies that det(rQ − I) = 0
or equivalently, det(Q − r−1 I) = 0. Hence, r−1 is an eigenvalue of Q. Suppose v is an
20 JIE DU

associated eigenvector, i.e., Qv = r−1 v. Since Q is orthogonal, we have v and r−1 v have
the same length, forcing r = 1. Hence, α is an isometry. 

Definition 9.2. (1) A stretch reflection on R2 is a non-identity stretch about some

point C followed by a reflection about a line through C.
(2) A stretch rotation on R2 is a non-identity stretch about some point C followed by
a non-identity rotation about C.

Exercise 9.3. For stretch reflections and stretch rotations, we have δC,r σl = σl δC,r and
δC,r ρC,θ = ρC,θ δC,r , where C ∈ l. It can be prove that the product of two stretch rotations
is a stretch rotation. The set of all stretch rotations and the identity forms a group. [See
Duzin and Chebatarevsky’s book].

Theorem 9.4. A non-identity planes similarity is exactly one of the following:

Isometry, Stretch of ratio r 6= 1, Stretch reflection, Stretch rotation.

Proof. In this proof, we use the theorem we just proved about similarities fixing a point,
and we use the decomposition of similarities.
Suppose α is a similarity which is not an isometry. Then by Theorem 9.1, α fixes at
least one point C. By Theorem 7.4, we may write α = τ ◦ δC,r , where τ is an isometry,
δC,r is a stretch. Then τ (C) = αδC,r−1 (C) = C. Thus τ is an isometry that fixes at least
one point, and we can fall back on our earlier classification of plane isometry theory: τ
is either a reflection in a line through C, a rotation about C, or the identity. Hence, α is
either a stretch, stretch reflection, or a stretch rotation. 
 GEOMETRIES AND TRANSFORMATION GROUPS 21

10. More on Groups

Recall the transformation groups T , I , H , S , D ≤ B.

from Proposition 7.9 that H = H (Rn ) is the subgroup generated by all point reflec-
tions ρC , C ∈ Rn and, since a dilation δC,r that is an isometry is the identity for r = 1 or
point reflection for r = −1, we have

H = T ∪ {ρC | C ∈ Rn } = D ∩ I .

Theorem 10.1. Suppose α ∈ S is a similarity, and G ∈ {I , D, H , T }. Then ατ α−1 ∈

G, for all τ ∈ G. In other words, each of the groups I , D, H , T is a “normal subgroup”
of S .

Proof. Case 1: G = I . If τ is an isometry, then τ is a similarity of ratio 1. If α

is a similarity of ratio r, then ατ α−1 is a similarity of ratio r × 1 × 1
r
= 1. Hence
ατ α −1
∈ I = G.
Case 2: G = D. If τ is a dilatation, then α−1 (l) k τ (α−1 (l)), for any line l. But
α = isometry ◦ dilation. Hence l = αα−1 (l) k ατ α−1 (l). Hence, ατ α−1 is also a dilatation.
Case 3: G = H . Since D = T ∪ {dilations} and a dilation δC,r that is an isometry
is the identity for r = 1 or a point reflection for r = −1, it follows that

H = T ∪ {ρC | C ∈ Rn } = D ∩ I .

Thus, Case 3 follows form Cases 1 and 2.

Case 4: G = T .
If τ ∈ T , ατ α−1 ∈ H (since T ≤ H ). But ατ α−1 is not an involution since
(ατ α−1 )2 = ατ 2 α−1 6= 1. Hence, ατ α−1 ∈ T . 

Corollary 10.2. For α ∈ S , a point C and a hyperplane H,

αρC α−1 = ρα(C) , ασH α−1 = σα(H) , αδC,r α−1 = δα(C),r .

In particular, in R2 , αρC,θ α−1 = ρα(C),±θ .

Definition 10.3. Let G be a group.

(1) For H, K ⊆ G, define subset multiplication HK := {hk : h ∈ H, k ∈ K}.
(2) A subgroup K of G is called a normal subgroup if gK = Kg, for all g ∈ G.
Notation: K E G.
(3) For groups G, H, a bijective map φ : G → H is called a (group) isomorphism if
φ(gg 0 ) = φ(g)φ(g 0 ) for all g, g 0 ∈ G. Notation: G ∼
= H.

Proposition 10.4. (1) If H ≤ G, then G is a disjoint union of cosets gH, q ∈ G. (Q:

find a subset S ⊂ G such that G = ˙
S
gH is a disjoint union.)
g∈S
22 JIE DU

(2) If K E G, then G/K := {gK | g ∈ G} is a group with the subset multiplication.

(G/K is called the quotient group of G by K.)

Proof. (1) Define a relation ∼ on G by setting g ∼ g 0 ⇐⇒ g −1 g 0 ∈ H. Show that this is

an equivalence relation and the equivalence class containing g has the form gH.
(2) For g, g 0 ∈ G, (g 0 K)(gK) = g 0 g(g −1 Kg)K = g 0 gK. So it is a well defined multipli-
cation. The subset multiplication is associative. The identity is the coset 1K = K and
(gK)−1 = g −1 K. 

Examples 10.5. (1) The translation group T is isomorphic to the abelian group (Rn , +).
(2) For a fixed point P ∈ Rn , the subset {δP,r | r ∈ R − 0} forms a subgroup of D.
This subgroup is isomorphic to the abelian group (R× , ·). We have also D/T ∼ = (R× , ·).
(3) S = ˙ r∈R>0 δC,r I , where C is a fixed point. We claim that the union is a disjoint
S

union. Indeed, if δC,r I ∩ δC,s I 6= ∅, then there exists τ, τ 0 ∈ I such that δC,r τ = δC,s τ 0 ,
which implies that δC,r δC,s−1 = τ 0 τ −1 . Now, the right hand product is an isometry, so the
product on the left hand side must also be an isometry. This means that r × s−1 = 1 or
r = s. Thus, if the intersection is non-empty, then δC,r I = δC,s I .
Consider now the abelian group (R>0 , ·). We now prove that the quotient group
S /I := {δC,r I : r ∈ R>0 } is isomorphic to (R>0 , ·).
There is a bijective map from S /I to (R>0 , ·), sending δC,r I to r. Since

(δC,r I )(δC,s I ) = δC,r (I δC,s )I = (δC,r δC,s )I I = δC,rs I ,

this map is a group isomorphism.

