(Problem Books in Mathematics) Răzvan GelcaIonuţ OnişorCarlos Yuzo Shine - Geometric Transformations-Springer, Cham (2022)

Problem Books in Mathematics
Răzvan Gelca
Ionuţ Onişor
Carlos Yuzo Shine
Geometric
Transformations
Series Editor
Peter Winkler
Department of Mathematics
Dartmouth College
Hanover, NH
USA
Books in this series are devoted exclusively to problems - challenging, difficult, but
accessible problems. They are intended to help at all levels - in college, in graduate
school, and in the profession. Arthur Engels "Problem-Solving Strategies" is good
for elementary students and Richard Guys "Unsolved Problems in Number Theory"
is the classical advanced prototype. The series also features a number of successful
titles that prepare students for problem-solving competitions.
More information about this series at https://link.springer.com/bookseries/714

Răzvan Gelca • Ionuţ Onişor • Carlos Yuzo Shine
Geometric Transformations
Răzvan Gelca Ionuţ Onişor
Mathematics and Statistics Colegiul National de Informatica Tudor
Texas Tech University Vianu
Lubbock Bucharest, Romania
TX, USA
Carlos Yuzo Shine

Colégio Etapa
São Paulo
São Paulo, Brazil
ISSN 0941-3502 ISSN 2197-8506 (electronic)

ISBN 978-3-030-89116-9 ISBN 978-3-030-89117-6 (eBook)
https://doi.org/10.1007/978-3-030-89117-6
Mathematics Subject Classification: 97G50, 97U40, 54H15, 51M04
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
In memory of Rodica Gelca
Preface
There is a true geometry which is not [. . . ] intended to be merely an illustrative form of

more abstract investigation. Its problem is to grasp the full reality of the figures of space,
and to interpret—and this is the mathematical side of the question—the relations holding
for them as evident results of the axioms of space-perception.
This thought was penned by Felix Klein in his notes on the Erlangen Program.
The Erlangen Program proposed a view of geometry that shifts the focus from
geometric objects to groups of transformations that act upon them. The accent
is placed on understanding the transformations and the properties of space they
preserve. Our book does embrace this perspective, but is more of a hybrid between
the old Euclidean geometry and Klein’s ideas. We let transformations act on the
entire configuration or just part of it, then exercise our intuition on the result, or we
recognize a geometric transformation hidden inside the configuration itself.
We have selected mostly Olympiad problems, because in mathematical
Olympiads the geometry of lines and circles is still alive. Elementary geometry
is a valuable instrument for building space-perception and offers probably the best
introduction to the concept of a group: a set of transformations that contains the
compositions and inverses of all of its elements. And familiarity with groups of
geometric transformations is nowadays a must, since they play a central role in
physics, both classical and quantum, and with it in geometry. With this book, we
want to shape the reader’s mind into thinking about geometry in motion as opposed
to the static view of Euclid’s Elements.
As for the method of proof, we use both the synthetic and the analytic, where
appropriate, though we put the accent on the first. To motivate this, we refer to
another quote from the Erlangen Program:
The distinction between modern synthesis and modern analytical geometry must no longer
be regarded as essential, inasmuch as both subject-matter and methods of reasoning have
gradually taken a similar form in both. [. . . ] Although the synthetic method has more to
do with space-perception and thereby imparts a rare charm to its first simple developments,
the realm of space-perception is nevertheless not closed to the analytic method, and the
formulae of analytic geometry can be looked upon as a precise and perspicuous statement
of geometrical relations.
vii
viii Preface
It is the “rare charm” of the synthetic method that we aim to reveal with most of
the problems; it is also unimaginable to practice geometry without a good spatial
intuition, and that is why we favor the synthetic method. However, the coordinate-
based approach is easier to generalize to other realms of geometry and to relate to
other areas of mathematics. Thus, in many problems we have included analytic and
synthetic solutions side by side, so that the reader can see how numbers and figures
interact. And many a time we integrate the analytic and the synthetic in the same
argument.
Besides building good geometric intuition, this text opens a window towards
other parts of mathematics, giving thus a mild introduction to groups of trans-
formations and illustrating how symmetry groups appear in combinatorics and
number theory. There is a description of circular transformations; they are useful
at elementary level in inversive geometry, and are an essential tool in contemporary
research on hyperbolic geometry in two and three dimensions. It is important to
point out that some groups of transformations present in the book, and their three-
dimensional counterparts, are the simplest examples of Lie groups, and have proved
essential in modeling classical and quantum physics.
There is a vast body of mathematics related to geometric transformations, too
voluminous to be enclosed in the confines of a single book. We had to be selective, so
we have decided to reduce our scope to just the Euclidean plane, and there, to those
transformations that can be modeled with complex coordinates. All transformations
appearing in this book can be placed within the framework of complex affine
transformations and complex linear fractional transformations, and their conjugates.
There is no discussion of two-dimensional real projective geometry (such as conics)
or of two-dimensional real affine geometry.
To teach the tools and tricks of geometric transformations, we apply the following
structure to each of the first three chapters (“Isometries,” “Homotheties and Spiral
Similarities,” and “Inversions”). We begin with a discussion of theoretical results,
followed by a few theoretical questions. Next, several applications of the methods
are explained in detail, including some classical theorems in Euclidean geometry.
This is followed by what you, the reader, await with excitement: problems to
solve. The problems are listed in some increasing order of difficulty, but to keep
the element of surprise and to stimulate ingenuity, there is no grouping based on
common ideas. If the challenge is too big, hints for all problems can be found
in the middle of the book. Additionally, to help with the learning, all problems
have detailed solutions at the end of the book, often multiple solutions, some
of which have been discovered by experienced problem solvers. Even if you are
successful in solving a problem, and it would be good if you could explore and find
more approaches, you should always read the solutions from the end of the book.
Not only because they might teach you new tricks, but also because they contain
commentaries about the method, and for problems whose authors and sources are
known, these are mentioned there. The last chapter, “A Synthesis,” shorter but more
challenging than the others, puts all transformations from this book on common
ground, and contains problems that can be tackled with diverse techniques.
Preface ix
Geometric Transformations was carefully crafted by three experienced mathe-

matical Olympiad coaches on three continents. Răzvan Gelca has trained the United
States International Mathematical Olympiad Team for many years, and has also
served as deputy leader of this team, as well as leader of the team that represented
the United States at the Romanian Master of Mathematics. He has helped organize
the USA Mathematical Olympiad. Ionuţ Onişor has coached Romanian students
for the International Mathematical Olympiad and has taken Romanian teams to
international mathematics competitions. Carlos Yuzo Shine has been the head
coach of the Brazilian International Mathematical Olympiad Team as well as the
academic chair of the Brazilian Mathematical Olympiad, and has taken several times
the Brazilian team to the International Mathematical Olympiad. The authors have
benefited from the support, encouragement, and advice of their parents and of their
colleagues, as well as of Georgiana Onişor and Titu Andreescu, for which they are
deeply grateful. They have authored books on mathematical Olympiads in the past,
and they have now come together to write a book on geometry. As for the need of
such a book, Johannes Kepler once said: At ubi materia, ibi Geometria.1
Lubbock, TX, USA Răzvan Gelca
Bucharest, Romania Ionuţ Onişor
São Paulo, Brazil Carlos Yuzo Shine
1 Where there is matter, there is geometry.

Contents
Part I Problems
1 Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Theoretical Results About Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Definition and Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Translations, Rotations, Reflections. . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.3 Isometries as Composition of Reflections . . . . . . . . . . . . . . . . . . 16
1.1.4 Compositions of Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.1.5 Discrete Groups of Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.1.6 Theoretical Questions About Isometries . . . . . . . . . . . . . . . . . . . . 31
1.2 Isometries in Euclidean Geometry Problems . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.2.1 Some Constructions and Classical Results in
Euclidean Geometry That Use Isometries . . . . . . . . . . . . . . . . . . 32
1.2.2 Examples of Problems Solved Using Isometries . . . . . . . . . . . 46
1.2.3 Problems in Euclidean Geometry to be Solved
Using Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.3 Isometries Throughout Mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
1.3.1 Geometry with Combinatorial Flavor . . . . . . . . . . . . . . . . . . . . . . . 57
1.3.2 Combinatorics of Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
1.3.3 Number Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.3.4 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2 Homotheties and Spiral Similarities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.1 A Theoretical Introduction to Homotheties . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.1.1 Definition and Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.1.2 Groups Generated by Homotheties. . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.1.3 Problems About Properties of Homotheties . . . . . . . . . . . . . . . . 78
2.2 Problems in Euclidean Geometry That Use Homothety . . . . . . . . . . . . . 78
2.2.1 Theorems in Euclidean Geometry Proved Using
Homothety. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
xi
xii Contents
2.2.2 Examples of Problems Solved Using Homothety . . . . . . . . . . 88

Using Homothety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.3 Homothety in Combinatorial Geometry; Scaling . . . . . . . . . . . . . . . . . . . . 97
2.4 A Theoretical Study of Spiral Similarities. . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
2.4.1 The Definition and Properties of Spiral Similarities . . . . . . . 100
2.4.2 The Center of a Spiral Similarity: The Generic Case . . . . . . 102
2.4.3 The Center of a Spiral Similarity: The Case
A = B, Symmedians Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
2.4.4 Spiral Similarities and Miquel’s Theorem . . . . . . . . . . . . . . . . . . 107
2.4.5 Compositions of Spiral Similarities . . . . . . . . . . . . . . . . . . . . . . . . . 109
2.4.6 Groups Generated by Spiral Similarities. . . . . . . . . . . . . . . . . . . . 110
2.4.7 Theoretical Questions About Spiral Similarities . . . . . . . . . . . 113
2.5 Spiral Similarity in Euclidean Geometry Problems . . . . . . . . . . . . . . . . . . 113
2.5.1 Similar Figures and the Circle of Similitude . . . . . . . . . . . . . . . 113
2.5.2 Examples of Problems Solved Using Spiral Similarities. . . 120
Using Spiral Similarities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3 Inversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.1 Theoretical Results About Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.1.1 The Definition of Inversion and Some of Its Properties . . . . 129
3.1.2 Inverses of Lines and Circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
3.1.3 Möbius Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.1.4 Möbius Transformations Versus Isometries,
Spiral Similarities, and Inversions; Inversion and
Circular Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
3.1.5 Linear Fractional Transformations of the Real Line . . . . . . . 145
3.1.6 The Invariance of Angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
3.1.7 Inversion with Negative Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3.1.8 Circles Orthogonal to the Circle of Inversion. . . . . . . . . . . . . . . 153
3.1.9 The Limiting Points of Two Circles . . . . . . . . . . . . . . . . . . . . . . . . . 155
3.1.10 Problems with Theoretical Flavor About
Properties of Inversion and Möbius Transformations . . . . . . 158
3.2 Inversion in Euclidean Geometry Problems . . . . . . . . . . . . . . . . . . . . . . . . . . 159
3.2.1 Applications of Inversion to Proving Classical Results . . . . 159
3.2.2 Examples of Problems Solved Using Inversion . . . . . . . . . . . . 178
with Inversion (or with Möbius Transformations). . . . . . . . . . 198
4 A Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
4.1 Bringing Together All Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
4.1.1 Some Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4.1.2 Some Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Contents xiii
4.2 A Story of Complete Quadrilaterals

√ .................................. 234
4.2.1 Miquel’s Theorem and bc Inversion . . . . . . . . . . . . . . . . . . . . . . 234
4.2.2 Some Classical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
4.2.3 Problems About Complete Quadrilaterals . . . . . . . . . . . . . . . . . . 244
Part II Hints
5 Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
7 Inversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
8 A Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Part III Solutions

9 Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
11 Inversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
12 A Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
Part I
Problems
Chapter 1
Isometries
1.1 Theoretical Results About Isometries
Our discussion begins with those transformations that lie at the heart of the notion of
equality in geometry. We tend to identify two geometric figures and call them equal
(or congruent when we are very rigorous) if we can place one on top of the other to
coincide exactly. This means that we take one figure and move it across the plane
without breaking or deforming it until it overlaps the other figure. The motions that
we perform, to which we add the reflection into a mirror, are called isometries (from
the Greek words isos meaning “equal” and metron meaning “measure”). Figure 1.1
illustrates several instances of planar figures being mapped into one another by
isometries.
We have promised in the introduction to draw a parallel between the synthetic
and the analytic method. Because we work in the plane, we have two options
for coordinates: real or complex. There is logic in opting for the latter. Complex
numbers have been introduced for solving polynomial equations, as such they
come endowed with a multiplication, and this multiplication is somehow related
to the richness of Euclidean geometry. For us, the plane is identified with the set
C of complex numbers, and we convene to pass from synthetic to analytic by
replacing uppercase letters by lowercase letters: the point P in the Euclidean plane
becomes complex number p ∈ C (said differently p is the complex coordinate
of P ). Figure 1.2 shows an instance of how points in the plane acquire complex
coordinates. We will use the notation z and z for the real and the imaginary parts
of z and |z| and arg z for the absolute value and argument of z.
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 3

R. Gelca et al., Geometric Transformations, Problem Books in Mathematics,
https://doi.org/10.1007/978-3-030-89117-6_1
4 1 Isometries
Fig. 1.1 How figures are transformed under isometries
A a = 3 + 5i
B b = 5+i
Fig. 1.2 Passing from synthetic to analytic geometry
1.1.1 Definition and Basic Properties
It is desirable to reach a good level of understanding of the isometries before starting

to solve problems. For that reason we begin with a theoretical discussion.
Definition An isometry of the plane is a transformation that preserves distances
between points, meaning that if A maps to A and B maps to B , then the segments
AB and A B have equal lengths.
In complex coordinates, the distance between the points A and B is |a − b|. So,
by switching to analytic geometry, we have the following definition.
Definition An isometry in C is a function f : C → C such that
|f (z) − f (w)| = |z − w| for all z, w ∈ C.

1.1 Theoretical Results About Isometries 5
As we will see below, isometries preserve not just distances, but every property
that can be phrased in the language of Euclidean geometry. They lie behind the
notion of congruence, being the most natural transformations of Euclidean geom-
etry. We now embark on the task of understanding their structure and properties.
The first steps are easier in coordinates, as the functional equation |f (z) − f (w)| =
|z − w| can be solved explicitly, though with some effort, and yields a surprisingly
simple solution.
Theorem 1.1 The function f is an isometry of C if and only if f (z) = rz + s or
f (z) = rz + s, where r, s are complex constants with |r| = 1.
Proof If f (z) = rz + s then
|f (z) − f (w)| = |r(z − w)| = |r||z − w| = |z − w|;
while if f (z) = rz + a, then
|f (z) − f (w)| = |r(z − w)| = |r||z − w| = |z − w|,
so in each case f is an isometry.

Now suppose that f is an isometry. Let f (0) = s. Since 1 = |1 − 0| = |f (1) −
f (0)| = 0, if we set r = f (1) − f (0), then r = 0. We can define thus g such that
f (z) = r · g(z) + s, namely, g(z) = f (z)−s
r . Notice that

f (z) − f (w) |f (z) − f (w)|
|g(z) − g(w)| = =
= |z − w|,
r |r|
so g is an isometry, too. We have g(0) = 0 and g(1) = 1.

The equality |g(z) − g(0)| = |z − 0| is equivalent to |g(z)| = |z|, while |g(z) −
g(1)| = |z − 1| is equivalent to |g(z) − 1| = |z − 1|. Using the fact that |z|2 = z · z,
we can transform these two conditions into
g(z) · g(z) = z · z and (g(z) − 1) · (g(z) − 1) = (z − 1)(z − 1).
Subtracting the first relation from the second, we can transform this into
g(z) · g(z) = z · z and g(z) + g(z) = z + z.
We now deduce that {z, z} and {g(z), g(z)} are the solution sets to the same
quadratic equation (recall Viète’s relations for a quadratic equation), so the two
pairs must be equal. Hence {z, z} = {g(z), g(z)}, that is, for every z ∈ C, either
g(z) = z or g(z) = z.
It remains to prove that there do not exist nonreal numbers z, w such that g(z) =
z and g(w) = w simultaneously (we exclude z or w real because r = r for every
real r). Were this to happen, then
6 1 Isometries
|z − w| = |g(z) − g(w)| = |z − w|,
which means that z is at the same distance from w and w; alas, z lies on the
perpendicular bisector of w and w. But this perpendicular bisector is the real axis, so
z must be a real number, which is ruled out by our assumption. So either g(z) = z for
all z or g(z) = z for all z. From here we deduce that f (z) = rz+s or f (z) = rz+s,
and the theorem is proved.

A simple functional equation with an elegant algebraic solution. But in order
to decrypt the geometric meaning of the solution, we need more work. In the
next section, we will understand geometrically the multiplication by a number of
absolute value 1, the addition of a complex number, and the taking of the conjugate,
and this will shed light on the geometric interpretation of Theorem 1.1. This will
be a recurring theme: to discover the geometric meaning hidden inside algebraic
manipulations.
For the moment let us notice that, as an immediate consequence, we obtain the
following result.
Theorem 1.2 An isometry maps a segment to a segment equal to it, a line to a line,
a circle to a circle of the same radius, and a triangle to a triangle congruent to it.
An isometry also preserves angles.
Proof We only discuss the case f (z) = rz + s for some r, s ∈ C with |r| = 1; the
case f (z) = rz + s is left to the reader. In this case, for every a, b ∈ C and t ∈ R,
the image of ta + (1 − t)b is
rta + r(1 − t)b + s = t (ra + s) + (1 − t)(rb + s).
For t ∈ [0, 1], this means that the segment with endpoints a, b is mapped to the
segment with endpoints ra + s, rb + s, and note that |ra + s − rb − s| = |r||a − b| =
|a − b|. By allowing t to roam freely in R, we deduce that the line passing through
a and b is mapped to the line passing through ra + s and rb + s.
A complex number z satisfies the equation |z − a| = R if and only if it satisfies
the equation |(rz + s) − (ra + s)| = R (here we use the fact that |r| = 1). So z
belongs to the circle of center a and radius R if and only if f (z) = rz + s belongs to
the circle whose center is f (a) = ra + s and radius is R. In other words, the image
through an isometry of a circle is a circle of the same radius, and whose center is
the image of the center of the circle.
Finally, let ABC be a triangle, and let A , B , C be the images of ABC through
the isometry. Because the isometry maps AB to A B , BC to B C , and AC to
A C , the triangle A B C is the image of the triangle ABC and is congruent to it.
Therefore the angles of ABC and the angles of A B C are equal. Since every angle
can be placed in a triangle, isometries preserve angles.

By decomposing polygons into triangles, we obtain that an isometry maps a
polygon to a polygon congruent to it. It is not hard to see that all special points,
segments, lines, or circles (e.g., orthocenter, centroid, circumcenter, medians, angle
bisectors, altitudes, incircle) of a triangle are mapped to the same special points,
segments, lines, or circles of the image.
As a by-product of Theorem 1.1, we obtain that isometries are bijections from
the plane to itself. Indeed, the inverse of f (z) = rz + s is
1 s
f −1 (w) = w− ,
r r
and the inverse of f (z) = rz + s is
1 s
f −1 (w) = w− .
r r
We conclude that every isometry has an inverse, and the inverse is an isometry
as well; it is also immediate to see that the composition of two isometries is an
isometry, as distances are preserved (see Fig. 1.3). These properties characterize
one of the most fundamental concepts in mathematics, which we will now introduce
because it will allow us to formulate certain results from this book in a more concise
language and because a new concept always gives rise to new ideas.
Definition A group of transformations of a given set is a set G of bijective maps
from that set to itself with the property that the composition of any two elements of
G is in G and the inverse of every element of G is in G.
Isometries form a group of transformations of the plane.
f −1
g
f
g f
Fig. 1.3 Inverses and compositions of isometries

8 1 Isometries
1.1.2 Translations, Rotations, Reflections
In this section we introduce some particular classes of isometries.

Definition Given a vector − →v , the translation τ by the vector − →v is the transforma-
−−→ →
tion of the plane that maps a point A to the point A = τ (A) such that AA = −
v.
A translation is depicted in Fig. 1.4. If we write the translation vector −
→v = (a, b)
in complex coordinates as v = a + bi, then the translation is given by the equation
τ (z) = z + v.
This clarifies the geometric meaning of the transformation defined by adding a

complex number to the variable: it is the translation by the vector described by
that number. Translations are therefore particular cases of isometries.
You can immediately check that
• all line segments of the form XX are parallel and congruent;
−−→
• a single point A and its image A = τ (A) uniquely determine τ , since − →v = AA
is the only parameter of the translation;
• the line segments AB and A B are parallel and of equal lengths, consequently
translation maps a line to a line parallel to it;
• if AB is not parallel to −
→v and if −
→v = 0, then AA B B is a parallelogram.
Definition Given a point O and an angle α (measured counterclockwise), the
rotation about O by angle α is the transformation ρ of the plane that maps O to
itself and any other point A to a point A = ρ(A) such that AOA = α and A and
A are at the same distance from O.
Figure 1.5 shows an example of a rotation. You can check that
• the angle between AB and A B is α;
• the triangles OAB and OA B are congruent;
Fig. 1.4 A translation τ A = τ (A)

v
A
B = τ (B)
v
B
Fig. 1.5 A rotation ρ. The A = (A ) B

center is O and the angle is
α = 130◦
B = (B )
O
• the triangles OXX are all similar to an isosceles triangle with angles α, 90◦ −
α/2 and 90◦ − α/2;
• knowing the center of rotation O, a single point X, and its image X is enough to
determine the rotation; in fact, α = XOX ;
• when O is not given, two points A and B and their images A and B are
required in order to find the rotation; the center O lies on both the perpendicular
bisectors of AA and BB (since OA = OA and OB = OB ), so it lies at
their intersection, and if the two perpendicular bisectors coincide, then O is the
intersection of AB and A B ;
• if the rotation is not trivial, the center of rotation is its only fixed point.
Because the triangles OAB and OA B are congruent, the segments AB and
A B are equal. Consequently, rotations preserve distances; they are isometries.
When writing rotations in coordinates, we see the reason for using complex
numbers. To understand rotations, it is better to work with complex numbers in
trigonometric form
z = k(cos θ + i sin θ ).
This is because of the identity
arg(zw) = arg(z) + arg(w),
which follows from trigonometry
k1 (cos α + i sin α)k2 (cos β + i sin β)

= k1 k2 [(cos α cos β − sin α sin β) + i(cos α sin β + sin α cos β)]
= k1 k2 [cos(α + β) + i sin(α + β)].
Theorem 1.3 The rotation by α about z0 is then given by the formula
ρ(z) = r(z − z0 ) + z0 , where r = cos α + i sin α.

10 1 Isometries
Proof There is nothing to prove if α = 0. If α = 0, then the point z0 is the unique

fixed point of this transformation. The length of the segment joining z0 and z is
|z − z0 |, and the length of the segment joining z0 and ρ(z) is |r(z − z0 )| = |z − z0 |.
Also the segment joining z0 and z forms with the x-axis, an angle whose argument
is arg(z − z0 ), while the segment joining z0 and z forms with the x-axis, the angle
arg r(z − z0 ) = arg r + arg(z − z0 ) = α + arg(z − z0 ).
Thus the angle between the two segments is α, and the theorem is proved.

The formula for rotation can also be written as ρ(z) = rz + z0 (1 − r). On
the other hand, if f : C → C, f (z) = rz + s with |r| = 1 and r = 1, then
f defines a rotation; we can find the angle of rotation α = arg(r) and the center
z0 = s/(1 − r). The imaginary number i, which has been introduced for solving the
equation x 2 + 1 = 0, stands for the 90◦ rotation, and so it is related to the concept
of orthogonality in geometry.
There is an elegant way of looking at rotations which uses some elements of real
analysis. It is based on Euler’s number
1 1 1 1
e =1+ + + + + · · · = 2.71828 . . .
1! 2! 3! 4!
and the associated power series
x x2 x3 x4
ex = 1 + + + + + ··· .
1! 2! 3! 4!
This series converges to (i.e., approximates) a real number for all real numbers x.
But more is true. This definition can be extended to imaginary numbers, and we
have
α α2 α3 α4 α5
eαi = 1 + i− − i+ + i − ···
1! 2! 3! 4! 5!
Comparing this with two other series expansions
α2 α4 α6
cos α = 1 − + − + ··· ,
2! 4! 6!
α α3 α5 α7
sin α = − + − + ··· ,
1! 3! 5! 7!
where α is necessarily measured in radians, we deduce Euler’s formula
eαi = cos α + i sin α.

The reader uncomfortable with series expansions can take this as the definition of
eαi . The trigonometric identities
cos(α + β) = cos α cos β − sin α sin β, sin(α + β) = sin α cos β + cos α sin β
correspond to the multiplicative properties of the exponential function

n
e(α+β)i = eαi eβi , eαi = enαi .
A complex number is written in trigonometric form as
z = reiα , where r = |z| and α = arg z.
With this notation, we can write the counterclockwise rotation about the origin by
the angle α (measured in radians) as
z → z = eαi z.
Then the counterclockwise rotation about a point w by angle α is defined by the

equation
z − w
= eαi ,
z−w
which gives z = eαi z − eαi w + w.

To summarize, eiα can be used as a short-hand writing for cos α + i sin α, and
trigonometry dictates that it has the good multiplicative properties that powers have.
But you have to be careful and use radians, so the 60◦ rotation about the origin is
z → eπ i/3 z.
In the context of rotations, it is appropriate to open a parenthesis and discuss
directed angles modulo π (or modulo 180◦ if we use the old-fashioned Babylonian
measurements). Given two lines and , the angle of the rotation that maps into
is ambiguous; it can take two different values even in the interval [0, 2π ). There are
two ways of resolving this ambiguity. The first is to orient the lines, for example, by
choosing two points on each. As such, the angle (AB, CD) ∈ [0, 2π ) is the angle
−→
of the rotation that maps AB to CD such that the image of AB points in the same
−→
direction as CD. This is the convention that we adopt when we do not say anything
explicitly.
The second approach is to notice that two angles of rotation differ by multiples
of π , so the ambiguity disappears when working modulo π (see Fig. 1.6). Now the
angle (, ) is defined unambiguously. Note that ( , ) = − (, ).
Whenever we declare to be working with directed angles modulo π , when we
write ABC, we mean the angle (AB, BC) between the lines AB and BC, which
is the angle of the rotation that takes AB to BC, reduced modulo π .
12 1 Isometries
α α = β = γ(mod π)
Fig. 1.6 Definition of directed angles
If a, b, c are the complex coordinates of A, B, C, then the directed angle ABC

modulo π is
c−b
arg (mod π ).
a−b
The points A, B, C are collinear if and only if for some other point D we have
DAB = DAC, or in complex numbers, if and only if
b−a c−a
arg = arg (mod π ).
d −a d −a
In a triangle ABC,
BAC + CBA + ACB = (AB, AC) + (BC, AB) + (AC, BC) = 0.
Also, if A, B, C are not collinear, then a point D is on the circumcircle of ABC

if and only if, as directed angles modulo π , DAC = DBC. In complex numbers,
the condition that four points lie on a circle is

c−a c−b c−a c−b
arg − arg = arg : =0 (mod π ),
d −a d −b d −a d −b
that is,
c−a c−b
: ∈ R.
d −a d −b
And if A, B, C are on a line, then this is also the condition for D to lie on the same
line. The quantity
c−a c−b a−c b−c

: = : ,
d −a d −b a−d b−d
which we denote by (a, b, c, d) and call cross-ratio, seems to be important, if

only for checking that points are concyclic or collinear. Note that (a, b, c, d) =
(c, d, a, b) = (d, c, b, a) = (a, b, d, c)−1 . The cross-ratio depends only on the

distances between the points and the angles they form, so it does not depend on the
system of coordinates. We can therefore talk about the cross-ratio of four points in
the plane.
Before moving on, let us state what was explained above as a theorem.
Theorem 1.4 (Condition for Concyclicity) Four points A, B, C, D lie on a circle
or on a line if and only if the cross-ratio of their complex coordinates is a real
number.
Now let us return to our discussion of isometries.
Definition Given a point O in the plane, the reflection over the point O is the
transformation σ of the plane that maps O to itself and any other point A to a point
A such that O is the midpoint of AA .
A reflection over a point z0 in the complex plane is therefore a function σ :
C → C, such that σ (z)+z
2 = z0 . This clearly implies that σ (z) = −z + 2z0 . This
transformation does not bring anything new as it is the rotation about the point z0
by 180◦ .
Definition Given a line in the plane, the reflection σ over the line is the
transformation of the plane that maps a point A on to itself, and a point A that
does not belong to to a point A = σ (A) such that AA is perpendicular to , A
and A are in opposite half-plane determined by , and the distances from A and A
to are equal.
The reader can visually notice (on Fig. 1.7) that
• all segments XX have as perpendicular bisector;
• a single point X that does not lie on and its image X determine a reflection
over a line ; is the perpendicular bisector of XX ;
• the lines AB and A B are either parallel to or meet at a point on ;
• every reflection σ is an involution, that is, σ ◦ σ is the identity. This implies that
lines AB and A B are either parallel to or meet at a point on .
It is somewhat complicated to write reflections over lines in complex coordinates,
because one can write the equation of a line in many ways. We nevertheless have
the following result:
Theorem 1.5 Let f be an isometry. Then f is a reflection over a line if and only if
s
f (z) = rz or f (z) = − z + s
s
where r, s ∈ C, |r| = 1 and s = 0.

Proof The simplest reflection is over the real axis, σ (z) = z.
Next, let σ be the reflection over some line given by the parametric equation
z = z0 + vt, t ∈ R being the parameter and z0 and v being constants with |v| = 1.
14 1 Isometries
A = σ(A)
B = σ(B)
Fig. 1.7 A reflection σ over a line
We transform into the real axis by means of translations and rotations: first we
translate by −z0 , so z and σ (z) become z − z0 and σ (z) − z0 , respectively; the line
becomes given by z = t · v, t ∈ R. Then we rotate by − arg(v): becomes the
real axis, z − z0 becomes z−z v , and σ (z) − z0 becomes
0 σ (z)−z0
v . In other words, the
z−z0 σ (z)−z0
reflection over the real axis maps v to v . Therefore

σ (z) − z0 z − z0
=
v v
which yields, using the fact that v/v = v 2 , the following formula for the reflection:
σ (z) = v 2 z + z0 − v 2 z0 . (1.1)
So σ (z) must be of the form σ (z) = rz+s, |r| = 1. If we identify the coefficients
of σ (z), we find r = v 2 and z0 − v 2 z0 = s, that is, s = z0 − rz0 . Since z0 can be any
point of , the line must have the equation z−rz = s. Conjugation gives z−z/r = s
which is equivalent to z − rz = −rs, so s = −rs. This happens if either s = 0 or
r = −s/s. The case s = 0 gives σ (z) = v 2 z = rz, and the other case gives
σ (z) = − ss z + s.
Notice that we can recover the equation of the line from the formula for σ : it is
z − rz = s, which is equivalent to sz + sz = |s|2 , if s = 0, and z = rz if s = 0.
Notice also that passes through s/2 and that v is any square root of − ss , that is,
v = ±iω, ω = cos arg(s) + i sin arg(s).

As a consequence of the proof, we obtain the formula for the reflection over a
line determined by two points.
Proposition 1.6 Let a, b be two distinct complex numbers. Then the reflection of a
point z over the line determined by a and b is
a−b ab − ab
σ (z) = z+ .
a−b a−b
Proof The parametric equation of the line is z = a + (b − a)t. We have seen in the
proof of the previous result that

σ (z) − a z−a
= .
b−a b−a
Hence the formula.

Once reflections have been introduced, it is appropriate to talk about the
orientation of polygons, a concept that plays a major role in this book. A (nonskew)
polygon is oriented counterclockwise if, when reading the names of its vertices in
order, the interior is on the left, and it is oriented clockwise if, when reading the
names of its vertices in order, the interior is on the right. In Fig. 1.8, the triangle
ABC is oriented counterclockwise, and the triangle A B C is oriented clockwise. It
is not hard to see that rotations and translations preserve orientation, while reflection
changes orientation (e.g., in Fig. 1.8, the two triangles are images of each other
through the reflection over the vertical dotted line).
A A
B C C B
Fig. 1.8 Orientation of triangles

16 1 Isometries
1.1.3 Isometries as Composition of Reflections
It looks like the aforementioned transformations are only particular cases of

isometries of the plane and that there might exist more general types of isometries.
But Theorem 1.1 can now be read in the dialect of synthetic geometry, and it tells
that every isometry is either a rotation centered at the origin, a translation, the
reflection over the x-axis, a composition of a translation and a rotation centered
at the origin, or the composition of the reflection over the x-axis with a translation
and/or a rotation centered at the origin. We can do better than that, for example, we
can write all isometries of the plane in terms of reflections alone, as the next result
shows.
Theorem 1.7 Every isometry of the plane is the composition of two or three
reflections over lines.
Proof Let f be an isometry of the plane. Then f transforms any ray into a ray, by
preserving the order of the points, and every half-plane into a half-plane.
Let A and B be two points and let h be a half-plane bounded by the line AB. With
the standard notation, let A = f (A), B = f (B), and h = f (h). We distinguish
the following cases:
Case 1. The rays |AB and |A B coincide. Then A = A and B = B, so the entire
line AB is fixed by f . Then either the entire plane is fixed by f , in which case
we can write f as the square of any reflection, or f is the reflection over AB, in
which case f is the cube of this reflection, as well.
Case 2. The rays |AB and |A B are mapped into each other by a reflection σ over
a line. Compose f with this reflection to reduce the problem to Case 1. In this
situation f ◦ σ is either a reflection in which case f is the composition of this
reflection and σ or the identity map in which case f = σ = σ ◦ σ ◦ σ .
Case 3. The rays |AB and |A B do not coincide, nor do they map one to the other
by a reflection. Let σ1 be the reflection over a line that takes A to f (A) = A ,
and let B1 = σ1 (B). Let also σ2 be the reflection, over some line , that takes B1
to B . The two reflection lines are drawn as dotted in Fig. 1.9. Notice that is the
perpendicular bisector of B1 B , and since A B1 = AB = A B , A lies on , so
σ2 (A ) = A . It follows that σ2 ◦ σ1 (A) = A and σ2 ◦ σ1 (B) = σ2 (B1 ) = B .
Then σ2 ◦ σ1 takes A to A and B to B , so either f = σ2 ◦ σ1 (two reflections)
or f is the composition of σ2 ◦ σ1 and a reflection over A B (three reflections).

The theorem shows that reflections generate the group of isometries. Let us
observe that the composition of an even number of reflections preserves the orien-
tation of polygons, while the composition of an odd number of reflections changes
orientation meaning that what was originally counterclockwise, is now clockwise.
The isometries obtained as compositions of an even number of reflections form
therefore a group, the group of orientation preserving isometries. As a corollary of
Theorem 1.7 we obtain:
A
B1
A
B
Fig. 1.9 Case 3
Theorem 1.8 Every orientation-preserving isometry is the composition of two

reflections over lines.
But we have a better result:
Theorem 1.9 Every orientation-preserving isometry is either a translation or a
rotation.
Proof 1 Let f be an orientation-preserving isometry of the plane. First notice that
we only need to find a translation or a rotation that takes two points A and B to
A = f (A) and B = f (B), respectively. Indeed, if P is a point not belonging to
the line AB, then f (P ) is determined by A , B and the orientation of ABP . If P
is on the line AB, then the image P = f (P ) of a point P on line AB is uniquely
determined by the conditions P ∈ A B , P A = P A , and P B = P B .
Having in mind the technique of finding the center of a rotation, consider the
perpendicular bisectors a of AA and b of BB . We distinguish the following cases:
Case 1. (Fig. 1.10). The lines a and b have a non-empty intersection. Let O be
the intersection point of a and b. Then OA = OA , OB = OB , and, since
AB = A B , the triangles OAB and OA B are congruent. If these two triangles
have the same orientation, we have the following equalities of angles between
lines:
(OA, OA ) = (OA, OB) + (OB, OA )

= (OA , OB ) + (OB, OA ) = (OB, OB ).
From here we deduce that f is a rotation about O by the angle α = (OA, OA ).
If the two triangles have inverse orientations, then there is a reflection that takes
AB to A B , and so we must have a = b. If AB and A B are both parallel to
18 1 Isometries
a
a=b a=b
O A A O
A
B
A A
A
b B B
B B B
Fig. 1.10 Case 1
A A
B B
B
A A
B
Fig. 1.11 Case 2
a = b, there is a translation that takes A to A and B to B’; otherwise, AA B B

is an isosceles trapezoid with non-parallel sides AB and A B and a rotation with
center on the intersection of lines AB and A B does the job.
Case 2. (Fig. 1.11) The lines a and b do not intersect. Then a and b are parallel and
distinct. In this case AA and BB are also parallel. If A, A , B, B lie on the same
−−→ −−→ →
line and A and B are in the same order as A and B , then AA = BB = − v and
f is a translation by −→v ; if A and B are in reversed order as compared to A and
B , then the midpoints of AA and BB coincide, but then a ∩ b = ∅, which is
ruled out by the hypothesis.
On the other hand, if A, A , B, and B do not lie on the same line, they determine
a trapezoid. If the trapezoid is not a parallelogram, then AB and A B are either
its diagonals or the non-parallel sides. Either way, since AB = A B , the trapezoid
is isosceles and then a = b, a contradiction. So the trapezoid is a parallelogram,
−−→ −−→ → −−→ −−→
meaning that either AA = BB = − v and f is a translation by −
→v , or AA = −BB ,
and f is a reflection over the midpoint of AA and BB , which is a rotation by 180◦ .

Proof 2 There is a slick proof of the theorem using complex coordinates that the
reader might have noticed. Since conjugation reverses orientation, an orientation-
preserving isometry in C can only be of the form f (z) = rz + s, |r| = 1. If r = 1
we have a translation by s; if r = 1, we have a rotation by arg(z) about s/(1 − r).

Unfortunately, we cannot describe an orientation-reversing isometry as a single
reflection, but we still have the following result:
Theorem 1.10 Every orientation-reversing isometry is the composition of a unique
reflection over a line and a unique translation parallel to the line of reflection.
Proof 1 Let f be an orientation-reversing isometry, and let A = f (A) and B =
f (B) be the images of two distinct points A and B under f .
First we find a reflection σ over a line and a translation τ by a vector − →
v

parallel to such that f = τ ◦ σ . Again, it suffices to show that τ ◦ σ (A) = A and
τ ◦ σ (B) = B . Let M and N be the midpoints of AA and BB , respectively (if A
coincides with A, we take A = A as the midpoint).
If M = N (see Fig. 1.12), then is the line through M and N . Notice that
d(A, ) = d(A , ) and d(B, ) = d(B , ), so the reflection σ over takes
A to A1 and B to B1 such that A A1 and B B1 are both parallel to . Also, A
and A are on opposite sides of , and the same is true about B and B . Hence
(AB, ) = (, A B ), and thus, because of the reflection, lines A1 B1 and A B
are parallel. Since A1 B1 = A B , we find that A1 B1 B A is a parallelogram, and
−−→
thus there is a translation τ by the vector − →v = A1 A which is parallel to . We
conclude that f = τ ◦ σ .
If M = N (see Fig. 1.13), then ABA B is a parallelogram. In this case, we
take to be the line perpendicular to both AB and A B and passing through M.
Then σ takes A to A1 ∈ AB and B to B1 ∈ AB, and d(A1 , ) = d(A , ) and
d(B1 , ) = d(B , ), with A1 and A on the same side of , and B1 and B on the
Fig. 1.12 The case M = N B
M N
A
A1
B
B1
20 1 Isometries
Fig. 1.13 The case M = N

A B1 B A1
M =N
B A
same side of . Therefore, A1 A and B1 B are both parallel to , and we can define
−−→
τ to be the translation by A1 A .
Now we prove that the composition is unique. Let f = τ ◦ σ be a composition
of a reflection σ over a line and a translation τ parallel to . We have d(A , ) =
d(A, ) and d(B, ) = d(B , ), which means that both midpoints M of AA and
N of BB lie on . If M = N, we have = , that is, σ = σ . Now τ = f ◦σ −1 =
τ , and we are done. If M = N, then again ABA B is a parallelogram. Since τ is
a translation, the line through A1 = σ (A) and B1 = σ (B) is parallel to A B , so
A1 B1 is parallel to AB. However, a reflection maps a line to another parallel line if
and only if the lines are perpendicular or parallel to the line of reflection. The latter
case makes σ map AB to A B , but with the points A and B in reversed order, as
A1 and B1 , so we cannot translate A1 to A and B1 to B simultaneously; the former
case leads to σ = σ and, as before, τ = τ .

Proof 2 The complex number proof is again shorter, but it obscures completely
the geometric intuition. An orientation-reversing isometry f in C is of the form
f (z) = rz + s. Our goal is to find a complex t and a real number u such that f is a
composition of g(z) = −(t/t)z + t and h(z) = z + iωu, where ω = cos arg(t) +
i sin arg(t). We compute
h(g(z)) = −(t/t)z + t + iωu = −ω2 z + |t|ω + iωu.
Identifying the coefficients, we obtain r = −ω2 and s = |t|ω + iωu. From here
u and t can be uniquely determined in terms of r and s: pick ω as the square root
of −r that makes |t| = (sω−1 ) ≥ 0; then |t| is defined that way, and thus t is
uniquely determined; also, u = (sω−1 ) is uniquely determined.

We conclude this section with the following result:
Theorem 1.11 Let f be an isometry. Suppose that the points P1 , P2 , . . . , Pn on a
line are mapped to the points P1 , P2 , . . . , Pn on a line . Then the midpoints of
P1 P1 , P2 P2 , . . . , Pn Pn are collinear.
Proof 1 First, suppose that f reverses orientation. Since the midpoint of P and
f (P ) lies on the reflection line, then all midpoints of Pi Pi lie on the reflection line.
If f preserves orientation, compose f with the reflection σ over , that is,

consider f ◦ σ . Since σ () = , f ◦ σ maps Pi to Pi and reverses orientation,
and we apply the previous result to f ◦ σ instead.

Proof 2 This can also be seen with complex numbers: If f (z) = rz + s reverses
orientation, the midpoint of zi and f (zi ) is zi +f2 (zi ) = zi +rz
2
i +s
. Let ω be a square
root of r; then
zi + f (zi ) zi + ω 2 zi + s ωzi + ωzi s s

= =ω + = + ω(ωzi ),
2 2 2 2 2
and hence all midpoints lie on the line z = 2s + ωt, t ∈ R.

If f (z) = rz + s preserves orientation, then g(z) = z+f2 (z) is an affine function1
of z that takes to the set of midpoints of points from . Every affine function g
maps a line to another line. So the midpoints are on a line.

−−→ −−→
Remark This result is also true if we consider points Qi such that Pi Qi = t Pi Pi ,
t being a fixed real number. But this is a job for another transformation: homothety
(stay tuned!).
At this moment we understand well enough the structure of isometries them-
selves. The next section is devoted to understanding the outcome of composing
isometries.
1.1.4 Compositions of Isometries
It is clear that a composition of several isometries is an isometry itself, since it

preserves distances. In the previous section, we have learned that an isometry is
a rotation or a translation (when it preserves orientation) or the composition of
a unique reflection and a unique translation by a vector parallel to the line of
reflection (when it reverses orientation). We have also seen that if we interpret
points as complex numbers, the isometries are the functions defined by the formulas
f (z) = rz + s or f (z) = rz + s, |r| = 1.
Our next goal is to specify the outcome of all possible compositions of two
isometries of the plane. Let us first deal with the composition of two orientation-
preserving isometries. We already know that this composition preserves orientation,
and therefore it is either a translation or a rotation. Let us be more specific. We begin
with a result that was mentioned before.
Theorem 1.12 The composition of two translations τ1 and τ2 by − →u and − →
v is a
−
→
translation by u + v . −
→
1 Meaning a function of the form z → az + b.

22 1 Isometries
Proof Let A be any point in the plane, A1 = τ1 (A) and A = τ2 ◦ τ1 (A) = τ2 (A1 ).
−−→ −−→ −−→ → −
Thus AA = AA1 + A1 A = − u +→ v . Notice that this also proves that translations
commute, that is, τ2 ◦ τ1 = τ1 ◦ τ2 .

To summarize, translations form a commutative group: the compositions of the
−−−→
translations by vectors −
→v and −
→
w are the translation of vector v + w, and the inverse
−
→ −
→
of the translation of vector v is the translation of vector − v . The elements of this
group are parametrized by the complex numbers.
Theorem 1.13 Let ρ be the rotation about point O by angle α, and let τ be the
translation by −
→
v . Then
(i) the composition ρ ◦ τ is a rotation by α about the unique point O1 such that
O1 O = O1 O and O O1 O = α, where O = τ −1 (O). The point O1 lies
at the intersection of the perpendicular bisectors of OO and OO , where
O = ρ ◦ τ (O);
(ii) the composition τ ◦ ρ is a rotation by α about the unique point O2 such that
O2 O = O2 O3 and OO2 O3 = α, where O3 = τ (O). The point O2 is the
intersection of the perpendicular bisector of OO3 and the line O3 O ;
−−−→ →
(iii) if O1 and O2 are defined as above, then O1 O2 = − v.
Proof 1 The argument can be followed on Fig. 1.14. First we prove that both
compositions are rotations. To this end, let A = B be points in the plane, and
let A and B be the images of A and B, respectively, under either composition.
Because the angle between a segment and its translation is zero, and because
(AB, A B ) = α, neither of the two compositions can be a translation, both having
to be rotations by the angle α.
The rest is straightforward. We have
ρ ◦ τ (O ) = ρ ◦ τ ◦ τ −1 (O) = σ (O) = O,
Fig. 1.14 Composition of a O2 O

rotation and a translation and O3
vice versa
v v
α
O1 O
O
because every rotation fixes its center. It follows that O1 must be such that
O O1 O = α and O1 O = O1 O . In particular, O1 must lie on the perpendicular
bisector of OO . Because O = ρ ◦ τ (O), O1 must also lie on the perpendicular
bisector of OO , so (i) is proved.
The proof of (ii) is similar: τ ◦ ρ(O) = τ (O) = O3 , so O2 O = O2 O3 and
OO2 O3 = α. Also, O O3 O = 90◦ − α/2 = O2 O3 O, so the points O2 , O3 ,
and O are collinear.
For (iii), τ (O1 ) = τ (ρ ◦ τ (O1 )) = τ ◦ σ (τ (O1 )), so τ (O1 ) is a fixed point of
ρ ◦τ . However, the only fixed point of a rotation is its center, therefore τ (O1 ) = O2 .
−−−→ →
And this is equivalent to O1 O2 = − v.

Proof 2 In complex numbers, let ρ(z) = rz + s and τ (z) = z + v be a rotation and
a translation, respectively. Then ρ ◦ τ (z) = r(z + v) + s = rz + rv + s is a rotation
with center (rv + s)/(1 − r) = rv/(1 − r) + s/(1 − r), and τ ◦ ρ(z) = rz + s + v
is a rotation with center (s + v)/(1 − r) = s/(1 − r) + v/(1 − r). One can decrypt
from these numbers the geometric properties listed in the statement.

The composition of two rotations about the same center O is a rotation by the
sum of their angles (or the identity if this sum is zero). What if the centers are
different?
Theorem 1.14 Let O1 = O2 be points in the plane. The composition ρ2 ◦ ρ1 of two
rotations ρ1 about O1 by α and ρ2 about O2 by β is
(i) A rotation by α + β about the point O that is the intersection of the
perpendicular bisectors of O1 O1 , O1 = ρ2 (O1 ), and O2 O2 , O2 = ρ1−1 (O2 ), if
α + β is not a multiple of 360◦ ;
−−−→
(ii) A translation by −
→
v = O1 O1 , O1 = σ2 (O1 ) if α + β is a multiple of 360◦ .
Proof 1 By Theorem 1.9, the resulting composition is either a rotation or a
translation. Let A and B be distinct points in the plane, and define A1 = ρ1 (A),
B1 = ρ1 (B), A = ρ2 ◦ ρ1 (A) = ρ2 (A1 ) and B = ρ2 ◦ ρ1 (B) = ρ2 (B1 ). Then
(AB, A B ) = (AB, A1 B1 ) + (A1 B1 , A B ) = α + β. So if α + β is not a
multiple of 360◦ , we have a rotation by α+β (as all figures are rotated by this angle),
−→ −−→
and if α + β is a multiple of 360◦ , then AB = A B , and we have a translation. Now
we only need to find the parameters of the composition map.
If α + β is not a multiple of 360◦ , then we take advantage of the fact that
the rotation fixes its center, and compute ρ2 ◦ ρ1 (O1 ) = ρ2 (O1 ) = O1 and
ρ2 ◦ ρ1 (O2 ) = ρ2 ◦ ρ1 (ρ1−1 (O2 )) = ρ2 (O2 ) = O2 . So the center of rotation O lies
on the perpendicular bisectors of O1 O1 and O2 O2 . Notice also that since O1 = O2 ,
O1 = ρ2 (O1 ) = O1 and O2 = ρ −1 (O2 ) = O2 , the perpendicular bisectors are well
defined.
If α + β is a multiple of 360◦ , and we are in the presence of a translation, then
we only need to find the image of one point. We choose O1 , and then σ2 ◦ σ1 (O1 ) =
−−−→
σ (O ) = O . So the translation vector is −
2 1 1
→v = O O .
1 1

24 1 Isometries
Proof 2 For the analytical proof, if ρ1 (z) = r1 z + s1 and ρ2 (z) = r2 z + s2 are

rotations, ρ2 ◦ ρ1 (z) = r1 r2 z + r2 s1 + s2 is a translation by r2 s1 + s2 = ρ2 (s1 ) if
r1 r2 = 1 and a rotation by arg(r1 r2 ) about (r2 s1 + s2 )/(1 − r1 r2 ) otherwise.

We conclude that rotations do not form a group. The rotations about a fixed point
O do form a group (here you have to include the rotation of angle zero, which is the
identity map). This is the famous (multiplicative) group U (1) = {eαi | α ∈ R} that
plays an important role in electromagnetism (because eiα is written in terms of sine
and cosine, which are used for representing waves).
Now we turn to orientation-reversing isometries. Let us study first the composi-
tion of two reflections.
Theorem 1.15 The composition σ2 ◦ σ1 of two reflections σ1 , over line 1 , and σ2 ,
over line 2 , is
(i) the identity if 1 = 2 ;
(ii) a translation by a vector orthogonal to both 1 and 2 , directed from 1 to 2 ,
and length 2d(1 , 2 ) if 1 and 2 are parallel and distinct;
(iii) a rotation with center the intersection of 1 and 2 and angle 2 (1 , 2 ) if 1
and 2 are not parallel.
Proof 1 Part (i) is immediate. For (ii), arguing on Fig. 1.15, we introduce a signed
distance from a point P to 1 as being positive if P and 2 are in different half-
planes with respect to 1 and negative otherwise. Then we let d = d(1 , 2 ) be the
signed distance from any point of 2 to 1 . We have d(P , σ1 (P )) = 2d(P , 1 ) and
P
Q1
1
Q
P1
d( 1 , 2)
Fig. 1.15 Composition of two reflections over parallel lines

Fig. 1.16 Composition of 1

two reflections over P
non-parallel lines
α
O α P1
β 2
β
P
d(σ1 (P ), σ2 ◦σ1 (P )) = 2(d −d(P , 1 )). Therefore d(P , σ2 ◦σ1 (P )) = 2d(P , 1 )+

−−→
2(d −d(P , 1 )) = 2d, so if we let P = σ2 ◦σ1 (P ) then the vector P P is a constant.
If 1 and 2 are not parallel, they meet a unique point O (Fig. 1.16). Let P be any
point in the plane, P1 = σ1 (P ) and P = σ2 ◦ σ1 (P ) = σ2 (P1 ). We know that 1 is
the perpendicular bisector of P P1 , so OP = OP1 . Also, 2 is the perpendicular
bisector of P1 P , so OP1 = OP . Therefore, OP = OP , and moreover,
P OP = (OP , OP1 ) + (OP1 , OP ) = 2 (1 , OP1 ) + 2 (OP1 , 2 ) =
2 (1 , 2 ), which proves (iii).
Proof 2 For the complex number proof, if 1 and 2 are two lines that meet at z0 , we
can express their equations as z = z0 + tv1 and z = z0 + tv2 . So the reflections over
1 and 2 are, respectively, σ1 (z) = v12 z + z0 − v12 z0 and σ2 (z) = v22 z + z0 − v22 z0 .
Then
σ1 ◦ σ2 (z) = v12 σ2 (z) + z0 − v12 z0 = v12 (v2 2 z + z0 − v2 2 z0 ) + z0 − v12 z0

= v12 v2 2 z − v12 v2 2 z0 + z0 = z0 + v12 v2 2 (z − z0 ),
which is a rotation of angle arg(v12 v2 2 ) = 2 (1 , 2 ) about z0 .

If 1 and 2 are parallel, then we can suppose that 1 and 2 have equations
(z) = k and (z) = −k. For z = a + bi, we have σ1 (z) = a + (2k − b)i and
σ2 ◦ f1 (z) = a + (−2k − 2k + b)i = z − 4ki, which is a translation by −4ki.
Theorem 1.16 Two reflections σ1 and σ2 commute (i.e., σ1 ◦ σ2 = σ2 ◦ σ1 ) if and

only if σ1 = σ2 or the lines of reflection are perpendicular.
Proof 1 If σ1 ◦ σ2 = σ2 ◦ σ1 , then
(σ1 ◦ σ2 ) ◦ (σ1 ◦ σ2 ) = σ1 ◦ σ2 ◦ σ2 ◦ σ1 = σ1 ◦ 1 ◦ σ1 = σ1 ◦ σ1 = 1,
so applying σ1 ◦ σ2 twice yields the identity transformation. However, σ1 ◦ σ2

is a rotation, a translation, or the identity map (by Theorem 1.9). It cannot be a
translation because a translation by − →v applied twice is a translation by 2−
→
v . If
it is a rotation by α, then applying it twice yields a rotation by 2α. This is the
identity only if α = 180◦ , which means that the angle between the reflection lines
is 180◦ /2 = 90◦ . And if it is the identity map, then σ2 = σ1−1 = σ1 .

26 1 Isometries
Proof 2 The complex number proof is more involved. By Theorem 1.5, a reflection
is either of the form σ (z) = rz, |r| = 1 or σ (z) = − ss z + s for some s ∈ C, s = 0.
We look only at the situation where the reflections are of the latter form, which can
always be made the case by placing the origin off the lines. The composition of
reflections σ1 (z) = − ss11 z + s1 and σ2 (z) = − ss22 z + s2 is
s1 s1 s2 s1 s2
σ1 ◦ σ2 (z) = − σ2 (z) + s1 = z− + s1 .
s1 s1 s2 s1
The composition σ2 ◦ σ1 is obtained by interchanging the indices, so σ1 and σ2

commute if and only if
s2 s1 s1 s2 s2 s1 s1 s2
= and − + s2 = − + s1
s2 s1 s1 s2 s2 s1
and this is equivalent to
s2 s1 s2 s1 s1 s2
= ± and − + s2 = − + s1 .
s2 s1 s2 s1
If s2 /s2 = s1 /s1 , then
s1 s1 s2 s2
− + s2 = − + s1
s1 s2
is equivalent to −s1 + s2 = −s2 + s1 , and this of course is equivalent to s1 = s2 .

The latter is equivalent to σ1 = σ2 .
If s2 /s2 = −s1 /s1 , the second equation is an identity:
s1 s1 s2 s2
+ s2 = + s1 .
s1 s2
So s2 /s2 = −s1 /s1 is equivalent to 2 arg(s2 ) = ±π + 2 arg(s1 ), that is, arg(s2 ) =

±π/2 + arg(s1 ). This last relation means the lines of reflection are orthogonal.
Let us study what happens when we compose a reflection with a rotation and
then with a translation.
Theorem 1.17 Let ρ be the rotation about the point O by the angle α, and let σ be
the reflection over the line . Then
(i) ρ ◦ σ is the composition of a reflection and a translation. The reflection line
r passes through the midpoint O1 of the segment determined by O and O =
−−−→
σ (O) such that (, r) = α/2, and the translation vector is −
→
v = O1 O2 , O2
being the intersection of r and ρ();
(ii) σ ◦ ρ is the composition of a reflection and a translation. The reflection line
s passes through the midpoint O1 of the segment determined by O and O =
Fig. 1.17 Composition of a reflection and a rotation
−−−→
σ (O) such that (, s) = −α/2, and the translation vector is −
→
v = O3 O1 , O3
−1
being the intersection of s and ρ ().
Proof Because both ρ ◦ σ and σ ◦ ρ reverse orientation, both are compositions of a
reflection and a translation (Theorem 1.10). We need to find the reflection line and
the translation vector in each case. The reasoning can be followed on Fig. 1.17.
For ρ ◦ σ , let O = σ (O). Then
ρ ◦ σ (O ) = ρ ◦ σ ◦ σ (O) = ρ(O) = O,
because O is fixed by ρ. So, because in Theorem 1.10 the translation vector is

parallel to the reflection line, the reflection line passes through the midpoint O1 of
OO . Since O1 belongs to both and the reflection line r, the translation vector is
−−−→
O1 O2 , where O2 = ρ ◦ σ (O1 ) = ρ(O1 ). We want to better understand O2 . Because
O1 lies on , ρ(O1 ) lies on ρ(), and thus O2 = ρ(O1 ) is the intersection point of r
−−−→
and ρ() and − →v = O1 O2 . Finally, if A is any other point of , then the ray |O1 A is
rotated by α, so the reflection line is the angle bisector of (|O1 A, ρ(|O1 A)), and it
makes an angle of α/2 with . Since (, r) = α/2 = α = (, σ ()), r and ρ()
meet at a single point, so O2 is unique.
Now we examine σ ◦ρ. Notice that σ ◦ρ(O) = σ (O) = O , so again O1 belongs
to the reflection line s. Instead of looking for σ ◦ ρ(O1 ), we determine
(σ ◦ ρ)−1 (O1 ) = ρ −1 ◦ σ −1 (O1 ) = ρ −1 ◦ σ (O1 ) = ρ −1 (O1 ),
to be able to take advantage of the fact that σ (O1 ) = O1 . But O1 ∈ , so ρ −1 (O1 ) ∈

ρ −1 (), and O3 is the intersection point of s and ρ −1 (). By a similar argument as
for the proof of (i), we show that (, s) = −α/2.

With Theorem 1.10 in mind, let us find out what happens if we compose a
reflection with a translation that is not parallel to the reflection line.
28 1 Isometries
A1 B1 τ −1 ( )
v v⊥
A B
v⊥ v A v B
v A B τ( )
Fig. 1.18 Composition of a reflection and a translation
Theorem 1.18 Let τ be the translation by vector − →v , and let σ be the reflection
over the line , with and v not parallel. Let v = −
−
→ −
→ →
v + −v→ −
→
⊥ , with v parallel to
−
→
and v⊥ orthogonal to .
(i) The composition τ ◦ σ is a composition of a reflection over line 1 obtained by
translating by 12 −
v→ −
→
⊥ and a translation by v .
(ii) The composition σ ◦ τ is a composition of a reflection over line 2 obtained by
translating by − 12 −
v→ −
→
⊥ and a translation by v .
Proof Because both transformations τ ◦ σ and σ ◦ τ reverse orientation, both are

compositions of a reflection and a translation (Theorem 1.10). We need to determine
the reflection line and the translation vector for each case. Let A and B be two points
on (see Fig. 1.18).
We have τ ◦ σ (|AB) = τ (|AB), so the line is mapped to a line parallel to it.
The reflection line of τ ◦ σ is then the line that is halfway between and τ (), which
is obtained by translating by 12 − v→
⊥ . The translation vector is then the projection of
−−→
AA , where A = τ ◦ σ (A), onto . However, since A lies on , τ ◦ σ (A) = τ (A),
−−→ →
and AA = − v , so the translation vector is −
→
v .
For (ii), consider τ −1 (|AB), so σ ◦ τ (τ −1 (|AB)) = σ (|AB) = |AB. Thus the
reflection line of σ ◦τ is halfway between and τ −1 () and is obtained by translating
−−−→
by − 12 −
v→⊥ . The translation vector is again the projection of A1 A2 onto , where
−−−→ →
A1 = τ −1 (A) and A2 = σ ◦ τ (τ −1 (A)) = σ (A) = A. This means that A1 A2 = − v,
and so the translation vector is − →
v .

Notice that this theorem also proves that a translation and a reflection commute if
and only if the translation vector is parallel to the reflection line. With this at hand,
it becomes easier to compose an orientation-reversing isometry with some other
isometry. For instance, the composition of a rotation and an orientation-reversing
isometry is another orientation-reversing isometry equivalent to composing a rota-
tion, translation, and reflection, in order. This is the same as composing a rotation
(obtained by composing the rotation and the translation) and a reflection. We leave
the remaining cases to the reader.
In complex numbers, it is often easier to work directly with orientation-reversing
isometries as functions of the form z → rz + s and compose them with other
isometries. For instance, if f (z) = r1 z + s1 and g(z) = r2 z + s2 are two orientation-

reversing isometries, then f ◦ g(z) = r1 g(z) + s1 = r1 r2 z + r1 s2 + s1 , which is a
translation by v = r1 s2 + s1 if r1 r2 = 1 (which means r1 = r2 ), and a rotation by
arg(r1 r2 ) about (r1 s2 + s1 )/(1 − r1 r2 ) otherwise.
1.1.5 Discrete Groups of Isometries
There are some groups of geometric transformations of importance in combinatorial

geometry. We will examine those that come in handy when solving the combinato-
rial problems listed later in this chapter.
First, there are the groups of translations that keep invariant the tilings of the
plane by equilateral triangles, squares, or hexagons that are shown in Fig. 1.19. We
prove now that each of these groups is isomorphic to the group Z × Z endowed with
the composition law (m, n) + (m , n ) = (m + m , n + n ).
Theorem 1.19 Assume that the plane is tiled with equilateral triangles such that
πi
one of these triangles has vertices of complex coordinates 0, 1, e 3 . Then the group
of translations that maps this tiling to itself is isomorphic to Z × Z with the
isomorphism defined by τ1 → (1, 0) and τ2 → (0, 1) where τ1 is the translation by
the vector −
→
v1 defined by the complex number 1 and − →
v2 is the vector defined by the
πi
complex number e 3 .
Proof An arbitrary translation τ that maps the tiling to itself is determined by τ (T ),
πi
where T is the triangle with vertices 0, 1, e 3 mentioned in the statement. We want
to write
τ = τ1m τ2n .
The integer number n is uniquely determined by counting how many rows we go

up or down from the level of T to that of τ (T ). The integer number m is uniquely
determined by counting how many steps we go left or right from τ2n (T ) to τ (T ).
The isomorphism is τ1m τ2n → (m, n), and the theorem is proved.

Similarly one can prove the following two results:
Fig. 1.19 The tessellations of the plane by regular polygons

30 1 Isometries
Theorem 1.20 The group of translations that keeps invariant the tiling of the plane
by squares whose vertices are at the points of integer coordinates is isomorphic to
Z × Z with the isomorphism defined by τ1 → (1, 0) and τ2 → (0, 1) where the
translation vectors of τ1 and τ2 are −
→
v1 = (1, 0) and −
→
v2 = (0, 1).
Theorem 1.21 Assume that the plane is tiled with regular hexagons such that one of
πi πi
these hexagons has three consecutive vertices of complex coordinates e− 3 , 0, e 3 .
Then the group of translations that maps this tiling to itself is isomorphic to Z × Z
with the isomorphism defined by τ1 → (1, 0) and τ2 → (0, 1) where τ1 is the
√ πi
translation by the vector −
→
v1 defined by the complex number 3e 6 and − →
v2 is the
√ − πi
vector defined by the complex number 3e 6 .
A finite group that is omnipresent in applications is the one comprising the
rotations that map a regular polygon with n vertices into itself. It is not hard to
see (to boost your intuition use Fig. 1.20) that this group is
G = {1, ρ, ρ 2 , . . . , ρ n−1 },
where ρ is the rotation by 2π

n about the center of the polygon. This group is cyclic,
meaning that is has an element with the property that any other element is some
power of this element.
The last group that we will need is the group of symmetries of a nonsquare
rectangle (Fig. 1.21). This is a four-element group known as the Klein 4-group,
defined by
K = {a, b, c, e | a 2 = b2 = c2 = e, ab = ba = c, bc = cb = a, ac = ca = b}.
Here of course e is the identity element, namely, the identity transformation of the
rectangle, a is the reflection over a vertical axis, b is the reflection over a horizontal
axis, and c is the reflection over the center of the rectangle.
Fig. 1.20 The group of

rotations of a regular polygon ρ
Fig. 1.21 The Klein 4-group a
1.1.6 Theoretical Questions About Isometries
Our narrative continues with a number of questions of theoretical nature. Their aim
is to test and deepen your understanding of the structure of the group of isometries,
and to endow you with some useful lemmas. They will be followed in the next
section by more inviting problems about applications of isometries to Euclidean
geometry.
1 The triangles ABC and A B C are mapped into each other by the reflections
over two distinct lines. Prove that both triangles are equilateral.
2 Does there exist a set of points in the plane that has exactly two centers of
symmetry, meaning that it is mapped into itself by the reflections over exactly two
points?
3 Show that if f is an isometry such that f ◦ f = 1, then f is either the identity
map, the reflection over a point, or the reflection over a line.
4 Let O1 , O2 , . . . , O2n be points in the plane, and let σk be the reflection over Ok
for all k = 1, 2, . . . , 2n. Prove that σ1 ◦ σ2 ◦ · · · ◦ σ2n−1 ◦ σ2n is the identity map if
−−−→ −−−→ −−−−−−→ − →
and only if O1 O2 + O3 O4 + · · · + O2n−1 O2n = 0 .
5 Given the reflections σ1 , σ2 , σ3 , σ4 , σ5 over five points in the plane, prove that
σ1 ◦ σ2 ◦ σ3 ◦ σ4 ◦ σ5 = σ5 ◦ σ4 ◦ σ3 ◦ σ2 ◦ σ1 .
6 Let a, b, c be three distinct lines in the plane, and let σa , σb , σc be the reflections
over these lines. Prove that
σa ◦ σb ◦ σc = σc ◦ σb ◦ σa
if and only if the lines a, b, c either have a common point or are parallel to one
another.
7 Show that the composition of the reflections over three lines that intersect
pairwise at three distinct points has no fixed points.
8 Let a, b, c, d be four distinct lines, and let σa , σb , σc , σd be the reflections over
these lines, respectively. How are the lines arranged in the plane if
32 1 Isometries
σa ◦ σb ◦ σc ◦ σd = σb ◦ σa ◦ σd ◦ σc ?
9 Let 1 and 2 be two parallel lines, and let be a line perpendicular to them.
Denote by M the midpoint of the segment determined by the intersection points of
1 and 2 with , and consider A, B ∈ such that M is the midpoint of AB. Let a, b
be two lines parallel to 1 and 2 that pass through A and B, respectively. Denote by
σ1 , σ2 , σa , σb , σA , and σB the reflections over the lines 1 , 2 , a, b and the points
A, B, respectively. Prove that
σ2 = σb ◦ σ1 ◦ σa = σB ◦ σ1 ◦ σA .
10 Show that all axes of symmetry of a polygon intersect at one point. (An axis of
symmetry is a line such that the reflection over that line maps the polygon to itself.)
11 Using a straight edge and a compass, construct a pentagon knowing the
midpoints of its sides.
12 Two cars travel at equal speeds on two roads that are nonparallel straight lines.
Show that there is a point from which the two cars are, at any moment, at equal
distances.
1.2 Isometries in Euclidean Geometry Problems
Isometries are ubiquitous in Euclidean geometry. Many a time they show up

explicitly, such as when noticing that the orthocenter lies at the intersection of the
three reflections of the circumcircle over the sides, or that the rotation of a segment
by 60◦ around one of its endpoints gives rise to an equilateral triangle.
It is hence time to show how isometries come into play in the old-fashioned, but
beautiful realm of Euclidean geometry.
1.2.1 Some Constructions and Classical Results in Euclidean

Geometry That Use Isometries
We introduce below two objects from the geometry of a triangle that are constructed
using reflections over the angle bisectors of a triangle.
I. The Symmedians of a Triangle
Given two lines that pass through the same vertex of a triangle (i.e., two cevians of
the same vertex), these lines are called isogonal if they reflect into each other over
the bisector of the angle at that vertex.
1.2 Isometries in Euclidean Geometry Problems 33
Fig. 1.22 The definition of A

the symmedian
B M M C
Definition A symmedian of a triangle is the reflection of a median of the triangle

over the corresponding angle bisector.
So the median and symmedian of a given vertex provide a particular example
of isogonal lines. The definition of the symmedian is illustrated in Fig. 1.22, where
M is the midpoint of the side BC and the line AM is the reflection of the line
of support of the median AM over the angle bisector of A. A triangle has three
symmedians, one for each median. The next result shows how to construct them.
Theorem 1.22 (Construction of Symmedians) Let ABC be a triangle. Then the

symmedian from A passes trough the point where the tangents to the circumcircle at
B and C intersect.
We will give two proofs of this result:
Proof 1 Let T be the intersection of the tangents to the circumcircle at B and C, let
M be the midpoint of BC, and let α, β, γ denote the measures of the three angles
of the triangle. Examining Fig. 1.23, we notice that BM = MC and BT = T C.
We can put these equalities to work using the Law of Sines in the triangles ABM,
ACM, ABT , and ACT
AM BM AM CM
= , = ,
sin β sin BAM sin γ sin CAM
AT BT AT CT
= , = ,
sin(α + β) sin BAT sin(α + γ ) sin CAT
where for the last two equalities we used the fact that ABT = α + β and ACT =
α + γ.
Using the abovementioned equalities of segments, we obtain
sin BAM sin CAT

= .

sin CAM sin BAT
And we have BAM + CAM = CAT + BAT = α, so by denoting the

measures of CAM and BAT by x and y, respectively, we obtain
34 1 Isometries
Fig. 1.23 The construction A

of the symmedian
B M C
sin(α − x) sin(α − y)
= .
sin x sin y
Using the subtraction formula for sine, this further yields
sin α cot x − cos α = sin α cot y − cos α.
But the cotangent function is one-to-one on (0, π ), so this equality implies x = y,

that is, CAM = BAT , showing that AT is symmedian.

