Affine transformations (Affinities)

A file of the Geometrikon gallery by Paris Pamfilos

I am among those who think that science has great

beauty . . . A scientist in his laboratory is not only a
technician but also a child placed in front of natural
phenomena which impresses him like a fairy tale.

E. Curie, Madame Curie p.341

Contents (Last update: 27-08-2024)

1 Definition and group property 2

2 Determination of an affinity 4

3 Additional fundamental properties of affinities 5

4 The associated ellipse of an affinity 7

5 The conic of intersections of an affinity 9

6 The conic of intersections of an affinity 13

7 Special kinds of affinities 15

8 Axial affinities or homologies, shears, strains 17

9 Affinities leaving invariant a triangle 19

10 The period of an affinity 22

11 Translations 24
12 Dilatations 25
13 Shears 26
14 Strains 27
15 Invariant pencils 28

16 Affinities defined by three elements 30

17 The case of parabolic conics of intersections 33

18 Equiaffinities as products of reflections 34

19 Conjugacy for affinities with fixed points 36

20 Elliptic affinities 37
 1 Definition and group property 2

21 Orbits of points under equiaffinities 39

22 Classification of equiaffinities 40

23 Analytic description of orbits of equiaffinities 43

24 Remarks and exercises on orbits of equiaffinities 44

25 Affine equivalence of conics 47

26 Affinities preserving a non-degenerate conic 48

1 Definition and group property

Affinity Considering the case of two dimensions, “Affine transformations” or “Affinities” are invert-
ible transformations of the plane onto itself, which, fixing a coordinate system (not neces-
sarily orthogonal or having equal unit-lengths on the axes), are defined by an invertible
matrix {𝐴, | 𝐴| = 𝑎 11 𝑎 22 − 𝑎 12 𝑎 21 ≠ 0} and a vector 𝑣(𝑣 1 , 𝑣 2 ) ([Cox61, p.203]):
 
𝑎 11 𝑎12
𝑌 = 𝑓 (𝑋) = 𝐴𝑋 + 𝑣, with 𝐴 =
𝑎 21 𝑎22

𝑦1 = 𝑎 11 𝑥 1 + 𝑎 12 𝑥 2 + 𝑣 1 ,
(1)
𝑦2 = 𝑎 21 𝑥 1 + 𝑎 22 𝑥 2 + 𝑣 2 .
Using the three dimensional extensions 𝑋 ′ = (𝑥 1 , 𝑥 2 , 1) of points 𝑋(𝑥 1 , 𝑥 2 ) ∈ R2 and the
matrix
𝑎11 𝑎 12 𝑣 1  
© ª 𝐴 𝑣
𝐴𝑣 = ­ 𝑎21 𝑎 22 𝑣 2 ® = , (2)
0 1
« 0 0 1 ¬
equation (1) is equivalent with

𝑦1 𝑥1
′ ′ © ª © ª
𝑌 = 𝐴𝑣 · 𝑋 ⇔ ­ 𝑦2 ® = 𝐴 𝑣 ­ 𝑥 2 ® . (3)
«1¬ «1¬
The determinants of 𝐴𝑣 and 𝐴, are equal and the inverse of 𝐴𝑣 is of the same form:

𝑎 −𝑎 12 𝑎 12 𝑣 2 − 𝑎 22 𝑣1  −1 
1 © 22 ª 𝐴 −𝐴−1 · 𝑣
𝐴−1
𝑣 = · ­−𝑎 21 𝑎 11 𝑎 21 𝑣 1 − 𝑎 11 𝑣2 ® = ,
| 𝐴| 0 1
« 0 0 | 𝐴| ¬
which guarantees that this represents again an affine transformation. Also the product
(composition) of two such transformations, represented by the matrices {𝐴𝑣 , 𝐵𝑤 }, is of
the same form:      
𝐴 𝑣 𝐵 𝑤 𝐴𝐵 𝐴𝑤 + 𝑣
· = .
0 1 0 1 0 1
These remarks imply easily the important property formulated as a theorem:
 1 Definition and group property 3

Group Theorem 1. The affinities of the plane form a group 𝐺 𝐴 .

property
We say two shapes of the plane {𝐴, 𝐵} are “affine equivalent”, if there is an affinity
affine mapping the first onto the second. The affine geometry of the plane deals with properties
equivalent of shapes that remain the same (invariant) by affinities. For example, the middle of a
segment. We’ll see below that by an affinity a segment 𝜇 maps to a segment 𝜇′ and its
middle 𝑀 ∈ 𝜇 maps correspondingly to the middle 𝑀 ′ ∈ 𝜇′. We say in short: affinities
“preserve” the middles of segments.
Remark 1. There are successive extensions {𝐺𝐸 ⊂ 𝐺 𝐴 ⊂ 𝐺𝑃 } of the group of transfor-
mations defining the “equality” or “equivalence” of two shapes of the plane, the symbols
standing respectively for the euclidean, the affine and the projective group of transforma-
tions. These groups can be represented by matrices in the form

 

 ©cos(𝜙) ∓ sin(𝜙) 𝑎 ª
 

Euclidean isometries: 𝐺𝐸 = ­ sin(𝜙) ± cos(𝜙) 𝑏 ® , for 𝜙, 𝑎, 𝑏 ∈ R ,

 

« 0 0 1¬ 

 

©𝑎 𝑏 𝑐 ª
 

Affinities: 𝐺 𝐴 = ­ 𝑘 𝑙 𝑚 ® , for 𝑎, 𝑏, 𝑐, 𝑘, 𝑙, 𝑚 ∈ R ,

 0 0 1 

« ¬ 

 𝑎 𝑏 𝑐 

©
 ª 

Projectivities: 𝐺𝑃 = ­ 𝑘 𝑙 𝑚 ® , for 𝑎, 𝑏, 𝑐, 𝑘, 𝑙, 𝑚, 𝑝, 𝑞, 𝑟 ∈ R .

 𝑝 𝑞 𝑟 

« ¬ 
In the two last groups we require also from the matrices to be invertible i.e. to have non-
zero determinants. In the last group also we consider that two matrices differing by a
non-zero multiplicative constant 𝐵 = 𝜆 · 𝐴, 𝜆 ≠ 0, define the same (projective) transfor-
mation.
Analogous to the “affine equivalence” is the “congruence” of two shapes, meaning that
there is an euclidean isometry mapping one to the other. Finally, “projective equivalent”
are called two shapes for which there is a projectivity mapping one to the other.
The fewer parameters has the group, the more entities, the group acts upon, can be dis-
tinguished as non-equivalent. Thus, the euclidean group, having only three parameters,
distinguishes several kinds of hyperbolas, of ellipses and parabolas. The affine group,
with six parameters, cannot distinguish that many entities. For it all hyperbolas are the
same, i.e two hyperbolas, which from the euclidean aspect are different, can be trans-
formed, one to the other, via an affine transformation 𝑓 ∈ 𝐺 𝐴 and thus, can be considered
to be the same thing. Ultimately, under this aspect, there is only one hyperbola, one el-
lipse and one parabola. Finally, under the projective aspect, the previous three different
kinds are unified and, ultimately, there is only one conic, the circle. In the following sec-
tions we confine our study to the elements of the group 𝐺 𝐴 of affine transformations and
properties of shapes preserved by affinities.
projective Remark 2. From the coordinates (𝑥, 𝑦) we pass to (𝑥, 𝑦, 1) and from there to the cor-
coordinates responding “projective coordinates” (𝑥, 𝑦, 𝑧), in which {(𝑥, 𝑦, 𝑧), (𝑘𝑥, 𝑘 𝑦, 𝑘𝑧), 𝑘 ≠ 0} repre-
sent the same point (𝑥/𝑧, 𝑦/𝑧, 1).
The lines 𝑎𝑥 + 𝑏 𝑦 + 𝑐 = 0 are described in projective coordinates by 𝑎𝑥 + 𝑏𝑦 + 𝑐𝑧 = 0
line at and 𝑧 = 0 represents the “line at infinity” containing the far out points {(𝑥, 𝑦, 0)} of the
infinity plane. From equation (1) we see that affinities map a point (𝑥, 𝑦, 0) to a point of the same
kind (𝑥 ′ , 𝑦 ′ , 0) i.e. affinities preserve the line at infinity, consequently, ordinary points of the
plane, characterized by coordinates of the form (𝑥, 𝑦, 1), map by affinities to points of the
same kind (𝑥 ′ , 𝑦 ′ , 1).
 2 Determination of an affinity 4

This has an important consequence concerning the image of a circle by an affinity.

circles map This is a conic, since quadratic equations transform to quadratic equations by affinities.
to ellipses The conic though cannot have points at infinity, since if it had, the affinity would send
an ordinary point of the circle to a point at infinity and the inverse affinity would send
that point at infinity to an ordinary point of the plane, which cannot be done. Thus, the
images of circles via affinities are ellipses. This is further exploited in section 4.

2 Determination of an affinity
Given three points of the plane {𝑋 , 𝑌, 𝑍}, the determinant of the corresponding matrix

𝑥1 𝑦1 𝑧 1
© ª
𝑋𝑌𝑍 = ­ 𝑥 2 𝑦2 𝑧 2 ® , | 𝑋𝑌𝑍| := (𝑦1 𝑧 2 − 𝑦2 𝑧1 ) + (𝑧 1 𝑥 2 − 𝑧 2 𝑥 1 ) + (𝑥 1 𝑦2 − 𝑥 2 𝑦1 ),
«1 1 1¬

if the coordinate frame is “orthonormal”, expresses twice the signed area of the triangle
𝑋𝑌𝑍 ([Str88, p.239]). Otherwise, in an oblique coordinate frame, it is a constant multiple of
this area, the constant factor depending on the frame. This implies that the three points are
collinear precisely when this determinant vanishes. Also applying the affinity to {𝑋 , 𝑌, 𝑍}
we obtain three other points {𝑈 , 𝑉 , 𝑊} and we can describe this operation with a matrix
multiplication:

𝑈𝑉𝑉 = 𝐴𝑣 · 𝑋𝑌𝑍 ⇒ |𝑈𝑉𝑊| = | 𝐴𝑣 | · | 𝑋𝑌𝑍| = | 𝐴| · | 𝑋𝑌𝑍|.

This implies that if {𝑋 , 𝑌, 𝑍} are non-collinear(collinear) the same is true for {𝑈 , 𝑉 , 𝑊}. In
addition, it follows that the quotient of the areas of two triangles is preserved by affinities.
By splitting a polygon in triangles, we conclude that affinities preserve the quotient of
areas of two polygons. There is even a kind of affinities preserving the area. This, by
the last formula, happens when the determinant of the affinity is | 𝐴| = | 𝐴𝑣 | = 1. These
equiaffinity special affinities are called “equiaffinities”. The following theorems formulate the previous
remarks.

preserve Theorem 2. An affine transformation maps collinear points to collinear and non-collinear to non-
collinearity collinear points as well.

lines map to Corollary 1. An affine transformation maps a line of the plane onto a line of the same.
lines
Corollary 2. An affine transformation maps a genuine triangle to a genuine (non-degenerated)
triangle.

Theorem 3. For an affine transformation the image 𝑓 (𝑝) of a polygon 𝑝 is a polygon with the
same number of sides and the ratio of the signed areas | 𝑓 (𝑝)|/| 𝑝| is a constant independent of the
particular polygon.

quotient Corollary 3. An affine transformation preserves the quotient of areas of two polygons {𝑝, 𝑝 ′}, i.e.
of areas | 𝑓 (𝑝 ′)|/| 𝑓 (𝑝)| = | 𝑝 ′|/| 𝑝|.
preserved
The inverse of the matrix 𝑋𝑌𝑍 , for non-collinear points, is found to be

𝑦 − 𝑧2 𝑧 1 − 𝑦1 𝑦1 𝑧 2 − 𝑦2 𝑧 1
1 © 2 ª
(𝑋𝑌𝑍)−1 = ­ 𝑧2 − 𝑥2 𝑥1 − 𝑧1 𝑥2 𝑧1 − 𝑥1 𝑧2 ® .
| 𝑋𝑌𝑍|
« 𝑥 2 − 𝑦2 𝑦1 − 𝑥 1 𝑥 1 𝑦2 − 𝑥 2 𝑦1 ¬
 3 Additional fundamental properties of affinities 5

Thus, given two triples of non-collinear points {(𝑋 , 𝑌, 𝑍), (𝑈 , 𝑉 , 𝑊)} the matrix equation

𝐵 · 𝑋𝑌𝑍 = 𝑈𝑉𝑊 ⇔ 𝐵 = 𝑈𝑉𝑊 · (𝑋𝑌𝑍)−1 ,

has a unique solution and defines the matrix 𝐵, which, a short calculation shows to be of
the form of equation (2), thus defining an affinity. We formulate this as a theorem.

affinity from Theorem 4. Two triples of non-collinear points {(𝑋 , 𝑌, 𝑍), (𝑈 , 𝑉 , 𝑊)} define a unique affinity
3 to 3 pts 𝑓 mapping the first to the second: { 𝑓 (𝑋) = 𝑈 , 𝑓 (𝑌) = 𝑉 , 𝑓 (𝑍) = 𝑊}.

This implies several other properties expressed by the next corollaries.

Corollary 4. Given two triangles 𝐴𝐵𝐶 and 𝐴′𝐵′𝐶 ′ there is precisely one affinity 𝑓 mapping
the first onto the second, in the sense { 𝑓 (𝐴) = 𝐴′ , 𝑓 (𝐵) = 𝐵′ , 𝑓 (𝐶) = 𝐶 ′}.

identity: Corollary 5. An affinity mapping the vertices of a triangle 𝐴𝐵𝐶 onto the same vertices is the
fix 3 pts identity transformation.

Corollary 6. Two affinities coinciding at three non collinear points (or the vertices of a triangle)
coincide at every other point of the plane and define the same transformation.

Corollary 7. Two lines of the plane are “affine equivalent”.

equilateral’s Corollary 8. Every triangle of the plane is “affine equivalent” to the equilateral.
universality
In section 25 we will show that two non-degenerate conics of the same type are affine
equivalent.

3 Additional fundamental properties of affinities

parallels to Corollary 9. Two different parallel lines map by an affinity to two different parallel lines.
parallels
Proof. In fact, assume that the parallels {𝛼, 𝛽} have images {𝛼′ = 𝑓 (𝛼), 𝛽′ = 𝑓 (𝛽)} inter-
secting at the point 𝐶. Then, 𝑓 −1 (𝐶) is a point on 𝛼 ∩ 𝛽, which is a contradiction. □
Corollary 10. A parallelogram maps by an affinity to a parallelogram.

parallelogram Corollary 11. Two arbitrary parallelograms of the plane are affine equivalent, equivalently, every
property parallelogram is affine equivalent to a square.
preservation
Proof. Consider the affinity 𝑓 mapping the triangle 𝐴𝐵𝐶 onto 𝐴′𝐵′𝐶 ′. Then see that
𝑓 (𝐷) = 𝐷 ′ (See Figure 1). □
D A' D'
A

B C
B' C'

Figure 1: Parallelograms are affine equivalent

ratios Corollary 12. If the affinity 𝑓 maps the line 𝛼 onto the line 𝛽, then the ratio of three points on
preservation 𝛼 : 𝐶𝐴/𝐶𝐵 is the same with the ratio of their images on 𝛽 : 𝑓 (𝐶) 𝑓 (𝐴)/ 𝑓 (𝐶) 𝑓 (𝐵) = 𝐶𝐴/𝐶𝐵.
 3 Additional fundamental properties of affinities 6

Proof. In fact, the point 𝐶 ∈ 𝛼 can be expressed as a linear combination

𝐶 = (1 − 𝜆)𝐴 + 𝜆𝐵 ⇔ 𝐶 ′ = (1 − 𝜆)𝐴′ + 𝜆𝐵′ ,

the primes denoting the extended 3-dimensional points, and applying the affinity 𝑓 on
𝐶 means multiplying 𝐶 ′ by a matrix 𝐴𝑣 leading to

𝐴𝑣 · 𝐶 ′ = (1 − 𝜆)𝐴𝑣 · 𝐴′ + 𝜆𝐴𝑣 · 𝐵′ ⇔ 𝑓 (𝐶) = (1 − 𝜆) 𝑓 (𝐴) + 𝜆 𝑓 (𝐵).

The proof follows from this and the general relation of 𝐶𝐴/𝐶𝐵 with 𝜆 :

𝐶𝐴 𝜆
𝐶 = (1 − 𝜆)𝐴 + 𝜆𝐵 ⇔ = . □
𝐶𝐵 𝜆−1
middles Corollary 13. If the affinity 𝑓 maps the line-segment 𝐴𝐵 onto the line segment 𝐴′𝐵′ , then it
preservation maps also the middle 𝑀 of 𝐴𝐵 onto the middle 𝑀 ′ of 𝐴′𝐵′. If the affinity interchanges the end-
points of a segment, i.e. { 𝑓 (𝐴) = 𝐵, 𝑓 (𝐵) = 𝐴}, then it fixes the middle 𝑀 of 𝐴𝐵 : 𝑓 (𝑀) = 𝑀.

Given 𝑛 points of the plane {𝐴 𝑖 , 𝑖 = 1...𝑛} and 𝑛 numbers {𝑡 𝑖 , 𝑖 = 1...𝑛,} the weighted
sum
1
𝐴0 = Í (𝑡1 𝐴1 + · · · + 𝑡 𝑛 𝐴𝑛 ), (4)
𝑡𝑖
is called “barycenter” or “centroid” of the system of masses {𝑡 𝑖 } attached to the points
{𝐴 𝑖 }. Using the vectors {𝐴′𝑖 = (𝑥 𝑖 , 𝑦 𝑖 , 1)} instead of {𝐴 𝑖 (𝑥 𝑖 , 𝑦 𝑖 )} and applying the method
of corollary 12, we prove easily next property formulated as a theorem.

Theorem 5. Affinities 𝑓 preserve the barycenter of systems of weighted points of the plane, i.e.
1 1
𝐴0 = Í (𝑡1 𝐴1 + · · · + 𝑡 𝑛 𝐴𝑛 ) ⇒ 𝑓 (𝐴0 ) = Í (𝑡1 𝑓 (𝐴1 ) + · · · + 𝑡 𝑛 𝑓 (𝐴𝑛 )).
𝑡𝑖 𝑡𝑖
centroid or Remark 3. Setting {𝑡1 = 𝑡2 = ... = 𝑡 𝑛 = 1}, point 𝐴0 = (𝐴1 + . . . 𝐴𝑛 )/𝑛 is the “barycenter”
barycenter or “centroid” of the polyon 𝐴1 𝐴2 ...𝐴𝑛 and the theorem means that

“Affinities preserve the barycenter of polygons”.

Remark 4. Barycenters can be used to give an alternative definition of affinities. Accord-

ing to this: an affinity of the plane is a transformation which preserves barycenters.

convexity Exercise 1. Show that an affinity preserves the “convexity” property of a polygon, which means
preservation that if {𝑋 , 𝑌} are points inside the polygon, then the whole segment 𝑋𝑌 is also inside.

Exercise 2. Given 𝑛 points of the plane {𝐴 𝑖 , 𝑖 = 1...𝑛} consider the function, which for every
Í
point of the plane 𝑃 defines 𝑚(𝑃) = 𝑖 |𝑃𝐴 𝑖 | 2 . Show that 𝑚(𝑃) attains its minimum value at
the centroid 𝐴0 of the polygon 𝐴1 ...𝐴𝑛 ([Rya86, p.54]).