(4) Show that D/H ∼

= (R>0 , ·). This is S
seen from the fact that
[
D= δ0,r H , D I
r∈R>0
or
D/T ∼ × H
D/H ∼ = = R /{±1} ∼ = (R>0 , ·).
H /T
We draw the following Hasse diagram to see
T
the relationship between the groups:
 GEOMETRIES AND TRANSFORMATION GROUPS 23

11. Collineations

A transformation τ : Rn → Rn (n ≥ 2) is a collineation if τ (l) is a line for every line l.

Theorem 11.1. A transformation is a collineation iff the images of collinear points are
themselves collinear.

Proof. The “only if” is trivial; that is, it is obvious that a collineation sends collinear
points to collinear points.
We must prove the “if” part; that is, if the images of collinear points are themselves
collinear, then the map is a collineation (sends lines to lines).
Suppose that α is a transformation that preserves collinearity. We wish to show that
α(l) is a line whenever l is a line. Let A, B be distinct points on a line l. Then let
m = `(α(A), α(B)). By the definition of α, we know that α(l) ⊆ m. We must show
that α(l) = m. Suppose C 0 ∈ m, distinct from α(A), α(B). Let C be a point such that
α(C) = C 0 . Suppose C 6∈ l. Then the images of all points on lines `(A, B), `(A, C),
`(B, C) are on m (since α sends collinear points to collinear points, and α(A), C 0 , α(B)
are collinear.).

P
C

m
l
B C0
A B0 A0

However, any point P in the plane through A, B, C is on a line that intersects at least
two of the three previously mentioned lines, and even stronger than that, can be placed on
a line that intersects two of the three sides of the triangle 4ABC. Then since the images
of those two intersecting points are on m, this point P must also be on m. Therefore, for
the plane π through A, B, C, α(π) ⊆ m. Thus, α maps n + 1 points including A, B, C in
generic position to points that must be contained in a hyperplane, contradicting the fact
that α is onto. 

Lemma 11.2. If α is a collineation, and l, m are parallel lines, then α(l) and α(m) are
parallel.

Proof. We first prove that α(l), α(m) are coplanar. Suppose A ∈ l and B ∈ m. Let
p = `(A, B). Then m lies in the plane spanned by l and p. Since α(l) intersect α(p) at
24 JIE DU

α(A), α(l) and α(p) span a plane Π. For any point X on m, there exists a line q through
X intersecting l at C and p at D. Hence, line α(q) intersects both α(l) and α(p), and
consequently, α(X) ∈ Π. This proves that α(l), α(m) are coplanar.
Suppose α(l), α(m) are not parallel. Let the intersection of α(l), α(m) be the point
Q. Then, since Q ∈ α(l), there exists a point A ∈ l such that α(A) = Q. Similarly,
since Q ∈ α(m), there exists a point B ∈ m such that α(B) = Q. But α is a bijection
Rn → Rn , so it must be that B = A, since α is one-to-one. Thus A lies on both l, m,
thus the lines l, m intersect, and so are not parallel. 

Theorem 11.3. A collineation takes the midpoint of points A, B to the midpoint of points
α(A), α(B).

Proof. Let A, B be two distinct points, and let α be a collineation. If P 6∈ `(A, B), we
form a parallelogram, AP BQ. Let A0 = α(A), etc. Then by Lemma 11.2, A0 P 0 B 0 Q0
is a parallelogram.
Further, define M := `(A, B) ∩ `(P, Q). This must be mapped to M 0 = `(A0 , B 0 ) ∩
`(P 0 , Q0 ) (since M is on line l(A, B), l(P, Q), M 0 must be on both `(A0 , B 0 ), `(P 0 , Q0 ),
thus it is the intersection of those two lines). Since diagonals of a parallelogram bisect
each other, we see that M is the midpoint of AB, and M 0 is the midpoint of A0 B 0 . Hence
α preserves midpoints. 

Corollary 11.4. For a collineation α, if n + 1 points P0 , ..., Pn divide the segment P0 Pn

into n congruent segments Pi−1 Pi , and Pi0 = α(Pi ), then the n + 1 points P00 , ..., Pn0 divide
the segment P00 Pn0 into n congruent segments Pi−1
0
Pi0 .
In particular, if a point P is between A and B, and AP
PB
= r is rational, then P 0 = α(P )
A0 P 0
is between α(A) and α(B) and P 0B0
= r.

Proof. Since Pi is the midpoint of Pi−1 Pi+1 , applying Theorem 11.3 gives the first asser-
P0 Pp
tion. If AP
PB
= pq , then take n = p + q so that Pp Pp+q
= pq , where P0 = A, Pp = P , and
P 0P 0
Pp+q = B. Now applying the first assertion gives P 0 P0 0 p = pq , where α(Pi ) = Pi0 . Hence,
p p+q

α(P ) is between α(A) and α(B). 

 GEOMETRIES AND TRANSFORMATION GROUPS 25

12. Darboux’s Theorem

AP
Lemma 12.1. Let t > 0, t 6= 1, and P, Q be points on `(A, B) such that PB
= t,
AQ AC
QB
= −t. Then C is the midpoint of P, Q if and only if CB
= −t2 .

Proof. (Exercise, PS4) Denote the position vector of P by p etc. By the hypothesis, we
have (1) p − a = t(b − p); (2) q − a = −t(b − q), or

(3) (1 + t)p = a + tb, (4) (1 − t)q = a − tb.

AC
Suppose CB
= −t2 . Then (1 − t) × (3) + (1 + t) × (4) gives (1 − t2 )(p + q) = 2a − 2t2 b or
p+q a−t2 b
2
= 1−t2
= c since c − a = −t2 (b − c). Hence, C is the midpoint of P, Q. Conversely,
suppose C is the midpoint of P, Q. (1) + (2) gives p + q − 2a = t(q − p), and (2) − (1)
gives q − p = −2tb + t(p + q) or t(q − p) = −2t2 b + t2 (p + q). Hence, c − a = −t2 (b − c),
AC
proving CB
= −t2 . 

Theorem 12.2. If α is a collineation, and point P is between points A, B, then α(P ) is

between α(A), α(B).