Proof 2 Draw a line P Q through T (P ∈ AB, Q ∈ AC) that is antiparallel to

BC, namely, such that ACB = AP Q and (consequently) ABC = AQP (see
Fig. 1.24). We have
T BP = 180◦ − ABC − CBT = 180◦ − ABC − BAC

= ACB = BP T
and similarly T CQ = T QC. It follows that the triangles T P B and T QC are

isosceles, and hence T B = T P and T C = T Q. But T B = T C, so T is the
midpoint of P Q. And the triangle AP Q reflects over the angle bisector of BAC
to a triangle AP Q whose side P Q is parallel to BC (the antiparallel becomes
parallel). In that case the image T of T under this reflection is on the line AM.
Consequently the line AT is the reflection of AM; it is the symmedian.

This construction of the symmedian leads naturally to the concept of a harmonic
quadrilateral. We say that four points A, B, D, C that appear in this order on a circle
form a harmonic quadrilateral if
AB CB
: = 1 ⇐⇒ AB · CD = AD · BC.
AD CD
Fig. 1.24 Proof of the A

characterization of
symmedians
C
B M
In complex coordinates, a quadrilateral A, B, C, D is harmonic if and only if
a−b c−b
: = −1.
a−d c−d
This is the condition that the cross-ratio (a, c, b, d) is equal to −1 (and a little
algebra shows that this is equivalent to (a, b, c, d) = −1).
Proposition 1.23 Given a triangle, the point where the symmedian intersects the
circumcircle forms with the vertices a harmonic quadrilateral. Conversely, given a
harmonic quadrilateral, in the triangle formed by three vertices, the diagonal that
is not side of the triangle is a symmedian.
Proof We argue on Fig. 1.25. Let ABC be the triangle, and let D be the intersection
of the symmedian from A with the circumcircle. For the direct implication, note
that from the similarity of triangles T BA and T DB ( DT B = BT A, DBT =

BAT = 1
2 BD), we obtain
BA TB
= ,
BD TD
and from the similarity of T CA and T DC, we obtain
CA TC
= .
CD TD
36 1 Isometries
Fig. 1.25 Symmedians give A

rise to harmonic
quadrilaterals
B C
Given that T B = T C, we obtain
BA CA
= ,
BD CD
so ABDC is harmonic.
For the converse, note that the condition for ABDC to be harmonic can be
rewritten as
AB DB
=
AC DC
so D is the second intersection point of the Apollonian circle determined by the
points A and B and the ratio k = AB/AC with the circumcircle ABC. Since this
intersection point is unique, it must be the point where the symmedian intersects the
circumcircle. We recall that the Apollonian circle determined by the points A and B
and the ratio k > 0 is the locus of the points P with the property that P A/P B = k
(the proof that this locus is a circle makes the object of Problem 165).

We will visit harmonic quadrilaterals once more in the chapter about inversion.
II. The Isogonal Conjugate of a Point
The definition of the isogonal conjugate of a point is based on the following result.
Theorem 1.24 (Isogonal Conjugate of a Point) Given a triangle, the reflections
over the angle bisectors of three concurrent cevians are three cevians that are also
concurrent.
In other words, if three cevians of a triangle intersect, then the reflections of these
cevians over the respective bisectors also intersect. If P and Q are the intersections
of the two groups of cevians, then P and Q are called isogonal conjugates, and each
point is said to be the isogonal conjugate of the other. As an example, the orthocenter
is the isogonal conjugate of the circumcenter (prove it!). The isogonal conjugate of
the centroid is called the Lemoine point. We will revisit this point in Sect. 2.5.1.
Again, we give two different proofs.
Proof 1 Let us change slightly the phrasing of the theorem: In the interior of the
triangle ABC, there are the points P and Q such that the segments AP and AQ
form equal angles with the angle bisector of BAC and the segments BP and BQ
form equal angles with the angle bisector of ABC. Prove that CP and CQ form
equal angles with the angle bisector of BCA.
To prove this, reflect P over AB, AC, and BC to obtain the points PC , PB , and
PA , respectively, as shown in Fig. 1.26. Then, the fact that AP and AQ form equal
angles with the angle bisector of BAC implies that AQ is the angle bisector of
PB APC . Indeed, working with oriented angles, if R is a point on the angle bisector
of BAC, then
QAPB = P APB − 2 P AR = RAPB − P AR

= RAB + RAB + P AR − P AR = 2 RAB
= BAC
and
PC AQ = PC AP + 2 P AR = PC AR + P AR
= CAR + CAR − P AR + P AR = 2 CAR
= BAC.
So AQ bisects PB APC . Because APB = APC , the triangle APB P C is isosceles,

so the angle bisector AQ is also the perpendicular bisector of the segment PB PC .
Hence Q is on the perpendicular bisector of PB PC .
Fig. 1.26 Reflection of P C

over the sides PA
PB
P
A B
PC
38 1 Isometries
Repeating the argument, we deduce that the conditions from the statement imply
that Q is on the perpendicular bisectors of the segments PA PC as well. But then
Q is the circumcenter of triangle PA PB PC , and so Q must be on the perpendicular
bisector of PA PB , which is equivalent to the fact that CP and CQ form equal angles
with the angle bisector of C.

Proof 2 Let the cevian from A form the angles α1 and α2 with AB and AC,
respectively; the cevian from B form the angles β1 and β2 with BC and BA,
respectively; and cevian from C form the angles γ1 and γ2 with CA and CB,
respectively. By the trigonometric version of Ceva’s Theorem, the fact that the three
cevians intersect implies that
sin α1 sin β1 sin γ1

· · = 1.
If we reflect the cevians over the corresponding bisectors, then the first of them
forms the angles α2 and α1 with AB and AC, respectively; the second forms the
angles β2 and β1 with BC and BA, respectively; and the third forms the angles γ2
and γ1 with CA and CB, respectively (Fig. 1.27). And, again by Ceva’s Theorem
written in trigonometric form, the condition that the three cevians intersect is

· · = 1,
and this is just the above relation flipped over. Hence the conclusion.

We point out that a proof of Ceva’s Theorem can be found in the next chapter
(Problem 116). It is important to know that the construction of the isogonal
conjugate also works when the triangle ABC is degenerate, with the vertex A at
infinity. In that case the triangle consists of the segment BC and the two parallel
rays |BX and |CY (Fig. 1.28).
Then the isogonal conjugate of a point P is the intersection of the reflections of
P B and P C over the angle bisectors of XBC and Y CB and the reflection of the
A A
α2 α2 α1
α1
β1 γ2
β2 γ1
β2 γ1
β1 γ2
B C B C
Fig. 1.27 Trigonometric proof of the existence of the isogonal conjugate

B X
P
A=
Q
C Y
Fig. 1.28 The existence of the isogonal conjugate in the case where the triangle is degenerate
line through P that is parallel to BX over the line parallel to BC that runs through
the midpoint of BC.
The existence of Q can be shown by adapting the first proof of Theorem 1.24, or
by using a limiting argument with A → ∞, where we notice that the angle bisector
of BAC becomes the line that is at equal distance from BX and CY , as by the
Bisector Theorem, the angle bisector of BAC cuts BC in the ratio AB/AC, and
this ratio tends to 1 when A tends to infinity.
As a direct corollary of the first proof of Theorem 1.24, we obtain the following
two characterizations of the isogonal conjugate of a point.
Proposition 1.25
(i) Let P be a point in the plane of the triangle ABC, and let PA , PB , PC
be reflections of P over the lines BC, CA, and AB, respectively. Then the
isogonal conjugate of P with respect to the triangle ABC is the circumcenter
of PA PB PC .
(ii) Let P be a point in the plane of the triangle ABC, and let L, M, N be
the projections of P onto the lines BC, CA, and AB, respectively. Then
the perpendiculars from A, B, and C onto the lines MN, LN , and LM,
respectively, are concurrent at the isogonal conjugate of P with respect to the
triangle ABC.
Note that (ii) follows from the fact that MN, LN, and LM are parallel to PB PC ,
PA PC , and PA PB , respectively.
III. Area in Complex Coordinates
Here is a slick proof, via translations and rotations, of the “shoelace” formula for
the area, which we phrase in complex coordinates, in the spirit of this book.
Theorem 1.26 (Gauss) Let a1 , a2 , . . . , an be the coordinates of the vertices of a
non-self-intersecting polygon that is oriented counterclockwise. Then the area of
the polygon is equal to
1
(a1 a2 + a2 a3 + · · · + an−1 an + an a1 ).
2
40 1 Isometries
Proof We begin by proving this formula for a triangle, where it falls apart when
attacked with geometric transformations. If we translate the triangle by z, then the
formula reads
1
[(a1 + z)(a2 + z) + (a2 + z)(a3 + z) + (a3 + z)(a1 + z)]
2
1 1
= (a1 a2 + a2 a3 + a3 a1 ) + [(a1 + a2 + a3 )z + (a1 + a2 + a3 )z]
2 2
3
+ |z|2 .
2
And this is equal to the area of the nontranslated triangle
1
(a1 a2 + a2 a3 + a3 a1 )
2
because the second and the third terms are real numbers. It follows that we can
translate the triangle so that a3 = 0, without changing the value of either the area
or of the formula from the statement. It suffices to prove that, in this case, the area
is equal to 12 a1 a2 . Rotate the triangle so that a1 becomes positive and real. The
rotation does not change the area, nor does it change the formula because
eiα a1 eiα a2 = ei(α−α) a1 a2 = a1 a2 .
And in this case the area is indeed the imaginary part of 12 a1 a2 , because this is 12 ×
base×height (Fig. 1.29). Note that because the triangle is oriented counterclockwise,
the imaginary part of a2 is positive (when it is oriented clockwise, the imaginary part
of a2 is negative, and consequently the formula needs a minus sign in front of it).
We have thus proved the formula for the area of a triangle. For a general polygon,
start by dissecting the polygon into triangles (Fig. 1.30). This can be done easily if
the polygon is convex; if the polygon is not convex, the existence of the dissection
can be proved by induction on the number of vertices once we show that interior
diagonals always exist. To show the existence of an interior diagonal, pick a vertex,
Fig. 1.29 Proof of the

formula for the area: the
simplest case a2 = k + ih
b a1 = b

formula for the area of a
polygon
and from it look inside a polygon. If you see another vertex, join it to the chosen
one by an interior diagonal. If you do not see another vertex, then the two vertices
adjacent to the chosen vertex can be joined by an interior diagonal.
Once the polygon has been decomposed into triangles, add the area formulas for
all these triangles. Note that every internal diagonal joining some aj and ak belongs
to two triangles, and in the sum it gives rise to a term of the form 12 aj ak and a term
of the form 12 ak aj . But aj ak + ak aj is a real number, so its imaginary part is zero.
Hence the contribution of internal diagonals is zero, and the formula is proved.
Remark An easy check shows that

1 1 1

(a1 a2 + a2 a3 + a3 a1 ) = − a1 a2 a3 ,
a a a
1 2 3
so we can also write the area in terms of determinants (in fact it is the behavior of
determinants under geometric transformations that was used in the solution). This
formula can be used to check the collinearity of three points a1 , a2 , a3 , as the points
are collinear if and only if the triangle they form has area zero, that is, if and only if
(a1 a2 + a2 a3 + a3 a1 ) = 0. We can thus write the equation of the line through two
points a1 , a2 as

1 1 1

(a1 a2 + a2 z + za1 ) = 0 or a1 a2 z = 0.
a a z
1 2
For completeness, we recall the formula for a 3 × 3 determinant

a11 a12 a13

a21 a22 a23 = a11 (a22 a33 − a23 a32 ) + a12 (a23 a31 − a21 a33 )

a a a
31 32 33
+a13 (a21 a32 − a31 a22 ).

42 1 Isometries
IV. Morley’s Theorem

We conclude our series of theoretical examples with Alain Connes’ proof of
Morley’s Theorem.
Theorem (Morley’s Theorem) The three points of intersection of the adjacent
trisectors of the angles of any triangle form an equilateral triangle.
Proof Let the original triangle be ABC, whose angles A, B, C are oriented
counterclockwise, and let A B C be the triangle determined by the trisectors, as
shown in Fig. 1.31. Throughout the solution, we measure angles in radians.
The first observation that Connes made was that the intersections of consecutive
trisectors are the fixed points of pairwise products of rotations ρA , ρB , ρC , around
the vertices A, B, C of the triangle, of angles equal to two thirds of the correspond-
ing angles of the triangle. More precisely:
• A is the unique fixed point of ρB ◦ ρC ,
• B is the unique fixed point of ρC ◦ ρA ,
• C is the unique fixed point of ρA ◦ ρB .
To see this, note, with the aid of Fig. 1.32, that ρC (A ) is the reflection A of A over
BC and ρB (A ) = A , so ρB ◦ ρC (A ) = A .
His second observation is that
ρA3 ◦ ρB3 ◦ ρC3 = 1.
Fig. 1.31 Morley’s Theorem A
B
C
A
B C
Fig. 1.32 Alain Connes’ first A

observation
ρC
B C
ρB
A
Indeed, ρA3 is the rotation about A by 2 A, which is the composition of the reflection
σAB over AB followed by the reflection σAC over AC, and similarly for the other
two rotations. Then
ρA3 ◦ ρB3 ◦ ρC3 = σAC ◦ σAB ◦ σAB ◦ σBC ◦ σBC ◦ σAC = 1,
because σAB2 = σ 2 = σ 2 = 1.
BC AC
Another way to see this is to first observe that the transformation ρA3 ◦ ρB3 ◦ ρC3
rotates figures by 2 A + 2 B + 2 C = 2π . This means that it does not rotate
figures, and therefore it is either the identity map or a translation. Also
ρA3 ◦ ρB3 ◦ ρC3 (C) = ρA3 ◦ ρB3 (C) = ρA3 (σAB (C))
= σAB (σAB (C)) = C.
Since the transformation has a fixed point, it cannot be a translation; hence, it is the
identity map.
Note also that ρA ◦ ρB ◦ ρC is an isometry which rotates figures by 23 × π = 2π 3 ,
so it is a rotation by 2π3 , and Connes first tried to see if this is the rotation that maps
A B C to itself. Unfortunately, it is not.
This is the moment to switch to complex numbers. Let α, β, and γ be the complex
coordinates of A , B , and C , respectively. We have the following characterization
of equilateral triangles, which is of interest in itself.
Lemma 1.27 The points α, β, and γ are the vertices of an equilateral triangle
oriented counterclockwise if and only if
2π i 4π i
α+e 3 β +e 3 γ = 0.
πi
Proof The condition that γ rotates to α around β by e 3 is that
α−β πi
=e3.
γ −β
Rewrite this as
πi πi
α + (e 3 − 1)β − e 3 γ = 0.
Using
√ √
πi 1 3 2π i πi 1 3 4π i
e 3 −1=− + i = e 3 and − e 3 = − − i=e 3 ,
2 2 2 2
we obtain the equivalent condition
44 1 Isometries
2π i 4π i
α+e 3 β +e 3 γ = 0,
and the lemma is proved.

Returning to the proof of Morley’s Theorem, using Theorem 1.1, we write
ρA (z) = a1 z + b1 , ρB (z) = a2 z + b2 , ρC (z) = a3 z + b3 ,
where
2i A 2i B 2i C
a1 = e 3 , a2 = e 3 , a3 = e 3 .
Note that because A + B + C = π

2π i
a1 a2 a3 = e 3 .
Now Morley’s Theorem follows by applying Connes’ Theorem below to

ρA , ρB , ρC and then using Lemma 1.27.

Theorem (Connes’ Theorem) Consider the transformations
fj : C → C, fj (z) = aj z + bj , j = 1, 2, 3.
Assume that a1 a2 , a1 a3 , a2 a3 , and a1 a2 a3 are all different from both 0 and 1. The
following are equivalent:
(i) f13 ◦ f23 ◦ f33 = 1,
(ii) ω3 = 1 and α + ωβ + ω2 γ = 0 where ω = a1 a2 a3 and α, β, γ are the (unique)
fixed points of f2 ◦ f3 , f3 ◦ f1 , and f1 ◦ f2 , respectively.
Proof Condition (i) reads
a13 a23 a33 z + (a12 + a1 + 1)b1 + a13 (a22 + a2 + 1)b2 + (a1 a2 )3 (a32 + a3 + 1)b3 = z,
for all z. Identifying coefficients we obtain ω3 = a13 a23 a33 = 1 and
(a12 + a1 + 1)b1 + a13 (a22 + a2 + 1)b2 + (a1 a2 )3 (a32 + a3 + 1)b3 = 0. (1.2)
All that remains to show is that this equality is equivalent to
α + ωβ + ω2 γ = 0.
For this we have to compute the coordinates of α, β, γ . Note that
f1 ◦ f2 (z) = a1 a2 z + a1 b2 + b1 ,
f2 ◦ f3 (z) = a2 a3 z + a2 b3 + b2 ,
f3 ◦ f1 (z) = a3 a1 z + a3 b1 + b3 ,
where we point out that each of these is a rotation so it has a unique fixed point. We
compute the fixed points to be
a2 b3 + b2 a1 a2 b3 + a1 b2
α = fix(f2 ◦ f3 ) = = ,
1 − a2 a3 a1 − ω
a3 b1 + b3 a2 a3 b1 + a2 b3
β = fix(f3 ◦ f1 ) = = ,
1 − a3 a1 a2 − ω
a1 b2 + b1 a3 a1 b2 + a3 b1
γ = fix(f1 ◦ f2 ) = = .
1 − a1 a2 a3 − ω
The reader should notice that we transformed the denominators from quadratic
expressions to linear expressions with the goal of simplifying the equation α +ωβ +
ω2 γ = 0 as much as possible. Now substitute the values of α, β, and γ into this
equation and then multiply by the common denominator to obtain the equivalent
equation
(a1 a2 b3 + a1 b2 )(a2 − ω)(a3 − ω) + ω(a2 a3 b1 + a2 b1 )(a1 − ω)(a3 − ω)

+ω2 (a1 a3 b2 + b1 a3 )(a1 − ω)(a2 − ω) = 0.
This looks discouraging, but after multiplying out
a2 b3 ω + b2 ω − a1 a22 b3 ω − b3 ω2 − a1 a2 b2 ω − a1 a3 b2 ω + a1 a2 b3 ω2 + a1 b2 ω2
+b1 a3 ω2 + b3 ω2 − a2 a32 b1 ω2 − b1 − a2 a3 b3 ω2 − a2 a1 b3 ω2 + a2 a3 b1 + a2 b3
+a1 b2 + b1 − a12 a3 b2 − b2 ω − a3 a1 b1 − a2 a3 b1 + a1 a3 b2 ω + a3 b1 ω
we notice that quite a few terms cancel out. We are left with
a2 b3 (1 + ω) + a1 b2 (1 + ω2 )a3 b1 (ω + ω2 ) − a1 a22 b3 ω − a1 a2 b2 ω − a2 a32 b1 ω2

−a2 a3 b3 ω2 − a12 a3 b2 − a1 a3 b1 = 0.
This can be further transformed using the equality 1+ω+ω2 = 0 into the equivalent
identity
a2 b3 ω2 + a1 b2 ω + a3 b1 + a1 a22 b3 ω + a1 a2 b2 ω − a2 a32 b1 ω2 − a2 a3 b3 ω2
−a12 a3 b2 − a1 a3 b1 = 0.
46 1 Isometries
In order to turn this into (1.2), we group the terms with respect to b1 , b2 , b3
(a3 + a1 a3 + a2 a32 ω2 )b1 + (a1 ω + a1 a2 ω + a12 a3 )b2

(1.3)
+(a2 ω2 + a2 a3 ω2 + a1 a2 ω)b3 = 0.
Comparing the coefficient of b1 in (1.2) and (1.3), we notice that

1
a3 + a1 a3 + a2 a3 ω2 = a3 + a1 a3 + a12 a23 a34 = a3 1 + a1 +
a1
a3 2
= (a1 + a1 + 1).
a1
And now it is straightforward to check that (1.3) becomes (1.2) after multiplying by
a1 /a3 . The theorem is proved.
1.2.2 Examples of Problems Solved Using Isometries
Let us now present a few actual mathematics competition problems. The first has
appeared in the All-Russian Mathematical Olympiad in 2002.
Problem 1.1 Let O be the circumcenter of the triangle ABC. Choose M and N
on the sides AB and AC, respectively, such that MON = A. Prove that the
perimeter of the triangle MAN is greater than or equal to the length of the side BC.
Solution 1 The aim is to rearrange the sides of the triangle MAN so that they form a
polygonal line connecting B to C. To this end consider the rotations ρ1 and ρ2 about
O, the first mapping A to B, and the second mapping A to C, and let M = ρ1 (M)
and N = ρ2 (N ) (Fig. 1.33).
Fig. 1.33 A transformation A

of MAN into a polygonal
line
N
M
B C
N
M
The triangle OBM is the image of the triangle OAM through ρ1 , so BM =

AM, and the triangle OCN is the image of the triangle OAN through ρ2 , so CN =
AN. But
MOM = AOB = 2 C and NON = AOC = 2 B.
Hence
M ON = 360◦ − MON − MOM − NON

= 360◦ − A − 2 C − 2 B = A.
So the triangles MON and M ON are congruent, having two pairs of sides and
the angle between them equal. It follows that M N = MN. As the shortest path
between two points is the segment joining them, we can write
AM + MN + NA = BM + M N + N C ≥ BC.
This is the desired inequality.

Solution 2 Reflect A over the lines OM and ON to the points B and C ,
respectively. Then AM = B M and AN = C N. Therefore
AM + MN + NA = B M + MN + NC ≥ B C ,
where for the last inequality we have used again the fact that the shortest path
between two points is the segment joining them (Fig. 1.34).
On the other hand
B OC = B OM + MON + NOC = AOM + MON + AON

= 2 MON = 2 A = BOC.
Fig. 1.34 Another A

transformation of MAN into
C
a polygonal line
N
O
B
B C
48 1 Isometries
Also, B O = AO = C O, and since AO = BO = CO, we obtain that BO = B O

and CO = C O. Consequently the triangles BOC and B OC are congruent, and
therefore BC = B C . The conclusion follows.

This is a standard method for proving metric equalities or inequalities, which can
be applied in some of the problems below: map by isometries the segments to turn
them into the sides of a triangle (or polygon), and then use metric relations in a
triangle (or polygon).
If we rotate a point A about O by 60◦ to a point A , then the triangle AOA is
equilateral, being isosceles and having a 60◦ angle (see Fig. 1.35). So equilateral
triangles and 60◦ rotations go hand in hand, and this is the subject of the next
problem, which was communicated to us by Onofre Campos.
Problem 1.2 Let ABCD be a convex quadrilateral such that there exists a point M
in the plane satisfying AM = MB, CM = MD, and AMB = CMD = 120◦ .
Prove that there exists a point N in the plane such that the triangles AND and CN B
are both equilateral.
Solution There is a 60◦ rotation hidden in this problem. To bring it to light, note
that the point M is the center of a 120◦ rotation that takes A to B and C to D
(this is where we use the fact that ABCD is convex, so that A, B, C, and D are
properly cyclically ordered). This rotation takes therefore the segment AC to the
segment BD. So the two diagonals of the quadrilateral are equal and make an angle
of 120◦ . But then the diagonals also make an acute angle of 180◦ − 120◦ = 60◦ (see
Fig. 1.36).
If now we focus on the equal segments AC and DB, and consider the orientation
preserving isometry that maps A to D and C to B, then by Theorem 1.9, this
transformation is a rotation by 60◦ , which is the angle formed by AC and DB.
Fig. 1.35 A 60◦ rotation A

yields an equilateral triangle
Fig. 1.36 Equal diagonals C

forming a 60◦ angle D
60°
A B
Fig. 1.37 The translation A D

P → P
P E
P
B C
We can therefore choose N to be the center of this 60◦ rotation, as on the one
hand N A = N D and AND = 60◦ and on the other hand NC = NB and
BNC = 60◦ , and so the triangles AND and CN B are equilateral, as desired.

Problem 1.3 Let ABCD be a square and let P be a point inside it. Prove that
AP B + CP D = 180◦ if and only if P lies on at least one of the diagonals AC
and BD.
Solution When the problem involves parallelograms (a square is a very special
parallelogram!) and angles that sum up to nice angles, like 180◦ or 90◦ , it might
be useful to perform a translation. And indeed, this is what we do here. We translate
−→
P by AD to obtain P , as shown in Fig. 1.37. Then CP D = BP A, and
AP B + CP D = 180◦ is equivalent to CP D + CP D = 180◦ . Since P
is inside the square, CP DP is convex, and so AP B + CP D = 180◦ if and only
if CP DP is cyclic.
Now let us assume that CP DP is cyclic. Note that P P = AD = CD, so
P P and CD are equal chords in the same circle. The two arcs they respectively
subtend are either equal or add up to the whole circle. So we either have CP D =
P CP or CP D + P CP = 180◦ . In the first case, we would have CP D +
P DP = 180◦ . Thus in both cases CP DP is a trapezoid, with either P D CP
or P C DP . But, because of the translation, CP BP and DP AP , so either
P D BP or P C AP , which means that either P is on BD or it is on AC.
Conversely, if P is on say BD, then by symmetry, CP D = AP D, and the
latter is the supplement of AP B.

So in this problem you noticed that the sum of two angles is 180◦ , and you
know that this leads to cyclic quadrilaterals. However, the angles were facing the
opposite ways, but the square ABCD, which is a parallelogram, yields at least two
translations, and each translation transports one angle to make it face the other! Of
course, there are other possible transformations. Try reflecting across the diagonal
BD, for example.
We conclude our discussion with a more difficult problem, where reflection over
a line is used.
Problem 1.4 Let ABC be a triangle and let D be a point on the segment BC
different from B and C. The circumcircle of ABD meets the segment AC again
50 1 Isometries
at an interior point E. The circumcircle of ACD meets the segment AB again at an

interior point F . Let A be the reflection of A over the line BC. The lines A C and
DE meet at P , and the lines A B and DF meet at Q. Prove that the lines AD, BP
and CQ are concurrent, or are all parallel.
This problem was given at the Romanian Master of Mathematics in 2016. We
present the three official solutions, all of which rely on the reflection σ over the line
BC. This reflection has been introduced already in the statement of the problem, so
it is natural to expect that it will play an important role in the solution.
Solution 1 To see why we should expect σ to be useful, we start with the
observation that
1 1
BDF = BAC = CDF and CDE = BAC = BDE,
2 2
from where we deduce that BDF = CDE, and so the lines DE and DF are
images of each other under σ . On Fig. 1.38, we have marked a few more elements
that arise from the action of σ such as the point P = σ (P ), where the lines AC =
σ (A C) and DF = σ (DE) meet, the point Q = σ (Q) where the lines AB =
σ (A B), and DE = σ (DF ) meet. We have also marked the point M where the line
AD intersects for the second time the circumcircle of ABC together with M =
σ (M). To solve the problem, we will check that the lines DM = σ (AD), BP =
σ (BP ) and CQ = σ (CQ) are concurrent.
A carefully drawn figure should disclose additional information about M ,
namely, that it lies at the intersection of the circumcircles of BDF and CDE. And
indeed, using the cyclic quadrilaterals ABMC and AF DC, we can write
(BM , M D) = − (BM, MD) = (BM, MA) = (BC, CA) = (BF, F D),
which shows that M lies on the circumcircle of BDF . A similar argument shows
that M lies on the circumcircle of CDE. We deduce that the line DM is the radical
axis of the circumcircles of BDF and CDE.
On the other hand, P lies on both lines AC and DF , so it must be the radical
center of the circumcircles of ABC, ADC, and BDF , and hence the line BP is
the radical axis of the circumcircles of BDF and ABC. Similarly, the line CQ
is the radical axis of the circumcircles of CDE and ABC. We conclude that the
lines DM , BP , and CQ are concurrent at the radical center of the circumcircles
of ABC, BDF and CDE, or are all parallel. Consequently, the lines AD = DM,
BP , and CQ are also concurrent or all parallel, as desired.

The other two solutions are based on Desargues’ Theorem (Problem 115) and
Pappus’ Theorem (proved in Sect. 2.2.1), respectively. The solutions were found by
the member of the problem committee Ilya Igorevich Bogdanov.
Solution 2 You can follow this solution on Fig. 1.39. We start again with the
reflection σ and conclude as above that the lines AC and DF meet at P = σ (P )
F
Q E
M
B D C
M
Q
Fig. 1.38 First solution to Problem 1.4
and the lines AB and DE meet at Q = σ (Q). From here we deduce that the lines
P Q and P Q = σ (P Q) meet at some point R on the line BC (possibly at infinity).
Since the pairs of lines (CA; QD), (AB; DP ), (BC; P Q) meet at three collinear
points, namely, P , Q , and R, respectively, the triangles ABC and DP Q are
perspective, i.e., the lines AD, BP , CQ are concurrent (or parallel), by Desargues’
Theorem.

Solution 3 Arguing on Fig. 1.40, we let the lines BE and CF meet at X. Using
inscribed angles in the cyclic quadrilaterals BDEA and CDF A, we have XBD =
EAD = XF D, so the quadrilateral BF XD is cyclic. Similarly, the quadrilateral
CEXD is cyclic. Then X is the point M from the first solution, and consequently
X = σ (X) = M, showing that X is on the line AD.
Let E = σ (E) and F = σ (F ). The points Q, B, F are collinear, and the points
P , C, E are also collinear, so we can apply Pappus’ Theorem (Theorem 2.14 from
Sect. 2.2.1) to infer that X = BE ∩ CF , D = F P ∩ QE , and BP ∩ CQ are
collinear points. It follows that the lines BP , CQ, DX are concurrent. But the lines
DX and AD coincide, and we are done.

52 1 Isometries
F
Q E
R B
D C
Fig. 1.39 Second solution to Problem 1.4
1.2.3 Problems in Euclidean Geometry to be Solved Using

Isometries
Below is list of geometry problems for you to solve. Many of these problems can
be solved by other methods, of course, but we insist that you look at them from the
point of view of geometric transformations and find solutions that use isometries.
We remind you that all problems have hints in the middle of the book and solutions
at the end of the book.
13 Inside a square A1 A2 A3 A4 consider a point P . Take the perpendiculars from
A2 to A1 P , from A3 to A2 P , from A4 to A3 P , and from A1 to A4 P . Show that
these four perpendiculars intersect at one point.
14 Let ABCD be a quadrilateral, and let M and N be the midpoints of the sides
AD and BC, respectively. Prove that if
AB + CD
MN =
2
F
Q E
X
B D C
X E
F
Fig. 1.40 Third solution to Problem 1.4
then ABCD is a trapezoid.

15 On the sides AB and AC of a triangle ABC, construct in the exterior the
squares ABDE and ACF G. Let the point I be such that AGI E is a parallelo-
gram.
(a) Prove that EG is perpendicular to the median from A and is twice its length.
(b) Prove that AI is perpendicular to BC.
(c) Prove that DC and BF intersect on the altitude from A.
(d) Let O be the circumcenter of the triangle AEG. Prove that the line AO is a
symmedian in the triangle ABC.
16 Consider a square inscribed in a parallelogram in such a way that each vertex
of the square is on another side of the parallelogram. Show that the perpendiculars
from the vertices of the parallelogram onto the sides of the square form another
square.
17 Do there exist noncongruent rectangles ABCD and MNP Q that do not have
the same center, such that AM = BN = CP = DQ?
54 1 Isometries
18 (The Pappus Area Theorem) Let ABC be a triangle, and let the parallelograms
ABMN and ACP Q be erected externally to ABC. Let R be the point where the
lines MN and P Q meet. Construct the parallelogram BCST externally having the
sides CS and BT parallel and equal to segment RA. Show that the area of BCST
is equal to the sum of the areas of ABMN and ACP Q.
19 In a right triangle ABC, the midpoint of the hypotenuse AC is denoted by O.
The points M and N are chosen on the legs AB and BC so that MON = 90◦ .
Prove that AM 2 + CN 2 = MN 2 .
20 Let P and Q be isogonal conjugates in a triangle ABC. Show that the reflection
of the line AP over the internal angle bisector of BP C and the reflection of the
line AQ over the internal angle bisector of BQC are reflections of each other over
the line BC.
21 Let n > 1 and let A1 A2 . . . A2n , be a (nonskew) polygon with 2n
sides. Translate the vertices A1 , A3 , . . . , A2n−1 by the same vector to the
points A1 , A3 , . . . , A2n−1 so that A1 A2 A3 A4 . . . , A2n−1 A2n is still a (non-
selfintersecting) polygon. Show that the polygons A1 A2 . . . A2n and A1 A2 A3 A4 . . . ,
A2n−1 A2n have the same area.
22 (a) Let ABCD be a cyclic quadrilateral, and let Ha , Hb , Hc , Hd be the
orthocenters of the triangles BCD, ACD, ABD, and ABC, respectively. Prove
that the quadrilaterals ABCD and Ha Hb Hc Hd are mapped into each other by a
reflection over a point.
(b) Show that the four perpendiculars constructed from the midpoints of the sides of
ABCD onto the opposite sides meet at one point.
23 Let be a semicircle of diameter AB. The point C lies on the diameter AB,

and the points D and E lie on the arc AB, with E between B and D. The tangents
to at D and E meet at F . Suppose that ACD = ECB. Prove that EF D =
ACD + ECB.
24 Show that if one can inscribe three equal squares in a triangle, then the triangle
is equilateral.
25 Let ABC be a triangle and let be a line. Let also A0 , B0 , C0 be the projections
onto of the vertices A, B, C, respectively. Prove that the perpendiculars from
A0 , B0 , C0 to the lines BC, AC, AB intersect at one point.
26 On the sides AB, BC, CD, DA of an arbitrary convex quadrilateral ABCD
construct in the exterior four squares, and let M1 , M2 , M3 , M4 be their respective
centers. Prove that the segments M1 M3 and M2 M4 are perpendicular and have equal
lengths.
27 Let ABC be an equilateral triangle. A line parallel to AC intersects the sides
AB and BC at M and P , respectively. Let D be the center of the equilateral triangle
BMP , and let E be the midpoint of the segment AP . Find the angles of the triangle
DEC.
28 Let M and N be the midpoints of the sides CD and DE, respectively, of a

regular hexagon ABCDEF of center O. Let L be the intersection point of the lines
AM and BN . Prove that
(a) the triangle ABL and the quadrilateral DMLN have equal areas;
(b) ALO = OLN = 60◦ ;
(c) OLD = 90◦ .
29 Let M be a point inside the convex quadrilateral ABCD such that ABMD is a
parallelogram. Prove that if CBM = CDM then ACD = BCM.
30 Let ABC be an acute triangle. Let also AD, BE, and CF be the altitudes (with
D on BC, E on AC, and F on AB), and let M be the midpoint of the side BC.
The circumcircle of the triangle AEF meets the line AM at A and X. The line AM
meets the line CF at Y . Let AD and BX meet at the point Z. Prove that the lines
Y Z and BC are parallel.
31 (The Butterfly Theorem) Let A be the midpoint of a chord c of a circle. Choose
B and C on c such that AB = AC. Let MN and P Q be two arbitrary chords
through B and C, respectively. Let R and S be the respective intersections of P M
and N Q with c. Prove that AR = AS.
32 Let ABC and BCD be two equilateral triangles that share one side. An
arbitrary line through D intersects AB at F and AC at E. Find the angle between
the lines BE and CF .
33 On the sides of an arbitrary triangle ABC, construct the equilateral triangles
BCA1 , ACB1 , and ABC1 such that A1 and A are separated by the line BC, B1 ,
and B are separated by the line AC, but C1 and C are on the same side of the line
AB. Let M be the center of the triangle ABC1 . Prove that the triangle MA1 B1 is
isosceles with A1 MB1 = 120◦ .
34 Given a convex n-gon P with no parallel sides, and a point O interior to it,
prove that there are no more than n lines through O that bisect the area of P.
35 Let ABCD be a convex quadrilateral with BAD = BCD = 120◦ .
Construct in the exterior equilateral triangles ABM, BCN , CDP , and DAQ. Prove
that CQ is parallel to AN and CM is parallel to AP .
36 Let ABCDE be a convex pentagon such that BC AE, AB = BC + AE, and
ABC = CDE. Let M be the midpoint of CE, and let O be the circumcenter of
the triangle BCD. Given that DMO = 90◦ , prove that 2 BDA = CDE.
37 Let M, N, P be points on the sides BC, CA, AB of a triangle ABC, respec-
tively. Let M1 be reflection of M with respect to the midpoint of BC, let N1 be the
reflection of N with respect to the midpoint of AC, and let P1 be the reflection of P
with respect to the midpoint of AB. Prove that if the perpendiculars to the respective
sides of the triangle ABC erected at the points M, N, and P meet at one point, then
the perpendiculars to the sides erected at M1 , N1 , P1 also meet at one point.
56 1 Isometries
38 A convex hexagon AC1 BA1 CB1 satisfies the relations AB1 = AC1 , BC1 =
BA1 , CA1 = CB1 , and A + B + C = A1 + B1 + C1 . Prove that the area
of the triangle ABC is half the area of the hexagon.
39 Six points are chosen on the sides of an equilateral triangle ABC as follows:
A1 , A2 on BC, B1 , B2 on CA, and C1 , C2 on AB, in such a way that they are the
vertices of a convex hexagon A1 A2 B1 B2 C1 C2 with equal side lengths. Prove that
the lines A1 B2 , B1 C2 , and C1 A2 are concurrent.
40 Let ABC be an isosceles triangle with AB = AC, and let I be its incenter.
A line r passing through I meets the sides AB and AC at the points D and E,
respectively. Let F and G be points on the side BC such that BF = CE and
CG = BD. Show that the angle F I G does not depend on r.
41 Let A, B, C be fixed points in the plane. A man starts from a certain point P0
and walks directly to A. At A, he turns his direction by 60◦ to the left and walks to P1
such that P0 A = AP1 . He repeats the same action starting at P1 and using the vertex
B to arrive at P2 . After he does the same action 1986 times successively around the
points A, B, C, A, B, C, . . ., he returns to the starting point (this means that P1986 =
P0 ). Prove that the triangle ABC is equilateral and oriented counterclockwise.
42 Let C1 and C2 be two congruent circles centered at O1 and O2 , which intersect

at A and B. Take a point P on the arc AB of C2 that lies inside C1 . The line AP
meets again C1 at C, the line CB meets C2 again at D, and the angle bisector of
CAD intersects C1 and C2 at E and L, respectively. Let F be the reflection of D
with respect to the midpoint of the segment P E. Prove that there exists a point X in
the plane satisfying XF L = XDC = 30◦ and CX = O1 O2 .
43 (Fagnano’s Theorem) Prove among all triangles inscribed in a given acute
triangle, the one determined by the feet of the altitudes has minimal perimeter.
44 Let ABCD be an isosceles trapezoid with ABCD. Let E be the midpoint
of AC. Denote by ω and the circumcircles of the triangles ABE and CDE,
respectively. Let P be the crossing point of the tangent to ω at A with the tangent to
at D. Prove that P E is tangent to .
45 (The Erdős-Mordell Theorem) Let ABC be a triangle and let P be a point in its
interior. Denote by da , db , and dc the respective distances from P to BC, AC, and
AB. Show that
P A + P B + P C ≥ 2(da + db + dc ).
46 You are given a rusty compass whose opening became stuck at a fixed width.
You are also given three arbitrary points A, B, C in the plane. Using the rusty
compass, and no other tool, construct a fourth point D such that the lines AB and
CD are perpendicular.
1.3 Isometries Throughout Mathematics 57
47 A convex polygonal surface P lies on a flat wooden table. You are allowed to
drive some nails into the table. The nails must not go through the polygonal surface
P, but they may touch its boundary. We say that a set of nails blocks P if the nails
make it impossible to move P without lifting it off the table. What is the minimum
number of nails that suffices to block any convex polygon P?
1.3 Isometries Throughout Mathematics
At this moment, with isometries in mind, we take a turn toward other fields of
mathematics.
1.3.1 Geometry with Combinatorial Flavor
Let us make a first stop in combinatorial geometry, where we commence with

a problem from a 2005 Romanian Team Selection Test for the International
Mathematical Olympiad.
Problem 1.5 Prove that if any two vertices of a polygon√are at distance at most 1
from each other, then the area of the polygon is less than 23 .
Solution The solution that we present parallels the one found by the student Alin
Purcaru during the test. It goes like this: Let us call the polygon K. The condition
from the hypothesis implies that every band of width 1 determined by two parallel
lines can be translated to a position that contains K inside the union of the interior
and the boundary of the band.
Now consider three such bands that make angles of 2π 3 with each other and that
contain K in their intersection. These bands form an equiangular hexagon with
distances between opposite sides equal to 1. We claim that
√
(i) the area of the hexagon is less than or equal to 23 ;
(ii) the diagonals joining opposite vertices are equal to √2 .
3
Both properties hold for the regular hexagon with side equal to √1 , in which the
3
distance between opposite sides is 1. To prove the two claims in general, note that
the geometric figure determined by the intersection of two of the bands is always the
same (a parallelogram with an angle of 60◦ ), so it is the location of the third band
that determines the particular figure.
For (i) it suffices to prove that the area decreases when we modify the regular
hexagon by translating just one of the bands in the direction perpendicular to the
lines that bound it. Examining Fig. 1.41, we see that in the process of translation,
we lose one trapezoid and we gain another. Both trapezoids have equal altitudes and
58 1 Isometries
Fig. 1.41 Proof of the bound

for the area

invariance of the length of the
diagonals
equal angles, but the bases of the one that we lose are larger than the bases of the
one that we gain. So the area decreases, as desired.
The proof of (ii) becomes transparent when we glimpse at Fig. 1.42. When
translating one of the bands, the diagonals themselves are being translated (watch
the parallelograms determined by the diagonals!). √
From (i) we deduce that K lies inside a figure of area 23 , so the area of K cannot
exceed this number. But (ii) shows that K cannot be the entire hexagon (for then it
would contain two points at distance √2 > 1), so the inequality is strict.

3
The second problem that we present was published by Yu.P. Lysov in the journal
Kvant (Quantum). It employs rotations.
Problem 1.6 Let S be the union of n closed connected subsets of the unit circle.
It is known that for every rotation ρ about the center of the circle, the sets S and
ρ(S) overlap. What is the smallest value that the sum of the measures of the sets can
have?
A closed connected subset of the unit circle is either a point, a closed arc, or the
whole circle. The measure of a point is simply 0, while the measure of an arc is its
length.
Solution Let σ be the sum of the measures of the n connected sets, which sets we
label in counterclockwise order. Rotate the circle by the angle φ, and denote by Aj k
those values of φ for which the j th set of the original circle overlaps with the kth
set of the rotated one. It is clear that Aj k is a set whose measure is the sum of the
measures of the j th and kth sets (Aj k is itself either a point, an arc, or the whole
circle). Moreover, the union of the sets Aj k should contain all angles, since for every
rotation some j th and kth arcs overlap. Thus, for the sum of the measures of the sets
Aj k , we have the inequality

m(Aj k ) ≥ 2π.
j,k
But

m(Aj k ) = 2nσ,
j,k
as each connected set is counted 2n times, once when it pairs with the rotates of each
of the n sets and once when it is itself a rotate and pairs with each of the original n
sets. Hence
2nσ ≥ 2π,
so σ ≥ πn .
The value σ = π
n is attained when one set is the arc of length π
n centered at 1, and
2π ik
each of the other n − 1 sets consists of one of the points e , k = 1, 2, . . . , n − 1.
n
This example is shown in Fig. 1.43. And if you are not completely convinced that
the sum of the measures of the sets Aj k equals 2nσ , you can compute it explicitly
for this example.

Here are more problems that you should solve with isometries in mind.
48 Given a triangular cake and a box in the shape of its mirror image, show that
the cake can be cut into three slices so that the slices now fit inside the box. (The
cake has icing, so we are not allowed to flip it over and place it upside down inside
the box.)
Fig. 1.43 Example with

minimal sum of measures for
n=6
60 1 Isometries
49 Two equal squares overlap to determine an octagon. The sides of the first square
are colored red, while the sides of the second square are colored blue. Prove that the
sum of the lengths of the red sides of the octagon is equal to the sum of the lengths
of its blue sides.
50 Let n, p be natural numbers such that 6 ≤ n and 3 ≤ p ≤ n − p. The vertices
of a regular n-gon are colored either red or black: p vertices red and n − p vertices
black. Show that there are two congruent polygons, each having at least p/2
vertices, one with all vertices red and the other with all vertices black.
51 (a) Inside the unit square there is a set of points that is the union of finitely
many polygonal surfaces and has the property that the distance between any two
points of the set is never equal to 0.001. Prove that the area of the set is less than
0.34.
(b) Show that the area of this set is actually less than 0.287.
52 Every point of integer coordinates in the plane is colored either red or blue.
Prove that there is an infinite set of points of the same color that has a center of
symmetry.
53 Given a finite number of disks of radius 1 in the plane, denote by S the area of
their union. Prove that one can choose finitely many of these disks that are pairwise
disjoint and have the sum of their areas greater than 29 S.
54 Several chords are constructed in a circle of radius 1. Prove that if every
diameter intersects at most k chords, then the sum of the lengths of the chords is
less than kπ .
55 Let n ≥ 2 be an integer. In some group of 2n people, each person has at most
n − 1 enemies. Show that the people can be seated at a round table so that no person
sits next to an enemy.
56 On a cylindrical surface of radius r, unbounded in both directions, consider n
points and a surface S of area strictly less than 1. Prove that by rotating around the
axis of the cylinder and then translating in the direction of the axis by at most 4πn r
units, one can transform S into a surface that does not contain any of the n points.
57 Let n ≥ 3 be a positive integer; and let us consider n guests that sit at a round
table. A move consists of two neighbors exchanging seats. What is the smallest
number of moves one has to perform so that after performing these moves the guests
sit in reverse order.
58 Let S1 , S2 , . . . , Sn be a collection of equilateral triangles with one side parallel
to the x axis and all pointing up. The triangles may overlap. For each Si , let Ti be
its medial triangle (formed by the midpoints of its sides). Finally, let S be the union
of all Si triangles, and let T be the union of all Ti triangles. Prove that
area(S) ≤ 4 area(T ).
59 (The Chord Theorem)

(a) Let be a polygonal line in the plane with endpoints A and B. Prove that for
each positive integer n there exists a segment parallel to AB with endpoints in
having length n1 AB.
(b) Show that for every number α ∈ (0, 1) that is not of the form n1 , there always
exists a polygonal line with endpoints A and B having the property that there is
no segment parallel to AB and with endpoints in whose length is αAB.
60 A flea jumps between two intersecting lines in the plane, a and b, on a path
A1 B1 A2 B2 . . . such that:
(i) Ai ∈ a and Bi ∈ b for all i;
(ii) A1 B1 = B1 A2 = A2 B2 = B2 A3 = · · · ;
(iii) for all j , Aj = Aj +1 if and only if Aj Bj ⊥a, and similarly Bj = Bj +1 if and
only if Bj Aj +1 ⊥b.
Prove that the flea’s path is periodic if and only if the lines a and b make a rational
angle (when measured in degrees).
And some problems about chessboards.
61 On an n × n square lattice, two players mark alternatively one (length 1)
segment of the lattice. The game is won by the player who is the first to create
some closed polygonal line (formed by segments marked by both). Does any of the
two players have a winning strategy?
62 On an arbitrarily large chessboard, consider a generalized knight which can
jump p squares in one direction and q in the other. Show that such a knight can only
return to the initial position after an even number of jumps.
63 A rectangular bar of chocolate is divided by longitudinal and transversal
hollows into 50 small equal square portions. Two people play the following game:
The first breaks the bar along some hollow into two rectangular parts. Then the
players take turns breaking one of the resulting pieces along a hollow. The player
who is the first to break off a piece with no hollows (a) loses, (b) wins. Which of the
two players can guarantee winning?
64 Can one place the letters a, b, c in the boxes of a 100 × 100 chessboard so that
every 3 × 4 rectangle of the lattice contains three as, four bs, and five cs?
65 (a) Consider the tessellation of the plane by equal squares. For what n ≥ 2 it
is possible to color the squares by n colors such that the centers of the squares of
the same color form a square lattice of the plane and all these lattices have equal
squares and parallel sides?
(b) Consider a tessellation of the plane by equal hexagons. For what n ≥ 2 it is
possible to color the hexagons by n colors so that the centers of the hexagons of
the same color form an equilateral triangular lattice and these lattices have equal
triangles and parallel sides?
62 1 Isometries
1.3.2 Combinatorics of Sets
More combinatorics problems for you to solve follow after this example from the
Bulgarian Mathematical Olympiad.
Problem 1.7 There are 2000 white balls in a box. There are also unlimited supplies
of white, green, and red balls, initially outside the box. At each turn, we can replace
two balls in the box with one or two balls as follows: two whites with a green; two
reds with a green; two greens with a white and a red; a white and a green with a red;
and a green and a red with a white.
(a) After finitely many of the above operations there are three balls left in the box.
Prove that at least one of them is green.
(b) Is it possible that after finitely many operations only one ball is left in the box?
Solution We consider the group of rotations of a square (mentioned in Sect. 1.1.5),
consisting of the rotations 1, ρ, ρ 2 , ρ 3 by 0◦ , 90◦ , 180◦ , and 270◦ . This group is
exhibited in Fig. 1.44.
Assign the rotation ρ to each white ball, ρ 2 to each green ball, and ρ 3 to each
red ball. A quick check shows that the given operations preserve the product of
the rotations assigned to all balls in the box. This product is initially ρ 2000 = 1.
If three balls were left in the box, none of them green, then the composition of
their associated transformations would be either ρ or ρ 3 , a contradiction. Hence, if
three balls remain, at least one is green, proving the claim in part (a). Furthermore,
because no ball has the assigned rotation equal to 1, the box must contain at least two
balls at any time. This shows that the answer to the question in part (b) is negative.

Fig. 1.44 The group of C B

rotations of a square
ρ
D A
D C B A
ρ2
C D
A B
ρ3 A D
B C
Associating algebraic structures to combinatorial configurations and the con-

struction of mathematical objects that are invariant under transformations is a
fundamental idea in mathematics.
66 There are n trees arranged in a circle, and initially on each tree there is a bird.
At every moment, two birds fly in opposite directions to the neighboring trees. For
which n it is possible for all birds to meet on one tree?
67 Let A and B be two sets of real numbers. Define
A + B = {a + b | a ∈ A and b ∈ B}.
Prove that |A + B| ≥ |A| + |B| − 1, and find all cases where equality occurs (here
and below |A| denotes the number of elements of the set A).
68 Let a, b be fixed positive integers, and let A and B be finite sets so that:
(i) A and B are disjoint;
(ii) if an integer i is in either A or B, then i + a is in A or i − b is in B.
Prove that a|A| = b|B|.
69 Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be real numbers such that ai < bi for all
i = 1, 2, . . . , n and b1 + b2 + · · · + bn < 1 + a1 + a2 + · · · + an . Prove that there
is a real number c such that for all i = 1, 2, . . . , n and for every integer k
(ai + k + c)(bi + k + c) > 0.
70 Let A = {1, 2, . . . , n}, n > 1. A permutation σ : A → A is called an involution

if σ ◦ σ = 1A . Find the minimal number k such that every permutation is the
composition of k involutions.
71 Two subsets of the positive integers are called congruent if one is obtained from
the other by a translation. Is it possible to partition the set of positive integers into
infinitely many subsets that are all infinite and congruent to one another?
72 Let ← denote the left arrow2 key on a standard keyboard. If one opens a text
editor and types the keys “ab ← cd ←← e ←← f ,” the result is “faecdb.” We say
that a string B is reachable from a string A if it is possible to insert some amount of
←’s into A, such that typing the resulting characters produces B. So, our example
2 Here is a short explanation of how the ← key works. A computer’s text editor starts with an empty
screen and a cursor which we denote by “|”. When you type a letter x, the cursor | is replaced by
x|. So if the screen shows “m|th,” and you press the “o” key, then the result is “mo|th.”
The ← key moves the cursor one space backward. That is, “mo|th” becomes “m|oth” and finally
“|moth.” If the cursor is at the beginning of the string, then the ← key has no effect.
Finally, the cursor is not considered to be part of the final string. That is, if the string displays
“f |aecdb,” we just take the result to be faecdb.
64 1 Isometries
shows that “faecdb” is reachable from “abcdef.” Prove that for any two strings A
and B, A is reachable from B if and only if B is reachable from A.
1.3.3 Number Theory
In this section, by letting groups of transformations act on discrete geometric

combinatorial structures, we will derive some classical results in number theory.
We start with the famous proof of Wilson’s Theorem discovered by Julius Petersen.
Theorem (Wilson) If p be a prime number, then p divides (p − 1)! + 1.
Proof The result is obviously true for p = 2, so we only have to do the proof in
the case where p > 2. Consider p equally spaced points on a circle, which form the
vertices of a regular p-gon. Let us count the number of polygons (meaning closed
polygonal lines) whose vertices are the p points. There are p ways to choose the
first vertex, p − 1 ways to choose the second, p − 2 ways to choose the third, and
so on, so the total number of polygons is p!. However, in a given polygon, the first
vertex can be chosen in p ways, and the direction in which the polygon is traveled
can be chosen in two ways. It follows that each polygon is counted 2p ways. Thus,
the actual number of polygons is
p! (p − 1)!
= .
2p 2
If a polygon A1 A2 . . . Ap has the property that A1 A2 = A2 A3 = · · · = Ap A1 ,

we call it a regular stellated polygon (it might be a true regular polygon, or it may
have self-intersections). A regular stellated polygon is determined by the choice of
two vertices, since all other vertices are determined by the condition that the sides
are equal. Hence, there are p(p − 1) regular stellated polygons. But we counted
each regular stellated polygon 2p times, so the true number is
p(p − 1) p−1
= .
2p 2
The three regular stellated polygons for p = 7 are shown in Fig. 1.45.
We deduce that the number of polygons that are not regular stellated is
(p − 1)! p − 1 [(p − 1)! + 1] − p

− = .
2 2 2
Now let us declare two polygons to be equivalent if they coincide under a
rotation about the center of the circle. The regular stellated polygons are invariant
under rotations by multiples of 2π/p, so they are only equivalent to themselves.
Conversely, assume that for some integer k, the rotation of angle 2kπ/p keeps
A3 A2 A4
A2 A5 A6
A4 A6 A2
A1 A1 A1
A5 A3 A7
A7 A4 A3
A6 A7 A5
Fig. 1.45 Regular stellated polygons
the polygon A1 A2 . . . Ap invariant. Because p is prime, this rotation generates the

group of all rotations, so A1 A2 . . . Ap is invariant under all rotations. The rotation
that maps A1 to A2 should necessarily map A2 to A3 and so on, which implies that
A1 A2 . . . Ap is regular stellated.
So if a polygon is not regular stellated, then it is not invariant under any
nontrivial rotation. It follows that when rotating each such polygon by 2kπ/p,
k = 0, 1, 2, . . . , p − 1, p distinct polygons are obtained. An example is illustrated
in Fig. 1.46. We deduce that the orbit of each polygon under the action of the group
of rotations (namely, the polygon itself and its rotates) has p elements. It follows
that the polygons that are not stellated can be divided into disjoint groups of p. So
the number of equivalence classes of polygons that are not stellated is

1 (p − 1)! p − 1 1 [(p − 1)! + 1] − p
− = · ,
p 2 2 p 2
and for this number to be an integer, the numerator (p −1)!+1−p must be divisible
by p. But then (p − 1)! + 1 itself is divisible by p and the theorem is proved.

Try similar ideas in order to solve the following problems.
73 (Fermat’s Little Theorem) Let p be a prime number, and let n be a positive
integer. Prove that
np − n ≡ 0(mod p).
74 (Lucas’ Theorem) Let p be prime, and let a = a0 + a1 p + a2 p2 + · · · + ak pk ,

b = b0 + b1 p + b2 p2 + · · · + bk pk with aj , bj integers such that 0 ≤ aj , bj < p,
j = 1, 2, . . . , k. Prove that

a a0 a1 ak
≡ ··· (mod p).
b b0 b1 bk
66 1 Isometries
A7 A2 A1
A2 A1 A4
A5 A7 A2
A1 A4 A6
A3 A5 A7
A4 A6 A3
A6 A3 A5
A4
A6
A1
A3
A2
A5
A7
A6 A3 A5
A3 A5 A7
A4 A6 A3
A5 A7 A2
A1 A4 A6
A7 A2 A1
A2 A1 A4
Fig. 1.46 The seven rotates of a 7-gon
75 (Sylvester’s Theorem) Given two coprime positive integers a and b, show that
c = ab − a − b is the largest integer that cannot be represented as ax + by with x, y
non-negative integers. Also, show that for each positive integer n < c, exactly one
of the numbers n or c − n can be written as ax + by with x, y non-negative integers.
76 (The Cauchy-Davenport Theorem) Let p be a prime number. Show that if A
and B are two non-empty subsets of Zp then
|A + B| ≥ min(|A| + |B| − 1, p).

1.3.4 Functions
We will conclude the first chapter with problems about functions. The example
presented below is a problem from the 2013 Romanian Master of Mathematics,
which was proposed by Alexander Betts from the United Kingdom. Its solution was
found during the competition by Mark Sellke, member of the team that represented
the United States.
Problem 1.8 Does there exist a pair of functions g, h : R → R such that the only
function f : R → R satisfying f (g(x)) = g(f (x)) and f (h(x)) = h(f (x)) for all
x ∈ R is the identity function f (x) = x?
Solution The answer is yes! This is the example

x + 1 if {x} < 1 √ x
g(x) = 2 and h(x) = 2g √ ,
x + 12 if {x} ≥ 1
2 2
where {x} denotes the fractional part of x. We have to prove that g and h have the
in the range of g, then g(x) = x + 1, and if
required property. First, note that if x is √
x is in the range of h, then h(x) = x + 2.
At the heart of the solution lies the study of the behavior of a function f satisfying
the condition from
√ the statement under translations in the variable by numbers of
the form m + n 2, with m and n integers. We will make use of the following well-
known result, which we now prove for sake of completeness.

Theorem (Kronecker) A nontrivial subgroup of the additive group of real numbers
is either cyclic (meaning that it has one element so that any other element is an
integer multiple of it) or it is dense in the set of real numbers.
Proof Let the group be G. It is either discrete, or it has an accumulation point on
the real axis. If it is discrete, let a be its smallest positive element. Then any other
element is of the form b = ka + α with 0 ≤ α < a. But b and ka are both in G;
hence α is in G is well. By the minimality of a, α can only be equal to 0, and hence,
the group is cyclic.
If there is a sequence (xn )n in G converging to some real number, then ±(xn −xm )
approaches zero as n, m → ∞. Choosing the indices m and n appropriately, we can
find a sequence of positive numbers in G that converges to 0. Thus for any > 0,
there is an element c ∈ G with 0 < c < . For some integer k, the distance between
kc and (k + 1)c is less than ; hence, any interval of length contains some multiple
of c. Varying we conclude that G is dense in the real axis.

Applying this theorem we conclude that the set of numbers
√
{m + n 2 | m, n ∈ Z},
68 1 Isometries
√
is dense, being a group that is not cyclic because 1 and 2 cannot both be integer
multiples of the same number.
√ Consequently, when n ranges among all positive
integers, the numbers {n 2} are dense in the unit interval.
Lemma 1 The following properties hold:
(g1) If x is in the range of g then so is f (x).
(g2) If x is in the range of g then for all positive integers m, f (x + m) = f (x) + m.
Proof To prove (g1), let x = g(x0 ), then f (x) = f (g(x0 )) = g(f (x0 )), which is
therefore in the range of g. For (g2) start with
f (x + 1) = f (g(x)) = g(f (x)) = f (x) + 1,
then use induction to show that f (x + m) = f (x) + m.

A similar argument proves the next result.
Lemma 2 The following properties hold:
(h1) If x is in the range of h then so is f (x). √
(h2) If√x is in the range of h then for all positive integers n, f (x + n 2) = f (x) +
n 2.
We prove by contradiction that f (x) = x, by assuming that there are a = b with
f (a) = b.
Case 1: a is in the range of both g and h. Then so is b and f (a + n)= b+n for all
positive integers n. Since a and b are in the range of h, we have √a , √b <
2 2
1
2 . Then, unless the number
a−b
√ is an integer (and so √ a
= √ ), by using
b
2 2 2
Kronecker’s Theorem, we can find a positive integer n such that n · √1 is in
2
one of the intervals

1 b 1 a a b
− √ , − √ or 1 − √ , 1 − √ ,
2 2 2 2 2 2
(depending on which of the fractional parts is larger, either

the
first or the second
interval does not exist). If the first interval exists and n · √ belongs to it, then
1
2
a+n
√ 1
< 2 and b+n
√ ≥ 2 , so a + n is in the range of h while b + n is not.
1
2 2
But b + n = f (a + n), which contradicts (h1). The same reasoning rules out the
other case, so a−b
√ is an integer.
2
Similarly byKronecker’s Theorem, we obtain that there exists a positive integer
√ √
m such that a + m 2 < 12 and b + m 2 ≥ 12 , which violates (g1) because
√ √
f (a + m 2) = b + m 2 for all m, unless a − b is an integer.
Thus, both a − b and a−b√ should be integers, which is clearly impossible.
2
Case 2: a is in the range of g but not in the range of h. Then f(a + n) = b + n

for all positive integers n. There exists an n such that n · √1 is in the interval
2
1 − √a , 32 − √a , so a+n
√ < 12 and a + n is in the range of h. Then we
2 2 2
can use a + n as our a in Case 1 to rule out this case. The case where a is in the
range of h but not g is similar.
Case 3: a is neither of the ranges of g and h. We have f (g(a)) = g(a) = a + 12
and f (g(a)) = g(f (a)), which by the definition of g is equal to either f (a) + 1
or f (a) + 12 . But since f (a) = a, we must have a + 12 = f (g(a)) = f (a) + 1,
√
so f (a) = a − 12 . And we also have f (h(a)) = h(a) = a + 22 and f (h(a)) =
√ √ √
h(f (a)) = f (a) + 2 or f (a) + 22 . By the same reasoning f (a) = a − 22 ,
and this is a contradiction that rules out this case as well.
As we have ruled out all possible cases, we conclude that it is impossible for
f (a) not to be equal to a, and hence f (a) is always equal to a.

Try your hand at the following problems.

77 Prove that any function f : R → R can be written as the sum of two functions
whose graphs admit centers of symmetry.
78 Give an example of a function f : R → R whose graph is invariant under a
90◦ rotation about the origin. Show that for any such function f (0) = 0.
79 Let n > 2 be an integer, and let f : C → C be a function such that for every
regular polygon with n vertices, A1 A2 . . . An , in the plane
f (A1 ) + f (A2 ) + · · · + f (An ) = 0.
Prove that f is the zero function.

80 Let f : R → R be a continuous function whose graph has two axes of
symmetry. Prove that
f (x)
lim
x→∞ x
exists and is finite.
81 Find all polynomial functions P (x) with real coefficients that satisfy
√
P (x 2) = P (x + 1 − x 2 )
for all x with |x| ≤ 1.

82 (a) Let f : R → R be a function such that |f (x) − f (y)| = 1 for every
x, y ∈ R with |x − y| = 1. Is it necessarily true that f is an isometry?
(b) Let f : C → C be a function such that |f (z) − f (w)| = 1 for every z, w ∈ C
with |z − w| = 1. Is it true that f is an isometry?
Chapter 2
Homotheties and Spiral Similarities
2.1 A Theoretical Introduction to Homotheties
This chapter is devoted to transformations that dilate or shrink geometric objects by

a fixed factor. Such transformations occur naturally in real life, for example, when
changing scale, with or without rotation. Several situations are depicted in Fig. 2.1.
It is customary, when presenting this subject, to split the narrative into two parts:
first the story of the transformations that dilate without rotating, known as homoth-
eties (from the Greek words homo meaning “similar” and thesis meaning “position”)
and then that of the spiral similarities, which dilate and rotate simultaneously.
2.1.1 Definition and Properties
We commence with the definition of homothety, which is illustrated in Fig. 2.2.

Definition Given a point O and a nonzero number k, the homothety of center O
and ratio k is a transformation h of the plane that sends every point P to a point
P = h(P ) such that
−−→ −→
OP = k OP .
The homothety is called direct if k > 0, in which case P is on the ray |OP , and
inverse if k < 0, in which case O belongs to the segment P P . Note that in both
cases OP /OP = |k|. The center O is a fixed point of the homothety, and it is the
only fixed point (unless k = 1, when the homothety is the identity map).
Two geometric figures are called homothetic if there is a homothety that maps
one into the other.

72 2 Homotheties and Spiral Similarities
Fig. 2.1 Examples of polygons being shrunk or dilated
Fig. 2.2 The definition of P = h(P)

homothety
O Q Q = h(Q)
Being truthful to our plan to enlace synthetic and analytic reasoning, we rephrase
homothety in the language of numbers. If the homothety h is centered at the origin
and its ratio is k ∈ R, then it is defined by the equation h(z) = kz. In general, if the
complex coordinate of the center O is a, then the homothety is defined by
h(z) = a + k(z − a) = kz + (1 − k)a.
Indeed, this formula arises from the equation
h(z) − a
= k,
z−a
and the latter expresses the fact that the segment z − a is enlarged by a factor of k
to the segment h(z) − a. Conversely, every map
h : C → C, h(z) = kz + b
for k ∈ R\{0, 1} is a homothety of ratio k centered at the point a = b/(1 − k).

2.1 A Theoretical Introduction to Homotheties 73
Fig. 2.3 The image of a N

segment through a homothety
N
O M M
Most theoretical statements bellow are easier to attack analytically, since this
avoids configuration-dependent proofs. But with the problems, it will be quite the
opposite; geometric intuition yields sudden breakthroughs.
Theorem 2.1 If M and N are the images of M and N through the homothety of
center O and ratio k, then M N = kMN, and M N is parallel to MN. Moreover,
the image through the homothety of the segment MN is the segment M N , and the
image of the line MN is the line M N .
Proof The situation is exhibited in Fig. 2.3. Let a, m, n, m , n be the complex
coordinates of O, M, N, M , N , respectively. Then m = km + (1 − k)a and
n = kn + (1 − k)a. So n − m = k(n − m), which expresses analytically the
fact that M N is parallel to MN and has length kMN.
If z = tm + (1 − t)n, then the image z of z satisfies
z = k[tm + (1 − t)n] + (1 − k)a = t[km + (1 − k)a] + (1 − t)[kn + (1 − k)a].
For t ∈ [0, 1] we interpret this as saying that the segment MN is mapped to the
segment M N , while for t ∈ R, we interpret this as saying that the line MN is
mapped to the line M N . And so homothety maps lines to lines.

An important corollary of this result is that homothety dilates (or shrinks)
distances by a factor of |k|. The converse of Theorem 2.1 is also true, as the next
result shows.
Proposition 2.2 If a transformation h of the plane maps every segment to a segment
parallel to it and of length r times the length of the original segment, with r > 0
and r = 1, then h is a homothety.
Proof A solution in complex coordinates is again simpler, since it avoids the
discussion of cases. You may want to try the synthetic solution and compare. Let
h(z) = z , and h(w) = w . Then h(z) − h(w) = ±r(z − w) where the sign depends
on how the two segments are oriented with respect to each other. Fix w (and then
also w ) and write
h(z) = ±rz ∓ rw + w .
If the sign is always the same, then we are done, because we just obtained the
formula for a homothety (direct if the sign is plus and inverse if the sign is minus).
Now, assume that h(z1 ) = rz1 − rw + w and h(z2 ) = −rz2 + rw + w for some z1
and z2 . Then h(z2 ) − h(z1 ) = −r(z2 + z1 ) + 2rw. But h(z2 ) − h(z1 ) = ±r(z2 − z1 ).
If the sign is plus, then we must have z2 = w, while if the sign is minus, then we
must have z1 = w. Consequently there is a real number k = 0 such that r = |k| and
h(z) = kz − kw + w for all z = w. But then h(w) = w = kw − kw + w , so the
formula actually holds for all z. Thus h is a homothety.

Theorem 2.3
(i) Given a homothety, the image of a triangle is a triangle similar to it with the
similarity ratio equal to the absolute value of the ratio of the homothety. The
image of every polygon is a polygon similar to the original one with similarity
ratio equal to the absolute value of the ratio of the homothety.
(ii) Homothety preserves angles.