G E

B C
D

Figure 2: Fixing the centroid 𝐺 of the triangle

centroid Corollary 14. An affinity of the plane mapping the triangle 𝐴𝐵𝐶 onto itself leaves the centroid
preservation 𝐺 of the triangle fixed and maps each median line to another median line of the triangle.
 4 The associated ellipse of an affinity 7

Proof. It is obvious that an affinity leaving invariant a triangle permutes the vertices of
the same. Assume now that the affinity 𝑓 defines such a permutation recycling the ver-
tices: { 𝑓 (𝐴) = 𝐵, 𝑓 (𝐵) = 𝐶, 𝑓 (𝐶) = 𝐴}. Then, the middle 𝐷 of 𝐵𝐶 maps to the middle
𝐸 = 𝑓 (𝐷) of 𝐶𝐴 and the median 𝐴𝐷 maps to the median 𝐵𝐸 (see figure 2). Then the
equality of ratios 𝐺𝐴/𝐺𝐷 = 𝐺𝐵/𝐺𝐸 = −2 shows that 𝐺 maps onto itself. Analogous is
the proof for the other permutations of the vertices of 𝐴𝐵𝐶 introduced by 𝑓 . Alterna-
tively, the result follows from theorem 5 and the representation of the centroid:
1
𝐺 = (𝐴 + 𝐵 + 𝐶), □
3

4 The associated ellipse of an affinity

An affinity 𝐹 may possess a center, i.e. an “isolated” point 𝑋(𝑥, 𝑦) such that
 
𝑎 11 𝑎12
𝑋 = 𝑓 (𝑋) = 𝐴𝑋 + 𝑣, with 𝐴 =
𝑎 21 𝑎22

𝑥 1 = 𝑎 11 𝑥 1 + 𝑎 12 𝑥 2 + 𝑣 1 ,
(5)
𝑥 2 = 𝑎 21 𝑥 1 + 𝑎 22 𝑥 2 + 𝑣 2 .
This is a linear system of equations and it has a unique solution when its determinant is
non zero:
𝑎 −1 𝑎12
𝐷 = 11 = | 𝐴| − 𝑡𝑟(𝐴) + 1 ≠ 0, (6)
𝑎 21 𝑎 22 − 1
where | 𝐴| is the determinant of 𝐴 and 𝑡𝑟(𝐴) = 𝑎 11 + 𝑎 22 is its “trace”. If condition (6) is
satisfied, then the solution is
1 1
𝑥0 = (𝑎 12 𝑣2 + (1 − 𝑎 22 )𝑣 1 ) , 𝑦0 = (𝑎 21 𝑣1 + (1 − 𝑎 11 )𝑣 2 ) . (7)
𝐷 𝐷
The condition | 𝐴| − 𝑡𝑟(𝐴) + 1 = 0 characterizes the lack of fixed points or the existence of
a whole line of fixed points of the affinity. In fact if this condition holds, then the system
(5) either has no solution or has a whole line of points 𝑋 = 𝑋0 + 𝑡𝑋1 satisfying it. In the
latter case 𝑋1 must be an eigenvector to the eigenvalue 1 of 𝐴. This is immediately seen
by applying the transformation:

𝐴𝑋 + 𝑣 = 𝑋 ⇔ 𝐴(𝑋0 + 𝑡𝑋1 ) + 𝑣 = 𝑋0 + 𝑡𝑋1

⇔ (𝐴𝑋0 + 𝑣 − 𝑋0 ) + 𝑡(𝐴𝑋1 − 𝑋1 ) = 0 ⇒ 𝐴𝑋1 = 𝑋1 ,

since the before to last equation is true for 𝑡 = 0 and 𝑡 = 1.

Corollary 15. If the affinity 𝑓 possesses an “axis” of fixed points, then the matrix 𝐴 has 1 as
eigenvalue and also holds:
| 𝐴| − 𝑡𝑟(𝐴) + 1 = 0 . (8)

Corollary 16. The affinity 𝑓 has a unique fixed point, if and only if, 1 is not an eigenvalue of its
matrix.

The “generic” affinity satisfies | 𝐴| − 𝑡𝑟(𝐴) + 1 ≠ 0 and consequently has a center. In

associated any case the behavior of the affinity can be inspected by watching its effect on a circle. In
ellipse fact, the affine image of any circle is an ellipse and the images of various circles under the
same affinity are pairwise homothetic ellipses. The reason for this is illustrated in figure 3.
The conjugate ℎ ′ = 𝑓 ◦ ℎ ◦ 𝑓 −1 of a homothety ℎ by an affinity 𝑓 is a homothety. In fact,
4 The associated ellipse of an affinity 8

a homothety is represented in cartesian coordinates by the equation 𝑌 = 𝑘𝑋 + (1 − 𝑘)𝑋0 ,

where 𝑋0 is the center of the homothety and 𝑘 its ratio. It is then readily seen that
𝑓 ◦ ℎ ◦ 𝑓 −1 is represented by 𝑌 = 𝑘𝑋 + (1 − 𝑘)(𝐴𝑋0 + 𝑣), which is also a homothety with
the same ratio and center at the point 𝑌0 = 𝐴𝑋0 + 𝑣 = 𝑓 (𝑋0 ). Since any two circles are
homothetic by the ratio 𝑘 = 𝑟/𝑟 ′ of their radii, the claim is true and the resulting ellipses
are homothetic by the same ratio 𝑘.

h'

f -1

Figure 3: Affine images of circles

Thus, an affinity defines an infinity of pairwise homothetic ellipses and by abstracting

to the proper shape and orientation of such an ellipse we can speak of the “ellipse associ-
ated to an affinity”. If we are not afraid to do calculations, we can represent this ellipse by
identifying it with the image via 𝑓 of the unit circle 𝑥 2 + 𝑦 2 = 1, the coordinates now as-
sumed to be ordinary cartesian referred to orthogonal unit vectors of length 1. Assuming
the affinity 𝑓 in the form
 
𝑥 ′ = 𝑎𝑥 + 𝑏𝑦 + 𝑣 1 , 𝐷𝑥 = 𝑑𝑥 ′ + −𝑏𝑦 ′ + 𝑏𝑣 2 − 𝑑𝑣 1 ,
⇒ , 𝐷 = 𝑎𝑑 − 𝑏𝑐. (9)
𝑦 ′ = 𝑐𝑥 + 𝑑𝑦 + 𝑣 2 , 𝐷𝑦 = −𝑐𝑥 ′ + 𝑎𝑦 ′ + 𝑐𝑣 1 − 𝑎𝑣 2 ,

A matrix representation of the ellipse is obtained through the product of matrices:

−1 0 0 𝑑 −𝑏 𝑏𝑣 2 − 𝑑𝑣 1
© ª © ª
𝑁 = 𝑀 𝑡 · ­ 0 −1 0® · 𝑀 with 𝑀 = ­−𝑐 𝑎 𝑐𝑣 1 − 𝑎𝑣2 ® , (10)
«0 0 1¬ «0 0 𝑎𝑑 − 𝑏𝑐 ¬

𝑀 being the 𝐷−multiple of the inverse of the matrix representing 𝑓 and 𝑀 𝑡 denoting
the transposed of 𝑀. Doing the multiplication we find:

𝐴 𝐵 𝐴𝑣 1 + 𝐵𝑣 2
© ª
𝑁 = ­ 𝐵 𝐶 −𝐵𝑣 1 − 𝐶𝑣 2 ® , (11)
«𝐴𝑣1 + 𝐵𝑣2 −𝐵𝑣 1 − 𝐶𝑣 2 𝐴𝑣 1 + 2𝐵𝑣 1 𝑣 2 + 𝐶𝑣 2 + 𝐷 ¬
2 2 2

with {𝐴 = −𝑐 2 − 𝑑 2 , 𝐵 = 𝑎𝑐 + 𝑏𝑑 , 𝐶 = −𝑎 2 − 𝑏 2 }. The invariants of this conic are

(see file Quadratic equations in the plane)

𝐽1 = 𝐴 + 𝐶 , 𝐽2 = 𝐴𝐶 − 𝐵2 = 𝐷 2 , 𝐽3 = 𝑑𝑒𝑡(𝑁) = 𝐷 4 .

The center of the ellipse is the image (𝑣 1 , 𝑣 2 ) = 𝑓 (0, 0) and translating there the coordinate
system, the equation of the ellipse obtains the next form, in which the coefficients do not
contain the constants {𝑣1 , 𝑣 2 }:

𝐴𝑥 2 + 2𝐵𝑥 𝑦 + 𝐶 𝑦 2 + 𝐷 2 = 0 .
 5 The conic of intersections of an affinity 9

3'
C 4
3 D F
4'

A
1 1

3 4 1' 2'

2 2
B

Figure 4: Determination of the fixed point of the affinity

Exercise 3. Consider the pencil of lines {𝛼 𝑡 } which are parallel to a fixed line 𝛼 0 . Show that for
the affinity 𝑓 the intersections {𝐼𝑡 = 𝛼 𝑡 ∩ 𝑓 (𝛼 𝑡 )} describe a line passing through the fixed point
of the affinity, if any.

Remark 5. The preceding exercise and the preservation of parallels by affinities (parallels
map to parallels) can be used to locate the isolated fixed point of the affinity, if any. In fact
consider the pairs {(1, 2), (3, 4)} of opposite side-lines of an arbitrary parallelogram (see
figure 4). Lines {1, 2} map via 𝑓 to paralles {1′ , 2′} and {3, 4} map to parallels {3′ , 4′}. Ac-
cording to the exercise, the line through the points {𝐴 = 1 ∩ 1′ , 𝐵 = 2 ∩ 2′} passes through
the fixed point 𝐹 of the affinity. Analogously passes through 𝐹 the line through the
points {𝐶 = 3 ∩ 3′ , 𝐷 = 4 ∩ 4′}. Thus 𝐹 can be found through a simple procedure of in-
tersections of lines.

5 The conic of intersections of an affinity

Figure 5 shows a circle 𝜅(𝑂) and its image 𝜅′ under the affinity 𝑓 . It also shows the image
œ
𝑋 š centered at 𝑂. As the right angle 𝑋𝑂𝑌
′ 𝑂 ′𝑌 ′ of a right angle 𝑋𝑂𝑌 š revolves about 𝑂

λ I

I'
D
Y
C
X X'
Y'
O Β Β'
C' Κ
O'
κ κ'
J

Figure 5: The image 𝜅′ = 𝑓 (𝜅) of a circle and the conic 𝜆 of the affinity 𝑓

affinity’s the image angle defines two conjugate directions {𝑂𝑋 ′ , 𝑂𝑌 ′} of the ellipse 𝜅 ′. In general
conic of the corresponding under 𝑓 lines intersect at points {𝐼 = 𝑂𝑋 ∩ 𝑂 ′𝑋 ′ , 𝐼 ′ = 𝑂𝑌 ∩ 𝑂 ′𝑌 ′}
intersections
describing a conic 𝜆. In fact, from the “Chasles-Steiner theorem” ([Cox87, p.77]), lines such
5 The conic of intersections of an affinity 10

as {𝑂𝑋 , 𝑂𝑋 ′} of two pencils {𝑂 ∗ , 𝑂 ′∗ } corresponding under a homography have an in-

tersection 𝐼 describing a conic passing through the centers {𝑂, 𝑂 ′} of the pencils. The
conic 𝜆 passes, of course, also through the fixed point 𝐷 of the affinity. It can be easily
seen that the tangent 𝑡 𝑂 to 𝜆 at 𝑂 maps via 𝑓 to line 𝑂𝑂 ′ and this maps in turn to
the tangent 𝑡 𝑂 ′ of 𝜆 at 𝑂 ′. Further the tangent at 𝐼 is the harmonic conjugate line of
the point 𝐽 w.r.t. the line pair {𝑂𝐼, 𝑂 ′𝐼}. In the case of the fixed point 𝐼 = 𝐷 the ratios
𝑂𝑋/𝑋𝐼 = 𝑂 ′𝑋 ′/𝑋 ′𝐼 and the line 𝑋𝑋 ′ becomes parallel to 𝑂𝑂 ′.
ε

Χ Κ' Χ'
ε' Ο Ο'
Κ
κ κ'

Figure 6: The axis 𝜀 of fixed points of the strain 𝑓

The shape of the conic 𝜆 reflects some basic properties of the corresponding affinity
𝑓 . Figure 6 illustrates the particular case of “strains” and “shears”, which are affinities pos-
sesing an entire line 𝜀 of fixed points. In this case the conic 𝜆 degenerates to a product of
two lines: the line 𝜀 containing all intersections {𝐼 = 𝑂𝑋 ∩ 𝑂 ′𝑋 ′} and the line 𝜀′ = 𝑂𝑂 ′ .
The lines 𝑋𝑋 ′ are parallel to 𝑂𝑂 ′ and if these parallels intersect the line 𝜀, then the ratio
𝑋𝐾 ′/𝐾 ′𝑋 ′ = 𝑂𝐾/𝐾𝑂 ′ is constant and the affinity is a “strain”. This proves the following.
Theorem 6. If the affinity 𝑓 has a whole line 𝜀 of fixed points, then there is a direction of the
plane, such that every line 𝜀′ parallel to that direction remains invariant under 𝑓 . If these parallels
intersect 𝜀, then for every pair of points {𝑋 , 𝑋 ′ = 𝑓 (𝑋)} the ratio 𝑘 = 𝑋𝐾/𝐾𝑋 ′ is constant, for
𝐾 = 𝑋𝑋 ′ ∩ 𝜀.
The direction referred to in this theorem is called “conjugate direction of the strain” and
𝑘 is called the “ratio of the strain”. Section 14 contains more on this kind of affinity.
Exercise 4. Show that if an affinity possesses an invariant line 𝜀′ and a fixed point 𝐷 not con-
tained in that line, then every line parallel to 𝜀′ is invariant under 𝑓 . If in addition there are two
corresponding under 𝑓 lines {𝜇, 𝜇′ = 𝑓 (𝜇)} intersecting at a point 𝐷 ′ and the parallel 𝜀0 to 𝜀′
from 𝐷 contains no other fixed point, then 𝑓 is a strain.

μ'
D' Β'
Α'
C'
Β
D C
ε0 δ ε ε' Α μ

Figure 7: Affinity with an invariant line 𝜀′ and a fixed point 𝐷 not in 𝜀

Hint: Since 𝐷 maps via 𝑓 to itself and the point at infinity of 𝜀0 which is also point at
infinity of 𝜀′ mapping to itself (because of the invariance of 𝜀′), the line 𝜀0 remains also
invariant under 𝑓 (see figure 7). Consider then an arbitrary line 𝜀 parallel to {𝜀′ , 𝜀0 }
and a secant 𝜇 intersecting the three lines at the points {𝐵 ∈ 𝜀0 , 𝐶 ∈ 𝜀, 𝐴 ∈ 𝜀′}. By the
5 The conic of intersections of an affinity 11

preservation of the ratios on lines by affinities the ratio 𝑘 = 𝐵𝐶/𝐶𝐴 equals the ratio of
the corresponding image points via 𝑓 : 𝑘 = 𝐵′𝐶 ′/𝐶 ′𝐴′ , implying that 𝐶 ′ ∈ 𝜀 and that 𝜀
is invariant too.
Consider now the point 𝐷 ′ = 𝜇 ∩ 𝜇′ and the parallel 𝜀′′ to 𝜀′ through 𝐷 ′ by the
invariance of 𝜀′′ point 𝑓 (𝐷 ′) will be 𝜀′′ ∩ 𝜇′ = 𝐷 ′. Hence 𝐷 ′ is fixed and consequently
(applying prematurely corollary 19) the whole line 𝛿 = 𝐷𝐷 ′ consists of fixed points of 𝑓
and is not parallel to the direction of the invariant lines.

ε
C'

C B' O'
E
κ'
κ X'
O B F

Ε' X λ
T
F'
I
ε'
S

Figure 8: Affinity with an invariant line

Figure 8 shows an affinity which has also an invariant line 𝜀 and also a fixed point
𝐷, now contained in the invariant line and assumed to be unique. The figure shows also
a circle 𝜅(𝑂), its image, the ellipse 𝜅′(𝑂 ′) and the corresponding conic 𝜆 containing
the intersections {𝐼 = 𝑂𝑋 ∩ 𝑂 ′𝑋 ′ , 𝑋 ∈ 𝜅}. Since affinities preserve the line at infinity,
the point at infinity 𝑋∞ of the line 𝜀 maps via 𝑓 to itself and lines parallel to 𝜀 map
also to parallels to 𝜀. The conic 𝜆 in this case is a hyperbola or parabola with an asymp-
tote/axis parallel to 𝜀. This, because for a point 𝐶 ∈ 𝜅 such that 𝑂𝐶 is parallel to 𝜀, the
corresponding image point 𝐶 ′ = 𝑓 (𝐶) defines 𝑂 ′𝐶 ′ also parallel to 𝑂𝐶. In the case of
hyperbola the parallel 𝜀′ to the other asymptote though 𝐷 is also an invariant line. An
important property in this case is the following.

Theorem 7. Let 𝑓 be an affinity which possesses a center, a unique fixed point 𝐷 and an invariant
line 𝜀 through it. When the affinity possesses a second invariant line 𝜀′ , then for every point 𝑋
of the plane and its image 𝑋 ′ = 𝑓 (𝑋) the line 𝑋𝑋 ′ intersects the two invariant axes {𝜀, 𝜀′} at
points {𝐹, 𝐹 ′} such that {𝐹𝑋/𝐹𝑋 ′ = 𝑘1 , 𝐹 ′𝑋/𝐹 ′𝑋 ′ = 𝑘 2 } are constants.