Proof. Let A0 = α(A), B 0 = α(B), etc. Since α−1 is a collineation (as we have proved the
set of collineations forms a group), it suffices to prove that, for C ∈ `(A, B), if C 6∈ AB,
then C 0 6∈ A0 B 0 .
AC −→ −−→
Suppose C ∈ l(A, B), but C 6∈ AB. Then it is true that CB
< 0 as AC,CB are in
AC 2
opposite directions. Thus, there exists t > 0, such that CB
= −t . Let P, Q be points on
AP AQ
`(A, B) such that PB
= t, QB
= −t. Then by Lemma 12.1, C is the midpoint of P Q.
D

A P R B C Q

Choose points D, E so that `(A, B) is the perpendicular bisector of DE at the point

A. Let the line through B which is parallel to DE intersect `(D, Q) at F , and intersect
`(E, Q) at G. Then (by high school geometry/similar triangles), B is the midpoint of
AR AD AQ AQ
F G. Since, for R = DG ∩ EF , RB
= BF
= BQ
= − QB = −(−t) = t, it follows that
AR AP
RB
=t= PB
, forcing R = P .
26 JIE DU

Consider the image under α of the geometrical figure above. Then α sends collinear
points to collinear points, midpoints to midpoints, parallel lines to parallel lines, and
concurrent lines to concurrent lines, etc. So what we know are: (1) A0 , B 0 , P 0 , C 0 , Q0 are
collinear; (2) D0 E 0 k F 0 G0 ; (3) A0 is the midpoint of D0 E 0 , B 0 is the midpoint of F 0 G0 ,
and C 0 is the midpoint of P 0 Q0 ; (4) D0 , E 0 , F 0 , G0 are off l(A0 , B 0 ), and (5) Preservation of
concurrency.
By considering which of D0 , F 0 , Q0 is between the other two, there are 3 cases.
Case 1: D0 − F 0 − Q0 ; Case 2: F 0 − Q0 − D0 ; Case 3: F 0 − D0 − Q0 .
D0 G0 F0
F0 D0 D0

A0 B0 Q0 P0 B0 B0 A0 Q0
0
P0 C0 C0 A Q0 P0 C0
E0 E0
G0 F0 G0
E0

(1) In this case, the points A0 , P 0 , B 0 , Q0 on l0 are the same as A, P, B, Q on l. By

D 0 A0
similar triangles, there exists a number s ∈ R − {0} where s = G0 B 0
, such that
A0 P 0 A0 Q0 A0 Q0 D 0 A0 D 0 A0
P 0B0
= s, Q0 B 0
= −B 0 Q0 = − B 0 F 0 = − G0 B 0 = −s. (Note in cases 1,3, s > 0, case 2,
A0 C 0
s < 0) Since C 0 is necessarily the midpoint of P 0 , Q0 , then C 0B0
= −s2 , by Lemma
12.1. The negativity of −s2 implies that C 0 is off A0 B 0 .
(2) Now the ordering on l0 is P 0 , A0 , Q0 , B 0 (as pictured above) or A0 , Q0 , B 0 P 0 . The
picture is the same again after appropriate permutation of points. The proof is
similar.
(3) This is the case by moving Q0 to the left of A0 , or points A0 , D0 , E 0 are swapped
with B 0 , F 0 , G0 . So the same argument show that C 0 is off A0 B 0 .
(Since Q is off DF , applying the proof the points D, F, Q instead of A, B, C yields the
conclusion that Case 2 is impossible.) 

Corollary 12.3. (1) A collineation fixing two points on a line fixes the line pointwise.
(2) A collineation fixing 3 non-collinear points fixes the plane through them pointwise.
(3) A colineation in Rn fixing n + 1 points in generic position is the identity.

Proof. To come. 
 GEOMETRIES AND TRANSFORMATION GROUPS 27

13. Affine Transformations

Proof of Corollary 12.3. Suppose α is a collineation that fixes A, B on line l. Suppose

that C ∈ l, but that C 0 = α(C) 6= C. Without loss, we may assume that C is on the ray
from A through B. Let B0 = B and define Bi+1 ∈ l, so that Bi is the midpoint of ABi+1 .
Since α fixes A and B = B0 , α fixes B1 . This is because if α did move B1 , then it
would move B0 by Theorem 11.3. Inductively, α fixes every Bi .

A B0 B1 B2 C

Let D = Bk where k is an integer such that ABk > 2k AB > AC. Then, C lies on AD.
By Darboux’s theorem, α fixes the whole line segment AD (not necessarily pointwise,
yet!). Thus since C is on the line segment and the line segment is fixed, C 0 must also be
on AD. So we have C 0 , C both on AD. Thus, there exists n ∈ Z so that 2n |CC 0 | > AD.
Now, define P0 = A, P2n = D, and the n + 1 points P0 , P1 , ..., P2n divide AD into 2n
congruent segments Pi−1 Pi such that P2n−1 is the midpoint of AP2n , P2n−2 is the midpoint
of AP2n−1 , ..., P1 is the midpoint of AP2 , etc. Thus, each of the points Pi is fixed, by
Corollary 11.4. So each APi , and Pi D is fixed by α.
However, since |CC 0 | > AD
2n
= Pi−1 Pi , there exists Pj such that Pj is between C and
C 0 . So C, C 0 are in different fixed segments APj and Pj D, a contradiction.
(2) If α fixes non-collinear A, B, C, then α fixes `(A, B), `(B, C), `(C, A) point wise.
Every point P on the plane lies in a line that intersects two of the three lines in (at least)
two distinct points. So α fixes that line. Hence, α fixes P .
(3) Apply induction. (Exercise.) 

Definition 13.1. An affine transformation α = TA,b : Rn → Rn is one that has an

equation of the form α(x) = Ax + b, where A ∈ GLn (R), b ∈ Rn .

Lemma 13.2. The set A of all affine transformations forms a group. Moreover, S is
a subgroup of A .

−1
Proof. The following can be checked easily: TA2 ,b2 TA1 ,b1 = TA2 A1 ,A2 b1 +b2 , TA,b =
TA−1 ,−A−1 b . 

Theorem 13.3. Let τ be a transformation. Then the following are equivalent:

(1) τ is an affine transformation;

(2) τ is a collineation.
28 JIE DU

Proof. (1) ⇒ (2): Let l be a line, defined by: l : x = a + λv, λ ∈ R. Let α be an affine
transformation, α(x) = Ax + b. Then α(l) = {Aa + b + λ(Av) : λ ∈ R}, which is clearly
a line (a line parallel to Av, that passes through Aa + b). Hence, α is a collineation.
(2) ⇒ (1): Let τ be a collineation and let P = τ (0), Qi = τ (ei ). Then P, Q1 , ..., Qn
are in generic position. Let p, q1 , ..., qn be the position vectors of P, Q1 , ..., Qn . Then
q1 − p, ..., qn −
 p are linearly independent. 
Define A = (q1 − p)|(q2 − p)|...|(qn − p) . Since the column vectors are all linearly
independent, then we know that nullity(A) = 0, and A is invertible. Let α = TA,p . Then
α(0) = p, α(ei ) = qi . Hence ατ −1 fixes P1 , Q1 , ..., Qn . By Corollary 12.3, ατ −1 = id,
thus α = τ . 