Proof Note that the sides of a triangle are mapped to segments parallel and in the
ratio |k| to them. Thus, a triangle is mapped to a triangle similar to it, with similarity
ratio |k|. Since every polygon can be decomposed into triangles, the same is true for
polygons. This proves (i).
For (ii), place the angle in a triangle. The property follows from the fact that
similar triangles have equal angles.

As a corollary, if a homothety maps one polygon to another, then the ratio of the
areas of the two polygons is equal to the square of the ratio of the homothety.
Theorem 2.4 Homothety maps a circle to a circle of radius equal to the radius of
the original circle multiplied by the absolute value of the ratio of the homothety.
Proof Let the homothety be h(z) = kz + (1 − k)a. The equality
|z − α| = R
that defines the circle is equivalent to
|(kz + (1 − k)a) − (kα + (1 − k)a)| = kR.
This means that z satisfies |z−α| = R if and only if z = h(z) satisfies |z −h(α)| =
kR. So z belongs to the circle of center α and radius R if and only if h(z) belongs
to the circle of center h(α) and radius kR.

The image ω of a circle ω through a homothety of center O is illustrated in
Fig. 2.4. Note that if the center of homothety is exterior to ω, it must also be exterior
to ω , and in this case the common tangents of the circles pass through the center
of homothety (this is a consequence of the fact that the line joining a point and its
image passes through the center of homothety).
Fig. 2.4 The image of a

circle through a homothety
ω
ω
O
2.1.2 Groups Generated by Homotheties
Following the pattern from the previous chapter, we analyze the behavior of
homotheties under compositions and likewise interpret the results in the language
of group theory. First, a simple observation.
Theorem 2.5 For a fixed point O, the homotheties with center O form a group,
the inverse of a homothety with ratio k is the homothety of ratio 1/k, and the
composition of the homotheties of ratios k1 , k2 is the homothety of ratio k1 k2 .
But if we look at all homotheties, then they do not form a group. In fact, we have
the following result.
Theorem 2.6 (Composition of Homotheties) If h1 and h2 are homotheties of
centers A and B and ratios k1 and k2 , respectively, then h2 ◦ h1 is
(i) the homothety of ratio k1 k2 and center a point O on the line AB so that
OA/OB = (1 − k2 )/[(1 − k1 )k2 ], the bar denoting oriented segments
−→
(ii) the translation of vector (1 − k2 )AB if k2 = 1/k1 .
Consequently, if h2 ◦ h1 is a homothety, then its center is collinear with the centers
of h1 and h2 .
Proof 1 Let a and b be the coordinates of A and B, respectively. The two
homotheties are
h1 (z) = k1 z + (1 − k1 )a and h2 (z) = k2 z + (1 − k2 )b,
and their composition is
h2 ◦ h1 (z) = k2 [k1 z + (1 − k1 )a] + (1 − k2 )b = k1 k2 z + (1 − k1 )k2 a + (1 − k2 )b.
If k1 k2 = 1, this is further equal to


(1 − k1 )k2 1 − k2
k1 k2 z + (1 − k1 k2 ) a+ b ,
1 − k1 k2 1 − k1 k2
whose center has coordinate

(1 − k1 )k2 1 − k2
a+ b
1 − k1 k2 1 − k1 k2
which proves (i). If k1 k2 = 1, then

1 − k1 1 k1 − 1
h2 ◦ h1 (z) = z + a+ 1− b=z+ (b − a)
k1 k1 k1
= z + (1 − k2 )(b − a),
proving (ii).

Proof 2 Here is a synthetic proof, albeit without the precise location of the center
of homothety. Let us first choose a point C that does not belong to the line AB.
Let B = h1 (B) and C = h1 (C); then B C is parallel to BC. Set A = h2 (A),
B = h2 (B), and C = h2 (C). Then the sides of A B C are parallel to the sides
of AB C and hence to the sides of ABC (Fig. 2.5).
Assume first that k1 k2 = ±1. Problem 84 below shows that there is a homothety
h that maps ABC to A B C . Its ratio is k1 k2 . Note that A and B belong to the
line AB, and so h(AB) = AB. This means that the center O of the homothety is
on AB. The point O is a fixed point of h = h2 ◦ h1 . But h can only have one fixed
point, since for any two points, the distance from h(X) to h(Y ) is |k1 k2 |XY = XY .
Thus, the same homothety appears for all points C that are not on AB. If C is on
AB, let D be a point that does not belong to AB. Set C = h(C) and D = h(D).
Then the triangles A C D and ACD have parallel sides, so they are homothetic.
And the homothety has ratio k1 k2 and center at the intersection of AA and DD , so
it must be h.
Fig. 2.5 The composition of C

two homotheties
A A B B B
If k1 k2 = ±1, then the composition is an isometry, and we are in the environment

of the previous chapter. The composition is therefore a translation (when k1 k2 =
1), or reflection over a point (when k1 k2 = −1), the latter case being an inverse
homothety.

The collinearity proved in this theorem will play a major role in our story; we
will highlight its significance later (stay alert!).
Theorem 2.7 Let h be the homothety of center A and ratio k = 1, and let τ be the
translation of vector −
→
v . Then
k −→
(i) h◦τ is the homothety of ratio k and whose center is the translate of A by 1−k v,
1 −→
(ii) τ ◦h is the homothety of ratio k and whose center is the translate of A by
1−k v.
Proof Let the complex coordinate of A be a, and let b be the complex number
representing −
→
v , so that τ (z) = z + b. Also, let h(z) = kz + (1 − k)a. Then

k
h ◦ τ (z) = k(z + b) + (1 − k)a = kz + (1 − k) a + b .
1−k
Also

1
τ ◦ h(z) = kz + (1 − k)a + b = kz + (1 − k) a + b .
1−k
The proposition is proved.

You can also write a synthetic proof similar to that of Proposition 2.6. We are
now able to state a fundamental result.
Theorem 2.8 The group of transformations of the plane that take every line to a
line parallel to it consists of all homotheties and all translations.
Proof By Theorems 2.5–2.7, the set consisting of all homotheties and translations
is a group. Since the image of a line through a translation or a homothety is a line
parallel to it, every element of this group maps a line to a line parallel to it.
Conversely, let f be a transformation of the plane that maps every line to a line
parallel to it, and let A and B be two distinct points. Compose f with a translation τ
such that τ ◦ f (A) = A (if f (A) = A take τ to be the identity map). Then τ ◦ f (B)
belongs to the line AB and is a point different from A. Compose with the homothety
h centered at A such that h ◦ τ ◦ f (B) = B (if τ ◦ f (B) = B then h is just the
identity map).
Now let g = h ◦ τ ◦ f . Then g(A) = A, and g(B) = B, and g also maps every
line to a line parallel to it. Let C be a point that does not belong to AB. Then each
of the lines AC and BC is mapped into itself, and so the intersection of these lines
must be mapped into itself. Thus g(C) = C for all C that do not belong to AB.
Next, let C ∈ AB and let D, E be points that are not on AB such that C does not
belong to DE. Then, as seen above, g(D) = D, and g(E), and applying the same
reasoning for the triple D, E, C, we deduce that g(C) = C. Thus g is the identity
map. This shows that f = τ −1 ◦ h−1 , so f is a composition of a translation and a
homothety.

A consequence of this theorem is that if a transformation maps every line to
a line parallel to it, and if one can find two points A and B with images A
and B , respectively, such that AA and BB intersect at some point O, then the
transformation is a homothety of center O.
In view of Theorem 2.8 and the results preceding it, the smallest group that
contains all homotheties necessarily contains all translations. Note also that direct
homotheties and translations form a subgroup of this group.
2.1.3 Problems About Properties of Homotheties
We continue with a short list of theorems given as exercises, some of which will
play an important role in what follows.
83 Given two distinct circles, how many homotheties map them one into the other?
84 (a) For two noncongruent triangles with parallel sides, show that there exists a
homothety that maps one into the other.
(b) For two noncongruent polygons whose sides and diagonals are respectively
parallel, show that there exists a homothety that maps one into the other.
85 What is the largest number of homotheties that can map two polygons into each
other?
86 Let ω1 , ω2 , ω3 be three nonconcentric, noncongruent circles. Prove that:
(a) (Monge’s Theorem) The centers of the direct homotheties of the pairs (ω1 , ω2 ),
(ω1 , ω3 ), and (ω2 , ω3 ) are collinear.
(b) (d’Alembert’s Theorem) The six centers of the homotheties of the pairs (ω1 , ω2 ),
(ω1 , ω3 ), and (ω2 , ω3 ) lie on four lines.
2.2 Problems in Euclidean Geometry That Use Homothety
2.2.1 Theorems in Euclidean Geometry Proved Using

Homothety
We now explain how to apply what we have learned about homotheties.

2.2 Problems in Euclidean Geometry That Use Homothety 79
Fig. 2.6 Euler’s line A
H
G
O
B A C
I. Euler’s Line and the Nine-Point Circle

The following theorem was proved by Leonard Euler in 1765.
Theorem 2.9 (Euler’s Line) In a triangle the orthocenter, the centroid, and the
circumcenter are collinear. Moreover, the centroid lies on the segment joining the
circumcenter and the orthocenter and divides this segment in the ratio 1 : 2.
Proof Figure 2.6 describes the situation. Let the triangle be ABC, and let G, H, O
be its centroid, orthocenter, and circumcenter, respectively.
Consider the homothety h of center G and ratio − 12 . Then h maps the vertex
A to the midpoint A of BC, and it maps the altitude from A to the perpendicular
bisector of BC. Reasoning similarly for the other altitudes, we deduce that h maps
the intersection of the altitudes to the intersection of the perpendicular bisectors of
the sides, thus h(H ) = O. Since a point and its image are collinear with the center
of homothety, it follows that H, O, and G are collinear. Moreover, since h is an
inverse homothety, G is between H and O, and GO/GH is equal to the absolute
value of the homothety ratio, which is 12 . The theorem is proved.

Remark If we place the origin of a complex coordinate system at the circumcenter
O, with the coordinates of the vertices being a, b, c, then the coordinate of the
orthocenter H is a + b + c. This is because H is the image of G under a homothety
of center O and ratio 3, and the coordinate of G is a+b+c
3 .
Theorem 2.10 (The Nine-Point Circle) In a triangle, the midpoints of the sides,
the feet of the altitudes, and the midpoints of the segments that join the orthocenter
and the vertices are on a circle. Moreover, the center of this circle lies on Euler’s
line, and the orthocenter, the center of the nine-point circle, the centroid, and the
circumcenter form a harmonic division.
Four points X, Y, Z, W that appear in this order on a line form a harmonic
division if
YX WX XY ZY
: = 1 ⇐⇒ : = 1.
YZ WZ XW ZW
In complex coordinates, the four collinear points form a harmonic division if their
coordinates satisfy
Fig. 2.7 The nine-point A

circle
E
P N
F H
B D C
M
D M
y−x w−x x−y z−y

: = −1 ⇐⇒ : = −1
y−z w−z x−w z−w
namely, if their cross-ratio is −1.

Proof The proof of the theorem can be followed on Fig. 2.7. The midpoints of the
segments that join the orthocenter with the vertices are on a circle that is mapped
to the circumcircle by the homothety h centered at the orthocenter and of ratio 2.
Thus it is natural to show that this homothety maps the other points in question to
the circumcircle, too.
Let therefore the triangle be ABC, H its orthocenter, M, N, P the midpoints
of BC, CA, AB, and D, E, F the feet of the altitudes from A, B, C, respectively.
Consider a system of complex coordinates centered at the circumcenter O, and let
a, b, c be the coordinates of A, B, C, respectively. In view of the above observation,
the orthocenter has coordinate a + b + c. Then M = h(M) is the reflection of H
over M, and since M has the coordinate equal to b+c
2 , the complex coordinate of M

is b + c − (a + b + c) = −a. Thus, M is the point diametrically opposite to A in
the circumcircle.
The point D = h(D) is the reflection of H over BC, and this reflection is
the composition of the reflection over the perpendicular bisector of BC with the
reflection over M. Thus D is the reflection of M over the perpendicular bisector of
the chord BC of the circumcircle, so it is itself on the circumcircle. We conclude that
h(M) and h(D) are on the circumcircle, and the same is true for the other points.
It follows that the nine points are on the image of the circumcircle through h−1 , so
they are concyclic.
Moreover, the image of the circumcenter is the center of this nine-point circle,
and it is the midpoint of the segment OH . It therefore has the complex coordinate
a+b+c
2 . Thus the orthocenter, center of the nine-point circle, the centroid, and the
circumcenter have the coordinates
a+b+c a+b+c
a + b + c, , , 0.
2 3
We compute
(a + b + c)/2 − (a + b + c) 0−a+b+c −1/2 −1

: = : = −1,
(a + b + c)/2 − (a + b + c)/3 0 − (a + b + c)/3 1/6 −1/3
showing that the four points form a harmonic division.

And here is another proof of
Theorem 1.22 Let ABC be a triangle. Then the symmedian from B passes through
the point where the tangents to the circumcircle at A and C intersect.
Proof Let W be the intersection point of the tangents. We step aside and explore the
nine-point circle of some triangle ABC with orthocenter H . We have just learned
that the nine-point circle passes through the midpoints M, N , and P of the sides
BC, CA, and AB; the midpoints X, Y , and Z of AH , BH , and CH ; and the feet D,
E, and F of the altitudes through A, B, and C, respectively. Because Y EN = 90◦ ,
Y N is a diameter. In particular, Y DN = Y F N = 90◦ (Fig. 2.8).
Now consider the circle of diameter BH , whose center is Y . Because BDH =
BF H = 90◦ , it is also the circumcircle of DBF . Since Y D ⊥ DN and Y F ⊥
F N, ND and NF are tangents from N to this circle.
We are ready to finish the proof: since AF C = ADC = 90◦ , AF DC is
cyclic, and the power of the point B with respect to its circumcircle gives BA·BF =
BC · BD. Hence BD/BF = AB/BC. This means that DBF is similar to ABC,
and this similarity takes the median BN to its reflection over the bisector of B,
and this reflection is BN. Consequently BN is a symmedian in the triangle DBF ,
and N D and N F are tangent to the circumcircle of BDF . Now just change the
notation, trading D with A, F with C, and N with W .
Fig. 2.8 Symmedians and B

tangents
Y
D
P M
F
H
X
Z
A E N C
The nine-point circle will play an important role in many of the problems from
this chapter. We will encounter it again briefly when we talk about inversion.
II. The Incircle and the Excircle
Homothety provides the natural setting for introducing the excircles of a triangle.
Let ABC be the triangle, let I be its incenter, and let D, E, F be the points of
tangency of the incircle with the sides BC, AC, AB, respectively.
Consider the point J on the incircle that lies diametrically opposite to D and,
with it, the homothety ha of center A that maps the tangent to the incircle at J to
the line BC. Such a homothety exists because the two lines are parallel. Then the
incircle is mapped to a circle that is exterior tangent to BC and is also tangent to
AB and AC. This is one of the three excircles of the triangle, the other two are
obtained by performing the same construction with respect to the sides AC and AB.
The excircle corresponding to side BC is shown in Fig. 2.9. It is standard to denote
its center by Ia , with the centers of the other two excircles being denoted by Ib and
Ic .
The ratio of the homothety ha is the result of a short computation. To do it, let
AB = c, AC = b, and BC = a. Let also E and F be the images of E and F
through the homothety, and let K be the point where the excircle is tangent to BC.
Because the tangents from a point to a circle are equal, we can set AE = AF = x,
BD = BF = y, CD = CF = z, BK = BF = u, and CK = CE = v. We have
x + y = c, x + z = b, y + z = a, u + v = a, and u + c = v + b. Solving the system
of equations, we obtain
E K
Ia
J I D
A
F
B
F
Fig. 2.9 The excircle corresponding to the vertex A

b+c−a a+c−b a+b−c a+b−c a+c−b

x= ,y = ,z = ,u = ,v = .
2 2 2 2 2
The homothety ratio is therefore
AF AF x b+c−a
= = = .
AF AB + BF c+u b+c+a
This is the ratio of the inradius by the radius of the excircle. But from this
computation, we can also infer that BD = CK, so D and K are symmetric with
respect to midpoint of the side BC.
Proposition 2.11 Let ABC be a triangle, let D, K ∈ BC be the points where the
incircle and the excircle are tangent to the side, let I and Ia be the centers of these
circles, and let R be the midpoint of the altitude from A. Then R, I, K and R, D, Ia
are collinear.
Proof We add to the picture the foot N of the altitude from A, the point J that
is diametrically opposite to D in the incircle, and the point L that is diametrically
opposite to K in the excircle. Then the homothety ha defined above maps DJ to
LK, and so A, D, L and A, J, K are collinear. It follows that the triangles KDJ
and KNA are homothetic under a homothety of center K, and so the midpoint R of
AN, the midpoint I of J D, and K are collinear. Similarly, the triangles DAN and
DKL are homothetic under a homothety of center K, so the midpoint R of AN, the
midpoint Ia of LK, and D are collinear.

III. The Theorems of Pappus and Menelaus
We begin with the first version of Pappus’ Theorem.
Theorem 2.12 (Pappus) Let 1 and 2 be two parallel lines, let A1 , B1 , C1 be
distinct points on 1 , and let A2 , B2 , C2 be points on 2 . If P = B1 C2 ∩ C1 B2 ,
Q = C1 A2 ∩ A1 C2 , and R = A1 B2 ∩ A2 B1 , then P , Q, R are collinear.
Proof The theorem is illustrated in Fig. 2.10. We need the following lemma.
Fig. 2.10 Pappus’ Theorem A2 B2 C2

2
P
R Q
1 A1 B1 C1
Lemma 2.13 Let hP , hQ , hR be three homotheties of centers P , Q, R and ratios

p, q, r, respectively, such that
hP ◦ hQ ◦ hR = hR ◦ hP ◦ hQ or hP ◦ hQ ◦ hR = hR ◦ hQ ◦ hP .
Assume additionally that pq = 1. Then the points P , Q, R are collinear.

Proof At the heart of the proof lies Theorem 2.6. Let us assume, for example, that
hP ◦ h Q ◦ h R = h R ◦ h Q ◦ h P .
Because pq = 1, by the abovementioned theorem, hP ◦ hQ and hQ ◦ hP are

homotheties; denote their respective centers by S and T . By the same theorem, the
points S, T are on the line P Q. We have two possibilities.
If pqr = 1, then the transformation hP ◦ hQ ◦ hR is the translation of vector
−
→ −→
(1 − pq)RS = (1 − r)T R,
and it follows that R ∈ ST = P Q.

If pqr = 1, let U be the center of the homothety hP ◦ hQ ◦ hR = hR ◦ hQ ◦ hP .
Then U is on both lines SR and RT and if R ∈ ST = P Q, then R = U . On the
other hand, from Theorem 2.6, (i) we deduce that
−→ −→ −→ −→
pq(1 − r)U R = −(1 − pq)U S and r(1 − pq)U T = −(1 − r)U R.
But this can only happen if U = S = T , which is false. It follows that R ∈ ST =

P Q, and the lemma is proved.

Returning to the proof of the theorem, let
P C1 P B1 QA2 QC2 RA1 RB1

p= = , q= = , r= = ,
P B2 P C2 QC1 QA1 RB2 RA2
where the overline notation stands for oriented segments (so that if P is between
A1 and B2 , then the ratio is negative; in the situation from Fig. 2.10 p, q, r < 0).
Next, consider the homotheties hP , hQ , hR of centers P , Q, R and ratios p, q, r,
respectively. Note, for example, that hP (B2 ) = A1 and hQ (A1 ) = C2 . If pq =
qr = 1, then it follows easily that the lines P Q and QR are parallel to 1 and 2
respectively, thus P , Q, R are collinear. Otherwise, if say pq = 1, we have two
cases.
If pqr = 1, then the transformations hP ◦ hQ ◦ hR and hR ◦ hQ ◦ hP have dilation
factor 1 so they are translations, and they both take B2 to B1 . So both must be the
−−−→
translation of vector B2 B1 , and in particular they are equal.
hP ◦ hQ ◦ hR = hR ◦ hQ ◦ hP .
If pqr = 1, then hP ◦ hQ ◦ hR and hR ◦ hQ ◦ hP are homotheties of ratio pqr

that again take B2 to B1 . But there is only one point O on the line B1 B2 such that,
with oriented segments, OB2 /OB1 = pqr. This point O must be the center of both
homotheties, so again
hP ◦ h Q ◦ h R = hR ◦ h Q ◦ h P .
In both cases we can apply the lemma to conclude that P , Q, R are collinear.
If, instead, qr = 1, we are in the other hypothesis of the lemma, and we obtain
the same conclusion.

This proof requires that the lines are parallel, but the theorem is true even when
the lines intersect.
Theorem 2.14 (Pappus) Let 1 and 2 be two intersecting lines, let A1 , B1 , C1 be
distinct points on 1 , and let A2 , B2 , C2 be points on 2 . If P = B1 C2 ∩ B2 C1 ,
Q = A1 C2 ∩ A2 C1 , and R = A1 B2 ∩ A2 B1 , then P , Q, R are collinear.
The most natural approach is by two-dimensional real projective geometry. In
that setting, the fact that the lines intersect is irrelevant, as we can move the
intersection point at infinity. We land in the case we just proved, and we are
done. But to be true to the scope of this book, we must write a proof that avoids
transformations of the real projective plane.
The geometric configuration of the general Pappus’ Theorem hides many
homotheties in a subtle way. The geometry afficionado probably already thinks
about Menelaus’ Theorem, which is the most common tool for checking collinearity.
And this theorem finds its most natural setting in the realm of homotheties.
Theorem 2.15 (Menelaus’ Theorem) Let ABC be a triangle, and let A1 , B1 , C1
be points on BC, CA, AB, respectively. Then the points A1 , B1 , C1 are collinear if
and only if, with oriented segments, the following equality holds:
A1 B B1 C C1 A
· · = 1,
A1 C B 1 A C1 B
We can rephrase this using complex numbers without mentioning oriented

segments by saying that
b − a1 c − b1 a − c1
· · = 1,
c − a1 a − b1 b − c1
with the lowercase letter being the coordinate of the uppercase point. The configu-
ration of the Menelaus’ Theorem is shown in Fig. 2.11.
Proof To avoid oriented segments, let us stay in the world of complex numbers.
Consider the homothety ha of center A1 and ratio ka that maps C to B, the
homothety hb of center B1 and ratio kb that maps A to C, and the homothety hc
of center C1 and ratio kc that maps B to A. Then
Fig. 2.11 Menelaus’ A

Theorem
C1 B1
A1
B C
b − a1 c − b1 a − c1
ka = , kb = , kc = .
c − a1 a − b1 b − c1
By Theorem 2.6, the composition ha ◦ hb ◦ hc is either a homothety or a translation.

It is not a translation, because it has a fixed point: ha ◦ hb ◦ hc (B) = B. So it is a
homothety. By the same theorem, the center of the homothety is on the line of the
points A1 , B1 , C1 and so is different from B and is another fixed point. A homothety
with two fixed points is the identity map. Consequently ka kb kc = 1, that is,
b − a1 c − b1 a − c1
· · = 1.
c − a1 a − b1 b − c1
For the converse, consider the same homotheties, and notice that the given
relation implies that ha ◦ hb ◦ hc has ratio 1. As this transformation has B as a fixed
point, it is not a translation, so it is the identity map. This means that h−1
c = ha ◦ hb ,
and, by Theorem 2.6, ha , hb , hc have collinear centers.

By applying repeatedly Menelaus’ Theorem, we produce a proof of Pappus’
Theorem with a large number of homotheties embedded in it.
Proof 1 Let X, Y, Z be the intersection points of the pairs of lines (A1 C2 , C1 B2 ),
(B1 A2 , A1 C2 ), and (C1 B2 , B1 A2 ), respectively, as shown in Fig. 2.12. Using
complex coordinates with the usual convention for upper-/lowercase letter, we write
the relation from Menelaus’ Theorem for the triangle XY Z and the transversals
A1 B2 , B1 C2 , C1 A2 , 1 , and 2
x − a1 y − r z − b2 y − b1 z − p x − c2
· · = 1, · · = 1,
y − a1 z − r x − b2 z − b1 x − p y − c2
z − c1 x − q y − a2
· · = 1,
x − c1 y − q z − a2
x − c1 z − b1 y − a1 y − c2 x − b2 z − a2
· · = 1, · · = 1.
z − c1 y − b1 x − a1 x − c2 z − b2 y − a2
Multiply the five relations to obtain
y−r z−p x−q

· · = 1,
z−r x−p y−q
C2
B2
X
P
A2
Q
1 R Y
A1 B1 C1
2 Z
Fig. 2.12 The second version of Pappus’ Theorem
which, by applying Menelaus’ Theorem in reverse to the same triangle, implies that
P , Q, R are collinear.
Proof 2 We can decrypt the meaning of these algebraic computations. Like in the
proof of Menelaus’ Theorem, behind each fraction lies a homothety. For example,
behind the first fraction lies a homothety ha1 (w) = ra1 w + sa1 of center A1 and that
maps Y to X. Here ra1 is a nonzero real number and sa1 is a complex number. We
use the same convention to associate to each fraction in the first three products a
homothety indexed by the coordinate of the center written as a linear function with
coefficients indexed by the same coordinate. The five equalities correspond, via the
same argument as in the proof of the direct implication of Menelaus’ Theorem, to
ha1 ◦ hr ◦ hb2 = 1, hb1 ◦ hp ◦ hc 2 , hc1 ◦ hq ◦ ha2 = 1,

−1 −1
h−1 −1
c1 ◦ hb1 ◦ ha1 = 1, h−1 −1
c2 ◦ hb2 ◦ ha2 = 1,
where 1 stands for the identity transformation. By composing with the appropriate
transformations on the left and right, we can change these equalities into
−1
hr = h−1
a1 ◦ hb2 , hp = h−1 −1
b1 ◦ hc 2 , hq = h−1 −1
c 1 ◦ ha2 ,
h−1
c 1 = ha1 ◦ hb1 , h−1
c 2 = ha2 ◦ hb2 .
Substitute h−1 −1
c1 and hc2 in the first equations, and then compose to obtain
hp ◦ hq ◦ hr = h−1 −1 −1 −1
b1 ◦ ha2 ◦ hb2 ◦ ha1 ◦ hb1 ◦ ha2 ◦ ha1 ◦ hb2 .
For arbitrary functions, this is not the identity map. But here we work with affine
transformations of the plane, and so the right-hand side is a affine transformation of
the form f (w) = rw + s with
r = rb−1
1
ra2 rb2 ra1 rb1 ra−1 r −1 r −1 = 1.
2 a1 b2
But hp ◦ hq ◦ hr maps Z to Z, so f has a fixed point. This implies s = 0, and

therefore f is the identity map. We conclude that hp = h−1 −1
r ◦ hq , so the centers
P , Q, R of the three homotheties are collinear, by Theorem 2.6.

It is worth pausing for a moment to allow the meaning of these examples to sink
in. Homotheties can be parametrized by a pair consisting of a real number and a
complex number. For a synthetic point of view, we choose as parameters the ratio
k and the (coordinate of) the center a. For the analytic point of view, we choose
different parameters, k and b, such that the homothety is written as h(z) = kz + b.
The composition rule is simpler in the second parametrization, for if h1 (z) =
k1 z + b1 , h2 (z) = k2 z + b2 , and h2 ◦ h1 (z) = kz + b, then k = k1 k2 , and b =
k2 b1 + b2 . It is significantly more complicated to write the composition rule in
the first parametrization, as Theorem 2.6 tells us that if the centers and ratios of
h1 , h2 , and h2 ◦ h1 are k1 , k2 , k and a1 , a2 , a, respectively, then k = k1 k2 , while
a is characterized by the geometric conditions of lying on the line passing through
a1 and a2 and forming with a1 and a2 segments that are in a certain ratio. But
it is this more geometrically intuitive parametrization, or rather the composition
rule transcribed for the parameters, that empowers us with one of the best tools for
proving collinearity, in the guise of Theorem 2.6. Both the Menelaus’ Theorem and
the Monge-d’Alembert Theorem (Problem 86) are immediate consequences of the
composition rule for homotheties.
Groups that admit (local) parametrizations by real numbers in which the compo-
sition rule and the taking of the inverse are continuous maps are called Lie groups
(after Marius Sophus Lie). The group of homotheties and translations is therefore a
Lie group (parametrized by R3 ). We see how just by writing the composition rule in
the appropriate parametrization we obtain powerful theorems in classical geometry.
2.2.2 Examples of Problems Solved Using Homothety
We start with an easy example.

Problem 2.1 The trapezoids ABCD and AP QD share the base AD, while BC
and P Q have different lengths. Show that the intersection points of the pairs of
lines (AB, CD), (AP , DQ), and (BP , CQ) are collinear.
Solution Denote the intersections of the three pairs by K, L, M, respectively, as
in Fig. 2.13. They are the centers of the direct homotheties h1 , h2 , and h3 that
transform the segments BC into AD, AD into P Q, and BC into P Q, respectively.
Then h3 = h2 ◦ h1 , so by Theorem 2.6, the centers K, L, M of these homotheties
are collinear.

A
B
P
K L M
Q
C
D
Fig. 2.13 Trapezoids sharing a side
Fig. 2.14 The orthic triangle A

of MN P
N
D
P
I
O
H
E F
B M C
The next problem has appeared in the Iranian Mathematical Olympiad in 1995,
the Kürschák Competition in 1997, and the Hungary-Israel Competition in 2000.
Problem 2.2 In a triangle ABC, let O be the circumcenter, let I be the incenter,
and let M, N, P be the points where the incircle touches the sides BC, CA, and
AB, respectively. Let also H be the orthocenter of MNP . Prove that I, O, and H
are collinear.
Solution 1 The first solution that we present was found when this problem was
discussed at the training program of the Bulgarian International Mathematical
Olympiad Team in 2002. Let D, E, F be the feet of the altitudes from M, N, P ,
respectively, of the triangle MNP , as shown in Fig. 2.14. The points E and F are
on the circle of diameter NP , so P NM = MEF . Using inscribed angles in
the incircle, we have P NM = BMP . Hence BMP = MEF , showing that
EF and BC are parallel. Similarly, DF AC and DEAB. The triangles ABC and
DEF have parallel sides, so by Problem 84, they are homothetic. Let S be the center
of the homothety.
The incenter I of ABC is mapped by the homothety to the incenter of DEF ,
which is H , so S, I, H are collinear. Since I is the circumcenter of MNP , the
line I H is the Euler line of this triangle, whose line contains also the circumcenter
Fig. 2.15 The triangle A

M N P
N
N
P
P
H O
I
B M C
O of this triangle. As the homothety maps O to O , S, O, O are also collinear,

showing that O belongs to the same Euler line. Thus I, O, H are on the Euler line
of MNP , so they are collinear.

Solution 2 Let M , N , P be the points where the angle bisectors of A, B, C
intersect the circumcircle, respectively (Fig. 2.15). Then
1 1
(BN , M P ) = P AN + B M = A/2 + B/2 + C/2 = 90◦ .
2 2
Hence M P is perpendicular to the angle bisector of B. But so is MP . So

M P MP . Similarly, M N MN, N P NP . From here we also deduce that I
is the orthocenter of M N P .
The triangles MNP and M N P have parallel sides, so by Problem 84, they
are homothetic. Let T be the center of the homothety. Then O (the circumcenter
of M N P ), I (the circumcenter of MNP ), and T are collinear. Also I (the
orthocenter of M N P ), H (the orthocenter of MNP ), and T are collinear.
Consequently, T , O, I, H are collinear, and the problem is solved.

One should point out that in this configuration, the lines MM , NN , P P
intersect at the center S of the first homothety, while the lines AD, BE, CF intersect
at the center T of the second homothety, and these intersection points are on the
Euler line of MNP .
The method of these two solutions is to be remembered: If two triangles have
parallel sides, then they are homothetic, and the lines connecting their corresponding
vertices intersect at the center of homothety.
The next problem was given at the International Zhautykov Olympiad in 2011 in
Kazakhstan.
Problem 2.3 Given a convex quadrilateral, draw two segments that join points on a
pair of opposite sides and do not intersect. Draw two other segments that join points
on the other pair of opposite sides and do not intersect. We know that the points of
intersection of the first two segments with the last two segments lie on the diagonals
of the quadrilateral, and that, of the 9 quadrilaterals they determine, those in three
corners and the one in the middle admit inscribed circles. Show that the quadrilateral
in the fourth corner also admits an inscribed circle.
Solution The solution needs two auxiliary results.
Lemma 2.16 Let ABC be a triangle, and let ω and ω1 be two circles that are
tangent to the sides AB and AC and intersect the side BC. If the arcs on the circles
ω and ω1 that lie outside the triangle ABC have equal measures, then the two
circles coincide.
Proof Examining Fig. 2.16, we can see that there is a homothety of center A that
maps ω and ω1 . Arguing by contradiction, let us assume that the circles do not
coincide and that ω has smaller radius than ω1 . Then the arc of ω that lies outside

of triangle ABC (DE in the figure) is mapped by the homothety to an arc that lies

strictly in the interior of the arc of ω1 that lies outside of triangle ABC (D1 E1 in the
figure). Homothety preserves the measures of arcs (since it preserves the measure
of angles), so the image has the same measure as the original arc, and hence the
arc of ω that lies outside of the triangle ABC has measure strictly less than the arc
of ω1 that lies outside of the triangle. This is a contradiction, and the conclusion
follows.
Lemma 2.17 (Igor Voronovich) Let ABCD be a convex quadrilateral, and let ω1
be a circle tangent to the sides AB and AD and ω2 a circle tangent to the sides
CB and BD. Then the quadrilateral ABCD can be circumscribed to a circle if and
only if the arcs of the circles ω1 and ω2 that lie on the same side of diagonal AC
have equal measures.
Fig. 2.16 Proof that ω and A

ω1 coincide
ω C
E1
E
B D
D1 ω1
Fig. 2.17 Proof of the ω2

Voronovich Lemma A D
ω1
ω2
ω1
B C
Proof Consider the homothety of center A that maps ω1 to a circle ω1 that is tangent
to BC and the homothety of center C that maps ω2 to a circle ω2 that is tangent to
AB. These new circles are shown in Fig. 2.17.
If ABCD is a circumscribed quadrilateral, then ω1 = ω2 , so the arcs that lie
on the same side of AC on ω1 and ω2 are equal. Then, since homothety preserves
the measure of arcs, the arcs on the same side of AC of ω1 and ω2 have also equal
measures. This proves the direct implication.
For the converse implication, if the arcs of ω1 and ω2 that lie on the same side of
the diagonal AC as the vertex D have the same measure, then the same is true for
the circles ω1 and ω2 , since again homothety preserves the measure of arcs. Using
Lemma 2.16 (applied to the triangle BAC and the circles ω1 and ω2 ), we obtain
that ω1 = ω2 , and this is the circle inscribed in ABCD. The Voronovich Lemma is
proved.

Returning to the problem, we reason on Fig. 2.18 and notice that the circles 2 and
3 are inverse homothetic with respect to the point of intersection of two segments
that join opposite sides that lies on the diagonal between the two circles. The same
is true for the circles 3 and 4. So the circles 2 and 4 are directly homothetic, and
the center of the homothety lies on the line of support of the diagonal that joins
the vertices that correspond to these circles. This enforces the conditions of the
Voronovich Lemma (Lemma 2.17) for these two circles, and by the lemma, the
original quadrilateral can be circumscribed to a circle.
Draw the circles 5 and 5 so that each is tangent to a pair of adjacent sides
of the quadrilateral where the fifth circle is supposed to be inscribed, as shown in
the figure. Like before, the circles 1 and 3 are inverse homothetic with respect to
the point that lies at the intersection of the corresponding diagonal with the two
segments that join opposite sides, and the same is true for the circles 3 and 5 , so the
circles 1 and 5 are directly homothetic, and the center of the homothety lies on the
line of support of the diagonal. It follows that the measures of the arcs of circles 1
and 5 that lie on the same side of the diagonal are equal, being the images of each
other by a direct homothety.
Fig. 2.18 Proof that the fifth

quadrilateral has an inscribed 2
circle
1
3
4 5
On the other hand, by applying the Voronovich Lemma to the big quadrilateral,
we deduce that the measures of the arcs of circles 1 and 5 that lie on the same side
of the diagonal are equal. Thus, the measures of the arcs of circles 5 and 5 that
lie on the same side of the diagonal are equal. Now we can apply the Voronovich
Lemma to the quadrilateral that arises in the lower right corner of the figure and
conclude that this quadrilateral admits an inscribed circle. The problem is solved.


Homothety
This section lists problems that you should try to solve using homothety.
87 The circle ω2 is interior tangent to the circle ω1 at A. A line intersects ω1 at P
and Q and ω2 at M and N . Show that P AM = QAN .
88 Show that the projections of the vertex A of the triangle ABC onto the interior
and exterior bisectors of B and C are four collinear points.
89 Let I be the incenter of the triangle ABC, and let D be the point of tangency
of the incircle with the side AC. Show that if B1 is the midpoint of AC, then B1 I
cuts the segment BD in half.
90 Let AB be a segment with K a fixed point on it. Let M be a variable point on
the circle of diameter AB, and let N be the point diametrically opposite to M.
Find the locus of the intersection P of AN and KM when M varies on .
91 (Boutin’s Theorem) Let ABC be a triangle with circumcenter O. Let A , B , C
be the midpoints of BC, AC, and AB, respectively, and let A , B , C be points on
the rays |OA , |OB , and |OC such that
OA OB OC
= = .
OA OB OC
Show that AA , BB , and CC intersect at a point that lies on the Euler line of the
triangle ABC.
92 Prove that if in a trapezoid the point of intersection of the diagonals is at equal
distance from the two nonparallel sides, then the trapezoid is isosceles.
93 Let A1 A2 . . . An be a polygon, n ≥ 3, and let Gj be the centroid of the polygon
with n − 1 sides obtained by removing the vertex Aj . Show that the polygons
A1 A2 . . . An and G1 G2 . . . Gn are homothetic.
94 A polygon has the property that if its sides are pushed outward by one unit and
then extended until they meet, the resulting polygon is similar to the original. Show
that one can inscribe a circle in this polygon.
95 Let ABCD be a convex quadrilateral, and let M, N, P , Q be the midpoints
of the sides AB, BC, CD, DA, respectively. Construct the parallelogram XY ZW
by taking the parallel lines through the vertices of the quadrilateral ABCD to its
diagonals. Let I be the intersection of the diagonals of ABCD, J the intersection of
the diagonals of MNP Q, and K the intersection of the diagonals of XY ZW . Prove
that I, J, K are collinear.
96 Let ABC be a triangle, let A1 B1 C1 be its orthic triangle, let A B C be the
triangle determined by the midpoints of the sides, and let A2 B2 C2 be the orthic
triangle of A B C . Assume that the nine-point circles of A1 B1 C1 and A2 B2 C2 are
exterior to each other. Show that the interior tangents of these two nine-point circles
intersect on the Euler line of the triangle ABC.
97 Let ABCD be a quadrilateral. Consider the reflection of each of the lines
AB, BC, CD, DA over the respective midpoints of the opposite sides. Prove that
these four lines determine a quadrilateral that is homothetic to ABCD, and find the
ratio and the center of the homothety.
98 The incircle of the triangle ABC touches the sides BC, CA, AB at the points
A , B , C , respectively. Prove that the perpendiculars from the midpoints of A B ,
B C , and C A to AB, BC, and CA, respectively, are concurrent.
99 Inside a triangle ABC, there are four circles α, β, γ , δ, of the same radius ρ,
such that each of α, β, γ is tangent to two sides of the triangle and δ is tangent to α,
β, and γ .
(a) Show that the center of δ lies on the line passing through the incenter and the
circumcenter of ABC.
(b) Find ρ in terms of the circumradius and the inradius of ABC.
100 Prove that if the diagonals of a cyclic quadrilateral are perpendicular, then the
midpoints of its sides and the feet of the perpendiculars drawn from the intersection
point of the diagonals onto the sides lie on a circle.
101 Let ABC be a right triangle ( A = 90◦ ) such that the vertex A is fixed and
the vertices B and C vary on two fixed circles ωb and ωc that are exterior tangent at
A. Find the locus of the foot of the altitude AD.
102 For an angle measure α > π/3, consider all triangles ABC with BC = 1
and BAC = α. Among these triangles, find the one that minimizes the distance
between the centroid and the incenter, and compute this distance in terms of α.
103 In the acute triangle ABC, let A1 , B1 , and C1 be the feet of the altitudes from
A, B, and C, respectively, and let H be the orthocenter. The perpendiculars from
H onto A1 C1 and A1 B1 intersect the lines AB and AC at P and Q, respectively.
Prove that the line that joins A with the midpoint of the segment P Q is orthogonal
to B1 C1 .
104 Let ABC be a triangle, and let H , O, and R be its orthocenter, circumcenter,
and circumradius, respectively. Let D be the reflection of A across BC, E that of B
across CA, and F that of C across AB. Prove that D, E, and F are collinear if and
only if OH = 2R.
105 Let ABC be a triangle, and let A2 B2 C2 be the triangle formed by the tangents
to the circumcircle at A, B, and C (with A2 on the tangents at B and C, B2 on the
tangents at A and C, and C2 on the tangents at A and B). Let also AA1 , BB1 , and
CC1 be the altitudes of the triangle ABC. Show that the lines A1 A2 , B1 B2 , and
C1 C2 intersect on the Euler line of the triangle ABC.
106 Let ABC be a scalene, acute triangle, and let N be the center of its nine-point
circle. The lines r and s that are tangent to the circumcircle of the ABC at B and C,
respectively, meet at the point D. Prove that A, D, and N are collinear if and only
if BAC = 45◦ .
107 A circle passing through the vertices A and C of the triangle ABC intersects
the side AB at its midpoint D and the side BC at E. A circle that is tangent to AC at
C and passes through E intersects the line DE again at F . Let K be the intersection
of the lines AC and DE. Prove that AE, BK, and CF are concurrent.
108 Let ABC and DBC be two acute triangles, with A = D, such that the
vertices A and D are separated by the line BC. Consider E ∈ AC and F ∈ BD
such that BE⊥AC and CF ⊥BD, and let H1 and H2 be the orthocenters of the
triangles ABC and DBC. Prove that the lines AD, EF , and H1 H2 intersect at one
point, and moreover, this point is on the nine-point circles of both triangles.
109 The trapezoid ABCD has AB CD. A point P on the line BC, which does
not coincide with B or with C, is joined with D and with the midpoint M of the
segment AB. Let X be the intersection of P D and AB, Q the intersection of P M
and AC, and Y the intersection of DQ and AB. Show that M is the midpoint of the
segment XY .
110 Let ω be a circle, and let be a line tangent to it on which we fix a point M.
Find the locus of the points P in the plane with the property that there exist points
Q, R ∈ such that M is the midpoint of QR and ω is the incircle of P QR.
111 Let A, B, C, D be four collinear points (in this order). The circles of diameters
AC and BD intersect at X and Y , and the line XY intersects the segment BC at Z.
Let P be a point on the line XY different from Z. The line CP intersects the circle
of diameter AC for the second time at M, and the line BP intersects the circle
of diameter BD for the second time at N. Prove that the lines AM, DN , and XY
intersect at one point.
112 In the triangle ABC, let A , B , and C be the midpoints of the sides BC, CA,
and AB, respectively, let H be the orthocenter, and let O9 be the center of the nine-
point circle. Let also A1 , B1 , C1 be the intersections of the respective pairs of lines
AO9 and H A , BO9 and H B , and CO9 and H C . Show that the triangles ABC
and A1 B1 C1 are homothetic and have the same Euler line.
113 Let ω be a circle, and let ABCD be a square in its interior. Construct the circle
ωa that is tangent to the lines AB and AD, does not intersect the lines BC and CD,
and is interior tangent to ω at A . Construct analogously the points B , C , and D .
Prove that the lines AA , BB , CC and DD intersect at one point.
114 Let ABC be a triangle, and let AM and AN be the median and the angle
bisector of A (M, N ∈ BC). Let P and Q be the points where the line
perpendicular to AN at N intersects AB and AM, respectively, and let R be the
point where the line perpendicular to AB at P intersects the line AN. Prove that
RQ is perpendicular to BC.
115 (Desargues’ Theorem) Let A1 B1 C1 and A2 B2 C2 be two triangles. Show that
the lines A1 A2 , B1 B2 , and C1 C2 are concurrent or parallel if and only if the pairs of
lines (A1 B1 , A2 B2 ), (A1 C1 , A2 C2 ), and (B1 C1 , B2 C2 ) intersect at three collinear
points.
116 (Ceva’s Theorem) Let ABC be a triangle, and let A1 , B1 , C1 be points on the
lines BC, CA, and AB, respectively. Give a proof based on homothety of the fact
that AA1 , BB1 , and CC1 are concurrent if and only if their complex coordinates
satisfy
b − a1 c − b1 a − c1
· · = −1.
c − a1 a − b1 b − c1
117 The circle ω is interior tangent to the circle at K. The chords AB and BC
of are tangent to ω at M and N, respectively. Let P and Q be the midpoints of

the arcs AB and BC, respectively. The circumcircles of BP M and BQN intersect
the second time at L. Prove that the quadrilateral BP LQ is a parallelogram.
118 Let D be an arbitrary point on the side BC of the triangle ABC. Denote by
I and J the incenters of the triangles ABD and ACD and by E and F the centers
2.3 Homothety in Combinatorial Geometry; Scaling 97
of the exscribed circles of these triangles tangent to BC. Prove that I J and EF
intersect on the line BC.
119 In the plane are given three circles ω1 , ω2 , and ω3 that are exterior to each
other, of radii r1 , r2 , r3 , such that r1 is larger than both r2 and r3 . From the point of
intersection of the common exterior tangents of ω1 and ω2 , one constructs the two
tangents to ω3 , and from the point of intersection of the common exterior tangents
of ω1 and ω3 , one constructs the two tangents to ω2 . Prove that these two pairs
of tangents determine a quadrilateral in which one can inscribe a circle, and then
compute the radius of this circle in terms of r1 , r2 , and r3 .
120 (Lucas’ circles) Let ABC be an acute triangle. On the side BC, construct in
the exterior the square BCDE. Let M and N be the intersections of AE and AD
with the side BC, respectively. Let Q be the intersection of the perpendicular at M
to BC with AB, and let P be the intersection of the perpendicular at N to BC with
AC.
(a) Prove that MNP Q is a square.
(b) Let a be the circumcircle of AP Q; construct similarly the circles b and c .
Prove that the circles a , b , and c are pairwise tangent.
121 Let A1 A2 A3 A4 be a quadrilateral with no parallel sides. For each i =
1, 2, 3, 4, define ωi to be the circle tangent to the lines Ai−1 Ai and Ai+1 Ai+2 and
also tangent externally to the side Ai Ai+1 of the quadrilateral, and let Ti be this point
of tangency to the side (the indices are considered modulo 4, so A0 = A4 , A5 = A1 ,
and A6 = A2 ). Prove that the lines A1 A2 , A3 A4 , and T2 T4 are concurrent if and only
if the lines A2 A3 , A4 A1 , and T1 T3 are concurrent.
122 Let ABCD be a convex quadrilateral with AB = BC. Denote by ω1 and ω2
the incircles of ABC and ADC, respectively. Assume that there is a circle ω that is
tangent to the ray |BA at a point beyond A, to the ray |BC at a point beyond C, as
well as to the lines AD and CD. Prove that the common exterior tangents of ω1 and
ω2 intersect at a point on ω.
123 Given is a convex quadrilateral ABCD with an inscribed circle ω such that
the rays |AB and |DC meet at the point P and the rays |AD and |BC meet at the
point Q. The lines AC and P Q intersect at the point R. Let T be the point on the
circle ω that is the nearest to the line P Q. Prove that the line T R passes through the
incenter of the triangle P QC.
2.3 Homothety in Combinatorial Geometry; Scaling
In this section, we demonstrate the presence of homothety in combinatorial geom-

etry. The definition of homothety leads naturally to the idea of scaling, namely, of
reducing or enlarging shapes or of changing the scale at which they are viewed.
Scaling shows up in fractal geometry, where fractals, according to one of the
Fig. 2.19 The construction of the Sierpiński triangle
definitions given by Benoit Mandelbrot, are shapes made of parts similar to the
whole in some way. An example is the Sierpiński triangle, named after Wacław
Sierpiński. The Sierpiński triangle is some subset S of a closed equilateral triangular
surface T , having the property that if h1 , h2 , h3 are the homotheties of ratio 1/2
centered at the vertices of T , then
S = h1 (S) ∪ h2 (S) ∪ h3 (S).
To construct S, you divide T into four equilateral triangles, remove the interior of
the triangle in the middle, and call the resulting set S1 . Do the same with each of
the four triangles in S1 , to obtain S2 . Repeat for S2 to obtain S3 and so on. The first
three steps of the process are illustrated in Fig. 2.19. The Sierpiński triangle is the
intersection of all the sets Sn , n ≥ 1.
To illustrate how the idea of scaling can be used in solving mathematical
Olympiad problems, we apply it to a question that was published by S.V. Konyagin
in the journal Kvant (Quantum).
Problem 2.4 In a kingdom, whose territory has the shape of a square with a side
of 2 km, the king decides to invite all residents to his palace for a ball that starts at
7 pm. To this end he sends, at noon, a messenger who can give this information to
any resident, who, in his turn, can take this information to any other resident, and
so on. Every resident, before receiving the news, is located at home (in a known
place) and can travel at 3 km/h in any direction. Show that the king can organize the
transmission of messages so that all the residents can arrive at the court in time for
the opening of the ball.
Solution The solution reminds of the construction of the Sierpiński triangle, and
in general of the structure of fractals. To organize the notification, start by dividing
the kingdom into four squares of side 1 km (the squares of first rank). Each of these
squares is further divided into four squares of side 1/2 km (the squares of second
rank). The process is continued until reaching the largest rank with the property that
in each square there is no more than one residence of the kingdom (the ones on the
boundary can be distributed to one of the neighboring squares).
The notification is organized in stages. At the first stage, one residence of each
square of rank 1 is notified, after which all messengers return to the original position.
This process, illustrated in Fig. 2.20, is done as follows: say the castle, which is
2.3 Homothety in Combinatorial Geometry; Scaling 99
Fig. 2.20 The notification

process C
B
D
A
itself a residence, is at point A, and the other three residences are at B, C, D.

The messenger from the castle travels to B then to C and then back to A; in the
meanwhile, someone from B travels to D and back. In squares with no resident, no
one
√ is notified. The distances between any two of the points A, B, C, D are at most
2 2 < 3, so they can be traveled in less than 1 hour. Thus, the entire process took
at most 3 hours.
At the next step, we repeat the same strategy in each of the four squares, at scale
1/2. Then scale again, and repeat. After the nth stage, all residents will have been
notified. Because of the scaling, the time is halved at each stage. Therefore, in at
most

1 1
3 1 + + + · · · + 1/2n < 3 × 2 = 6
2 4
hours, everyone has been notified. In another hour every guest can reach the palace.
So by 6 + 1 = 7 o’clock, the ball can start.

The following problems are based either on this idea of scaling, or just on plain
homothety.
124 Let P1 = A1 A2 A3 A4 A5 be a convex pentagon, and let Pj be its translate
−−−→
by the vector A1 Aj , j = 2, 3, 4, 5. Show that the interiors of at least two of the
pentagons P1 , P2 , P3 , P4 , P5 overlap.
125 Let 1 , 2 , . . . , n be some lines in the plane, no two perpendicular and not
all parallel. Show that for every k there is a unique point xk on k such that the
perpendicular to k at xk passes through xk+1 for every k = 1, 2, . . . , n (where
xn+1 = x1 ).
126 It is given an equilateral triangle, with a flower and a grasshopper in its interior.
Every second, the grasshopper picks a vertex and jumps in the direction of that
vertex 1/10 of the distance to it. Show that regardless of the initial position of the
grasshopper, it can choose a sequence of jumps that will bring it arbitrarily close to
the flower.
127 Several turtles walk in the plane with constant velocities. Their velocities are
equal but have different directions. Show that after sometime, the turtles will be at
the vertices of a convex polygon.
128 In an infinite lattice, N squares have been colored black. Prove that from the
lattice one can cut a finite number of squares such that
(i) all the black squares lie in these squares
(ii) if K is one of the squares that was cut, the area of black squares inside K is
between 1/5 and 4/5 of the area of K.
129 (The Gohberg-Marcus Theorem) Show that every convex polygon M that is
not a parallelogram can be covered by three polygons homothetic to M and smaller
than M. Show that for the parallelogram, the minimal number is four.
2.4 A Theoretical Study of Spiral Similarities
2.4.1 The Definition and Properties of Spiral Similarities
We now proceed to the more general situation that has been foretold at the beginning
of the chapter.
Definition A spiral similarity about a point O by an angle α (counterclockwise)
and scaling factor (i.e., ratio) k > 0 is the composition of a rotation about O by α
and a homothety with ratio k and center O (Fig. 2.21).
We agree to include in the class of spiral similarities, as particular examples,
rotations, when the homothety has scaling factor k = 1, and homotheties, when the
rotation has angle α = 0.
In complex numbers, a spiral similarity maps the point z to the point f (z) defined
by the equation
f (z) − z0
= keiα .
z − z0
Fig. 2.21 A spiral similarity A

with center O, angle α =√45◦ ,
and scaling factor k = 2 2
O A
2.4 A Theoretical Study of Spiral Similarities 101
This gives the equation of the spiral similarity of center z0 , angle α, and ratio k > 0
as
f (z) = keiα z + (1 − keiα )z0 .
Conversely, if f (z) = rz + s with r = 1, then we can recover z0 = s/(1 − r),

α = arg(r), and k = |r|, so f is a spiral similarity.
Definition A transformation of the form f : C → C, f (z) = rz+s where r, s ∈ C,
r = 0 is called a complex affine transformation.
Adding the translations to the fold, we conclude that the affine transformations
of C represent all possible spiral similarities and translations. This also proves that
spiral similarities and translations together form a group. It is important to keep the
distinction between the affine transformations of C, the complex line, and the real
affine transformations of the real two-dimensional plane, R2 , which do not make
their presence in this book.
We can derive immediately a number of useful properties of spiral similarities. If
s is a spiral similarity of center O, (counterclockwise) angle α, and scaling factor k,
and if we denote, as usual, P = s(P ), then we have the following:
• (Automatic similarities) All triangles OP P are similar, with the same orienta-
tion. In fact, they are all similar to a triangle with sides of lengths 1 and k that
form an angle α; these similarities are “automatic” in the sense that you do not
need to visualize them; the similar triangles are always there!
• (Conserved properties) A spiral similarity maps lines to lines, circles to circles,
and segments to segments; it also preserves angles. If S is a polygon in the plane,
S and s(S) are similar, and corresponding sides in S and s(S) form angles equal
to α. These are consequences of the fact that a spiral similarity is the composition
of a rotation and a homothety.
Endowed with the understanding of the analytic form of spiral similarities, we
can state a principle that comes in handy when solving problems.
Theorem 2.18 (Averaging Principle) Let F1 and F2 be two geometric figures with
the property that there is a spiral similarity s such that s(F1 ) = F2 . Let t be a
fixed real number, and for each P ∈ F1 , consider a point Q in the plane such that
−→ −−−→
P Q = t P s(P ); let also F3 be the geometric figure consisting of all such points Q
when P varies in F1 . If F3 does not consist of a single point, then there is a spiral
similarity or translation that maps F1 to F3 .
Proof Let the spiral similarity be s(z) = rz + s. If p, q are the coordinates of P , Q,
respectively, then the condition from the statement reads
q − p = t (s(p) − p) = trp + ts − tp,
that is,
q = (tr − t + 1)p + ts.
This is not a constant function of p when tr − t + 1 = 1, and in that case it is either

a spiral similarity or a translation.

A simple observation, but a powerful problem-solving tool! The analytic proof
can be adapted to several figures and to linear combinations of points, even with
complex coefficients, giving rise to a more general statement.
Theorem 2.19 (Averaging Principle) Let F1 , F2 , . . . , Fn be several figures that
can be mapped into one another by spiral similarities, and let t1 , t2 , . . . , tn be any
complex numbers. Let also F be the figure consisting of the points of the form t1 z1 +
t2 z2 + · · · + tn zn , where zj ∈ Fj and zj is mapped to zk by the spiral similarity that
maps Fj to Fk , j, k = 1, 2, . . . , n, and assume that F is not a point. Then F can be
mapped by spiral similarities or translations to each of the n figures F1 , F2 , . . . , Fn .
Proof Let sk (z) = rk z + sk be the spiral similarity that maps F1 to Fk „ k =
2, 3, . . . , n. Then
t1 z1 + t2 z2 + · · · + tn zn = (t1 + t2 r2 + · · · + tk rk )z1 + t2 s2 + t3 s3 + · · · + tk sk ,
showing that there is a spiral similarity mapping F1 to F .

Remark The figure F defined by the numbers t1 , t2 , . . . , tn depends on the coordi-
nate system in use, except in the case where t1 + t2 + · · · + tn = 1.
The name of the principle comes from the case where t1 , t2 , . . . , tn are real and
add up to 1, for in that case F is the “average” (i.e., weighted mean) of the figures
F1 , F2 , . . . , Fn . This startling result is a straightforward consequence of the fact that
spiral similarities are complex affine transformations.
2.4.2 The Center of a Spiral Similarity: The Generic Case
Now, suppose we run into a spiral similarity, but do not have its parameters (center
O, angle α, and scale factor k) at hand. How do we recover them? The last two are
easy to find: if we know the images A and B of two distinct points A and B, then
α = (AB, A B ), the angle between the lines AB and A B , and k = A B /AB.
All that is left is to find the center O.
Theorem 2.20 Given four points A, B, A , and B in the plane such that A = B
and A = B , there is a unique spiral similarity s or translation that takes A to A
and B to B (i.e., A = s(A) and B = s(B)).
Proof Complex numbers yield a short proof, as it usually happens with theoretical
results. The center can be found using simple interpolation as follows. Let s be a
spiral similarity such that s(a) = a and s(b) = b . Then
b − a
s(z) − a = (z − a)
b−a
which gives
a − b ab − a b
s(z) = z+ .
a−b a−b
Writing s(z) = rz + (1 − r)z0 , we have
a − b a − b − a + b
1−r =1− = ,
a−b a−b
and hence
ab − a b
z0 = ,
a − b − a + b
For this to work, we should have a − a = b − b , but a − a = b − b means that

−−→ −−→
AA = BB , and then s is a translation.

There is an immediate corollary of this result. To formulate it, we need the
concept of orientation of a polygon introduced in the first chapter. Recalling what
was said in that chapter, a polygon is oriented counterclockwise if when traveling
from one vertex to the next the polygon is on the left and clockwise if the polygon
is on the right.
Definition Two polygons are called directly similar if they are similar and oriented
the same way.
Proposition 2.21 Given two directly similar polygons that are not translates of
each other, there is a unique spiral similarity that maps one into the other.
Proof Every polygon can be dissected into triangles whose vertices are also vertices
of the polygon. If we perform the same dissection in each polygon (namely, we
group corresponding vertices into triangles), then the polygons are similar if and
only if the triangles from the dissection are similar. So it suffices to prove the result
for triangles.
If ABC and A B C are two directly similar triangles, take the unique spiral
similarity that maps the segment AB to the segment A B . Then the points C is
mapped to some point C such that ABC and A B C are similar and oriented the
same way. But there is only one such point C in the plane, namely, C . Thus, the
spiral similarity maps C to C , and we are done.

This proposition helps extend the concept of direct similarity to geometric figures
other than polygons. Two geometric figures will be called directly similar if they are
mapped into each other by a transformation of the form f (z) = az + b, a, b ∈ C.
Simple as the complex number approach is, it gives no clue of how to locate
geometrically the center of the spiral similarity. And we do want to know the answer
to this geometric question. Two particular cases should be taken out of our way. The
case where the line AB is parallel or coincides with A B corresponds to AB being
mapped to A B by either a translation or a homothety (the latter is a spiral similarity
with angle 0 or 180◦ ). The translation has no center, and the center of the homothety
is the intersection of AA and BB . The case of a true spiral similarity is clarified by
the following result.
Theorem 2.22 (Center of Spiral Similarity) Given four points A, B, A , and B in
the plane such that A = B and A = B and AB is not parallel to A B , the center of
the unique spiral similarity that takes A to A and B to B is the second intersection
point of the circumcircles of P AA and P BB where P is the intersection point of
the lines AB and A B .
Proof Let O be the center of the spiral similarity that maps AB to A B , and let us
assume that O is different from P . Because the spiral similarity maps the line AB
to the line A B and the line OA to the line OA , and also because spiral similarities
preserve angles
(AB, OA) = (A B , OA ).
It follows that P , A, A , O lie on a circle. For a similar reason, P , B, B , O lie on a

circle, and so O is the second intersection point of the circumcircles of P AA and
P BB (Fig. 2.22).
If O coincides with P , let us show that the two circles are tangent at P (in which
case O is the “other” point of intersection because the tangency point is a double
point). The segment AB is mapped to A B by a rotation about P followed by a
homothety about P . Consequently
PA P A
= .
PB P B
So the triangle P AA is mapped to the triangle P BB by a homothety of center P ,

and hence the circumcircle of P AA is mapped to the circumcircle of P BB by a
Fig. 2.22 Finding the center

A B
of a spiral similarity. Try to P
draw the points A, B, A , B
in other positions to see what
happens! A B
O
Fig. 2.23 Intuitive aid for

understanding how to find the
center of a spiral similarity
Fig. 2.24 Spiral similarity: A

what if B lies in AB?
B =P
B
A
homothety of center P . Therefore, the two circles are tangent at P . The theorem is
proved.

This result can be easily remembered and understood by looking at Fig. 2.23.
There are some degenerate cases to consider. One such case is when three of the
four points are collinear. Suppose, without loss of generality, that A, B, and B are
these points. Since AB and A B are not parallel, A does not lie on this line. In this
case, the intersection P of AB and A B coincides with B , so the circumcircle of
P BB degenerates to a circle that passes through B and B and is tangent to the line
A B . You can interpret the equality B = P as a double root, so taking a tangent
circle makes sense. This situation is illustrated in Fig. 2.24
In fact, the angle equality (AB, OB) = (A B , OB ) from Theorem 2.22 can
be interpreted exactly as saying that A B is tangent to the circumcircle of OBB .
2.4.3 The Center of a Spiral Similarity: The Case A = B,

Symmedians Revisited
Looking at Theorem 2.22, we realize that there is an ambiguity when A = B, in

which case P also coincides with B and the construction degenerates. We could
address this case similarly with tangent lines, but instead, we study it separately, as
it unveils a surprising relationship between spiral similarities and symmedians. Let
therefore A = B and B = C, so that the spiral similarity s takes A to B and B to
C and the triangle ABC is nondegenerate.
Theorem 2.23 (Spiral Similarity and Symmedians) Let ABC be a triangle. The
center of the spiral similarity s that takes A to B and B to C lies on the B-
symmedian of the triangle ABC, namely, on the symmetric of the median from B
with respect to the bisector of ABC.
Proof Let K be the center of s. Then
KA KB AB
= = .
KB KC BC
Let BK meet the circumcircle of ABC again at L = B (Fig. 2.25). Notice that
ALB = ACB, so the spiral similarity s with center B that takes L to C takes
the line AL to the line AC. Let M = s (A). By the automatic similarity, the triangles
BAM and BLC are similar, so
AB AM BM
= = ,
BL LC BC
and hence
LC
AM = · BM.
BC
By automatic similarity, the triangles ABL and MBC are also similar, so
AB BL AL
= = ,
BM BC MC
and hence
AL
MC = · BM.
AB
Finally, using inscribed angles and s, LAC = LBC = KBC = KAB,

and LCA = LBA = KBA, so the triangles ALC and AKB are also similar,
and
AL AK AB
= = ,
LC KB BC
which implies
AL LC
= ,
AB BC
Fig. 2.25 Finding the center B

of a spiral similarity, A = B
A M C
and using the relations derived above, we find that AM = MC. We have proved that
M is the midpoint of AC and MBC = ABK. So the line BK is the symmetric
of the median BM with respect to the internal bisector of B, showing that it is the
B-symmedian.

Let us put symmedians more solidly in the context of spiral similarity. Because
the spiral similarity s has ratio AB/BC, the ratio of the distances from K to AB
and AC is dA /dC = AB/BC. This, of course, is true for every point lying on the
symmedian, in particular for the point T where the symmedian BK meets the side
AC. Then
AT area AT B AB · dA AB 2
= = = .
TC area AT C AC · dC BC 2
2.4.4 Spiral Similarities and Miquel’s Theorem
We now interpret a classical result about complete quadrilaterals from the point of
view of spiral similarities. A complete quadrilateral is the configuration formed by
four lines, no two parallel. The name comes from the fact that it is the configuration
obtained when extending the lines of support of the sides of a quadrilateral that is
not a trapezoid.
Theorem 2.24 (Miquel’s Theorem) Consider four lines 1 , 2 , 3 , and 4 , no two
of them parallel. For i = 1, 2, 3, 4, define Ci as the circumcircle of the triangle
determined by the three lines different from i . Then the four circles C1 , C2 , C3 , and
C4 have a common point, known as the Miquel point of 1 , 2 , 3 , and 4 .
Proof There are several proofs using simple angle chasing (the reader is encouraged
to try), but we present a very short approach based on spiral similarities, which has
the advantage of avoiding configuration issues. One possible construction is shown
in Fig. 2.26.
Fig. 2.26 The Miquel point

1
3
A13
M
A14
A34
2
A24
A12 A23 4
Let Aij be the intersection of the lines i and j , for 1 ≤ i < j ≤ 4. Consider the
spiral similarity that takes A12 to A14 and A23 to A34 . By Theorem 2.22, its center
is the second intersection point M of the circle that passes through A12 , A14 , and
the intersection of A12 A23 = 2 and A14 A34 = 4 , that is, the circumcircle C3 of
A12 A14 A24 , and the circle through A23 , A34 , and A24 , which is C1 .
Now the automatic similarity between the triangles MA12 A14 and MA23 A34
induces another spiral similarity with center M that takes A12 to A23 and A14 to
A34 . Repeating the argument for these new pairs of points, we deduce that M is also
the second intersection point of C2 and C4 , and so M lies on all circles, as claimed.

The trick of the proof can be phrased as follows.
Theorem 2.25 (Spiral Similarities Come in Pairs) Let A, B, C, and D be points
in the plane. Then the spiral similarity that takes A to C and B to D induces another
spiral similarity with the same center that takes A to B and C to D.
Proof The angle of the spiral similarity that maps A to B and C to D is the sum of
COB and the angle of the initial spiral similarity. Its ratio is AO/BO, which is
equal to CO/DO because AO/CO = BO/DO.

We can rethink this result more intuitively and visualize it in Fig. 2.27: if the
triangle OAB is mapped to the triangle OCD by a spiral similarity, then the triangle
OAC is mapped to the triangle OBD by a spiral similarity.
We will revisit Miquel’s Theorem at the end of Chap. 4, where we will employ a
variety of geometric transformations to obtain detailed knowledge about complete
quadrilaterals.
Fig. 2.27 Spiral similarities A A

come in pairs C C
O O
B B
D D
2.4.5 Compositions of Spiral Similarities
We want to understand compositions of spiral similarities. To this end, let us first

look at the simpler case of the composition of a rotation and a homothety (both
particular cases of spiral similarities), when the centers of rotation and homothety
do not coincide.
Theorem 2.26 The composition of a rotation and a homothety with different
centers is a spiral similarity.
Proof Let s be the resulting composition. It preserves the shape and orientation of
geometric figures. We have learned that, given A and B, A = s(A), and B = s(B),
there exists a unique spiral similarity or translation τ that takes A to A and B to
B . Let P be any point in the plane, and let P = s(P ). Then ABP and A B P are
similar with the same orientation and thus s(P ) = τ (P ). We deduce that s = τ .
We just have to rule out the possibility that s is a translation. If it were a
−→ −−→
translation, then A B AB, and AB = A B , so the homothety ratio is k = 1,
and the rotation angle is 0; this means that s is the identity, which can be considered
a spiral similarity with any center, angle 0, and ratio 1.

Theorem 2.27 (Composition of Spiral Similarities) The composition of two spi-
ral similarities is a spiral similarity or a translation.
Proof 1 Surprisingly, the above particular case essentially solves the general case.
The reason is that we now can break each spiral similarity into two transformations
and handle them separately.
Specifically, let s1 = ρ1 ◦ h1 and s2 = h2 ◦ ρ2 be two spiral similarities, ρi
being rotations and hi being homotheties, i = 1, 2, whose centers coincide for each
i (notice that we can commute ρi with hi for each i = 1, 2 because of that). Then
s1 ◦ s2 = ρ1 ◦ h1 ◦ h2 ◦ ρ2 . The composition of two homotheties h1 and h2 is another
homothety or a translation.
If the composition h1 ◦ h2 were a translation, then ρ1 ◦ h1 ◦ h2 ◦ ρ2 is an
orientation preserving isometry, which is either a translation or a rotation (hence
a spiral similarity).
If the composition of h1 and h2 is a homothety h, then s1 ◦ s2 = ρ1 ◦ h ◦ ρ2 .