Proof. The proof follows by applying “Menelaus’ theorem” (see file Menelaus’ theorem).
For example, in the case of 𝐹, consider the triangles {𝑂𝑂 ′𝐼, 𝑋𝑋 ′𝐼} and the secant 𝜀 in-
tersecting their sides at the points {𝐸, 𝐹, 𝑇, 𝑆}:

𝐸𝑂 𝑇𝑂 ′ 𝑆𝐼 𝐹𝑋 𝑇𝑋 ′ 𝑆𝐼
· · = 1 and · · =1 ⇒
𝐸𝑂 ′ 𝑇𝐼 𝑆𝑂 𝐹𝑋 ′ 𝑇𝐼 𝑆𝑋
𝐸𝑂 𝑇𝑂 ′ 𝑆𝐼 𝐹𝑋 𝑇𝑋 ′ 𝑆𝐼
· · = · · ⇒
𝐸𝑂 ′ 𝑇𝐼 𝑆𝑂 𝐹𝑋 ′ 𝑇𝐼 𝑆𝑋
𝐸𝑂 𝐹𝑋 𝑇𝑋 ′ 𝑆𝑂 𝐹𝑋
′
= ′
· · ′
= .
𝐸𝑂 𝐹𝑋 𝑆𝑋 𝑇𝑂 𝐹𝑋 ′
5 The conic of intersections of an affinity 12

The last equality being valid because of the preservation by 𝑓 of ratios of points on a
line, according to which 𝑇𝑋 ′/𝑇𝑂 ′ = 𝑆𝑋/𝑆𝑂, since 𝑇 = 𝑓 (𝑆). The proof of the analogous
property for 𝐹 ′ is similar. □

X'
ε Y'

Y μ'
F
F'
ε' X μ
D

Figure 9: The ratio on the two invariant lines

Corollary 17. With the assumptions of the preceding theorem the ratios on the invariant lines are

𝐷𝑋 𝐷𝑋
= 𝑘 2 for 𝑋 ∈ 𝜀 and = 𝑘 1 for 𝑋 ∈ 𝜀′ .
𝐷𝑋 ′ 𝐷𝑋 ′
Proof. We show the first relation, the second being analogously provable. For this con-
sider the lines {𝜇, 𝜇′} through {𝑋 , 𝑋 ′ ∈ 𝜀} which are parallel to 𝜀′. They are related by
𝑓 : 𝑓 (𝜇) = 𝜇′. For an arbitrary point 𝑌 ∈ 𝜇 and 𝑌 ′ = 𝑓 (𝑌) ∈ 𝜇′ consider 𝐹 ′ = 𝑌𝑌 ′ ∩ 𝜀′.
By the preceding theorem 𝐹 ′𝑌/𝐹 ′𝑌 ′ = 𝐷𝑋/𝐷𝑋 ′ = 𝑘 2 . □
Remark 6. Affinities with a single fixed point and two invariant lines, like those of the
preceding theorem are characterized by matrices 𝐴 possessing two real eigenvalues: the
{𝑘 1 , 𝑘 2 } of the preceding corollary. Taking the fixed point as origin and the corresponding
eigenvectors as a basis, the affinity is described by the equations:

𝑥 ′ = 𝑘1 · 𝑥 , 𝑦 ′ = 𝑘2 · 𝑦 .

In section 17 we discuss the case of affinities whose corresponding conic 𝜆 is a parabola.

They are described by affinities represented by matrices with a double eigenvalue but with
one only corresponding eigenvector, pointing in the direction of the axis of the parabola.

μ'
μ κ ν'
ν
κ'
X X'

Figure 10: Two pairs of orthogonal directions corresponding under 𝑓 for any 𝑋

Exercise 5. For any affinity 𝑓 there are two pairs of orthogonal directions {(𝜇, 𝜈), (𝜇′ , 𝜈′)}, such
that for any point 𝑋 of the plane, the parallels to {𝜇, 𝜈} through 𝑋 , map via 𝑓 to two parallels
to {𝜇′ , 𝜈′} through 𝑋 ′ = 𝑓 (𝑋).

Hint: Consider a circle 𝜅 centered at 𝑋 and its image, the ellipse 𝜅′ centered at 𝑋 ′. The
pairs of orthogonal radii of the circle map to conjugate diameters of the ellipse. Consider
the lines which map via 𝑓 on the axes of the ellipse (see figure 10).
6 The conic of intersections of an affinity 13

6 The conic of intersections of an affinity

The conic 𝜆 of the preceding section, created from the intersections {𝐼 = 𝑂𝑋 ∩ 𝑂 ′𝑋 ′} of
a line 𝑂𝑋 rotating about 𝑂 and its image 𝑂 ′𝑋 ′ = 𝑓 (𝑂𝑋) via a general affinity 𝑓 , is in
some sense independent of the selected point 𝑂. Next theorem prepares the ground for
the discussion of this property.

Theorem 8. Let the conic 𝜆 be created by the intersections {𝐼 = 𝐴𝑋 ∩ 𝐵𝑋 ′}, where {𝐴, 𝐵} are
fixed points and points {𝑋 , 𝑋 ′} are points varying on the line 𝜀 related by a homography 𝑓 and
{𝐴, 𝐵} are not contained in 𝜀. Let also 𝐶 be a third point not contained in 𝜀 and for each 𝑋 ′ ∈ 𝜀
let 𝐶𝑌 by the parallel to 𝐵𝑋 ′ from 𝐶 intersecting 𝜀 at 𝑌. Then {𝑋 , 𝑌} are also homographically
related and the corresponding conic 𝜆′ of intersections {𝐼 ′ = 𝐴𝑋 ∩ 𝐶𝑌} is homothetic to or a
translate of 𝜆.

I
D A'

I'
λ
A
B

C λ'

ε
X Y X'

Figure 11: Homothetic conics generated by homographic relations

Proof. We may assume that 𝜀 is the 𝑥−coordinate axis and the homography is repre-
sented by the function 𝑡 ′ = (𝑎𝑡 + 𝑏)/(𝑐𝑡 + 𝑑) with 𝐷 = 𝑎𝑐 − 𝑏𝑑 ≠ 0. Then, the two lines
{𝐴𝑋(𝑡, 0), 𝐵𝑋 ′(𝑡 ′ , 0)} and their intersection point 𝐼 (see figure 11) are represented by the
equations

𝐴𝑋 : 𝑎 2 𝑥 + (𝑡 − 𝑎 1 )𝑦 − 𝑎 2 𝑡 = 0 for 𝐴(𝑎 1 , 𝑎 2 ) ,
 
𝑎𝑡 + 𝑏 𝑎𝑡 + 𝑏
𝐵𝑋 ′ : 𝑏2 𝑥 + − 𝑏1 𝑦 − 𝑏2 = 0 for 𝐵(𝑏1 , 𝑏 2 ) ,
𝑐𝑡 + 𝑑 𝑐𝑡 + 𝑑
𝐼 : (𝑎2 𝑏2 𝑐)𝑥 2 + (𝑏1 (𝑑 + 𝑎 1 𝑐) − 𝑏 − 𝑎𝑎 1 )𝑦 2 − (𝑏2 (𝑑 + 𝑎 1 𝑐) + 𝑎 2 𝑏 1 𝑐 − 𝑎𝑎 2 )𝑥𝑦
+ 𝑎 2 𝑏 2 (𝑑 − 𝑎)𝑥 − (𝑎 2 𝑏1 𝑑 − 𝑏2 (𝑏 + 𝑎𝑎 1 ) − 𝑎 2 𝑏)𝑦 − 𝑎 2 𝑏2 𝑏 = 0 .

The parallel from 𝐶(𝑐1 , 𝑐 2 ) to 𝐵𝑋 ′ intersects the line 𝜀 at 𝑌(𝑡 ′ , 0) with

(𝑏2 𝑐1 𝑐 + 𝑎𝑐 2 − 𝑏1 𝑐 2 𝑐)𝑡 + (𝑏2 𝑐1 𝑑 + 𝑏𝑐 2 − 𝑏1 𝑐2 𝑑)

𝑡′ = ,
𝑏2 (𝑐𝑡 + 𝑑)

confirming the claim that 𝑡 ′ = 𝑓 (𝑡) is a homographic relation. Without loss of general-
ity we may use a homothety with center on 𝜀 and assume that point 𝐵 coincides with
𝐵(0, 1). This, reduces somewhat the calculations and simplifies the representation of the
 6 The conic of intersections of an affinity 14

homography and the corresponding two conics to:

(𝑎𝑐 2 + 𝑐𝑐 1 )𝑡 + (𝑐 1 𝑑 + 𝑏𝑐 2 )
𝑡′ = ,
𝑐𝑡 + 𝑑
𝐼 : (𝑎 2 𝑐)𝑥 2 − (𝑏 + 𝑎𝑎 1 )𝑦 2 + (𝑎𝑎 2 − 𝑑 − 𝑎 1 𝑐)𝑥𝑦 + 𝐷1 𝑥 + 𝐸1 𝑦 + 𝐹1 = 0 ,
𝐼 ′ : (𝑎 2 𝑐)𝑐 2 𝑥 2 − (𝑏 + 𝑎𝑎 1 )𝑐 2 𝑦 2 + (𝑎𝑎2 − 𝑑 − 𝑎 1 𝑐)𝑐2 𝑥𝑦 + 𝐷2 𝑥 + 𝐸2 𝑦 + 𝐹2 = 0 .

The coefficients of the linear terms can be computed but the condition ([Cas93, p.326])
guarantying the homothety of the two conics does not need them, since this can be read
from their quadratic terms, which are respectively multiples by the same constant 𝑐 2 . If
𝑐2 = 1 then 𝜆′ is a translate of 𝜆. This is the case when point 𝐶 lies on the parallel to 𝜀
through 𝐵. □
Exercise 6. With the notation and conventions of the preceding theorem, show that line 𝐵𝐶 passes
through an intersection point 𝐴′ of the two conics {𝜆, 𝜆′} (see figure 11).

Theorem 9. For two fixed points {𝐴, 𝐵} the corresponding conics {𝜆, 𝜆′} generated respectively
through the intersections of lines {𝐼 = 𝐴𝑋 ∩ 𝐴′𝑋 ′ , 𝐽 = 𝐵𝑌 ∩ 𝐵′𝑌 ′}, where 𝐴′𝑋 ′ = 𝑓 (𝐴𝑋) and
𝐵′𝑌 ′ = 𝑓 (𝐵𝑌) for an affinity 𝑓 and {𝑋 , 𝑌} varying on respective circles {𝜅(𝐴), 𝜇(𝐵)}, are ho-
mothetic (see figure 12) or each is a translate of the other.

ν Η
κ
A
B'
μ'
X I
Y'
μ
B J
λ' λ
A'
Y
X' κ'

Figure 12: Homothetic conics generated by affinities

Proof. We reduce the proof to that of the preceding theorem. For this we consider the
line 𝐵𝑌 to be parallel to 𝐴𝑋. Since 𝑓 preserves parallelity, the lines {𝐴′𝑋 ′ , 𝐵′𝑌 ′} are also
parallel and 𝐵′𝑌 ′ intersects the line 𝐴𝑋 at a point 𝐻. According to theorem 8 point 𝐻
describes a conic 𝜈 homothetic to or a translate of 𝜆. For the same reason the conic 𝜈 is
also homothetic to or a translate of 𝜆′. Since homotheties+translations build a group this
proves that the two conics {𝜆, 𝜆′} are homothetic or each is a translate of the other. □
affinity’s We call the conic 𝜆 the “conic of intersections” of the affinity 𝑓 . From the discussion in
conic of this section follows that this conic is well defined from 𝑓 up to homothety.
intersections
Remark 7. In this and the preceding section we assumed that the conic 𝜆 exists. This
means, that we had at least one point 𝐴 on the plane, such that the lines 𝜀 = 𝐴𝑋 through
it do not map via 𝑓 to lines 𝐴′𝑋 ′ parallel to 𝐴𝑋 , thus defining indeed intersection points
𝐼 = 𝐴𝑋 ∩ 𝐴′𝑋 ′. If there is a point 𝐴 such that every line 𝐴𝑋 maps to a “parallel” line
𝐴′𝑋 ′ , then we cannot define the conic 𝜆 and the affine map is a special one, called “di-
latation” handled in section 12 (exercise 24, see also section 15).
 7 Special kinds of affinities 15

X'

O'
λ

X
φ O
D
I

Figure 13: Rotation: the conic of the affinity is a circle

Exercise 7. Show that for a rotation 𝐷(𝜑), its conic 𝜆 of intersections is a circle through its
center 𝐷 (see figure 13). For the converse see exercise 29.
Exercise 8. Show that the affinity 𝑓 preserves the set of its conics of intersections 𝛬 ={𝜆}, i.e.
the image 𝜆′ = 𝑓 (𝜆) for 𝜆 ∈ 𝛬 is also a conic of 𝛬.
Hint: This is trivial, since 𝐼 = 𝑂𝑋 ∩ 𝑂 ′𝑋 ′ = 𝑂𝑋 ∩ 𝑓 (𝑂𝑋) ⇒ 𝑓 (𝐼) = 𝑂 ′𝑋 ′ ∩ 𝑓 (𝑂 ′𝑋 ′).

7 Special kinds of affinities

Using the general properties of affinities, discussed in the preceding sections, we can de-
fine affinities in several ways, mainly using their determination by prescribing the images
{ 𝑓 (𝐴), 𝑓 (𝐵), 𝑓 (𝐶)} of three non-collinear points, which we may consider as the vertices of
a triangle. An important role in this endeavour play the “fixed points”, i.e. points such
that 𝑓 (𝑃) = 𝑃 and the “invariant lines”, i.e. lines such that 𝑓 (𝜇) = 𝜇. By the preceding
discussion, an affinity different from the identity can have zero or one or infinite many
fixed points (invariant lines). From the preceding discussion we deduce also the following
corollaries:

μ'
B'
O A'
A B
μ

Figure 14: Property of the fixed point for 𝜇′ = 𝑓 (𝜇) ≠ 𝜇

property of Corollary 18. If the affinity has one fixed point 𝑂, then it maps every line 𝜇 through 𝑂 onto
the center a line 𝑓 (𝜇) = 𝜇′ through 𝑂, and for two points {𝐴, 𝐵 ∈ 𝜇} and their images 𝐴′ , 𝐵′ ∈ 𝜇′ , if
𝜇′ ≠ 𝜇, the segments {𝐴𝐴′ , 𝐵𝐵′} are parallel (see figure 14).
two points Corollary 19. If the affinity 𝑓 fixes two points {𝐴, 𝐵} of the line 𝛼 then, it leaves the whole line
fixing invariant “fixing every point of the line”.
Corollary 20. If the affinity has one invariant line 𝜇 then it maps every line 𝜇′ parallel to 𝜇 to
a line 𝜇′′ also parallel to 𝜇.
Corollary 21. If the affinity has two non-parallel invariant lines {𝜇1 , 𝜇2 }, their intersection point
𝑂 = 𝜇1 ∩ 𝜇2 remains fixed and every parallelogram with sides parallel to these two lines maps to
a parallelogram with the same property.
Corollary 22. If the affinity has three pairwise non-parallel invariant lines {𝜇1 , 𝜇2 , 𝜇3 }, then it
is the identity.
7 Special kinds of affinities 16

Besides the “equiaffinities” alluded to in section 2, the following special kinds of affini-
ties are often used as “fundamental”, to which decompose all other general affinities, in the
sense that the general 𝑓 is written as a composition 𝑓 = 𝑓𝑛 ◦ ... ◦ 𝑓1 .

1. “Axial affinities” characterized by having a line of fixed points ([ea83, p.80], [Tar11,
p.116]).

2. “Affine reflections”, a kind of “axial affinities”.

3. “Strains”, a kind of “axial affinities” with a 𝐵 = 𝑓 (𝐴) and 𝐴𝐵 intersecting the axis.

4. “Shears”, a kind of “axial affinities” with a 𝐵 = 𝑓 (𝐴) and 𝐴𝐵 parallel to axis.

5. “Dilatations” mapping “every line to a parallel to it”.

6. “Symmetries” or “Half-turns” usual point-symmetries, a kind of “Dilatation”

7. “Translations”, a kind of “Dilatation” without fixed points.

8. “Homotheties”, a kind of “Dilatation” with a fixed point.

9. “Affine glide reflections”, a product of an “affine reflection” and a “translation”.

10. “Hyperbolic rotations”, 1 of 5 classes of “equiaffinities” ([Cox67, p.22]).

11. “Crossed hyperbolic rotations”, 1 of 5 classes of “equiaffinities”.

12. “Elliptic rotations”, 1 of 5 classes of “equiaffinities”.

13. “Parabolic rotations”, 1 of 5 classes of “equiaffinities”.

14. “Focal rotations”, 1 of 5 classes of “equiaffinities”.

Affinities

Direct Affinities Products of reflections

Dilatations
Equiaffinities

Translations and Half-turns

Translations

Figure 15: Inclusion diagram of main normal subgroups of the group of affinities

Figure 15 given also by Coxeter [Cox67, p.15] shows the inclusion diagram of the most
important “normal subgroups” of the group 𝐺 of affinities. These are sets 𝐺 ′ of special
affinities closed under composition, inversion and conjugation, latter meaning that:

𝑔 ◦ 𝑓 ◦ 𝑔 −1 = 𝑓 ′ ∈ 𝐺 ′ for every 𝑓 ∈ 𝐺′ and every 𝑔 ∈ 𝐺.

“Direct affinities” are those that preserve the orientation of the triangles, equivalently in the
description of affinities with matrices {𝐴, 𝐴𝑣 }, the determinants are positive |𝐴| = | 𝐴𝑣 | > 0.
An important subcategory of direct affinities are the “direct” or “rotational similarities” han-
dled in the file Homotheties and Similiarities. Affinities reversing the orientation are
called “reflective” or “indirect” and are characterized by | 𝐴| < 0. An example of the latter
kind represent the indirect or reflective similarities handled in the aforementioned file.
 8 Axial affinities or homologies, shears, strains 17

8 Axial affinities or homologies, shears, strains

axial By definition, “axial affinities” or “homologies”([Tar11, p.116]) leave a single line point-wise
affinity fixed. They are usually defined by prescribing their images at three points:
𝑓 (𝐴) = 𝐴, 𝑓 (𝐵) = 𝐵, (𝐴𝐵 = the axis,) and 𝑓 (𝐶) = 𝐷 ≠ 𝐶.
They could be represented by a quadrangle and its diagonals. If the diagonals intersect
𝐴𝐵 ∩ 𝐶𝐷 ≠ ∅, then the image 𝐷 ′ = 𝑓 (𝐶 ′) of any other point 𝐶 ′ can be found using the

(I) (II)
A A
D' C' D'
C' E'
E'
C E D E D
C

F F

B B

Figure 16: Strains: axial affinities with {𝐴 = 𝑓 (𝐴), 𝐵 = 𝑓 (𝐵), 𝐷 = 𝑓 (𝐶), 𝐸 = 𝐶𝐷 ∩ 𝐴𝐵}

line 𝐶𝐶 ′ and its intersection 𝐹 = 𝐶𝐶 ′ ∩ 𝐴𝐵 (see figure 16). Point 𝐷 ′ = 𝑓 (𝐶) will be on
line 𝐹𝐷 = 𝑓 (𝐹𝐶) and from the preservation of ratios by corollary 12, {𝐶 ′𝐷 ′ , 𝐶𝐷} must
be parallel, giving a geometrical construction of the image point for all 𝐶 ′ of the plane,
except for those 𝐶 ′ ∈ 𝐶𝐷, which, by the analogous constancy of ratio are seen to map to
points on the same line 𝐶𝐷. From the similar triangles involved, follows that, introducing
strains line coordinates with x-axis along 𝐴𝐵 and y-axis along 𝐶𝐷 the affinity 𝑓 is described
by a real constant 𝑘, different from 0 and 1 and the matrix
 ′      
𝑥 1 0 𝑥 𝑥
′ = · = . (12)
𝑦 0 𝑘 𝑦 𝑘𝑦
axis and Transformations of this kind are called “strains” or “skew compressions”. The line 𝐴𝐵 is the
conjugate “axis” the direction of 𝐶𝐷 is called “conjugate direction” and the constant 𝑘 the “ratio” of
direction
the strain. In the special case 𝑘 = −1 they are called “affine reflections” or “skew reflections”
of a strain
and the axis is often referred to as mirror. “Euclidean reflections” are a special case of affine
affine reflections: their conjugate direction is orthogonal to the axis.
reflections If the diagonals {𝐴𝐵, 𝐶𝐷} do not intersect (See Figure 17), then the same procedure leads

C D

C' D'

F A B

Figure 17: Shears: axial affinities with 𝐶𝐷 ∩ 𝐴𝐵 = ∅

to an analogous construction of the image 𝐷 ′ = 𝑓 (𝐶 ′) for an arbitrary point of the plane.