Definition 13.4. An affine transformation α as given in Definition 13.1 is called an equi-

affine transformation if det(A) = ±1. A plane equi-affine transformation is called is an
equiareal transformation.

Theorem 13.5. (1) If α be a plane affine transformation as given in Definition 13.1 and
α(P ) = P 0 , etc., then area(4P 0 Q0 R0 ) = | det A|area(4P QR).
(2) If Ω is the parallelepiped spanned by the vectors a, b, c in R3 and α is an affine
transformation on R3 , then vol(αΩ) = | det(A)|vol(Ω).

Proof. (1) If p, q, r are position vectors of P, Q, R, then area(4P QR) = 21 | det(D)|, where
 
p1 p2 1
D = q1 q2 1 .
r1 r2 1
Let M = (p − r, q − r) and M 0 = (p0 − r0 , q0 − r0 ). Then
1 1
area(4P 0 Q0 R0 ) = | det(M 0 )| = | det(A)|| det(M )| = | det A|area(4P QR),
2 2
as desired.
(2) Since translation does not change the volume, we simply assume that α is linear
(i.e., b = 0). The perpendicular height of the parallelepiped is
|a  (b × c)| |a  (b × c)|
kprojb×c (a)k = = .
kb × ck area of base
Hence, vol(Ω) = |a  (b × c)|. Since a  (b × c) = det(a|b|c), it follows that

vol(αΩ) = | det(Aa|Ab|Ac)| = | det(A)|| det(a|b|c)|,

as desired. 
 GEOMETRIES AND TRANSFORMATION GROUPS 29

14. The Real Projective Line RP 1 , Plane RP 2 and Space RP n

In the real number system R, every non-zero real number has at most two square
roots. Why not exactly two? The reason is that negative numbers do not have any
square root.
This was considered to be a defect of the real number system. To make the number
system more perfect, “imaginary numbers” were invented to provide square roots for
negative numbers. What mathematician did was to increase the dimension by 1 to form
the so-called complex numbers, which are two dimensional numbers! In other words, the
imaginary numbers are actual element in a higher dimensional space. In this new number
system C, we can say that every non-zero number has exactly two square roots.
In analogy to this scenario, the Euclidean plane R2 has the following defect: Given any
two lines, there are at most one point lying in both. However, given any two distinct
points, there is exactly one line passing through them.
We are going to invent some imaginary points, where parallel lines can meet. We call
these extra points ideal points. All ideal points form a line called the ideal line `∞ .
The new space R2 ∪ `∞ will be called the real projective plane.
By inceasing dimension, the ideal point can be realised as actual elements in a higher
dimensional space.
We start with the construction of the real projective line—the ideal line for the real
projective plane.
Observe the following link:
R −→ L0 = {lines through 0 in R2 }
(14.0.1)
m 7−→ y = mx
This map is not surjective as the line x = 0 has no preimage! If we add an imaginary
point ∞ to R, we obtain a bijection R ∪ {∞} ↔ L0 . This new space RP 1 := R ∪ {∞} is
called the real projective line.
On the projective line, there is only one ideal point, all other point are ordinary. Note
this new structure we’ve created is not a field. You may think of as a line whose two ends
meet at the ideal point ∞.
We now can define the real projective plane.

Definition 14.1. (1) The real projective plane RP 2 is defined as the extended Euclidean
plane
RP 2 := R2 ∪ RP 1 (disjoint union).
The points in R2 (resp., RP 1 ) are called ordinary (resp., ideal) points and `∞ := RP 1 is
called the ideal line.
30 JIE DU

(2) In general, for n ≥ 2, define the real projective space

[
RP n = R n
|{z} RP n−1
| {z } (disjoint union)
ordinary points ideal points

as a disjoint union of the ordinary part and the ideal part — the ideal hyperplane.

Homogeneous coordinates: What are the coordinates of points in these new space?
We modify the correspondence (14.0.1) by identifying a line equation with its solution
set:
RP 1 −→ L homodeneous coordinates
m 7−→ y = mx [1, m]
∞ 7−→ x = 0 [0, 1]
Here [a, b] represents the (non-zero) solution set [a, b] := {(λa, λb) | λ ∈ R − 0} of such
a line for any non-zero solution (a,b). In other words, we have the following points-vs-
coordinates correspondence:
RP 1 = R ∪ {∞} −→ {[a, b] | (a, b) ∈ R2 − (0, 0)}
ord. pt: m 7−→ [1, m]
ideal pt: ∞ 7−→ [0, 1]
This map is bijective! Similarly, for the real projetive plane, we have
RP 2 = R2 ∪ RP 1 −→ {[a, b, c] | (a, b, c) ∈ R3 − 0}
ord pts: (a, b) 7−→ [1,a,b]
ideal pts: [c, d] ∈ RP 1 7−→ [0,c,d]
The homogeneous coordinates for real projective space RP n can be constructed in
a similar fashion and write P [x0 , x1 , . . . , xn ] ∈ RP n for a point P with homogeneous
coordinates [x0 , x1 , . . . , xn ].

What are the equations of lines in RP 2 ?

x1 x2
For ordinary lines in R2 : a + bx + cy = 0, where (b, c) 6= (0, 0). Let x = x0
,y = x0
(x0 6= 0). A point [1, x, y] on the line has the form [x0 , x1 , x2 ] and equation becomes
ax0 + bx1 + cx2 = 0. The ideal line l∞ = {[0, x1 , x2 ] | (x1 , x2 ) ∈ R2 − (0, 0)}. So
its equation is x0 = 0. Thus, all lines have the form ax0 + bx1 + cx2 = 0 for every
[a, b, c] ∈ RP 2 .
Two non-parallel ordinary lines intersect an ordinary point, while two parallel lines
intersect at an ideal point. This is seen as follows: For parallel ordinary lines, a+bx+cy =
0, a0 + bx + cy = 0, their versions in homogeneous coordinates are
( (
ax0 + bx1 + cx2 = 0 ax0 + bx1 + cx2 = 0
0
which is equivalent to
a x0 + bx1 + cx2 = 0 (a − a0 )x0 = 0

Proposition 14.2. Two distinct projective lines have exactly one point of intersec-
tion.
 GEOMETRIES AND TRANSFORMATION GROUPS 31

15. The Principle of Duality in RP 2

We continue to have another closer look at the real projective plane RP 2 , where we
simply “invented” ideal points/the ideal line, and make the artificial distinction between
the ordinary and ideal.
In order to free us from having to maintain the artificial distinction between ordinary
and ideal, we introduce the following algebraic version of the real projective space.