The composition of h and ρ2 is, by Theorem 2.26, a spiral similarity s = ρ ◦ h , so
s1 ◦ s2 = ρ1 ◦ ρ ◦ h . Finally, ρ1 ◦ ρ is a composition of rotations, which is another
rotation ρ, and we find s1 ◦ s2 = ρ ◦ h , which, by the previous theorem again, is a
spiral similarity, finishing the proof.

Proof 2 In complex numbers, this is equivalent to stating that the composition of
two affine functions on C is another affine function, which is immediate. Explicitly,
if f1 (z) = r1 z + s1 and f2 (z) = r2 z + s2 , then
(f2 ◦ f1 )(z) = f2 (r1 z + s1 ) = r2 (r1 z + s1 ) + s2 = (r1 r2 )z + (r2 s1 + s2 ).
If r1 r2 = 1, then the composition is a translation; otherwise, it is a spiral similarity.

2.4.6 Groups Generated by Spiral Similarities
With the tools for studying spiral similarities that we have developed so far at hand,
we are ready to prove a general result about geometric transformations of the plane.
Theorem 2.28 Every transformation of the plane that maps lines to lines and
circles to circles is either an isometry, a spiral similarity, or the composition of
a spiral similarity and a reflection.
Proof Let f be the transformation in question, and let A and B be two distinct
points in the plane. Consider a translation τ that maps f (A) to A and a spiral
similarity s of center A that maps τ (f (B)) to B. The transformation g = s ◦ τ ◦ f
still maps lines to lines and circles to circles, and g(A) = A, g(B) = B, so g has
two fixed points. We will show that g is either the identity map or the reflection over
AB.
Note that because g is bijective, g maps the intersection of two lines to the
intersection of their images. If the lines do not intersect, their images do not intersect
either. So g maps parallel lines to parallel lines. For the same reason, g maps the
intersection of two circles, to the intersection of their images. In particular, if the
circles are tangent, their images must be tangent, too, and the tangency point of the
original two circles is mapped to the tangency point of their images. The same is
true if one of the circles is replaced by a line.
Take the lines 1 and 2 that are perpendicular to AB at A and B, respectively,
and take also the circle ω of diameter AB (see Fig. 2.28). The images g(1 ) and
g(2 ) are parallel lines through A and B as well, but they might be tilted, thus
not forming an angle of 90◦ with AB. But g(ω) is a circle through A and B that
is tangent to both g(1 ) and g(2 ), and this can only happen if these two lines are
orthogonal to AB (as you can notice on Fig. 2.28). In particular g(1 ) = 1 , g(2 ) =
2 , and g(ω) = ω.
1 2
g( 1)
g( 2)
A B A B
ω
g(ω)
Fig. 2.28 The lines orthogonal to AB at A and B are mapped into themselves
Fig. 2.29 Construction of a

lattice whose nodes are
invariant under g
D C E
A B
F
The line tangent to ω and parallel to AB is mapped to a line tangent to ω and

parallel to AB, as well. There are two such lines: itself and its reflection over AB.
By eventually replacing g by σ ◦ g, where σ is the reflection over AB, we may
assume that g() = . Let C be the tangency point of ω and , D the intersection
of and 1 , and E the intersection of and 2 . Let also F be the midpoint of AB
(Fig. 2.29). Then g(C) = C, g(D) = D, and g(E) = E. But by replacing A and
B by C and D in the above argument, and arguing that the circle of diameter CD is
invariant under g, we conclude that g(F ) = F (as F plays the role of C in this new
configuration).
A similar argument shows that the reflections of C and D over AB are invariant
under g. Repeating the argument, we find that there exists a square lattice of size
AB/2 whose nodes are invariant under g. Next, replace A and B by A and F , and
repeat the argument. This proves that there is a square lattice of size AB/4 whose
nodes are invariant under g. Inductively, we obtain that for every n > 0, there is a
square lattice of size AB/2n whose sides are parallel to AB and whose nodes are
invariant under g. The lines through these nodes that are parallel and perpendicular
to AB are mapped into themselves by g.
Next, let be an arbitrary circle. Let u and d be parallel to AB, l and r be
perpendicular to AB, and all four tangent to (Fig. 2.30). Then g() must intersect
all lines that belong to each of the lattices and lie between u and d , respectively,
Fig. 2.30 g maps any circle

into itself l r
u
between l and r , and not intersect any other line. Since we can find lines from the
union of lattices as close as we wish to any of u , d , l , and r , this is only possible
if g() = . So g maps any circle into itself.
Finally, for an arbitrary point P , let 1 and 2 be two circles that are tangent at
P . Since g(1 ) = 1 and g(2 ) = 2 , g(P ) = P . Hence, g is the identity map.
Returning to f , it is either a translation, a spiral similarity, the composition of
one of these and a reflection, or the composition of all three. But the composition
of a translation and a spiral similarity is itself a spiral similarity. The theorem is
proved.

In conclusion, the group generated by spiral similarities consists of spiral simi-
larities (including rotations and homotheties as particular cases) and translations. In
complex coordinates, it is
{f : C → C | f (z) = az + b, a, b ∈ C, a = 0}.
The group of transformations that map lines to lines and circles to circles is
generated by spiral similarities and reflections, and it is
{f : C → C | f (z) = az + b or f (z) = a z̄ + b, a, b ∈ C, a = 0}.
Remark It is important to compare Theorem 2.28 to Theorem 2.8. It is not true that
every transformation that maps lines to lines is an isometry, a spiral similarity, or a
spiral similarity composed with a reflection. In real coordinates, the map (x, y) →
(x + y, y) maps lines to lines, but it is unlike any transformation discussed in this
book. It is what we call a real affine transformation; while such transformations are
important in mathematics, we do not include them in our discussion.
2.5 Spiral Similarity in Euclidean Geometry Problems 113
2.4.7 Theoretical Questions About Spiral Similarities
Here are a few problems that will improve your understanding of these conceptual
facts.
130 Let AB1 C1 , AB2 C2 , and AB3 C3 be three directly similar triangles. Show that
if B1 , B2 , B3 are collinear, then C1 , C2 , C3 are collinear.
131 Let ω be a circle, and let s be a spiral similarity whose center M is on ω.
Show that all the lines XX with X ∈ ω and X = s(X) pass through the second
intersection point of the circles ω and s(ω).
132 Let ABC be a triangle, and let X be the intersection of the circle through B
tangent to AC at A with the circle through C tangent to AB at A. Prove that AX is
symmedian in the triangle ABC.
133 Let 1 and 2 be two lines that intersect at P , let A1 be on 1 , A2 on 2 , and
let k be a positive number. Show that there are exactly two spiral similarities of ratio
k that map 1 to 2 so that A1 is mapped to A2 . Moreover, the centers of these spiral
similarities are at the intersection of the circumcircle of the triangle P A1 A2 and
the Apollonian circle determined by the points A1 and A2 and the ratio k. (For the
definition of the Apollonian circle, see Problem 165 in Sect. 3.1.10.)
134 Let ω1 and ω2 be two nonconcentric circles. How many spiral similarities that
map one circle to the other and have the angle equal to 90◦ exist? (We allow both
clockwise and counterclockwise 90◦ rotations.)
2.5 Spiral Similarity in Euclidean Geometry Problems
2.5.1 Similar Figures and the Circle of Similitude
We will demonstrate the power of spiral similarity by revisiting an exposition

of the second author from Transformari Geometrice. Omotetia si Inversiunea
(Matrixrom). While the theorems presented in this section are not necessarily useful
in solving problems, the techniques employed in proving them are.
Proposition 2.21 has shown that two polygons that are directly similar and are
not translates of each other can be mapped one into the other by a unique spiral
similarity. And we have generalized the notion of direct similarity of figures by
agreeing that it is decided by the existence of a spiral similarity or translation
that maps one figure to the other. In this section we will assume that figures that
are directly similar are not mapped into each other by translations, only by spiral
similarities.
Definition Let F1 , F2 , F3 be three directly similar figures; let O1 be the center of
the spiral similarity that maps F2 to F3 , O2 the center of the spiral similarity that
maps F3 to F1 , and O3 the center of the spiral similarity that maps F1 to F2 . If the
points O1 , O2 , and O3 are not collinear, the triangle they form is called the triangle
of similitude of F1 , F2 , and F3 , and the circumcircle of O1 O2 O3 is called the circle
of similitude of these three figures.
If O1 , O2 , and O3 coincide, we say that we have a center of similitude, and if
they are collinear, then we have an axis of similitude. In what follows, we assume
that neither of these two degenerate cases happen, so the three figures have a triangle
of similitude.
Theorem 2.29 Let A1 B1 , A2 B2 , and A3 B3 be three segments in the plane, and let
P1 , P2 , P3 be the intersection points of the pairs of lines A2 B2 and A3 B3 ; A1 B1
and A3 B3 ; and A1 B1 and A2 B2 , respectively.
(a) The circumcircles of A1 A2 P3 , A2 A3 P1 , and A3 A1 P2 intersect at a point that
lies on the circle of similitude of A1 B1 , A2 B2 , and A3 B3 , and the same is true
for the circumcircles of B1 B2 P3 , B2 B3 P1 , and B3 B1 P2 .
(b) Let Oi be the center of the spiral similarity that maps Aj Bj to Ak Bk , {i, j, k} =
{1, 2, 3}. Then P1 O1 , P2 O2 , P3 O3 intersect at a point that is on the circle of
similitude of A1 B1 , A2 B2 , and A3 B3 .
Part (b) is usually phrased as: In a system of three directly similar figures, the
triangle formed by three homologous lines is in perspective with the triangle of
similitude, with the center of perspective lying on the circle of similitude.
Proof We first prove an auxiliary result.
Lemma 2.30 On the sides of the triangle ABC, we construct the triangles A BC,
AB C, and ABC such that the sum of the angles (or their supplements) at the
vertices A , B , C is a multiple of 180◦ . Then the circumcircles of these triangles
intersect at one point.
Proof For simplicity, let us assume that the circumcircles of A BC and AB C are
not tangent so that they intersect at a second point, P (Fig. 2.31).
Using directed angles modulo 180◦ , we can write
(P A, P B) = (P A, P C) + (P C, P B) = (B A, B C) + (A C, A B)
= (C A, C B),
showing that P is also on the circumcircle of ABC . The case where the circles are
tangent at C is similar, with the line P C replaced by the tangent at C. The lemma is
proved.

A particular case of the lemma is the configuration where A ∈ B C , B ∈ A C ,
and C ∈ A B , since in that case the three angles in question are BA C =
B A C , AB C = A B C , and AC B = A C B , and, as angles of a triangle,
they add up to 180◦ . This particular case is used for proving the theorem. Such a
A
C
C
B
Fig. 2.31 The circumcircles of A BC, AB C, ABC intersect
Fig. 2.32 Configuration with P1

the lines A1 B1 , A2 B2 , A3 B3
intersecting at P1 , P2 , P3
B3 B2
A2
A3
V
A1
P2 B1 P3
configuration can be discovered in Fig. 2.32, where the roles of A, B, C are played
by A1 , A2 , A3 and the roles of A , B , C are played by P1 , P2 , P3 .
Returning to the proof of the theorem, note first that by Theorem 2.22, the
points O3 , O2 , O1 are on the circumcircles of the triangles A1 A2 P3 , A1 A3 P2 , and
A2 A3 P1 , respectively. Then, by applying Lemma 2.30 to the configuration where
A1 ∈ P2 P3 , A2 ∈ P1 P3 , and A3 ∈ P1 P2 , we deduce that the circumcircles of
A1 A2 P3 , A1 A3 P2 , and A2 A3 P1 intersect at some point V . For the same reason, the
circumcircles of B1 B2 P3 , B1 B3 P2 , and B2 B3 P1 intersect at some point V .
Let U be the intersection point of P2 O2 and P3 O3 . We claim that V is on the

circumcircle of U O2 O3 . Indeed, using inscribed angles, we can write
(O2 V , V O3 ) = (V O2 , O2 P2 ) + (O2 P2 , P3 O3 ) + (P3 O3 , O3 V )

= (V A1 , A1 P2 ) + (O2 U, U O3 ) + (P3 A1 , A1 V )
= (O2 U, U O3 ),
so O2 , O3 , U, V are concyclic. Similarly V is on the circumcircle of U O2 O3 .

As a consequence, we obtain that O2 , O3 , V , V are concyclic, and similarly
O1 , O2 , V , V are concyclic, proving that V , V are on the circle of similarity. This
proves (a).
For (b), notice that the lines P2 O2 and P3 O3 intersect at U , which is also on the
circle of similarity. But P1 O1 and P2 O2 also intersect at a point on the circle of
similarity; this point is the second intersection point of the circle of similarity with
P2 O2 , so it must be U . Hence, the three lines intersect at U , and the theorem is
proved.

Let F1 and F2 be two directly similar figures, so that there exists a spiral
similarity s that maps F1 to F2 . If S1 and S2 are subsets of F1 and F2 , respectively,
we say that S1 and S2 are corresponding subsets of F1 and F2 if s(S1 ) = s(S2 ).
Theorem 2.31 Let F1 , F2 , F3 be three directly similar figures, and let A1 B1 , A2 B2 ,
A3 B3 , and A1 C1 , A2 C2 , and A3 C3 be corresponding segments of F1 , F2 , F3 . Then
the triangle formed by the lines A1 B1 , A2 B2 , and A3 B3 and the triangle formed by
the lines A1 C1 , A2 C2 , and A3 C3 are directly similar, and the center of the spiral
similarity that maps one into the other is on the circle of similitude of F1 , F2 , F3 .
Proof Let Pi be the intersection of Aj Bj and Ak Bk , and let Pj be the intersection
of Aj Cj and Ak Ck , {i, j, k} = {1, 2, 3} (Fig. 2.33). The spiral similarity that takes
F1 into F2 takes the lines A1 B1 and A1 C1 into A2 B2 and A2 C2 , respectively. Hence,
(A1 B1 , A2 B2 ) = (A1 C1 , A2 C2 ), as directed angles modulo π . This is the same
as P1 P3 P2 = P1 P3 P2 , and we have similar equalities for the other two angles.
We deduce that the triangles P1 P3 P2 and P1 P3 P2 are similar and also (because we
are working with directed angles) that they have the same orientation. Hence, by
Proposition 2.21, they are mapped into each other by a spiral similarity s.
Applying Theorem 2.22 to P2 P3 and P2 P3 , we deduce that the center of s is the
second point of intersection of the circumcircles of A1 P3 P3 and A1 P2 P2 . By the
same argument applied to P1 P3 and P2 P3 , we deduce that the center of s is the
second intersection point of the circumcircles of A2 P3 P3 and A2 P1 P1 . Thus, the
circumcircles of A1 P3 P3 and A2 P3 P3 have three points in common: P3 , P3 , and the
center of s. But this can only happen if the circles coincide, and thus A1 , A2 , P3 , P3
are concyclic. It follows that the center of s is on the circumcircle of A1 A2 P3 .
Similarly, the center of s is on the circumcircles of A2 A3 P1 and A1 A3 P2 . But then
by Theorem 2.29 (a) applied to the configuration consisting of A1 B1 , A2 B2 , A3 B3 ,
we deduce that the intersection point of these three circles, which is the center of s,
is also on the circle of similitude of F1 , F2 , F3 .

Fig. 2.33 Proof that P1 P2 P3 P1

and P1 P2 P3 are directly
similar C2
A2
P3
P1 B2
C3 P3
B3
A1
C1
A3
P2 B1
P2
Fig. 2.34 Construction of the

3
invariable points J1 , J2 , J3 J3
O1
O3
P1
O2 1
P3
2
P2
J2 1
2
W
J1
3
Theorem 2.32 Let 1 , 2 , and 3 be three lines that intersect at a point W and
are corresponding lines of the directly similar figures F1 , F2 , F3 . Then W is on the
circle of similitude of F1 , F2 , F3 . If J1 , J2 , J3 are the other points where 1 , 2 ,
and 3 intersect the circle of similitude, respectively, then J1 , J2 , J3 depend only on
F1 , F2 , F3 and not on the choice of 1 , 2 , and 3 .
Before starting the proof, let us point out that the figures F1 , F2 , F3 might not
contain entire lines, such as in the case where they consist of polygons and circles.
But we can choose any two points of F1 and then add the line through them to F1
and the corresponding lines to F2 and F3 to obtain larger figures that are still directly
similar.
Proof The lines 1 , 2 , 3 are not very helpful, because they intersect at one point,
so they do not allow us to use Theorem 2.29 or Theorem 2.31. To bring these
theorems into the story, we add the lines 1 , 2 , 3 to F1 , F2 , F3 , respectively, such
that i i , i = 1, 2, 3, and 1 , 2 , 3 also correspond by the spiral similarities. We
denote by Pi the intersection of j and k , {i, j, k} = {1, 2, 3}. The situation is
shown in Fig. 2.34.
Denote by si the spiral similarity that takes Fj to Fk , and let Oi be its center,
where (i, j, k) is a cyclic permutation of (1, 2, 3). Let h3 be the homothety of center
O3 that maps 1 to 1 . Then h3 and s3 commute (because they are spiral similarities
of the same center), so
h3 (2 ) = (h3 ◦ s3 )(1 ) = (s3 ◦ h3 )(1 ) = s3 (1 ) = 2 .
It follows that h3 (W ), which is the image of the intersection of 1 and 2 , is the

intersection of h3 (1 ) = 1 and h3 (2 ) = 2 , and this intersection point is P3 .
But O3 , W, h(W ) = P3 are then collinear, showing that O3 P3 passes through W .
Similarly O2 P2 and O1 P1 pass through W , so the three lines intersect at W . But
Theorem 2.31 shows that O1 P1 , O2 P2 , O3 P3 intersect on the circle of similitude;
hence, W is on the circle of similitude of F1 , F2 , F3 .
The second claim is a consequence of the first. Note that the angles formed by
1 , 2 , and 3 are the angles of the spiral similarities. Because the lines intersect
at W on the circle of similarity, the arcs determined by J1 , J2 , J3 have the fixed
measures, namely, twice these rotation angles. And the ratio of the distances from
Oi to j and k , {i, j, k} = {1, 2, 3} is the ratio of the spiral similarity with center
Oi . Since the angle between j and k is fixed, this means that the angles Oj W Jk ,

k = j are fixed. So then the measures of the arcs Oj Jk are also fixed for all j = k,
and since the points Oj are fixed, the points Jk are fixed as well, for j, k = 1, 2, 3.
This completes the proof of the theorem.

The points J1 , J2 , J3 are called the invariable points of the directly similar figures
F1 , F2 , F3 , and the triangle J1 J2 J3 is called the invariable triangle.
Proposition 2.33 Given three directly similar figures:
(a) the triangle formed by three corresponding lines is similar to the invariable
triangle but has opposite orientation;
(b) the three invariable points are corresponding points of the three figures.
Proof We reason once more on Fig. 2.34. We have
(J1 J2 , J2 J3 ) = (J1 W, W J3 ) = (P3 P2 , P2 P1 ),
where for the first equality we used inscribed angles and for the second we used the
fact that (J1 W, W J3 ) is the angle of the spiral similarity that maps F1 to F3 . This
proves (a). For (b), note that
(J2 O1 , O1 J3 ) = (J2 W, W J3 ),
which is the angle of the spiral similarity of center O1 that maps F2 to F3 . By

inscribed angles
(O1 J2 , J2 W ) = (O1 J3 , J3 W ),
which shows that O1 J2 /O1 J3 equals the ratio k of the distances from O1 to 2 and
3 . But k is the scaling factor of the spiral similarity of center O1 that maps 2 to 3 .
Hence J2 is mapped to J3 by the spiral similarity that maps F2 to F3 . The situation
is the same for J1 , J3 and J1 , J2 . The theorem is proved.

For a triangle ABC, the circle of similitude of the segments AB, BC, CA is
called the circle of similitude of the triangle ABC.
Theorem 2.34 The circle of similitude of a triangle is the circle of diameter KO
where K is the Lemoine point (i.e., the intersection of the symmedians), and O is
the circumcenter.
Proof Let Oa , Ob , Oc be the centers of the spiral similarities that map AB to CA,
BC to AB, and CA to BC, respectively. Then by Theorem 2.29, the lines AOa ,
BOb , and COc intersect at a point on the circle of similitude .
Now we are in the situation described in Sect. 2.4.3, where the construction of the
center of the spiral similarity degenerates, and Oa is the point where the circle that
is tangent to AC at A and passes through B intersects for the second time the circle
that is tangent to AB at A and passes through C. And we know that this is related
to symmedians, more precisely; Theorem 2.23 shows that AOa is a symmedian,
and the same is true for BOb and COc , so they intersect at the Lemoine point K
(Fig. 2.35). Thus, we know that K ∈ .
Ob
K
Oa Oc
Σ
B C
Fig. 2.35 Construction of the Lemoine point

The perpendicular bisectors of the sides are corresponding lines, and they
intersect at one point, namely, O, and Theorem 2.32 implies that O is on the circle
of similitude , as well. The points A1 , B1 , C1 where the perpendicular bisectors
of BC, CA, AB intersect are the invariable points of the three sides.
If we take the three lines a , b , c through K parallel to AB, BC, CA, respec-
tively, then we are in a converse situation of Theorem 2.32. We claim that these
parallels correspond through the spiral similarities that map the sides of the
triangle into each other. Indeed, consider the spiral similarities sa , sb , sc of centers
Oa , Ob , Oc that map b to c , c to a , and a to b , respectively, and let BC,
sc (BC), and sa sc (BC) be the corresponding lines that form the triangle A , B , C .
Then B , C ∈ BC, and Ob , K, B are collinear, and also Oc , K, C are collinear.
It follows that B = B, C = C. But because the triangles A B C and ABC
have parallel sides and are oriented the same way, A = A. Therefore, sa , sb , sc
are the spiral similarities that map the sides into each other, and so a , b , c are
corresponding lines. These lines must therefore pass through the invariable points
A1 , B1 , C1 . The line OA1 , being the perpendicular bisector of BC, is perpendicular
to a . Thus, KA1 O = 90◦ , showing that KO is a diameter in the circle of
similitude, and we are done.
2.5.2 Examples of Problems Solved Using Spiral Similarities
Let us present some problems that use spiral similarities and have shown up in
mathematics competitions. Our first example is a problem that was given at the
Asian Pacific Mathematical Olympiad in 1998, being proposed by the first author
of the book.
Problem 2.5 Let ABC be a triangle and let D be the foot of the altitude from A. Let
E and F be two points on a line passing through D such that AE is perpendicular
to BE, AF is perpendicular to CF , and E and F are different from D. Let M and
N be the midpoints of the segments BC and EF , respectively. Prove that AN is
perpendicular to NM.
Solution 1 The right angles imply that A, B, D, E lie on the circle of diameter AB
and A, C, D, F lie on the circle of diameter AC (see Fig. 2.36). From here, using
inscribed angles and working with directed angles modulo 180◦ , we deduce that
(AB, BC) = (AB, BD) = (AE, ED),
and
(AC, CB) = (AC, CD) = (AF, F D).
It follows that the triangles ABC and AEF are directly similar, so there is a spiral
similarity of center A taking B to E and C to F . This spiral similarity takes M,
B D M C
Fig. 2.36 Proof that AN M = 90◦
the midpoint of BC, to N , the midpoint of EF . So B → E, and M → N, but as

spiral similarities come in pairs (Theorem 2.25), there is a spiral similarity of center
A that maps the triangle ABE to the triangle AMN . We conclude that ANM =
AEB = 90◦ , as required.

Solution 2 We have seen in the first solution that ABC is mapped to AEF by a
spiral similarity, and since spiral similarities come in pairs (Theorem 2.25), there is
a spiral similarity of center A that takes ABE to ACF . Since M is the midpoint
of BC and N the midpoint of EF , the Averaging Principle (Theorem 2.18) implies
that the triangles ABE and ACF are also similar to AMN . Therefore, ANM =
AEB = 90◦ .

By comparing these two solutions, you should be able to derive a synthetic proof
of the Averaging Principle (Theorem 2.18).
Our second example is a problem of Cosmin Pohoaţă that was given at the United
States of America Mathematical Olympiad in 2014.
Problem 2.6 Let ABC be a triangle with orthocenter H , and let P be the second
intersection point of the circumcircle of the triangle AH C with the internal bisector
of the angle BAC. Let X be the circumcenter of the triangle AP B, and let Y be
the orthocenter of the triangle AP C. Prove that the length of the segment XY is
equal to the circumradius of the triangle ABC.
Solution The configuration is shown in Fig. 2.37. Because Y is the orthocenter
of AP C, the circumcircles of AP C and AY C are mapped into each other by the
reflection over AC. But the circumcircle of AP C is the same as the circumcircle of
AH C, and it reflects to the circumcircle of ABC. Hence, the circumcircle of AY C
coincides with the circumcircle of ABC, so Y is on the circumcircle of ABC.
X
H
B C
Fig. 2.37 Y is on the circumcircle of ABC
Next, let O be the circumcenter of ABC, and let O1 be the circumcenter of AP C,

which is the reflection of O over AC. We will show that there is a spiral similarity
that maps triangle O1 OX to triangle AOY . We continue our reasoning on Fig. 2.38.
Because OO1 is orthogonal to AC and OX is orthogonal to AB, O1 OX is the
supplement of BAC. On the other hand, ACY = 90◦ − CAP , because AP
is orthogonal to CY , so AOY = 2 CAP is the supplement of BAC as well.
Therefore, O1 OX = AOY .
Also, XO perpendicular to AB and XO1 perpendicular to AP implies
XO1 O = P AC = A/2. But then O1 XO = 180◦ − (180◦ − A) − A/2 =
A/2, so the triangle OO1 X is isosceles. But OA = OY because Y is on the
circumcircle of triangle ABC, so the triangle AOY is also isosceles. We thus have
two isosceles triangles OAY and OO1 X with equal angles at O, and they are
mapped into each other by a spiral similarity.
But spiral similarities come in pairs (Theorem 2.25), so the triangle OO1 A is
mapped into the triangle OXY by a spiral similarity. But OO1 = OX, so the two
triangles are congruent, and therefore XY = O1 A. Now the conclusion follows
from the fact the circumcircles of AH C and ABC have the same radius, as they are
reflections of each other over AC. Hence O1 A is equal to the circumradius of ABC,
and so is XY .
It is worth observing that point Y can be located easily with complex numbers.
Place the triangle in a coordinate system so that the origin is at the circumcenter O
and A has coordinate 1. Let the coordinates of B and C be eiβ and eiγ , respectively.
Then BAC = BOC = (γ − β)/2. By using inscribed angles in the circle of
O1
X
P
O
B C
Fig. 2.38 The existence of the spiral similarity
center O1 , we obtain P O1 C = 2 P AC = BAC, so P is the rotate of C about

O1 by −(γ −β)/2. The reflection of P over AC, which we denote by P , is therefore
obtained by rotating C about O by (γ − β)/2. Its coordinate is
γ −β 3γ −β
p = ei 2 eiγ = ei 2 .
We obtain P by reflecting P over AC; by using Proposition 1.6 and performing the
computations, we deduce that the coordinate of P is
β−γ
p = −ei 2 − eiγ − 1.
The triangle AP C is inscribed in a circle centered at the origin, so the coordinate of

the orthocenter is the sum of the coordinates of the vertices (see the discussion on
the Euler line). Hence the coordinate of Y is

−i γ −β i π − γ −β
y = −e 2 =e 2
.
From this formula we read that Y is on the circumcircle of ABC and that the angle
AOY is the supplement of the angle BAC.

The third example is from the short list of the 1978 International Mathematical
Olympiad, proposed by Bulgaria.
Fig. 2.39 SA N and SB M B

are similar
A
S
B A
N
Problem 2.7 Let AOB be an equilateral triangle with centroid S, and let A OB
be another equilateral triangle such that A = S, B = S and such that the angles
A OB and AOB have the same orientation. Let M be the midpoint of A B, and
let N be the midpoint of AB . Prove that the triangles SA N and SB M are similar.
Solution We consider the triangles placed as in Fig. 2.39. The composition of the
homothety of center A and ratio 2 and the clockwise rotation about O is a spiral
similarity that maps N to A and keeps S fixed. The center of the spiral similarity
is the fixed point S. Also, the homothety of center B and ratio 2 followed by the
counterclockwise rotation about O maps M to B , and S is again its fixed point.
Thus we have another spiral similarity, of center S, with the same ratio, but with
opposite angle. We therefore have SN/SA = SM/SB = 1/2 and NSA =
MSB = 60◦ . This proves that triangles SA N and SB M are similar, as desired.

We conclude the discussion with a simple example that illustrates how to use the
general form of the Averaging Principle.
Problem 2.8 Two circles ω1 and ω2 intersect at M and N . Through M we take
a line that intersects ω1 and ω2 again at A and B, respectively. In the half-plane
determined by AB that does not contain N, construct a point C such that the triangle
ABC is similar to a given triangle. What is the locus of C when the line varies?
Solution 1 The configuration is shown in Fig. 2.40. Assign to points complex

coordinates using the convention that the lowercase complex number corresponds
to the uppercase point. Because the triangle ABC is similar to some fixed triangle
XY Z
c−a z−x
= w, where w = .
b−a y−x
Solving for c we deduce that there exist complex numbers t1 and t2 that do not
depend on such that c = t1 a + t2 b. By Problem 131, the spiral similarity that maps
Fig. 2.40 Finding the locus C

of C
B
M
ω2
ω1 N
ω1 to ω2 maps A to B. Now we can apply the Averaging Principle in its general

form, as stated in Theorem 2.19, to the figures ω1 and ω2 and the numbers t1 and t2
to deduce that as the point A traces the circle ω1 , the point C traces a figure similar
to ω1 , hence a circle. This circle is the “weighted mean” of ω1 and ω2 , with the
complex weights t1 and t2 .

Solution 2 Of course, there is solution that does not use the Averaging Principle
explicitly. Again by using the result proved in Problem 131, we deduce that the
spiral similarity of center N that maps ω1 to ω2 maps A to B.
Since the ratios NA/AB and AC/AB do not depend on the position of ,
N A/AC does not depend on this position either. Combining this with the fact that
NAC = NAB + BAC does not depend on the position of , we deduce that
all triangles N AC are similar and oriented the same way. In particular, the angle
CNA and the ratio CN/AN do not depend on the position of .
We deduce that the spiral similarity of center N, angle CN A, and ratio CN/AN
maps A to C. Hence the locus is the image of the circle ω1 through this spiral
similarity. Because the spiral similarity is a bijection, it establishes a one-to-one
correspondence between the points of ω1 and the locus, so the locus is actually the
whole circle.

Spiral Similarities
Endowed with the methods and ideas presented above, you are now invited to solve
the following problems.
135 Given a triangle ABC and a polygon, show that there exist points M, N, P on
the sides of the polygon such that the triangles MNP and ABC are similar.
136 Given a square ABCD, let P and Q be two points on the sides AB and BC,
respectively, such that BP = BQ. Let H be the projection of B onto P C. Prove
that DH Q = 90◦ .
137 (a) Given two regular pentagons A1 A2 A3 A4 A5 and A1 A2 A3 A4 A5 oriented
the same way, prove that the midpoints of A1 A1 , A2 A2 , A3 A3 , A4 A4 , and A5 A5
form a regular pentagon.
(b) Given two regular pentagons A1 A2 A3 A4 A5 and A1 A2 A3 A4 A5 with common
vertex A1 and oriented the same way, show that the lines A2 A2 , A3 A3 , A4 A4 , and
A5 A5 pass through the same point.
138 Let ABCD be a square with center O, and let M and N be the midpoints of
BO and CD, respectively. Prove that AMN is an isosceles right triangle.
139 Let ABC and A B C be two equilateral triangles such that the segments BC
and B C have the same midpoint. Find the ratio between the segments AA and
BB and the angle that they form.
140 Given two directly similar triangles, ABC and A1 B1 C1 , and an arbitrary point
O in the plane, consider the points A2 , B2 , and C2 such that AA1 OA2 , BB1 OB2 ,
and CC1 OC2 are parallelograms. Prove that the triangles ABC and A2 B2 C2 are
similar.
141 Given a convex polygon A1 A2 . . . An of area S and a point M in the plane,
find the area of the polygon M1 M2 . . . Mn , where Mj is the image of M under the
rotation of center Aj and angle α, j = 1, 2, . . . , n.
142 Two circles intersect at the points A and B, and the chords AM and AN
are tangent to these circles. We consider the point C such that AMCN is a
parallelogram, and on the segments BN and CM, we consider the points P and
Q, respectively, such that BP /P N = MQ/QC. Prove that AP Q = ANC.
143 Let ω1 , ω2 , · · · , ωn be n circles passing through O. A grasshopper starts at
a point X1 ∈ ω1 and jumps to the points X2 , X3 , X4 , . . . such that Xi ∈ ωi and
Xi Xi+1 passes through the other point of intersection of ωi and ωi+1 for all i (where
ωn+1 = ω1 ). Show that the grasshopper returns to the initial position.
144 On the sides of the triangle ABC, construct the similar triangles BP A, AQC,
and BRC such that the first two are in the exterior of the triangle ABC and the third
is on the same side of BC as the point A. Prove that AP RQ is a parallelogram.
145 A circle passing through the vertex A and the circumcenter O of an acute
triangle ABC intersects the sides AB and AC at the points P and Q, respectively.
Prove that the orthocenter of the triangle P OQ lies on BC.
146 Let ABCD be a trapezoid with AD||BC and B > 90◦ . Let M be a point on
the side AB, let O1 and O2 be the circumcenters of the triangles MAD and MBC,
and let N be the second point of intersection of the circumcircles of MO1 D and
MO2 C. Prove that the line O1 O2 passes through N.
147 Let ABC be a triangle, and let be a line that intersects the lines BC, CA,
AB at D, E, and F , respectively. Construct the triangles A1 F E, F B1 D, and EDC1
that are similar to the triangle ABC and oriented the same way as ABC. Show that
the spiral similarities that map these triangles into one another have the same center.
148 The circles ω1 , ω2 centered at O1 , O2 , respectively, intersect at P and S. The
points A, B on ω1 and C, D on ω2 are chosen such that the segments AC and BD
intersect at P . Denote the midpoints of the segments AC, BD, O1 O2 by M, N, O,
respectively. Prove that O is the circumcenter of the triangle MNP .
149 Let ABC be a triangle in the plane. In the exterior of this triangle, construct
the triangles ABR, BCP , and CAQ such that
P BC = CAQ = 45◦ , BCP = QCA = 30◦ , ABR = RAB = 15◦ .
Prove that P RQ = 90◦ and that P R = QR.

150 Let ABC be a triangle, and let E and F be two arbitrary points on the
sides AB and AC, respectively. The circumcircle of the triangle AEF meets the
circumcircle of the triangle ABC again at the point M. Let D be the reflection of
M over EF , and let O be the circumcenter of the triangle ABC. Prove that D is on
BC if and only if O is on the circumcircle of the triangle AEF .
151 On the sides BC, CA, and AB of a triangle ABC, we consider the points
A1 , B1 , C1 , respectively, such that the triangles ABC and A1 B1 C1 are similar.
Denote by A2 , B2 , C2 the intersections of BB1 and CC1 , CC1 and AA1 , and AA1
and BB1 , respectively, and assume that these three points are distinct. Prove that
the circumcircles of the triangles ABC2 , BCA2 , CAB2 , A1 B1 C2 , B1 C1 A2 , and
C1 A1 B2 have a common point.
152 The points P and Q are chosen on the side BC of an acute-angled triangle
ABC so that P AB = ACB and QAC = CBA. The points M and N are
taken on the rays AP and AQ, respectively, so that AP = P M and AQ = QN.
Prove that the lines BM and CN intersect on the circumcircle of the triangle ABC.
153 Let ABC be a triangle with AB < AC; let D be the point where the
perpendicular bisector of the side BC intersects the side AC, and let E be the point
where the angle bisector of ADB intersects the circumcircle of ABC. Prove that
the angle bisector of AEB and the line that passes through the incenters of the
triangles ADE and BDE are orthogonal.
154 Consider a convex pentagon ABCDE such that
BAC = CAD = DAE and CBA = DCA = EDA.
Let P be the point of intersection of the lines BD and CE. Prove that the line AP
passes through the midpoint of the side CD.
155 Let ABC be a scalene triangle, and let X, Y, Z be points on the lines
BC, CA, AB, respectively, such that AXB = BY C = CZA. The circum-
circles of the triangles BXZ and CXY meet again at P = X. Prove that P lies on
the circle of diameter GH , where G and H are the centroid and the orthocenter of
the triangle ABC, respectively.
156 Let ABC be an acute triangle with orthocenter H , and let W be a point on
the side BC, lying strictly between B and C. The points M and N are the feet of
the altitudes from B and C, respectively. Denote by ω1 the circumcircle of BW N ,
and let X be the point on ω1 such that W X is a diameter of ω1 . Analogously, denote
by ω2 the circumcircle of triangle CW M, and let Y be the point such that W Y is a
diameter of ω2 . Prove that X, Y , and H are collinear.
Chapter 3
Inversions
3.1 Theoretical Results About Inversion
A great leap has to be taken from those geometric transformations of the first two
chapters to inversion, and the intuition behind the latter is more difficult to explain.
We will need a longer theoretical exposition to fully understand this transformation.
Inversion should be thought of as reflection over a circle, with the points inside the
circle being reflected outside and the points outside of the circle being reflected
inside. With inversion, the point at infinity makes its appearance; it is the image of
the center of the circle of inversion. There is just one point at infinity! The idea of
inversion is suggested in Fig. 3.1.
3.1.1 The Definition of Inversion and Some of Its Properties
We state the following definition.

Definition Given a circle of center O and radius R, the inverse of a point P = O
with respect to the circle is a point P such that P lies on the ray from O through
P and
OP · OP = R 2 .
The transformation defined this way is called inversion with respect to the circle of
center O and radius R, or inversion with center O and radius R. The quantity R 2 is
called the ratio of the inversion.
Inversion is represented schematically in Fig. 3.2.
Inversion is not defined at its center O, which can be an issue when composing
several inversions with different centers. The problem is solved by adding to the
130 3 Inversions
Fig. 3.1 The idea of

inversion
Fig. 3.2 The definition of

inversion with respect to a
circle
P
O P
plane a single point at infinity, denoted by ∞. And this means that in whichever
direction of the plane you look, you see this point at infinity, all lines pass through
it. Then O is mapped to the point at infinity and vice versa. This renders inversion a
bijective transformation of the plane enlarged with the point at infinity.
Inversion shares several properties with reflection over a line:
• Composing an inversion with itself yields the identity map: if P is mapped to P ,
then P must be mapped to P (because OP · OP = OP · OP ). This means
that inversion is an involution, exactly like reflection over a line is.
• The points on the circle of inversion are invariant under inversion, exactly how
the points on the line of reflection are invariant under the reflection.
• Inversion maps points that are on one side of the circle of inversion to points
on the other side (meaning that the interior is mapped to the exterior and the
exterior is mapped to the interior), exactly in the same way as reflection maps
one half-plane to the other.
• Inversion, like reflection, changes the orientation of figures.
3.1 Theoretical Results About Inversion 131
We will see below that these similarities are not accidental and that in fact inversion
should be thought of as reflection over a circle, as suggested in the preamble. We
will arrive later there, and for that we will have to switch to coordinates.
The name “inversion” makes sense if you consider R as the unit, that is, R = 1:
then the inverse P of P satisfies OP = OP 1
. In this sense, inversion in a plane (or
in space) generalizes the notion of the inverse (reciprocal) of a number on the real
line.
It is quite easy to write in complex coordinates the formula for inversion over the
unit circle centered at the origin
1
z → .
z̄
Indeed, if we write z = reiθ , then 1/z̄ = r −1 eiθ , so on the one hand z and 1/z̄ are
on the same ray from the origin (having the same argument) and on the other hand
|z| · |1/z̄| = r · r −1 = 1.
The inversion with respect to an arbitrary circle of center a and radius R maps a
2
point z = a + reiθ to the point z∗ = a + Rr eiθ . This means that
(z − a)(z∗ − a) = R 2 ,
which gives
R2
z∗ = + a.
z−a
So in order to obtain the inverse of z with respect to the circle of center a and radius
R 2 , you should translate z by −a, take its inverse over the unit circle centered at
the origin, map the new point by the homothety of center the origin and ratio R 2 ,
and translate the result by a. Thus any inversion can be obtained by composing the
inversion with respect to the unit circle with some translations and homotheties.
Proposition 3.1 Let P be a point that does not lie on the circle of inversion, and
let P be its inverse. Let also M and N be the two points where P P intersects the
circle of inversion. Then P , M, P , N form a harmonic division.
Proof We recall that the collinear points P , M, P , N form a harmonic division if
exactly one of P and P separates M and N and P M/P N : P M/P N = 1, which
is equivalent, in complex coordinates, to the fact that their cross-ratio satisfies
p − m p − m
: = −1,
p − n p − n
where p, m, p , n are the complex coordinates of P , M, P , N , respectively.

For the proof, place the center of inversion at the origin of a system of coordinates
for which P P is the x-axis, the radius of inversion is 1, and M and N have
132 3 Inversions
Fig. 3.3 Similarity of B

triangles obtained from
inversion
B
O
A
A
coordinates 1 and −1, respectively. If p is the real number which is the coordinate
of P , then by the above result, P has coordinate 1/p̄ = 1/p. Then
p − m p − m p−1
1
p −1
: = : = −1,
p − n p − n p+1 1
p +1
and the proposition is proved.

Now that we understand what inversion does to a single point, we can study what
it does to a pair of points.
Theorem 3.2 Consider the inversion about the circle of center O and radius R. If
A is the inverse of A and B is the inverse of B, then the triangles OAB and OB A
R2
are (inverse) similar, and the similarity ratio is OA·OB . In particular OAB =

A B O.
Proof Arguing on Fig. 3.3, we see that that AOB = B OA , and also the
equality OA · OA = R 2 = OB · OB gives
OA OB R2
= = .
OB OA OA · OB
So the triangles are indeed similar, and the similarity ratio is as said.

We have written the equality of angles so that it holds for directed angles modulo
π , too.
Remark As a corollary we obtain a formula that will be used extensively in what
follows, a formula which describes how distances are distorted by inversion. If A
is the image of A and B is the image of B through the inversion of center O and
radius R, then
R2
A B = AB. (3.1)
OA · OB
Next, we learn how to construct the inverse of a point.
Fig. 3.4 Image of a point

lying in the exterior of the T1
circle of inversion
O X
X
T2
Proposition 3.3
(i) If a point X lies outside of the circle of inversion, then its image X is the
midpoint of the segment formed by the tangency points of the two tangents from
X to the circle of inversion.
(ii) If a point X lies inside the circle of inversion, then its image X is the point
obtained as the intersection of the tangents to the circle of inversion at the
points where the perpendicular to OX intersects the circle of inversion.
Proof We follow the proof of (i) on Fig. 3.4, where we have denoted the tangency
points by T1 and T2 . We notice that the midpoint X of T1 T2 satisfies OX · OX =
OT12 , as given by the “Leg Theorem” in the right triangle T1 OX. So X is the inverse
of X. Part (ii) follows from (i) using the fact that inversion is an involution.
3.1.2 Inverses of Lines and Circles
It is not hard to see that the image of a line that passes through the center of inversion
is the same line. It is worth including the point at infinity, so that the image contains
the center of inversion as well, a convention that we will tacitly follow. Throughout
this section, the inversion has center O and radius R.
Theorem 3.4 (Inverse of a Line) Let be a line that does not pass through the
center O of inversion, A the projection of O onto , A the image of A, and ω the
circle of diameter OA . Then the line and the circle ω are mapped into each other
by the inversion.
Proof We need not check the pairs of points (O, ∞) and (A, A ). Arguing on
Fig. 3.5, let P and P be two points that correspond through the inversion. Then
P ∈ \{A} is equivalent to OAP = 90◦ . By Theorem 3.2, this is equivalent to
134 3 Inversions
Fig. 3.5 Lines and circles

that are mapped into each
other
P
P
O A
A
A P O = 90◦ , and this is further equivalent to P ∈ ω\{O, A }. The theorem is

proved.

Remark So a line that does not pass through the center of inversion is mapped to a
circle that passes through the center of inversion and vice versa.
As a corollary, a line and a circle that are not tangent are mapped into each other
by exactly two inversions, whose centers are the endpoints of the diameter that is
perpendicular to the line.
Theorem 3.5 (Inverse of a Circle) Let ω be a circle that does not pass through the
center O of inversion, and let A and B be the endpoints of the diameter whose line
of support passes through O. Let also A , B be the images of A and B, respectively.
Then the image of ω through the inversion is a circle ω of diameter A B . In
particular the center of inversion lies on the line connecting the center of the circle
and the center of its image.
Proof 1 This argument can be followed on Fig. 3.6. Certainly we do not have to
discuss the pairs (A, A ) and (B, B ). Assume that A is between O and B, and let
P and P be a point and its image through the inversion. Then
AP B = OP B − OP A = P B O − P A O = B P A ,
where for the last step we have used the Exterior Angle Theorem in the triangle
A P B . This shows that AP B = 90◦ if and only if B P A = 90◦ . In other
words, P is on ω if and only if P is on ω . The theorem is proved.
Proof 2 There exists also a short complex number argument. Choose a coordinate
system with origin at the center of inversion and with the x-axis containing the
A B B A
O
Q
P P
Fig. 3.6 Circles that are mapped into circles
center of the circle, and let x ∈ R be the coordinate of this center. The equation of
the inversion is z → R 2 /z, where R is the radius of inversion. The equation of the
circle is
|z − x| = r,
and if w is the image (and also the preimage) of z through the inversion, then
2
R

w − x = r.
So this is the equation of the image of the circle through the inversion. To check that
this is the equation of a circle as well, we rewrite it as
R 4 − R 2 xw − R 2 xw + (x 2 − r 2 )ww = 0,
and then as
R2x R2x R4x 2 R4x 2 R4

ww − w − w + = − .
x2 − r 2 x2 − r 2 (x 2 − r 2 )2 (x 2 − r 2 )2 x2 − r 2
This is the equation of the circle

2
R 2 x R2r
w − = ,
x2 − r 2 x2 − r 2
with center R 2 x/(x 2 − r 2 ) and radius R 2 r/|x 2 − r 2 |.

Remark We have shown that a circle that does not pass through the center of
inversion is mapped to a circle that does not pass through the center of inversion.
To summarize, while isometries and spiral similarities map lines to lines and
circles to circles, inversion maps a line or a circle to a line or a circle.
136 3 Inversions
It is important to note that the center of inversion is also the center of the unique
homothety with positive ratio that maps the circles one into the other. Indeed, if
in our figure we denote the other point of intersection of OP and ω with Q, then
OP · OP = R 2 and OP · OQ = ρ, where ρ is the power of O with respect
to ω. Consequently OP = (R 2 /ρ)OQ, showing that P is the image of Q under
the homothety of center O and ratio R 2 /ρ. This can be read immediately in the
analytic proof. In particular, if the two circles admit common exterior tangents, then
the center of inversion lies at the intersection of those tangents (see Fig. 3.6). And
we have
Proposition 3.6 If ω and ω are circles of radii r and r , respectively, that are
mapped into each other by an inversion of center O and radius R, and if the power
of O with respect to ω is ρ, then
R2
r = r.
ρ
Unfortunately, the center of a circle is not mapped to the center of its image. This
is to be expected because sometimes the image of the circle is a line, which has no
center. Nevertheless we can still locate the image of the center of a circle through
the inversion, as the following result shows.
Proposition 3.7 The inverse of the center C of a circle ω through an inversion χ of
center O is
(i) the reflection of O over χ (ω) if χ (ω) is a line;
(ii) the midpoint of the segment determined by the points of tangency to χ (ω) of
the tangents from O if χ (ω) is a circle with O in its exterior;
(iii) the intersection of the tangents to χ (ω) at the points where the perpendicular
to OC at O intersects χ (ω) if χ (ω) is a circle and O is in its interior.
Proof The three cases are shown in Fig. 3.7. For (i), denote by P the other end of
the diameter of ω through the center O of the inversion, and let C = χ (C) and
P = χ (P ). Then OP · OP = OC · OC , and since OP = 2OC, it follows that
OC = 2OP , as desired.
For (ii), with the same notation as above, if OT T is a common tangent of ω and
ω = χ (ω) (T ∈ ω, T ∈ ω ), then OT C = OC T , hence the conclusion.

Finally, for (iii), note that if T is at the intersection of ω with the perpendicular
through O to OC, then the circumcircle of OCT has diameter OT , so it is tangent
to ω. Its image through the inversion is a line that is tangent to ω , and the proposition
is proved (please read in Sect. 3.1.6 the explanation of why inversion maps tangent
curves to tangent curves).

On the other hand, given two circles ω1 and ω2 , there is almost always a unique
inversion χ that maps ω1 to ω2 .
T
ω
T
ω
O C C
ω
T ω
T
O C C
O C C
ω
Fig. 3.7 The image through inversion of the center of a circle
Proposition 3.8 Let ω1 and ω2 be two circles in the plane with distinct radii. There
is a unique inversion χ that maps ω1 to ω2 . The center of inversion coincides with
the center of the unique homothety with positive ratio that maps ω1 to ω2 , except
when one of the circles is inside the other; in this case, the center of inversion
coincides with the center of the unique homothety with negative ratio that maps ω1
to ω2 .
Proof We consider two cases, both shown in Fig. 3.8. If ω1 is inside ω2 , let O and
−k be, respectively, the center and ratio of the inverse homothety h that maps ω1 to
ω2 . The point O must be inside both circles. Let be a line through O that meets ω1
at A and B and ω2 at B = h(A) and A = h(B) (yes, we reversed the points). Then,
using oriented segments, OA · OA = OA · (−k · OB) = k · (−OA · OB); this is k
times the power of O with respect
√ to ω1 , which is a positive constant. We can take
χ with center O and radius k|ρ|, and then ρ is the power of O with respect to ω1 .
The other case, where neither of the circles is inside the other, is handled
analogously, with signs (doubly) inverted, as it follows from both the homothety and
the fact that the center of the direct homothety that maps ω1 to ω2 is the intersection
of the two common external tangent lines to both circles, which is outside both
circles.
Theorem 3.2 shows that two points and their inverses determine uniquely the
inversion, so the inversion χ is unique.

138 3 Inversions
A
A
A
B B B
O A
B O
Fig. 3.8 Finding the center of the inversion that maps a circle to another circle
3.1.3 Möbius Transformations
As we have said in the first chapter, by introducing the concept of a group, we

introduced a new body of ideas, and one such idea is that we must understand the
compositions of geometric transformations. This was an easy task in the previous
chapters, but not so now, and to explain compositions of inversions, we must take a
step into a wider context. It will be the goal of the fourth chapter to truly take a global
look at all the transformations discussed in this book; here we intend to capture
only some essential facts about inversion that become clearer when discussed in
this general setting.
What we are doing now is to place ourselves in the framework of complex
projective geometry. To this end, as before, we add to the complex plane C one
point at infinity, denoted by ∞. The result, C ∪ {∞}, is not the real projective plane
(in that case we have to add an entire line at infinity), but the complex projective line,
also called the Riemann sphere and denoted by CP 1 by algebraic geometers. It is a
line in the sense that it has complex dimension equal to one, it is parametrized by
one complex number, and it is referred to as a sphere because to be able to add the
point at infinity you have to bend the plane into a sphere (we only bend it mentally;
the plane should stay straight so that we can talk about lines, segments, triangles,
etc.). The convention is that all real lines loop through the point at infinity.
Definition A Möbius transformation is a map φ : C ∪ {∞} → C ∪ {∞} of the form
az + b
φ(z) = , if z = −d/c, φ(−d/c) = ∞, and φ(∞) = a/c,
cz + d
where a, b, c, d are complex numbers satisfying the condition ad − bc = 0.

Möbius transformations, named after August Ferdinand Möbius, are also known
as linear fractional transformations of C, or as homographic functions (from the
French fonctions homographiques). Inversion is not here yet; it will show up in a

few moments.
It is immediate to observe that a Möbius transformation is one-to-one and onto
(from the Riemann sphere to itself) and that it does not change when multiplying
a, b, c, d simultaneously by the same nonzero number.
Theorem 3.9 The Möbius transformations form a group.
Proof In other words, the composition of two Möbius transformations is a Möbius
transformation, and the inverse of every Möbius transformation is a Möbius
transformation. Indeed, if
az + b a z + b
φ(z) = and φ (z) = ,
cz + d c z + d
then
az + b
a + b
cz + d (aa + cb )z + (a b + db )
φ ◦ φ(z) = = ,
az + b (c a + cd )z + (c b + dd )
c + d
cz + d
and
1 dz − b
φ −1 (z) = · .
ad − bc −cz + a
The reader familiar with the theory of matrices will immediately notice that if we
arrange the four numbers defining the Möbius transformation in a 2 × 2 matrix, then
composition corresponds to matrix multiplication, of course with a caveat that one
Möbius transformation is represented by several matrices which can be transformed
into one another by multiplication by nonzero scalars.

The group of Möbius transformations plays a sufficiently important role to have
a notation and a name attached to it; it is denoted by P SL(2, C) and is called the
complex projective special linear group. It is important to observe that this group is
not commutative, namely, that in general φ ◦ ψ = ψ ◦ φ.
For a Möbius transformation that is not the identity map, the equation
az + b
=z
cz + d
has at most two distinct solutions in C ∪ {∞}. Indeed, ∞ is a fixed point if and only
if c = 0, in which case any other fixed point is the solution to a linear equation,
and if c = 0, then the above equation is equivalent to the quadratic equation cz2 +
(d − a)z + b = 0, which can have at most two distinct complex solutions. Hence,
a Möbius transformation that is not the identity map has at most two fixed points.
This means that such a transformation is determined by the images of three points.
140 3 Inversions
Indeed, if φ and ψ are Möbius transformations that coincide at three points, then
φ ◦ ψ −1 has three fixed points, implying that it is the identity map, so φ = ψ.
Another way to look at this fact is by noticing that a Möbius transformation is
determined by four parameters, a, b, c, d, but one degree of freedom is lost since
the multiplication by a nonzero constant does not change the transformation (this is
one reason why the group has the word “projective” in its name.)
It is standard to look at the Möbius transformations that map three given points
z2 , z3 , z4 to 1, 0, ∞, respectively. Such a transformation is defined by the formula
z − z 3 z 2 − z3
φz2 ,z3 ,z4 (z) = : .
z − z4 z 2 − z4
There are three degenerate versions of this formula, corresponding to the cases
where z2 , z3 , or z4 is ∞, and they are, respectively,
z − z3 z − z3 ∞ − z3 z2 − z4 z − ∞ z2 − ∞ z − z3 z − z3 z2 − z3
= : , = : , = : .
z − z4 z − z4 ∞ − z4 z − z4 z − z4 z2 − z4 z2 − z3 z − ∞ z2 − ∞
And we recognize the formula of φz2 ,z3 ,z4 (z) to be the cross-ratio (z, z2 , z3 , z4 ) of
z, z2 , z3 , z4 ! As a rule, we equate all fractions of the form (a − ∞)/(b − ∞) with 1.
Having the above standard transformations at hand, we can express every Möbius
transformation as a composition of such standard transformations and their inverses.
In fact, given three distinct points z2 , z3 , z4 in C ∪ {∞} and three more distinct
points w2 , w3 , w4 , there is a unique Möbius transformation ψ for which z2 → w2 ,
z3 → w3 , and z4 → w4 given by ψ = φw −1 ◦ φz2 ,z3 ,z4 . In other words, we map
2 ,w3 ,w4
z2 , z3 , and z4 to the points 1, 0, and ∞ and then map these to the desired images
w2 , w3 , and w4 .
Theorem 3.10 (Invariance of Cross-Ratio) The cross-ratio is invariant under
Möbius transformations, that is if ψ is a Möbius transformation, then
(z1 , z2 , z3 , z4 ) = (ψ(z1 ), ψ(z2 ), ψ(z3 ), ψ(z4 )).
Proof It is worth changing back to the functional notation and writing this as
φz2 ,z3 ,z4 (z1 ) = φψ(z2 ),ψ(z3 ),ψ(z4 ) (ψ(z1 )).
This equality states that z1 is mapped by φz2 ,z3 ,z4 to the same point where ψ(z1 ) is
mapped by φψ(z2 ),ψ(z3 ),ψ(z4 ) . And this follows from
φψ(z2 ),ψ(z3 ),ψ(z4 ) = φz2 ,z3 ,z4 ◦ ψ −1 ,
which is true because both transformations map ψ(z2 ), ψ(z3 ), ψ(z4 ) to 1, 0, and
∞, respectively.

Recall that the cross-ratio of four points in the plane is the cross-ratio of
their complex coordinates, which is independent of the choice of coordinates. As
coordinate changes are Möbius transformations, the cross-ratio does not depend on
the coordinate system; it is an invariant of the configuration consisting of the four
points. Let us place in our context Theorem 1.4 from Chap. 1.
Theorem 3.11 (Condition for Concyclicity) Four points z1 , z2 , z3 , z4 in C ∪ {∞}
lie on a circle or on a line if and only if (z1 , z2 , z3 , z4 ) ∈ R.
And we have the following result.
Theorem 3.12 Möbius transformations map lines or circles into lines or circles.
Proof Let φ be a Möbius transformation. By Theorem 3.10,
(z1 , z2 , z3 , z4 ) = (φ(z1 ), φ(z2 ), φ(z3 ), φ(z4 )).
The first of these is real if and only if z1 , z2 , z3 , z4 are on a circle or line. The
second is real if and only if φ(z1 ), φ(z2 ), φ(z3 ), φ(z4 ) are on a circle or line. So
z1 , z2 , z3 , z4 are on a circle or line if and only if φ(z1 ), φ(z2 ), φ(z3 ), φ(z4 ) are on a
circle or line. Done.
3.1.4 Möbius Transformations Versus Isometries, Spiral

Similarities, and Inversions; Inversion and Circular
Transformations
We now want to understand the so-called circular group, obtained by adding

inversions to the group consisting of spiral similarities and translations. A good
description of the elements of P SL(2, C) is necessary. Some Möbius transforma-
tions we recognize immediately:
• a = d = 1, c = 0, z → z + b is a translation;
• b = c = 0, d = 1, z → az is a spiral similarity centered at the origin;
• a = d = 0, b = c = 1, z → 1z is the inversion over the unit circle centered at the
origin, composed with reflection over the x-axis.
We will call this last transformation complex inversion.
Theorem 3.13 Every Möbius transformation is the composition of translations,
spiral similarities, and the complex inversion.
Proof If c = 0, the map is a spiral similarity, a translation, or just the identity map.
If c = 0, scale a, b, c, d so that bc − ad = c (e.g., by multiplying each of them
by bc−ad
c ). Now take the composition
142 3 Inversions
1 1 1 1
z → z + α → → · → + γ.
z+α β z+α βz + αβ
Then β = c, α = dc , and γ = ac , and we are done.

Note that translations, spiral similarities, and the complex inversion preserve
orientation, so Möbius transformations also preserve orientation.
We have seen in Sect. 3.1.1 that the inversion χ about the circle whose center has
complex coordinate a and whose radius is R is given by the formula
R2
χ (z) = + a.
z−a
This formula shows that χ = σx ◦ φ, where φ is the Möbius transformation
āz + (R 2 − a ā)
φ(z) = ,
z−a
and σx is the reflection over the x-axis. So you can obtain any inversion from a
Möbius transformation and the reflection over the x-axis, z → z̄. It is therefore
natural to make the following definition.
Definition A circular transformation is a transformation of the Riemann sphere
that is either a Möbius transformation or the composition of a Möbius transforma-
tion and the reflection over the x-axis.
We leave it to the reader to check that circular transformations form a group. This
group contains, among many other transformations, the maps of the form f (z) =
az + b or f (z) = a z̄ + b that were studied in the previous chapters.
By Theorem 3.12, Möbius transformations map circles or lines into circles or
lines, and so does the reflection σx . We have thus produced the coordinate-based
proof of the fact that circular transformations, and in particular inversions, map
circles or lines to circles or lines.
We will now look at inversion from a modified point of view, which will then
allow us to conclude that inversion plays indeed the role of reflection over a
circle. We will conclude that the group of circular transformations is generated by
inversions and reflections over lines, or, said differently, by reflections over circles
and lines.
Let z2 , z3 , z4 be three distinct points in the Riemann sphere (i.e., plane with the
point at infinity). They can be either collinear, in which case they determine a line,
or noncollinear, in which case they determine a circle. If one of the three points is at
infinity, then they are collinear, since we have agreed that all lines pass through the
point at infinity. In this sense we can think of lines as circles that have one point at
infinity. With this convention, inversion always maps circles to circles.
Theorem 3.14 Let z2 , z3 , z4 be three distinct points in C ∪ {∞}, and let χ be
either the reflection over the line determined by z2 , z3 , z4 if they are collinear or
the inversion over the circle determined by them if they are not. Then
(χ (z), z2 , z3 , z4 ) = (z, z2 , z3 , z4 ).
Proof Let us verify first the case where z2 , z3 , z4 are noncollinear. Because the
cross-ratio is invariant under Möbius transformations (Theorem 3.10), it is invariant
under isometries and spiral similarities. Inversion is well behaved under translations
and spiral similarities (meaning that the images of a point and its inverse correspond
through the inversion over the image of the circle). So we may assume that the circle
of inversion is the unit circle centered at the origin, and so that z2 z2 = z3 z3 =
z4 z4 = 1. Applying Theorem 3.10 to the Möbius transformation w → 1/w, and
using the fact that χ (z) = 1/z, we can write

1 1 1
(χ (z), z2 , z3 , z4 ) = z, , , = (z, z2 , z3 , z4 ) = (z, z2 , z3 , z4 ).
z2 z3 z4
If z2 , z3 , z4 are collinear, we may assume that they lie on the real axis, so zj = zj ,
j = 2, 3, 4. In this case the equality is
(χ (z), z2 , z3 , z4 ) = (z, z2 , z3 , z4 ) = (z, z2 , z3 , z4 ).

Remark In other words, the reflection over a line passing through z2 , z3 , z4 is
defined by the formula (χ (z), z2 , z3 , z4 ) = (z, z2 , z3 , z4 ), and the same formula
defines inversion over the circle passing through z2 , z3 , z4 . That is why we interpret
inversion as reflection over a circle. And this result allows us to find the formula for
the reflection χ (z) over a circle determined by three points z2 , z3 , z4 by solving
χ (z) − z3 z2 − z3 z − z3 z 2 − z3
: = : .
χ (z) − z4 z2 − z4 z − z4 z 2 − z4
In light of this result, we say that χ (z) and z are symmetrical with respect to the
line or circle determined by z2 , z3 , z4 .
Theorem 3.15 (The Symmetry Principle) If f is a circular transformation, and
if P and P are points that are symmetrical with respect to a line or circle C, then
f (P ) and f (P ) are symmetrical with respect to the line or circle f (C).
Proof This is easy to check if f is the reflection over the x-axis, so the problem
reduces to checking this property for Möbius transformations. We do this with
coordinates. Let f be a Möbius transformation, and let z and z be symmetrical
with respect to the circle or line passing through z2 , z3 , z4 . Then, by Theorem 3.14
(z , z2 , z3 , z4 ) = (z, z2 , z3 , z4 ).
144 3 Inversions
This implies, by Theorem 3.10, that
(f (z ), f (z2 ), f (z3 ), f (z4 )) = (f (z), f (z2 ), f (z3 ), f (z4 )),
Hence, again, by Theorem 3.14, f (z) and f (z ) are symmetrical with respect to
the circle or line through f (z2 ), f (z3 ), f (z4 ), which is the image through f of the
circle or line through z2 , z3 , z4 . The theorem is proved.

So, if two points are the image of each other through an inversion with respect to
some circle, and you perform another inversion that maps this circle to some circle
(or line), then the images of the points correspond under the inversion over this new
circle (or under reflection over the line).
Remark The Symmetry Principle allows us to give another proof to Proposition 3.7.
Assume that we are given an inversion χ of center O, and a circle ω of center C.
Then C and ∞ are reflections of each other over ω. So χ (C) and O = χ (∞) should
be reflections of each other over χ (ω). Using Proposition 3.3, and depending on
whether χ (ω) is a line or a circle, and in the latter case depending on the relative
position of the circles, we obtain the three cases from Proposition 3.7.
Now we can describe the structure of the circular group.
Theorem 3.16 Every circular transformation is the composition of reflections over
lines and circles.
Proof Every circular transformation is either a Möbius transformation or the
composition of a Möbius transformation and the reflection over the x-axis. So
it suffices to prove the property for Möbius transformations. By Theorem 3.13,
every Möbius transformation is the composition of translations, spiral similarities,
and the complex inversion, and spiral similarities are compositions of rotations
and direct homotheties. Translations and rotations are compositions of reflections
(Theorem 1.7), and so is the complex inversion. To complete the argument, we will
prove synthetically that direct homotheties are compositions of inversions.
Let h be the homothety of center O and ratio k > 0. Consider the inversions
√ χ1
of center O and radius 1 and the inversion χ2 of center O and radius k. Let also
P = O be a point in the plane. Then P = χ1 (P ) ∈ |OP and OP = 1/OP . If
we let P = χ2 (P ) = χ2 (χ1 (P )), then P ∈ |OP and OP = k/OP = kOP .
Consequently χ2 ◦ χ1 = h, and the theorem is proved.

This result should be compared with Theorem 1.7.
While Möbius transformations preserve cross-ratios, the argument of the cross-
ratio changes to its negative when reflecting over a line or circle. So although cross-
ratio is not invariant under circular transformations, its absolute value
|z1 − z3 | |z2 − z3 |
:
|z1 − z4 | |z2 − z4 |
is. Real valued cross-ratios are invariant under circular transformations. So on the
one hand, we have the following result:
Proposition 3.17 If A, B, C, D are four points in the plane, and A , B , C , D are
their images through an inversion (or in general a circular transformation), then
AC BC A C B C
: = : .
AD BD AD BD
On the other hand, if the four points lie on a circle or a line, then by combining
Theorems 3.10, 3.11, and 3.14, we obtain the following result:
Theorem 3.18 (Invariance of Cross-Ratio) If four points lie on a line or circle,
then the cross-ratio of their complex coordinates is invariant under circular
transformations.
As a final observation, we recall that the condition for the points a, b, c, d to
form a harmonic quadrilateral or a harmonic division on a line is (a, c, b, d) = −1.
As a consequence of Theorem 3.18, circular transformations, and in particular
inversion, map harmonic quadrilaterals to harmonic quadrilaterals (or to harmonic
divisions when the image of the circle is a line). It is this invariance under all
circular transformations of the property of a quadrilateral to be harmonic that makes
harmonic quadrilaterals so organically related to the geometric transformations
studied in this book and explains why they show up in so many problems.
3.1.5 Linear Fractional Transformations of the Real Line
Cross-ratios are probably a more familiar subject when talking about segments on
a line, so let us open briefly a parenthesis, which will explain why we use the word
“projective” when talking about Möbius and circular transformations. On a line
we choose a coordinate system that identifies it with the x-axis and add to it the
point at infinity to obtain what is called the projective real line. The linear fractional
transformations1 that map R ∪ {∞} into itself are of the form
ax + b
φ(x) = , a, b, c, d ∈ R, ac − bd = 0.
cx + d
These linear fractional transformations form what is called the real projective special
linear group P SL(2, R). As we have seen, the cross-ratio of four points on a line is
invariant under linear fractional transformations of the line.
1 The name Möbius transformation is restricted to the case of complex numbers.

146 3 Inversions
Theorem 3.19 Given two lines 1 and 2 in the plane and a point P on neither of
them, define a map f : 1 → 2 by associating to a point X ∈ 1 the point f (X) on
2 such that P , X, f (X) are collinear; if no such point exists, associate the point at
infinity of 2 , and if for some X ∈ 2 , P X is parallel to 1 , then X is the image of
the point at infinity of 1 . If we fix an origin on each line, thus identifying each line
with R, then f is a linear fractional transformation of R.
Proof 1 If the lines do not intersect, then the map is just a homothety, so let us
consider the case where the lines intersect (Fig. 3.9). As translations and dilations
are linear fractional transformations, we may assume that on each line the origin is
fixed at the intersection O of the two lines, some orientation is chosen, and the real
coordinate of a point is specified by its (oriented) distance to the origin. Consider
the points X ∈ 1 , X = f (X) ∈ 2 as given by the statement, and let OP = a,
OX = x, OX = x , P OX = α, P OX = β, XP O = θ . Then the Law of
Sines in the triangles P OX and P OX gives
a x a x
= and = .
sin(α + θ ) sin θ sin(β + θ ) sin θ
Using the addition formula the sine function we obtain
a sin α cos θ + cos α sin θ

= = sin α cot θ + cos α
x sin θ
and similarly
a
= sin β cot θ + cos β.
x
Substituting cot θ from the first equation into the second, we obtain
Fig. 3.9 Definition and P

computation of the map f
θ
1
X
α
β
O
2
X
a a
x = f (x) = = a .
sin β cot θ + cos β sin β − cot α + cos β
x sin α
Therefore
(a sin α)x
f (x) = ,
(sin α cos β − cos α sin β)x + a sin β
which is a linear fractional transformation of the real line.