In this case introducing line coordinates with x-axis along 𝐴𝐵 and y-axis along 𝐴𝐶, we
see that the affinity is described by a constant 𝑘 ≠ 0 and the matrix
 ′      
𝑥 1 𝑘 𝑥 1 + 𝑘𝑦
= · = . (13)
𝑦′ 0 1 𝑦 𝑦
 8 Axial affinities or homologies, shears, strains 18

shears Transformations of this kind are called 𝑠 ℎ𝑒 𝑎𝑟𝑠 . Line 𝐴𝐵 is the “axis” and the constant 𝑘
is the “ratio” of the shear. Next theorems summarize the discussion.

Theorem 10. Every axial affinity is defined by two triangles {𝐴𝐵𝐶, 𝐴𝐵𝐷} with a common side
consisting of fixed points, or equivalently, by four pairwise different arbitrary points {𝐴, 𝐵, 𝐶, 𝐷}
with {𝐶 ∉ 𝐴𝐵, 𝐷 ∉ 𝐴𝐵} and prescribing the values:
𝑓 (𝐴) = 𝐴, 𝑓 (𝐵) = 𝐵 and 𝑓 (𝐶) = 𝐷.

Theorem 11. The only affine transformations which leave a single line 𝜀 point-wise fixed are
axial affinities, i.e. shears and strains having 𝜀 for axis.

X
Y
A
A'

Z B C X'

Figure 18: Axial affinity defined by two triangles with a common base

Exercise 9. Show that two trianlges {𝐴𝐵𝐶, 𝐴′𝐵𝐶} with a common side define an axial symme-
try 𝑓 fixing point-wise the common base 𝐵𝐶 and mapping 𝐴 ↦→ 𝐴′. Show also that for each
point of the plane 𝑋 ∉ 𝐵𝐶 the image 𝑌 = 𝑓 (𝑋) defines a line 𝑋𝑌 parallel to 𝐴𝐴′ and two lines
{𝐴𝑋 , 𝐴′𝑋 ′} intersecting at a point 𝑍 ∈ 𝐵𝐶. Show finally, that if 𝑋 ′ is the projection of 𝑋 on
𝐵𝐶, the triangles 𝑋𝑋 ′𝑌 have constant angles (See Figure 18).

Exercise 10. Show that a quadrangle 𝐴𝐵𝐶𝐷 defines through its diagonals two strains. The first
strain 𝑓1 has for axis the diagonal 𝐴𝐶 and conjugate direction that of the other diagonal 𝐵𝐷.
Its ratio is 𝑘 = 𝐸𝐵/𝐸𝐷, where 𝐸 = 𝐴𝐶 ∩ 𝐵𝐷 (see figure 19). Analogously is defined the other
strain 𝑓2 with axis 𝐵𝐷 and ratio 𝑘 ′ = 𝐸𝐶/𝐸𝐴. Find some conditions, so that these strains leave
the quadrangle invariant.
Show also that in a coordinate system with (𝑥, 𝑦) -axes the pair (𝐸𝐷, 𝐸𝐴) the composition
𝑓2 ◦ 𝑓1 of the two strains is described by the equations:
  
𝑥 ′ = 𝑘 · 𝑥, ′ 𝑘 0
⇔ 𝑋 = 𝐴𝑋 with 𝐴= . (14)
𝑦 ′ = 𝑘 ′ · 𝑦. 0 𝑘′

E D
B
C

Figure 19: Strains defined by a quadrangle 𝐴𝐵𝐶𝐷

Exercise 11. Show that the axial affinities with the same axis 𝛼 form a group. Show further that
this group acts “transitively” on the plane, in the sense that for any two different points {𝐴, 𝐵},
there is an affinity of this group mapping 𝐴 to 𝐵.
 9 Affinities leaving invariant a triangle 19

9 Affinities leaving invariant a triangle

From the affinities 𝑓 , different from the identity and leaving invariant the triangle 𝐴𝐵𝐶,
of particular importance is an affinity leaving one vertex, 𝐴 say, fixed (see figure 20). Then

X ε X1 Y1 Y
N

B C
D

Figure 20: Affine reflection

it must interchange the other two: { 𝑓 (𝐵) = 𝐶, 𝑓 (𝐶) = 𝐵} leaving also fixed the middle 𝐷
of 𝐵𝐶 and, by corollary 19, fixing the whole line 𝐴𝐷, while mapping line 𝐵𝐶 onto itself.
By applying corollary 12 we see that 𝑓 acts on 𝐵𝐶 as the symmetry about 𝐷. The same
behaviour can be seen to happen on every parallel line 𝜀 to 𝐵𝐶. In fact, if the parallel
intersects the median 𝐴𝐷 at the point 𝑁 and the sides {𝐴𝐵, 𝐴𝐶} at {𝑋1 , 𝑌1 }, then, again
by corollaries 9, 19 and 12, follows that a) 𝑋1𝑌1 is parallel to 𝐵𝐶, and b) 𝑓 (𝑋) = 𝑌 is the
symmetric of 𝑋 w.r. to 𝑁 . Next theorem summarizes the short discussion and justifies
reflection the notation 𝐴(𝐵𝐶) used by Coxeter for the affine reflection interchanging {𝐵, 𝐶} and
𝐴(𝐵𝐶) fixing point 𝐴 ∉ 𝐵𝐶.
Theorem 12. If an affinity, different from the identity, leaves invariant a triangle and fixes a vertex
of it, then it is an affine reflection with axis the median from that vertex and conjugate direction the
opposite side to that vertex.

A C
D
D
B
ε ε

A B
C

Figure 21: Affine reflections leaving invariant a quadrangle

Exercise 12. Show that if an affine reflection leaves invariant a quadrangle, then its axis either
coincides with a diagonal or it is the line joining the middles of two opposite sides. Show also that
in the first case the conjugate direction is that of the other diagonal, whose middle must lie on the
axis. In the second case the quadrangle is a trapezium and the conjugate direction is that of the
parallel sides (See Figure 21).
Consider now the product (composition) of two reflections as these of the preceding
section. First 𝑓𝐵𝐴 then 𝑓𝐶𝐵 . The first with axis the median 𝜀1 = 𝐶𝐶 ′ and the second the
reflection with axis the median 𝜀2 = 𝐴𝐴′. Their product transformation 𝑓 = 𝑓𝐶𝐵 ◦ 𝑓𝐵𝐴 is an
𝑓 𝑓 𝑓
affinity recycling clockwise the vertices {𝐴 ↦→ 𝐶 ↦→ 𝐵 ↦→ 𝐴} (See Figure 22). Since each of
the factors leaves invariant the triangle, the same happens also with their composition 𝑓 .
Since the centroid 𝐺 is fixed by each reflection, it is fixed also by their composition. Every
other point 𝑋 ≠ 𝐺 maps to a point 𝑌 = 𝑓 (𝑋) and if the line 𝐺𝑋 intersects a side, 𝐶𝐵
 9 Affinities leaving invariant a triangle 20

ε2
Y=f(X)
A
Y1
ε1
C'
G

B X1 A' C

Figure 22: Composition of two reflections

say at a point 𝑋1 , then this line maps to 𝑓 (𝐺𝑋) = 𝐺𝑌, which intersects the image-side
𝐵𝐴 at a point 𝑓 (𝑋1 ) = 𝑌1 . From the discussed properties we deduce easily the relations

𝑋1 𝐶 𝑌1 𝐵 𝑋1 𝐺 𝑌1 𝐺
= and = .
𝑋1 𝐵 𝑌1 𝐴 𝑋1 𝑋 𝑌1𝑌
From this and similar relations w.r. to the other sides of the triangle we deduce that there
is no other fixed point of 𝑓 .
affine Affinities which can be represented as product of two affine reflections and have a
rotation unique fixed point, are called “affine rotations” and their fixed point is called “center” of
the rotation. Next theorem formulates what we proved so far.

ε2

A
X2
ε1
C'
G
X1

B A' X C

Figure 23: The centroid of triangles with vertices {𝑋 , 𝑋1 = 𝑓 (𝑋), 𝑋2 = 𝑓 2 (𝑋)} is 𝐺

Theorem 13. The product of two affine reflections leaving invariant the triangle 𝐴𝐵𝐶 is an affine
rotation with center at the centroid 𝐺 of the triangle (see figure 23).

same Exercise 13. Show explicitly that the affine rotation 𝑓 , defined above, has no other fixed point
centroid except 𝐺. For the same rotation consider an arbitrary point 𝑋 and show that 𝑓 3 (𝑋) = 𝑋. Show
also that the centroid of the triangle with vertices {𝑋 , 𝑋1 = 𝑓 (𝑋), 𝑋2 = 𝑓 2 (𝑋)} is always the
fixed point 𝐺 of 𝑓 (see figure 23).

symmetry Exercise 14. Consider the affine rotation, which is the product of reflections 𝑓 = 𝑓2 ◦ 𝑓1 , where
{ 𝑓1 , 𝑓2 } have for axes the diagonals {𝐴𝐶, 𝐵𝐷} of the parallelogram 𝐴𝐵𝐶𝐷 and conjugate direc-
tions respectively the other diagonals {𝐵𝐷, 𝐴𝐶}. Show that 𝑓 coincides with the symmetry at the
center 𝐸 of the parallelogram (See Figure 24).

Exercise 15. Determine all the affinities leaving invariant a parallelogram and show that they form
a group.
9 Affinities leaving invariant a triangle 21

A D

E
X Y

B C

Figure 24: Symmetry at 𝐸 represented as an affine rotation

Exercise 16. Show that a pentagon has in general no affinities preserving it, i.e. mapping it onto
itself. Show also that if the pentagon has each diagonal parallel to a side, then it has affinities leaving
it invariant. Describe then all these affinities leaving the pentagon invariant.

Remark 8. The construction of a pentagon having each diagonal parallel to a side can be
done by selecting the triangle 𝐷𝐴𝐵 of a regular pentagon 𝐴𝐵𝐶𝐷𝐸 and applying to it an
affinity 𝑓 . The image pentagon of 𝐴𝐵𝐶𝐷𝐸 via the affinity is a pentagon as required (see
figure 25). Last figure suggests an alternative construction of such a pentagon:

ε D ζ
D

E K F
E C C

Y X

A B η
A B

Figure 25: Affine image of a regular pentagon

1. Starting with an arbitrary triangle 𝐴𝐵𝐷 draw from {𝐴, 𝐵} respectively parallels
{𝜀 ∥ 𝐷𝐵 , 𝜁 ∥ 𝐷𝐴} intersecting at 𝐽.
2. From a moving point 𝑋 ∈ 𝜁 draw the parallel 𝑋𝑌 ∥ 𝐴𝐵, 𝑌 ∈ 𝜀.
3. Consider the intersection 𝑍 = 𝑌𝐷 ∩ 𝑋𝐴.
4. As point 𝑋 varies on 𝜁 point 𝑍 describes a hyperbola.
5. For a point 𝑋 = 𝐶 ∈ 𝜁 the lines {𝑋𝐴, 𝑌𝐷} become parallel to an asymptote 𝜂 of the
hyperbola.
6. From 𝐶 draw the parallel 𝐶𝐸 ∥ 𝐴𝐵, 𝐸 ∈ 𝜀. This defines the requested pentagon
𝐴𝐵𝐶𝐷𝐸.
The side 𝐷𝐸 is then parallel to 𝐶𝐴. To show that the remaining side 𝐷𝐶 is also parallel
to 𝐵𝐸 it suffices to see that 𝐹𝐷/𝐹𝐵 = 𝐹𝐶/𝐹𝐸 for 𝐹 = 𝐸𝐶 ∩ 𝐵𝐷 and 𝐾 = 𝐸𝐶 ∩ 𝐴𝐷. This
follows from the relations due to the created parallels:

𝐹𝐷 𝐾𝐷 𝐾𝐸 𝐾𝐸 𝐴𝐸 𝐹𝐶 𝐹𝐶 𝐵𝐶 𝐴𝐸
= = = = and = = = .
𝐹𝐵 𝐾𝐴 𝐾𝐶 𝐴𝐵 𝐴𝐽 𝐹𝐸 𝐵𝐴 𝐵𝐽 𝐴𝐽
 10 The period of an affinity 22

Notice also the relation 𝐽𝐴2 = 𝐽𝐸 · 𝐴𝐸 due to:

𝐴𝐽 𝐵𝐽 𝐵𝐽 𝐷𝐴 𝐸𝐶 𝐸𝐽
= = = = = .
𝐴𝐸 𝐵𝐶 𝐴𝐾 𝐴𝐾 𝐶𝐾 𝐽𝐴

Exercise 17. Using the preceding relation show that the pentagons having each diagonal parallel
to a corresponding side are precisely the images by affinities of regular pentagons.

This discussion proves the following theorem.

Theorem 14. If a pentagon has four of its sides parallel to four corresponding diagonals, then
the fifth side is also parallel to the fifth diagonal and the pentagon is an affine image of the regular
pentagon.

10 The period of an affinity

period The affine reflections have period 2. This means, they satisfy 𝑓 2 = 𝑓 ◦ 𝑓 = 𝑒, where 𝑒
represents the identity transformation. More general the “period” of a transformation is
the smallest integer 𝑛, such that 𝑓 𝑛 = 𝑒. The transformation 𝑓 = 𝑓𝐶𝐵 ◦ 𝑓𝐵𝐴 of section 9 is
of period 3. This can be immediately seen by considering its effect on the vertices, which,
𝑓 𝑓 𝑓
as we already noticed, is {𝐴 ↦→ 𝐶 ↦→ 𝐵 ↦→ 𝐴}. Thus, 𝑓 3 leaves the three vertices fixed,
hence, by corollary 5, it is the identity.
Here we should notice, that not all affinities have a period. It is possible that no 𝑛 ex-
infinite ists, such that 𝑓 𝑛 = 𝑒. We say then that 𝑓 has “infinite period”. The product of affine reflec-
period tions 𝑓 = 𝑓𝐶𝐵 ◦ 𝑓𝐵𝐴 of period 3 is a special one related to a triangle. In general, the product
of two reflections is of infinite period. Figure 26 shows such an affine rotation of infinite

X4 X

X1 X3
O

X5 C
A X2
B β
α

Figure 26: Composition 𝑓 = 𝑓𝛽 ◦ 𝑓𝛼 of two reflections of infinite period

period. There the affinity 𝑓 is the product of the reflection 𝑓𝛼 with axis 𝛼 and conjugate
direction 𝐴𝐵 and the reflection 𝑓𝛽 with axis 𝛽 and conjugate direction 𝐵𝐶. The sequence
of points is {𝑋 , 𝑋1 = 𝑓 (𝑋), 𝑋2 = 𝑓 (𝑋1 ) = 𝑓 2 (𝑋), 𝑋3 = 𝑓 3 (𝑋), ..., 𝑋𝑛 = 𝑓 𝑛 (𝑋), ...} which
never recurs to the starting point 𝑋 . In section 18 we’ll see that all these points making
the “orbit” of 𝑋 under the composition of affine reflections 𝑓 = 𝑓𝛽 ◦ 𝑓𝛼 and the iterates
of 𝑓 lie on a conic, as suggested by the figure.

period 2 = Theorem 15. An affinity is of period 2, if and only if it is an affine reflection or a symmetry at a
reflection or point 𝑀.
symmetry
Proof. That an affine reflection is of period 2, follows from the definition of the affine re-
flection in section 9. To show the converse, consider a point 𝑋 with 𝑌 = 𝑓 (𝑋) ≠ 𝑋 hence
𝑓 (𝑌) = 𝑋. Then the line 𝛼 = 𝑋𝑌 is invariant under 𝑓 , i.e. 𝑓 (𝛼) = 𝛼 and the middle 𝑀
of the segment 𝑋𝑌 is fixed by 𝑓 . Take also another point 𝑋 ′ , consider again 𝑌 ′ = 𝑓 (𝑋 ′),
 10 The period of an affinity 23

α'
X' M' Y'
α
X M Y

Figure 27: Affinity of period 2 is a reflection

the line 𝛼′ = 𝑋 ′𝑌 ′ and the middle 𝑀 ′ of 𝑋 ′𝑌 ′ , which again remains fixed under 𝑓 (See
Figure 27). Assume now that 𝑀 ′ ≠ 𝑀. Then, by corollary 9 the line 𝛽 = 𝑀 𝑀 ′ remains
fixed under 𝑓 . For the intersection 𝑂 = 𝑋𝑋 ′ ∩ 𝑀 𝑀 ′ follows then that also {𝑌, 𝑌 ′ , 𝑂} are
collinear and from Thales’ theorem follows that lines {𝛼, 𝛼′} are parallel, thereby proving
the reflective property of the transformation. The reflection’s axis is line 𝛽 and the con-
jugate direction is that of the line 𝛼 = 𝑋𝑋 ′ for an arbitrary point 𝑋 not lying on 𝛽.

Y
Z'
ε
Y'
X' ε'
M
Z
X

Figure 28: A point symmetry is an affinity of period 2

In the case 𝑀 = 𝑀 ′ we have the situation of figure 28, in which 𝑋𝑋 ′𝑌𝑌 ′ is a parallelo-
gram and 𝑀 is its center and line 𝜀 = 𝑋𝑋 ′ maps under 𝑓 onto 𝜀′ = 𝑌𝑌 ′. By the preser-
vation of ratios it is seen that points 𝑍 ∈ 𝜀 maps to its symmetric 𝑍 ′ ∈ 𝜀′ . Using again
the preservation of ratios, we see easily that each point on 𝑀𝑍 maps to its symmetric w.r.
to 𝑀. This proves that 𝑓 is the symmetry at 𝑀. □
half-turns Remark 9. Symmetries about a fixed point 𝑀 are often called “half-turns” about 𝑀.