Definition 15.1. A projective point in RP n is a 1-dimensional subspace of Rn+1 . For

P [x0 , x1 , . . . , xn ] ∈ RP n , we also write P = hxi, the 1-dimesional subspace spanned by x
which is the column vector (x0 , x1 , . . . , xn )T .
A projective line in RP n is a 2-dimensional subspace of Rn+1 . If P = hpi, Q = hqi are
distinct projective points then p`(P, Q) = hp, qi, the subspace spanned by p, q.
A projective plane in RP n is a 3-dimensional subspace of Rn+1 .
A projective hyperplane in RP n is a n-dimensional subspace of Rn+1 .
A projective point P = hxi lies on a projective line h = hp, qi if the one dimensional
subspace hxi is a subspace of a two dimensional subspace hp, qi.
The Real Projective Plane RP 2 is the set of all projective points and lines, together
with the above incidence structure.

Proposition 15.2. In RP 2 , any two projective points lie on exactly one projective line,
and any two projective lines intersect in exactly one projective point.

Proof. For distinct projective points P = hpi, Q = hqi, p`(P, Q) = hp, qi is the projective
line passing through P and Q.
For distinct projective lines l, m, l∩m must be a one-dimensional subspace. Thus l∩m
is a projective point. (Recall dim(U + V ) = dim(U ) + dim(V ) − dim(U ∩ V ). For distict
projective lines l, m, dim(l ∩ m) = dim(l) + dim(m) − dim(l + m) = 2 + 2 − 3 = 1. Hence,
if l = hai⊥ , m = hbi⊥ , i.e. a, b are the corresponding normals, then l ∩ m = ha × bi.) 

Equations in homogeneous coordinates:

If l = hai⊥ , then the equation of l is a0 x0 + a1 x1 + a2 x2 = 0. We also write l =
[a0 , a1 , a2 ]⊥ . Thus, l is an affine (or ordinary) line if (a1 , a2 ) 6= (0, 0), l is the ideal line l∞
if (a1 , a2 ) = (0, 0),

Proposition 15.3. Three distinct projective points P = hpi, Q = hqi, and R = hri in
RP n , are collinear if and only if the vectors p, q, r are linearly dependent. Moreover, in
this case, the representatives p, q, r can be selected to satisfy p = q + r.
32 JIE DU

Similarly, four distinct projective points P = hpi, Q = hqi, R = hri and S = hsi in
RP n , no three of which are collinear, are coplanar if and only if the vectors p, q, r, s are
linearly dependent. Moreover, in this case, the representatives p, q, r, s can be selected to
satisfy p = q + r + s.

Proof. For hpi, hqi, hri to be collinear, they must lie on a common projective line. That
means they must lie in a two dimensional subspace (as projective lines are 2d subspaces).
Hence, they must be linearly dependent.
Conversely, if p, q, r are linearly dependent, they span a subspace of dimension 2. If
hp, q, ri has dimension 2, then it’s a projective line passing through the points P, Q, R.
In particular, ∃λ, µ, ν such that λp + µq + νr = 0. Since any two of p, q, r are linearly
independent, then all µ, λ, ν are non-zero. Thus P = hλpi, Q = h−µqi, R = h−νri.
Hence we may assume p = q + r. 

For any subspace W of Rn , let

W ⊥ = {x ∈ Rn | x  v = 0 ∀v ∈ W }.

Lemma 15.4. For subspaces U, V of Rn , we have

(1) (U + V )⊥ = U ⊥ ∩ V ⊥ , (2) (U ∩ V )⊥ = U ⊥ + V ⊥ .

Proof. For (1), U, V ≤ U + V implies (U + V )⊥ ⊆ U ⊥ ∩ V ⊥ .

Conversely, if x ∈ U ⊥ ∩ V ⊥ , then x · u = 0, x · v = 0 and so x · (u + v) = 0 for all
u ∈ U, v ∈ V . Hence, x ∈ (U + V )⊥ , proving (1).
(2) follows from (1) by the fact (U ⊥ )⊥ = U . 

Observe that the map V 7→ V ⊥ defines a bijection between

{projective points on RP 2 } and {projective lines on RP 2 }.

Principle of duality
In RP 2 , any true statement involving points and straight lines remains true if the words
“points” and “lines” are interchanged (i.e., [x0 , x1 , x2 ] ←→ [x0 , x1 , x2 ]⊥ ). E.g.,
• Any two projective points lie on exactly one projective line.
• Any two projective lines intersect in exactly one projective point.

Theorem 15.5. Three projective points in RP 2 are collinear (or “are on a line”) if and
only if three projective lines are concurrent (or “ are on a point”).

Proof. Points hpi, hqi, hri are collinear ⇐⇒ dimhp, q, ri = 2, or equivalently, hp, q, ri⊥
is a projective point. But hp, q, ri⊥ = (hpi + hqi + hri)⊥ = hpi⊥ ∩ hqi⊥ ∩ hri⊥ . Hence,
hpi⊥ , hqi⊥ , hri⊥ are concurrent. 
 GEOMETRIES AND TRANSFORMATION GROUPS 33

16. Desargues’ Theorem and Pappus Theorem

Theorem 16.1 (Desargues’ Theorem). Let A, B, C, A0 , B 0 , C 0 be distinct points in RP 2 ,

such that the projective lines p`(A, A0 ), p`(B, B 0 ), p`(C, C 0 ) are distinct and concurrent.
Then the projective points of intersections C 00 = p`(A, B) ∩ p`(A0 , B 0 ), A00 = p`(B, C) ∩
p`(B 0 , C 0 ), B 00 = p`(A, C) ∩ p`(A0 , C 0 ) are collinear.