The map X → X defined in this theorem is called a central projection, and the
theorem can be phrased as follows: The central projection of one line to another
defines a linear fractional transformation of R ∪ {∞}. This is why the adjective
“projective” is associated with linear fractional transformations.
Proof 2 We will now give an alternative proof borrowing some ideas from algebraic
geometry. The real projective line R ∪ {∞} can be constructed as follows: place an
observer at the origin O of the coordinate plane R2 , whose points are parameterized
by two real numbers. Each point (x1 , x2 ) ∈ R2 different from the origin can be
interpreted as a direction of view of the observer, and these directions of view allow
the observer to perform a central projection from one line that lies in R2 and does
not pass through the origin to another. Even the point at infinity of one line can
be projected to the other. So the points of the real projective line are in one-to-
one correspondence with the directions of view. To enhance your intuition, we have
illustrated this in Fig. 3.10, with the directions of view represented as dotted lines
(note the direction parallel to 1 that allows us to project ∞ ∈ 1 onto 2 ).
Each direction of view is a line passing through the origin, of the form
{(tx1 , tx2 ) | t ∈ R}. Because of this, the real projective line can be identified with
the set of equivalence classes of pairs of real numbers (x1 , x2 ), with x1 and x2 not
Fig. 3.10 A construction of the real projective line R ∪ {∞}

148 3 Inversions
simultaneously equal to zero, where (x1 , x2 ) is equivalent to (x1 , x2 ) if and only if
there is a nonzero real number t such that x1 = tx1 and x2 = tx2 .
The equivalence class of (x1 , x2 ) is denoted by [x1 : x2 ]. With the reference line
1 of equation x2 = 1, we notice that whenever x2 = 0, the line {(tx1 , tx2 ) | t ∈ R}
intersects 1 at (x1 /x2 , 1). So the map
[x1 : x2 ] → (x1 /x2 : 1), [1 : 0] → ∞
is a bijection from the set of directions of view to the real projective line, and it is
realized geometrically by intersecting the directions of view with the line 1 .
Now suppose that 2 is another line in R2 that does not pass through the origin,
and let us perform the central projection from O of 1 onto 2 . We can do this
as follows: map 1 back to the set of directions of view, rotate the directions of
view and dilate until 2 becomes the line x2 = 1, and then project onto 2 . The
rotation+dilation is realized by a linear map
R2 → R2 , (x1 , x2 ) → (ax1 + bx2 , cx1 + dx2 ), where ad − bd = 1,
which realizes the central projection of 1 onto 2 as the map
R ∪ {∞} → R ∪ {∞}, [x1 : x2 ] → [ax1 + bx2 : cx1 + dx2 ].
For x2 = 0 this map is [x1 /x2 : 1] → [(ax1 + bx2 )/(cx1 + dx2 ) : 1], which after
setting x = x1 /x2 becomes [x : 1] → [(ax + b)/(cx + d) : 1], a linear fractional
transformation.

Here is an immediate corollary of the invariance of cross-ratios under linear
fractional transformations.
Theorem 3.20 Let 1 and 2 be two lines, and let P be on neither. Let also
A, B, C, D be four points on 1 , and let A , B , C , D be their images through
the projection of center P . Then the cross-ratio of A, B, C, D and the cross-ratio of
A , B , C , D are equal.
Cross-ratio is invariant under central projections. We deduce that for a pencil of
four concurrent lines like the one shown in Fig. 3.11, the cross-ratio of the four
points obtained by intersecting the pencil with another line is independent of this
intersecting line. It is an intrinsic invariant of the pencil itself. A pencil whose cross-
ratio is equal to −1 is called harmonic.
As an aside for the curious reader, let us point out that the framework of the
second solution can be constructed for the complex projective line C ∪ {∞} as
well. This line can be identified with the set of equivalence classes of pairs of
complex numbers (z1 , z2 ) not both equal to zero, which are the directions of view
of an observer sitting at the origin of the two-dimensional complex space C2 , where
(z1 , z2 ) is equivalent to (z1 , z2 ) if and only if there is a nonzero complex number λ
such that z1 = λz1 and z2 = λz2 . The equivalence class of (z1 , z2 ) is denoted by
Fig. 3.11 A pencil of four

lines crossed by two lines
[z1 : z2 ], and it is important to notice that whenever z2 = 0, [z1 : z2 ] = [z1 /z2 : 1].
Hence, C is identified with the set of points of the form [z : 1] and ∞ is the
point [1 : 0]. In this framework, the Möbius transformation z → az+b cz+d assumes
the form [z1 : z2 ] → [az1 + bz2 : cz1 + dz2 ]. It arises from “projecting” a linear
transformation of the two-dimensional complex space onto the complex projective
line!
3.1.6 The Invariance of Angles
Like isometries and spiral similarities, inversions preserve angles. But because lines
are usually transformed into circles, we have to look at angles in a more general
sense. Given two curves intersecting at some point, if the tangent lines to the curves
at that point exist, then the angle between the curves is, by definition, the angle
formed by their tangents. For two intersecting lines, this recovers the classical
definition of angles. And the angle between two intersecting circles is the angle
between the tangents at any of the two intersection points (the two angles are equal
by symmetry). This is illustrated in Fig. 3.12.
In general, one defines the angle between two intersecting curves using coordi-
nates as follows. Parametrize the curves as γ1 , γ2 : [0, 1] → C, so they intersect at
some t0 ∈ (0, 1), meaning that γ1 (t0 ) = γ2 (t0 ). The curves have tangents at t0 if
they are differentiable with respect to t and have nonzero derivative at t0 , meaning
that
dγj γj (t0 + h) − γ (t0 )
(t0 ) = lim , j = 1, 2,
dt h→0 h
exist and are nonzero numbers. The angle between the curves is the angle between
the tangents and is
150 3 Inversions
Fig. 3.12 The angle between

two circles
Fig. 3.13 Inversion γ

preserves the angle between a γ
curve and a ray through O N
N
M M
O
tM γ tM γ
dγ2 dγ1
arg (t0 ) − arg (t0 ).
dt dt
In this book we are only concerned with the angles formed by circles and lines, in
which case we can rely on synthetic geometry; we use this general definition only
for the proof of the results in this section. We denote the tangent line to the curve γ
at the point X by tX γ .
Theorem 3.21 (Invariance of Angles) The angle formed by two curves is invariant
under inversion, but the orientation of angles is reversed.
Proof Let the center of inversion be O. Let also γ : [0, 1] → C be a curve that has
a tangent at M = γ (t0 ) for some t0 ∈ (0, 1). We will show that the angle between γ
and OM is invariant under inversion. The argument can be followed on Fig. 3.13.
Using dashes for inverted points and curves, we have
(OM, tM γ ) = lim (OM, MN ) = lim OMN = lim ON M

N →M N →M N →M
= lim (ON , N M ) = (OM , tM γ ).

N →M
Here we have used tacitly the property that if γ is differentiable, then its image γ
is differentiable as well; in other words, if γ admits a tangent line at a point, the so
γ1
γ2
t M γ1
O M
t M γ2
Fig. 3.14 Proof that inversion preserves angles
does its image. And we used the fact that if N ∈ γ and N approach M, then N ∈ γ
and N approach M .
Next, consider the line OM connecting the center of inversion with the intersec-
tion point of the two curves, as shown in Fig. 3.14. Using what we just proved, we
obtain
(tM γ2 , tM γ1 ) = (tM γ2 , OM) + (OM, tM γ1 )

= (tM γ2 , OM ) + (OM , tM γ1 ) = (tM γ2 , tM γ1 ).

If an inversion maps M, N, P to M , N , P , is it true that
MNP = M N P ?
Absolutely not. The segments MN and NP are mapped into arcs of circles. What
we have proved is that the angle between these arcs is equal to the angle MNP .
As a corollary of Theorems 3.21 and 3.16, we obtain the following.
Theorem 3.22 (Invariance of Angles) Circular transformations preserve angles
(if the orientation of angles is ignored).
Here is a short proof of this result using complex analysis; the reader unfamiliar
with this technique can skip it.
Proof Since every circular transformation is either a Möbius transformation, or the
composition of a Möbius transformation and the reflection over the x-axis, it suffices
to check the result for Möbius transformations.
Let γ1 , γ2 : [0, 1] → C be the parametrization of two curves that intersect at
z0 = γ1 (t0 ) = γ2 (t0 ), t0 ∈ (0, 1). Let also φ be a Möbius transformation. The angle
between γ1 and γ2 at t0 is
dγ2 dγ1
arg (t0 ) − arg (t0 ),
dt dt
while the angle between their images is
152 3 Inversions
dφ ◦ γ2 dφ ◦ γ1
arg (t0 ) − arg (t0 ).
dt dt
The chain rule gives
dφ ◦ γj dφ dγj
(t0 ) = (z0 ) (t0 ), j = 1, 2,
dt dz dt
so
dφ ◦ γj dφ dγj
arg (t0 ) = arg (z0 ) + arg (t0 ), j = 1, 2.
dt dz dt
Thus φ rotates both the tangent to γ1 and the tangent to γ2 by the same angle
arg dφ
dz (z0 ). This means that the angle between the tangents is preserved under φ.

We point out that if
az + b
φ(z) = ,
cz + d
then
dφ ad − bc
= .
dz (cz + d)2
For Möbius transformations, we have the stronger result.

Theorem 3.23 Möbius transformations preserve angles, including their orienta-
tion.
Circular transformations, and in particular inversion, map tangent curves to
tangent curves and orthogonal curves to orthogonal curves. Two curves are called
orthogonal if their tangents at the intersection point form a right angle.
3.1.7 Inversion with Negative Ratio
Let us point to a particular type of circular transformation, known as inversion

with negative ratio. For R > 0, the inversion of ratio −R 2 and center O is the
transformation that maps every point P in the plane to a point P such that O
belongs to the segment P P and OP · OP = R 2 . This circular transformation
is the composition of
• the inversion of center O and radius R, which maps P to P ∈ |OP such that
OP · OP = R 2
H
A
B C
B A
C
H
B C B C
A A
Fig. 3.15 The nine-point circle is the image of the circumcircle through an inversion
• the reflection over O.

This transformation is of interest because it appears in the geometry of the triangle.
Let ABC be a triangle that is not right, and let H be its orthocenter. Then
AH · H A = H B · H B = H C · H C = 4R 2 cos A cos B cos C,
where R is the circumradius. Thus A , B , C are the images of A, B, C, respec-

tively, through an inversion of center H and ratio equal to −4R 2 cos A cos B cos C.
This ratio is negative if the triangle is acute and positive if the triangle is obtuse. We
thus have the following result (see also Fig. 3.15).
Proposition 3.24 The nine-point circle of a triangle that is not right is the image
of the circumcircle through an inversion whose center is the orthocenter and whose
ratio is negative if the triangle is acute and positive if the triangle is obtuse.
3.1.8 Circles Orthogonal to the Circle of Inversion
As said above, two circles are orthogonal if their tangent lines at either of the two
intersection points form a right angle. Two circles are orthogonal if the radii of one
circle at the points of intersection are tangent to the other circle.
Proposition 3.25 A circle that is not the circle of inversion is mapped into itself by
the inversion if and only if it is orthogonal to the circle of inversion.
154 3 Inversions
Proof In order for a circle ω to be invariant under inversion about , it cannot lie
entirely inside or entirely outside of , so it intersects at some point O. Take
an inversion χ of center O. By the Symmetry Principle (Theorem 3.15), χ (ω) is
invariant under inversion about χ (). Both χ (ω) and χ () are lines, so this is
better phrased as saying that χ (ω) is invariant under reflection over χ (). And this
is equivalent to the fact that χ (ω) and χ () are orthogonal. As χ preserves angles,
ω is invariant under inversion about if and only if ω and are orthogonal.

The situation is illustrated in Fig. 3.16. And here is an application of this fact.
Theorem 3.26 Let χ be an inversion about the circle , and let ω be a circle such
that ω = χ (ω) is also a circle. Then , ω, ω are coaxial, meaning that they have
a common radical axis.
Proof The proof is simple if ω and intersect. The two intersection points belong
also to ω , and the common radical axis passes through these points.
If and ω do not intersect, let T be the point where their radical axis intersects
the line of centers (Fig. 3.17). The four segments from T that are tangent to and
√
ω have equal lengths p, where p is the power T with respect to the two circles.
So the four tangency points lie on a circle γ that is centered at T and orthogonal
to and ω. This circle is invariant under inversion and remains orthogonal to ω
because inversion preserves angles. Hence, the segment from T that is tangent to ω
Fig. 3.16 Circle orthogonal

to the circle of inversion
invariant points
R
O
Fig. 3.17 The circle of

inversion, a circle, and its
inverse are coaxial ω
ω T
Ω
√
is a radius in γ , so it has length p. This implies that the power of T with respect
to ω is also p, and so T lies on the common radical axis of the three circles.
3.1.9 The Limiting Points of Two Circles
A surprising result:
Theorem 3.27 Given two nonintersecting circles, there is an inversion that maps
them into two concentric circles.
Proof 1 Two concentric circles are characterized by the fact that there are infinitely
many lines but no circles that are orthogonal to both. The lines orthogonal to the
two circles pass through the common center, and they actually have two common
points: the center of the circles and the point at infinity. From this observation, we
deduce that it is possible to map a pair of nonintersecting circles ω1 and ω2 by an
inversion into a pair of concentric circles if and only if all circles orthogonal to both
ω1 and ω2 have two common points (in which case one of the points is the center of
inversion). A configuration is shown in Fig. 3.18.
We will thus prove that for every two nonintersecting circles, ω1 of center O1 and
ω2 of center O2 , the circles that are orthogonal to both have two common points.
Any circle that is orthogonal to both ω1 and ω2 must intersect the line O1 O2 at two
points, and for such a circle ω, whose center we denote by O, let P and Q be its
intersections with O1 O2 (Fig. 3.19). Then OP 2 is the power of O with respect to
both circles. This is because the radius of ω at the point where it intersects ω1 is
tangent to ω1 (by the orthogonality condition), and the same is true for ω2 . Hence,
O is on the radical axis of ω1 and ω2 . This radical axis is therefore OM, where
M is the projection of O onto O1 O2 . So the projection M of O on O1 O2 does not
Fig. 3.18 The circles that are

orthogonal to two
nonintersecting circles have
two common points
ω1
ω2
156 3 Inversions
Fig. 3.19 Finding the center

of inversion that makes two
circle concentric
P M Q
ω1
ω2
O
depend on ω, only on the original circles, because it is where the radical axis of the
two circles intersects the line of centers.
The power of O with respect to ω1 and ω2 , being equal to OP 2 , is also the power
of O with respect to the degenerated circle of center P and radius 0. So OM is in
fact the radical axis of the triple of circles ω1 , ω2 , and P . This means that MP 2 is
the power of M with respect to both ω1 and ω2 , and the same is true if we replace
P by Q. Hence, the pair {P , Q} is uniquely determined by ω1 and ω2 , and so all
orthogonal circles pass through both these points, and the theorem is proved.

Proof 2 Let us see the proof by complex numbers. Instead of varying the center
of inversion, we keep it fixed at the origin 0 of the coordinate system and vary the
centers of the circles on the x-axis while keeping the distance between their centers
fixed. Let the distance between the centers be d, and let the first circle have center x
and radius r1 and the second circle have center d + x and radius r2 . Let the radius
of inversion be R. Then the inversion is the map z → R 2 /z̄. The two circles have
the equations
|z − x| = r1 and |z − (x + d)| = r2
and, as seen in the proof of Theorem 3.5, their inverses have the equations
2 2
R 2 x R 2 r1 R 2 (x + d) R 2 r2

w − 2 = and w − = .
x − r12 x 2 − r12 (x + d)2 − r22 (x + d)2 − r22
The two centers coincide if
R2x R 2 (x + d)
= .
x 2 − r1
2 (d + x)2 − r22
This yields the quadratic equation in the real variable x
dx 2 − (d 2 + r12 − r22 )x + dr12 = 0.

The discriminant of this equation is
(d 2 + r12 − r22 ) − 4d 2 r12 = (d 2 + r12 − r22 + 2dr1 )(d 2 + r12 − r22 − 2dr1 )
= [(d + r1 )2 − r22 ][(d − r1 )2 − r22 ]
= (d + r1 + r2 )(d + r1 − r2 )(d − r1 + r2 )(d − r1 − r2 )
= (d + r1 + r2 )[d 2 − |r1 − r2 |2 ](d − r1 − r2 ).
The circles do not intersect if either d > r1 + r2 or d < |r1 − r2 |. In the first case,
all factors are positive; in the second case, the last two factors are negative. So if
the circles do not intersect, the equation has exactly two solutions. As the center of
inversion must be on the line of circles, this algebraic computation shows that there
are exactly two centers of inversion that map the circles into concentric circles. Any
radius of inversion would work. This concludes the analytic proof.

Definition The two points that are the centers of the inversions that map the two
nonintersecting circles to concentric circles are called the limiting points of the two
circles.
This theorem has an important corollary. Given two nonintersecting circles ω1
and ω2 , a Steiner chain is a sequence of circles tangent to both ω1 and ω2 , and such
that each circle is also tangent to the previous one. If the sequence is finite, and
the first and the last circles are also tangent, the Steiner chain is called closed. An
example is shown in Fig. 3.20.
Theorem 3.28 (Steiner’s Porism) If for two nonintersecting circles ω1 and ω2 a
closed Steiner chain of n circles exists, then any circle that is tangent to both ω1 and
ω2 is part of such a closed Steiner chain of n circles.
Proof We can map ω1 and ω2 by an inversion to concentric circles, without
changing the conclusion. But now the statement is trivial, since the closed Steiner
chains are mapped into one another by rotations about the common center of ω1 and
ω2 .
Fig. 3.20 Example of a

closed Steiner chain
158 3 Inversions
3.1.10 Problems with Theoretical Flavor About Properties of

Inversion and Möbius Transformations
The understanding of inversion, and more generally of Möbius and circular

transformations, can be enhanced by solving the following problems.
157 (a) What is the composition of two inversions with the same center?
(b) What is the composition of an inversion and a homothety with the same center?
(c) What is the composition of a homothety and an inversion with the same center?
158 Show that two points are reflections of each other over a circle if and only if
the circle is orthogonal to every circle passing through the two points.
159 Let A, B, P be distinct points in the plane such that P is not the midpoint of
the segment AB. Show that there is an inversion that maps A, B, P to A , B , P ,
respectively, such that P is the midpoint of A B .
160 Let AB be a chord of a circle ω, and let χ be an inversion. Set ω = χ (ω),
A = χ (A), and B = χ (B). Show that AB and A B intersect on the radical axis
of ω and ω .
161 Show that if two circles are tangent to each other and orthogonal to a third
circle, ω, then their tangency point lies on ω.
162 Given the pairwise exterior tangent circles ω1 , ω2 , and ω3 , show that one and
only one of the following situations occurs:
(i) there is a unique circle which is exterior tangent to all three circles and a unique
circle to which the three circles are interior tangent;
(ii) ω1 , ω2 , and ω3 have a common tangent and there is a unique circle that is
tangent to all three circles, this circle being exterior tangent to them;
(iii) there are exactly two circles that are exterior tangent to the three circles and no
circle is interior tangent.
163 Let A, B, C, D be four points, no three of them collinear. Prove that the angle
between the circumcircles of ABC and ABD is equal to the angle between the
circumcircles of ACD and BCD.
164 Prove that if a circle is orthogonal to two circles ω1 and ω2 , then it is
orthogonal to the circle of the inversion that maps ω1 to ω2 (or to the line over
which ω1 reflects to ω2 if the circles have equal radii).
165 (Apollonian Circles) Given two points A and B and a positive number k, let
k (A, B) be the locus of points P in the plane such that P A/P B = k.
(a) Show that k (A, B) is a line for k = 1 and a circle for k = 1.
(b) Show that A and B are reflections of each other over k (A, B).
(c) Conversely, show that if A and B are reflections of each other over the circle (or
line) , then there is k > 0 such that = k (A, B).
3.2 Inversion in Euclidean Geometry Problems 159
166 Find all Möbius transformations that map the upper half-plane {z | Im z > 0}
to itself.
167 Prove that every transformation of C∪{∞} that maps every line and circle to a
line or a circle is a circular transformation. (It is understood that every line contains
{∞}).
168 To a polynomial P (x) = ax 3 + bx 2 + cx + d with real coefficients, of degree
at most 3, one can apply the following two operations: (i) swap simultaneously a
and d, respectively, b and c, and (ii) translate the variable x to x + t, where t is a
real number.
(a) Can one transform, by a successive application of these operations, the polyno-
mial P1 (x) = x 3 + x 2 − 2x into the polynomial P2 (x) = x 3 − 3x − 2?
(b) What if we replace P2 (x) by x 3 − 3x − 3?
(c) Allow a third operation in which all coefficients of a polynomial can be
multiplied simultaneously by the same nonzero real number, and replace P2 (x)
by x 3 − 3x − 1. Can we now transform P1 (x) into this polynomial?
3.2 Inversion in Euclidean Geometry Problems
3.2.1 Applications of Inversion to Proving Classical Results
We present one important construction and two groupings of theorems, as excur-

sions through old-fashioned geometry, guided by inversion. While the results do
have well-known proofs by other means, we think that the inversive proofs shown
here can deepen their understanding.
I. Pole and Polar with Respect to a Circle
In this section we need a result which, by itself, does not involve inversion.
Lemma 3.29 (Poncelet) Let be a circle of center O, and let M be a point that
does not belong to the circle. Through M, we take a variable line that intersects the
circle at M1 and M2 , and let P be the harmonic conjugate of M with respect to M1
and M2 (meaning that the points M, P , M1 , M2 form a harmonic division). Then
the locus of P is
(i) a line pM perpendicular to OM if M lies inside ;
(ii) the segment that lies inside of a line pM that is perpendicular to OM.
Proof The two situations are shown in Fig. 3.21. Let N be the midpoint of the
segment MP . Then the relation that expresses the fact that M, P , M1 , M2 form a
harmonic division
MM1 P M1
=
MM2 P M2
160 3 Inversions
M2 P
P
N
M1 N M1
M O
M O
pM
M2
Fig. 3.21 Proof of Poncelet’s lemma
can be rewritten as
MN ± NM1 P N ∓ NM1
= ,
NM2 ± MN NM2 ∓ P N
with the signs depending on whether M is outside or inside the circle .

Using the fact that MN = P N, we obtain NM 2 = NM1 · NM2 , showing that
N is on the radical axis of the circle and the point M (the point is a degenerate
circle). But P is the image of N under a homothety of center M and ratio 2, so
the locus is the image pM of this radical axis, or more precisely the intersection of
pM with the interior of the circle if M is exterior to , and the entire line if M is
exterior. And pM is orthogonal to OM, as the radical axis is.

Definition The line pM is called the polar of M with respect to , and M is called
the pole of pM with respect to . If M is on the circle, its polar with respect to the
circle is the tangent at M.
It is not hard to see that given a circle every line has a pole and every point has
a polar with respect to the circle. It is at this moment that we exhibit the connection
with inversion, with the following straightforward corollary to Proposition 3.1.
Proposition 3.30 The polar of a point M with respect to a circle of center O is
the line perpendicular to OM passing through the inverse of M with respect to .
From here we deduce immediately that the polar of a point that lies exterior to a
circle passes through the tangency points of the tangents from the point to the circle.
Here is another consequence.
Theorem 3.31 (La Hire) Considering polars with respect to a circle, A ∈ pB if
and only if B ∈ pA .
Proof The proof can be followed on Fig. 3.22. Let A , B be the inverses of A and
B with respect to , respectively. Then the similarity of the triangles OAB and
OB A implies that the quadrilateral AA B B is cyclic. Then AB B = BA A, so
Fig. 3.22 Proof of La Hire’s

pA
theorem
O A
A
pB
B
the fact that one of these angles is right is equivalent to the fact that the other is right.
But AA B = 90◦ means that B ∈ pA and BB A = 90◦ means that A ∈ pB . The
theorem is proved.

Proposition 3.32 Three points are collinear if and only if their polars with respect
to a circle are concurrent.
Proof Using La Hire’s Theorem, we deduce that the polars of three collinear points
pass through the pole of the line determined by the points. Conversely, if three lines
are concurrent, the polar of their intersection passes through their poles.

Theorem 3.33 Let A, B, C, D be four concyclic points, and let M = AC ∩ BD,
N = AB ∩ CD, and P = AD ∩ BC. Then pM = NP , pN = MP , and pP =
MN , the polars being considered with respect to the circle passing through the four
points.
Proof 1 To prove that pM = NP amounts to show that N ∈ pM and P ∈ pM .
Moreover, showing that N ∈ pM is the same as showing that NM is perpendicular
to OM, where M is the inverse of M with respect to the circle, and O is the center
of the circle.
Consider a system of coordinates with O at the origin so that the circle passing
through A, B, C, D is the unit circle. The coordinates of these four points satisfy
aa = bb = cc = dd = 1.
In complex coordinates, the point of intersection of the lines passing through the
pairs of points (z1 , z2 ) and (z3 , z4 ) has the coordinate
((z2 − z1 )z1 − (z2 − z1 )z1 )(z4 − z3 ) − ((z4 − z3 )z3 − (z4 − z3 )z3 )(z2 − z1 )
.
(z4 − z3 )(z2 − z1 ) − (z2 − z1 )(z4 − z3 )
Using the fact that a, b, c, d have absolute value 1, we can simplify the formulas for
m and n as
162 3 Inversions
a
(ba − ba)(d − c) − (dc − dc)(b − a) − a (d − c)
b
− c − c (b − a)
d
m= = b d
(d − c)(b − a) − (b − a)(d − c) (d − c) b1 − a1 − d1 − 1c (b − a)
(a − b)(c − d)(c + d − a − b) abcd

1
c+d −a−b
= = ,
(a − b)(c − d)(ab − cd) abcd
1 ab − cd
and
b+d −a−c
n= .
ac − bd
The inverse of m with respect to the unit circle is
1 ab − cd
m = = .
m c+d −a−b
To show that N M is perpendicular to OM is equivalent to showing that the ratio of

n − m and m is imaginary, which is the same as showing that the product of n − m
and m is imaginary (because (n − m )/m = (n − m )m/(mm), whose denominator
is real). We compute
1 b+d −a−c c+d −a−b

(n − m )m = nm − ·m= · −1
m ac − bd ab − cd
bc + bd − ba + dc − da − db − ac − ad + ab − cd + ac + cb
= − 1,
cb − ad − da + bc
where we have used the fact that a, b, c, d have absolute value 1. Now bring to the
common denominator, reduce terms, and rearrange to obtain
bd − bd + ab − ab + dc − dc + ca − ca
.
cb + cb − ad − da
The numerator is imaginary because it is a sum of numbers of the form z − z, while

the denominator is real being the sum of numbers of the form z + z. Hence, the
quotient is imaginary, and we are done.

Proof 2 There is a quick proof that has homothety at its heart. Let us assume that
we are in the configuration from Fig. 3.23, and let us show that MP is the polar of
N (the other two cases are similar). Denote by R and Q the intersections of P M
with AB and CD, respectively. Using again complex coordinates, the theorems of
Menelaus (Theorem 2.15) and Ceva (Problem 116) imply that
n−b q −b
=− ,
n−a q −a
Fig. 3.23 MP is the polar of N

N
A R
M
Q
C
P B
which in the context of those theorems means that the direct homothety of center N
that maps A to B and the inverse homothety of center Q that maps A to B have the
same ratio in absolute value. This means that N, A, Q, B form a harmonic division,
and the same is true for N, D, R, C. By definition, QR is the polar of N with respect
to the circle, and we are done.

The second proof shows that N, A, Q, B form a harmonic division on AB
regardless of whether the quadrilateral is cyclic or not. It is worth mentioning that a
more appropriate placement of the notion of pole and polar, as well as of the above
theorem, is within the field of real two-dimensional projective geometry. But that
lies outside of our discussion.
II. The Theorems of Simson-Wallace and Salmon
Let us turn our attention to some classical theorems, which we place in the context
of inversion.
Theorem 3.34 (Simson-Wallace) Let ABC be a triangle and let M be an arbi-
trary point in the plane. Let also N, P , Q be the projections of M on the lines BC,
AC, and AB, respectively. Then the points N, P , Q are collinear if and only if M
is on the circumcircle of ABC.
Proof In the proof, which can be followed on Fig. 3.24, we use the fact that
inversion preserves the angles but changes their orientation. In this figure, on the
left is the original configuration, and on the right is the inverted one. To avoid
case dependency, everywhere in the proof we will be working with directed angles
modulo 180◦ .
Consider an inversion of center M and arbitrary radius, and use the standard
notation for the inverse of a point. We have
N C P = N C M + MC P = − CN M − MP C = 0,
which shows that the points C , N , P are collinear. Similarly the points A , P , Q
are collinear. The fact that N, P , Q is collinear is equivalent to NP Q = 0. We
164 3 Inversions
Q
M Q
A
P
A M
B C P
N N C
Fig. 3.24 Proof of the Simson-Wallace theorem
have the following sequence of equivalences:
NP Q = 0 ⇐⇒ MP N = MP Q ⇐⇒ MN P = MQ P
⇐⇒ (MN , MC ) + (MC , C P ) = (MQ , MA ) + (MA , A P )
⇐⇒ N MC = Q MA ⇐⇒ NMC = QMA
⇐⇒ (MN, BC) + (BC, MC) = (MQ, AB) + (AB, MA)
⇐⇒ BCM = BAM.
The last condition is equivalent to the fact that A, B, C, M are concyclic, as these
points are not collinear.

If M is on the circumcircle of ABC, so that the points N, P , Q are collinear, then
the line they determine is called the Simson line corresponding to M with respect to
the triangle ABC.
This proof is more complicated than the standard proof based on cyclic quadrilat-
erals and angle chasing. But, by examining the inverted figure, we are led naturally
to another theorem. In the inverted figure, N , P , and Q lie diametrically opposite
to M in the circles that are the images of BC, AC, and AB, respectively. This is
because MN , MP , and MQ are orthogonal to the respective sides, and so MN ,
MP , and MQ are orthogonal to the respective circles, and hence are diameters in
these circles.
Examining thus the figure, we observe the four endpoints of the diameters MN ,
MP , and MQ of some three circles that lie on a circle themselves, and also we
notice that the pairwise intersection points of these three circles are collinear. This
brings us to the next theorem.
Theorem 3.35 (Salmon) Let ABCD be a cyclic quadrilateral, and let M, N, P be
the points where the circles of diameters AB and AC, AB and AD, and AC and
AD intersect the second time, respectively. Then M, N, P are collinear.
Fig. 3.25 Proof of Salmon’s M

theorem
B N
M
B
C
N
C
A
P P
Proof Of course this is the inverted figure of the Simson-Wallace Theorem, and so
the result should follow immediately by inversion from that theorem. Arguing on
Fig. 3.25, take an inversion of center A and arbitrary radius. Then the circumcircle
of ABCD transforms into the line B C D , and the circles of diameters AB, AC,
and AD transform into the lines M N , M P , and N P , respectively, and these
lines are orthogonal to the corresponding diameters. The perpendiculars from A to
the sides of the triangles M N P , which are the points B , C , D , are collinear. By
the Simson-Wallace Theorem, A is on the circumcircle of triangle M N P , which
implies that M, N, P are collinear, and we are done.
III. The Theorems of Euler, Tzitzeica, and Poncelet

We commence our next essay with Euler’s relation.
Theorem 3.36 (Euler’s Relation) Let R, r, O, and I be the circumradius, the
inradius, the circumcenter, and the incenter of a triangle. Then
OI 2 = R 2 − 2rR.
Proof In the triangle ABC, let D, E, F be the tangency points of the incircle and
the sides BC, CA, AB, respectively (see Fig. 3.26). Consider the inversion of center
I and radius r. We use the convention that the inverse of a point X is denoted by X .
In the right triangle EAI , we have
I A · I A = r 2 = I E 2 ,
166 3 Inversions
Fig. 3.26 Proof of Euler’s A

relation
A E
F
I
C
B
B D C
which, by the Leg Theorem, implies that A is the foot of the altitude from E in this
triangle. Thus A is the midpoint of EF . Similarly B is the midpoint of F D, and
C is the midpoint of ED. From here we deduce that the inverse of the circumcircle
of ABC is the nine-point circle of DEF . The radius of the nine-point circle is r/2,
namely, half of the circumradius. Using Proposition 3.6, we obtain
r r2
= R,
2 ρ
where ρ is the power of I with respect to the circumcircle of ABC. By computing

the power of I using the diameter of the circumcircle that passes through it, we
obtain ρ = (R + OI )(R − OI ) = R 2 − OI 2 . Therefore
r r2
= 2 R,
2 R − OI 2
from which we derive R 2 − OI 2 = 2rR, and finally Euler’s relation: OI 2 = R 2 −

2rR.

Here is one of the many corollaries of this result.
Proposition 3.37 Let ABC be a triangle, let I be its incenter, and let A , B , C be
the points where the angle bisectors of the triangle intersect again the circumcircle.
Then the circumcircles of the triangles I A B , I B C , and I A C have the same
radius as the circumcircle of the triangle ABC.
Proof We consider the circular transformation that is the composition
√ of the
reflection over I and the inversion with center I and radius ρ = R 2 − OI 2 , where
O is the circumcenter of ABC. (Sometimes this transformation is referred to as an
inversion of center I and negative ratio OI 2 − R 2 .) A point P on the circumcircle

of ABC is mapped to the point P in such a way that I is between P and P and
I P · I P = R 2 − OI 2 .
And this quantity is the power of I with respect of the circumcircle (see the proof
of the previous result). Consequently P is on the circumcircle, and so the circular
transformation maps the circumcircle of ABC into itself. The points A , B , C are
therefore the images of A, B, C, respectively, so the notation is appropriate!
We now follow the argument on Fig. 3.27. As a corollary of what we have just
shown is the fact that the circumcircles of I A B , I B C , I C A are the images of
the lines AB, BC, AC, respectively. The lines AB, BC, AC are at distance r from
the center of inversion, so the diameters of the circles that are their images are given
by the formula
ρ2 R 2 − OI 2 R 2 − R 2 + 2Rr
d= = = = 2R.
r r r
The last expression is the diameter of the circumcircle of triangle ABC, and the
proposition is proved.

Remark We have solved this problem by means of a special circular transformation.

This transformation is of interest in itself, so let us examine it in more detail. Let ω
be a circle and let O be a point that does not belong to ω. Using O we can define a
circular transformation that maps ω to itself:
• If O is outside the circle, then the transformation is the inversion with center O
and radius equal to the square root of the power of O with respect to ω.
• If O is inside the circle, then the transformation is the composition of the
reflection over O with the inversion of center O whose radius is the square root
Fig. 3.27 Proof of the D

equality of the three circles
B
C
B D C
A
168 3 Inversions
of the power of O with respect to ω (which therefore is an inversion with negative

ratio).
We will use this circular transformation to prove the two results below.
Theorem 3.38 (Tzitzeica’s Five-Lei Coin Problem) Three circles of equal radii
have a common point and intersect pairwise at three distinct points. Then the circle
passing through these three points is equal to any of the three circles.
Proof 1 Let I be the intersection of the three circles, and let A , B , C be the
pairwise intersection points. The reason for the notation is that if we consider the
circular transformation φ that is obtained by composing the reflection over I with
the inversion of center I and radius the square root of the power of I with respect
to the circumcircle of A B C , we arrive at the configuration of Proposition 3.37,
which can be seen by examining Fig. 3.28.
Indeed, the respective images A, B, C of A , B , C are on the circumcircle of
A B C . Also, the images of the three equal circles are the lines AB, BC, and AC,

and because the circles are equal, these lines are at equal distance from I . Hence I
is the incenter of ABC, and by Proposition 3.37, the circumcircle of ABC (which
is the same as the circumcircle of A B C ) is equal to any of the three circles.

Proof 2 With the notation from the first proof, consider an inversion of center I ,
and let A1 , B1 , and C1 be the images of A , B , C , respectively. Because the circles
have equal radii and pass through the center of inversion, their images, which are
the lines A1 B1 , B1 C1 , C1 A1 , are at equal distance from I , so I is the incenter of
the triangle A1 B1 C1 . Let us focus on this triangle and study Fig. 3.29. An easy
computation yields
B1 I C1 = 180◦ − A1 B1 C1 /2 − A1 C1 B1 /2 = 90◦ + B1 A1 C1 ,
and for a similar reason
Fig. 3.28 First proof of

Tzitzeica’s theorem
B
C
B C
A
Fig. 3.29 Second proof of A1

Tzitzeica’s theorem
B1 C1
A1 I C1 = 180◦ − B1 A1 C1 /2 − A1 C1 B1 /2 = 90◦ + A1 B1 C1 /2.
Let O be the circumcenter of B1 I C1 . Then, in the circumcircle of this triangle,
B1 OC1 = 360◦ − 2 B1 I C1 = 180◦ − B1 A1 C1 ,
and hence in the isosceles triangle B1 OC1 , OC1 B1 = B1 A1 C1 /2. Then in the
isosceles triangle OI C1
OI C1 = OC1 I = OC1 B1 + B1 C1 I = B1 A1 C1 /2 + A1 C1 B1 /2
= 180◦ − A1 I C1 .
It follows that A1 , I, O are collinear, so A1 I passes through the circumcenter

of B1 I C1 . From here we deduce that A1 I and the circumcircle of B1 I C1 are
orthogonal. But then, so are their preimages, which are the lines A I and B C .
Thus, A I is an altitude in the triangle A B C , and for a similar reason, so are B I
and C I . Then I is the orthocenter of A B C , and in this situation, the circumcircles
of A I B , B I C , and C I A are equal to the circumcircle of the triangle A B C .
The theorem is proved.

The name of this result comes from the fact that the Romanian mathematician
George Tzitzeica (also spelled Gheorghe Ţiţeica) discovered it by playing with a 5-
lei coin (leu, plural lei, being the currency of Romania). The first proof demonstrates
that this theorem is a direct consequence of Euler’s relation. We suggest that you try
to work backward from Tzitzeica’s Theorem via Proposition 3.37 to prove Euler’s
relation.
Tzitzeica’s Theorem can be proved easily using complex numbers, albeit without
inversion. Here is how. Place the origin of the coordinate system at I , and let
w1 , w2 , w3 be the coordinates of the centers O1 , O2 , O3 , of the three circles,
respectively; |w1 | = |w2 | = |w3 | because O1 , O2 , O3 are at equal distance from
170 3 Inversions
I . Then I is the circumcenter of O1 O2 O3 , and the circumradius of O1 O2 O3 is the

same as the radius of the three circles.
Because A, B, C are the reflections of I over the lines that connect the centers
of the circles, they have coordinates w1 + w2 , w1 + w3 , and w2 + w3 , so the side
lengths of the triangle ABC are |w1 + w2 − w1 − w3 | = |w2 − w3 |, |w1 − w3 |,
and |w2 − w3 |, which are the side lengths of O1 O2 O2 . Thus the two triangles are
congruent, and they have equal circumradii.
Let us build on what we have proved so far and apply the circular transformation
φ once more to prove another classical result in elementary geometry.
Theorem 3.39 (Poncelet’s Closure Theorem) Let and ω be two circles such
that ω is inside . If it is possible to find one triangle that is simultaneously
inscribed in and circumscribed to ω, then it is possible to find infinitely many
such triangles, and each point of or ω is a vertex or tangency point, respectively,
of one such triangle.
Proof Let ABC be the triangle inscribed in and circumscribed to ω, and let I
be the center of ω. We will show that by starting with any point A1 on , we can
construct a triangle A1 B1 C1 inscribed in and circumscribed to ω. The proof can
be followed on Fig. 3.30.
We start by choosing A1 arbitrarily on and constructing the tangents A1 B1
and A1 C1 to ω, with B1 , C1 ∈ . We then consider the circular transformation φ
obtained by composing the reflection over I with an inversion of center I that keeps
invariant, and let A1 = φ(A1 ), B1 = φ(B1 ), and C1 = φ(C1 ).
By applying Proposition 3.37 to the triangle ABC, we deduce that the images
of the sides are three equal circles through I that are tangent to the circle of center
I and have radii equal to the diameter of . Now take another line tangent to
ω. Because is the rotate about I of one of the sides of ABC, and because φ has
circular symmetry, being the composition of a reflection over I with an inversion
centered at I , it follows that φ() is a circle through I of diameter equal to the
Fig. 3.30 Proof of Poncelet’s

closure theorem
A1
A
B1
C1
B
I
B1 C1
C
A1
diameter of . We have therefore shown that φ maps lines tangent to ω to circles

through I that are equal to , and vice versa.
Then the circumcircles 1 and 2 of I A1 B1 and I A1 C1 are the images of A1 B1
and A1 C1 , respectively, and hence are equal to . We can apply Tzitzeica’s Theorem
to the circles , 1 , and 2 , which pass through A1 and intersect pairwise at I , B1 ,
and C1 , to deduce that the circumcircle 3 of I B1 C1 is equal to the three circles ,
1 and 2 . Hence, the line B1 C1 = φ(3 ) is tangent to ω. We conclude that the
triangle A1 B1 C1 is also inscribed in and circumscribed to ω, and the theorem is
proved.

This result is also known as Poncelet’s porism. It has the following general form.
Let 1 and 2 be two plane conics. If it is possible to find, for a given n > 2,
an n-sided polygon that is simultaneously inscribed in 1 (meaning that all of its
vertices lie on 1 ) and circumscribed to 2 (meaning that all of its edges are tangent
to 2 ), then it is possible to find infinitely many such n-sided polygons, and each
point of 1 or 2 is a vertex or tangency point, respectively, of such a polygon.
We will not prove this general result in this book, as its proof uses advanced
tools of algebraic geometry. One should note the similarity between the Poncelet
and Steiner porisms.
√
IV. The “ bc Inversion” and the Mixtilinear Incircles
In this section we introduce a special Möbius transformation associated to a triangle
and a vertex and present a construction motivated by this transformation. To define
the transformation, which we denote by φA , we let the triangle be ABC, with
BC = a, AC = b, and AB = c, and let the chosen √ vertex be A. Then φA is
the composition of the inversion of center A and radius bc with the reflection over
the internal bisector of BAC. This is indeed a Möbius transformation because it is
the composition of two reflections,
√ one over a circle the other over a line. By abuse
of language, it is called the “ bc inversion.” As we will see in what follows, this
Möbius transformation combining the reflections over a circle and a line passing
through its center is a mighty tool for solving problems.
√ origin and such that B = ce

Considering a system of coordinates with A at the iθ
and C = be −iθ (so that AB = c and AC = b), the bc inversion is given by
bc
φA : C ∪ {∞} → C ∪ {∞}, φA (z) = .
z
Proposition 3.40 The transformation φA has the following properties:

(i) φA (A) = ∞, φA (B) = C, and φA (C) = B;
(ii) the circumcircle of ABC and the line BC are mapped into one another, and

consequently the arc BC and the segment BC are mapped into each other;
(iii) if P is a point that is not on AB or AC, then the triangle ABP is similar to
the triangle AP C, and consequently P = φA (P ) is the image of B under the
spiral similarity of center A that maps P to C;
172 3 Inversions
(iv) if P and Q are two points such that A, P , Q are not collinear, then the
triangles AP Q and AQ P are similar, where P = φA (P ) and Q = φA (Q);
(v) the image of the circumcenter O of ABC through φA is the reflection of A over
BC, and the image of the orthocenter H of ABC is the second intersection
point of AO with the circumcircle of BOC.
Proof Property (i) follows from the fact that AB · AC equals the square of the
radius of inversion. For (ii) note that φA (BC) is a circle that passes through A, and
by (i) it also passes through B and C. Parts (iii) and (iv) are direct consequences
of Theorem 3.2 (see Fig. 3.31). In fact (iii) can be checked quickly with complex
coordinates, if we place A at the origin, with B = ceiθ , C = be−iθ , and b, c > 0. If
P has coordinate z, then φA (z) = bc z , and
c c z
= = ,
φA (z) bc b
z
showing that triangles ABP and AP C are directly similar, and hence are mapped
into each other by a spiral similarity.
For (v), note that O is the reflection of ∞ over the circumcircle, so, by the
Symmetry Principle (Theorem 3.15), O = φA (O) is the reflection of A = φA (∞)
over the line BC which is the image of the circumcircle. H is at the intersection
of AO and the circumcircle of BO C (because this circle is the reflection of the
circumcircle of ABC over BC), so H = φA (H ) is at the intersection of the image
of AO , which is AO, and the image of the circumcircle of BO C, which is the
circumcircle of BOC.
Fig. 3.31 The

√ image of P A
through the bc inversion
B C
P
√
What happens if we compose two bc inversions of the same triangle? The result
is surprising.
√
Proposition 3.41 Let φB and φC be the bc inversions determined by triangle
ABC in vertices B and C, respectively. Then φB ◦ φC = φA .
Proof Let φ = φB ◦ φC . This is a Möbius transformation, so we have to find the
images of three points. The natural choices are A, B, and C. Indeed
φ(A) = φB (φC (A)) = φB (B) = ∞, φ(B) = φB (φC (B)) = φB (A) = C,

φ(C) = φB (φC (C)) = φB (∞) = B.
These coincide with φA (A), φA (B), and φA (C), respectively, hence the conclusion.

√
The bc inversion allows for the following slick proof of
Proposition 1.23 The point where the symmedian of a triangle intersects the
circumcircle forms with the vertices a harmonic quadrilateral.
Proof Let the triangle
√ be ABC, let P be the midpoint of BC, and let P be its
image through the bc inversion corresponding to the vertex A in the triangle ABC.
Denote
√ the complex coordinates of points by the same letter, but lowercase. Because
the bc inversion is a Möbius transformation, it preserves the cross-ratio. As A is
mapped to ∞ and B is mapped to C, we have
(a, p , b, c) = (∞, p, c, b).
The fact that ABP C is harmonic is equivalent to (a, p , b, c) = −1, and the fact
that (∞, p, c, b) = −1 is equivalent to the fact that P is the midpoint of BC. The
proposition is proved.

Inversion and Möbius transformations become more interesting when they
involve circles and tangencies,
√ so let us look at incircles and excircles. To reca-
pitulate: if φA is the bc inversion determined by ABC at the vertex A, the side
AB is mapped to the ray r of the line AC that starts at C and does not contain A, the
side AC is mapped to a similar ray s of the line AB, and the side BC is mapped to

the arc BC of the circumcircle of ABC that does not contain A. Then the incircle ω
is mapped to a circle that is tangent to another circle that is tangent to the rays r and
s and the circumcircle of ABC. And this is the A-mixtilinear excircle, illustrated in
Fig. 3.32.
If we switch to the A-excircle ωA , then φA (ωA ) is a circle that is tangent to sides
AB and AC and to the circumcircle internally. This is the A-mixtilinear incircle
(see Fig. 3.33).
Let us first find a suitable way to construct the A-mixtilinear incircle A with
ruler and compass. One possible way to do that is to locate the tangency points of
A to AB, AC, and the circumcircle of ABC.
174 3 Inversions
Ma
A
I
ω
B C
Ta
φA (ω)
Ia
Fig. 3.32 The A-mixtilinear excircle
Proposition 3.42 (Finding Tangency Points, Part I) Let A be the A-mixtilinear

incircle of triangle ABC. Let be the circumcircle of ABC, and let I be the
incenter of this triangle. Let A touch AB at D, AC at E, and at Ua . Let also
Ya be the tangency point of the A-excircle to BC. Then the lines AYa and AUa are
isogonal and I is the midpoint of DE.
√
Proof You can follow our reasoning on Fig. 3.33. Let φA be the bc inversion
determined by the triangle ABC and the vertex A. The first part is immediate from
the fact that φA (Ya ) = Ua and the fact that φA inverts about a circle centered at A
and then reflects across the bisector of BAC.
The second part is more interesting: Proposition 3.7 implies that the A-excenter
Ia of ωA is mapped to the midpoint of the segment determined by the points of
tangency of the tangents to φA (ωA ) = A through A, that is, Ia is mapped to the
midpoint of DE. It remains to show that φA (Ia ) = I , which is true because
1 1
ACIa = ACB + (180◦ − ACB) = 90◦ + ACB = AI B,
2 2
which implies that the triangles AI B and ACIa are similar. And then we apply
Proposition 3.40, part (iii).

Ma
A
D I
B C
Ya
Ua
ωA La
Ia
Fig. 3.33 The A-mixtilinear incircle
It should be mentioned that proving that I is the midpoint of the segment DE

has made the object of a problem given at the International Mathematical Olympiad
in 1978. It is also true that each excenter is the midpoint of the segment determined
by the tangency points of the corresponding mixtilinear excircle.
We can now see how to construct A : find the incenter I (which can be done
by drawing the bisector of BAC and then drawing the circle with center at the

midpoint La of the arc BC of that does not contain A) and then draw the
perpendicular to AI through I to obtain the points D and E. Unfortunately, locating
Ua requires too much work, because we need to find the A-excircle, or, at least,
Ya . When we are using pencil and paper, this leads to loss of precision in the
construction. But the following result makes the construction of Ua easy and precise.
Proposition 3.43 (Finding Tangency Points, Part 2) Let A be the A-mixtilinear
incircle of triangle ABC. Let be the circumcircle of ABC, and let I be the

incenter. Let A touch at Ua , and let Ma be the midpoint of BC that contains
A. Then Ua lies on the line I Ma .
176 3 Inversions
Proof Continue looking at Fig. 3.33. The proof is based again on Proposition 3.40,
part (iii), but this time by taking into account that φ(Ya ) = Ua and φ(I ) = Ia , there
is a spiral similarity that takes Ya to I and Ia to Ua , so AUa I = AIa Ya .

As above, let La be the midpoint of the BC that does not contain A. Since Ia Ya
and La Ma are perpendicular to BC, they are parallel. Then
AIa Ya = ALa Ma = AUa Ma ,
and so AUa I = AUa Ma . The conclusion follows.

Now the problem of constructing A is simple: Once you have found D and E as
above, draw the line Ma I to find Ua on . Then A is the circumcircle of DEUa .
Mixtilinear circles bring forth countless little facts. Here are two of them.
Proposition 3.44 (Arc Midpoints) Let A be the A-mixtilinear incircle of the

triangle ABC, and let Lb and Lc be, respectively, the midpoints of the arcs CA

and AB that do not contain the other vertex of ABC. As before, let A touch the
line AB at D, the line AC at E, and the circumcircle of ABC at Ua . Then the
line Ua D passes through Lc and (analogously) the line Ua E passes through Lb .
Moreover, DE is parallel to Lb Lc .
Proof We argue on Fig. 3.34. The homothety that maps A to , with center Ua ,
maps AB to a line that is tangent to and parallel to AB. This line is tangent to at
Lc . So D is mapped to Lc , and Ua , D, and Lc are collinear. Similarly, E is mapped
to Lb , and so the line DE is mapped to the line Lc Lb . Hence, DE Lb Lc .

Proposition 3.44 can also be used to obtain another proof of the fact that I is the
midpoint of DE: by Pascal’s Theorem applied to the hexagon Ua Lc CABLb , the
Fig. 3.34 Midpoints of arcs A

and a mixtilinear circle
Lb
Lc
D
ΩA
B C
Ua Ω
points Ua Lc ∩ AB = {D}, Lc C ∩ BLb = {I }, and Lc C ∩ Lb Ua = {E} are collinear,

and since AI is the angle bisector of DAE in the isosceles triangle ADE, I is the
midpoint of DE.
It is natural to interpret the lines AUa , BUb , and CUc as cevians. The symmetry
of the configuration makes us expect that these lines are concurrent. This is the case,
and the common point is actually special!
Proposition 3.45 (Concurrent Cevians) Let A be the A-mixtilinear incircle
of the triangle ABC. Let A touch the circumcircle of ABC at Ua , and
define Ub and Uc similarly. Then AUa passes through the center of the direct
homothety that maps the incircle of ABC to the circumcircle. Consequently, the
lines AUa , BUb , CUc are concurrent.
Proof Arguing on Fig. 3.35, let h be the (direct) homothety that takes A to , and
let h be the (direct) homothety that takes the incircle ω to A . Then h has center
Xa , h has center A, and the composition h ◦ h takes ω to , because h (h(ω)) =
h (A ) = . This composition is the direct homothety between ω and , so we are
done by Proposition 2.6. You can also use Monge’s Theorem (see Problem 86 (a))
to prove this result.

Analogous results are true for the mixtilinear excircles. The reader should
explore, discover, and prove these results.
Fig. 3.35 Lines with the A

same opinion; they concur!
Ub
Uc
I
O
B C
Ua
178 3 Inversions
3.2.2 Examples of Problems Solved Using Inversion
The first example of our discussion is a problem given at the Israel Mathematical
Olympiad in the year 1995, whose first two solutions that are presented below were
suggested to us by Leandro Maia.
Problem 3.1 Let P Q be the diameter of a semicircle . The circle ω is tangent
internally to and is also tangent to the segment P Q at C. Let A be a point on ,
and let B be a point on P Q such that AB is tangent to ω and perpendicular to P Q.
Prove that AC is the angle bisector of P AB.
Solution 1 There is an inversion present in the picture, which can be revealed by

examining closely Fig. 3.36. The Leg Theorem in the right triangle AP Q gives
QB · QP = QA2 . So the inversion χ1 of center Q and radius QA maps P and B
into each other. But there is more to it. To discover this hidden fact, note first that
χ1 ( ) is a ray that is perpendicular to P Q, and this ray is |BA. It follows that χ1 (ω)
is a circle that is tangent to = χ1 (|AB), |AB = χ1 ( ), and P Q = χ1 (P Q).
There are four such circles, but because the image of χ lies entirely in the union of
rays |QX, with X ∈ ω, it follows that χ1 (ω) must be ω itself. But then χ1 (C) = C;
this is the hidden fact that solves the problem. From here we obtain QA = QC, so
the triangle QAC is isosceles. We thus have
CAB = CAQ − BAQ = CAQ − AP B = ACQ − AP B = P AC,
where for the second equality we have used the fact that AB is altitude in the
right triangle AP Q and for the fourth equality we have used the Exterior Angle
Theorem in the triangle ABC. This proves that AC is the angle bisector of P AB,
as required.

Solution 2 An inversion χ2 of center A (Fig. 3.37) solves the problem with a
“position argument,” meaning that the conclusion is derived from the understanding
of the positions of the images of various objects with respect to one another.
The image of the circle of which is a semicircle is the line P Q , while the
image of the line P Q is the circumcircle of P AQ . And because P AQ =
P AQ = 90◦ , has diameter P Q , so its center O is on P Q .
Fig. 3.36 ω is tangent to A

AB, P Q, and
P C B Q
Q
P B
C Q
O
P
C
B
Fig. 3.37 The inversion χ2
Additionally, χ2 (ω) is a circle that is (exterior) tangent to = χ2 (P Q) and is

also tangent to AB = χ (AB) and P Q . Note that because AB is perpendicular
to P Q, the center O of χ2 (P Q) = is on AB as well (Theorem 3.4). So the
lines P Q and AB pass through the center O of , and χ2 (ω) is tangent to at
C = χ2 (C) and is also tangent to AB and P Q . By the symmetry of the figure
formed by the two lines and the two circles, C must be the midpoint of the arc P B
(B = χ2 (B)), and consequently P AC = C AB , that is, P AC = CAB, as
desired.

Solution 3 Let χ3 be an inversion of center C. As CAP = CP A and
CAB = CB A (Theorem 3.2), the line AC bisects P AB if and only if
CP A = CB A .
The inversion maps the circle ω to a line ω that is parallel to the line P Q,
while the semicircle is mapped to a semicircle that is tangent to ω and
whose endpoints are P = χ3 (P ) and Q = χ3 (Q). We want to determine the
image of the segment AB. For that, let A = χ3 (A) and B = χ3 (B). Because
AB is perpendicular to P Q, by Theorem 3.2, CA B = CBA = 90◦ , so the
segment AB is mapped into an arc on the semicircle with diameter B C. Also, as
inversion preserves tangencies, this semicircle is tangent to the line ω . So we are
in the situation from Fig. 3.38. The two semicircles in this figure are equal, and
they are mapped into each other by the reflection over the line through A that is
180 3 Inversions
Fig. 3.38 The inversion χ3 ω

yields two equal semicircles
Γ A
P C Q B
perpendicular to P Q . Consequently CP A = CB A , as they are images of

each other through the reflection.
This solution is an instance of the interplay between affine and projective geom-
etry. The statement of the problem is phrased in the language of one-dimensional
affine complex geometry, where lines are lines and circles are circles and the point
at infinity is not present. Then we place the configuration on the Riemann sphere,
by adding ∞, which then allows us to act by Möbius transformations, or circular
transformations, or just inversion. We act by such a transformation to move ∞
somewhere else and then eliminate ∞ and read the new configuration once more
in the language of Euclidean geometry. By making the right choice for where to
move the point at infinity, the problem becomes significantly simpler. Note that the
original configuration and the final one are identical on the Riemann sphere, the only
difference is that we postulate a different point at infinity for each of them. When
we remove these different points at infinity and look at the problem in the Euclidean
setting, we see different things.

Solution 4 This last solution emphasizes once more this method of “moving the
point at infinity,” but we go further, by stating the conclusion of the problem so
that it is invariant under Möbius or circular transformations, and then transform the
configuration to one where the conclusion is easy to verify. And indeed, using the
Bisector Theorem, we write the conclusion of the problem as saying that
AB CB
= .
AP CP
In complex coordinates this states that the absolute value of the cross-ratio of
A, C, B, P is equal to 1, that is, |(a, c, b, p)| = 1.
The cross-ratio is invariant under Möbius transformations, so we consider the
Möbius transformation φ = χ2 ◦ χ1 , where χ1 is the inversion of center P and
radius P A and χ2 is the inversion of center Q and radius QA. Then
φ(P ) = χ2 (χ1 (P )) = χ2 (∞) = Q and φ(A) = χ2 (χ1 (A)) = χ2 (A) = A.
The Leg Theorem applied twice in the right triangle AP Q yields P A2 = P B · P Q

and QA2 = QB · QP , so
φ(Q) = χ2 (χ1 (Q)) = χ2 (B) = P , and φ(B) = χ2 (χ1 (B)) = χ2 (Q) = ∞.

Let C = φ(C). Then, in coordinates
|(a, c, b, p)| = 1 ⇐⇒ |(a, c , ∞, q)| = 1 ⇐⇒ |(∞, q, a, c )| = 1.
We are left with showing that

∞−a q −a

∞ − c q − c = 1.
:
The first fraction is 1, so we have to check that the second fraction has absolute
value equal to 1, and geometrically this means that QA = QC .
Since φ(P ) = Q, φ(Q) = P , φ(A) = A, we have φ( ) = . Note also that
A and the reflection of A over P Q are invariant under φ and also φ(B) = ∞
which means that φ(AB) is a line and this line is AB. Thus, φ(AB) = AB and
φ(|AB) = |AB. And because φ maps the line P Q into itself, φ(ω) is tangent to ,
the ray |BA, and the line P Q. But χ1 (ω) satisfies the same tangency conditions.
Note that χ1 maps ω to the other side of , and there is only one circle on that
side of tangent to , the ray |BA, and the line P Q simultaneously, namely, the
circle denoted by ω in Fig. 3.39. And χ2 maps ω = χ1 (ω) to the same side of .
But then χ2 (ω ) = ω . In particular, χ2 (χ1 (C)) = χ1 (C), which means that C is
invariant under χ2 . This can only happen if C is on the circle of inversion, that is,
if AQ = QC . The problem is solved.

Here is a short-listed problem of the 1995 International Mathematical Olympiad,
proposed by Turkey.
Problem 3.2 The incircle of the triangle ABC touches the sides BC, CA, and AB
at D, E, and F , respectively. A point X is chosen inside the triangle ABC such that
the incircle of XBC touches BC at D as well, and let Y and Z be the points where
this incircle touches CX and BX, respectively. Prove that EF ZY is cyclic.
Solution The solution presented below was published by Dan Brânzei in Gazeta
Matematică (Mathematics Gazette, Bucharest) in 1996. It can be followed on
Fig. 3.40.
Fig. 3.39 The solution based

on an invariant
A
Γ
ω
P B Q C
182 3 Inversions
A Γ2
Γ1 Γ2
Γ1 Ω2
E
X Z Y
F
Z Y
Ω1
Ω2
F E
B D C Ω1
Fig. 3.40 Proof that EF ZY is cyclic
Denote the incircles of ABC and XBC by 1 and 2 , respectively. Because the
tangents from a point to a circle are equal, BZ = BD = BF and CY = CD = CE,
so there are two more circles in the picture, namely, the circle 1 of center B and
radius BD and the circle 2 of center C and radius CD. Because the radii of 1
and 2 are tangent to 1 and 2 at the points of contact, the circles 1 and 2 are
orthogonal to both 1 and 2 .
Consider an inversion of center D. The circles 1 , 2 , 1 , 2 are mapped into
four lines 1 , 2 , 1 , 2 such that 1 is parallel to 2 and 1 is parallel to 2
and j makes an angle of 90◦ with k , j, k = 1, 2. Consequently, the images
E , F , Z , Y of E, F, Z, Y form a rectangle, which is cyclic. As inversion maps
the circumcircle of E F Z Y to a line or a circle, and as the points E, F, Z, Y are
not collinear, these points must therefore be concyclic.

The third example is a problem that was given at the Romanian Mathematical
Olympiad in 1997.
Problem 3.3 Given n circles (n ≥ 3) that have a common point O, if the circles
cut on two rays starting at O equal segments, then they cut on any ray starting at O
equal segments.
Solution The conclusion follows from the following.
Lemma Let ω1 , ω2 , ω3 be three circles that pass through a point O, and let |OX
and |OY be two rays that cut the circles a second time at A1 , A2 , A3 and B1 , B2 , B3 ,
respectively (with A2 between A1 and A3 and B2 between B1 and B3 ), such that
A1 A2 /A2 A3 = B1 B2 /B2 B3 . Then
(a) ω1 , ω2 , ω3 have a second common point or are tangent at O;
(b) if |OZ is a ray that cuts ω1 , ω2 , ω3 at C1 , C2 , C3 , respectively, then
C1 C2 /C2 C3 = A1 A2 /A2 A3 .
A3 A3
X
A2 X
A2 A1
A1 B3 B1 B2 B3 = B3
B1 B2 O
O Y Y
C1
C2
C1 C3
C2
C3
Z
Z
P
P
Fig. 3.41 Möbius transformation acting on the configuration of rays and circles
Proof The proof of the lemma combines linear fractional transformations in both
one-dimensional real and one-dimensional complex projective geometry.
First, we use a Möbius transformation φ to map O (where we locate the origin
of the coordinate system) to the point at infinity and the point at infinity to O, and
then ω1 = φ(ω1 ), ω2 = φ(ω2 ) and ω3 = φ(ω3 ) are lines, while the rays |OX,
|OY and |OZ are still mapped into rays that start at φ(∞) = O. The configurations
before and after the Möbius transformation are shown on the left and on the right of
Fig. 3.41.
Denote the complex coordinate of a point by the corresponding lowercase letter,
and use a dash to denote the image of a point through φ. The collinearity and order
of the points allow us to write the metric relation from the statement as
a1 − a2 b1 − b2
= ,
a3 − a2 b3 − b2
both quotients being negative real numbers. Because the cross-ratio is invariant
under Möbius transformations (Theorem 3.10), we have
a1 − a2 a1 − ∞ a1 − a2
(a1 , a3 , a2 , 0) = (a1 , a3 , a2 , ∞) = : =
a3 − a2 a3 − ∞ a3 − a2
b1 − b2 b1 − b2 b1 − ∞
= = : = (b1 , b3 , b2 , ∞) = (b1 , b3 , b2 , 0).
b3 − b2 b3 − b2 b3 − ∞
Let P be the intersection point of A1 B1 and A2 B2 . Let also B3 be the intersection
of P A3 with OY . Then by Theorem 3.20 applied to the pencil of four lines
P O, ω1 , ω2 , P A3
(b1 , b3 , b2 , 0) = (a1 , a3 , a2 , 0) = (b1 , b3 , b2 , 0),
184 3 Inversions
hence b3 = b3 (cross-ratio is a Möbius transformation in each variable; hence it

is injective), so B3 = B3 . This means that P ∈ A3 B3 , and since P ∈ A1 B1 and
P ∈ A2 B2 by construction, it follows that φ −1 (P ) ∈ ωj , j = 1, 2, 3, proving (a).
Part (b) is now straightforward. The equality
c1 − c2 a1 − a2
=
c3 − c2 a3 − a2
follows from (c1 , c3 , c2 , ∞) = (a1 , a3 , a2 , ∞), since by the invariance of

cross-ratio under Möbius transformations this is equivalent to (c1 , c3 , c2 , 0) =
(a1 , a3 , a2 , 0), and the latter follows from Theorem 3.20 applied to the pencil
P O, ω1 , ω2 , ω3 .

To solve the original problem, pick all triples of circles out of the n given circles,
and apply the lemma.

This is a common technique for proving the existence of common intersection
points of curves: map them by a Möbius (or circular) transformation, and check that
their images have a common point. This is a consequence of the fact that Möbius
(and circular) transformations are injective. In this problem we had to show that the
circles have two common points, and this is equivalent to the fact that their images
have two common points on the Riemann sphere, which are P and the point at
infinity.
The next example is a problem proposed by Poland which was given at the 59th
International Mathematical Olympiad in 2018.
Problem 3.4 A convex quadrilateral ABCD satisfies AB ·CD = BC ·DA. A point
X is chosen inside the quadrilateral so that XAB = XCD and XBC = XDA.
Prove that AXB + CXD = 180◦ .
Solution 1 The metric relation from the statement can be rewritten as
AB CB
: = 1,
AD CD
which is a condition about the cross-ratio of segments, and in view of the discussion
from Sect. 3.1.4, we should think about inversion and circular transformations. A
circular transformation will preserve this cross-ratio, and it might yield a nicer
quadrilateral, such as one with equal angles placed in better locations. Inversion
centered at X does the job, and the second official solution starts with an inversion
of center X and radius 1.
With the usual convention that A → A , B → B , C → C , and D → D , we
immediately have that
A B C B
: = 1,
A D C D
and this yields A B · C D = B C · D A .

Fig. 3.42 The image of D

ABCD through an inversion
of center X
X
C
A A C
As for the angles, Theorem 3.2 gives
XB A = XAB = XCD = XD C
and
XC B = XBC = XDA = XA D .
Nothing seems to have changed! But by examining Fig. 3.42 carefully, we notice
also that XB C = XCB, and hence
A B C = A B X + XB C = BAX + BCX = BCD.
Similarly B C D = CDA, C D A = DAB, and D A B = ABC. So not

only that two quadrilaterals satisfy the same properties specified in the statement of
the problem, but they have equal angles, too.
The metric relations give
AB CB DC B C
: = 1 = : ,
AD CD DA BA
which combined with the equalities of angles DAB = C D A and BCD =

A B C implies that the cross-ratio of the points A, B, C, D is equal to the cross-
ratio of the points D , A , B , C . But then the Möbius transformation that maps
A, B, C to D , A , B also maps D to C . And we have the following result.
186 3 Inversions
Lemma Assume that in the convex quadrilaterals XY ZT and X Y Z T are

mapped into each other by a Möbius transformation, and assume that we have the
following equalities of angles: X = X , Y = Y , Z = Z , and T = T .
Then the Möbius transformation is a spiral similarity or a translation.
Proof All we have to show is that XY ZT and X Y Z T are directly similar, since
in that case, there is a spiral similarity or a translation that maps the first into
the second, and because the Möbius transformation is uniquely determined by the
images of three vertices, it must be this spiral similarity or translation.
Construct the quadrilateral XY Z1 T1 similar to X Y Z T and sharing side XY
with XY ZT such that Z1 and T1 lie on the rays Y Z and XT , respectively, and
Z1 T1 ZT , as shown in Fig. 3.43. We need to prove that Z1 = Z and T1 = T .
Assume the contrary. Without loss of generality T X > T1 X (the way Fig. 3.43 is
drawn). Let the segments XZ and Z1 T1 intersect at U . Using the similarities of the
triangles XT1 U and XT Z we obtain
T X Y X T1 X Y X T1 X Y X T X YX

: = : < : = : ,
T Z YZ T1 Z Y Z1 T1 U Y Z T Z YZ
which contradicts the fact that the cross-ratio is invariant under Möbius transforma-
tions. The lemma is proved.

It follows from the lemma that the quadrilaterals ABCD and D A B C are
similar. Using this similarity and the fact that, by formula (3.1)
AB DA
A B = and D A = ,
XA · XB XD · XA
we have
AB D A DA XA · XB DA XB
= = · = · .
BC AB XD · XA AB AB XD
Therefore
Fig. 3.43 The construction Y

of the quadrilateral XY Z1 T1
X
U Z1
T1
T Z
XB AB 2 AB 2 AB
= = = .
XD BC · DA AB · CD CD
A similar computation shows that
XA DA
= .
XC BC
Now using the Law of Sines in the triangles AXB and CXD and the first of these
two relations, we can write
sin AXB AB CD sin CXD

= = = .
sin XAB XB XD sin XCD
The denominators of the first and last fraction are equal by hypothesis, hence so are
the numerators: sin AXB = sin CXD. The other relation yields sin DXA =
sin BXC. If at least one of the pairs ( AXB, CXD) and ( BXC, DXA)
consists of supplementary angles, we are done. Otherwise, AXB = CXD and
DXA = BXC. In this case X = AC ∩ BD, and the conditions about angles
from the statement imply that ABCD is a parallelogram. The metric relation shows
that it is in fact a rhombus, so the diagonals make a right angle, and the sum of the
two angles is again 180◦ .