ε1
C' ε2
G

Β A' C

Figure 29: Affinity of period 3 is an affine rotation

Theorem 16. An affinity is of period 3, if and only if it is an affine rotation with axes two medians
of a triangle and the corresponding sides as conjugate directions.
Proof. That an affinity with the above properties is an affine rotation has been established
in section 9. To show the converse, consider a point 𝐴 with 𝐶 = 𝑓 (𝐴) ≠ 𝐴 and 𝑓 (𝐶) = 𝐵,
11 Translations 24

having necessarily 𝑓 (𝐵) = 𝐴. Assume for the moment that {𝐴, 𝐵, 𝐶} are not collinear and
form a genuine triangle 𝐴𝐵𝐶. In this triangle consider the reflections { 𝑓𝐵𝐴 , 𝑓𝐶𝐵 } and their
product 𝑓 ′ = 𝑓𝐶𝐵 ◦ 𝑓𝐵𝐴 , defined as in the preceding section. The two affinities 𝑓 and 𝑓 ′
recycle the vertices in the same way {𝐴 ↦→ 𝐶 ↦→ 𝐵 ↦→ 𝐴} (see figure 29). Thus, by corollary
6, they coincide everywhere, thereby proving the theorem.
It remains to show that there is a point 𝐴, such that {𝐴, 𝐶 = 𝑓 (𝐴), 𝐵 = 𝑓 (𝐶)} are not
collinear. To see this we work with the matrices 𝐹𝑣 representing the affinity 𝑓 as ex-
plained in the first two sections. The collinearity of {𝐴, 𝑓 (𝐴), 𝑓 2 (𝐴)} means for the corre-
sponding matrix representation, the existence of a number 𝜆 such that:

𝐹𝑣2 · 𝐴 = 𝜆𝐴 + (1 − 𝜆)𝐹𝑣 · 𝐴 ⇒
𝐴 = 𝐹𝑣3 · 𝐴 = 𝜆𝐹𝑣 · 𝐴 + (1 − 𝜆)𝐹𝑣2 · 𝐴
= 𝜆𝐹𝑣 · 𝐴 + (1 − 𝜆)[𝜆𝐴 + (1 − 𝜆)𝐹𝑣 · 𝐴] ⇒
(𝜆 − 𝜆 + 1)𝐴 = (𝜆 − 𝜆 + 1)𝐹𝑣 · 𝐴.
2 2

Since the factor never vanishes, last equation is equivalent with 𝐹𝑣 · 𝐴 = 𝐴 and if this is
valid for every 𝐴, then 𝐹𝑣 is the identity matrix, which is contrary to the assumption. □

11 Translations
Theorem 17. The product 𝑓 = 𝑓2 ◦ 𝑓1 of two reflections { 𝑓1 , 𝑓2 } with parallel axes {𝜀1 , 𝜀2 } and
common conjugate direction 𝛽 is a “translation” i.e. a map defined by a fixed vector 𝑣, which to
every point 𝑃 corresponds the point 𝑃 ′ = 𝑃 + 𝑣.

Proof. The proof follows from figure 30. For an arbitrary point 𝐴 and 𝑓1 (𝐴) = 𝐵, 𝑓2 (𝐵) = 𝐶
the segment 𝐴𝐶 is the double of 𝐷𝐸 in the common conjugate direction 𝛽. This defines
a vector of constant direction and constant length, hence proves the claim. □

ε'1 ε'2
A

ε1 D
β
B
ε2 E

Figure 30: Translation as a product of two affine reflections

Remark 10. Notice that the relevant ingredient in the representation of the translation
as a product of reflections is the “conjugate direction” 𝛽. In fact we can “turn” the two
parallels{𝜀1 , 𝜀2 } to a new position taking care only that their distance in the 𝛽 -direction
remains the same and equal to half the measure of 𝑣. Then the translation can be again
represented as a product of the two reflections w.r. to the two parallel lines in their new
position. Thus, turning 𝑣1 about 𝐷 in the new position 𝜀1′ and turning also 𝜀2 about
𝐸 in the new position 𝜀2′ so that it is parallel to 𝜀1′ , we obtain two other affine reflections
{ 𝑓𝜀1′ ,𝛽 , 𝑓𝜀2′ ,𝛽 }, whose composition is the same translation 𝑓 = 𝑓𝜀2′ ,𝛽 ◦ 𝑓𝜀1′ ,𝛽 by the double 𝑣
of the vector 𝐷𝐸. Theorem 24 handles the case of different conjugate directions.
 12 Dilatations 25

Remark 11. Translations build a normal subgroup of the group of all affinities of the plane.
The composition 𝑓 = 𝑓2 ◦ 𝑓1 of two translations by the vectors {𝑣 1 , 𝑣 2 } is the translation
by the vector 𝑣 1 + 𝑣 2 . This is not so for the set of reflections. The product of two reflec-
tions is not a reflection. The previous representation of a translation as a product of two
reflections gives a counterexample.

X Z

A B
Y

Figure 31: Translation as composition of two symmetries

Exercise 18. Show that the composition 𝑆𝐵 ◦ 𝑆𝐴 of two symmetries w.r. to the points {𝐴, 𝐵} is
the translation by the vector 𝑣 = 2𝐴𝐵 (See Figure 31).

12 Dilatations
dilatation Translations are a particular case of “dilatations”. Latter are affinities characterized by the
property to map every line 𝛼 to a line 𝛼′ parallel to 𝛼. A dilatation leaving at least one point
central fixed is called a “central dilatation”. The symmetry or half-turn about a point 𝑂 is obviously
dilatation a central dilatation.

Theorem 18. A central dilatation 𝑓 , different from the identity, has precisely one fixed point 𝑂,
called “center” and there is a constant 𝑘, called “ratio” of the dilatation, such that for every other
point 𝑃 ≠ 𝑂 it is
𝑓 (𝑃) = 𝑂 + 𝑘 · (𝑃 − 𝑂).

P'
Q'
P
Q
ε

O O'

Figure 32: Central dilatation

Proof. Assume first that besides the fixed point 𝑂 there is a second one 𝑂 ′ remaining
also fixed under 𝑓 . Consider also a line 𝜀 through 𝑂. Since, per definition, 𝜀 maps
to a parallel 𝜀′ = 𝑓 (𝜀) containing also 𝑂, line 𝜀 remains invariant under 𝑓 : 𝑓 (𝜀) = 𝜀.
If a point 𝑃 ∈ 𝜀, 𝑃 ≠ 𝑂 were fixed by 𝑓 , then we would have three non-collinear fixed
points and 𝑓 would be the identity transformation, contrary to the hypothesis. Thus,
𝑃 ′ = 𝑓 (𝑃) ≠ 𝑃 and the lines {𝑂 ′𝑃, 𝑂 ′𝑃 ′} have the property 𝑓 (𝑂 ′𝑃) = 𝑂 ′𝑃 ′ without to be
parallel, contrary to the hypothesis for 𝑓 .
Having that, the claim about the constant 𝑘 is verified by selecting another point 𝑄
and its image 𝑄 ′ = 𝑓 (𝑄), which, by the previous reasoning is on line 𝑂𝑄. By assumption
the line 𝑃𝑄 maps then to a parallel 𝑃 ′𝑄 ′ and by Thales’ theorem 𝑂𝑃 ′/𝑂𝑃 = 𝑂𝑄 ′/𝑂𝑄,
completing the proof. □
 13 Shears 26

Exercise 19. If 𝑓 is a central dilatation and for a point 𝑋 , 𝑌 = 𝑓 (𝑋), show that the line 𝜀 = 𝑋𝑌
is invariant under 𝑓 i.e. 𝑓 (𝜀) = 𝜀.
Exercise 20. Show that if an affinity leaves invariant every line through the point 𝑂, then it is a
central dilatation with center at 𝑂.
Remark 12. Central dilatations coincide with usual euclidean “homotheties” and half-turns
or symmetries about a point 𝑂 of the plane are dilatations with ratio 𝑘 = −1.

A A'

B B'

Figure 33: Non-central dilatation is a translation

Exercise 21. Show that a non-central dilatation maps two points {𝐴, 𝐵} to {𝐴′ , 𝐵′}, such that
𝐴𝐵𝐵′𝐴′ is a parallelogram and therefore is a translation and has no fixed points.
Exercise 22. Show that a dilatation is completely determined by giving the images { 𝑓 (𝐴), 𝑓 (𝐵)}
of two points {𝐴, 𝐵}.
Exercise 23. Show that the dilatations build a normal subgroup of the group of all affinities of the
plane. Show also that the translations build a normal subgroup of the group of all dilatations. If 𝑇𝑣
denotes the translation by the vector 𝑣 and 𝑔 is a dilatation, what kind of affinity is 𝑔 −1 ◦ 𝑇𝑣 ◦ 𝑔?
Exercise 24. Show that if there is a point 𝑂 such that every line 𝑂𝑋 through 𝑂 maps via the
affinity 𝑓 to a parallel line 𝑂 ′𝑋 ′ through the point 𝑂 ′ ≠ 𝑂, then 𝑓 is a dilatation.
Hint: Consider the oriented segment 𝑣 = 𝑂 ′𝑂 and the corresponding translation 𝑡𝑣 . The
composition of affinities 𝑓 ′ = 𝑡𝑣 ◦ 𝑓 leaves invariant every line through 𝑂 and exercise
20 applies.

13 Shears
shear Shears are products 𝑠 = 𝑔 ◦ 𝑓 of two affine reflexions { 𝑓 , 𝑔} whose “mirrors” are identi-
cal. Figure 34 shows the image 𝑠(𝑃) = 𝑔( 𝑓 (𝑃)) of a point 𝑃 under such a composition
of transformations. Line 𝜀 is the common mirror of the two reflections. The segment
𝐴𝐵 defines the conjugate direction of 𝑓 and segment 𝐵𝐶 defines the conjugate direc-
tion of the affine reflection 𝑔. A fundamental property of this transformation is that the

P'' P
C

ε = ε'
N M A

P' B

Figure 34: Shear as a composition of two reflections

points of the mirror 𝜀 are left fixed by the shear and that the triangle 𝑃𝑃 ′𝑃 ′′ with vertices
{𝑃, 𝑃 ′ = 𝑓 (𝑃), 𝑃 ′′ = 𝑔(𝑃 ′)} has fixed angles. This implies that all these triangles resulting
 14 Strains 27

for various positions of 𝑃 are similar to each other. Thus, also the direction of the median
𝑀𝑃 ′′ is fixed and all triangles 𝑃𝑀𝑃 ′′ are also similar to each other. This enables a quick
construction of 𝑃 ′′ once the direction of the median MP” has been determined:
1. Project 𝑃 to the point 𝑀 on the mirror 𝜀 along the conjugate direction 𝐴𝐵 of 𝑓 .
2. Draw from 𝑀 parallel to the direction of the median and find its intersection 𝑃 ′′
with the parallel from 𝑃 to the mirror 𝜀.
This implies immediately that each parallel to 𝜀 is left invariant by the shear which, re-
stricted on this parallel, coincides with a translation by a vector of fixed length, depending
only of the distance of the parallel from 𝜀. Thus, moving 𝑃 on the parallel to 𝜀 through
that point gives a point 𝑃 ′′ = 𝑠(𝑃) such that 𝑃𝑃 ′′ is parallel to 𝜀 and has a constant signed
measure, which is the double of that of 𝑀𝑁 .
Notice that in the product representing the shear 𝑠 = 𝑔 ◦ 𝑓 the order plays a role and
𝑓 ◦ 𝑔 is different from 𝑠, latter defining the “inverse” shear 𝑠 −1 = 𝑓 ◦ 𝑔 .
From the similarity of all triangles 𝑃𝑃 ′𝑃 ′′ and all triangles 𝑃𝑀𝑃 ′′ follows that the
translation vector effecting the shear along the parallel to 𝜀 from 𝑃 is a constant multiple
of the distance of 𝑃 from 𝜀. This implies that in a coordinate system with x-axis identical
with 𝜀, the shear is described by a constant 𝑘 and a matrix in the form:
  
𝑥 ′ = 𝑥 + 𝑘 · 𝑦, ′ 1 𝑘
⇔ 𝑋 = 𝐴𝑋 with 𝐴= . (15)
𝑦 ′ = 𝑦. 0 1

Exercise 25. Show that two reflections with axes intersecting at a point 𝑃 and the same conjugate
direction 𝑢 define a shear with axis parallel to 𝑢 and passing through 𝑃.

Exercise 26. Given a shear 𝑓 , write it as a composition 𝑓 = ℎ ◦ 𝑔 of two reflections. How many
possibilities are there for this?

O x

Figure 35: Shear’s “canonical” representation

Exercise 27. Given a shear 𝑓 , show that selecting for x-axis its axis and an appropriate transversal
to it y-axis (See Figure 35), the transformation can be represented by the matrix
 
1 1
𝑓 : .
0 1

14 Strains
strain A “strain” is a transformation 𝑓 somewhat more general than an affine reflection. It can be
defined by describing three points and their images (theorem 4). For the first two points
{𝐴, 𝐵} we assume that they remain fixed by 𝑓 , so that the whole line 𝜀 = 𝐴𝐵 remains
also fixed by 𝑓 (See Figure 36). For the third point and its image {𝑃, 𝑃 ′ = 𝑓 (𝑃)} we assume
that their line is not parallel to 𝜀 = 𝐴𝐵. To describe the effect of 𝑓 on any other point X of
the plane, consider the intersection point 𝑄 of line 𝑃𝑃 ′ with 𝜀. Notice first that, by the
preservation of ratios, line 𝑃𝑃 ′ is invariant by 𝑓 , i.e. it maps onto itself. Then, for any
 15 Invariant pencils 28

P'
X'

ε Q Y Z QP'/QP = k

P X

Figure 36: Strain’s axis 𝜀 and ratio 𝑘

point 𝑋 and its image 𝑋 ′ = 𝑓 (𝑋) draw the parallel 𝑋𝑌 to 𝑃𝑃 ′ intersecting 𝜀 at 𝑌. By

the preservation of parallels by affinities and the constancy of 𝑌 follows that line 𝑋𝑌 is
also left invariant by 𝑓 . Hence 𝑋 ′ is on 𝑋𝑌. Using Thales’ theorem is then easy to see
that 𝑃𝑋 and 𝑃 ′𝑋 ′ intersect at a point 𝑍 of 𝜀 and 𝑋 ′𝑌/𝑋𝑌 = 𝑄𝑃 ′/𝑄𝑃 = 𝑘 is constant.
Thus the image 𝑋 ′ is found in this case by a simple recipe:
1. Draw from 𝑋 a parallel to 𝑃𝑃 ′.
2. Take 𝑋 ′ on this parallel so that 𝑌𝑋 ′/𝑌𝑋 = 𝑄𝑃 ′/𝑄𝑃, where 𝑌 is the intersection
point of 𝑋𝑋 ′ with 𝜀.
Taking the line 𝜀 for 𝑥−axis and the line 𝑃𝑃 ′ for 𝑦−axis the strain is represented by the
matrix  ′   
𝑥 1 0 𝑥
= . (16)
𝑦′ 0 𝑘 𝑦

C''
A' C'
A f

B C B'

Figure 37: Affinity = similarity + strain

Exercise 28. Any affinity 𝑓 can be written as a composition 𝑓 = ℎ ◦ 𝑔 of a strain or shear 𝑔

and a similarity ℎ.

Hint: ([MP65, p. 122]) Construct a triangle 𝐴𝐵𝐶 ′′ similar to 𝐴′𝐵′𝐶 ′ on the side 𝐴𝐵 and
consider the strain or shear 𝑔 fixing {𝐴, 𝐵} and sending 𝐶 to 𝐶 ′′. Then, 𝑓 = ℎ ◦ 𝑔, where
ℎ is the similarity sending 𝐴𝐵𝐶 ′′ to 𝐴′𝐵′𝐶 ′.

15 Invariant pencils
pencils A “pencil” of lines consists of all the lines passing through a point 𝑃, real or at infinity,
called “center” of the pencil. If 𝑃 is at infinity, then the pencil consists of all the lines
pencils parallel to a given direction. As remarked by Coxeter [Cox67, p.17] in his marvellous talk,
invariant for every affinity which is a shear, strain, translation or central dilatation there is a pencil of
under
lines, the members of which remain invariant under the affinity. A shear leaves invariant
affinities
all lines parallel to its axis. A strain leaves invariant all lines parallel to its conjugate axis.
A translation leaves invariant all lines parallel to the translation vector and a central di-
latation leaves invariant all lines through its center. Next theorem uses this property to
characterize axial symmetries and dilatations.
15 Invariant pencils 29

Theorem 19. An affinity 𝑓 , for which there is a pencil of lines 𝒫 , such that each member-line 𝜀
of 𝒫 remains invariant, is an axial affinity or a dilatation and vice versa.

Proof. We distinguish three cases: 1) 𝑓 has no real fixed points. 2) 𝑓 has precisely one
fixed point. 3) 𝑓 has at least two fixed points.

Q Q'

P P'

Figure 38: No fixed point for translation

1) If 𝑓 has no real fixed point, then two invariant lines {𝑃𝑃 ′ , 𝑄𝑄 ′} cannot intersect, since
their common point would be a fixed point for 𝑓 . Thus, pencil 𝒫 consists of parallel
lines. Take then points {𝑃, 𝑄} (See Figure 38) on two such lines and their images {𝑃 ′ , 𝑄 ′}
contained in the same lines. Then, {𝑃𝑄, 𝑃 ′𝑄 ′} cannot intersect at a finite point 𝑆. This,
because then, by Thales’ theorem, the ratios 𝑄𝑆/𝑄𝑃 = 𝑄 ′𝑆/𝑄 ′𝑃 ′ would imply that 𝑆 is a
fixed point of 𝑓 , contrary to the hypothesis. Thus, in this case 𝑃𝑃 ′𝑄 ′𝑄 is a parallelogram
and the transformation is a translation.
2) If 𝑓 has one only fixed point, then the lines of the invariant pencil 𝒫 must go through
that point. This because if they didn’t, then they would be parallel, since the intersection
of two invariant lines is a fixed point contained in both lines. Assuming then that the

P'
P Q'
Q

Figure 39: One only fixed point for central dilatation

invariant lines are parallel and do not pass through the fixed point, take points {𝑃, 𝑃 ′} on
two of them (See Figure 39) and their images {𝑄, 𝑄 ′} contained each in the same invariant
line. Then applying again Thales’ theorem, we see that the intersection 𝑂 = 𝑃𝑃 ′ ∩ 𝑄𝑄 ′ is
a real fixed point, except the case in which 𝑂 is at infinity. Later though cannot happen,
since then, as is easily seen, 𝑓 would be a translation, without real fixed points. Thus
𝑂 must be a real point and varying the locations of {𝑃, 𝑃 ′} on the same lines we obtain
infinite many fixed points {𝑂}, contrary to the hypothesis. Thus, in any case, assuming
that the invariant lines do not pass through the fixed point we land to contradiction. Now
assuming that the invariant lines pass through the same real point 𝑂 and arguing as in
theorem 18, we see that 𝑓 is a central dilatation.
3) If 𝑓 has at least two real fixed points, then the line of these points is fixed and the result
follows from theorem 11. □
16 Affinities defined by three elements 30

16 Affinities defined by three elements

Any affinity 𝑓 can be defined by giving two triangles {𝐴𝐵𝐶, 𝐴′𝐵′𝐶 ′} whose vertices are
𝑓 𝑓 𝑓
supposed to correspond under 𝑓 : {𝐴 ↦→ 𝐴′ , 𝐵 ↦→ 𝐵′ , 𝐶 ↦→ 𝐶 ′}. Replacing some of the
points with lines, we can analogously ask “if giving three elements and their images define an
affinity”. Here “elements” stands for points or lines. Obviously giving three lines {𝜆, 𝜇, 𝜈} in
general position and their images {𝜆′ , 𝜇′ , 𝜈′} also in general position, defines completely
an affinity. In fact, considering the intersections of pairs of these lines, this reduces to
the standard case of three points and their images. Next case leads also to a complete
determination.

Theorem 20. Two triples {(𝐴, 𝜇, 𝜈), (𝐴′ , 𝜇′ , 𝜈′)} of a point and two lines in general position
uniquelly determine an affinity 𝑓 mapping the first to the second.