A
C
B
A00

B0 C 00
B 00

C0
A0

Dual Desargues’ Theorem Let l, m, n, l0 , m0 , n0 be distinct lines in RP 2 such that their

intersections l ∩ l0 , m ∩ m0 , n ∩ n0 are distinct projective points, and collinear. Then the
projective lines joining l ∩ m, l0 ∩ m0 , and m ∩ n, m0 ∩ n0 , and n ∩ l, n0 ∩ l0 are concurrent.

Proof. Let P be the common point of intersection of projective lines p`(A, A0 ), p`(B, B 0 ),
p`(C, C 0 ). Let us first assume that P , the common point of intersection, is not one of
the points in the set {A, B, C, A0 , B 0 , C 0 }. Since P, A, A0 lie on a projective line and
are distinct, then, by Theorem 15.3, there are vectors p, a, a0 such that P = hpi, A =
hai, A0 = ha0 i, and p = a + a0 . Similarly, we can find representatives b, b0 for B, B 0 and
c, c0 for C, C 0 such that p = b + b0 , p = c + c0 . It follows that a + a0 = b + b0 , and so
c00 = a − b = a0 − b0 . Similarly, a00 = b − c = b0 − c0 , b00 = c − a = c0 − a0 .
But then we have c00 + a00 + b00 = a − b + b − c + c − a = 0. So we see that the repre-
sentative vectors c00 , a00 , b00 are linearly dependent, hence dimha00 , b00 , c00 i = 2. The points
A00 , B 00 , C 00 in RP 2 , which are/can be represented by vectors a00 , b00 , c00 , are collinear.
On the other hand, c00 = a − b ∈ ha, bi, so C 00 ∈ p`(A, B), and also c00 = a0 − b0 ⇒
C 00 ∈ p`(A0 , B 0 ). So C 00 = p`(A, B) ∩ p`(A0 , B 0 ). Similarly, A00 = p`(B, C) ∩ p`(B 0 , C 0 ),
and B 00 = p`(A, C) ∩ p`(A0 , C 0 ).
Now, suppose instead that P ∈ {A, B, C, A0 , B 0 , C 0 }.
Suppose (wlog) P = A. Then a = b + b0 = c + c0 (by same arguments as earlier that
p = b + b0 , using Theorem 15.3). Thus, ha, bi = ha, b0 i (since both equal hb, b0 i), and
also ha, ci = ha, c0 i. Hence, substituting using ha, bi = ha, b0 i, we have ha, bi ∩ ha0 , b0 i =
ha, b0 i ∩ ha0 , b0 i = hb0 i (we see b0 is common to both proj lines there, so it’s the unique
proj point of intersection). Similarly, ha, ci ∩ ha0 , c0 i = hc0 i.
34 JIE DU

Now, again using p = a = b0 + b = c + c0 , we find that b − c = c0 − b0 . Thus it

follows that (by similar arguments used in the proof when P 6= A), hb, ci ∩ hb0 , c0 i =
hb0 − c0 i = hb − ci.
Now we see that p`(A, B)∩p`(A0 , B 0 ) = hb0 i, p`(A, C)∩p`(A0 , C 0 ) = hc0 i, and p`(B, C)∩
p`(B 0 , C 0 ) = hb0 − c0 i. And so the representative vectors of the intersections are linearly
dependent. And so the the projective points C 00 , A00 , B 00 themselves are collinear. 

Theorem 16.2 (Pappus’ Theorem). Let A, B, C and A0 , B 0 , C 0 be two pairs of collinear

triples of distinct points in a projective plane. Then the three points A00 = p`(B, C 0 ) ∩
p`(B 0 , C), B 00 = p`(C, A0 ) ∩ p`(C 0 , A) and C 00 = p`(A, B 0 ) ∩ p`(A0 , B) are collinear.
C
B
A

A0 B0 C0

Dual Pappus’ Theorem Let l, m, n, l0 , m0 n0 be two pairs of concurrent projective lines

in RP 2 . Then the projective lines p`(m ∩ n0 , m0 ∩ n), p`(n0 ∩ l, n ∩ l0 ), p`(l ∩ m0 , l0 ∩ m) are
concurrent.
Note that a classical proof needs Menelaus’s Theorem. With Theorem 18.1, an alge-
braic proof is left as an exercise.

Proof. By Theorem 18.1, the quadrilateral ACC’A’ is projectively equivalent to the square
with vertices [1, 0, 0], [1, 1, 0], [1, 1, 1], [1, 0, 1]. Thus, we may assume A = (0, 0), C =
(1, 0), C 0 = (1, 1), A0 = (0, 1). Thus, B = (u, 0) and B 0 = (v, 1) where 0 < u, v < 1. Thus,
1/2
• The intersection `(A, C 0 ) ∩ `(A0 , C) is at the centre 1/2


of the square.
• For the intersection `(A, B 0 ) ∩ `(A0 , B), solving 01 + λ −1 u
= µ v1 gives u+v
1 u





v
.
1 v−1 1 u−1 1−u
• For `(C, B 0 ) ∩ `(C 0 , B), solving 0 + λ 1 = 1 + µ −1 gives 2−u−v 1






1−v
.
Now      
1 u 1/2 1 u−v
− =
u+v v 1/2 2(u + v) v − u
     
1 1−u 1/2 1 u−v
− = .
2−u−v 1−v 1/2 2(2 − u − v) v − u
Hence, the three intersections are collinear. 
 GEOMETRIES AND TRANSFORMATION GROUPS 35

17. Projective Transformations in RP n

Definition 17.1. A map π : RP n → RP n is called a projective transformation if there

exists an invertible matrix A ∈ GLn+1 (R) such that πhxi = hAxi.
Recall that TA : Rn+1 → Rn+1 , x 7→ Ax is a linear isomorphism.
Since A(λx) = λAx ∀λ ∈ R, it induces a projective transformation, denoted by πA .

Proposition 17.2. For A, A0 ∈ GLn+1 (Rn ), πA = πA0 ⇐⇒ A = λA0 for some λ ∈

R − {0}.

Proof. The “if” part is seen above. Conversely, suppose π = πA = πA0 . Take a ba-
sis {v0 , v1 , . . . , vn } for Rn+1 , then, since hAvi i = πhvi i = hA0 vi i, by properties of
span, we have A0 vi = λi Avi , for some λi ∈ R − {0}. Also, A0 (v0 + v1 + · · · + vn ) =
λA(v0 + v1 + · · · + vn ), for some λ ∈ R − {0}. Thus, we see that
X n X n  n
X  X n
0
λ(Avi ) = λA vi = A vi = λi Avi .
i=0 i=0 i=0 i=0
Now, since A is invertible, Avi are linearly independent. So this means that it must be
that λ = λi ∀i (as there must be a unique way to write any vector as a linear combination
of the linear independent vectors Avi ). Hence, Avi = λAvi ∀i, and so A0 = λA. 