Solution 2 Inversion about X is not the only possibility. The contestant Jonas
Walter of the German team used the inversion whose center is the point K where
the perpendicular bisector of the segment BD intersects the line AC and whose
radius is KB. This is motivated by the properties of the Apollonian circles (see
Problem 165). Let us explain this approach, with the aid of Fig. 3.44.
The metric relation from the statement can be rewritten as
AB CB
= ,
AD CD
and from it we read that A and C lie on the same Apollonian circle defined by B and
D and this ratio. Let us assume that we are not in the degenerate case where AC,
and thus the Apollonian circle, is the perpendicular bisector of BD, and let us also
assume that B is inside the Apollonian circle (otherwise swap B and D). Call this
Apollonian circle ω, and let its intersections with the line BD be U and V , with the
points U, B, V , D appearing in this order on the line BD. By Proposition 3.1 and
by Problem 165, the points U, B, V , D form a harmonic division on the line BD.
Consider the circle γ of diameter BD, and take an inversion χ of center D. Then
U ∗ = χ (U ), B ∗ = χ (B), V ∗ = χ (V ), and ∞ still form a harmonic division (by the
invariance of real valued cross-ratio), so B ∗ is the midpoint of the segment U ∗ V ∗ .
Note that χ (γ ) is the perpendicular bisector of U ∗ V ∗ , and U ∗ and V ∗ are reflections
of each other over χ (γ ). By the Symmetry Principle (Theorem 3.15), U and V are
188 3 Inversions
X
B N D
C
F
Fig. 3.44 The inversion of ABCD with center at K
symmetric with respect to γ . So if the center of γ is N, then NU · NV = NB 2 , the

later being the ratio of the inversion χ .
From this metric relation we deduce that the perpendicular bisector of BD is the
radical axis of ω and the degenerate circle of center B and radius 0. It follows that
the point K defined above satisfies
KB 2 = KA · KC.
This equality implies that the inversion of center K and radius KB = KD maps any
circle through A and C into itself. We now choose a particular such circle. For that
we make a second assumption that ABCD is not a trapezoid. Let the intersection
of AB and CD be E and the intersection of AD and BC be F . The circumcircle of
ACE is the circle we have in mind. Working with directed angles modulo 180◦ , so
that the argument does not depend on the configuration, we translate the condition
XAB = XCD from the statement into XAE = XCE, which then implies
that X belongs to the circumcircle of ACE.
So we have determined that the circumcircle of AEC contains X and is invariant
under the inversion. The other angle equality from the statement implies that X also
lies on the circumcircle of BDF . An important observation is that the circumcircle
of BDF is the image through the inversion of the circumcircle of BDE. Indeed, if
E is the image of E, then by working once more with directed angles modulo 180◦
and using Theorem 3.2, we have
BE D = BE K + KE D = KBE + EDK = KBA + CDK
By the same proposition the triangles KCD and KDA are similar, and KBC and
KAB are also similar, and so this is further equal to
BCK + KAD = F CA + CAD = F CA + CAF = CF A = BF D.
So, as directed angles modulo 180◦ , BE D = BF D showing that B, F, E , D

are concyclic, which proves our claim.
Because X lies at the intersection of the circumcircles of ACE and BDF , its
image X must lie at the intersection of the circumcircles of ACE and BDE.
This point cannot be E, for X and E lie on different sides of line AC which
passes through the center of inversion. Hence X is the point of intersection of the
circumcircles of ACE and BDE different from E. Using the formula (3.1) we
compute
KB 2 KB 2 KB 2
X B = XB, X D = XD = XD,
KX · KB KX · KD KX · KB
and so
XB X B
= .
XD XD
Examining the configuration consisting of the circumcircles of ACE and BDE,
we see that these circles intersect at E and X, and we know that the points
E, D, C are collinear and the points E, B, A are collinear. Using the result proved
in Problem 131 we deduce that there is a spiral similarity of center X that maps the
circles into each other such that A is mapped to B and C is mapped to D. Then the
triangles X AB and X CD are similar, so
X B AB
= .
X D CD
We have obtained that
XB AB
= .
XD CD
Switching A, C with B, D we deduce that
XA DA
= .
XC DC
And we have seen in the first solution that this yields the conclusion.
190 3 Inversions
Of course, we have left out a few cases. If BC and AD are parallel, then F lies
at infinity, so the circumcircle of BDF becomes the line BF . The argument adapts
mutatis mutandis. It might happen that AB is parallel to CD. Again, circumcircles
become lines, and the argument works. Finally, it might happen that AC is the
perpendicular bisector of BD, so that ABCD is a kite. We can switch A, C with
B, D if BD is not the perpendicular bisector of AC. If it is, then we have a rhombus,
and X is necessarily its center, and then the conclusion is obvious.

Solution 3 During the discussions of the jury, a solution arose that uses inversion
with respect to the Apollonian circle ω defined by the points B and D and the ratio
AB/AD = CB/CD. Let O be the center of ω. As the circle of diameter BD is
orthogonal to the Apollonian circle, by Proposition 3.25, it is invariant under the
inversion, so the inversion maps
A → A, C → C, B → D.
From here, the argument can be followed on Fig. 3.45. Again, let E = AB ∩ CD,
F = AD ∩BC, and we have seen above that X is on the circumcircles of both ACE
and BDF . The other intersection point of this two circles is F , the inverse of F .
Indeed, by OF · OF = OB · OD, we have that F is on the circumcircle of BDF
(by power-of-a-point). To show that it is on the circumcircle of ACE, we notice
first that Theorem 3.2 gives BCO = ODC, OAD = ABO, AF O =
OAF , and OF C = CF O. In the following computation, and for the rest of
the solution, we work with directed angles modulo 180◦
AF C = AF O + OF C = OAF + F CO = OAD + BCO

= ABO + ODC = ABD + BDC = AEC,
where for the last step we have used the Exterior Angle Theorem in the triangle
EBD. So F is indeed on the circumcircle of AEC.
Let T be the intersection of AC and BD, and let T be its image through the
inversion. Define X1 as the intersection of the circumcircles of BT C and AT D. We
will show that the image X1 of X1 through the inversion is X.
Indeed, using inscribed angles and the Exterior Angle Theorem in ACF
BX1 D = BX1 T + T X1 D = BCT + T AF = AF C = DF B.
Therefore X1 is on the circumcircle of BDF , and so is X1 , its image through the
inversion, because inversion preserves this circle (as OB · OD = OA2 ). A similar
angle chasing argument gives
CX1 A = CX1 T + T X1 A = CBT + T DA = CBD + BDA = BF A

= CF A.
D
X O
T T
F
B C
Fig. 3.45 The inversion about the circle of Apollonius
From here we deduce that A, F, C, X1 lie on a circle, and using the inversion, we
obtain that A, F , C, X1 lie on a circle as well. So X1 lies on the circle through
A, C, F , which we have seen is the circumcircle of ACE. We conclude that X1 is
the intersection point of ACE and BDF that is distinct from F , and thus it is X.
As the quadrilateral DX1 T A is cyclic, so is the quadrilateral formed by the
images of its vertices through the inversion, and the latter is ABT X. Exchanging
the roles of A and C, as well as B and D, we deduce that DCT X is also cyclic.
Using the fact that OAC is isosceles and that OT A = T AO by the inversion,
together with the two quadrilaterals that were proved to be cyclic, we can write
BXA= BT A= OT A= T AO = OCT = CT O = CT D = CXD,
We are working with directed angles modulo 180◦ , so
AXB = − BXA = − CXD = 180◦ − CXD,
and the problem is solved.

The lesson learned from this problem is that whenever you are in the presence a
configuration of four points such that the cross-ratio of their coordinates has absolute
value 1 (which is the case with the vertices of the quadrilateral), it is worth thinking
about the circles of Apollonius.
Next, we present an example from a 2003 Team Selection Test for the Moldovan
International Mathematical Olympiad Team because it makes use of the polar of a
point with respect to a circle.
Problem 3.5 Let ABCD be a quadrilateral inscribed in a circle of center O, and
let M and N be the midpoints of the diagonals AC and BD, respectively. Prove
192 3 Inversions
Fig. 3.46 The images of M

and N
O D
A
M N
B C
M N
that the points O, M, B, D are concyclic if and only if the points O, N, A, C are
concyclic.
Solution 1 Consider the inversion with respect to the circumcircle of ABCD. Then
the circumcircle of OAC becomes the line AC, and the circumcircle of OBD
becomes the line BD. Let M and N be the images of M and N , respectively
(Fig. 3.46), which, as we know from Proposition 3.3, lie the first at the intersections
of the tangents to the circle of inversion at A and C and the second at the intersection
of the tangents at B and D. Then M is on the circumcircle of OBD if and only if
M is on the line BD and N is on the circumcircle of OAC if and only if N is on
the line AC.
But the line BD is the polar pN of N with respect to the circle of inversion, and
the line AC is the polar pM of M with respect to the same circle. By La Hire’s
Theorem (Theorem 3.31), M ∈ pN if and only if N ∈ pM , and the problem is
solved.

Solution 2 It is worth showing the complex number solution just because it is
thoroughly elementary. Four points lie on a line or circle if and only if their cross-
ratio is real. Let A, B, C, D have coordinates a, b, c, d of absolute value 1, so that
O is at the origin. The complex number translation of the equivalence from the
statement is
a+c b+d
−b b −a a
2 : ∈ R ⇐⇒ 2 : ∈ R.
a+c b+d
−d d −c
c
2 2
Setting x = a/b, y = c/b, z = a/d, and w = c/d, we translate this equivalence
into
1 1
+ −2
x+y−2 x z
∈ R ⇐⇒ ∈ R.
z+w−2 1 1
+ −2
y w
Using the fact that x, y, z, w have absolute value one, by taking the complex
conjugate of the right-hand side, we can transform this equivalence into
x+y x+z
−1 −1
2 ∈ R ⇐⇒ 2 ∈ R.
z+w y+w
−1 −1
2 2
Note that xw = yz, so the points X, Y, Z, W with coordinates x, y, z, w form an
isosceles trapezoid inscribed in the unit circle with XZ parallel to Y W . The first
condition expresses the fact that the line joining the midpoints of XY and ZW passes
through 1, while the second condition expresses the fact that the line joining the
midpoints of XZ and Y W passes through 1. But the two lines coincide with the
midline of the trapezoid, and the problem is solved.

The next example uses two inversions simultaneously. It is a short-listed problem
which was proposed by Poland for the 1998 International Mathematical Olympiad,
and the first solution that we present belongs to Octav Drăgoi.
Problem 3.6 Let ABCDEF be a convex hexagon such that B + D+ F = 360◦
and
AB CD EF
· · = 1.
BC DE F A
Prove that
BC AE F D
· · = 1.
CA EF DB
Solution 1 Using the hypothesis of the problem, we can rewrite the conclusion as
the more complicated equality
AB CD DE F A
· = · .
CA DB F D AE
We now consider two inversions, the first of center A and the second of center D,
and both of arbitrary radii. With the convention that we denote, the image of a point
P under the first inversion by P and under the second inversion by P (Fig. 3.47),
by applying the formula (3.1), we obtain
AB CD C D DE F A F A
· = and · = .
CA DB DB F D AE A E
194 3 Inversions
C B A D
B
D F
A C
E D
F E
A
Fig. 3.47 The hexagon ABCDEF and the two inversions
The problem reduces to showing that
C D F A
= ,
DB A E
and this is somewhat simpler. We prove this equality by placing the segments in the
two similar triangles B C D and E F A .
So let us show that the two triangles are similar. Rewrite the hypothesis as
AB · CD DE · F A
= ,
BC EF
and apply again formula (3.1) to obtain
AB · CD C D DE · F A F A
= and = .
BC · AD BC EF · AD E F
We obtain the desired equality of the ratios of the sides.
C D F A

= .
BC E F
As for the angles formed by those sides, we first use Theorem 3.2 to obtain
B C D = B C A + AC D = B + ADC
and
E F A = E F D + DF A = E + DAF = 2π − F − ADE.
And then we use the angle equality from the hypothesis of the problem to conclude
that B C D = E F A . This combined with C D /B C = F A /E F
proved above implies that the triangles B C D and E F A are similar, as claimed,
which then implies
C D F A
= ,
DB A E
completing the proof.

Solution 2 The above solution can be encoded in complex numbers, where it
becomes, surprisingly, more natural. Using lowercase letters for the coordinates of
the uppercase points, we translate the hypothesis of the problem into the fact that
the number
a−b c−d e−f
· ·
c−b e−d a−f
is both real positive (because the arguments of the three fractions add up to 2π ) and
has absolute value equal to 1 (the metric relation from the statement). Therefore this
number is equal to 1, and we can rewrite this fact in terms of cross-ratios as
b−a d −a e−d a−d

: = : .
b−c d −c e−f a−f
The conclusion of the problem, as rewritten in the first solution, states that the cross-
ratios
a−b d −b d −e a−e
: and :
a−c d −c d −f a−f
have equal absolute values. In fact
a−b d −b d −e a−e
: = : ;
a−c d −c d −f a−f
these cross-ratios are equal! To be able to see this, we examine separately the left-
hand sides and the right-hand sides of the two equalities of cross-ratios that comprise
the hypothesis and the conclusion of the problem. To this end we choose a Möbius
transformation that maps a to ∞, and by using the invariance of the cross-ratios
under Möbius transformations, we infer that the two left-hand sides transform into
d − c d − c
and .
b − c d − b
p
If we call the first fraction p, then the second is p−1 . Similarly, if we choose a
Möbius transformation that maps d to ∞, the cross-ratios on the right sides become
a − f a − f

and ,
e −f a − e
196 3 Inversions
q
and if we denote the first fraction by q, the second is q−1 . Of course p = q
p q
implies p−1 = q−1 , so by using the invariance of cross-ratios under Möbius
transformations, we deduce the desired equality of cross-ratios. Certainly, we could
have used the more general circular transformations (and in particular inversions),
as we did in the first solution, and work with complex conjugates.

√
Let us illustrate the use of the bc inversion with two Olympiad problems.
The first has appeared in the 2015 United States of America Junior Mathematical
Olympiad, being proposed by Kim Sungyoon.
Problem 3.7 Let ABCD be a cyclic quadrilateral. Prove that there exists a point X
on the segment BD such that BAC = XAD and BCA = XCD if and only
if there exists a point Y on segment AC such that CBD = Y BA and CDB =
Y DA.
Solution We
√ argue on Fig. 3.49. Because AC and AX are isogonals in the triangle
ABD, the bc inversion corresponding to the vertex A in this triangle swaps B and
D as well as C and X. Hence the triangles ABC and AXD are similar, which gives
AD AC
= .
XD BC
√
Analogously, the bc inversion corresponding to vertex C in the triangle CBD
swaps B and D as well as A and X. Hence the triangles ABC and DXC are similar
and
CD CA
= .
XD BA
Substituting XD from one equation into the other, we obtain
AD AB
= .
CD CB
Thus, the quadrilateral ABCD is harmonic. Moreover, in √ a harmonic quadrilateral,
the midpoint of BD is swapped with point C by both bc inversions, so it plays
the role of the point X. Thus, the existence of X is equivalent to ABCD being
harmonic. But so is the existence of Y , so the two conditions from the statement are
equivalent.

We have seen here yet another characterization of
√ harmonic quadrilaterals.
The second example that illustrates the use of bc inversion is a problem that
was given at the Romanian Master of Mathematics in 2011, being proposed by
Vasily Mokin and Feodor Ivlev from Russia.
Problem 3.8 The triangle ABC is inscribed in the circle ω. A variable line
parallel to BC intersects the segments AB and AC at D and E, respectively, and
intersects ω at K and L such that D is between K and E. The circle γ1 is tangent to
Fig. 3.48 Finding the locus A

of the intersection point of the
interior tangents
X
B D
Fig. 3.49 The quadrilateral A

ABCD is harmonic
ω
K D E L
γ1 O 1
X O2 γ2
B C
the segments KD and BD and to ω, and the circle γ2 is tangent to the segments LE
and CE and to ω. Find the locus of the point of intersection of the common interior
tangents of γ1 and γ2 when varies.
Solution Denote the intersection of the interior tangents of γ1 and γ2 by X. The
situation is illustrated
√ in Fig. 3.48.
Let φA be the bc inversion determined by the triangle AKL and the vertex A.
Then
φA (ω) = KL, φA (KL) = ω, φA (AB) = AB, φA (AC) = AC.
Also φA (C) = D (the intersection of the image of ω with AB), and φ(B) = E.
Because tangencies are preserved
φA (γ1 ) = γ2 and φA (γ2 ) = γ1 .

198 3 Inversions
Denote by O1 the center of γ1 and by O2 the center of γ2 . Because the inversion

that defines φA keeps AO1 and AO2 invariant, after composing with the reflection,
the two lines are mapped into each other. Hence, AO1 and AO2 are symmetric with
respect to the angle bisector of KAL, which is the same as the angle bisector of
angle BAC (since KL and BC are parallel, the bisectors of KAL and BAC

cut the arc BC at the same point, so they coincide).
Let r1 and r2 be the radii of γ1 and γ2 , respectively. Because BAO1 = CAO2 ,
we have
AO1 r1
= .
AO2 r2
Also, if X is the intersection of the interior common tangents of γ1 and γ2 , then
XO1 r1
= .
XO2 r2
Consequently
AO1 XO1
= ,
AO2 XO2
which implies, by the Bisector Theorem, that X is on the bisector of O1 AO2 . But,
again because AO1 is mapped to AO2 by φA , this angle bisector coincides with the
bisector of BAC.
So the locus is a subset of the bisector of ABC. To understand which part of
the locus, we examine the limiting cases. These limiting cases are when passes
through A, when X = A, and when = BC, when X is on BC. And, by continuity,
any point on the angle bisector of BAC that lies on the open segment of this
bisector with one endpoint A and the other endpoint on BC is part of the locus. So
the locus is this open segment.
3.2.3 Problems in Euclidean Geometry to be Solved with

Inversion (or with Möbius Transformations)
And now, a rich list of practice problems.

169 Let ω1 and ω2 be two circles that intersect at A and B. The tangents at A to the
two circles intersect these circles a second time at M and N . Let C be the reflection
of A over B. Show that the quadrilateral AMCN is cyclic.
170 (Steiner’s Theorem) Let ABC be a triangle, and let D and E be two points on
the side BC such that DAB = EAC. Prove that
DB EB AB 2
· = .
DC EC AC 2
171 Let ABCD be a quadrilateral that admits an inscribed and a circumscribed

circle. Let M, N, P , Q be the tangency points of the inscribed circle with the sides
AB, BC, CD, DA, respectively. Prove that MP is perpendicular to NQ.
172 Consider two lines 1 and 2 that intersect at P and the circles ω1 , ω2 , ω3 , ω4
through P such that ω1 and ω3 are tangent to 1 and ω2 and ω4 are tangent to 2 .
Let Aj be the intersection of ωj and ωj +1 , j = 1, 2, 3, 4 (with ω5 = ω1 ).
(a) Prove that the following relation holds:
A1 A2 A3 A2 P A22
· = .
A1 A4 A3 A4 P A24
(b) Prove that A1 , A2 , A3 , A4 are concyclic if and only if 1 and 2 are orthogonal.
173 Let ω be a circle and let ST be a chord. The circle ω1 is orthogonal to ω and
has the center on ST . Let ω1 intersect the line ST at A1 and A2 and the circle ω at
B1 and B2 , such that A1 lies on the chord ST . Let also M be the midpoint of the

arc ST of ω that does not contain the point B1 . Prove that the lines A1 B1 and A2 B2
intersect at M.
174 (Shoemaker’s Knife Problem) Let 1 and 2 be two circles that are interior
tangent, and let ωn , n ≥ 0, be circles such that the center of ω0 is collinear with the
centers of 1 and 2 , and for each n ≥ 1, the circle ωn is tangent to 1 , 2 , and
ωn−1 . Denote by hn the distance from the center of ωn to the line of centers of 1
and 2 and by dn the diameter of ωn . Prove that
hn
= n.
dn
175 Given a non-isosceles triangle ABC, consider the medians AA1 , BB1 , CC1
and the circumcenter O. Show that the circumcircles of AA1 O, BB1 O, and CC1 O
have a second intersection point besides O.
176 Let and ω be two concentric circles with common center O, with ω inside
. Consider a point A on , and let B and C be the points of intersection of ω with
the circle of diameter OA. Let also D be the point where the ray |OA intersects ω,
and let E and F be the points where the rays |OB and |OC intersect , respectively.
Show that D, E, F are collinear.
177 Show that the three perpendiculars taken from the vertices of a triangle onto
the corresponding sides of its orthic triangle intersect at the circumcenter of the
triangle.
200 3 Inversions
178 Let be a circle of diameter AB, let P be a point outside of that does not
lie on the tangent at A, and let t be the tangent at B. Consider the points C and D
on the circle such that P C and P D are tangent to the circle, and let C , D , P be
the intersections of AC, AD, and AP with the line t, respectively. Prove that P is
the midpoint of the segment C D .
179 Two points, A and B, are chosen on a circle k of center S such that ASB =
90◦ . The circles k1 and k2 are internally tangent to k at A and B, respectively, and
are also externally tangent to each other. The circle k3 lies in the interior of the angle
ASB and is internally tangent to k at C and externally tangent to k1 and k2 at X
and Y , respectively. Prove that XCY = 45◦ .
180 (a) Let ω and ω1 be two circles such that ω1 is interior tangent to ω at A. A

chord ST of ω is tangent to ω1 at B. Let M be the midpoint of the arc ST of ω that
does not contain A. Prove that A, B, M are collinear.
(b) Let ω be a circle, ST a chord of ω, and ω1 and ω2 two circles that are tangent to
both ω and ST and intersect
at P and Q. Prove that the line P Q passes through the
midpoint of the arc ST of ω that lies on the other side of ST from ω1 and ω2 .
(c) Let ABC be a triangle, D a point on the side BC, ω1 a circle tangent to the
circumcircle and to the segments AD and BD, and ω2 a circle tangent to the
circumcircle and to the segments AD and CD. Prove that ω1 and ω2 are tangent
if and only if BAD = DAC.
181 A circle ω is tangent to the parallel lines 1 and 2 , and the circles ω1 and
ω2 are exterior tangent to each other and to ω, also ω1 is tangent to 1 , while ω2 is
tangent to 2 . Let A be the point where ω1 and ω are tangent, and let B be the point
where ω2 and 2 are tangent. Prove that the line AB is the common interior tangent
of ω and ω1 .
182 Let ABC be a triangle with circumcenter O. On the sides, AC and BC are
constructed; in the exterior, the rectangles ACDE and BCF G of equal areas. Let
M be the midpoint of the segment DF . Prove that the points O, C, M are collinear.
183 A right triangle ABC ( B = 90◦ ) is inscribed in a circle. We let K and N

be the midpoints of the arc BC and the segment AC, respectively, and let M be the
second intersection point of KN with the circle. The tangents to the circle at B and
C intersect at E. Prove that EMK = 90◦ .
184 Let ABC be a triangle that is not isosceles and let H be its orthocenter. Denote
by A1 , B1 , C1 the feet of the altitudes AH, BH, CH , respectively. Let A2 , B2 , C2
be the projections of H onto B1 C1 , C1 A1 , and A1 B1 , respectively. Prove that:
(a) the circumcircles of the triangles H A1 A2 , H B1 B2 , H C1 C2 have a second
common point besides the orthocenter H .
(b) the circumcircles of the triangles H AA2 , H BB2 , H CC2 have a second common
point besides the orthocenter H .
185 Let C, C1 , C2 , C3 be four circles in the plane, and let be a line that does not
intersect C, such that each of the circles C1 , C2 , C3 is tangent to the other two, as
well as to the circle C and to the line . Knowing that the radius of C is 1, find the
distance from its center to .
186 Let ABC be a triangle for which A < C, with two points, D and E, chosen
on the sides AC and AB, respectively, so that BED = C. Let also F be a
point inside the quadrilateral BCDE such that the circumcircles of BCF and DEF
are tangent and the circumcircles of BEF and CDF are tangent. Prove that the
quadrilateral ACF E is cyclic.
187 In the triangle ABC, consider the excircle tangent to side BC at A1 and
to the lines AC and AB at B1 and C1 , respectively, and let Ia be the center of
this circle. Denote by A , B , C the midpoints of the segments B1 C1 , A1 C1 , and
A1 B1 , respectively. Let a be the line that passes through Ia and the circumcenter of
A B C . Using the excircles corresponding to the other two sides, define similarly
the lines b and c . Prove that the lines a , b , and c are concurrent.
188 In a triangle ABC, D is the foot of the altitude from A, E is the point
where the symmedian from A intersects the circumcircle, and F is the point where
the diameter from A of the circumcircle intersects the side BC. Prove that the
circumcircles of the triangles ABC and DEF are tangent.
189 In the triangle ABC, consider a point X on the altitude from A. The circles
C1 and C2 have centers on the line BC, and C1 passes through B and X, while
C2 passes through C and X. Let M and N be the intersections of C1 with the side
AB and with the altitude from B, and let P and Q be the intersections of C2 with
the altitude from C and with the side AC. Prove that the points M, N, P , Q are
collinear.
190 Let ABC be a triangle, and let K and L be the midpoints of the sides AB and
AC, respectively. Let P be the second intersection point of the circumcircles of the
triangles ABL and ACK, and let Q be the second intersection point of the line AP
with the circumcircle of the triangle AKL. Prove that 2AP = 3AQ.
191 Let ω be the circumcircle of the triangle ABC, and let be the line through A
that is tangent to ω. The circles ω1 and ω2 are tangent to and to the line BC and
are exterior tangent to ω. Let D and E be the points where ω1 and ω2 are tangent to
BC, respectively. Prove that the circumcircles of ABC and ADE are tangent.
192 Let ABCD be a quadrilateral satisfying:
(i) ABC = ADC = 135◦
(ii) AC 2 · BD 2 = 2AB · BC · CD · DA.
Prove that the diagonals of the quadrilateral are orthogonal.
193 Connect a point M with the vertices of the triangle ABC. Prove that the
perpendiculars from the orthocenter H onto MA, MB, and MC meet BC, CA,
202 3 Inversions
and AB, respectively, at three collinear points. Show, additionally, that the line
determined by these three points is orthogonal to MH .
194 Let ABC be a triangle, and let D be the second intersection point of the circle
that passes through B and is tangent to the line AC at A with the circle that passes
through C and is tangent to the line AB at A. Let E be a point on the line AB such
that B is the midpoint of AE, and let F be the second point of intersection of the
ray |CA with the circumcircle of ADE. Prove that AF = AC.
195 Let ABC be a triangle and let O be its circumcenter. The lines AB and AC
intersect the circumcircle of the triangle BOC again at B1 and C1 , respectively. Let
D be the intersection of BC and B1 C1 . Prove that the circle tangent to AD at A and
with center on B1 C1 is orthogonal to the circle of diameter OD.
196 In the acute triangle ABC, let M be the midpoint of the angle bisector AD
(D ∈ BC). Consider a point X on the segment BM such that MXA = DAC.
Show that AXC = 90◦ .
197 Can one find 1975 points on the unit circle such that the distance between any
two is rational?
198 Let P be a point inside the triangle ABC such that
AP B − C = AP C − B.
Let D and E be the incenters of the triangles AP B and AP C, respectively. Prove

that AP , BD, CE meet at one point.
199 Let BD be the internal bisector of B in the triangle ABC (D ∈ AC). The
line BD intersects the circumcircle of the triangle ABC at E = B. The circle
ω of diameter DE intersects again at F . Prove that BF is a symmedian in the
triangle ABC.
200 Let be the circumcircle of the triangle ABC, and let ω be a circle that passes
through B and C, whose center we denote by O1 . We also denote by E and F the
second intersection points of ω with the lines AB and AC, respectively, by Q the
intersection of the lines EF and BC, and by P the second intersection point of AQ
with . Prove that AP O1 = 90◦ .
201 Let ABC be a triangle, let be its circumcircle, let ω be the C-mixtilinear
incircle, and let be the line that is parallel to the side AB, is tangent to the circle
ω, and crosses the sides AC and BC. Denote by P and Q the tangency points of the
circle ω with the circle and the line , respectively. Prove that ACP = BCQ.
202 Prove that the perpendiculars from the vertices of a triangle onto the lines
determined by the orthocenter and the midpoints of the opposite sides meet the
corresponding sides along a line orthogonal to Euler’s line.
203 Let ABCD be a cyclic quadrilateral, and let M and N be the midpoints of the
diagonals AC and BD. Prove that AC is the angle bisector of BMD if and only if
BD is the angle bisector of ANC.
204 Let A be the A-mixtilinear incircle of the triangle ABC, and let A touch
the circumcircle of ABC at Ua . Invert the circumcircle of ABC about its incircle,
obtaining . Prove that A and Ua are mapped to diametrically opposite points in
.
205 Let and ω be two circles interior tangent at P (with ω inside ), and let
AB be a chord of that is tangent at some point C to ω. The line P C intersects a
second time the circle at Q, and the tangents from Q to ω intersect again at R
and S. Let I, X, Y be the incenters of triangles ABP , ABR, and ABS, respectively.
Show that P XI + P Y I = 90◦ .
206 A circle with center O passes through the points A and C and intersects the
sides AB and BC of the triangle ABC at the points K and N , respectively. The
circumcircles of the triangles ABC and KBN intersect at two distinct points B and
M. Prove that OMB = 90◦ .
207 Let ABC be an acute triangle with circumcenter O, and let BE be the altitude
from B (E ∈ AC). The circumcircle of AOC intersects AB at P = A and BC
at Q = C. Let ω be the circle of center E and radius EB. Prove that P , Q, E are
collinear if and only if the circumcircle of AOC is tangent to ω.
208 Two circles 1 and 2 intersect at the points X and Y . The line through Y
that is parallel to the closer common tangent to 1 and 2 intersects 1 and 2 a
second time at A and B, respectively. Let O be the center of the circle that is exterior
tangent to 1 and 2 and interior tangent to the circumcircle of XAB. Prove that
the line XO is the angle bisector of AXB.
209 Let ω be a circle in the plane, and let A and B be two points on it. Let M be
the midpoint of the chord AB, and let P be another point on this chord. Construct
the circles γ and δ tangent to AB at P and to ω at C and D, respectively. Let E be
the point that is diametrically opposite to D in ω. Prove that the circumcenter of the
triangle BMC lies on the line BE.
210 Let ABC be an acute triangle, with AB > AC, let be its circumcircle, let
H be the orthocenter, and let F be the foot of the altitude from the vertex A. Let
also M be the midpoint of the side BC, and let Q, K be the points on the circle
for which H QA = H KQ = 90◦ . We assume that the points A, B, C, K, Q are
distinct and show up on the circle in this order. Prove that the circumcircles of the
triangles KH Q and F KM are tangent to each other.
211 Let ABCD be a convex quadrilateral with B = D = 90◦ . The point H is
the foot of the perpendicular from A to BD. The points S and T are chosen on the
sides AB and AD, respectively, in such a way that H lies inside triangle SCT and
SH C − BSC = 90◦ , T H C − DT C = 90◦ .

204 3 Inversions
Prove that the circumcircle of triangle SH T is tangent to the line BD.

212 Let ABC be a triangle with incenter I and circumcenter O. A circle is tangent
to the side AB at point K, to the side BC at point L, and is externally tangent to the
circumcircle of AOC. Prove that the midpoint of BI lies on the line KL.
213 Let ABC be a triangle with circumcenter O. The circle tangent to the side AB
that passes through the points A and C meets the circumcircle of BOC at T = C.
The lines T O and BC meet at the point K. Show that the line AK is tangent to the
circumcircle of the triangle ABC.
214 Let P be a point on the circumcircle ω of the triangle ABC. The lines P A and
BC meet at A1 , the lines P B and CA meet at B1 , and the lines P C and AB meet
at C1 . The inversion about ω maps A1 to A2 , B1 to B2 , and C1 to C2 . Prove that the
lines AA2 , BB2 , and CC2 have a common point whose isogonal conjugate is on the
nine-point circle of ABC.
215 Let ABC be a triangle with AD as altitude (D ∈ BC). Let M be the midpoint
of AD. The points X and Y are the orthogonal projections of D onto CM and BM,
respectively. The lines BX and CY meet at Z. Show that the circumcircle of the
triangle XY Z and the circle with diameter AD are tangent.
216 Let ω be the circumcircle of the triangle ABC. The mixtilinear incircle γ
is tangent to the sides AB and AC at the points P and Q, respectively, and to
the circumcircle ω at the point S. The lines AS and P Q meet at T . Prove that
BT P = CT Q.
217 Let Ob and Oc be the centers of the B- and C-mixtilinear excircles of the
triangle ABC, respectively. These mixtilinear circles touch the circumcircle of
ABC at R and S, respectively, and the line BC at D and E, respectively. The
lines Ob S and Oc R meet at X, and the lines Ob E and Oc D meet at Y . Prove that
BAX = CAY .
218 The incircle of a scalene triangle ABC has center I and touches the side BC

at the point D. The point X lies on the arc BC of the circumcircle of ABC that
does not contain A, and is such that if E and F are the orthogonal projections of
X onto BI and CI , and if M is the midpoint of EF , then MB = MC. Prove that
BAD = CAX.
219 Let ABC be a scalene triangle with circumcircle , and suppose the incircle
of ABC touches BC at D. The angle bisector of A meets BC and at E and F ,
respectively. The circumcircle of the triangle DEF intersects the A-excircle at the
points S1 and S2 and the circumcircle the second time at T . Prove that the line
AT passes through either S1 or S2 .
220 (Descartes’ Theorem) Let ω1 , ω2 , ω3 , ω4 be four circles that are pairwise
exterior tangent, and let r1 , r2 , r3 , r4 be their radii, respectively. Show that
2
1 1 1 1 1 1 1 1
2 2+ 2+ 2+ 2 = + + + .
r1 r2 r3 r4 r1 r2 r3 r4
221 Let ABC be a triangle, and let D, E, and F be the tangency points of the
incircle with BC, CA, and AB, respectively. Let EF meet the circumcircle
of ABC at X and Y . Furthermore, let T be the second intersection point of the
circumcircle of DXY with the incircle. Prove that AT passes through the tangency
point of the A-mixtilinear incircle with .
Chapter 4
A Synthesis
4.1 Bringing Together All Transformations
In the first three chapters, we have put each geometric transformation in its own
spotlight. And we have observed that each has its own character, leading to a specific
mindset. Now that we understand the transformations well enough, it is the time to
allow them to work together. As different as they look synthetically, in the analytic
perspective, the main actors of this book are very much alike. Just take a look at the
list of orientation-preserving transformations:
Transformation General equation in C

Translation f (z) = z + b
Rotation f (z) = az + b, |a| = 1, a = 1
Homothety f (z) = az + b, a ∈ R \ {0, 1}
Spiral similarity f (z) = az + b, a ∈ C \ R
az + b
Möbius transformation f (z) = , ad − bc = 0
cz + d
The first four are particular cases of the affine transformation f (z) = az + b,
and all transformations are, in fact, Möbius transformations! And we have seen
in Sect. 3.1.5 that Möbius transformations themselves are closely related to linear
transformations. If you throw in reflections over lines and circles, you have the
complete picture of all the transformations discussed in this book. The point
is that all these transformations are naturally intertwined; this is why so many
problems admit solutions based on different transformations. We should not think
of a problem as a “translation problem” or an “inversion problem”; but as a
“transformation problem”!
Then the reader might ask, “why think of synthetic geometry at all and not
focus on the analytic?” The answer is that, while complex geometry is potentially
effective, it masks the geometric insight under lengthy, and sometimes fruitless,
208 4 A Synthesis
computations. Moreover, a good synthetic observation leads organically to the right

transformation, and points to a more concise and elegant solution.
Finally, there is nothing wrong with integrating synthetic observations with
complex computations. Some of these computations have synthetic meanings that
are harder to convey through synthetic means. Better be pragmatic than purist.
4.1.1 Some Examples
Let us visit mathematics’ most ubiquitous theorem.

Theorem 4.1 (Pythagorean Theorem) In a right triangle, the square of the length
of the hypotenuse is equal to the sum of the squares of the lengths of the other two
sides.
Proof 1 We present first the proof by rotation published by Euclid in his Elements.
Let the right triangle be ABC with A = 90◦ . Construct in the exterior of the
triangle the squares ABMN , BCP Q, and CARS as shown in Fig. 4.1. We are
supposed to prove that the areas of ABMN and CARS add up to the area of
BCP Q.
Construct the altitude AD and let E be the intersection of the lines AD and P Q.
The triangles MBC and ABQ are mapped into each other by a 90◦ rotation about
B, so they have the same area. Similarly, the triangles SCB and ACP are mapped
into each other by a 90◦ rotation about C, so they have the same area. The area of the
triangle MBC is half the area of the square ABMN , because they have the same
base MB and the same height MN. For a similar reason, the area of the triangle
ABQ is half the area of the rectangle BDEQ. Thus, the area of the square ABMN
is equal to the area of the rectangle BDQE. The same argument shows that the area
of the square CARS is equal to the area of the rectangle CDEP . But the rectangles
Fig. 4.1 Proof of the R

Pythagorean Theorem using
rotations
S
N
A
B D C
Q E P
4.1 Bringing Together All Transformations 209
BDQE and CDEP make up the rectangle BCP Q. Hence, the area of the latter is
equal to the sum of the areas of ABMN and CARS, and the Pythagorean Theorem
is proved.
Proof 2 The second proof, by translation, has appeared in Roger B. Nelsen’s book
Proofs without Words. It can be followed on Fig. 4.2. On the sides BC, AC, and
AB of the right triangle ABC ( A = 90◦ ), construct in the exterior the squares
S1 , S2 , S3 , respectively. We want to show that S1 can be dissected into pieces that
can be translated so as to produce the squares S2 and S3 . That would imply that
the area of S1 is the sum of the areas of S2 and S3 , which proves the Pythagorean
Theorem.
To this end, translate the sides of S1 to segments whose midpoints coincide with
the center of S3 . They divide S3 into four quadrilaterals, α, β, γ , and δ, which are
equal because they map into one another by rotations about the center of S3 . Let
x and y be the lengths of the two segments that the side of S3 is divided into (see
Fig. 4.2). Translate α, β, γ , and δ inside S1 as shown in the figure. A square S is left
uncovered at the center of S1 . We are to show that S is equal to S2 . The side-length
of S is y − x. By examining the parallelogram having BC as one side that appears
in the figure (this parallelogram has a side equal to y and the opposite side equal to
x + AC), we deduce that the side-length of S2 is also equal to y − x. Thus, S and
S2 are equal, and the theorem is proved.

Pythagorean Theorem using
translations γ
S1
δ
S
C β
S2 α
A
x B
α
β
y δ
S3
γ
210 4 A Synthesis
A A
C D B D B
C
A
A A
C D B
Fig. 4.3 Proof of the Pythagorean Theorem using spiral similarity
Proof 3 But the simplest proof is probably the one by spiral similarity shown in
Fig. 4.3. As above, D is the foot of the altitude from A, and let us reflect A over BC
to a point A . A spiral similarity with center B maps triangle BAD to BCA . The
similarity ratio of these triangles is BA/BC and the ratio of their areas is BA2 /BC 2 .
Another spiral similarity with center C maps triangle CAD to CBA . The
similarity ratio of these triangles is BA/BC and the ratio of their areas is CA2 /BC 2 .
But the triangles BAD and CAD add up to the triangle ABC which is congruent to
A BC. Thus,
BA2 CA2
2
+ = 1,
BC BC 2

We continue with another famous result.
Theorem 4.2 (Feuerbach) In every triangle, the nine-point circle is tangent to the
incircle and the excircles.
Proof Let ABC be a triangle, let ω9 be its nine-point circle, let I be the center and
r the radius of the incircle ω, and let Ia be the center and ra the radius of the excircle
ωa corresponding to the vertex A. We denote by D and Da the tangency points of
Fig. 4.4 Proof of

Feuerbach’s Theorem
ω9
I χ(ω9)
B D Da
C
A
L M
Ia
ω and ωa with the side BC, respectively. Finally, let A , M, L ∈ BC be such that
AA , AM, AL are the altitude and median from A and the angle bisector of BAC,
respectively. The proof can now be followed on Fig. 4.4.
There is a direct homothety of center A and an inverse homothety of center L that
map ω to ωa , and consequently
LD r AI
= = .
LDa ra AIa
The latter is further equal to the ratio of the distances from A to I D and Ia Da ,
hence to A D/A Da . From the equality,
LD A D
=
LDa A Da
we can derive
LDa + LD A Da − A D
= .
LDa − LD A Da + A D
212 4 A Synthesis
This is the same as
DDa 2A M
= .
2LM DDa
By the discussion from Sect. 2.2.1 II, we know that M is the midpoint of DDa , so
MD = DDa /2. The above equality of fractions yields
DM A M
= .
LM DM
Therefore, MA · ML = MD 2 . There is an inversion χ hidden in the figure, of

center M and radius MD, which maps A and L into each other.
Because the radius MD of the circle of inversion is orthogonal to the radius
I D of ω and the circle ω is orthogonal to the circle of inversion, hence by
Proposition 3.25, ω is invariant under the inversion. For the same reason, ωa is
invariant under the inversion. The nine-point circle ω9 on the other hand passes
through the midpoint M of BC, so χ (ω9 ) is a line that is parallel to the tangent to
ω9 at M. Since A ∈ ω9 , L = χ (A ) is on χ (ω9 ). We deduce that the image χ (ω9 )
of the nine-point circle is a line through L that is parallel to the line tangent to the
nine-point circle at M.
On the other hand, by Theorem 2.10, the nine-point circle is the image of the
circumcircle by an inverse homothety of center the centroid of ABC and of ratio
−1/2. This homothety maps A to M, and consequently it maps the line tangent
to the circumcircle at A to the line tangent to the nine-point circle at M. The
tangent to the circumcircle at A forms with AC an angle equal to ABC, and
consequently its reflection over the angle bisector AI of BAC is parallel to BC
(the antiparallel is mapped to the parallel). The line through L that is parallel to
is therefore just BC, and this is the common tangent of ω and ωa . The line through
L that is the reflection of BC with respect to the angle bisector AI (which is also
the line of centers of ω and ωa ) is therefore the other common tangent of the two
circles.
So χ (ω9 ) is a line tangent to ω = χ (ω) and ωa = χ (ωa ). Because inversion
preserves tangencies, ω9 is tangent to both ω and ωa . By symmetry, the nine-point
circle is tangent to the other two excircles, and the theorem is proved.

The geometric transformations mindset brings on a system of coordinates
associated with a (convex) cyclic quadrilateral that can be quite handy. If ABCD is
the cyclic quadrilateral and P is the intersection of the diagonals AC and BD, then
the triangles AP B and DP C are similar. The triangle DP C can be obtained from
AP B by an inverse homothety of center P (and ratio −k, for some k > 0) followed
by a reflection over the angle bisector of AP B (Fig. 4.5). We can therefore choose
a system of coordinates with P at the origin, such that the coordinate of A is 1,
while the coordinate of B is aω, for some a > 0 and |ω| = 1. Then the coordinates
of C and D are −ka and −kω, respectively. This is illustrated in Fig. 4.6.
P
A
B
C
B
B
P
P
A
A
Fig. 4.5 Constructing a cyclic quadrilateral from a triangle using homothety and reflection
Fig. 4.6 Coordinates for a − ka

cyclic quadrilateral aω
0
1
− kω
Let us show how this coordinate system simplifies the solution to a problem of
Titu Andreescu that was given at the USA Mathematical Olympiad in 1999.
Problem 4.1 Let ABCD be a cyclic quadrilateral. Prove that
|AB − CD| + |AD − BC| ≥ 2|AC − BD|.
Solution In this system of coordinates, we compute
AB = |1 − ωa| = |ω| · |ω − a| = |ω − a|, CD = k|ω − a|,

AD = |1 + kω| = |ω| · |ω + k| = |ω + k|, BC = a|ω + k|,
AC = 1 + ka, BD = a + k.
214 4 A Synthesis
We are left with proving the inequality
|1 − k| · |ω − a| + |1 − a| · |ω + k| ≥ 2|1 − a| · |1 − k|,
which is equivalent to
|ω − a| |ω + k|
+ ≥ 2.
|1 − a| |1 − k|
The distance from a point on the positive x-semiaxis to a point on the unit circle
is minimized when the point on the unit circle is at 1, which is a consequence, as it
can be seen on Fig. 4.7, of the fact that in a triangle the longer side is opposed to
the larger angle. So each of the terms on the left side of the last inequality is greater
than or equal to 1. The inequality is proved.

Let us present now a problem of Vyacheslav Viktorovich Proizvolov that
has appeared in the Russian journal Kvant (Quantum), whose solution combines
inversion and homothety.
Problem 4.2 Let ω1 , ω2 , ω3 , ω4 be four circles such that ωj is exterior tangent to
ωj +1 , j = 1, 2, 3, 4, (ω5 = ω1 ). Show that the four tangency points are either
concyclic or collinear.
Solution Let Aj be the tangency point of ωj and ωj +1 , j = 1, 2, 3, 4. There are
many possible configurations, some of which are illustrated in Fig. 4.8. What is
important is that for every j , the circles ωj −1 and ωj +1 are outside ωj , so the
tangency points of these two latter circles with ωj +2 are outside ωj .
We simplify the configuration by an inversion of center A1 , and use the standard
convention that the inverse of a geometric object is denoted by adding a dash to its
name. The circles ω1 and ω2 are mapped to two parallel lines ω1 and ω2 , which are
tangent to the circles ω3 and ω4 . Because the circles are exterior tangent, the point
A3 lies between the two lines. Two possible configurations are shown in Fig. 4.9.
Fig. 4.7 Minimizing the distance from a point on the positive x-semiaxis to a point on the unit
circle
ω4
A3
A4
ω4 ω3
ω1
A2 ω3
A4 A1
ω1
A3
A1 ω2
A2
A3
ω4
A2 A4
A1
ω2
ω2
ω3
ω1
Fig. 4.8 Concyclic tangency points
ω4
A2 ω2
ω2 A2
A3
A3
ω3
ω4
A4 ω1
A4 ω1
ω3
Fig. 4.9 Collinear tangency points
The inverse homothety of center A3 that maps ω3 to ω4 must map A2 to A4
because it maps tangents to tangents and parallel lines to parallel lines. Thus, A2
and A4 are collinear with the center of homothety. In other words, A2 , A3 , A4 are
collinear, and as inversion maps, a line to a line through the center of inversion
or a circle through the center of inversion, A1 , A2 , A3 , A4 , is either collinear or
concyclic.

Certainly, this problem can also be solved by an angle chasing argument. But
there are many possible configurations and the solution is case-dependent. The
combined use of inversion and homothety addresses all cases at once. The property
does not necessarily remain true if we relax the condition for the circles to only be
216 4 A Synthesis
tangent, not exterior tangent, as Fig. 4.10 shows. What goes wrong in the inverted
figure?
Our last example is the most elaborate; it is Problem 6 from the 2019 Interna-
tional Mathematical Olympiad, proposed by Anant Mudgal from India. During the
Olympiad, this problem has become the playground of many geometric transforma-
tions, all invited into the game to help connect somehow the many points that show
up in the configuration. We have collected some of the solutions, and present them
below.
Problem 4.3 Let I be the incenter of the acute triangle ABC with AB = AC.
The incircle ω of ABC is tangent to the sides BC, CA, and AB at D, E, and F ,
respectively. The line through D perpendicular to EF meets ω again at R. The
line AR meets ω again at P . The circumcircles of the triangles P CE and P BF
meet again at Q. Prove that the lines DI and P Q meet on the line through A
perpendicular to AI .
Solution 1 The situation described in the statement is shown in Fig. 4.11. We

present first the solution discovered by the coordinator Žarko Rand̄elović, which
uses spiral similarity. Both the line DI and the line through A that is perpendicular
to AI are easy to understand; the first is perpendicular to BC at D, and the second
is the exterior angle bisector of A. It is therefore natural to start by letting L be
the intersection of these lines and try to prove that P , Q, L are collinear. Note
additionally that the exterior angle bisector of A is the line that joins A to the

midpoint Y of the arc BAC of the circumcircle. Throughout this solution, we
assume that B > C, so that Fig. 4.12 is an accurate representation.
This solution will use a considerable amount of angle chasing, and so it is worth
computing the angles of the triangle DEF . Because AI and CI rotate by 90◦ to EF
and ED, respectively, the angle DEF is the acute angle between AI and CI . The
latter angle can be computed using the Exterior Angle Theorem in the triangle AI C,
and it is equal to A/2 + C/2 = 90◦ − B/2. Therefore, DEF = 90◦ − B/2,
and similarly DF E = 90◦ − C/2, EDF = 90◦ − A/2.
If we add to the figure the circumcircle of BI C, then it appears that Q lies on it
(see Fig. 4.12). This is indeed true, so let us prove it with the help of angle chasing.
Because B, F, Q, P are on a circle, BQP = BF P , and because C, E, Q, P are
on a circle, P QC = P EC. Combining these facts and using inscribed angles in
the circumcircles of BP F , CP E, and DEF , we obtain
Fig. 4.10 A counterexample

Fig. 4.11 Statement of

Problem 4.3
A
R E
I
Q
P
B D C
1 1
BQC = BQP + P QC = BF P + P EC = BP + P E
2 2
= F EP + P F E = 180◦ − EP F = 180◦ − EDF = 90◦ + A/2 = BI C.
This shows that BQI C is a cyclic quadrilateral, as claimed.

Next, note that ALD is the angle between AL and LD, which is the same as
the angle between AI and BC because AL and LD rotate by 90◦ to AI and BC.
And the angle bisector AI forms with the side BC an angle equal to A/2 + C.
Therefore,
ALD = A/2 + C.
Also, using inscribed angles in the incircle and the above equalities, we can compute
AP D = RP D = RF D = RF E + EF D = RDE + EF D
= 90◦ − DEF + EF D = 90◦ − (90◦ − B/2) + 90◦ − C/2
= 90◦ + B/2 − C/2 = A/2 + B.
Hence, ALD + AP D = A + B + C, so AP DL is a cyclic quadrilateral.

Proving that BQI C and AP LD are cyclic is one of two major steps toward our
goal of connecting P and Q to the rest of the configuration. The other step involves
a geometric transformation, and for that we let T and X be the second intersection
points of P Q and I D with the circumcircle of BI C, respectively.
218 4 A Synthesis
Fig. 4.12 Configuration with Y

the circumcircles of ABC
A
and BI C L
R E
I
Q
P
B C
D
We will show that there is a spiral similarity that takes
X → D, T → P , B → F, I → R, C → E, Y → A.
We proceed to Fig. 4.13, from which we read
F ER = F DR = 90◦ − DF E = C/2 = BCI,
and similarly EF R = CBI , so that the triangles BI C and F RE are directly

similar. By Proposition 2.21, there is a spiral similarity s that maps the triangle
BI C to the triangle F RE, that is, s(B) = F , s(I ) = R, s(C) = E. The spiral
similarity s is the one we have predicted.
Because I X is perpendicular to BC, and RD is perpendicular to EF , the point
X plays in the configuration consisting of the triangle BI C and its circumcircle the
same role that D plays in the configuration consisting of the triangle F RE and its
circumcircle. From this we deduce that s(X) = D.
Also, both triangles BY C and EAF are isosceles, and BY C = A = EAF ,
so they are directly similar. Hence, they are mapped one into the other by a spiral
Fig. 4.13 Proof that Y

X, T , B, I, C, Y are mapped
A
to D, P , F, R, E, A L
R E
F
I
Q
P
B C
D
similarity, and this spiral similarity maps BC to EF , so it coincides with s. Thus,

s(Y ) = A.
Finally, using inscribed angles in the circumcircles of P F B and BI C, we have
P EF = P F B = P QB = T QB = T CB.
Similarly, using the circumcircles of P CE and BI C, we obtain
P F E = T BC.
It follows that the triangles T BC and P F E are similar, the spiral similarity that
maps the first into the second is s, and hence s(T ) = P . Therefore, s maps the
points as specified.
Let us see how we can use this spiral similarity to complete the solution. Spiral
similarities preserve angles, so
Y T X = s(Y )s(T )s(X) = AP D = 180◦ − ALD = Y LX,

220 4 A Synthesis
where for the third equality, we have used the fact that AP DL is cyclic. Hence,
Y LT X is cyclic too. Using the fact that these two quadrilaterals are cyclic, and that
spiral similarities preserve angles, we can write
T LX = T Y X = s(T )s(Y )s(X) = P AD = P LD = P LX.
We deduce that T , P , L are collinear. But Q belongs to the line P T , so P Q passes

through L. The problem is solved.

Solution 2 Here is a solution that combines reflections over lines and circles,
which was found by Ilya Igorevich Bogdanov, member of the Problem Selection
Committee. The midpoint A where AI intersects EF and the point K where DI
intersects ω appear naturally in the figure. Drawing several situations, we may guess
that K, A , P are collinear (Fig. 4.14). The solution begins with the proof of this
fact.
First, notice that K is the image of R under the reflection over AI . Indeed
DRK = 90◦ , because KD is a diameter in ω, and so RK is parallel to EF .
And since AI is perpendicular to the chord EF , it is also perpendicular to the chord
RK; thus, it is the perpendicular bisector of RK. From here we deduce that P R and
P K are isogonal in the triangle P RK. But P R is symmedian in this triangle, by
Theorem 1.22, so P K is median, and therefore it passes through A .
We invert with respect to the incircle ω, and use the convention that the images
of the points bear dashes (Fig. 4.15). There are some points that are fixed by
the inversion: D, E, F, P , R, K. By Proposition 3.3, A is the image of A, which
explains the notation. For the same reason, B is the midpoint of DF and C is the
Fig. 4.14 K, A , P are

collinear
A
L
R K E
A
F
I
Q
P
B D C
Fig. 4.15 Proof that

P Q I = P L I K
F
P1
R
L A
ω E
G I
B
Q P
C
D1
midpoint of DE. Some cyclicities are automatic: (E, P , C , Q ), (F, P , B , Q ),

(I, R, P , A ). Additionally, B C is parallel to EF , being midline in DEF .
The collinearity of P , Q, L is equivalent to the cyclicity of P , Q , L , I . To avoid
configuration issues, we will work with directed angles modulo 180◦ . We are left
with showing that
P Q I = P L I.
The latter angle is the same as P L D.

The point L is easy to locate; it is the projection of A onto I D, because
A L I = LAI = 90◦ (Theorem 3.2). Notice that this fact combined with
DP A = DP K = 90◦ implies that A L DP is cyclic. So we obtain P A D =
P L D. This reduces the problem to showing that
P Q I = P A D.
Locating the point Q , which lies at the intersection of the circumcircles of F P B

and EP C , is slightly more difficult. We use Fig. 4.15, from where we guess that
Q ∈ B C . This is equivalent to the fact that B, Q, I, P are concyclic, which was
proved in the first solution. Or we can compute directly with inscribed angles
B Q P = B F P = DF P = DEP = C EP = C Q P ,
222 4 A Synthesis
a faster argument that proves Q ∈ B C . We can locate Q even better. Introduce

the point D1 ∈ ω such that DD1 EF . Then D1 R is the reflection of DK over A I ,
so D1 and R are diametrically opposite in ω. We claim that Q ∈ P D1 .
To prove this, introduce also the point P1 ∈ ω such that P P1 EF . Then P1 R is
the reflection of P K over A I , so A ∈ P1 R. Then
(P P1 , Q P ) = (B Q , Q P ) = B Q P = DEP = DP1 P = P1 P D1
= (P P1 , D1 P ),
where the third equality was checked above, and the last follows from the symmetry
with respect to A I . Hence, Q ∈ P D1 , as claimed. It should be noted that
Q ∈ P D1 is equivalent to the fact that P , Q, I, D1 are concyclic in the original
configuration, but this is difficult to check without inversion.
Now we can complete the solution. Let A D1 meet B C at G. Because B C is
midline in DEF , G is the midpoint of A D1 . Consequently, GI is midline in the
triangle D1 RA . We thus obtain
D1 I G = D1 RA = D1 RP1 = P KD,
where for the last equality, we use reflection over A I . Now using the midline GI in
the triangle D1 RA as well as inscribed angles in the circle ω and the circumcircle
of P B F , we have
D1 I G = P KD = P F D = P F B = P Q B = P Q G
showing that D1 , G, I, Q are concyclic. This yields
P Q I = D1 Q I = D1 GI = D1 A R = D1 A P1 .
Using the reflection over A I , we deduce that D1 A P1 = P A D. Combining, we

obtain that P Q I = P A D, which is the equality that we had to prove.

Solution 3 And here is a solution that combines inversion and homothety, also due
to Ilya Igorevich Bogdanov. Like in the previous solution, we introduce the midpoint
A of EF , which is the image of A through the inversion over ω, K which is the
second intersection point of DI with ω, and L which is the intersection point of the
external bisector of A with DI , and with this auxiliaries aim to prove that P , Q, L
are collinear.
Let DA intersect ω for the second time at S, as shown in Fig. 4.16. Let us take
a look at the cyclic quadrilateral SKDP with Theorem 3.33 in mind. First, note
that EF is the polar of A with respect to ω, and since the polar of A with respect
to ω is perpendicular to AA , La Hire’s Theorem (Theorem 3.31) implies that the
polar of A must be AL. Note that A is the intersection of P K and DS (see the
second solution), so by Theorem 3.33 the lines P S and DK intersect on the polar
of the point A , which is AL. But DK intersects AL at L, so L is the intersection
Fig. 4.16 The construction

of S
A
L
S
R K E
A
F
I
Q
P
B D C
of P S and DK. We deduce that P , L, S are collinear, and the problem is reduced to
showing that P , Q, S are collinear, an easier task since S lies on ω.
Now we perform the inversion with respect to ω. As explained in the second
solution, the inverses A , B , C of A, B, C are the midpoints of EF , DF , and DE,
respectively, and Q is at the intersection of the circumcircles of F B P and EC P
and lies on B C . We maintain the convention about denoting inverses by adding
dashes.
Let T be the second intersection point of B C with the circumcircle of Q P I
(Fig. 4.17). When working with directed angles modulo π , we can phrase the angle
chasing in terms of rotating lines (as explained in Chap. 1); let us adopt this style
here. On the one hand I, T , Q , P are concyclic, and on the other hand, E, P , C , Q
are concyclic, so we can write
(T I, I P ) = (T Q , Q P ) = (Q C , Q P ) = (EC , EP ) = (ED, EP ).
Furthermore, because K, E, P , D are concyclic,
(ED, EP ) = (KD, KP ) = (I P , KP ),
the latest because I KP is isosceles. Combining these equalities, we obtain

(T I, I P ) = (I P , KP ); therefore, T I and KP are parallel.
Consider the homothety of center D and ratio 2. It maps I to K, and hence it
maps the line I T to KP . Also B C is mapped to EF . Consequently, T is mapped
to the intersection of EF and KP , which is A . So T ∈ DS. Now using the fact that
K, S, P , D lie on a circle, we deduce that (SD, SP ) = (KD, KP ), and we can
read above that the latter is equal to (T Q , Q P ). We conclude that I, T , Q , P , S
224 4 A Synthesis
Fig. 4.17 The construction K

of T and angle chasing F S
ω E
I
B
Q P
T C
lie on a circle, in particular S is on the circumcircle of I Q P . And this is equivalent

to the fact that P , Q, S are collinear, as desired.

Solution 4 The fourth solution that we present was discovered by Régis Prado
Barbosa, member of the Brazilian delegation, and is based on inversion and spiral
similarity. This solution can be followed on Fig. 4.18. It starts with the construction
of the points L, A , R, and S as before, and, like in the third solution, proves that
B, Q, I, C are concyclic, and by proving that P , S, L are collinear, reduces the
problem of showing that P , Q, L are collinear to that of showing that P , Q, S are
collinear. Again we can assume without loss of generality that B > C, so that
the figure is as drawn.
Next, we introduce X as the second intersection point of AD and ω, Y as the
intersection of DR and EF , and Z as the second intersection point of the circle
passing through B, Q, I, C with QP (point T in the first solution). For simplicity,
let α = P F E, β = P EF, γ = SDE, A = BAC. There are some automatic
angle equalities obtained using inscribed angles:
CBZ = CQZ = CQP = CEP = EF P = α,

BCZ = BQZ = BQP = BF P = F EP = β,
and we have seen in the first solution that
α + β = 180◦ − EDF = 180◦ − (90◦ − A/2) = 90◦ + A/2.

Fig. 4.18 Proof that P , Q, S

are collinear
A L
O
X S
E
R
F β
Y
α I
β Q
γ
P
B β C
α D
One more angle can be computed easily. By the symmedian construction (Theo-
rem 1.22), DA is a symmedian in the triangle DEF . Also the line DS is the same
as DA , hence is a median in the same triangle. Thus, γ = SDE = XDF .
Let us focus for a moment on the quadrilateral P RXD. By Theorem 3.33, the
polar of the point that is the intersection of the diagonals DR and P X passes through
A, thus by La Hire’s Theorem (Theorem 3.31), the polar of A passes through this
point. But the polar of A is EF , and DR intersects EF at Y , so Y is the intersection
of DR and P X. Therefore, P , Y, X are collinear.
A few more computations of angles before we switch to geometric transforma-
tions:

P I X = P I F + F I X =P F + F X= 2 P EF + 2 XDF = 2β + 2γ ,
and since the triangle I P X is isosceles, we have
180◦ − 2β − 2γ
IPY = IPX = = 90◦ − β − γ = α − A/2 − γ .
2
226 4 A Synthesis
Also I EY = I EF = I AE = A/2, where for the second equality we used the

right triangle EAI . Combining with the previous computation, we obtain
I P Y + I EY = α − γ .
The triangles P F E and ZBC have equal angles, so they are directly similar.
By Proposition 2.21, there is a spiral similarity s that takes the triangle P F E to
ZBC. Since s takes EF to BC, Theorem 2.22 implies that the center O of s is
the intersection of the circumcircles of AEF and ABC (see Fig. 4.18). But spiral
similarities come in pairs (Theorem 2.25), so there is a second spiral similarity
s that maps the triangle OP Z to OEC. Hence, (P Z, EC) = (OP , OE) =
P OE.
And again we invert over ω. This inversion takes the circumcircle of AEF to the
line EF and the circumcircle of ABC to a circle that passes through the midpoints
A , B , C of EF , DF , and DE, respectively. This is the nine-point circle of DEF .
The center O of s, which is the intersection of the circumcircles of AEF and ABC,
is mapped to the intersection of the nine-point circle with EF other than A . And
this is the leg Y of the altitude DY . By Theorem 3.2, P OI = Y P I and I OE =
I EY . Therefore,
P OE = P OI + I OE = Y P I + I EY = α − γ .
Consequently, (P Z, EC) = α − γ . If we show that P S forms the same angle

with EC, we are done. And indeed, let W be the intersection of P S with AC. Then
W P E = SP E = SDE = γ and the Exterior Angle Theorem in the triangle
W P E shows that
P W C = CEP − W P E = α − γ .
The problem is solved.

Solution 5 The last solution that we present has been put together by the US
contestants Luke Robitaille and Brandon Wang, by the member of the US delegation
Evan Chen, and by Michael Ren. It uses a Möbius transformation φ = χ2 ◦χ1 , where
χ1 is the inversion about ω and χ2 is an inversion of center P and arbitrary radius.
We mark by a dash the images of points under χ1 and by a star the images under φ.
As in the second solution, we let K be the intersection of I D with ω and let
A be the midpoint of EF ; we know that A ∈ P K; hence, A∗ ∈ P K ∗ . Note that
R ∗ = χ2 (R) and K ∗ = χ2 (K) are both on φ(ω) = χ2 (ω) = E ∗ F ∗ . We claim that
P K ∗ = P A∗ is a symmedian in the triangle P E ∗ F ∗ . Indeed, by Proposition 1.23,
the cyclic quadrilateral P ERF is harmonic, so, in coordinates,
p−e r −e
: = −1.
p−f r −f
As Möbius transformations preserve cross-ratios, and as P is mapped to ∞, we have
r ∗ − e∗ ∞ − e∗ r ∗ − e∗
= : = −1.
r∗ − f ∗ ∞ − f ∗ r∗ − f ∗
So R ∗ is the midpoint of E ∗ F ∗ , that is, P R ∗ is a median in P E ∗ F ∗ . But the lines

P R ∗ and P K ∗ (which are the same as P R and P K) form equal angles with P F ∗
and P E ∗ , respectively, showing that P K ∗ is a symmedian in the triangle P E ∗ F ∗ .
Also, the first inversion maps A to A , and the second inversion maps A to A∗
which is on the image of EF through this second inversion. We deduce that A∗ is
at the intersection of the line P K ∗ with the circumcircle of P E ∗ F ∗ .
Because D and K are diametrically opposite in ω and they are invariant under
χ1 , and because the lines P D and P K are invariant under χ2 , D ∗ P K ∗ = 90◦ .
As above, P B ∗ and P C ∗ are symmedians in the triangles P D ∗ F ∗ and P D ∗ E ∗ ,
respectively, and the quadrilaterals P D ∗ B ∗ F ∗ and P D ∗ C ∗ E ∗ are harmonic.
Additionally, as Q is the intersection of the circumcircles of BP F and CP E, Q∗
is the intersection of the lines C ∗ E ∗ and B ∗ F ∗ , which are the images of these two
circles.
Let us add to the picture the point L where DI = DK intersects the exterior
angle bisector of A; if L = χ1 (L), then A L D = LAD = 90◦ ; thus, L is the
projection of A on DK. Applying Theorem 3.2 to χ2 and working with directed
angles modulo 180◦ , we obtain
K ∗ L∗ D ∗ = K ∗ L∗ P + P L∗ D ∗ = P KL + L DP = P KD + KDP
= KP D = 90◦ ,
where the fourth equality uses the sum of (directed angles) in the triangle KDP .
Also, P A L D is cyclic ( DP A = DL A = 90◦ ), so L∗ ∈ A∗ D ∗ . We deduce
that L∗ is the projection of K ∗ onto D ∗ A∗ .
In the perspective of the inversion χ1 , showing that that P , Q, L are collinear
is equivalent to showing that P , Q , I, L are concyclic. In the perspective of the
second inversion, showing that P , Q , I, L are concyclic is equivalent to showing
that I ∗ = χ2 (I ), Q∗ , L∗ are collinear. Note that I ∗ is the reflection of P over EF
(Proposition 3.7). Dropping the stars, we have reduced the problem to the following:
Problem In the triangle P EF , the P -symmedian meets the side EF at K and the
circumcircle at A. Let D be a point on the line EF such that DP K = 90◦ and
let L be the foot of the perpendicular from K to AD. Denote by I the reflection of
P over EF . Construct also the cyclic harmonic quadrilaterals P DCE and P DBF .
Prove that the lines EC, F B, and LI are concurrent.
The solution to this new problem can be followed on Fig. 4.19. Suppose that the
line through A perpendicular to EF meets the line EF at W and the circumcircle
of P EF again at Z.
228 4 A Synthesis
D E K
F
W
C L A
I
Fig. 4.19 Proof that CE, BF , and I L are concurrent
Because DP A = DW A = 90◦ , the points D, P , W, A are concyclic, so

EDP = W DP = W AP = ZAP . But because Z, P , E, A are concyclic,
the latter angle is equal to ZEP . Hence, EDP = ZEP , showing that ZE is
tangent to the circumcircle of P DCE. Similarly, ZF is tangent to the circumcircle
of P DBF .
The orthogonalities DP ⊥KA, KL⊥DA, AW ⊥DK show that P LW is the
orthic triangle of DKA. So W D is the angle bisector of P W L. Consequently,
the line W L is the reflection of the line W P over EF , and so I ∈ W L.
The lines W Z, W P , W D, W I form a harmonic pencil, meaning that they
determine on any line that intersects them four points whose cross ratio is −1.
Indeed, in view of what was discussed in Sect. 3.1.5, we have to check that one
particular line intersects this pencil at four points that form a harmonic division, and
this is the line P I , which intersects W D at the midpoint of DI and W Z at the point
at infinity.
We will discover two more harmonic pencils in the figure with the aid of a lemma.
Lemma 4.3 Let ABCD be a cyclic quadrilateral and let M be a point on the
circumcircle. Then the quadrilateral ABCD is harmonic if and only if the lines
MA, MB, MC, MD form a harmonic pencil.
Proof Consider an inversion of center M. Then the four lines in question are
mapped into themselves, while the circumcircle of the quadrilateral becomes a line
parallel to the tangent at M (Fig. 4.20). As Theorem 3.18 shows, the cross-ratio of
the points A, B, C, D is equal to that of their images A , B , C , D . So A, B, C, D
form a harmonic quadrilateral if and only if A , B , C , D form a harmonic division
on the line that is the image of the circumcircle. But that is what characterizes the
pencil as being harmonic. It should be noted that when A = M, then the line AM
becomes the tangent to the circle at A.