B'
ν''
ξt
B
A'
D'
ξ C'
A Χ
D ξ'
C μ'
Z
μ ν ν'
Υ

Figure 40: Affinity defined by a point and two lines and their images

Proof. Suppose we have succeeded in constructing the affinity 𝑓 . Consider the intersec-
tions of an arbitrary line through 𝐴 : {𝜉 : 𝐶 = 𝜉 ∩ 𝜇 , 𝐷 = 𝜉 ∩ 𝜈} (see figure 40) and
the points 𝐵 = 𝜇 ∩ 𝜈, 𝐵′ = 𝜇′ ∩ 𝜈′. Certainly 𝑓 (𝐵) = 𝐵′ and to reduce the construction
to that of three points and their images, it suffices to find the intersections 𝐶 ′ = 𝜉′ ∩ 𝜇′ ,
𝐷 ′ = 𝜉′ ∩ 𝜈′ , where 𝜉′ = 𝑓 (𝜉). Since the ratio 𝑘 = 𝐴𝐷/𝐷𝐶 is known, it suffices to find the
line 𝜉′ through 𝐴′ which intersects {𝜇′ , 𝜈′} at points {𝐶 ′ , 𝐷 ′} such that 𝐴′𝐷 ′/𝐷 ′𝐶 ′ = 𝑘.
But this is easily achieved by taking an arbitrary point 𝑋 ∈ 𝜈′ and drawing the parallel
𝜈′′ to 𝜈′ such that the variable lines 𝜉𝑡 through 𝐴′ intersect it at points 𝑍 such that
𝐴′𝑋/𝑋𝑍 = 𝑘. The requested point 𝐶 ′ = 𝜈′′ ∩ 𝜇′ and 𝐷 ′ = 𝜈′ ∩ 𝐴′𝐶 ′. □
Next theorem shows that the situation is radically different if we are given two points
and a line and their images.

A A'
B B' D'

C
μ μ' C'
D

Figure 41: Affinities defined by two points {𝐴, 𝐵} and a line 𝜇 and their images
16 Affinities defined by three elements 31

Theorem 21. Two triples {(𝐴, 𝐵, 𝜇), (𝐴′ , 𝐵′ , 𝜇′)} of two points and a line in general position
either fail to determine an affinity mapping the first to the second or they define a double infinity of
affinities 𝑓 doing that transformation.

Proof. Suppose we have again succeeded in constructing the affinity 𝑓 (see figure 41). By
the preservation of ratios by affinities, if {𝐶 = 𝐴𝐵 ∩ 𝜇 , 𝐶 ′ = 𝐴′𝐵′ ∩ 𝜇′} the ratios must be
equal 𝐴𝐵/𝐵𝐶 = 𝐴′𝐵′/𝐵′𝐶 ′ = 𝑘. Thus, if this condition is not satisfied by the given data,
there is no affinity mapping the first triple onto the second.
On the other side, if this condition is satisfied, then selecting two arbitrary points
{𝐷 ∈ 𝜇 , 𝐷 ′ ∈ 𝜇′}, we see easily that the affinity mapping 𝐵𝐶𝐷 onto 𝐵′𝐶 ′𝐷 ′ satisfies
the requirements. □
Theorem 20 leads to the determination of an affinity from some other elements. First
we consider the case of three points on a conic.

Theorem 22. Consider three points {𝑂, 𝑂 ′ , 𝐷} on a non degenerate conic 𝜆. There is a unique
affinity 𝑓 mapping 𝑓 (𝑂) = 𝑂 ′ , leaving the point 𝐷 fixed and having the conic 𝜆 as a conic of
intersections.

λ
D
μ''
μ
B

μ' A μ'
O O' C
ε
J E

Figure 42: Affinity defined by three points on a conic

Proof. The theorem reverses the arguments used for the construction of 𝜆 in section 5.
The two points {𝑂, 𝑂 ′} define a chord 𝜇′ and two tangents {𝜇, 𝜇′′} at its endpoints (see
figure 42). By theorem 20 there is an affinity 𝑓 mapping the triple {𝜇, 𝜇′ , 𝐷} to {𝜇′ , 𝜇′′, 𝐷}.
It is then trivially seen that the given conic 𝜆 is an intersection conic of this affinity.
Figure 42 illustrates the construction of two triangles corresponding under the affinity.
Point 𝐴 ∈ 𝜇 can be selected arbitrarily and we take it to be the intersection 𝐴 = 𝐷𝐽 ∩ 𝑂𝑂 ′.
Then, by the method of theorem 20 we construct a secant 𝐷𝐵𝐶 such that 𝐷𝐵/𝐵𝐶 = 𝐷𝐴/𝐴𝐽.
The affinity is the one mapping the triangle 𝑂𝐴𝐽 to 𝑂 ′𝐵𝐶. □
Remark 13. The conic 𝜆 in figure 42 is uniquely determined by the triangle 𝐽𝑂𝑂 ′ and
the point 𝐷. In fact, the triangle defines a “bitangent pencil of conics” 𝜆𝑡 = 𝜇′2 − 𝑡 · 𝜇 · 𝜇′′,
passing through {𝑂, 𝑂 ′} and tangent there respectively to its sides {𝜇 = 𝑂𝐽, 𝜇′′ = 𝐽𝑂 ′}.
The distinguished conic 𝜆 of the pencil is the one passing through point 𝐷 and definingh
𝑡 through the equation 𝜇′2 (𝐷) − 𝑡 · 𝜇(𝐷) · 𝜇′′(𝐷) = 0.
From this remark we deduce the possibility to define four affinities associated to a
quadrangle. For this we distinguish one vertex of the quadrangle and let it play the role of
𝐷 , the remaining three vertices defining the triangle 𝐽𝑂𝑂 ′ as before (see figure 43). The
affinity associated to the vertex 𝐷 leaves this point fixed and maps {𝑂𝐽 ↦→ 𝑂𝑂 ′ ↦→ 𝐽𝑂 ′}.
16 Affinities defined by three elements 32

D
B
A O'
O C

J E

Figure 43: Affinities defined by a quadrangle

Remark 14. The determination of the affinity producing a given conic as its conic of inter-
sections allows the description of this conic by a simple procedure which uses paralles to
a fixed direction. Figure 44 illustrates the case. In this, starting with the points {𝑂, 𝑂 ′ , 𝐷}
of the conic and its tangents {𝐽𝑂, 𝐽𝑂 ′} we constructed the lines {𝜀 = 𝐷𝐽, 𝜀′ = 𝐷𝐵} corre-

Y X

λ
Z D
μ''
μ
ε'
ε
B
μ' μ'
O A O' C

J E

Figure 44: Determination of a conic through parallels

sponding under the affinity and passing through its fixed point 𝐷. Then, drawing from
an arbitrary point 𝑋 ∈ 𝜀 the parallel 𝑋𝑌, 𝑌 ∈ 𝜀′ to 𝐴𝐵 we obtain the line pair (𝑋𝑂, 𝑌𝑂 ′)
intersecting at a point 𝑍 of the conic.

X
Y

Z
D

B
ε ε'

O A O' C

J E

Figure 45: When the conic of intersections is a circle

17 The case of parabolic conics of intersections 33

Exercise 29. Show that if the conic 𝜆 of intersections of the affinity 𝑓 is a circle, then the affinity
is a rotational similarity, i.e. a composition of a rotation 𝐷(𝜙) and a homothety with the same
center 𝐷 (see figure 45).
Hint: The angles š
𝐽𝑂𝐴 = ›𝐶𝑂 ′𝐵 and 𝑂 › ›
′ 𝑂𝐷 = 𝐵𝑂 ′ 𝐷. Also equal are per construction the

ratios 𝐴𝐷/𝐴𝐽 = 𝐵𝐷/𝐵𝐶. This implies that the triangles {𝐴𝑂𝐷, 𝐵𝑂 ′𝐷} are similar and
the triangles {𝐽𝑂𝐴, 𝐶𝑂 ′𝐵} are also similar etc.

17 The case of parabolic conics of intersections

Here, starting with a given parabola 𝜆 and three points {𝑂, 𝑂 ′ , 𝐷} on it, we examine
the kind of affinity 𝑓 for which this coincides with the conic of intersections of 𝑓 . The
method of investigation is that of theorem 22 using two tangents {𝜇, 𝜇′′} at the points
{𝑂, 𝑂 ′} (see figure 46). Using cartesian coordinates with 𝑥− axis the tangent at the vertex

D'

μ' λ

μ''

J
A
Y
μ E ε
O' F
Z
G I
C H X
B
D

Figure 46: Affinity with parabolic conic of intersections

and 𝑦−axis the axis of the parabola we describe the points of the parabola through the
equation 𝑦 = 𝑘𝑥 2 for a constant 𝑘. By their definition the points 𝑋(𝑥, 𝑘𝑥 2 ) of the parabola
define lines {𝑋𝑂, 𝑋𝑂 ′} corresponding under the affinity. By the preservation of ratios
under affinities and the fact, that the line 𝜀 = 𝐷𝐼 through the fixed point 𝐷 of 𝑓 parallel
to the axis of the parabola is invariant under 𝑓 , results the basic relation for corresponding
points {𝑌, 𝑍 = 𝑓 (𝑌)} for 𝑌 ∈ 𝑂𝑋 and {𝐼 = 𝑂𝑋 ∩ 𝜀, 𝐻 = 𝑂 ′𝑋 ∩ 𝜀}:
𝑌 = (1 − 𝑡) · 𝑂 + 𝑡 · 𝐼 ⇒ 𝑍 = 𝑓 (𝑌) = (1 − 𝑡) · 𝑂 ′ + 𝑡 · 𝐻 . (17)
We denote the coordinates of corresponding points by {𝐷(𝑑, 𝑘𝑑 2 ), 𝑂(𝑜, 𝑘𝑜 2 ), 𝑂 ′(𝑜 ′ , 𝑘𝑜 ′2 )}
and use also the coordinate system {𝑥 1 , 𝑦1 } with axes parallel to the former and center at
𝐷. In this we find that:
𝑌1 = (1 − 𝑡)(𝑜 − 𝑑) , 𝑌2 = − 𝑘(𝑜 − 𝑑)(𝑡(𝑥 + 𝑜) − (𝑜 + 𝑑)) ,
′ ′ ′ ′
𝑍1 = (1 − 𝑡)(𝑜 − 𝑑) , 𝑍2 = − 𝑘(𝑜 − 𝑑)(𝑡(𝑥 + 𝑜 ) − (𝑜 + 𝑑)) .
Expressing 𝑍 in terms of 𝑌 we find their relation through a matrix, which expresses also
𝑓 w.r.t. the selected coordinate system and proves next theorem:
    
𝑍1 𝑜′ − 𝑑 1 0 𝑌1
= ′ − 𝑜) 1 . (18)
𝑍2 𝑜−𝑑 𝑘(𝑜 𝑌2
18 Equiaffinities as products of reflections 34

Theorem 23. If the conic of intersections of an affinity is a parabola, the affinity is a composition
of a shear and a homothety and is represented by a matrix which has a double eigenvalue but only
one corresponding eigenvector.

Notice that the constant ratio 𝑟 = (𝑜 ′ − 𝑑)/(𝑜 − 𝑑) is equal to 𝑟 = 𝐹𝐷/𝐺𝐷 = 𝐼𝐷/𝐻𝐷

(see figure 46). The affinity in this case is a composition of a shear and a homothety with
center 𝐷 and ratio 𝑟 (see section 19).

μ' λ
Z
μ
O'
I
β
ζ
X
ε α

O
Y

Figure 47: Equiaffinities are products of affine reflections

Figure 47 shows the conic 𝜆 of intersections of a composition 𝑓 = 𝑟2 ◦ 𝑟1 of two affine

reflections {𝑟1 , 𝑟2 } having respectively parallel axes {𝜀, 𝜁} and arbitrary non-parallel con-
jugate directions {𝛼, 𝛽}. It is 𝑍 = 𝑟2 (𝑌) , 𝑌 = 𝑟1 (𝑋) and {𝜇 = 𝑂𝑋 , 𝜇′ = 𝑂 ′𝑍} are the vari-
able lines through {𝑂, 𝑂 ′ = 𝑓 (𝑂)}. The conic is a parabola and 𝑓 is a shear. This can be
easily seen by identifying 𝜀 with the 𝑥−axis of a cartesian coordinate system, the parallel
𝜁 to 𝜀 assumed to intersect the 𝑦−axis at 𝑏0 . Representing the directions by unit vectors
{𝛼(𝑎 1 , 𝑎 2 ) , 𝛽(𝑏1 , 𝑏 2 )} we obtain the matrix representation of 𝑓 :
 ′     
𝑥 1 𝑘 𝑥 𝑣
𝑍 = = + 1 with
𝑦′ 0 1 𝑦 𝑣2
𝑎1 𝑏2 − 𝑎2 𝑏1 𝑏0 𝑏1
𝑘 = −2 , 𝑣1 = 2 , 𝑣 2 = 2𝑏0 ,
𝑎2 𝑏2 𝑏2
leading to next theorem.

Theorem 24. The composition 𝑓 = 𝑟2 ◦ 𝑟1 of two affine reflections with parallel axes and non-
parallel conjugate directions is a shear and its conic of intersections is a parabola with axis parallel
to these axes.

18 Equiaffinities as products of reflections

Equiaffinities are affinities preserving the area of triangles. By the discussion in section 2,
equiaffinities are characterized by a matrix representation with determinant | 𝐴| = 1. By
theorem 19, an equiaffinity preserving a pencil of lines must be a shear or half-turn or a
translation, which, by the discussion so far, are affinities representable as a product of two
affine reflections. Next theorem, whose proof I take from Coxeter’s lecture [Cox67, p. 17],
guarantees this property for every equiaffinity.

Theorem 25. Every equiaffinity is the product of two affine reflections.

Proof. For equiaffinities 𝑓 preserving a pencil of lines, this results from the previous re-
marks. If 𝑓 does not preserve any pencil of lines, then there is a triangle 𝐴𝐵𝐶 with
18 Equiaffinities as products of reflections 35

A M D

f f

B f N C

Figure 48: Equiaffinities are products of affine reflections

{𝐵 = 𝑓 (𝐴), 𝐶 = 𝑓 (𝐵)} (see figure 48). For point 𝐷 = 𝑓 (𝐶) it follows that the triangles
{𝐴𝐵𝐶, 𝐵𝐶𝐷} have the same area and a common base, hence their vertices {𝐴, 𝐷} are
on a parallel to 𝐵𝐶. If {𝑀, 𝑁} are the middles of {𝐴𝐷, 𝐵𝐶} consider the affine reflections
{ 𝑔 = 𝑀(𝐵𝐶), ℎ = 𝐶(𝐵𝐷)} (see section 9 for the notation):
𝑔 ℎ 𝑔 ℎ 𝑔 ℎ
𝐴 ↦→ 𝐷 ↦→ 𝐵, 𝐵 ↦→ 𝐶 ↦→ 𝐶, 𝐶 ↦→ 𝐵 ↦→ 𝐷.

Thus, the equiaffinity 𝑓 coincides with ℎ ◦ 𝑔 at three non collinear points, hence it coin-
cides with it everywhere and 𝑓 = ℎ ◦ 𝑔, thereby proving the theorem. □
Since the composition of an even number of affine reflections is an equiaffinity, it can
be refactored to a product of two reflections and this proves next corollary.

Corollary 23. A product of 𝑛 affine reflections can be refactored to a product of two, if 𝑛 is even,
or three affine reflections, if 𝑛 is odd.

Exercise 30. Show that an affinity 𝑓 can be represented as a composition 𝑓 = 𝑠 ◦ 𝑟2 ◦ 𝑟1 , with

two affine reflections {𝑟1 , 𝑟2 } and a shear or strain 𝑠. Conclude that every affinity is the product,
𝑓 = 𝑠 ◦ 𝑔, of an equiaffinity 𝑔 = 𝑟2 ◦ 𝑟1 and a shear or strain 𝑠. In some particular cases one or
more of these factors of 𝑓 can be the identity. Investigate these cases and see when they happen.

A1
μ2 μ1 C
A
A2
C2 A3

C1
B
B1

Figure 49: General affinity decomposition

Hint: Consider a triangle 𝐴𝐵𝐶 and its image 𝐴1 𝐵1 𝐶1 under 𝑓 . Define an affine reflection
𝑟1 with conjugate axis 𝐵𝐵1 and axis 𝜇1 arbitrary (See Figure 49). This maps 𝐴𝐵𝐶 onto
a triangle 𝐴2 𝐵1 𝐶2 . Then, define the affine reflection 𝑟2 with conjugate axis 𝐶1 𝐶2 and
axis 𝜇2 the median of 𝐵1 𝐶1 𝐶2 from 𝐵1 . This maps 𝐴2 𝐵1 𝐶1 onto 𝐴3 𝐵1 𝐶1 . Then, the
two triangles with common base 𝐵1 𝐶1 define a shear or strain fixing the line 𝐵1 𝐶1 and
mapping 𝐴3 to 𝐴1 .
19 Conjugacy for affinities with fixed points 36

19 Conjugacy for affinities with fixed points

“Conjugate” are called the affinities { 𝑓 , 𝑓 ′} for which there is a third affinity 𝑔, such that

𝑔 ◦ 𝑓 ◦ 𝑔 −1 = 𝑓 ′.

Conjugate affinities have the same geometric properties and, converesly, affinities with
the same geometric properties are conjugate. Writing the relation with corresponding
matrices, as in equation 2,
     
𝐺 ℎ 𝐴 𝑣 ′ 𝐴′ 𝑣 ′
𝑔: , 𝑓 : , 𝑓 : ,
0 1 0 1 0 1

the conjugacy condition takes the form

       
𝐺 ℎ 𝐴 𝑣 𝐺 −1 −𝐺 −1 ℎ 𝐴′ 𝑣 ′
· · = ⇔
0 1 0 1 0 1 0 1
   
𝐺𝐴𝐺−1 −𝐺𝐴𝐺−1 ℎ + 𝐺𝑣 + ℎ 𝐴′ 𝑣 ′
= ⇔
0 1 0 1
(1) 𝐺𝐴𝐺 −1 = 𝐴′ and (2) 𝐺𝐴𝐺−1 ℎ − ℎ = 𝐺𝑣 − 𝑣 ′. (19)
Thus, if the affinities are conjugate, then the matrix parts are also conjugate, but the con-
verse is not valid in general, since the second condition is not necessarily a consequence
of the first. For example, the identity and the translation 𝑇𝑣 by the vector 𝑣 ≠ 0 are not
conjugate. In this case 𝐴 = 𝐴′ = 𝐼 is the identity matrix and (1) is satisfied for any 𝐺, but
the second becomes 0 = −𝑣 which is not valid. Notice that, if the first condition holds,
then the second can be written:

𝐴′ ℎ − ℎ = 𝐺𝑣 − 𝑣 ′. (20)

If the affinity 𝑓 has no fixed point, then selecting a point 𝑋 and its image 𝑌 = 𝑓 (𝑋) and
composing 𝑓 ′ = 𝑇𝑣 ◦ 𝑓 with the translation by the vector 𝑣 = 𝑌𝑋 we get the affinity 𝑓 ′ ,
which has 𝑋 as fixed point. Taking the fixed point as origin of coordinates, the affinity
reduces to a representation in the form
 
𝐴 0
,
0 1

and two affinities fixing that point 𝑋 are conjugate if the corresponding matrices satisfy

(1) 𝐺𝐴𝐺 −1 = 𝐴′ and (2) 𝐺𝐴𝐺 −1 ℎ − ℎ = 0 ⇔ 𝐴𝐺 −1 ℎ = 𝐺 −1 ℎ . (21)

This proves the following theorem.