For A ∈ GLn+1 (R), let [A] = {λA : λ ∈ R − 0} (equivalence class).

Also, let P GLn+1 (R) = {[A] : A ∈ GLn+1 (R)} (set of equivalence classes).
Introduce a multiplication on P GLn+1 (R) by setting [A][B] = [AB]

Theorem 17.3. (1) The set P GLn+1 (R) with the above multiplication forms a group,
the projective linear group.
(2) Let P = P(RP n ) be the set of all projective transformations on RP n . Then P
is a group.
(3) The map φ : P GLn+1 (R) → P, [A] 7→ πA is a bijection, satisfying: φ([A][B]) =
φ([A])φ([B]). That is, the map φ is a group isomorphism.

Proof. (1) The multiplication, [A][B] = [AB], is well-defined. The associative law
for this multiplication follows from associativity for matrix multiplication. The identity
element of the group is [In+1 ], and inverses are given by [A]−1 = [A−1 ]. Hence P GLn+1 (R)
is a group.
(2) This is also clear. Let’s do a little bit of checking. πA πB hxi = πA hBxi = hABxi =
πAB hxi. We see πA πB = πAB . Multiplication is well-defined. Associativity is also inher-
ited/follows from associativity of matrix multiplication.
Identity: id = πIn+1 .
36 JIE DU

Inverses: (πA )−1 = πA−1 .

(3) By Proposition 17.2, the map φ is well-defined (there is a unique map πA in P for
any input [A]).
Clearly, φ is onto.
Suppose that φ([A]) = φ([B]). Then πA = πB . But by Proposition 17.2, this means
that [A] = [B]. Thus φ is one-to-one. Finally, φ([A][B]) = φ([AB]) = πAB = πA πB =
φ([A])φ([B]). 

Theorem 17.4. Every affine transformation can be uniquely extended to a projective

transformation which stabilises the ideal part and the ordinary part and preserves multi-
plication and inverses. In group theory terminology, A is (isomorphic to) a subgroup of
P. That is, A ∼ = A 0 ≤ P.

Proof. Define ψ : A → P, τ 7→ φ(τ ), where, if τ (x) = Ax + b, i.e., τ = TA,b , then

φ(τ ) = π(1 0) .
bA

We first check that φ is one-to-one: suppose φ(TA,b ) = φ(TB,c ). Then

   
1 0 1 0
=λ , for some λ ∈ R − 0.
b A c B
Equating matrix entries gives 1 = λ, b = c, and A = B. Hence, TA,b = TB,c , proving the
injectivity.
−1
Since TB,c TA,b = TBA,Bb+c , and TA,b = TA−1 ,−A−1 b , and
      −1  
1 0 1 0 1 0 1 0 1 0
= , = .
c B b A Bb + c BA b A −A−1 b A−1
it follows that φ(τ τ 0 ) = φ(τ )φ(τ 0 ) and φ(τ −1 ) = φ(τ )−1 . Hence, φ : A → φ(A ) is a
group isomorphism. That is, A ∼ = φ(A ) ≤ P.
Note that for every ideal (resp., ordinary) point hxi with homogeneous coordinates
[0, x1 , . . . , xn ], (resp., [x0 , x1 , . . . , xn ], x0 6= 0) π(1 0) hxi = h b1 A0 xi is clearly an ideal


bA

(resp., ordinary) point. 

 GEOMETRIES AND TRANSFORMATION GROUPS 37

18. Projective Plane Transformations

Theorem 18.1. Let P, Q, R, S and P 0 , Q0 , R0 , S 0 be two sets of four points, no three of

which are collinear in RP 2 . Then there is a unique π ∈ P such that π(P ) = P 0 , π(Q) =
Q0 , π(R) = R0 , π(S) = S 0 .

Proof. Suppose P = hpi, Q = hqi, R = hri, S = hsi. By the hypothesis, p, q, r form a

basis for R3 and, by Theorem 15.3, we may assume that s = p + q + r. We do the same
for the projective points P 0 , Q0 , R0 , S 0 , and assume s0 = p0 + q0 + r0 .
Now, since both {p, q, r} and {p0 , q0 , r0 } are bases for R3 , there exists a unique linear
transformation or matrix A (i.e. a change of basis matrix!) such that:

Ap = p0 , Aq = q0 , Ar = r0 .

(In fact, this matrix is: A = (p0 | q0 | r0 )(p | q | r)−1 .)

Let π = πA . Then π(P ) = P 0 , π(Q) = Q0 , π(R) = R0 , and

π(S) = πhsi = hAsi = hA(p + q + r)i = hp0 + q0 + r0 i = hs0 i = S 0 .

So we have shown the existence. Now we want to show the uniqueness.

Suppose π 0 = πA0 is another projective transformation, satisfying π 0 (P ) = P 0 , π 0 (Q) =
Q0 , π 0 (R) = R0 , π 0 (S) = S 0 . Then A0 p = λp0 , A0 q = µq0 , A0 r = νr0 , and A0 s = ds0 , for
some λ, µ, ν, d ∈ R − {0}. But

ds0 = A0 s = (A0 p + A0 q + A0 r) = λp0 + µq0 + νr0 .

Thus rearranging, s0 = λd p0 + µd q0 + νd r0 . On the other hand, s0 = p0 + q0 + r0 . So it must

be that λ
d
= 1, µd = 1, νd = 1 or λ = µ = ν = d. Hence, A0 = λA, and thus π = π 0 , by
Proposition 17.2. 

Definition 18.2. A bijective map τ : RP 2 → RP 2 is called a (projective) collineation

if τ takes collinear points to collinear points. (Equivalently, τ sends every line to a line.)

Note that the inverse of a (projective) collineation is a (projective) collineation.

Now, define

P1 = [1, 0, 0], P2 = [0, 1, 0], P3 = [0, 0, 1], Q = [1, 1, 1].

Lemma 18.3. If τ is a collineation of RP 2 and τ fixes points P1 , P2 , P3 , Q, then τ is the

identity map.