As a consequence of Lemma 4.3, EZ, EP , ED, EC form a harmonic pen-
cil (P DCE is harmonic and EZ is tangent to its circumcircle) and so do
F Z, F P , F D, F B (P DBF is harmonic and F Z is tangent to its circumcircle).
Examining how these pencils intersect the line P Z, we notice that three of the
intersection points are the same, namely, P , Z, and the point where P Z intersects
EF . Because all three pencils determine harmonic divisions on P Z, it follows that
the fourth point is the same. This means that P Z, W I, EC, F B meet at one point.
But since L ∈ W I , W I = I L, so EC, F B, LI meet at one point, and the problem
is solved.
M
D
A D
B C
C
A
B
Fig. 4.20 Harmonic quadrilaterals yield harmonic pencils

230 4 A Synthesis
4.1.2 Some Problems
Discover which geometric transformations solve the following problems. Let us

point out that for each of these problems, we were able to write either several solu-
tions that employ different geometric transformations or one solution that combines
geometric transformations of different types. These types of transformations are
highlighted in italics in the solutions printed at the back of the book, in order to
emphasize their diversity.
Can you find more than one solution, and can you use more than one technique?
Which is the most elegant solution?
222 On the sides of a triangle ABC construct in the exterior the equilateral
triangles ABC1 , BCA1 , and CAB1 .
(a) (Napoleon’s Theorem) Prove that the centers of the triangles ABC1 , BCA1 , and
CAB1 form an equilateral triangle.
(b) Prove that the segments AA1 , BB1 , and CC1 are equal and their lines of support
are concurrent.
223 (Ptolemy’s Theorem) Show that in a cyclic quadrilateral ABCD
AB · CD + AD · BC = AC · BD.
224 Let A, B, C, D be four points on the circle ω, in this order, and let S be a point
inside ω such that SAD = SCB and SDA = SBC. The angle bisector of the
angle ASB intersects the circle ω at P and Q. Prove that SP = SQ.
225 Let A1 A2 A3 be a scalene triangle, and, for 1 ≤ i ≤ 3, let Mi be the midpoint
of the side opposite to Ai , and let Ti be the point where this side touches the incircle.
Let Si be the reflection of Ti over the angle bisector of Ai . Show that the lines
Mi Si , 1 ≤ i ≤ 3, are concurrent.
226 Let OA1 B1 , OA2 B2 , and OA3 B3 be three equilateral triangles that share the
vertex O. Prove that the midpoints of B1 A2 , B2 A3 , and B3 A1 form an equilateral
triangle.
227 Two circles 1 and 2 of radii r1 and r2 , respectively, with r1 < r2 , are exterior
tangent. The line t1 is tangent to 1 at A and to 2 at D. The line t2 is parallel to
t1 , is tangent to 1 , and intersects 2 at E and F . A line t3 passes through D and
intersects the line t2 at B and the circle 2 at C (with B and C different from E and
F ). Prove that the circumcircle of the triangle ABC is tangent to the line t1 .
228 Let ABCD be a quadrilateral inscribed in the circle ω, with AC⊥BD. Let E
and F be the reflections of D over the lines BA and BC, respectively, and let P be
the intersection point of the lines BD and EF . Suppose that the circumcircle of the
triangle EP D meets ω at D and Q, and the circumcircle of the triangle F P D meets
ω at D and R. Show that EQ = F R.
229 On the sides AB and AC of the triangle ABC, construct in the exterior the
equilateral triangles ABN and ACM. Let P , Q, R be the midpoints of BC, AM,
and AN, respectively. Prove that the triangle P QR is equilateral.
230 The points A, B, C, D are chosen such that C, D are on the same side of the
line AB, AC · BD = AD · BC, and ADB = 90◦ + ACB. Find the ratio
AB · CD
AC · BD
and prove that the circumcircles of ACD and BCD are orthogonal.
231 Let A be one of the two intersection points of the circles ω and ω , whose
centers are O and O , respectively. The common tangents touch ω at P and Q and
ω at P and Q , with P and P on the same side of OO . Let M and M be the
midpoints of the segments P Q and P Q . Prove that OAO = MAM .
232 Let ABC be an acute scalene triangle, and let H and O be its orthocenter and
circumcenter, respectively. The line OA crosses the altitudes from B and C of the
triangle ABC at P and Q, respectively. Show that the circumcenter of the triangle
H P Q lies on one of the medians of the triangle ABC.
233 (a) Consider five points A, B, C, D, E such that ABCD is a parallelogram
and BCED is a cyclic quadrilateral. A line through A intersects the line BC at K
and the side DC at L so that EL = EK = EC. Prove that is the angle bisector of
BAD.
(b) Conversely, let the bisector of the angle BAD of the parallelogram ABCD
intersect the lines BC and CD at K and L, respectively. Prove that the circumcenter
of the triangle CKL lies on the circumcircle of the triangle BCD.
234 Let ω1 and ω2 be two circles that intersect at A and B. A line passing through
A intersects ω1 and ω2 at C and D, respectively. Let M be the midpoint of the

segment CD, P the midpoint of the arc BC that does not contain A, and Q the

midpoint of the arc BD that does not contain A. Prove that P MQ = 90◦ .
235 (a) (Steiner’s Theorem) Prove that the Simson line of a point on the circum-
circle of a triangle passes through the midpoint of the segment that joins the point
with the orthocenter of the triangle.1
(b) (Lemoine’s Theorem) Consider for each vertex of a cyclic quadrilateral its
Simson line with respect to the triangle formed by the other three vertices. Show
that the four Simson lines obtained this way intersect at one point.
1 The Simson line of a point on the circumcircle of a triangle is the line determined by the
projections of the point onto the sides.

232 4 A Synthesis
236 Let ABCD be a cyclic quadrilateral, and let M and N be the midpoints of the
sides AB and CD, respectively. Let E be the point of intersection of the lines AD
and BC and let F be the point of intersection of the diagonals AC and BD. Let also
P and Q be the projections of F onto BC and AD, respectively.
(a) Prove that

MN 1 AB CD
= − .
EF 2 CD AB
(b) Prove that P Q is perpendicular to MN .

237 The diagonals AC and BD of a quadrilateral ABCD are perpendicular.
Four squares, ABEF , BCGH , CDI J , and DAKL, are erected externally on
its sides. The intersection points of the pairs of straight lines CL, DF ; DF, AH ;
AH, BJ ; BJ, CL are denoted by P1 , Q1 , R1 , S1 , respectively, and the intersection
points of the pairs of straight lines AI, BK; BK, CE; CE, DG; DG, AI are
denoted by P2 , Q2 , R2 , S2 , respectively. Prove that the quadrilaterals P1 Q1 R1 S1
and P2 Q2 R2 S2 are congruent.
238 Let ABCD be a parallelogram that is not a rhombus. The point Q is the
intersection of the reflections of the lines AB and CD across the diagonals AC
and BD, respectively. Prove that Q is the center of the spiral similarity that maps
the segment AO to OD, where O is the center of the parallelogram.
239 Let ABC be a triangle and let M be the midpoint of the side BC. The circle
centered at M and of radius MA intersects a second time the lines AB and AC at B
and C , respectively. Let D be the point where the tangents to this circle at B and
C intersect. Prove that the perpendicular bisector of the side BC passes through the
midpoint of the segment AD.
240 Let A1 A2 A3 be a scalene triangle, and let I be its incenter. Let Ci be the
smaller circle through I that is tangent to the sides Ai Ai+1 and Ai Ai+2 , i = 1, 2, 3
(where indices are taken modulo 3). Let Bi be the second point of intersection of
Ci+1 and Ci+2 , i = 1, 2, 3. Prove that the circumcenters of A1 B1 I , A2 B2 I , and
A3 B3 I are collinear.
241 Let be a circle, let AB be a diameter, and let be a line perpendicular to
AB and exterior to the circle so that B is between A and . Let C be a point of ,
different from A and B, and let D be the intersection point of the lines AC and .
The line DE is tangent to the circle at E so that E and B are on the same side of
the line AC. The line BE intersects the line at F , and the line AF intersects the
circle at G. Prove that the reflection of G over AB is on the line CF .
242 Let ABC be a triangle, let P , Q, and R be the tangency points of the incircle
with the sides BC, AC, and AB, respectively, and let I be the incenter. The lines
P Q and AB intersect at D and the lines QR and BC intersect at E. Prove that the
circumcircles of QEP , QDR, and QBI have a second common point.
243 Let R and S be distinct points on the circle , and let t denote the tangent
line to at R. The point T is the reflection of R over S. A point J is chosen on the

smaller arc RS of so that the circumcircle of the triangle J ST intersects t at
two different points. Denote by A the common point of and t that is closest to R.
The line AJ meets again at K. Show that the line KT is tangent to the circle .
244 Let ABC be a triangle and let I and O denote its incenter and circumcenter,
respectively. Let ωA be the circle through B and C that is tangent to the incircle of
the triangle ABC; the circles ωB and ωC are defined similarly. The circles ωB and
ωC meet at a point A different from A; the points B and C are defined similarly.
Prove that the lines AA , BB , and CC are concurrent at a point on the line I O.
245 Two circles 1 and 2 touch internally the circle at M and N, respectively,
and the center of 2 is on 1 . The common chord of the circles 1 and 2 intersects
at A and B. The lines MA and MB intersect 1 a second time at C and D,
respectively. Prove that 2 is tangent to CD.

246 Let ABC be a given triangle. Let X be a variable point on the arc AB of the
circumcircle of ABC that does not contain C, and let O1 and O2 the incenters
of the triangles CAX and CBX. Prove that the circumcircle of XO1 O2 meets the
circle at a fixed point.
247 Let M be an arbitrary point in the circumcircle of ABC. Suppose that the
lines through M that are tangent to the incircle of ABC meet BC at K1 and K2 .
Prove that the circumcircle of MK1 K2 meets the circumcircle of ABC again at the
tangency point with the A-mixtilinear incircle.
248 A circle centered at the vertex A of the triangle ABC intersects the side BC
at D and E so that B, D, E, C are distinct and lie on the line BC in this order. Let F
and G be the intersection points of with the circumcircle , so that A, F, B, C, G
are on in this order. Let K and L be the points at which the circumcircles of the
triangles BDF and CEG intersect the second time the sides AB and AC. Suppose
that the lines F K and GL are distinct and intersect at the point X. Prove that X lies
on the line AO, where O is the circumcenter of ABC.
249 Fix a circle , a line tangent to , and another circle disjoint from such
that and lie on opposite sides of . The tangents to from a variable point X
on cross at Y and Z. Prove that, as X traces , the circumcircle of the triangle
XY Z is tangent to two fixed circles.
250 Let ABCD be a cyclic quadrilateral whose diagonals AC and BD intersect
at E. Show that the Euler lines of the triangles AEB, BEC, CED, and DEA are
concurrent.
251 For every triangle T that is not right, we will denote by H (T ) its orthic triangle
(whose vertices are the feet of the altitudes of T ). Starting with a triangle T0 , we
define recursively Tn+1 = H (Tn ), n ≥ 0. For a given n > 1, find how many
triangles T0 , not similar to each other, exist such that Tn is similar to T0 .
234 4 A Synthesis
4.2 A Story of Complete Quadrilaterals
Complete quadrilaterals have their place in two-dimensional real projective geom-

etry, but we can already establish a number of beautiful facts about them with our
own tools. And we will see all the stars of this book performing on the stage of this
rich configuration.
We recall that a complete quadrilateral is any set of four lines, no three
passing through the same point, and no two parallel.2 An example of a complete
quadrilateral, determined by the lines 1 , 2 , 3 , 4 , is shown in Fig. 4.21. We agree
to define the points Aij as the intersections on the lines i and j , 1 ≤ i < j ≤ 4,
and call them the vertices of the complete quadrilateral.
√
4.2.1 Miquel’s Theorem and bc Inversion
√
We apply a bc inversion to. . . a quadrilateral! We have seen in Chap. 2 that
complete quadrilaterals have a lot to do with spiral similarities, Miquel’s Theorem
(Theorem 2.24) states that all four circumcircles of the triangles determined by three
of four lines 1 , 2 , 3 , 4 , no two of them parallel, pass through a common point: the
Miquel point M. In that context, we have uncovered the following transformation-
focused version of this result:
Theorem 4.4 (Miquel’s Theorem, Rephrased) Let 1 , 2 , 3 , 4 be the lines that
define a complete quadrilateral, and let Aij be its vertices. Then there is a spiral
similarity that maps Aij to Aik and Amj to Amk , for {i, j, k, m} = {1, 2, 3, 4},
Fig. 4.21 A complete

quadrilateral
A23
A24
A34
2 4
A14
A12 A13 1
2 The latter condition is dropped in real projective geometry, since parallelism does not exist in real
projective geometry.
4.2 A Story of Complete Quadrilaterals 235
whose center is the intersection of the four circumcircles of the triangles determined
by three of the four lines.
√
On the other hand, Proposition 3.40 establishes a connection between bc inver-
sions and spiral similarities. In fact, putting
√ together the aforementioned proposition
and Miquel’s Theorem gives rise to bc inversion on complete quadrilaterals.
Proposition 4.5 (Inverting Complete Quadrilaterals) Let 1 , 2 , 3 , and 4 be
the four lines of a complete
√ quadrilateral, whose vertices are Aij , and let M be its
Miquel point. Then the bc-inversion determined by M in the triangle MA12 A34
maps A13 to A24 and A14 to A23 .
√
Proof Let φ be the aforementioned bc inversion. Combining the proof of
Miquel’s Theorem given in Chap. 2 with Proposition 3.40, (iii), we find that there
is a spiral similarity centered at M that maps A12 to A14 and A23 to A34 , which
implies that φ(A14 ) = A23 and φ(A23 ) = A14 . Analogously, there is a spiral
similarity, also centered at M, that maps A12 to A13 and A24 to A34 , which implies
that φ(A13 ) = A24 and φ(A24 ) = A13 . This proof can be followed on Fig. 4.22.
A quick corollary of this proposition is that the diagram formed by the lines 1 ,
2 , 3 , and 4 , and the four circles determined by three of these four lines, is fixed
by the Möbius transformation φ. Lines and circles are swapped; more specifically,
each line is mapped to the circle corresponding to the other three lines.
A consequence of this result is that one can assume that the complex coordinates
of four vertices of the complete quadrilateral are a, b, 1/a, and 1/b, with the Miquel
point being at the origin of the coordinate system. The coordinates of the other two
vertices, with less elegant formulas, are
(ab − ab)(a − b)
x= and 1/x.
(ab − 1)(ab − ab)
4.2.2 Some Classical Results
We will now prove some well-known results about complete quadrilaterals.

Theorem 4.6 If the lines 1 , 2 , 3 , and 4 determine a complete quadrilateral with
Miquel point M, then
(i) the four orthogonal projections of M onto all four lines i are collinear;
(ii) (The Orthocentric Line) The orthocenters of all four triangles determined by
the four lines are collinear;
(iii) the circumcenters of all four triangles determined by the four lines are
concyclic, and M lies on this circle.
236 4 A Synthesis
A13
A14
3
A12 A34
M
2
4
A23
A24
√
Fig. 4.22 Complete quadrilaterals and bc inversion. The common angle bisectors are repre-
sented by dashed lines
Proof (i) Any two triangles formed in the complete quadrilateral share two sides,
so they share the Simson line of the Miquel point. So the Simson line of the Miquel
point is the same for all four triangles, and this is the line of the four orthogonal
projections.
(ii) By Steiner’s Theorem (Problem 235), the orthocenters are on the line that is the
image of the above Simson line by a homothety with center M and ratio 2. So the
orthocenters are collinear.
(iii) Consider the same homothety centered at M and with ratio 2, and let Pi be the
image of the projection of M onto i under this homothety, i = 1,√2, 3, 4; in other
words, Pi is the reflection of M across i . By Proposition 3.7, the bc inversion φ
that fixes the configuration maps the points Pi to the circumcenters Oi of Ti ; since
the points Pi are collinear, the points Oi are concyclic; moreover, the center M of φ
lies on this circle.

Here is another result focused on transformations.
Theorem 4.7 Let 1 , 2 , 3 , and 4 be four lines that form a complete quadrilateral,
and let M be the Miquel point. Denote by Ti the triangle formed by all lines except
i , and by Oi the circumcenter of Ti . Then there exists a spiral similarity, centered
at M, that takes the triangle Oj Ok Om to the triangle Ti , where {i, j, k, m} =
{1, 2, 3, 4}.
Proof Let Oi be the circumcenter of Ti . We prove that there is a spiral similarity
with center M that takes O1 O2 O3 to T4 . Specifically, we show that O1 O2 O3 and
A23 A13 A12 are similar.
Consider the spiral similarity with center M that takes A13 A14 to A23 A24 . It also
maps the circumcircle of MA13 A14 , to the circumcircle of MA23 A24 ; in particular,
O2 is mapped to O1 . So we have the automatic (spiral) similarity s between the
triangles MO1 O2 and MA23 A13 . This transformation has center M, and is such
that s(O1 ) = A23 and s(O2 ) = A13 . We can prove in the same way that triangles
MO1 O3 and MA23 A12 are similar; furthermore, it follows that the spiral similarity
s with center M that takes O1 to A23 also takes O3 to A12 . Since a spiral similarity is
determined by its center and the image of some other point, s = s. So s(O3 ) = A12
and we are done.

For the next result, we define the diagonals of a complete quadrilateral to be the
line segments that connect two of the vertices that are not on the same line i .
Theorem 4.8 (The Newton-Gauss Line) The midpoints of the three diagonals of
a complete quadrilateral are collinear.
Proof We will use a fact that is proved easily:
Lemma 4.9 In a parallelogram XY ZW , a point M is on the diagonal XZ if and
only if the two parallelograms that have Y and W as vertices and are determined
by the lines through M parallel to the sides have equal areas.
With this result at hand, let 1 , 2 , 3 , and 4 be the four lines, and, with the usual
conventions about vertices, let N1 , N2 , and N3 be the midpoints of A12 A34 , A13 A24 ,
and A14 A23 , respectively. Without loss of generality, we may assume that the four
lines form the configuration from Fig. 4.23. Through the points A13 , A34 , A24 , A23 ,
and A14 , take parallels to the sides A12 A13 and A12 A24 ; let Q, R, S, T , U, V be the
points shown in the figure.
The homothety of center A12 and ratio 2 maps:
• N1 to A34 ,
• the midpoint N2 of the diagonal A13 A24 of the parallelogram A12 A13 QA24 to
the opposite vertex Q, and
238 4 A Synthesis
Fig. 4.23 The homothety A23 S T R

proving the existence of the
Newton-Gauss line
A24 U
Q
A34
V
N3
N2
N1
A12 A13 A14
A23 S T R A23 S T R
A24 U A24 U
Q Q
V V
A34 A34
A12 A13 A14 A12 A13 A14
A23 S T R
Q
A24 U
V
A34
A12 A13 A14
Fig. 4.24 The area argument proving the existence of the Newton-Gauss line
• the midpoint N3 of the diagonal A14 A23 of the parallelogram A12 A14 RA23 to the
opposite vertex R.
As homothety maps lines to lines, the collinearity of the midpoints N1 , N2 , N3 is
equivalent to the collinearity of their images A34 , Q, and R.
Applying Lemma 4.9 to the parallelogram A34 V RS, we deduce that the fact that
Q is on the diagonal A34 R is equivalent to the equality of the areas of the shaded
parallelograms from the top left configuration in Fig. 4.24.
We can apply the direct implication of Lemma 4.9 to the parallelogram
A12 A13 T A23 because A34 is on the diagonal A13 A23 , so the shaded parallelograms
from the top right configuration in Fig. 4.24 have equal areas. Also, in the
parallelogram A12 A14 U A24 , A34 is on the diagonal A14 A24 , and so, by the
same lemma, the shaded parallelograms from the configuration at the bottom of
Fig. 4.24 have equal areas. The two equalities of areas that we have proved imply,
by subtraction, the equality of areas for the shaded parallelograms on the top left,
and this implies that the points A34 , Q, R are collinear. The theorem is proved.
We have seen in Theorem 4.6 that the orthocenters of the four triangles formed in
a complete quadrilateral are collinear; they determine what is called the orthocentric
line. A proof of this collinearity based on power-of-a-point is also possible, and
it proceeds as follows. Let Hi be the orthocenter of Ti , and let Apq Br be the
altitudes.We know that Apq Hi ·Hi Br is the same for all three choices of indices. But
these products are the powers of Hi with respect to the three circles that have the
diagonals as diameters. Consequently, the orthocenters have the same power with
respect with the three lines, so on the one hand, the circles with the diagonals as
diameters are coaxial (the Gauss-Bodenmiller Theorem), and on the other hand, the
orthocentric line is the common radical axis, which is therefore orthogonal to the
Newton-Gauss line.
After this discussion, let us solve an enticing problem from the 2016 Brazilian
Mathematical Olympiad. This problem was proposed by the third author of the book
and, as we will discover, is about complete quadrilaterals.
Problem 4.4 Let ABCD be a noncyclic, convex quadrilateral with no parallel
sides. The lines AB and CD meet at the point E. The circumcircles of the triangles
ADE and BCE meet again at a point M = E. The internal bisectors of ABCD
determine a cyclic convex quadrilateral with circumcenter I , and the external
bisectors of ABCD determine a cyclic convex quadrilateral with circumcenter J .
Prove that I , J , and M are collinear.
Solution 1 We argue on Fig. 4.25, in which we define E and F to be the intersection
points of the pairs of lines (AB, CD) and (AD, BC), respectively. We have
constructed
√ a complete quadrilateral! Our first solution uses the aforementioned
bc inversion and some facts about Möbius transformations.
Denote by IXY the intersection of the internal bisectors of the angles X and
Y , and by JXY the intersection of the external bisectors of the angles X and Y ,
where X and Y range among the consecutive angles of the original quadrilateral.
Notice that IAB , ICD , JAB , and JCD are incenters and excenters of the triangles
ABF or CDF , so they lie on the internal bisector of F ; similarly, IBC , IDA , JBC ,
and JDA are the incenters and excenters of the triangles ADE and BCE, so they lie
on the internal bisector
√ of E.
Consider the bc inversion φ that fixes the complete quadrilateral ABCDEF .
We will prove a stronger statement:
The circumcircles ω of IAB IBC ICD IDA and of JAB JBC JCD JDA are fixed by
φ, so the centers of these circles lie on the common bisector of AMC, BMD,
and EMF .
As a consequence of this fact, since ω and are fixed by φ, both are orthogonal
to the circle of inversion, and M lies on the radical axis of ω and . This radical
axis is therefore the common external bisector of AMC, BMD, and EMF .
240 4 A Synthesis
JAB B JBC
M
IBC C
ICD IAB
IDA JCD
A
D
F
JDA
Fig. 4.25 Incenters and excenters
We now proceed with proving this stronger statement. To this end, let us find

JAB = φ(JAB ). Observe that JAB is the excenter of triangle ABF , and focus on
this triangle. At this moment, we apply the fact that the cross-ratio is invariant under
Möbius transformations, and we use complex coordinates. Recall that
x−y y−x
XY Z = arg = arg .
z−y y−z
From φ(A) = C, φ(B) = D, φ(C) = A, φ(D) = B, φ(E) = F , and φ(F ) = E,

, c, e, d), which is the same as
it follows that (jAB , a, f, b) = (jAB
jAB − f a − f j − e c − e
: = AB
−d : c−d.
jAB − b a − b jAB
Taking arguments, this yields

EJAB D + ECD = F JAB B − F AB,
where the angles are directed and taken modulo π . Because C, D, E are collinear,
ECD ∈ {0, π }. Working with directed angles modulo π and using the cyclic
quadrilaterals IAB AJAB B and IDA AJDA D (the incenter, excenter, and adjacent
vertices form a cyclic quadrilateral), we obtain

EJAB D = F JAB B − F AB = IAB AB − F AB
= − F AIAB = − DJDA E = EJDA D.
This means that JAB lies on the circumcircle of DEJDA . Analogously, by inter-
changing A and B, and C and D, we find that JAB lies on the circumcircle of

CEJBC . This implies that JAB is the Miquel point of the complete quadrilateral
obtained by completing CDJDA JBC , and thus it also lies on the circumcircle of
the triangle determined by the lines JBC JDA , CJBC , and DJDA , that is, on the
circumcircle of the triangle JBC JDA JCD . This circumcircle is. . . . So JAB lies on

, and similarly JBC , JCD , and JDA lie on too. This means that φ() = .
For the proof that φ(ω) = ω, we could invoke the same argument, but instead let
us show how to perform the angle chase with complex numbers instead. Again we
, c, e, d) = (i
have an equality of cross-ratios: (iAB AB , a, f, b). In all that follows,
we work with angles modulo π . The points C, D, E are collinear, so the equality of
cross-ratios implies

e − iAB e − iAB e−c f − iAB f − a
arg = arg : = arg : .
d − iAB d − iAB d −c b − iAB b − a
From the fact that F, IAB , JAB are collinear and A, IAB , B, JAB are concyclic, it
follows that
f − iAB jAB − iAB jAB − a
arg = arg = arg .
b − iAB b − iAB b−a
So

f − iAB f − a jAB − a f − a jAB − a
arg : = arg : = arg .
b − iAB b − a b−a b−a f −a
Because A, D, F are collinear, and JAB , A, JDA are collinear,
jAB − a jAB − a jDA − a

arg = arg = arg .
f −a d −a d −a
Because A, IAD , D, JAD are concyclic, IAD , JAD , E are collinear,
jDA − a jDA − iDA e − iDA

arg = arg = arg .
d −a d − iDA d − iDA
Consequently,

e − iDA
e − iAB
e − iDA e − iDA
arg = arg ⇐⇒ : ∈ R.
d − iDA d − iDA d − iAB d − iDA
242 4 A Synthesis
,i
So (iAB
DA , e, d) is a real number, proving that iAB lies on ω. From here the
solution continues as above to conclude that ω, like , is fixed by φ, and the
conclusion follows.

Solution 2 The second solution, discovered by Régis Prado Barbosa, is perhaps a
bit more elementary, and is based on homotheties, namely, on the two homotheties
between the circumcircles ω of IAB IBC ICD IDA and of JAB JBC JCD JDA . We
argue on Fig. 4.26.
We use the same notation as for the previous solution and start by proving some
basic facts. Because JBC and IBC are the incenter and the E-excenter of the triangle
BCE, JBC and IBC lie on the internal bisector of BEC, and the quadrilateral
IBC BJBC C is cyclic, having as circumcenter the midpoint K of IBC JBC . Reasoning
in a similar way for the triangle ADE, we deduce that the points IDA and JDA lie
on the internal bisector of AED (which angle coincides with BEC), and the
quadrilateral IDA AJDA D is cyclic as well, having as circumcenter the midpoint L
of IDA JDA . The common bisector of AED and BEC seems important for the
solution, so let us denote it by E . We define F analogously—notice that the points
IAB , JAB , ICD , and JCD all lie on F .
Recall from the statement of the problem that I is the center of ω and J is the
center of . Our strategy is to use the centers O+ and O− of the two homotheties
M
JBC
IDA
JAB C
K O+
IAB JCD
O− ICD
I F
A D
J IBC
L
JDA
Fig. 4.26 Homotheties

that map these circles into each other. First, note that if we are in the degenerate
situation with O− = O+ , then I = J and there is nothing to prove. Else, since both
O+ and O− lie on the line I J , the lines O+ O− and I J coincide, and we are left
with proving that O+ , O− and M are collinear.
It seems unnatural to replace I and J with O− and O+ , but the latter two points
are easier to locate! To see why this is so, note that since JAB J JCD is an angle at
the center of the circle and IBC BJBC C is a cyclic quadrilateral, we have
JAB J JCD = 2 JAB JDA JCD = 2(180◦ − JCD JBC JAB ) = 2(180◦ − BJBC C)
= 2 CIBC B = 2 ICD IBC IAB = ICD I IAB .
Moreover, because IAB , JAB , ICD , JCD lie in F , I ICD J JCD and I IAB J JAB .
We conclude that the direct homothety that takes ω to takes IAB to JAB and ICD
to JCD . From here we infer that the center O+ of this homothety lies on the line
IAB JAB , that is, on F . Analogously, O− lies on E and the inverse homothety
centered at O− that maps the circles into each other takes IBC to JBC and IDA to
JDA .
Now that we can work with the lines E and F , how do we bring the Miquel
point M into the picture? Let us first connect the inverse homothety with M. The
key elements are the midpoints K and L of IBC JBC and IDA JDA , which are on the
circumcircles of BCE and ADE, and these circumcircles intersect at M. We will
show that O− M bisects the angle BMD.
Notice first that, in the circumcircles of BEMCK and AEMDL,
1 1
BMK = BEK = BEC = AED = LED = LMD.
2 2
We now show that MO− bisects KML, because then BMO− = BMK +
KMO− = LMD + O− ML = O− MC, and so MO− bisects BMD.
To prove that MO− bisects KML, we use ratios, which are intrinsic to
homotheties, and the Bisector Theorem. From the inverse homothety, we obtain
O− JBC O− IBC O− JBC + O− IBC IBC JBC KJBC

= = = = ,
O− JDA O− IDA O− JDA + O− IDA IDA JDA LJDA
so
IBC JBC O− JBC KJBC O− JBC − KJBC O− K
= = = = .
IDA JDA O− JDA LJDA O− JDA − LJDA O− L
We are almost done: the spiral similarity centered at M that takes BC to AD

also takes the circumcircle of BCE to the circumcircle of ADE. Now, IBC JBC is
the diameter of the circumcircle of BJBC CIBC , since JBC BIBC = 90◦ (angle

between internal and external bisectors). Also, the midpoint K of arc BC is taken

to the midpoint L of arc AD. Thus,
244 4 A Synthesis
O− K IBC JBC MK
= = ,
O− L IDA JDA ML
which is the relation from the Bisector Theorem, and so MO− bisects KML. A
similar argument shows that MO+ bisects the angle BMD, and so the lines MO+
and MO− coincide. The conclusion follows.
4.2.3 Problems About Complete Quadrilaterals
We end our book with some questions involving complete quadrilaterals that can be
solved with the geometric transformations techniques that we have developed and
with the aid some of the theorems proved above.
252 Let ABCD be a cyclic quadrilateral with no parallel sides. Let E be the
intersection point of the lines AB and CD, and let F be the intersection point of
the lines AD and BC (i.e., we complete the quadrilateral.) Let M be the Miquel
point of this complete quadrilateral. Prove that:
(a) the point M lies on the line EF ; √
(b) the circumcircle of ABCD is fixed by the bc inversion that fixes the
complete quadrilateral;
(c) if P is the intersection of diagonals AC and BD and O is the circumcenter of
ABCD, then O and P lie on the common internal bisector of AMC, BMD,
and EMF ;
(d) the line OP is perpendicular to the line EF ;
(e) the circle ωAC through M, A, and C passes through the circumcenter O of
ABCD; moreover, ωAC meets again each of the perpendicular bisectors of the
sides of ABCD on points belonging to the sides of the complete quadrilateral.
253 Let ABCD be a quadrilateral with no parallel sides. Let E be the intersection
point of the lines AB and CD, and let F be the intersection point of the lines AD and
BC (so we complete the quadrilateral). Let M be the Miquel point of the resulting
complete quadrilateral and let P be the intersection point of the diagonals AC and
BD. Prove that if P lies on the common internal bisector of AMC, BMD, and
EMF , then ABCD is cyclic.
254 Let ABCD be a quadrilateral inscribed in the circle ω, and let K and N be
the intersections of the pairs of lines (AB, CD) and (AD, BC), so that ABCDKN
forms a complete quadrilateral. Prove that the circumcircle of AKN is tangent to ω
if and only if the circumcircle of CKN is tangent to ω.
255 Let ABCD be a quadrilateral, and let E and F be points on the sides AD and
BC, respectively, such that
AE BF
= .
ED FC
The ray |F E meets the rays |BA and |CD at S and T , respectively. Prove that the
circumcircles of the triangles SAE, SBF , T CF, and T DE pass through a common
point.
256 Let ABCD be a fixed convex quadrilateral with BC = AD and BC not
parallel to AD. Two variable points E and F lie on the sides BC and AD,
respectively, such that BE = DF . The lines AC and BD meet at P , the lines AC
and EF meet at Q, and the lines BD and EF meet at R. Prove that the circumcircles
of the triangles P QR, as E and F vary, have a common point other than P .
257 Let MN be a line parallel to the side BC of the triangle ABC, with M on the
side AB and N on the side AC. The lines BN and CM meet at the point P . The
circumcircles of the triangles BMP and CN P meet at two distinct points P and Q.
Prove that BAQ = CAP .
258 The points P and Q lie on the diagonals AC and BD, respectively, of a
quadrilateral ABCD such that
AP BQ
+ = 1.
AC BD
The line P Q meets the sides AD and BC at the points M and N . Prove that the
circumcircles of the triangles AMP , BNQ, DMQ, and CN P intersect at one point.
259 The circles ω and meet at the points A and B. Let M be the midpoint of the

arc AB of the circle ω (M lies inside ). A chord MP of the circle ω intersects at
Q (Q lies inside ω). Let P be the tangent line to ω at P , and let Q be the tangent
line to at Q. Prove that the circumcircle of the triangle formed by the lines P ,
Q , and AB is tangent to .
260 Show that in the four triangles formed by a complete quadrilateral, the
perpendiculars to the Euler lines at the centers of the nine-point circles intersect
at one point.
261 Let D be an interior point of the acute triangle ABC with AB > AC so that
DAB = CAD. The point E on the segment AC satisfies ADE = BCD, the
point F on the segment AB satisfies F DA = DBC, and the point X on the line
AC satisfies CX = BX. Let O1 and O2 be the circumcenters of the triangles ADC
and EXD, respectively. Prove that the lines BC, EF , and O1 O2 are concurrent.
Part II
Hints
Chapter 5
Isometries
1 What is the composition of two reflections? Where does it map the first triangle?
2 The set is invariant under the composition of the reflections over the centers of
symmetry.
3 The isometry has fixed points. To prove this, examine the behavior under the
isometry of a segment determined by a point and its image.
4 The composition of two reflections is a translation.
5 Use (and prove) the fact that if σ1 , σ2 , σ3 , σ4 are the reflections over four points
in the plane, then
σ1 ◦ σ2 ◦ σ3 ◦ σ4 = σ3 ◦ σ4 ◦ σ1 ◦ σ2 .
6 Look at the images of the points of pairwise intersections of a, b, and c under

the reflections σa , σb , and σc .
7 What kind of transformation is (σa ◦σb ◦σc )2 , where σa , σb , σc are the reflections
over the three lines?
8 Multiply the relation from the statement on the left by σa ◦ σb and on the right
by σd ◦ σc . Examine the relative position of the lines a and b.
9 By examining what types of isometries the compositions should be, deduce that
σa ◦ σ1 ◦ σb = σA ◦ σ1 ◦ σB . Then check that σ2 = σa ◦ σ1 ◦ σb .
10 Look at a pair of axes of symmetry that minimize the angle between them.
Create new axes of symmetry by reflecting one over the other.
11 What geometric transformation is the composition of the five reflections over
the midpoints (taken in order) of the pentagon?
250 5 Isometries
12 Consider rotations that map one trajectory into the other. Can you find such a
rotation that maps the starting points into each other?
13 Look at the 90◦ rotation about the center of the square. Where is P mapped?
14 Translate the segment CD into a segment AE. Look at the triangle EAB.
15 Construct the parallelogram ABJ C. Do you notice that it rotates by 90◦ into
EAGI ? You can also try to use complex numbers, the vertices of the squares
being 90◦ rotates of the vertices of the triangle. For (d), find a metric relation that
characterizes the symmedian.
16 Consider the 90◦ rotation about the center of the square.
17 The answer is yes, and the easiest construction uses a translation.
18 Try shearing, that is, try to slide the first two parallelograms, without changing
their areas, into two parallelograms that partition the third parallelogram.
19 Turn the relation into the Pythagorean Theorem in some right triangle.
20 Examine the proof of Theorem 1.24.
21 Use Theorem 1.26.
22 For (a), use complex coordinates with the origin at the circumcenter, which
yield a nice expression for the coordinate of the orthocenter. For (b), reflect over the
circumcenter.
23 Reflect D over AB.
24 What happens when you reflect the squares over the angle bisectors of the
triangle?
25 What happens when you translate ? Translate so that A0 lies on the
circumcircle of ABC.
26 Look at compositions of the 90◦ rotations about the centers of the squares.
27 You can use either a 60◦ rotation or a 120◦ rotation.
28 Look at the 60◦ rotation about the center of the hexagon. What happens to the
quadrilateral ABCM?
29 Parallelograms go hand in hand with translations!
30 Reflect A over the side BC and over the midpoint of this side. The images are
on the circumcircle of H BC, where H is the orthocenter of ABC. What else is on
this circle?
31 Consider either the reflection over the perpendicular bisector of the chord c, or
the reflection over A.
32 The angle is 60◦ , and a 60◦ rotation about A might help you prove it.
5 Isometries 251
33 Look at 60◦ rotations about A and B.

34 Examine the figure formed by P and its reflection over the point O.
35 Use the fact that AN is mapped to CM by a 120◦ rotation.
36 Reflect over the midpoint M of CE. If A maps to A and B to B , what kind of
quadrilateral is ABA B ?
37 Use Problem 6. Or, for a different solution, use the observation that if two lines
and are the reflections of each other over the point X, then they are reflections
over any point that lies on the parallel through X to and .
38 Can you dissect the triangle ABC into some triangles congruent to ABC1 ,
AB1 C, and A1 BC?
39 Think “cut and paste”! You can translate AB2 C1 , A1 BC2 , A2 B1 C so that A1
coincides with A2 , B1 coincides with B2 , and C1 coincides with C2 . What figure do
you get? Can you use this configuration to deduce that A1 B1 C1 is equilateral?
40 There is a rotation about I that takes the triangle DBF to the triangle GCE.
41 Study the compositions of the clockwise 120◦ rotations about A, B, C. Or
if you use complex numbers, you will see that the coordinates of the successive
locations of the man after 3n steps have a nice formula.
42 Look at the translation that maps one circle into the other. This translation will
map some points that are important in the solution into one other. Also, 30◦ angles
can be obtained from equilateral triangles!
43 The minimum should be a billiard path. An elegant approach uses five
reflections of the triangle over its sides that turn the perimeter of the orthic triangle
into a straight line.
44 The main step of the solution does not explicitly use isometries; it uses
properties of the isogonal conjugate of a point. Let Q be the isogonal conjugate
of P with respect to triangle AED. Show that QE is parallel to CD.
45 Use the same idea as for the proof of the Pappus’ Area Theorem.
46 Show that if ABC is a triangle, then you can construct a point D such that
ABCD is a parallelogram (it is not hard to do this if ABC is equilateral with
side-length equal to the opening of the compass, but by using compositions of
translations, you can do it in general). Show also that if A, B are two points, then
you can construct a point C such that ABC is equilateral (again, this is not hard if
AB has length equal to the opening of the compass, but by using compositions of
isometries, you can do it in general as well). Try to combine these two constructions.
47 Use “inscribed circles” and put the nails at the tangency points.
48 Use three cuts that meet at a (special) interior point.
252 5 Isometries
49 One of the squares can be obtained from the other by a rotation about the center
followed by a translation. Show that the property holds true when the squares have
the same center. Check that the total length of the red (or blue) segments does not
change under translations.
50 Consider the rotations of the polygon by the angles 2kπ/n, k = 1, 2, . . . , n−1,
and examine the overlaps of red vertices with black vertices. What is the average
number of red-black overlaps?
51 For part (a), use two translations of the set of points so that the resulting sets
and the original set do not overlap, and so that the three sets lie in a region whose
area is as small as possible. For (b) you can allow overlaps.
52 If such a set did not exist, then every reflection over a point would change the
color of all but finitely many points of the plane. Consider the group generated by
two reflections.
53 Tessellate the plane by regular hexagons with the distance between the centers
of two neighboring hexagons slightly greater than 4. Let H be one of the hexagons.
Look at the problem “modulo H.”
54 Add to the figure the reflections of the chords over the center of the circle. Then
use the Pigeonhole Principle.
55 Reflect “arcs” of people in order to reduce the number of neighboring enemies.
56 Imagine instead that the figure is fixed and the points move on the cylinder, all
rigidly linked to each other. Now rotate and translate the cylinder so that the points
avoid the surface. What region does one of the points trace?
57 The “reversed” configuration is the reflection over a line of the original
configuration.
58 Each of the equilateral triangles is partitioned into four equal triangles: medial,
top, bottom left, and bottom right. Focus just on the union of the top triangles and
show that the part of it, A, that is in the complement of T , has area smaller than
T . Note that a top triangle turns into the medial triangle by a reflection over the
common side. Slice A by the lines of support of these common sides, and then
reflect the slices.
59 Use the following Lemma: Let be a polygonal line in the plane with endpoints
A and B, and consider a number α ∈ (0, 1). Then among all segments with
endpoints in that are parallel to AB, there is either one of length αAB or one
of length (1 − α)AB.
60 Call the angle between the lines α. The trajectory of the flea is determined by
the vectors that describe the jumps. Can you relate the angle between the first and the
third vector to α? What about the angle between the second and the fourth vector?
61 Play symmetrically.
5 Isometries 253
62 Think about invariants! Color the chessboard by elements of the group of

symmetries of a nonsquare rectangle in such a way that at every jump the color
of the final square is obtained by multiplying the color of the initial square by the
same element.
63 Can you reduce the problem to symmetric play?
64 There are two key observations: (1) if two 1 × 4 rectangles coincide by a
horizontal or vertical translation by 3 units in the direction of their shorter side,
then they must contain an equal number of a’s, b’s, and c’s, and (2) if two 1 × 3
rectangles coincide by a horizontal or vertical translation by 4 units in the direction
of their shorter side, then they must contain an equal number of a’s, b’s, and c’s.
65 For (a) consider the two orthogonal vectors that define the lattice of one given
color. Factor out the plane by identifying two points if and only if one is mapped
into the other by a linear combination with integer coefficients of the two vectors.
You obtain a torus. What is the surface area of the torus? What region of that torus
is colored by each color?
66 Arrange the trees at the nth roots of unity. When a bird flies from the kth to
mth tree, its position rotates by 2π i(m−k)
n . Define an invariant associated with each
configuration of birds using powers of the rotation about the center of the circle by
2π i
n .
67 Translate A and B such that max A = min B = 0.
68 Translate A → A = A + {−a} and B → B = B + {b}, and then compare
A ∪ B to A ∪ B .
69 In other words, we are supposed to show that the union of finitely many
nondegenerate intervals with total length less than 1 may be translated so as not
to intersect Z. Can the translates of this union by integers cover the entire interval
[0, 1]?
70 The number is 2.
71 Use binary expansions to construct an example.
72 Encode the strings by geometric objects, and then use symmetry.
73 Count the number of non-monochromatic colorings of the vertices of a regular
p-gon by n colors.
74 Do the case k = 1 first. Follow the idea in Problem 73 where you consider a1
regular p-gons that rotate about their centers, together with a0 fixed points that do
not belong to these polygons. Pick b0 + b1 p points among the a1 vertices and the
a0 fixed points. Count configurations that are not rotation invariant.
75 Represent the family of equations ax + by = n as lines in the plane.
76 Think about Problem 67, but this time with rotations about a point since you
work modulo p.
254 5 Isometries
77 Write f = g + h such that the center of symmetry of the graph of g is (0, 0)

and the center of symmetry of the graph of h is (1, 0). You can start by constructing
f and g on one small interval and then extend this construction inductively.
78 Make f (x) = x + 1 on (0, 1].
79 To prove f (A0 ) = 0, choose a regular n-gon, A0 A1 A2 · · · An−1 , and consider
the rotations ρk of center A0 and angles 2kπ n , k = 0, 1, . . . , n − 1. Define Akj =
ρk (Aj ), k, j = 0, 1, . . . , n − 1. Examine all regular polygons that arise in the figure
after the rotations. There are more regular polygons than just the rotations of the
original polygon.
80 What is the composition of two reflections?
81 By a change of variable, you can turn P into a polynomial in z and z−1 that is
πi
invariant under rotation of z by π4 about the origin z → e 4 z.
√
82 The answers to (a)√and (b) are different. For (b), show that if |z − w| = 3,
then |f (z) − f (w)| = 3. From
√ this and the hypothesis, deduce that if z and w are
such that |z − w| = |m + n 3| with m, n ∈ Z, then |f (z) − f (w)| = |z − w|.
Chapter 6
83 The answer is 1 or 2. There is a simple solution using coordinates. For a

synthetic solution, you can use the group of homotheties and translations.
84 Beware of the many possible configurations, and of the fact that the homothety
can be direct or inverse! Think about the group of homotheties and translations.
85 There can be one or two such homotheties. Use composition to rule out the
existence of a third homothety.
86 Let h12 , h13 , h23 be the direct homotheties of the pairs (ω1 , ω2 ), (ω1 , ω3 ),
(ω2 , ω3 ), respectively. Then h13 = h23 ◦ h12 .
87 What is the image of the line through the homothety that maps ω2 to ω1 ?
88 There is a special line on which these projections lie. Use a homothety of center
A and ratio 1/2 to find it.
89 Use the homothety that maps the incircle to the excircle.
90 Can you spot an inverse homothety of center K?
91 If h1 , h2 , h3 are homotheties such that h3 = h2 ◦ h1 , then their centers are
collinear.
92 Use homothety to show that, in a trapezoid, the line determined by the
midpoints of the parallel sides passes through the intersection of the diagonals and
also through the intersection of the nonparallel sides.
93 The center of homothety is the centroid of the polygon.
94 The two polygons are homothetic.
95 A homothety of center I maps MNP Q to XY ZW .
96 The two orthic triangles are homothetic. What is the center of homothety?
97 In complex coordinates, the line through a and b reflects to the line through
c + d − a and c + d − b, whose parametric equation is t → c + d − a + t (a − b),
t ∈ R. Similarly, the line through a and d reflects to the line s → b+c−d +s(a−d),
s ∈ R. Can you guess the intersection point?
98 The perpendiculars through A , B , C to the sides BC, AC, AB, respectively,
are concurrent.
99 The triangle formed by the centers of three of the circles is homothetic to ABC.
The two triangles have the same incenter.
100 A homothety whose center is the intersection of the diagonals maps the circle
through the eight points to a circle centered at the circumcenter of the original
quadrilateral.
101 How are B and C connected by the direct homothety that maps the circles into
each other?
102 Fix B and C in the plane, and let A vary so that ABC has the property from
the statement. What are the loci of the centroid and the incenter?
103 A homothety of center H and ratio 2 can help.
104 Use a homothety centered at the centroid and of ratio −1/2. The Simson-
Wallace Theorem (Theorem 3.34) might be helpful.
105 Given a homothety, the line joining a point with its image passes through the
center of homothety.
106 Consider a homothety of center A and ratio 1/2.
107 There are two parallel lines in the figure! Consider homotheties of centers K
and E.
108 Use the homotheties centered at the orthocenters that map the circumcircles
to the nine-point circles.
109 Consider the homotheties hQ of center Q and which maps A to C, and hP of
center P and which maps C to B. What kind of map is the composition hQ ◦ hP ?
110 Think about the relationship between the incircle and the excircle.
111 Use a homothety of center Z and ratio ZB/ZA.
112 Find a better description of the points A1 , B1 , C1 using the triangles H BC,
H AC, H AB.
113 Construct the incircle of ABCD and consider homotheties that map it to ω
and ωa .
114 Let D be the point where AN intersects the circumcircle. Consider a
homothety of center A that maps D to R.
115 Use the idea from the proofs of Menelaus’ and Pappus’ Theorems.
6 Homotheties and Spiral Similarities 257
116 To prove that if the lines intersect then the identity holds, use homotheties that
map B1 to C, C to B, B to A, and A to B1 .
117 Prove that each of the triples (K, M, P ), (K, N, Q), and (M, L, N) consists
of collinear points.
118 Use the Monge-d’Alembert Theorem (Problem 86).
119 Let A be the intersection of the common exterior tangents of ω1 and ω2 , and
let B be the intersection of the common exterior tangents of ω1 and ω3 . You should
show that there is a circle ω that is mapped to ω2 by a homothety of center B and to
ω3 by a homothety of center A.
120 The two squares are homothetic. Use the homothety to compute the radius of
a . Show that the radii of a and b add up to the distance between their centers.
121 Let P be the intersection of A1 A4 and A2 A3 and assume that P , T1 , T3 are
collinear. Use the Monge-d’Alembert Theorem (Problem 86) to conclude that there
is a circle ω that is tangent to A1 A2 at T1 and to A3 A4 at T3 . Use the same theorem
in reverse to prove the other collinearity.
122 Add to the picture the B-excircle of the triangle ABC and the D-excircle
of the triangle ACD. Use homotheties that map the circles of this enriched
configuration into one another.
123 The previous problem is actually a good hint. Consider the incircle and A-
excircle of AP Q, and also the C-excircle of CP Q. Play with composition of
homotheties to find several triples of collinear points.
124 The five pentagons lie inside a pentagon that is homothetic to each of them
and twice as big.
125 Compose the projection maps from each line to the next to obtain a transfor-
mation of a line to itself. What kind of transformation is it?
126 Use homotheties centered at the vertices of the triangle to increase the range
where the grasshopper can start in order to land close to the flower.
127 Change the scale at which you look at the turtles.
128 Scaling! If a square K has too little black, cut it into four parts.
129 Use the following result due to Isaak Moiseevich Yaglom: every convex
polygon that is not a parallelogram has three sides with the property that the polygon
itself lies inside the triangle formed by the lines of support of these sides.
130 There is a spiral similarity mapping the line through B1 , B2 , B3 to the line
through C1 , C2 , C3 .
132 Use Problem 131 and Theorem 2.23.

134 There are two such spiral similarities.
135 On a side of the polygon, choose a point M that is not a vertex and consider
a spiral similarity s with center M, angle BAC, and ratio AB/AC. Look at the
points where the image of the polygon crosses the polygon itself.
136 Consider a spiral similarity of center H , angle 90◦ , and ratio BP /BC.
137 For (a) use the Averaging Principle. For (b), think circles! Use Problem 131.
138 Use the Averaging Principle.
139 Let M be the common midpoint; use a spiral similarity of center M.
141 There is a spiral similarity of center M that maps A1 A2 . . . An to
M1 M2 . . . Mn .
142 Construct a point C such that a spiral similarity of center A maps AMB to
AC N . Show that C = C .
143 Use Problem 131.
144 Theorem 2.25 implies that there are spiral similarities that map the triangle
ABC to P BR and to QRC.
145 The property is true if P and Q are the midpoints of the sides.
146 There is a spiral similarity of center M that maps O1 to O2 and D to C, and
spiral similarities come in pairs.
147 Use the Miquel point.
149 Map P to Q using the composition of two spiral similarities. What transfor-
mation is this composition?
150 Can you use a spiral similarity that appears in the proof of Miquel’s Theorem
(Theorem 2.24)?
151 The common point of these circles is the center of the spiral similarity that
maps ABC to A1 B1 C1 .
152 If X is the intersection point of the circumcircles of AP C and AQB, then AX
is symmedian and AX, BM, CN are concurrent.
153 There is a spiral similarity that maps DAE to DEB.
155 Let D, E, F be the feet of the altitudes from A, B, C, respectively. Then there
are spiral similarities that map the triangles ADX, BEY , CF Z into one another.
These spiral similarities are paired with three others. Locate the centers of all these
spiral similarities.
156 There are spiral similarities that map the triangle NXW to BNC and the
triangle Y MW to BMC, and spiral similarities come in pairs. Use the new spiral
similarities and their composition.
Chapter 7
Inversions
157 Use coordinates.

158 Take a Möbius transformation (or just an inversion) that maps the given circle
to a line.
159 Use complex coordinates.
160 There is a hidden circle in the figure, and the intersection point is the radical
center of the three circles.
161 Where does the tangency point map through the inversion with respect to ω?
162 Apply an inversion that maps two of the circles into two parallel lines.
163 Take an inversion of center A, and use the invariance of angles.
164 Map the circle of inversion to a line, and then use the Symmetry Principle.
165 Reduce the problem to an easy particular situation by making use of the
invariance of the cross-ratio under circular transformations and of the Symmetry
Principle.
166 Such Möbius transformations map the real axis to itself.
167 Compose with a Möbius transformation to obtain a transformation that maps
lines to lines and circles to circles. Can you then use a result from a previous
chapter?
168 How do the zeros of the polynomial change when (a) and (b) are applied?
169 Invert about A; what kind of quadrilateral is AM B N ?
170 Use formula (3.1). For an analytic solution, write the identity as an equality of
cross-ratios.
171 Invert about the inscribed circle and use Proposition 3.3.
262 7 Inversions
172 Use an inversion of center P . For (a) use the formula (3.1) or the invariance
of cross-ratios.
173 Consider an inversion of center M that maps ω into the line ST .
174 Use an inversion that transforms 1 and 2 into two parallel lines.
175 Invert with respect to the circumcircle and use Ceva’s Theorem.
√
176 Take an inversion of center O and radius Rr, where R and r are the radii of
the two circles.
177 Let BE and CF be altitudes. Map, by an inversion, the line EF to the
circumcircle of ABC.
178 Use an inversion of center A and radius AB.
179 An inversion can help reduce the number of circles.
180 The following can help with (a): an inversion of center M, an inversion of
center B composed with reflection over B, or an inversion of center A. One of these
transformations helps with (b) and (c).
181 Work backward, let B1 be the intersection of the common interior tangent of
ω and ω1 with the line 2 , and show that B1 = B. For this consider an inversion of
center B1 and radius B1 A.
182 Can you reduce the question to Problem 15 by using an inversion?
183 Locate the second intersection point of the circle with EM using an inversion
of center E and radius EB = EC.
184 Part (a) follows from the fact that if ABC is a triangle that is not isosceles, I
is its incenter, and D, E, F are the tangency points of the incircle to the sides BC,
AC, and AB, respectively, and then the circumcircles of AI D, BI E, CI F have a
second intersection point besides I . Part (b) follows from the fact that if ABC is a
triangle, and if M is on BC, N is on AC, and P is on AB such that AM, BN, CP
are concurrent, and if moreover X is on NP , Y is on MP , and Z is on MN such
that MX, N Y , and P Z are concurrent, then AX, BY , and P Z are also concurrent.
185 Consider the inversion whose center is the tangency point of C3 and and has
the property that maps C into itself.
186 Use an inversion of center F .
187 Use Proposition 3.3.
188 Consider an inversion of center A. Then D is diametrically opposite to A in
the circumcircle of AB C , E is the midpoint of the segment B C , and F is the
point where the perpendicular bisector of B C intersects the circumcircle of AB C .
189 Power-of-a-point leads to inversion.
7 Inversions 263
190 There are three circles passing through A: the circumcircles of ABL, AKC,
and AKL. Turn them into lines.
√
191 Consider the bc inversion corresponding to the vertex A in the triangle
ABC.
192 Consider an inversion of center B. Do you see a right triangle?
193 A circular transformation that is the composition of an inversion of center H
and the reflection over H might help.
194 There are too many circles in the configuration.
195 Show first that AO is perpendicular to B1 C1 . Then use an inversion of center
D and radius DA.
196 Reduce the problem to the following: two circles and ω intersect at B and
X and are tangent to the line at A and D, respectively. Let S and T be points on
such that BS is parallel to and AT is a diameter. Then the lines BD, SA, and T X
intersect (at some point C).
197 Map the circle to a line.
198 Prove that the bisectors of ABP and ACP intersect at the same point on
the line AP by showing that AB/BP = AC/CP . Write this as BA CA : CP = 1.
BP
√
199 Does a bc inversion help?
200 Take an inversion with respect to a circle centered at A and orthogonal to ω,
and use the fact that the polar of A with respect to ω is the line that passes through
Q and through the intersection of the diagonals of BCF E.
√
201 Modify the bc inversion corresponding to the vertex C.
202 Use a circular transformation that is the composition of an inversion of center
H and the reflection over H (i.e., an inversion of negative ratio).
203 The quadrilateral ABCD will have to be harmonic. Have you seen this before?
204 An inversion that takes one circle to another also induces a homothety between
the circles!
205 A, B, X, Y, I are on a circle. Invert about this circle.
206 Consider an inversion of center B, and then build an argument based on the
polar with respect to a circle.
√
207 Use a bc inversion defined by the triangle ABC and the vertex B. Reduce
the problem to the following: in triangle ABC, O is the circumcenter. The
perpendicular to the radius OB through O passes through the orthocenter H
of ABC if and only if the distance from the reflection of O over BC to this
perpendicular is equal to the circumradius R of ABC.
264 7 Inversions
208 Invert about a circle centered at X (to make things simple, keep the circle of
center O invariant).
209 ADBC is harmonic.
210 Consider the inversion of negative ratio centered at the orthocenter that maps
the circumcircle to the nine-point circle.
211 Invert about a circle centered at H . Use properties of symmedians.
√
212 Use a bc inversion. Theorem 4.2 from the next chapter turns out to be useful.
√213 We have written three solutions at the end of the book: one√that uses the
bc inversion corresponding to the vertex A, one that uses the bc inversion
corresponding to the vertex C, and one that uses inversion centered at K.
√
214 Use the bc inversion corresponding to the vertex A.
215 It is easier to prove that a line is tangent to a circle.
216 Let I be the incenter of ABC. Prove that SBP and SI Q are similar, and
deduce another similarity to finish the problem. There is also a symmedian in the
diagram!
√
217 Perform the bc inversion φA determined by the triangle ABC and the vertex
A. Use Proposition 3.41 to find the images of the B- and C-mixtilinear excircles.
218 First, prove that the point X is unique. Then work backward, showing that if
BAD = CAX then MB = MC. Consider the A-excenter Ia , and look at the
triangle BIa C. Think mixtilinear excircles!
√
219 First order of business is to take the bc inversion corresponding to the vertex
A. Where does the circumcircle of DEF go? Consider also separately the reflection
across the bisector of BAC.
220 Invert about the circle centered at the tangency point of ω1 and ω2 and of
√
radius 2 r1 r2 .
221 Reduce the problem to proving that A, T , the incenter I , and the point D0
diametrically opposite in the incircle to D are concyclic.
Chapter 8
A Synthesis
222 Use either spiral similarities or rotations. There is a short solution of (a) using
the Averaging Principle. For (b), use the spiral similarities centered at the vertices
A, B, C of angles 30◦ and ratios 2 cos 30◦ . Complex numbers work too.
223 For an inversion-based solution, use the invariance of the cross-ratio. There
is also an analytic solution that uses the system of coordinates introduced before
Problem 4.1 from the introduction.
224 Reflect the triangle SAD over the line through S that is orthogonal to P Q. Or
take an inversion centered at S.
225 Show that A1 A2 A3 and M1 M2 M3 have parallel sides.
226 Notice that the property is true if the three triangles are part of a regular
hexagon. To get to the general configuration, you can transform each triangle by a
spiral similarity centered at the common point. Show that the property is preserved
when changing one triangle at a time. Alternatively, the complex number solution is
quite easy.
227 An inversion centered at A yields a symmetrical figure.
228 The line through E, P , F is the image of the Simson line of D with respect to
ABC under a homothety of center D and ratio 2.
229 Midpoints lead to dilations; equilateral triangles lead to 60◦ rotations.
230 Cross-ratios, of course! But you can also solve this problem using just spiral
similarities.
231 Use a homothety and an inversion both centered at the intersection of the
common tangents of the two circles.
232 Use a transformation λ that is the composition of a spiral similarity and a
reflection over a line such that λ(H P Q) = ABC.
266 8 A Synthesis
233 For the direct implication, examine the image of through the homothety of
center C and ratio 1/2. For the converse, use the rotation about the circumcenter of
CKL that maps K to C.
234 Use Problem 131, the fact that spiral similarities come in pairs, and the
Averaging Principle for a proof via spiral similarities. For proof by inversion, check
that M, A, P , Q are concyclic by inverting with respect to a circle of center A and
proving that the images of M, P , Q are collinear.
235 For (a), prove that the orthocenter lies on the line determined by the reflections
of A over BC and DC, and for (b) use Steiner’s Theorem proved in (a).
236 The pairs of triangles F AB and F CD respectively F AD and F BC are
similar, but not directly similar. Use reflections over lines to make them directly
similar in order to use homothety for (a) and spiral similarity for (b).
237 There is a 90◦ rotation that maps P1 Q1 R1 S1 into P2 Q2 R2 S2 . Its center lies at
the intersection of the bisectors of the angles formed by AI and LC, BK and F D,
CE and H A, and DG and J B. What points do this angle bisectors pass through?
Can you use the fact that the diagonals of ABCD are orthogonal to find the points
where these angle bisectors meet?
238 Start with a spiral similarity that maps the triangle ABO to AOQ and prove
that Q = Q. Solutions based on isometries are also possible.
239 Let A be the point that is diametrically opposite to A in the given circle.
Prove that A D is perpendicular to BC. A spiral similarity of center C mapping the
triangle C CA to the triangle C MD might be useful.
240 Invert about I and observe that the lines Ai Ai+1 become equal circles.
241 Let O be the center of . Use a spiral similarity that maps EAF to EOD and
an inversion that maps to .
√
242 Start with the bc inversion defined by the triangle QRP and the vertex Q.
Theorem 1.22 might be useful for the rest of the argument.
243 Take an inversion of center R. Or, for a solution without inversion, reflect A
over S.
244 Let A1 , B1 , C1 the points where the incircle is tangent to the sides, and let

MA , MB , MC be the midpoints of the arcs BC, AC, AB of the circumcircle. Then,
the intersection point of AA , BB , CC is the center of the homothety that maps
A1 B1 C1 to MA MB MC .
245 One possibility is to use inversion and show that the line that joins C with the
intersection of AN and 2 is tangent to both 1 and 2 , and the same is true for the
line that joins D with the intersection of BN and 2 . Or you can use homothety to
show that CD is parallel to AB and attempt a computational solution to prove that
the distance from O2 to CD is equal to the radius of 2 .
8 A Synthesis 267
246 You can perform a regular inversion about C and do some computations, but
√
the bc inversion can circumvent these: consider the midpoints of the arcs AC and

BC that do not contain the other vertex, and find their images under the inversion
determined by the vertex C of the triangle ABC.
247 Invert about the incircle of ABC. Problem 204 can be used to locate the
inverse of the desired point.
248 Let Y be the intersection of BC and F G. Then, Y is the center of both an
inversion and a homothety that exchange the circumcircles of F BD and GCE.
249 Inverting about gives rise to the following question: fix a circle of center
G and radius r, a circle of radius r/2 which passes through G, and a circle
inside and disjoint from . The circles ω1 and ω2 of radii r/2 pass through G and
through a variable point on , and cross again at Y and Z. Prove that as X traces
, the circumcircle of XY Z is tangent to two fixed circles.
250 Recall the coordinate system introduced before Problem 4.1 from the essay
part of this chapter.
251 Parametrize the triangles (up to similarity) by points in an equilateral triangle
with altitude equal to π , so that the angles of a triangle are the distances from a point
in the equilateral triangle to the sides. How does H act on this equilateral triangle?
252 In (a), use
√ inscribed angles. How do O and P behave under the inversion that
is part of the bc inversion?
253 Try to use the coordinate system introduced at the beginning of Sect. 4.2.
√
254 Use homotheties that map the two circumcircles to ω. Or use a bc inversion.
255 This problem is a showcase for spiral similarities and the Miquel point.
256 Spiral similarities and the Miquel point.
257 Take advantage of√ spiral similarities and homotheties that arise in the config-
uration. Or consider a bc inversion defined by the vertex A in the triangle ABN.
258 There is a spiral similarity that takes A to D, C to B, and P to Q.
259 The tangency point is the Miquel point of the quadrilateral determined by
P , Q , AB, and MP . There is a solution based on inversion and one based on
homothety.
260 Use Theorems 4.7, 4.6, and Problem 23.
261 Turn BCEF into a complete quadrilateral and then consider an inversion of
center D.
Part III
Solutions
Chapter 9
Isometries
1 The composition of the two reflections is either a translation or a rotation, and

this transformation maps the triangle ABC to itself. It cannot be a translation
because its repeated applications would move the triangle away from itself.
Therefore, it is a rotation. The triangle that is invariant under some rotation is the
equilateral triangle. This can be proved as follows. Let ρ be the rotation. Then ρ(A)
is either B or C, say ρ(A) = B. Then ρ(B) = C and ρ(C) = A. It follows that
ρ 3 (A) = A, ρ 3 (B) = B, ρ 3 (C) = C, and since the vertices are three distinct
noncollinear points, ρ 3 must be the identity map. This implies that ρ is a rotation by
120◦ , and consequently A, B = ρ(A), C = ρ 2 (A) are the vertices of an equilateral
triangle. The image A B C of ABC through a reflection is also an equilateral
triangle.
Remark This will be a recurring theme in some problems: instead of exploring
the diagrams obtained after the transformations, we focus on the structure of the
transformations themselves.
Case in point: while we can work on the two triangles at hand, we prefer to
compose the reflections, because we already know well the structure of composition
of reflections. Since we know that the composition of two reflections is a rotation or
a translation, and we know what properties it must have, we can infer what kind of
triangle resonates with them.
Source D. Smaranda, N. Soare, Transformări Geometrice (Geometric Transforma-
tions), Ed. Academiei, Bucharest, 1988.
2 The answer is negative. Assume that such a set S exists and let O1 and O2 be the
two centers of symmetry. The set S must be invariant with respect to the reflections
σ1 and σ2 over the points O1 and O2 , respectively, so it is invariant under σ2 ◦ σ1 .
−−−→
But the latter is the translation τ by 2O1 O2 .
The composition of a translation and a reflection is a reflection, and the simplest
reflection that we can obtain this way that is not σ1 or σ2 is τ 2 ◦ σ1 . Indeed, this is
the reflection over O3 = σ2 (O1 ), for example, because
272 9 Isometries
τ 2 ◦ σ1 (O3 ) = τ 2 ◦ σ1 (σ2 (O1 )) = σ2 ◦ σ1 ◦ σ2 ◦ σ12 ◦ σ2 (O1 )

= σ2 ◦ σ1 ◦ σ22 (O1 ) = σ2 ◦ σ1 (O1 ) = σ2 (O1 ) = O3 .
So S has at least three centers of symmetry, O1 , O2 , O3 , and therefore it cannot

have exactly two.
Remark We see that in fact the figure must have infinitely many centers of
symmetry.
A good way to express the symmetries of a set is as invariance under a set of
transformations. In our case, an axis of symmetry corresponds to a reflection over
that axis and a center of symmetry corresponds to a reflection over the center. Then
we can compose these transformations, giving birth to new transformations under
which the set is invariant as well. In fact, these transformations form a group with
composition as operation, and can be fully characterized. With this in mind, finding
another reflection, and thus another center of symmetry, becomes an easy task.
Do you realize that there are several other options of compositions to find other
centers? You only need to make sure you compose an odd number of reflections.
3 First solution. Let A be a point such that f (A) = A. Then f (f (A)) = A, which
means that f maps the segment Af (A) to f (A)A. It follows that the midpoint M
of Af (A) is a fixed point of f . This property is trivially true for fixed points of f ,
so we conclude that for every point A, the midpoint of Af (A) is a fixed point of f .
We distinguish two situations. If f has exactly one fixed point, then f is just the
reflection over this point.
If f has at least two fixed points, say M and N, then the entire line MN is fixed.
In this case if f has a fixed point that does not lie on MN, then f is the identity
map. Otherwise, if A ∈ MN , then triangles AMN and f (A)MN are congruent, so
A and f (A) are symmetrical with respect to line MN . Hence, f is the reflection
over line MN .
Second solution. Suppose f is not the identity map. If f (z) = rz + s, then
f (f (z)) = z yields r 2 z + rs + s = z, for all z; thus, r 2 = 1 and s(r + 1) = 0. Since
f is not the identity map, this is only possible if r = −1, which corresponds to the
reflection over a point.
If f (z) = rz + s, then f (f (z)) = z yields rrz + rs + s = z, which means that
rs + s = 0. If s = 0, then we are done by Theorem 1.5. Otherwise, r = − ss , and
Theorem 1.5 shows again that f is the reflection over a line.
Remark A transformation f with the property that f ◦f is the identity map is called
an involution.
−−−−−−→
4 The composition σ2j −1 ◦ σ2j equals the translation by 2O2j O2j −1 . Thus,
σ1 ◦ σ2 ◦ · · · ◦ σ2n−1 ◦ σ2n = 1
−−−−−−→
is equivalent to the fact that the composition of the translations by O2j O2j −1 , j =
1, 2, . . . , n, is the identity map. Since the composition of the translations by −
→
v and
9 Isometries 273
−
→ −−−→
w is the translation by v + w, this is further equal to the translation by
−−−→ −−−→ −−−−−−→
O2 O1 + O4 O3 + · · · + O2n O2n−1 .
So the composition of the reflections is equal to zero if and only if this sum of
vectors is equal to zero. The sum from the statement is actually the negative of this
sum, and the conclusion follows.
Remark For n = 2 we obtain the following characterization of parallelograms: the
quadrilateral O1 O2 O3 O4 is a parallelogram if and only if σ1 ◦ σ2 ◦ σ3 ◦ σ4 = 1
where σk is the reflection over Ok , k = 1, 2, 3, 4.
5 First solution. The solution is based on the following result:
Lemma If σ1 , σ2 , σ3 , σ4 are reflections over the points P1 , P2 , P3 , P4 in the plane,
respectively, then
σ1 ◦ σ2 ◦ σ3 ◦ σ4 = σ3 ◦ σ4 ◦ σ1 ◦ σ2 .
−−→
Proof The composition σ1 ◦ σ2 is the translation by 2P2 P1 , and the composition
−−→
σ3 ◦ σ4 is the translation by 2P4 P3 . The conclusion of the lemma follows from the
fact that translations commute.
Using the lemma, we can write
σ1 ◦ σ2 ◦ σ3 ◦ σ4 ◦ σ5 = σ1 ◦ σ4 ◦ σ5 ◦ σ2 ◦ σ3 = σ5 ◦ σ2 ◦ σ1 ◦ σ4 ◦ σ3
= σ5 ◦ σ4 ◦ σ3 ◦ σ2 ◦ σ1 .
Second solution. Let σj be the reflection over the point of complex coordinate zj .
Then σj (z) = 2zj − z, so
σ1 ◦ σ2 ◦ σ3 ◦ σ4 ◦ σ5 (z) = 2(z1 − z2 + z3 − z4 + z5 ) − z
= σ5 ◦ σ4 ◦ σ3 ◦ σ2 ◦ σ1 (z)

Remark In general, if in a composition of reflections over points two pairs of
consecutive reflections are separated from each other by an even number of
reflections, then the two pairs can be swapped.
The lemma and the conclusion to the problem hold true for reflections over lines
as well, provided that we impose that those lines have a common point. In that case,
we use the fact that two rotations about the same point commute.
274 9 Isometries

6 We start with the computation
(σa ◦ σb ◦ σc )2 = (σa ◦ σb ◦ σc ) ◦ (σc ◦ σb ◦ σa ) = σa ◦ σb ◦ σc2 ◦ σb ◦ σa

= σa ◦ σb2 ◦ σa = σa2 = 1,
which shows that the equation from the statement is equivalent to the fact that σa ◦
σb ◦ σc is an involution (when composed with itself yields the identity map). And
this is equivalent to the fact that it is a reflection (see Problem 3), in fact reflection
over a line because it changes orientation.
To understand why if a, b, c intersect, then they satisfy the condition from the
statement, note that, by Theorem 1.10, σa ◦ σb ◦ σc is the composition of a reflection
and a translation parallel to the line of reflection. But the common point of a, b, c is
a fixed point for this transformation, so no translation is involved. Consequently, we
are in the presence a reflection, and the condition from the statement is satisfied.
For a, b, c parallel, σb ◦ σc is a translation in a direction perpendicular to c. Pick
a point at distance half the length of the translation vector from a. Then σa ◦ σb ◦ σc
keeps this point fixed, so again this transformation has to be a reflection.
For the converse implication, we give two proofs.
First proof. We know that σa ◦ σb ◦ σc is the reflection over a line. Let be this line
and let σ be the reflection over . Then
σa ◦ σb ◦ σc = σ
implies
σb ◦ σc = σa ◦ σ .
If a and intersect, then σa ◦ σ is a rotation about their intersection point. But then
σb ◦ σc is a rotation, and it must be a rotation about the intersection point of b and
c. It follows that a, b, c, intersect at one point and we are done.
If a and are parallel, then σa ◦ σ is a translation by a vector perpendicular to
a. But then σb ◦ σc is a translation by the same vector, so these two lines must be
parallel to each other, and must be parallel to a and . Thus, a, b, c, are parallel
lines.
Second proof. If b is parallel to both a and c, there is nothing to prove. So suppose,
without loss of generality, that a and b meet at the point P . Therefore, σa (P ) =
σb (P ) = P . Let Q = σc ◦ σb ◦ σa (P ) = σc (P ). Thus, Q = σa ◦ σb ◦ σc (P ) =
σa ◦ σb (Q). But σa ◦ σb is a rotation with center P , and the only fixed point by a
rotation is its center. Hence, P = Q if and only if σb (P ) = P . But the only points
that are fixed by a reflection over a line are the points on this line, so P is on c. We
conclude that P lies on a, b, and c.
9 Isometries 275
Remark The solution is based on understanding how reflections generate the group
of isometries. We use the very important fact that the composition of two reflections
is either a rotation or a translation.
7 Assume that the lines intersect at the three different points A, B, C and that
the lines are a = BC, b = AC, c = AB. Denote the reflections by σa , σb , σc ,
correspondingly. We are to prove that the transformation σa ◦ σb ◦ σc has no fixed
points.
Note that σa ◦σb is the rotation about C by 2 ACB, σb ◦σc is the rotation about A
by 2 BAC, and σc ◦ σa is the rotation about B by 2 CBA (the angles are directed).
Therefore,
(σa ◦ σb ◦ σc )2 = (σa ◦ σb ) ◦ (σc ◦ σa ) ◦ (σb ◦ σc )
is an orientation preserving isometry that rotates by 2 · 360◦ , so it does not rotate at

all. It must therefore be either the identity map or a translation.
The first case is excluded since the transformation does not keep A fixed. Hence,
(σa ◦ σb ◦ σc )2 is a translation by a nonzero vector. Because a translation does not
have fixed points, neither does its square root “σa ◦ σb ◦ σc ” have.
Remark We revisited here the idea of understanding first the square of the transfor-
mation in question. Note that if f (x) = x, then f (f (x)) = x, so if f ◦ f has no
fixed points, then neither does f .
8 Multiply the relation from the statement on the left by σa ◦ σb and on the right
by σd ◦ σc to obtain
(σa ◦ σb )2 = (σd ◦ σc )2 .
Now we distinguish three cases:

Case 1. If a is parallel to b, then there is a vector −
→
v such that σa ◦σb is the translation
−
→
by v . Then (σa ◦ σb ) = (σd ◦ σc ) is the translation by 2−
2 2 →v . This implies that
σd ◦ σc must be a translation (it cannot be a rotation or else its square would be a
rotation as well). And the translation vector must be − →v . Consequently, a, b, c, d
−
→
are parallel, being all perpendicular to v and the distance between a and b equals
the distance between c and d.
Case 2. If a and b are orthogonal, then (σa ◦σb )2 is the identity map, and so (σd ◦σc )2
is the identity map. Now σd ◦σc is an orientation-preserving isometry with square
equal to 1, so it is either the identity map or the reflection over a point. The first
case is ruled out by c = d, so it is the reflection over a point. But then c and d are
orthogonal. Note that if a and b are orthogonal and c and d are orthogonal, then
σa ◦ σb ◦ σc ◦ σd = σb ◦ σa ◦ σd ◦ σc = σd ◦ σc ◦ σb ◦ σa ,
276 9 Isometries
where for the last step we have used the lemma proved in the solution to
Problem 5.
Case 3. If a and b intersect at a point O and are not orthogonal, then by examining
Cases 1 and 2 and noticing that we can switch the pair (a, b) with the pair (d, c),
we conclude that c and d also intersect at one point and are not orthogonal.
Then σa ◦ σb is the rotation by twice the angle α = (a, b), so (σa ◦ σb )2 =
(σd ◦ σc )2 is the square of this rotation. This means that σd ◦ σc is a rotation around
O by some angle 2β such that
360◦ | 4α − 4β.
This divisibility condition holds only if either α = β or α = 90◦ + β. And in this

case, c and d must intersect at the same point O and they form the angle β.
The answer to the problem consists of the following three possibilities:
(i) a, b, c, d are parallel and the distance between a and b is the same as that
between d and c, measured in the same direction,
(ii) a and b are orthogonal and c and d are orthogonal,
(iii) a, b, c, d intersect at the same point and the angle between c and d either is
equal to that between a and b or exceeds this angle by 90◦ .
Remark We exploit the group structure on the set of isometries. And again we
decode the properties of an isometry from the properties of its square.
9 Note that the line is invariant under both transformations σb ◦ σ1 ◦ σa and
σB ◦ σ1 ◦ σA . The restrictions to this line of the two transformations are equal (see
Fig. 9.1), and they must be reflections over a point (prove it!).
Fig. 9.1 The reflections 1 2

restricted to
A M B
9 Isometries 277
Fig. 9.2 The transformation 1 2

σB ◦ σ1 ◦ σA
A M B
Now working back in the plane, it is not hard to see that both transformations
are reflections over lines perpendicular to (because the transformations reverse
orientation, and the line determined by a point and its image is parallel to , see
Fig. 9.2). Since the two transformations coincide on , they are reflections over the
same line, so they are equal.
Therefore, it suffices to show that σ2 equals either σb ◦ σ1 ◦ σa or σB ◦ σ1 ◦ σA .
We can check synthetically that σ2 = σb ◦ σ1 ◦ σa because this is the same as
σb ◦ σ1 = σ2 ◦ σa .
If L1 and L2 are the intersection points of 1 and 2 with , respectively, then σb ◦σ1
−−→ −−→
is translation by 2L1 B and σ2 ◦ σb is translation by 2AL2 . Since these two vectors
are equal, so are the corresponding translations, and we are done.
Or we can check analytically that σ2 = σB ◦ σ1 ◦ σA by choosing coordinates
such that A = α, B = β, α, β ∈ R, and 1 is given by the equation z = γ +it, where
γ ∈ R and t is a real parameter. Then σA = 2α − z, σB = 2β − z, σ1 = 2γ − z,
and
σB ◦ σ1 ◦ σA = 2(α + β) − γ − z.
The latter is σ2 .