Theorem 26. Two affinities having an isolated fixed point at the orgin and corresponding to the
matrices {𝐴, 𝐴′} are conjugate, if and only if their matrices are conjugate.

Proof. In fact, if the matrices are conjugate, then obviously the affinities are conjugate (tak-
ing ℎ = 0 ). If the affinities are conjugate with a ℎ ≠ 0, then the second of the equations
(21) must be satisfied and 1 will be an eigenvalue of the matrices, which by corollary 16
means that the fixed point is not isolated, contradicting the hypothesis. Thus, if the affini-
ties are isolated the corresponding ℎ = 0 and the matrices are conjugate. □
20 Elliptic affinities 37

Thus, the “classification” of affinities having an isolated fixed point results from the
conjugate classes of the matrices 𝐴. These, in turn, are distinguished by the correspond-
ing eigenvalues and their multiplicities. If the eigenvalues are (conjugate) complex, i.e.
the characteristic polynomial of 𝐴

𝑝(𝑥) = 𝑥 2 + 𝑏𝑥 + 𝑐,

has 𝑏 2 < 4𝑐 and “complex roots”. Then, selecting an arbitrary vector 𝑢 ≠ 0 and taking the
basis of {𝑢, 𝑢 ′ = 𝐴𝑢}, we have

𝐴𝑢 = 𝑢 ′ , 𝐴𝑢 ′ = 𝐴2 𝑢 = −𝑏𝐴𝑢 − 𝑐𝑢,

since 𝐴 satisfies its characteristic polynomial 𝐴2 + 𝑏𝐴 + 𝑐𝐼 = 0. Hence, w.r. to the basis

{𝑢, 𝑢 ′} the matrix 𝐴 is conjugate to
 
′ 0 −𝑐
𝐴 = .
1 −𝑏

If the eigenvalues are real and distinct {𝑎, 𝑏} and both different from 1, then, changing to
the basis of two eigenvectors, the matrix 𝐴 is conjugate to the diagonal one
     
′ 𝑎 0 𝑎 0 1 0
𝐴 = = · ,
0 𝑏 0 1 0 𝑏

representing a product of “strains” for 𝑎, 𝑏 ∉ {0, 1}.

If the characteristic polynomial has a double root 𝑎 ≠ 1, then in the case of one only
independent eigenvector, there is a basis such that 𝐴 is conjugate to
   
′ 𝑎 1 1 1/𝑎
𝐴 = = 𝑎 for 𝑎 ≠ 0.
0 𝑎 0 1

This, because there is then a vector 𝑣 : (𝐴 − 𝑎𝐼)𝑣 = 𝑣 ′ ≠ 0 and (𝐴 − 𝑎𝐼)2 𝑤 = 0 for every
vector 𝑤, implying that in the basis of {𝑣 ′ , 𝑣} the matrix has the preceding representa-
tion. In this case the affinity is the product of a shear and a homothety and the conic of
intersections of the affinity is a parabola (see section 17).
In the case of multiplicity two, with two independent eigenvectors, 𝐴 is conjugate to
the diagonal matrix representing a “homothety” for 𝑎 ≠ 1.
 
′ 𝑎 0
𝐴 = .
0 𝑎

20 Elliptic affinities
“Elliptic” are called the affinities 𝑓 fixing an isolated point and having no invariant lines.
By the discussion in section 19, follows that their matrix representation 𝐴 has a charac-
teristic polynomial with complex roots 𝜒𝐴 (𝑥) = 𝑥 2 + 𝑏𝑥 + 𝑐, with 𝑏 2 < 4𝑐 :
   
𝐴 0 0 −𝑐
with 𝐴 = .
0 1 1 −𝑏

The eigenvalues of 𝐴 are in this case complex-conjugate:

√
′ ′ 𝑏 ′4𝑐 − 𝑏 2
𝛼 = 𝜇 + 𝑖𝜇 , 𝛽 = 𝜇 − 𝑖𝜇 with 𝜇=− and 𝜇 =
2 2
20 Elliptic affinities 38

leading to two complex conjugate eigenvectors {𝑢𝛼 = 𝑣 + 𝑖𝑤, 𝑢𝛽 = 𝑣 − 𝑖𝑤} and

 
𝐴𝑣 = 𝜇 · 𝑣 − 𝜇′ · 𝑤,
𝐴𝑢𝛼 = 𝛼 · 𝑢𝛼 , 𝐴𝑢𝛽 = 𝛽 · 𝑢𝛽 ⇔
𝐴𝑤 = 𝜇′ · 𝑣 + 𝜇 · 𝑤.

Taking into account 𝜇2 + 𝜇′2 = 𝑑 2 = 𝑐 > 0 and setting

 
𝜇 𝜇′ 𝐴𝑣 = 𝑑(cos(𝜙) · 𝑣 − sin(𝜙) · 𝑤),
= cos(𝜙), = sin(𝜙) ⇒
𝑑 𝑑 𝐴𝑤 = 𝑑(sin(𝜙) · 𝑣 + cos(𝜙) · 𝑤),

we see that the affinity in this case is the composition 𝑓 = ℎ ◦ 𝑔 of an equiaffinity 𝑔 and a
homothety ℎ. These two affinities are represented in the frame of {𝑣, 𝑤} by matrices {𝐶, 𝐵}
and their composition by the product of matrices
   
𝑑 0 cos(𝜙) − sin(𝜙)
𝐴 = 𝐵·𝐶 = · .
0 𝑑 sin(𝜙) cos(𝜙)

Denoting by 𝑈 the affine transformation fixing the origin and mapping the standard
basis {𝑒1 , 𝑒2 } to {𝑣, 𝑤}, we see that the equiaffinity 𝑔 is conjugate to a rotation 𝑅 𝜙 , since
{𝑈 , 𝑅 𝜙 } can be identified with the matrices
   
𝑣1 𝑤1 cos(𝜙) sin(𝜙)
𝑈 = , 𝑅𝜙 =
𝑣2 𝑤2 − sin(𝜙) cos(𝜙)

and 𝑔 is represented w.r. to the standard base {𝑒1 , 𝑒2 } by the matrix

𝐺 = 𝑈 · 𝑅 𝜙 · 𝑈 −1 .

It is easy to see that 𝐺 leaves invariant the homothetic w.r. to the origin ellipses {𝑈(𝜅 𝑟 )},

κ
e2 κ'

w
v
Α
e1

Βφ Aφ

Figure 50: Equiaffinity leaving invariant a family of homothetic ellipses

where 𝜅 𝑟 denotes the circle 𝑥 2 + 𝑦 2 = 𝑟 2 (See Figure 50), mapping the orthogonal direc-
tions {𝑒1 , 𝑒2 } to two conjugate directions {𝑣, 𝑤} w.r. to the ellipses. The figure shows also
the images on the ellipse 𝑈(𝜅)

𝐵 = 𝑈(𝐴) and 𝐵 𝜙 = 𝑈(𝐴 𝜙 ) = 𝑈 𝑅 𝜙 (𝐴) = 𝑈 𝑅 𝜙 𝑈 −1 (𝐵) = 𝐺(𝐵).

of two points {𝐴, 𝐴 𝜙 = 𝑅 𝜙 (𝐴)} on the circle 𝜅.

Remark 15. The claim about the conjugate diameters is a simple consequence of the
preservation of middles of segments by affinities. Since in the circle two orthogonal di-
rections define diameters in the one direction whose middles are on a diameter parallel
to the second direction, the same property will hold with the images of these two families
of diameters by the affinities.
 21 Orbits of points under equiaffinities 39

21 Orbits of points under equiaffinities

In this and the following sections I study Coxeter’s ([Cox67, p.18]) classification of equi-
affinities, which he started with the following comment:
“Until two weeks ago, I thought the complete classification of equiaffinities might be
a rather awkward problem, but then (while sitting on a bench in Leicester Square,
London) I saw something very simple which I should have thought of long ago, ... ”
What Coxeter saw is the fact, used in section 18, that for every affinity 𝑓 , not possessing a
equiaffinity pencil of invariant lines (section 15), there is a point 𝐴 , such that {𝐴, 𝐵 = 𝑓 (𝐴), 𝐶 = 𝑓 (𝐵)}
invariant are non-collinear, and in the case of “equiaffinities” the point 𝐷 = 𝑓 (𝐶) is then on the par-
𝜆 = 𝐴𝐷
𝐵𝐶
Ε
F

ε A Μ

Ν D
G
I
P
B C
H
μ ν

Figure 51: The orbit of point 𝐴 lies on a conic

allel 𝜀 to line 𝐵𝐶 from 𝐴 (See Figure 51). Then the equiaffinity is characterized by the
ratio
𝐴𝐷
𝜆= (22)
𝐵𝐶
and the “orbit” under the action of 𝑓 of a point 𝐴 of the plane:
orbit
𝐴, 𝐵 = 𝑓 (𝐴), 𝐶 = 𝑓 2 (𝐴), 𝐷 = 𝑓 3 (𝐴), 𝐸 = 𝑓 4 (𝐴), . . . (23)
lies on a conic, which, depending on the value of 𝜆, can be an ellipse, parabola, hyper-
bola or a degenerate one. In the non-degenerate cases, the equiaffinity is correspondingly
named “elliptic, parabolic, hyperbolic, crossed-hyperbolic”, the last class relating to orbits that
have their points alternatively on a hyperbola and its conjugate. In the degenerate cases
the equiaffinity is called “focal” and the points of the corresponding orbit, as will be seen
below, alternate on two parallel lines.
In section 20, using analytic arguments we saw an “elliptic” case. The geometric ar-
gument applicable to any case and used by Coxeter is based on the fact that the orbit
𝐴𝐵𝐶𝐷..., which, in general, is an infinite polygon, is invariant not only under 𝑓 , but also
under the affine reflections { 𝑔 = 𝑀(𝐵𝐶), ℎ = 𝐶(𝐵𝐷)} whose axes are the lines {𝜇, 𝜈} and
which, by theorem 25, represent 𝑓 = ℎ ◦ 𝑔 . This is inductively seen by first noticing that
{ ℎ(𝐸) = 𝐴 ⇔ ℎ(𝐴) = 𝐸}, which follows from (See Figure 51):
ℎ(𝐴) = 𝐸 ⇔ ℎ(𝐴) = 𝑓 (𝐷) = ℎ(𝑔(𝐷)) ⇔ 𝐴 = 𝑔(𝐷),
implying that { 𝑔, ℎ} map the first four points of the orbit to points of the same:
𝑔(𝐴) = 𝐷 = 𝑓 (𝐶), ℎ(𝐴) = 𝐸 = 𝑓 (𝐷),
𝑔(𝐵) = 𝐶 = 𝑓 (𝐵), ℎ(𝐵) = 𝐷 = 𝑓 (𝐶),
𝑔(𝐶) = 𝐵 = 𝑓 (𝐴), ℎ(𝐶) = 𝐶,
𝑔(𝐷) = 𝐴, ℎ(𝐷) = 𝐵.
22 Classification of equiaffinities 40

Then, the first six points of the orbit map under { 𝑔, ℎ} to points of the same:

𝑔(𝐸) = 𝑔(ℎ(𝐴)) = 𝑓 −1 (𝐴) ⇒ 𝑔(𝐹) = 𝑔( 𝑓 (𝐸)) = 𝑔(ℎ(𝑔(𝐸)) = 𝑔(ℎ( 𝑓 −1 (𝐴)) = 𝑓 −2 (𝐴),

⇒ ℎ(𝐹) = ℎ(𝑔( 𝑓 −2 (𝐴)) = 𝑓 ( 𝑓 −2 (𝐴)) = 𝑓 −1 (𝐴), . . . and so on.

To show that the orbit is contained in a conic, consider the conic 𝜅 passing through
{𝐴, 𝐵, 𝐶} and having for diameters the lines {𝜇, 𝜈} (See Figure 51). By the assumed invari-
ance of 𝜅 w.r. to the affine reflections { 𝑔, ℎ}, the conic can be seen inductively to contain
also all other points of the orbit:

𝐷 = 𝑔(𝐴), 𝐸 = ℎ(𝐴), 𝐹 = 𝑓 (𝐸) = ℎ(𝑔(𝐸)) = ℎ(𝑔(ℎ(𝐴)), . . . and so on.

We summarize these facts in the form of a theorem:

Theorem 27. The orbit { 𝑓 𝑛 (𝐴), 𝑛 ∈ Z}, under the equiaffinity 𝑓 , for every point 𝐴 of the plane,
is invariant under 𝑓 and also invariant under the affine reflections { 𝑔, ℎ} representing 𝑓 = ℎ ◦ 𝑔.
It is also contained in a conic 𝜅, invariant under 𝑓 and invariant also under 𝑔 and ℎ.

Theorem 28. An affinity 𝑓 of finite period 𝑛 > 2 is necessarily an equiaffinity.

Proof. We notice first that the affinity has a fixed point, the centroid of an orbit. In fact, if
{𝑋1 , . . . , 𝑋𝑛 } is an orbit, then it is easily seen that 𝑂 = 𝑛1 (Σ𝑖 𝑋𝑖 ) satisfies 𝑓 (𝑂) = 𝑂. If 𝑂
is not isolated, then (section 8) the affinity is a strain or a shear and either is a reflection of
period 2 or has orbits with infinite many points, which are cases excluded by the hypoth-
esis. Thus, 𝑂 is an isolated fixed point and taking the origin at 𝑂, the hypothesis implies
that the matrix representing 𝑓 satisfies | 𝐴| 𝑛 = 1 and the translation vector 𝑣 = 0. Tak-
ing into account the discussion in sections 19, 20 and the resulting forms of the matrices
representing the corresponding cases of affinities, we see that the only possibility for 𝑓
to generate finite orbits is when it is elliptic with corresponding matrix
 
0 −1
𝐴 = ,
1 −𝑏

which represents an equiaffinity. □

22 Classification of equiaffinities
Continuining the discussion of the previous section, we examine how the kind of an orbit
{ 𝑓 𝑛 (𝐴), 𝑛 ∈ Z}, of a point under an equiaffinity and the kind of the conic 𝜅 containing

ε A Μ D
μ ν
Ν
Q D'
B P C

Figure 52: Center 𝑂 of the conic carrying the orbit. 𝜆 > 3 : hyperbolic rotation
 22 Classification of equiaffinities 41

such an orbit is influenced by the value of the signed-ratio invariant 𝜆 = 𝐴𝐷/𝐵𝐶. Draw-
ing the parallels {𝑀𝑄, 𝐷𝐷 ′} to 𝐴𝐵 from {𝑀, 𝐷} we see that the ratio

𝐴𝐷 𝐵𝐷 ′ 2𝐵𝑄 𝐵𝑄
𝜆 = = = = .
𝐵𝐶 𝐵𝐶 2𝐵𝑃 𝐵𝑃
Since {𝑀, 𝑁} are diameters of the conic 𝜅 containing the orbit of 𝐴, the intersection point
𝑂 = 𝜇 ∩ 𝜈 is the center of 𝜅. The location of 𝑂 depends on 𝜆 and figure 52 shows the

λ
-1 0 1 3

Figure 53: Location of the invariant 𝜆

hyperbolic case 𝜆 > 3. In this the center 𝑂 is on the other than 𝐴 side of 𝐵𝐶, 𝜅 is a hyperbola
rotation containing the entire orbit in one of its branches and the equiaffinity is called “hyperbolic
rotation”. In the case 𝜆 = 3 it is easily seen that the diameters {𝜇, 𝜈} are parallel, hence
parabolic the conic carrying the orbit is a parabola and the equiaffinity is called “parabolic rotation”
rotation (see figure 54).

μ ν
ε A D
Μ
Ν

Q D'
B P C

Figure 54: Center 𝑂 of 𝜅 at infinity. 𝜆 = 3 : parabolic rotation

Exercise 31. Show that the orbits of hyperbolic and parabolic rotations have infinite many points.

In the case −1 < 𝜆 < 3 the two diameters {𝜇, 𝜈} intersect on the side of 𝐵𝐶 con-
elliptic taining the points {𝐴, 𝐷}, the conic is an ellipse and the affinity is called “elliptic rotation”
rotation (see figure 55).

ν μ
ε A Μ D

O Ν

B P Q C D'

Figure 55: Invariant −1 < 𝜆 < 3 : elliptic rotation

Remark 16. Notice that the affinities of finite period are elliptic rotations, since they are
equiaffinities (see exercise 28) and their orbit is finite (see exercise 31).

Exercise 32. Show that an equiaffinity 𝑓 is of period 3 (theorem 16) precisely when 𝜆 = 0.
 22 Classification of equiaffinities 42

In the case 𝜆 < −1 the segments {𝐵𝐶, 𝐴𝐷} are inversely oriented and, by the discus-
sion in the previous section, 𝐴𝐷𝐵𝐸 is a trapezium and using the value of 𝜆 we see that
{𝐴𝐶𝐸, 𝐵𝐷} are a triangle, respectively a segment, inscribed in the conic with 𝐶 lying in
the interior of the trapezium. This can happen only if the conic is a hyperbola and the
two shapes are on different branches. Inductively, using the value of 𝜆 by means of the
analogous pairs { 𝑓 𝑛 (𝐴𝐶𝐸), 𝑓 𝑛 (𝐵𝐷), 𝑛 ∈ Z}, we see that the points of the orbit lie alterna-

ν ε D Μ A

Ν
O
D' Q B P C

Figure 56: Invariant 𝜆 < −1 : crossed hyperbolic rotation

crossed
hyperbolic tively on the two branches of a hyperbola (see figure 56). The equiaffinity in this case is
rotation called “crossed hyperbolic rotation”.
The case left 𝜆 = −1, leads to a degenerate conic of two parallel lines. In fact, it is then
easy to see that the orbit points lie alternatively on the two side-lines {𝐵𝐷, 𝐴𝐶} of the
focal parallelogram 𝐴𝐷𝐵𝐶 (see figure 57). The centers {𝑂, 𝑃} of the parallelogram, respec-
rotation tively of side 𝐵𝐶 are interchanged by 𝑓 which is called “focal rotation”. On each line the
orbit points are evenly distributed and successive points on each are at distance 2| 𝑂𝑃|

F
μ

D A
ν
O

C
B
P

Figure 57: Invariant 𝜆 = −1 : focal rotation

proceeding in opposite directions.