Proof (R. Hartshorne, D. Harvey). Since τ fixes P2 , P3 , it must map `∞ to `∞ . Also, τ −1

must map `∞ to `∞ . Hence, τ must map RP 2 \`∞ (which is just R2 ) to itself.
38 JIE DU

Let τ 0 be the restriction of τ to R2 . Then τ 0 is a collineation on R2 . (This is because

given any three collinear points in R2 , they are also collinear in RP 2 , and so τ sends them
to collinear points in RP 2 , which are then also collinear in R2 .) Thus, τ 0 = TA,b for some
A ∈ GL2 (R), b ∈ R2 . Note that τ 0 fixes (0, 0) (because τ fixes P1 ). So b = 0. Let us
write A = ac db .

We know τ 0 fixes (1, 1) (because τ fixes Q), so a + b = 1 and c + d = 1. Also, since τ

fixes P2 , the map tau must send the x-axis to itself. So ac = A 10 = λ0 . This yields

c = 0. A similar argument involving P3 and the y-axis shows b = 0. Thus, a = d = 1 and

τ 0 is the identity map on R2 .
Finally, for P ∈ `∞ , choose any two points in R2 defining a line with slope corresponding
to P ; then τ fixes those two points, so must fix P . 

Theorem 18.4. A bijective map τ on RP 2 is a projective collineation if and only if τ is

a projective transformation.

Proof. The “if” part is clear from the definition of projective transformation, since if a line
is given by `T x = 0, where `T = (a, b, c), then, for τ = πA , ((A−1 )T `)T (Ax) = `T x = 0.


(Or if a line in RP 2 is given by p`(P, Q) = hp, qi, then π p`(P, Q) = span{Ax : x ∈
span{p, q}} = hAp, Aqi.)
Conversely, suppose τ is a collineation. Let τ (P1 ) = P 0 , τ (P2 ) = Q0 , τ (P3 ) = R0 , τ (Q) =
S 0 . Then no three of the P 0 , Q0 , R0 , S 0 are collinear. By Theorem 18.1, there exists a
unique projective transformation π ∈ P such that π(P1 ) = P 0 , etc. Thus τ −1 π is a
collineation which leaves P1 , P2 , P3 , Q fixed. (The reason τ −1 π is a collineation is because
π is a collineation by the “if” proof, and τ −1 is a collineation, and the composition of two
collineations is a collineation.) By Lemma 18.3, τ −1 π = 1. Hence, τ = π. 

Example 18.5. Let π = πA : RP 2 → RP 2 be defined by the matrix

 
a b c
A = d1 a11 a12  .
d2 a21 a22
Then the restriction of π induces a map from `a,b,c : ax0 + bx1 + cx2 = 0 to the ideal line
`∞ : x0 = 0. Check: [x0 , x1 , x2 ] ∈ `a,b,c ⇐⇒ π[x0 , x1 , x2 ] ∈ `∞ .
Thus, π induces a bijection from RP 2 − `a,b,c to RP −
2
`∞ .  
b c
Note that π(`∞ ) is the line spanned by vectors a11  and a12 . So π induces a
a21 a22
bijection
π :RP 2 − (`∞ ∪ `a,b,c ) −→ RP 2 − (π(`∞ ) ∪ `∞ )
 
d1 + a11 x + a12 y d2 + a21 x + a22 y
[1, x, y] 7−→ 1, , .
a + bx + cy a + bx + cy
 GEOMETRIES AND TRANSFORMATION GROUPS 39

The Hasse diagram of the groups:

S Q

D I

T
Proof of Lemma 18.3
Since τ fixes P2 = [0, 1, 0] and P3 = [0, 0, 1], we see that τ fixes the ideal line, x0 = 0.
Thus τ induces a map τ 0 : RP 2 − l∞ → RP 2 − l∞ . (This is the map τ but restricted to
the domain RP 2 − l∞ )
Identify RP 2 − l∞ with R2 via [x0 , x1 , x2 ] → [ xx10 , xx20 ]
Then τ 0 : R2 → R2 is a collineation on R2 . Since τ fixes l∞ , τ 0 sends horizontal lines
to horizontal lines**.
τ fixes P1 , i.e. τ 0 fixes (0, 0), τ fixes x-axis.
For any point (a, 0) on the x-axis, ∃ aσ ∈ R such that τ 0 (a, 0) = (aσ , 0) **.
It can be proved that the map a 7→ aσ from R to R is a field automorphism. But
the only field automorphism of R is the identity.
Hence τ 0 (a, 0) = (a, 0) ∀a ∈ R. Hence τ 0 must be the identity map.
It remains to prove that τ fixes every ideal point also. Write the point [0, x1 , x2 ], a
point on the ideal line, as the intersection of the line x0 = 0 and the line joining the
points [1, 0, 0] and [1, x1 , x2 ]. Since points of the form [1, x1 , x2 ] are ordinary points, they
are fixed by τ , so the lines are fixed by τ . Hence, their intersection is fixed by τ .

Proof of Pappus Theorem. There exists a projective transformation sending the quadri-
lateral ACC 0 A0 to a square in homogeneous coordinates

P [1, −1, 1], Q[1, 1, 1], R[1, −1, 1], S[1, −1, −1].

Since a projective transformation preserves collinearity, it suffices to prove the case where
A = P, C = Q, C 0 = R, A0 = S is a square.
40 JIE DU

Consider their affine coordinates (−1, 1), (1, 1), (−1, 1), (−1, −1) for A, C, C 0 , A0 , re-
spectively. Thus, B 00 is the origin and B, B 0 have the form B = (b, 1) and B 0 = (b0 , −1).
Then        0 
00 −1 b+1 −1 b +1
c = +λ = +µ .
−1 2 1 −2
b0 +1
Solving gives c00 = −1 b+1



1
+ 0
b+b +2 −2
. Similarly,
       0 
00 1 b−1 1 b −1
a = +λ = +µ .
−1 2 1 −2
b0 −1
Solving gives a00 = 11 + b+b b−1
. Finally, it is direct to check that (b + b0 + 2)c00 =



0 −2 −2
(b + b0 − 2)a00 . Hence, A00 , B 00 , C 00 are collinear. 

References
[1] Curtis, C.W., Reiner, I., Representation Theory of finite groups and associative algebras, In-
tersceince Publishers, New York, (1962)

School of Mathematics and Statistics, University of New South Wales, Sydney 2052,
Australia.
Email address: [Link]@[Link]