Thus, the equiaffinities, which do not preserve a pencil of lines (see section 15), quoting
Coxeter ([Cox67, p.23]):

“To sum up, the equiaffinities, other than translations, half-turns, and shears,
may be classified into the following types:” (see figure 58)
 23 Analytic description of orbits of equiaffinities 43

λ < -1 λ = -1 -1 < λ < 3 λ=3 λ>3

crossed hyperbolic focal elliptic Parabolic Hyperbolic

Figure 58: Classification of equiaffinities not preserving a pencil of lines

23 Analytic description of orbits of equiaffinities

For the analytic description of the orbits we can use the fact that affinities do not distin-
guish the conics of the “same kind” (see section 25). Thus, all ellipses are affinely equivalent
to the circle represented in appropriate coordinates by 𝑥 2 + 𝑦 2 = 1, all hyperbolas are
affinely equivalent to the hyperbola 𝑥 𝑦 = 1, and all parabolas are equivalent to the one
represented by 𝑦 = 𝑥 2 . Using such an affinity the various “affine rotations” are equivalent,
i.e. conjugate 𝑓 = 𝑔 ◦ 𝑓 ′ ◦ 𝑔 −1 by an affinity 𝑔, to standard affinities described by:

elliptic 1. “Elliptic rotations” corresponding to 𝑓 ′ : rotation by an angle 𝜙

rotation
 ′    
𝑥 cos 𝜙 − sin 𝜙 𝑥
= · , (24)
𝑦′ sin 𝜙 cos 𝜙 𝑦

the orbit consisting of the points

(1, 0) , (cos 𝜙, sin 𝜙) , (cos 2𝜙, sin 2𝜙) , (cos 3𝜙, sin 3𝜙) , . . .

the ratio 𝜆 being given, as expected −1 < 𝜆 < 3, since by the figure 59 and the
well known identity sin(3𝜙) = 3 sin(𝜙) − 4 sin3 𝜙 ([Hob18, p.52]):

𝐴𝐷 sin 3𝜙 3 − 4 sin2 𝜙
𝜆 = = = = 2 cos 𝜙 + 1.
𝐵𝐶 sin 2𝜙 − sin 𝜙 2 cos 𝜙 − 1

C
B
D

φ
A

𝐴𝐷
Figure 59: 𝜆 = 𝐵𝐶

hyperbolic 2. “Hyperbolic rotations” corresponding to 𝑓 ′ : hyperbolic rotation by 𝑡

rotation
 ′    
𝑥 𝑒𝑡 0 𝑥
′ = −𝑡 · , (25)
𝑦 0 𝑒 𝑦

the orbit consisting of the points

(1, 1) , (𝑒 𝑡 , 𝑒 −𝑡 ) , (𝑒 2𝑡 , 𝑒 −2𝑡 ) , (𝑒 3𝑡 , 𝑒 −3𝑡 ) , . . .

and 𝜆 having the value

𝐴𝐷 𝑒 3𝑡 − 1 𝑒 2𝑡 + 𝑒 𝑡 + 1
𝜆 = = 2𝑡 = = 2 cosh 𝑡 + 1 > 3.
𝐵𝐶 𝑒 − 𝑒𝑡 𝑒𝑡
 24 Remarks and exercises on orbits of equiaffinities 44

parabolic 3. “Parabolic rotations” corresponding to 𝑓 ′ given by

rotation  ′       
𝑥 1 0 𝑥 1 𝑥+1
= + = , (26)
𝑦′ 2 1 𝑦 1 2𝑥 + 𝑦 + 1

the orbit consisting of the points

(0, 0) , (1, 1) , (2, 4) , (3, 9) , . . .

and 𝜆 having the value

9−0
𝜆 = = 3.
4−1
crossed 4. “Crossed hyperbolic rotations” corresponding to 𝑓 ′ given by the commutative product
hyperbolic of the hyperbolic rotation and the half-turn (𝑥 ′ , 𝑦 ′) = (−𝑥, −𝑦)
rotation
 ′    
𝑥 −𝑒 𝑡 0 𝑥
= · , (27)
𝑦′ 0 −𝑒 −𝑡 𝑦

the orbit consisting of the points

(1, 1) , (−𝑒 𝑡 , −𝑒 −𝑡 ) , (𝑒 2𝑡 , 𝑒 −2𝑡 ) , (−𝑒 3𝑡 , −𝑒 −3𝑡 ) , . . .

and 𝜆 having the value

𝐴𝐷 −𝑒 3𝑡 − 1 𝑒 2𝑡 − 𝑒 𝑡 + 1
𝜆 = = 2𝑡 = − = 1 − 2 cosh 𝑡 < −1.
𝐵𝐶 𝑒 + 𝑒𝑡 𝑒𝑡

focal 5. “Focal rotations” corresponding to 𝑓 ′ given by

rotation  ′   
𝑥 −1 −1 𝑥
= , (28)
𝑦′ 0 −1 𝑦

with focal points {𝑃, 𝑂 = (± 21 , 0)} and orbit consisting of the points

(0, 1) , (−1, −1) , (2, 1) , (−3, −1) , . . . .

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

O
(0,0) P (1,0)

(-5,-1) (-3,-1) (-1,-1)

Figure 60: Orbit of a focal rotation

24 Remarks and exercises on orbits of equiaffinities

Here I collect some properties of affinities related directly or indirectly to the much cited
talk by Coxeter [Cox67]. First, we should notice that equiaffinities 𝑓 preserving a non-
degenerate conic 𝜅 are intimately related to orbits of points by equiaffinities. This follows
by considering a point 𝐴 ∈ 𝜅 and its orbit under 𝑓 , i.e. the iterates { 𝑓 𝑛 (𝐴), 𝑛 ∈ Z}. The
whole orbit must be a subset of the conic and, if the orbit has more than 4 points in general
 24 Remarks and exercises on orbits of equiaffinities 45

D' C' C1
D1 D C
κ κ0
A' B' A B Β1

Α1

Figure 61: Orbit {𝐴, 𝐵, 𝐶, 𝐷} of equiaffinity contained in many conics

position, this determines uniquely 𝜅.

Figure 61 shows a simple example of four points which are the vertices of a parallelogram
and represent also the orbit {𝐴, 𝐵 = 𝑓 (𝐴), 𝐶 = 𝑓 (𝐵), 𝐷 = 𝑓 (𝐶)} of an equiaffinity 𝑓 . There
is a “pencil of conics” passing through these four points, but from its infinite many non-
degenerate members, it can be shown that only one conic 𝜅0 is invariant under 𝑓 . This
is the image-conic 𝜅 0 = 𝑔(𝜅) of the circumcircle 𝜅 of a square 𝐴′𝐵′𝐶 ′𝐷 ′ transformed to
𝐴𝐵𝐶𝐷 by the affinity 𝑔 mapping the {𝐴′ , 𝐵′ , 𝐶 ′} respectively onto {𝐴, 𝐵, 𝐶}. The figure
shows also a second orbit {𝐴1 𝐵1 𝐶1 𝐷1 } of 𝑓 for a point 𝐴1 on 𝜅0 . The conjugate affinity
𝑓 ′ = 𝑔 ◦ 𝑓 ◦ 𝑔 −1 is the rotation by 𝜋/2 about the center of the circle 𝜅.
Second, we should notice also that a conic 𝜅 has plenty of affine reflections leaving
it invariant. Every diameter 𝛼 of the conic and its conjugate direction 𝑢 define such an
affine reflection.
Third, we should notice also that the two reflections { 𝑔, ℎ} to which decomposes an
equiaffinity (theorem 25) 𝑓 = ℎ ◦ 𝑔, which preserves the conic 𝜅, preserve this conic too.
This, for the case of affinities of infinite period follows from the discussion in section 21.
For orbits of affinities of finite period this follows from the fact that they occur only in the
elliptic type. But ellipses are affinely equivalent to circles and there the equiaffinities are
ordinary rotations, represented by two ordinary reflections, whose axes (mirrors) are at
angle half the angle of rotation.
These remarks lead to the next two exercises.

Exercise 33. An affinity leaving invariant a non-degenerate conic 𝜅 is either an equiaffinity or a

product of an equiaffinity and an affine reflection.

Hint: This can be quickly seen using matrices and representing the conic in cartesian ho-
mogeneous coordinates by an invertible matrix 𝐵 and the corresponding quadratic equa-
tion:
𝜅 : 𝑋 𝑡 · 𝐵 · 𝑋 = 0.
An affine transformation 𝑓 expressed through a matrix 𝐴𝑣 (notation of section 1) leaves
invariant the conic if the first condition below is valid, implying the second one of respec-
tive determinants:

𝐴𝑡𝑣 · 𝐵 · 𝐴𝑣 = 𝐵 ⇒ | 𝐴𝑡𝑣 || 𝐵|| 𝐴𝑣 | = | 𝐵| ⇒ | 𝐴𝑣 | 2 = 1.

Thus, if | 𝐴𝑣 | = 1 the transformation 𝑓 is an equiaffinity, otherwise composing with an

affine reflection 𝑔 preserving the conic, the transformation 𝑔 ◦ 𝑓 is an equiaffinity, thereby
proving the claim.

affinity Exercise 34. Every affinity 𝑓 preserving a conic 𝜅 is a product of 1, 2 or 3 affine reflections, each
preserving of them preserving the conic too.
a conic
 24 Remarks and exercises on orbits of equiaffinities 46

Exercise 35. Assume that the orbits {𝑋𝑛 = 𝑓 𝑛 (𝑋), 𝑛 ∈ Z} and {𝑌𝑛 = 𝑓 𝑛 (𝑌), 𝑛 ∈ Z} are con-
tained in the same non-degenerate conic 𝜅. Show that there is an affinity 𝑔 preserving 𝜅 and
mapping the first orbit onto the second.

Hint: Use again matrices, 𝐴𝑣 for 𝑓 and 𝐶𝑤 for the affinity 𝑔 mapping {𝑋0 , 𝑋1 , 𝑋2 } re-
spectively to {𝑌0 , 𝑌1 , 𝑌2 }. Since 𝐴𝑣 satisfies its characteristic polynomial, there are con-
stants {𝑎, 𝑏, 𝑐} such that, with the unit 3 × 3 matrix 𝐼3 :

𝐴3𝑣 = 𝑎𝐼3 + 𝑏𝐴𝑣 + 𝑐𝐴2𝑣 .

By the assumption

𝐶𝑤 𝑋𝑖 = 𝑌𝑖 ⇔ 𝐶𝑤 𝐴𝑣𝑖 𝑋0 = 𝐴𝑣𝑖 𝑌0 for 𝑖 = 0, 1, 2

we have then

𝐶𝑤 · 𝑋3 = 𝐶𝑤 · (𝐴3𝑣 𝑋0 ) = 𝐶𝑤 · (𝑎𝑋0 + 𝑏𝐴𝑣 𝑋0 + 𝑐𝐴2𝑣 𝑋0 ) =

𝑎𝑌0 + 𝑏𝐴𝑣 𝑌0 + 𝑐𝐴2𝑣 𝑌0 = 𝐴3𝑣 𝑌0 = 𝑌3 .

In the same way, using 𝐴4𝑣 = 𝑎𝐴𝑣 + 𝑏𝐴2𝑣 + 𝑐𝐴3𝑣 we show 𝐶𝑤 𝑋4 = 𝑌4 and inductively the
equivalent to the claimed condition 𝐶𝑤 𝑋𝑛 = 𝑌𝑛 for all 𝑛. The preservation of the conic
𝜅 under 𝑔 follows immediately if the orbits contain more than 4 points and by reduction
to the circle in the other cases.

X1 X4

X5 X
O

X2 X3

Figure 62: Areas swept by the radius from the center

area swept Exercise 36. Show that for a central conic 𝜅 and the orbit {𝑋𝑛 = 𝑓 𝑛 (𝑋), 𝑛 ∈ Z} in it by an
affinity 𝑓 preserving 𝜅 the areas of the triangles 𝑂𝑋𝑖 𝑋𝑖+1 are equal, 𝑂 being the center of the
conic. Show also that the areas of the corresponding sectors of the conic have the same area (see
figure 62) and also the corresponding “conic segments” enclosed by the chord 𝑋𝑖 𝑋𝑖+1 and the
corresponding arc of the conic are equal.

affinities Exercise 37. Show that an affine reflection preserving the circle 𝑆1 = {(𝑥, 𝑦) : 𝑥 2 + 𝑦 2 = 1} is an
preserving ordinary euclidean reflection (with conjugate axis orthogonal to the axis). Show that the product of
a circle
three reflections preserving 𝑆1 is a reflection too. Conclude that the group of affinities preserving
𝑆1 consists of “rotations” about its center and reflections on its diameters.
25 Affine equivalence of conics 47

25 Affine equivalence of conics

(Ι) (ΙΙ)
B
B
λy κ' κ
B'
A X
y λB

A' X'

O
κ κ'

Figure 63: Central conics 𝜅′ resulting via a strain from conic 𝜅

Analogously to “homotheties = central dilatations” centerred at 𝑂, represented in cartesian

coordinates by {𝑥 ′ = 𝜆𝑥, 𝑦 ′ = 𝜆𝑦}, which map central conics 𝜅 to central conics 𝜅′ with
the same axes and the same eccentricity, the use of affinities and in particular strains creates
maps between conics of the same type and of different eccentricities. Figure 63-I illustrates
such a case of a strain applied to hyperbola 𝜅, which maps to hyperbola 𝜅 ′ by “bending”
the original hyperbola 𝜅 while maintaining fixed its vertices, i.e. preserving the transverse
axis 𝑎 and changing only the conjugate axis 𝑏. It shows also how the point 𝐴 of 𝜅 and 𝐴′,
on the asymptote of 𝜅, map correspondingly to points 𝐵 of 𝜅′ and 𝐵′ of the asymptote
of 𝜅 ′. By varying the parameter 𝜆 of the strain, we can obtain from a hyperbola with
eccentricity 𝑒 a hyperbola with any other value 𝑒 ′ of eccentricity we wish. Equivalently,
from a hyperbola 𝜅 with angle between the asymptotes 𝜔, a hyperbola 𝜅 ′ with arbitrary
angle 𝜔′ between its asymptotes. Figure 63-II shows the analogous procedure mapping a
circle 𝜅 to an ellipse.

κ' κ (λx,λy)
(x,y)

Figure 64: All parabolas are similar

The analogous property for parabolas is trivial, since all of them are pairwise similar and
similarities are special affinities. Figure 64 shows the parabola 𝜅′ : 𝑦 = 𝜆1 𝑥 2 , resulting
from 𝜅 : 𝑦 = 𝑥 2 by the homothety 𝑥 ′ = 𝜆𝑥, 𝑦 ′ = 𝜆𝑦.
26 Affinities preserving a non-degenerate conic 48

26 Affinities preserving a non-degenerate conic

By exercise 34, every affinity preserving a non-degenerate conic is the composition of at
most three affine reflections. The first property to show is the following concerning a
single reflection.

Exercise 38. An affine reflection 𝑓 preserving a non-degenerate conic has necessarily its axis 𝜇
coinciding with a diameter of the conic and its conjugate direction 𝑢 coinciding with the conjugate
direction of 𝜇.

X Y Z
μ
u

X'
Y' Z'

Figure 65: Affine reflection preserving a conic

Hint: For all points 𝑋 on the conic 𝑋 ′ = 𝑓 (𝑋) defines chords 𝑋𝑋 ′ parallel to 𝑢 whose
middles lie on the conjugate diameter 𝜇 to 𝑢.

Exercise 39. The product 𝑓 of two affine reflections preserving an ellipse/parabola/hyperbola is

respectively an elliptic/parabolic/hyperbolic or crossed hyperbolic rotation.

Hint: Take a point 𝑋 on the conic and apply repeatedly 𝑓 to create the orbit { 𝑓 𝑛 𝑋} of
points of the conic (section 22).
Next exercise can be given a solution valid for all three kinds of conics. This depends
however on the possibility to rewrite an affine rotation 𝑓 as a product 𝑓 = 𝑓2 ◦ 𝑓1 of two
reflections whose axes and respective conjugate axes are conveniently chosen. For non-
euclidean rotations this is not trivial, but this sort of argument can be avoided by examin-
ing the cases separately. For the elliptic case this is already handled in exercise 37. Thus,
the solution can be given by examining the remaining cases: (i) of the parabola 𝑦 = 𝑥 2
and (ii) the case of the hyperbola 𝑥𝑦 = 1.

Exercise 40. The product 𝑓 of three affine reflections preserving a non-degenerate conic is an
affine reflection preserving the conic.

Hint: Show that this is equivalent with the following geometric property of quadrangles
inscribed in conics.

Exercise 41. Given three directions {𝑢1 , 𝑢2 , 𝑢3 }, a conic 𝜅 and an arbitrary point 𝐴 on it, con-
struct the inscribed in it quadrangle 𝐴𝐵𝐶𝐷, such that {𝐴𝐵, 𝐵𝐶, 𝐶𝐷} are respectively parallel to
{𝑢1 , 𝑢2 , 𝑢3 }. Show that the fourth side 𝐷𝐴 is then parallel to a fixed direction 𝑢4 .

Hint: In the case of the parabola 𝑦 = 𝑥 2 project the vertices on the x-axis to the points
respectively {𝐴′ , 𝐵′ , 𝐶 ′ , 𝐷 ′}. Use the description of the symmetry of the x-axis about
its point 𝑚, given by 𝑥 ′ = 2𝑚 − 𝑥 and show that the coordinates {𝛼, 𝛿} respectively of
{𝐴′ , 𝐷 ′} are related by the equation (See Figure 66)

𝛿 = 2(𝑚3 − 𝑚2 + 𝑚1 ) − 𝛼,
REFERENCES 49

B D

C
ε
A' M1 B' M2 M4 C' M3 D'

Figure 66: Inscribed quadrangles with fixed directions of three sides

{𝑚1 , 𝑚2 , 𝑚3 } being respectively the (fixed) coordinates of the middles of {𝐴𝐵, 𝐵𝐶, 𝐶𝐷}.
In the case of hyperbola and a chord 𝐴𝐵 of it such that 𝐴𝐵 = (𝛼, 𝛽) and coordinates
{𝐴(𝑥 1 , 𝑦1 ), 𝐵(𝑥2 , 𝑦2 )} show first that for fixed direction 𝜆 = 𝛼/𝛽 of the chords 𝐴𝐵
     
𝑥2 0 −𝜆 𝑥
= · 1 ,
𝑦2 − 𝜆1 0 𝑦1

describes the affine reflection with conjugate direction the one of 𝐴𝐵. Calculate the com-
position of three analogous reflections and show that the resulting transformation is an

A
y1

B
y2
C D
x1 x2
μ

Figure 67: Affine reflection with conjugate direction defined by 𝐴𝐵

affine reflection (See Figure 67).

References
[Cas93] John Casey. A treatise on the analytical geometry of the point, line, circle and conic
sections. Hodges Figgis and Co., Dublin, 1893.

[Cox61] H Coxeter. Introduction to Geometry. John Wiley and Sons Inc., New York, 1961.

[Cox67] H Coxeter. CUPM Geometry Conference. Math. Assoc. Amer. Washington DC,
1967.

[Cox87] H Coxeter. Projective Geometry. Springer, New York, 1987.

[ea83] H. Behnke et al. Fundamentals of Mathematics, vol. II, Geometry. MIT Press, Mas-
sachusetts, 1983.

[Hob18] E.W. Hobson. A treatise on plane trigonometry. Cambridge university press, Cam-
bridge, 1918.

[MP65] P. Modenov and A. Parkhomenko. Geometric Transformations I, II. Academic

Press, New York, 1965.

[Rya86] P. Ryan. Euclidean and Non-Euclidean Geometry. Cambridge University Press,

Cambridge, 1986.
REFERENCES 50

[Str88] Gilbert Strang. Linear Algebra and its Applications. Thomson Learning, 3rd Edi-
tion, London, 1988.

[Tar11] A. Tarrida. Affine Maps, Euclidean Motions and Quadrics. Springer, Heidelberg,
2011.

Related topics
1. Menelaus’ theorem

2. Quadratic equations in the plane

3. Projective plane

4. Homotheties and Similiarities

Any correction, suggestion or proposal from the reader, to improve/extend the exposition, is welcome
and could be send by e-mail to: pamfilos@[Link]