Remark In the particular case where A = B = M, if we let m = a = b, then we
obtain σ2 = σM ◦ σ1 ◦ σM = σm ◦ σ1 ◦ σm .
10 The set of vertices of the polygon is invariant under the reflection over an axis
of symmetry, and a vertex and its image determine the reflection axis (provided that
the vertex is not on the axis of symmetry). Since there are finitely many pairs of
vertices, and each reflection maps some vertex to a different vertex, there can only
be finitely many axes of symmetry. As the axes of symmetry divide the polygon into
equal parts, no two axes of symmetry can be parallel, so any two intersect. Assume
there are more than two axes of symmetry, or else the problem has no content.
278 9 Isometries
Fig. 9.3 The axes of

3
symmetry 1 , 2 , . . . , n 2
4 1
6
5
Let 1 and 2 be two axes that minimize the angle between them. Consider the
reflection 3 of 1 over 2 , of 2 over 3 and so on. Then 1 , 2 , 3 , · · · are axes
of symmetry of the polygon. Note that they intersect at one point, namely, the
intersection of 1 and 2 . Because there are only finitely many axes of symmetry,
the sequence 1 , 2 , 3 , . . . is periodic. The figure formed by these lines is invariant
under the reflection over any of them. So they form a pencil of lines so that any
two neighboring lines form equal angles (which by minimality is (1 , 2 )). If
n is the period of our sequence, then in counterclockwise order, the sequence is
1 , 2 , . . . , n+1 = l1 (see Fig. 9.3).
Now let be a different axis of symmetry. Let be the translate of to the
common point of 1 , 2 , . . . , n . Then lies inside the angle formed by some j
and j +1 . But then forms with j an angle smaller than the angle between 1 and
2 . This contradicts minimality. Hence, 1 , 2 , . . . , n . are necessarily all the axes
of symmetry, and the conclusion follows.
Remark An n-gon can have at most n axes of symmetry (as one vertex can be
paired with itself or with any other vertex, and when it is paired with itself, then
the neighboring vertices are mapped into each other). The regular n-gon has exactly
n axes of symmetry, and is the only n-gon with this property.
Source M. Pimsner, S. Popa, Probleme de geometrie elementară (Problems in
elementary geometry), Ed. Didactică şi Pedagogică, Bucharest, 1979.
11 The solution can be followed on Fig. 9.4. Let M, N, P , Q, and R be the
given midpoints, of the sides AB, BC, CD, DE, and EA, respectively. The
transformation
σR ◦ σQ ◦ σP ◦ σN ◦ σM
is an isometry that rotates geometric figures by 180◦ , so it must be a reflection over

a point. Since it leaves A fixed, it is the reflection over A, σA . This can also be
checked easily using complex coordinates.
For the construction, start with a point X, and then reflect it successively over
M, N , P , Q, and R. The result is X = σA (X). Since X is the reflection of X over
9 Isometries 279
Fig. 9.4 The construction of

the pentagon E Q D
P
X R
C
A N
M
X B
A, we can construct A as the midpoint of XX . Then we reflect A successively over

M, N, P , and Q to obtain B, C, D, and E.
Remark The construction can be adapted to any polygon with an odd number of
sides. The midpoints do not determine uniquely the polygon if the number of sides
is even. Indeed, if a1 , a2 , . . . , am are the vertices and mj = 12 (aj + aj +1 ), j =
1, 2, . . . , m, (aj = aj +1 ) are the midpoints. The midpoint conditions
aj + aj +1 = 2mj , j = 1, 2, . . . , 2n + 1,
form a system of linear equations in the unknowns a1 , a2 , . . . , am . This can

be solved by Cramer’s rule provided that the determinant is nonzero. But the
determinant of the system is a particular case of a circulant determinant, and the
particular case of Cremona’s formula shows that it is equal to
2π i
4π i
2(m−1)π i

(1 + 1) 1 + e m 1 + e m ··· 1 + e m .
2(m/2)π i
This product is zero precisely when m is even, because then 1+e m = 1−1 = 0.
So the system can be solved if and only if m is odd.
Source I.M. Yaglom, Geometric Transformations I, (transl. by A. Shields), MAA,
1975.
12 Let P (t) and Q(t) be the locations of the two cars at time t, and let us orient
each line in the direction of the motion. If there exists a rotation that maps the
first line into the second, preserving the orientation and mapping P (0) to Q(0),
then its center O has the desired property. Indeed, in the triangles OP (0)P (t)
and OQ(0)Q(t), OP (0)P (t) = OQ(0)Q(t), because the angles are mapped
into each other by the rotation, OP (0) = OQ(0), because of the rotation, and
P (0)P (t) = Q(0)Q(t) because the cars travel with the same speed. So OP (t) =
OQ(t) and we are done.
280 9 Isometries
Fig. 9.5 The locus of the

centers of rotations mapping
one line into the other
Fig. 9.6 Finding the point P(0)

that is at equal distance from S
the cars
Q(0)
T
It remains to show that such a rotation exists. The locus of the points that
are centers of rotations that map the first line into the second by preserving the
orientation is one of the angle bisectors of the angles formed by the two lines. This
can be seen easily on Fig. 9.5, since the center of rotation should be at equal distance
from the two lines, and if a point is indeed at equal distance from the lines, then the
rotation that maps the perpendiculars to the lines from the point into one another
also maps the lines into one another. Note also that the rotations about the points on
only one of the bisectors preserve orientations.
On the other hand, the locus of the centers of the rotations that map P (0) to Q(0)
is the perpendicular bisector of the segment P (0)Q(0). This line cannot be parallel
to the angle bisector discussed above unless the two coincide. It follows that the two
lines always intersect. If the intersection point is unique, let it be O. Then O is the
desired center of rotation. Indeed, let us consider the rotation of center O that maps
the first line into the second, preserving orientations. We claim that the image of
P (0) is Q(0). Let S and T be the projections of O onto the first and second line,
respectively. Then the right triangles OSP (0) and OT Q(0) are congruent (Fig. 9.6),
because OS = OT and OP (0) = OQ(0). It follows that P (0)OS = Q(0)OT ,
and these angles are directed the same way (otherwise P (0) reflects to Q(0) over the
bisector of the angle formed by the two lines and O is not unique). Hence, SOT ,
which is the angle of rotation, is congruent to P (0)OQ(0). This combined with
OP (0) = OQ(0) implies that P (0) is mapped to Q(0) by the given rotation, as
claimed. Thus, O is the center of the rotation that maps one of the car trajectories
to the other. The image through the rotation of center O of P (t) is Q(t), and we are
done.
If O is not unique, that is, if the bisector of the angle formed by the two lines
coincides with the perpendicular bisector of P (0)Q(0), change the time reference,
9 Isometries 281
Fig. 9.7 The 90◦ rotation of A4 A3

the square
A1 A2
Fig. 9.8 Construction of a C

triangle in which MN is a D
midline
M N
E A
B
that is, work with the points P (t1 ) and Q(t1 ), where t1 = 0. This will make O
unique.
Remark There is only one point in the plane that satisfies the requirements of the
problem.
Source Kvant (Quantum), proposed by Igor Fedorovich Sharygin.
13 The 90◦ rotation about the center of the square maps A1 P , A2 P , A3 P , and
A4 P to the perpendiculars from A2 , A3 , A4 , A1 , respectively (see Fig. 9.7). Hence,
the intersection of the four perpendiculars is the point P which is the image of P
under the 90◦ rotation about the center.
Remark The 90◦ rotation is suggested by the construction of the perpendiculars.

Try to use the same idea in other problems where perpendicular lines are present.
Source Kvant (Quantum), proposed by A.N. Vilenkin.
14 Translate the segment DC to the segment EA, as shown in Fig. 9.8. Then
EACD is a parallelogram, so M is the midpoint of CE. In the triangle CEB,
MN is a midline, so EB = 2MN. The condition from the statement implies that
EB = EA + AB. But because of the triangle inequality in triangle EAB, this can
only happen if E, A, B are collinear. And this happens if and only if CD and AB
are parallel.
282 9 Isometries
E F
C
B
Fig. 9.9 Triangle with squares on its sides
Remark Perhaps the best inspiration for the solution is the well-known proof of the
trapezoid midline formula, in which we extend the base AB and intersect it with the
line CM, obtaining point E. However, doing exactly the same here is not nearly as
good, because we lose the fact that CDM and EAM are congruent. So, to preserve
this, we instead perform a translation that guarantees the congruence, and then see
that the key to the solution lies in the triangle inequality.
The lesson here is: when you try to replicate a proof, focus on why that proof
works rather than simply redoing it.
15 First solution. Complete the triangle ABC to a parallelogram ABJ C (see

Fig. 9.9). Because AB and AE are equal and form a 90◦ angle and AC and AG
are equal and form a 90◦ angle, it follows that there is a 90◦ rotation that maps the
parallelogram ABJ C to the parallelogram EAGI . Hence, the segments EG and
AJ , being transformed into each other by the rotation, are perpendicular and have
equal lengths. Because AJ is the diagonal in the parallelogram ABJ C, it passes
through the midpoint of BC and has length equal to twice the median from A. This
proves (a).
9 Isometries 283
Fig. 9.10 Three 90◦ rotations
Additionally, the segments AI and BC are transformed into each other by the
90◦ rotation, which proves (b).
Moreover, BC = AI . The segments AC and CF are also perpendicular and
of the same length. This shows that the triangle AI C is mapped to the triangle
CBF by a 90◦ rotation. We deduce that BF is perpendicular to I C. Similarly, CD
is perpendicular to I B. Thus, AI, CD, and BF are the altitudes of the triangle
I BC. They intersect at the orthocenter of this triangle, which proves (c). The three
rotations used for proving (a), (b), and (c) are sketched in Fig. 9.10.
(d) Denote by d(X, ) the distance from a point X to the line . Because O
is on the perpendicular bisectors of AE and AG, the distance from O to AB is
d(O, AB) = AE/2 = AB/2, and to AC is d(O, AC) = AG/2 = AC/2. Thus,
d(O, AB) AB
= .
d(O, AC) AC
On the other hand, if M is a point on BC, then the fact that M is the midpoint of
BC is characterized by the fact that the triangles AMB and AMC have the same
area, namely, by the equation d(M, AB) · AB = d(M, AC) · AC. This equation can
be written as
d(M, AB) AC
= .
d(M, AC) AB
Using similar triangles, we deduce that
d(X, AB) AC
=
d(X, AC) AB
characterizes the points that lie on the line of support of the median through A.
Reflecting over the angle bisector of A, we deduce that the equation of the points
that lie on the line of support of the symmedian is
d(O, AB) AB
= .
d(O, AC) AC
284 9 Isometries
As we have seen, O is one of these points, and the problem is solved.

Second solution. There is a quick complex number solution for (a), (b), and (c).
Denote, in the standard way, the complex coordinate of a point by the lowercase
letter of the uppercase letter that denotes the point. Assume that the triangle is
oriented counterclockwise. Then E is the clockwise 90◦ rotation of B about A,
and G is the counterclockwise rotation of C about A, so
e−a g−a
= −i and = i.
b−a c−a
It follows that e = i(a − b) + a and g = i(c − a) + a, so the segment EG has

length |g − e| = |b + c − 2a| and direction specified by the angle arg(g − e) =
arg(i(b + c − 2a)). On the other hand, the midpoint of BC has coordinate b+c
2 , so
the median has length

b + c 1

2 − a = 2 |b + c − 2a|
and its direction is specified by the angle

b+c
arg − a = arg(b + c − 2a).
2
Hence, the length of the median is half the length of the segment EG and the median
makes with this segment a 90◦ angle, proving (a).
For (b) note that I is the reflection of A over the midpoint of EG, so it suffices
to show that the segment formed by this midpoint and A is orthogonal to BC. This
is the same as showing that
e+g
−a
2
c−b
is imaginary. This fraction is equal to
i(a − b) + a + i(c − a) + a
−a
2 = i,
b−c
which is, indeed, imaginary.

For (c) we use the same observation that I A, BF , and CD are the altitudes of the
triangle I BC after checking with complex coordinates that CD is orthogonal to I B
and BF is orthogonal to I C. To check, for example, that BF is orthogonal to I C,
we compute the complex coordinate of I using the fact that I is the reflection of A
over the midpoint of EG. Hence, its coordinate is g + e − a = i(c − b) + a. The
point F is the clockwise rotation of A about C so is f = i(c − a) + c. Then the line
9 Isometries 285
BF has direction given by the angle arg(f − b) = arg[i(c − a) + c − b], and the line
CI has direction given by the angle arg[i(c − b) + a − c] = arg[i((c − b) + i(c − a)]
and the latter angle differs by 90◦ from the former. The conclusion follows.
Remark The idea of the problem is to prove congruence and orthogonality of
segments by including them in figures that rotate into each other by 90◦ , and these
90◦ rotations are suggested by the presence of the squares.
Also question (a) gives away the construction of the parallelogram ABJ C, since
constructing it is the best way to produce a segment twice as long as the median from
A. Part (d) exhibits another construction of the symmedian; see also the discussion
in Sect. 2.4.3.
16 It suffices to prove that the four perpendiculars transform into one another by
the 90◦ rotation ρ about the center of the square. Let the parallelogram be ABCD
and let the square be EF GH , as in Fig. 9.11.
If C = ρ(C), then ρ maps the triangle CF G to the triangle C GH . So C G
is perpendicular to CF and hence to AD, and C H is perpendicular to CG. It
follows that D is the orthocenter of triangle C GH . From this we deduce that C D
is perpendicular to GH . Hence, the perpendicular line from C onto F G is mapped
by the 90◦ rotation ρ to the perpendicular line from D onto GH . The same holds
true for the other perpendiculars from the vertices. This means that the rectangle
formed by the four perpendiculars is invariant under a 90◦ rotation, so it is a square.
Remark You should compare this solution to the one given to Problem 15. Again
you include the segments that you want to map into one another by 90◦ rotations into
figures that are mapped as such. And again a key step is the fact that the altitudes of
a triangle intersect at one point.
D
H G
A
C
E F
B
Fig. 9.11 The square inscribed in a parallelogram

286 9 Isometries
Fig. 9.12 Rectangles ABCD Q P

and MN P Q such that
AM = BN = CP = DQ
D C
M N
A B
Source Kvant (Quantum), proposed by Nikolai Borisovich Vassiliev.

17 Consider an isosceles trapezoid ABNM that is not a rectangle, and translate
it by a vector −→v that is perpendicular to AB. If C, D, P , Q are the images of
B, A, N, M, respectively, then ABCD and MNP Q have the desired property. The
construction is shown in Fig. 9.12.
18 This is a true “geometry in motion” solution, presented in Fig. 9.13. Translate

MN along its line of support to M R and P Q along its line of support to P R.
Then M B, RA, P C, CS, and BT are all parallel and equal to each other. The
parallelograms ABMN and ABM R have the same area, because they have the
same base and the same height. The same is true for the parallelograms ACP Q and
ACP R.
Now translate RA along its line of support to R A so that R is on BC.
Then BR A T and R CSA are parallelograms that partition BCST . Note also
that BR A T has the same area as ABM R, because they have the same base
(RA = R A ) and the same height (the distance between lines AR and BM ).
Similarly, the parallelograms ACP R and R CSA have the same area.
Thus, we have partitioned the parallelogram BCST into two parallelograms, one
whose area is equal to the area of ABMN and one whose area is equal to the area
of ACP Q. The conclusion follows.
Remark This is a generalization of the Pythagorean Theorem, and so this
translation-based method also proves the Pythagorean Theorem (see Chapter 4 for
other proofs of the Pythagorean Theorem that use geometric transformations). The
transformation that slides a shape between two parallel lines is called shearing. This
proof of Pappus’ Theorem uses shearing to transform the first two parallelograms
into a partition of the third parallelogram.
19 The metric relation in the statement should remind us of the Pythagorean
Theorem, and the trick is to arrange the three segments into a right triangle. For
that, reflect the point N over O to the point N (Fig. 9.14). Note that the entire
triangle CNO is reflected to the triangle AN O, and so CN = AN .
9 Isometries 287
R
R
Q
N M P
A
A
P
M
B C
B C
T S T S
R
B C
T S
A
Fig. 9.13 The proof of the Pappus’ Area Theorem
Moreover, because NOM = 90◦ , N is also the reflection of N over line OM,
so MN = MN . Thus, we have placed the three segments CN, AM, MN in the
triangle AMN .
The last observation is that
N AM = N AO + OAM = OCN + OAM = 90◦ ,
so the triangle AMN is right. The Pythagorean Theorem gives AN 2 + AM 2 =

MN 2 , and hence AM 2 + CN 2 = MN 2 , as desired.
Source Moscow Mathematical Olympiad, 1999.
20 Like in the first proof of Theorem 1.24, let PA , PB , PC be the reflections of
P over BC, CA, AB, respectively. Let also QA be the reflection of Q over BC
288 9 Isometries
N A N A N A
M M M
O O O
C N B C N B C N B
Fig. 9.14 The construction of a right triangle containing the three segments
Fig. 9.15 The reflections of

AP and AQ over the angle A
bisectors of BP C and
BQC are reflections of each PC
other over BC PB
Q P
B C
QA
PA
(Fig. 9.15). Then the lines QPA and P QA are the reflections of each other over
BC; they are symmetric with respect to BC.
By Proposition 1.25 (i), Q is the circumcenter of PA PB PC , so
QPA PC = 90◦ − PA PB PC ,
In other words, (QPA , PC PA ) and (PB PC , PA PB ) are complementary. In the

proof of Theorem 1.24, we have seen that AQ, BQ, CQ are the perpendicular
bisectors of the segments PB PC , PA PC , PA PB , respectively, so
(AQ, PB PC ) = (BQ, PC PA ) = (CQ, PA PB ) = 90◦ .
Consequently,
(BQ, QPA ) = 90◦ − (QPA , PC PA ) = (PB PC , PA PB ) = (CQ, AQ).

9 Isometries 289
Thus, the line QPA is the reflection of the line AQ with respect to the angle bisector
of BQC. Switching the roles of P and Q, we deduce that the line P QA is the
reflection of the line AP with respect to the angle bisector of BP C. So P QA and
QPA are the two lines from the statement, and since Q and PA are the reflections of
P and QA over BC, the two lines reflect into each other over BC.
Source Jean-Pierre Ehrmann.
21 We will use Theorem 1.26. Let a1 , a2 , . . . , a2n be the complex coordinates of
the vertices A1 , A2 , . . . , A2n . Translate a1 , a3 , . . . , a2n−1 by v ∈ C. Then the area
of the modified polygon is
1
[(a1 + v)a2 + a2 (a3 + v) + · · · + (a2n−1 + v)a2n + a2n (a1 + v)]
2
1 1
= (a1 a2 + a2 a3 + · · · + a2n a1 ) + (va2 + va2 + · · · + va2n + va2n )
2 2
1
= (a1 a2 + a2 a3 + a3 a4 + · · · + a2n a1 ),
2
because the second term of the sum is a real number, so its imaginary part is zero.
And this is the area of the original polygon. The problem is solved.
Source Kiril Dochev.
22 (a) The shortest solution uses complex coordinates. We place the origin at the
circumcenter of the triangle ABC, and then use the property that the orthocenter’s
reflection over the midpoint of one side lies on the circumcircle and is diametrically
opposite to the third vertex. Here is how to prove this fact (this proof can be followed
on Fig. 9.16). If A is the reflection of the orthocenter H of triangle ABC over BC,
then H BA C is a parallelogram; therefore, A BA is equal to (CH, AB), so it
is right. For the same reason, A CA is right, which implies that A lies on the
circumcircle of ABC diametrically opposite to A.
Fig. 9.16 Reflection of the A

orthocenter over the midpoint
of a side
B C
A
290 9 Isometries
Returning to the problem, consider a coordinate system with the origin at the
circumcenter of the quadrilateral ABCD, and let the coordinates of A, B, C, D
be a, b, c, d, respectively. Then the coordinate of a of the point A that lies
diametrically opposite to A (and so is the reflection of A over the origin) is −a.
Because Hd is the reflection of A over the midpoint of BC, its complex coordinate
is
b+c
hd = 2 − (−a) = a + b + c.
2
Similarly, the coordinates of Ha , Hb , Hc are
ha = b + c + d, hb = a + c + d, hc = a + b + d.
And we can see that Ha , Hb , Hc , Hd , are the respective reflections of A, B, C, D

over the centroid of ABCD. Indeed, the coordinate of the midpoint of AHa is
a + (b + c + d) a+b+c+d
= ,
2 2
which is the coordinate of the centroid, and the same is true for BHb , CHc , and
DHd . The problem is solved.
(b) First solution. Let ABCD be the cyclic quadrilateral and let M, N, P , Q be
the respective midpoints of AB, BC, CD, and DA (Fig. 9.17). Then MNP Q is a
parallelogram, and let S be its center. The perpendicular bisectors of the sides meet
at the center O of the circumcircle. The perpendiculars from M, N, P , Q onto the
opposite sides are reflections of the perpendicular bisectors over O. So they meet at
the reflection over O of S.
Second solution. With the notation from the proof of (a), if we choose the radius
of the circle to be 1, so that aa = bb = cc = dd = 1, then the fact that the line
through the midpoint M of AB (whose coordinate is a+b 2 ) and the anticenter (whose
coordinate is a+b+c+d ) rotates by 90◦ to line CD, means that the ratio
2
D P D
C C
Q
S O
A N A N
M M
B B
Fig. 9.17 The proof of the existence of the Mathot point

9 Isometries 291
a+b+c+d a+b
−
2 2
c−d
is a purely imaginary number. This is equivalent to
c+d
c−d
being an imaginary number. We compute
c+d c+d c−d cc − cd + dc − dd

= · =
c−d c−d c−d |c − d|2
1 i
= (1 − cd + dc − 1) = I(dc),
|c − d|2 |c − d|2
which is, indeed, an imaginary number. The problem is solved.

Remark As a consequence of (a), the quadrilateral formed by the orthocenters is
congruent to the original quadrilateral. The four perpendiculars from the midpoints
of the sides onto the opposite sides are also known as maltitudes. Their point of
intersection, which is also the circumcenter of Ha Hb Hc Hd , is known as the Mathot
point or the anticenter of ABCD.
It is worth remembering the fact that if the circumcenter of a triangle lies at the
origin of the coordinate system, then the coordinate of the orthocenter is the sum of
the coordinates of the vertices.
23 Reflect over AB to complete a circle, and let D be the reflection of D (see
Fig. 9.18). Then D CA = ACD = ECB, so E, C, D are collinear. We have
1
DEF = DE= DD E = CDD = 90◦ − ACD.
2
Given that the triangle F ED is isosceles, we have
EF D = 180◦ − 2 DEF = 180◦ − 2(90◦ − ACD) = 2 ACD

= ACD + ECB,
and we are done.
Remark The solution is built on the classical trick used for proving that, for a ray of
light reflected by a mirror, the angle of incidence equals the angle of reflection. This
is a consequence of Fermat’s principle that light travels on the fastest path (which
in a homogeneous atmosphere is also the shortest), and by “reflecting” the beam
into the mirror (which is the same as removing the mirror), the angle of reflection
becomes opposite, and hence equal, to the angle of incidence.
292 9 Isometries
Fig. 9.18 Reflection of D F

over AB
E
D
Γ
A B
C
Fig. 9.19 Triangle with two

equal inscribed squares
A
C
B
Source British Mathematical Olympiad, 2018-2019, proposed by D. Griller.

24 Let the triangle be ABC, and let us concentrate on the angle BAC and the two
squares that have sides on AB and AC (Fig. 9.19). Because the squares are equal,
the figure consisting of the squares and the angle is symmetric with respect to the
bisector of the angle. So the squares are mapped into each other by the reflection
with respect to this bisector. Consequently, the vertices of these squares that lie
on side BC are mapped into one another by the reflection; hence, the side BC is
perpendicular to the angle bisector of BAC.
The same argument shows that the angle bisectors of ABC and ACB are
perpendicular to the sides AC and AB, respectively. And because all three angle
bisectors are also altitudes, the triangle is equilateral.
9 Isometries 293

existence of the orthopole
M C
B
C1
A
A1
B1 C0
A0
B0
Remark The fact that two equal squares are inscribed in a triangle forces the
existence of a symmetry of the triangle.
25 If you translate the line , the points A0 , B0 , C0 are translated by the same
vector, and so are the perpendiculars from these points to the sides of the sides of
the triangle. Translation preserves concurrency, so it suffices to prove the result for
a special translate of .
If A1 , B1 , C1 are the images of A0 , B0 , C0 under the translation, we may enforce
that A1 be the second intersection point of AA0 with the circumcircle of the triangle
ABC. The proof of this particular case can be followed on Fig. 9.20. Let M be the
projection of A1 onto BC, and let A be the point that is diametrically opposite to
A, so that A A1 A is right. Then A must be on (because is orthogonal to AA1
at A1 ).
Since the quadrilateral A1 MCC1 is cyclic (having two opposite right angles),
(CB, C1 M) = (CM, C1 M) = (CA1 , C1 A1 ) = (CA1 , A A1 )

= 90◦ − (AA1 , CA1 ) = 90◦ − (AB, CB).
Thus, (CB, C1 M) + (AB, CB) = 90◦ , from where we infer that that C1 M is
orthogonal to AB. Similarly, B1 M is orthogonal to AC. Hence, the desired point
of concurrency for is M, and consequently the corresponding point for is the
preimage of M under the translation.
Remark The point where the three perpendiculars intersect is called the orthopole
of the line with respect to the triangle ABC.
Source This proof for the existence of the orthopole was communicated to us by
James Tao.
294 9 Isometries
26 First solution. We argue on Fig. 9.21. Let ρ1 , ρ2 , ρ3 , ρ4 be the (clockwise) 90◦

rotations about M1 , M2 , M3 , M4 , respectively, so that ρ1 (A) = B, ρ2 (B) = C,
ρ3 (C) = D, and ρ4 (D) = A. The composition ρ2 ◦ ρ1 rotates figures by 180◦ , so
it is the reflection over some point P . Because this reflection maps A to C, P must
be the midpoint of AC. Note that P is also the midpoint of the segment joining
the M1 with ρ2 ◦ ρ1 (M1 ) = ρ2 (M1 ), so it is the midpoint of the hypotenuse in the
right triangle formed by M2 , M1 , and ρ2 (M1 ). Consequently, the triangle P M1 M2 is
right isosceles, with the right angle at P . Similarly, the composition ρ4 ◦ ρ3 is is the
reflection over the same midpoint P of AC, and for the same reason, the triangle
P M3 M4 is right isosceles, with the right angle at P . We conclude that triangle
P M1 M3 is obtained from triangle P M2 M4 by a 90◦ rotation about P , and hence,
M1 M3 and M2 M4 are equal and perpendicular.
Second solution. For the complex number solution, note that M1 is the midpoint of
the segment joining A to the 90◦ rotation of A about B. If the coordinates of the
vertices are a, b, c, d, correspondingly, then the coordinate x of this 90◦ rotation of
A satisfies
x−b
= i,
a−b
so x = i(a − b) + b. Then M1 has coordinate
M4
A
M1 P
D
B M3
C
M2
Fig. 9.21 Quadrilateral with squares constructed on its sides

9 Isometries 295
x+a a+b a−b

m1 = = +i .
2 2 2
Similarly, the coordinates of M2 , M3 , M4 are, respectively,
b+c b−c c+d c−d d +a d −a

m2 = +i , m3 = +i , m4 = +i .
2 2 2 2 2 2
−−−→
We are left to check that m4 − m2 = i(m3 − m1 ), namely, that the vector M1 M3
−−−→
rotates by 90◦ to the vector M2 M4 . And indeed

c+d c−d a+b a−b
i(m3 − m1 ) = i +i − −i
2 2 2 2
−c + d + a − b c+d −a−b
= +i
2 2
d +a d −a b+c b−c
= +i − −i = m4 − m2 ,
2 2 2 2
which completes the solution.
Remark Can you find a complex number solution that runs parallel to the first
solution?
It should be noted that when two consecutive vertices of the quadrilateral
coincide, the problem degenerates into the following: On the sides BC, CA, AB
of the triangle ABC, one constructs in the exterior squares; let M1 , M2 , M3 be
their centers, respectively. Show that AM1 and M2 M3 are perpendicular and of
equal length. This configuration can be related to that from Problem 15. In this
situation, we challenge you with the following problem, which was published by
V.I. Dubrovsky in Kvant (Quantum): at every vertex, the two sides of squares that
are not sides of ABC determine a triangle, and in each of these triangles, consider
the side that lies opposite to the corresponding vertex of ABC. Show that the
perpendicular bisectors of these three sides are concurrent.
1975.
27 First solution. Rotate about C by 60◦ so that B is mapped to A (Fig. 9.22). Let
D be the image of D. The segment P D rotates by 120◦ to BD, and BD rotates by
60◦ to AD , so P D rotates by 180◦ to AD . It follows that the segments DP and
AD are parallel and equal, so DP D A is a parallelogram. As E is the midpoint
of the diagonal AP , it is also the midpoint of the diagonal DD . But CDD is an
equilateral triangle; hence, DEC is the 60◦ − 90◦ − 30◦ triangle.
Second solution. We argue on Fig. 9.23. Consider the rotation about D by 120◦ that
maps P to M and B to P . Then C is mapped to some point K on MP . We must have
MK = CP , because one segment is the image of the other through the rotation, and
296 9 Isometries
Fig. 9.22 Finding the angles B

of the triangle DEC using a
60◦ rotation
M P
E
A C
Fig. 9.23 Finding the angles B

of the triangle DEC using a
120◦ rotation
K M P
E
A C
since CP = AM, we have MK = AM. It follows that triangle AMK is equilateral

(being isosceles and having AMK = 60◦ ), so AK is parallel and equal to P C.
We find that KACP is a parallelogram, in which E is the midpoint of one diagonal,
AP , so it is the midpoint of the other diagonal, KC. We conclude that E is the
midpoint of the side KC in the isosceles triangle DKC having KDC = 120◦ .
Thus, DEC = 90◦ , ECD = 30◦ , and EDC = 60◦ .
Remark When you see an equilateral triangle, you should think about either a 60◦
rotation about a vertex or a 120◦ rotation about its center.
Source Kvant (Quantum), proposed by A. Kuptsov.
9 Isometries 297
E N D
F O C
L
A B
Fig. 9.24 Regular hexagon
28 The solution can be followed on Fig. 9.24. The key observation is that the
quadrilateral AMCB transforms into the quadrilateral BNDC by a 60◦ rotation
about the center O of the hexagon. By subtracting from the areas of these two
quadrilaterals the area of the overlap, we obtain (a).
For (b), note first that since M rotates to N, the triangle OMN is equilateral,
so OMN = 60◦ . Because AM rotates to BN, NLM = 60◦ , but also
NOM = 60◦ , and hence N, O, L, M, as well as D lie on a circle (the latter
because NDM = 120◦ ). We deduce that OLN = OMN = 60◦ , and also
ALO = 180◦ − OLN − NLM = 60◦ .
Finally, OLD = OMD = 90◦ , which proves (c).
Remark Regular polygons are invariant under rotations about their centers!
Source Kvant (Quantum), proposed by E. Gotman.
29 We argue on Fig. 9.25. The condition ACD = BCM does not look
very suggestive, except for the geometry expert who might think about isogonal
conjugates. But the parallelogram ABMD induces at least two translations. Of
these, let us work with the translation that takes A to B and D to M, and let C
be the image of C. Then ACD = BC M and now all we need to prove is that
BC M = BCM, namely, that the quadrilateral BC CM is cyclic. But DC is
mapped to MC by the translation, so MDCC is also a parallelogram. Therefore,
MC C = MDC = MBC, which proves that BC CM is indeed cyclic, and we
are done.
Remark You might have already noticed that parallelograms and translations are a
perfect match! In fact, a parallelogram XY ZW induces two translations, one that
takes X to Y and W to Z and the other that takes X to W and Y to Z.
298 9 Isometries
Fig. 9.25 The translation by B

−→
AB
A
M
But which translation should we use for this problem? Actually, both of them
work! Try performing the other translation to see what happens (in fact, you will
obtain a reflection of key quadrilateral BC CM).
Source Communicated by Cicero Thiago Magalhães.
30 The solution can be followed on Fig. 9.26. Let H be the orthocenter of the
triangle ABC. Notice that BH C = EH F = 360◦ − 2 · 90◦ − BAC = 180◦ −
BAC. But, even though the angles BH C and BAC sum up to 180◦ , they face
the same way. How do we fix that? One idea is performing a reflection σ across BC,
obtaining σ (A) = A . Then A BH C is cyclic and its circumcircle is the reflection
of the circumcircle of ABC.
Is there any other symmetry between and σ () = ? Yes, there is the
reflection ρ over the midpoint M of their common chord BC. Let A = ρ(A).
Since AD = DA and AM = MA , A A BC; in particular, H A A = 90◦ ,
so H A is a diameter of . Since X lies on the circumcircle of AEF , which has
diameter AH , H XA = 180◦ − H XA = 90◦ , and we conclude that X lies on
.
Now an angle chasing drives the point home: since A BCA is an inscribed

trapezoid, the arcs BA and CA are equal, so, with directed angles modulo 180◦ ,
ZXY = BXC − A XC = BH C − BH A = ZH Y,
proving that H, X, Y, Z are concyclic. Hence H ZY = H XY = 90◦ and so

Y Z BC, as desired.
Remark Keep in mind that the reflections of H over both the side AB and the
midpoint of the side AB are on the circumcircle.
9 Isometries 299
Fig. 9.26 Proof that A

Y Z BC
E
H X
F
Z
Y
B M C
D
A A
P N
D B A C E
R S
Q
M
Fig. 9.27 The Butterfly Theorem
Source Cono Sur Mathematical Olympiad, 2007, proposed by Y. Lima, Brazil.

31 Look at Fig. 9.27 and recognize the shape of the butterfly, which explains the
name of the theorem.
First solution. Consider the reflection over the perpendicular bisector of the chord
c, and, for a point X, denote by X its reflection. The solution can be followed on
Fig. 9.28.
300 9 Isometries
Fig. 9.28 Reflection over a

line for the proof of the N
Butterfly Theorem P
A C
D E
S
Q Q
M M
We are supposed to show that R coincides with S. To this end, we check that
both S and R lie on the circumcircle of the triangle CN P . That R is on this
circumcircle follows from the fact that the quadrilateral NP R C is cyclic, which is

true because CNR = CP R both being half the measure of the arc MQ.
On the other hand,
1 1 1
CSN = (ND + QE) = (ND + Q D) = Q N= Q P N = CP N.
2 2 2
This shows that N P SC is cyclic. Consequently, both R and S lie at the intersection
of a circle and a line. The other point of intersection being C, we must have R = S
(for R = C would mean B = R and that can only happen if M = Q in which case
S = C as well). The problem is solved.
Second solution. Denote the circle by . Consider the reflection over the point A,
and let , M , and P be the reflections of , M, and P , respectively (Fig. 9.29).
One observation is that D reflects to E and vice versa, so D, E, as well as M and
P are on .
We now show that the points M , P , N, Q lie on a circle. First, using power-of-
a-point, we have
BM · BN = BD · BE = CD · CE = CP · CQ,
and using the fact that BM = CM and CP = BP , we obtain
BN CQ
= .
BP CM
Also, since the reflection over A maps the line BN to CM and the line BP to CQ,
NBP = M CQ, which together with the above metric relation, implies the
triangles BNP and CQM are similar. It follows that BNP = CQM . Using
this equality, and working with directed angles modulo π , we have
9 Isometries 301
Fig. 9.29 Reflection over a

point for the proof of the
Butterfly Theorem
M
N
P
B A C
D E
R S
P
Q
M
P M Q = (MP , M Q) = MP Q + P QM
= MNQ + P QM = MNP + P NQ + P QM
= P NQ,
showing that M , N, P , Q are concyclic.

The lines DE, NQ, and M P are the pairwise radical axes of , and the
circumcircle of M , N, P , Q, so they intersect at the radical center of the three
circles. But M P intersects DE at the symmetric of R with respect to A, while
N Q and DE intersect at S. Thus, S is the reflection of R with respect to A, and the
theorem is proved.
Remark The fact that A is the midpoint of both BC and DE suggests that a
reflection that maps B → C and D → E might help, and two such reflections
exist, one over a line and one over a point. Both work.
When A = B = C, the second solution becomes simpler and was communicated
to us by Ilya Bogdanov. In that case, using once more directed angles modulo π ,
M NQ = MNQ = MP Q = QP M ,
where the last equality follows from M P ||MP , whence M , N, P , Q are con-
cyclic, and the solution ends as above.
32 Let P be the intersection of CF with the parallel through A to BC (Fig. 9.30).
Then the triangles AF P and BF C are similar, so AP /BC = AF /BF . Also the
triangles AF E and BF D are similar; thus, AE/BD = AF /BF , that is, AE/BC =
AF /BF . It follows that AP = AE.
302 9 Isometries
Fig. 9.30 Additional C

construction for finding
(BE, CF )
A
D
F
E
P B
Now consider the rotation by 60◦ about A that maps B to C. This rotation also
maps E to P , so it maps the segment BE to CP . It follows that the angle between
BE and CF = CP is 60◦ .
Remark Note that in order to prove that the segment BE rotates by 60◦ to the
segment CP , we have placed them in the triangles ABE and ACP that rotate into
each other.
Can you write an analytical solution?
Source A. Myller, Geometrie Analitică (Analytical Geometry), 3rd ed., Ed. Didac-
tică şi Pedagogică, Bucharest, 1972.
33 For easy reference, draw the triangle ABC so that it is oriented counterclock-
wise. Let ρA and ρB be the 60◦ counterclockwise rotations about and A and B,
respectively. Then ρB (A1 ) = C and ρA (C) = B1 (see Fig. 9.31). So
ρA ◦ ρB (A1 ) = B1 .
The composition of the two rotations is a rotation itself, by 120◦ . The center of this
rotation is its fixed point. Note that ρB (M) is the reflection M of M over AB and
ρA (M ) = M, and hence ρA ◦ ρB (M) = M. So M is this fixed point. We conclude
that A1 is mapped to B1 by a 120◦ rotation about M, and so MA1 = MB1 and
A1 MB1 = 120◦ , as desired.
Remark In order to prove that a triangle is isosceles, we show that there is a rotation
centered at one vertex that maps the second vertex to the third.
1975.
34 First note that two lines with the desired property divide the polygon P into
four parts, with the parts inside opposite angles having the same area. Reflect P
over O to P (see Fig. 9.32). The boundaries of P and P intersect inside each angle
formed by two lines that divide the area in half, or else the part of P inside this
angle would be either larger or smaller than the corresponding part of P , and the
9 Isometries 303
Fig. 9.31 Equilateral A1

triangles constructed on the C1
sides of a triangle
C
B1
M
A B
Fig. 9.32 Lines that bisect

the area of a polygon P
opposite equality would hold in the opposite angle. Of course this cannot happen
because then the part of P in one region would be strictly larger than the part of P
in the other.
If there are k lines, there are 2k angles formed by them, hence 2k intersection
points of the two polygons. However, one side of P cannot contain more than two
intersection points, or else P would not be convex. Therefore, 2k ≤ 2n, that is,
k ≤ n, as desired.
Remark By sliding a horizontal line continuously, we can always find a situation
where it bisects the area of the polygon. Same with a vertical line. So there is a
point in the interior where at least two lines that bisect the area exist. The reflection
of the polygon over this point intersects the original polygon in at least four points.
Source All-Union Mathematical Olympiad, 1973, solution from S. Savchev,
T. Andreescu, Mathematical Miniatures, MAA, 2003.
35 We assume that the quadrilateral is oriented counterclockwise and that all
rotations in this solution are counterclockwise as well. We will prove first that
AN = CM, by showing that they map into each other by a 120◦ rotation. The
argument can be followed on Fig. 9.33.
304 9 Isometries
Fig. 9.33 Quadrilateral with N

equilateral triangles on its
sides
P
C
A
Q
The 60◦ rotation ρ1 about A maps M to B, and the 60◦ rotation ρ2 about C maps
B to N . Note that if S = ρ1 (C), then ACS is equilateral and ρ2 (S) = A. Hence,
ρ2 ◦ ρ1 maps the segment AN to the segment CM. This composition is an isometry
that rotates every figure by 120◦ , and we can conclude that it is a 120◦ rotation. And
so AN = CM and the two segments make an angle of 120◦ .
Similarly, ρ2 ◦ ρ1 (AP ) = CQ so AP = CQ and they form a 120◦ angle. We
deduce that the angle between AP and CQ is equal to the angle between AN and
CM, both angles being equal to 120◦ .
On the other hand, BD maps to P N by the reflection over the exterior angle
bisector of BCD, and BD maps to QM by the reflection over the exterior angle
bisector of BAD. Hence, QM = BD = P N. We deduce that triangles AP N and
CQM are congruent having the corresponding sides equal. Consequently, P AN =
QCM. So the angle between CQ and CM equals the angle between AP and
AM, which combined with the fact that the angle between AN and CM equals the
angle between AP and CQ implying that the four lines AP , AN, CM, CQ form a
parallelogram, and the conclusion follows.
Remark The problem is based on the properties of the Fermat-Torricelli point. This
point can be defined for triangles whose angles are less than 120◦ , and is the point
in the interior of the triangle with the property that the sum of the distances to
the vertices is minimal. The Fermat-Torricelli point is obtained by constructing
equilateral triangles on the sides of the triangle and then taking the common
intersection point of the circumcircles of these triangles. It can also be obtained
9 Isometries 305
as the common point of the lines that join each vertex of the original triangle with
the exterior vertex of the equilateral triangle that lies opposite to it.
In the case of our problem, both triangles ABC and ADC have angles less than
120◦ (or else the sum of the angles of the quadrilateral would exceed 360◦ ). The
intersection F of AN and CM is the Fermat-Torricelli point of the triangle ABC.
Let ACT be the third equilateral triangle that defines the Fermat-Torricelli point
F . An easy angle chasing using the fact that F AT C is cyclic shows that BT passes
through F , and the above argument based on rotations shows that BT = AN = CM
and that they form 60◦ angles. Note also that by applying Ptolemy’s Theorem in the
quadrilateral F AT C (F A · CT +F C · AT = AC · F T ) implies F A +F C = F T , so
F A+F B +F C = BT = AN = CM. Note that if F is another point in the interior,
then F A+F C ≥ F B, by Ptolemy’s Theorem (F A·CT +F C ·AT ≥ AC ·F T )
in the quadrilateral F AT C, which is not necessarily cyclic, and T F + F B ≥ T B
by the triangle inequality, so F A + F B + F C > F A + F B + F C (you cannot
have equality in both of the above inequalities unless F = F ).
36 The condition AB = BC + AE is not very geometry-friendly unless we make
an additional construction. This construction is motivated by a number of factors:
that M is the midpoint of CE, the condition from the statement, and the fact that
BC is parallel to AE. Because of these considerations, we take the reflection over
M (Fig. 9.34). Let this reflection map A to A , B to B , and D to D . Notice that C
is mapped to E. Now the condition BC AE implies that C ∈ A B and E ∈ AB ,
and we obtain the parallelogram ABA B . Furthermore, since AB = BC + AE =
BC + CA = BA , the parallelogram ABA B is actually a rhombus.
O D
A M A
Fig. 9.34 The reflection of ABCD over M

306 9 Isometries
The condition DMO = 90◦ means that OM is the perpendicular bisector of

DD , which implies that D lies on the circumcircle of BCD. Keeping in mind the
cyclic quadrilateral BCDD and the reflection, we obtain
BDC = BD C = B DE,
so
BDB = BDE + EDB = BDE + BDC = CDE

= ABC = 180◦ − BAB .
This means that BDB A is a cyclic quadrilateral itself. Combining this with the fact
that ABA B is a rhombus, we obtain
1 1
BDA = BB A = ABC = CDE,
2 2
and we are done.
Remark Whenever the conditions of a problem include a sum of two segments, it is
good to think about an additional condition that creates a segment equal to the sum.
Source Short list of the International Mathematical Olympiad, 2010, proposed by
N. Serdyuk, Ukraine.
37 First solution. Let the perpendiculars to the sides BC, CA, AB at M and M1 ,
N and N1 , and P and P1 be m and m1 , n and n1 , p and p1 , respectively. The
configuration is shown in Fig. 9.35.
For a line l, we denote by σl the reflection across l, and for a point X, we denote
by σX the reflection over the point X.
Fig. 9.35 Concurrent lines A

that are orthogonal to the
sides of a triangle
p1 n
P1
N
n1
p
P N1
B M M1 C
m m1
9 Isometries 307
By the result proved in Problem 6, three nonparallel lines l1 , l2 , l3 intersect at one

point if and only if σl1 ◦ σl2 ◦ σl3 = σl3 ◦ σl2 ◦ σl1 , that is, if and only if
(σl1 ◦ σl2 ◦ σl3 )2 = 1.
Using this fact, we see that we have to show that
(σm ◦ σn ◦ σp )2 = 1
is equivalent to
(σm1 ◦ σn1 ◦ σp1 )2 = 1.
Of course we only have to check the direct implication (by symmetry). We can write
σm1 = σB ◦ σm ◦ σC , σn1 = σC ◦ σn ◦ σA , σp1 = σA ◦ σp ◦ σB .
Then
(σm1 ◦ σn1 ◦ σp1 )2

= σB ◦ σm ◦ σC2 ◦ σn ◦ σA2 ◦ σp ◦ σB2 ◦ σm ◦ σC2 ◦ σn ◦ σA2 ◦ σp ◦ σB
= σB ◦ (σm ◦ σn ◦ σp )2 ◦ σB = σB2 = 1.

Second solution. With the notation from the first solution, the reflections of m, n,
and p over the circumcenter O of triangle ABC are m1 , n1 , and p1 , respectively. So
if the lines m, n, p intersect at one point, then the lines m1 , n1 , p1 intersect at the
reflection of this point over O and vice versa.
Remark The first solution relies on algebraic manipulations in the group of
isometries.
The second solution relies on the following observation: if two lines and are
the reflections of each other over the point O, then they are reflections over any
point that lies on the parallel through O to and .
1975.
38 The sums of angles from the statement can be computed, because the sum of
the angles of a hexagon is 720◦ , so
A + B + C = A1 + B1 + C1 = 360◦ .
The hexagon is the juxtaposition of the triangles ABC, A1 BC, AB1 C, and ABC1 ,
so the problem actually requires us to show that the area of ABC is equal to the
308 9 Isometries
C
B1
A1 A
A
C1 B
Fig. 9.36 Dissection of the triangle ABC into the triangles ABC1 , AB1 C, A1 BC
sum of the areas of the triangles A1 BC, AB1 C, and ABC1 . As we are focused on
isometries, a first idea that should come to mind is to dissect the triangle ABC into
those three triangles. This is possible, indeed, and we will prove it with the aid of
Fig. 9.36.
We are given that B1 C = A1 C, so there is a rotation about C that maps B1 to
A1 . Let A be the image of A through this rotation. We compute
A A1 B = 360◦ − A A1 C − BA1 C = 360◦ − AB1 C − BA1 C

= 360◦ − B1 − A1 = C1 = AC1 B.
So we have AC1 = AB1 = A A1 , BA1 = BC1 , and AC1 B = A A1 B (for the

equality of segments, we have used the hypothesis of the problem). It follows that
the triangles AC1 B and A A1 B are congruent. We deduce that the triangle A BC
can be dissected into triangles congruent to A1 BC, AB1 C, and ABC1 .
Let us compare the triangles ABC and A BC. They share the side BC, while
CA = CA as one rotates into the other. But also AB = A B because of the
congruence of triangles AC1 B and A A1 B. So the triangles ABC and A BC are
congruent, having respectively equal sides. Therefore, the triangle ABC can be
dissected into triangles congruent to A1 BC, AB1 C, and ABC1 , and the problem
is solved.
Remark This “cut and paste” technique will appear in some other problems.
Source Moscow Mathematical Olympiad, 1997.
39 Translate the triangles A1 BC2 , A2 B1 C along their respective equilateral
triangle sides until C1 coincides with C2 and B1 coincides with B2 . Now complete
the figure to a triangle (see Fig. 9.37). The result is an equilateral triangle.
Next, let α, β, and 60◦ be the measures of the angles around the vertex P , defined
in Fig. 9.38. We have α + β + 60◦ = 180◦ , so in the upper triangle, the remaining
angle is β, and in the bottom rightmost triangle, the remaining angle is α. It follows
9 Isometries 309
B2
x x
x
C1 x
B1 x
x
C2 x
x
B A1 x A2 C
Fig. 9.37 Translating the small triangles to form an equilateral triangle
Fig. 9.38 Analysis of the

small equilateral triangle o
60
P
x o
60
x
o o
60 x 60
that the three shaded triangles are congruent. So what we have here are the translates
of the triangles AB2 C1 , A1 BC2 , A2 B1 C.
Back to the original diagram, we now deduce that the triangles A1 A2 B1 ,
B1 B2 C1 , and C1 C2 A1 are congruent (by side-angle-side), so the triangle A1 B1 C1
is equilateral.
But then the equalities C1 B1 = C1 A1 and A2 B1 = A2 A1 mean that the line
C1 A2 is the perpendicular bisector of A1 B1 (see Fig. 9.39). Analogously, the line
A1 B2 is the perpendicular bisector of B1 C1 and the line B1 C2 is the perpendicular
bisector of C1 A1 . Therefore, the lines A1 B2 , B1 C2 , and C1 A2 are concurrent at the
circumcenter of the triangle A1 B1 C1 .
Remark The main motivation behind the “cut and paste” idea is that we know much
more about triangles than hexagons; so we prefer working on a diagram with only
triangles! In this problem, the “cut and paste” translates into. . . translations!
The translations work well because they preserve many of the angles, so most
of the angle chasing that we have done can be carried back to the original diagram.
Finally, the many equal segments lead to perpendicular bisectors and to yet another
equilateral triangle, and this is what finishes the problem.
310 9 Isometries
Fig. 9.39 Analysis of the A

original equilateral triangle
B2
x
x
C1
B1
x
C2 x
x
B A1 x A2 C
Fig. 9.40 A rotation about I A

maps DBF to GCE
E
I
D
B F G C
Source International Mathematical Olympiad, 2005, proposed by Bogdan Enescu,

Romania.
40 Because BD = CG, BF = CE, and DBF = ABC = ACB = GCE,
the triangles DBF and GCE are congruent and have the same orientation (see
Fig. 9.40).
Since BI = CI , there is a rotation of center I and angle α = BI C that takes
B to C and, consequently, the triangle DBF to GCE. So DI G = F I E = α.
Since α does not depend on r, F I G = DI G + F I E − 180◦ = 2α − 180◦ ,
which also does not depend on r.
Remark With some angle chasing (try it!), you can prove that F I G = BAC.
Source Cono Sur Mathematical Olympiad, 2005.
41 First solution. For a point M in the plane, denote by fM the clockwise 120◦
rotation about M. Then P1 = fA (P0 ), P2 = fB (P1 ), P3 = fC (P2 ), etc., as you
can see in Fig. 9.41. The problem states that (fC ◦ fB ◦ fA )662 (P0 ) = P0 . But
9 Isometries 311
Fig. 9.41 The path of the P0

man
120º P1
P2
A
60º
B
P3
Fig. 9.42 C is a fixed point A

of fC ◦ fB ◦ fA .
fB
C
fC
fA
the transformation g = fC ◦ fB ◦ fA is an orientation-preserving isometry, so by

Theorem 1.9, it is either a rotation, a translation, or the identity map. It rotates figures
by 120◦ + 120◦ + 120◦ = 360◦ , so it is not a rotation. It cannot be a translation
because g 662 is a translation as well, which cannot have P0 as a fixed point. So it
must be the identity map.
It follows that fC = (fB ◦ fA )−1 . Now construct a point C such that ABC is
equilateral and oriented counterclockwise. Then, by the same argument, fC ◦fB ◦fA
is either a translation or the identity map. Figure 9.42 shows that fC ◦fB ◦fA (C ) =
C , so this map is not a translation; it is the identity map. But then
fC ◦ fB ◦ fA = fC ◦ fB ◦ fA ,
so fC = fC , and hence C = C . The problem is solved.

312 9 Isometries
Second solution. Let zk be the complex coordinate of Pk , k ≥ 0, and let w1 , w2 , w3

be the complex coordinates of A, B, C, respectively. Set also = e4π i/3 .
As P1 is obtained from P0 by a clockwise rotation about A of angle 2π/3, which
is the same as the counterclockwise rotation by 4π/3, we have z1 −w1 = (z0 −w1 ).
Hence
z1 = z0 + (1 − )w1 .
Then
z2 = z1 + (1 − )w2 = 2 z0 + (1 − )(w1 + w2 )
and
z3 = z2 + (1 − )w3 = 3 z0 + (1 − )( 2 w1 + w2 + w3 ) =

z0 + (1 − )( 2 w1 + w2 + w3 )
We recognize that z3 is obtained from z0 by a translation by (1 − )( 2 w1 + w2 +

w3 ). Iterating we obtain
z3n = z0 + n(1 − )( 2 w1 + w2 + w3 ), n ≥ 1.
In particular this is true for n = 662. But z1986 = z0 ; hence,
n(1 − )( 2 w1 + w2 + w3 ) = 0.
The only possibility is that 2 w1 + w2 + w3 = 0. And this is the same as

2π i 4π i
w3 + e 3 w1 + e 3 w2 = 0,
which, by Lemma 1.27, implies that w1 , w2 , w3 are the coordinates of the vertices
of an equilateral triangle.
Source International Mathematical Olympiad, 1986, proposed by China, first
solution from D. Djukić, V. Janković, I. Matić, N. Petrović, The IMO Compendium,
Springer, 2006, second solution from T. Andreescu, D. Andrica, Complex numbers
from A to . . . Z, Birkhäuser, Second Ed. 2014.
42 Arguing on Fig. 9.43, we begin with some angle chasing in which we take
advantage of the fact that C1 and C2 have equal radii:

AP B AEB
ADC = ADB = = = ACB = ACD.
2 2
9 Isometries 313
O1 O2
E
D
P
B
C L
Fig. 9.43 Angle chasing in C1 and C2
O1 O2
E
D
F P
B
C L
Fig. 9.44 Analysis of the translation τ
So the triangle ACD is isosceles which implies that the line AL is also the
perpendicular bisector of the segment CD, thus AL ⊥ CD. Also, the equal angles
BAE and BAL are respectively inscribed in C1 and C2 , so BE = BL. Moreover,
since EL ⊥ CD, BD is the perpendicular bisector of EL. This proves that CLDE
is a rhombus and, in particular, it is a parallelogram. As a consequence of this, there
is a translation τ that takes E to D and C to L.
Let us take a closer look at τ examining Fig. 9.44. Because τ takes the chord CE
to the chord LD and because the circles C1 and C2 have equal radii, τ also takes
C1 to C2 , and hence it takes the center O1 of C1 to the center O2 of C2 . Then the
distance between any two points Y and τ (Y ) is equal to O1 O2 .
314 9 Isometries
O1 X1 O2
E
D
F P
B
C L
X2
Fig. 9.45 The construction of X
The definition of F means that the midpoints of the segments P E and DF

−→ −→
coincide, so P DEF is a parallelogram, and hence ED = F P . It follows that τ
takes F to P ∈ C2 , which proves that F ∈ C1 .
We also have P AL = CAE, so P L = CE = CL. Therefore, CLP F is a
rhombus.
Now draw two circles S1 and S2 with centers C and L, respectively, and the
same radius r = O1 O2 , as shown in Fig. 9.45. Then L and F lie on S1 and C and
D lie on S2 . We claim that any of the two intersection points of S1 and S2 can
be chosen as point X (note that the circles intersect because the distance CL = r
between their centers is smaller than the sum of the radii, which is 2r, and greater
than the difference, which is zero). Indeed, the point X satisfies the desired property
because XL = XC = CL = r, so CLX is an equilateral triangle. Then XF L is
inscribed in S1 and is equal to XCL/2 = 30◦ and XDC is inscribed in S2 and
is equal to XLC/2 = 30◦ . The problem is solved.
Remark The first idea of the solution is that circles with equal radii are good
for angle chasing, because of the symmetry in arcs: not only angles inscribed in
the same arc are congruent, but angles inscribed in arcs of the same length are
congruent. Watch how effective this idea is in the way we have proved that triangle
ACD is isosceles and that BE = BL.
Also, two congruent circles can be mapped into each other by all three isometries:
−−−→
we can reflect across AB, rotate about A or B, or translate by the vector O1 O2 .
9 Isometries 315
The parallelograms CLDE and F CLP , which share side CL, suggest that the
translation is the best option here. Indeed, the translation is crucial to prove that
F lies on C1 . It also spreads the length O1 O2 across the diagram, causing both
rhombi to pop up.
Once we discover that these parallelograms are rhombi, the equality F C =
CL = LD gives us a good reason to construct the circles S1 and S2 : the centers
and any of the two intersection points are vertices of an equilateral triangle (aha, this
is where the 30◦ came from!), and both D and F are in at least one of the circles
(inscribed angles!).
Source Iberoamerican Mathematical Olympiad, 2009, proposed by A. Aguilar.
43 First solution. Let ABC be the triangle; reflect it successively over AC, BC,
AB, AC, BC, as shown in Fig. 9.46 (we have labeled the sides a, b, c so that you
can read the figure with ease). A careful examination of the configuration finishes
the proof.
Let us explain. First, it should be noted that each pair of sides of the orthic triangle
that meet at a vertex form with the side of the triangle that passes through that vertex
equal angles (the well-known fact that the altitudes of ABC are the angle bisectors
of the orthic triangle is a consequence of this). Moreover, this property characterizes
the orthic triangle. Indeed, assume that some triangle U V W inscribed in the triangle
ABC (with U ∈ BC, V ∈ AC, and W ∈ AB) has the property that
BU W = CU V , CV U = AV W, AW V = BW U.
Set BU W = CU V = u, CV U = AV W = v, AW V = BW U = w. Then
in the triangles AV W , BU W , CU V , we can write
v + w = 180◦ − A
u + w = 180◦ − B
u + v = 180◦ − C.
Adding we obtain u + v + w = 180◦ , so u = A, v = B, c = C, showing that

the sides of U V W and those of the orthic triangle are parallel. This forces the two
triangles to coincide. We say that the orthic triangle forms a billiard path, since if the
original triangle were a billiard table and a ball were shot from one of the vertices
of the orthic triangle toward another, then the ball would reflect off the sides of the
table and follow a periodic trajectory along the sides of the orthic triangle.
In Fig. 9.46, we have drawn the orthic triangle DEF and some other inscribed
triangle U V W . Because DEF is a billiard path, the successive reflection of its
sides yields the straight line F P from the figure. On the other hand, the successive
reflections of the sides of U V W yield the broken line from W to Q as seen in the
same figure. The length of F P is twice the perimeter of DEF , and the length of
the broken line is twice the perimeter of U V W (because each side of the triangle
316 9 Isometries
a c
b P
a
Q
c
A
c c
V
c a
F E
b
b
W
B D a U C
Fig. 9.46 A proof without words: Hermann Schwarz’s proof of Fagnano’s Theorem
appears twice in the sum). By the triangle inequality, the broken line is strictly
longer than the segment W Q. But the latter is equal to F P because F P QW is a
parallelogram. We infer that the perimeter of DEF , which is half of the length of
F P and hence of W Q, is strictly less than the perimeter of U V W . We conclude
that the orthic triangle has minimal perimeter among all triangles inscribed in a
given acute triangle.
Second solution. Orient the triangle counterclockwise, and let z1 , z2 , z3 be the
complex coordinates of the vertices A, B, C, respectively. The function S : [0, 1] ×
[0, 1] × [0, 1] → R,
S(s, t, u) = |z1 + s(z2 − z1 ) − z2 − t (z3 − z2 )|

+|z2 + t (z3 − z2 ) − z1 − u(z1 − z3 )| + |z3 + u(z1 − z3 ) − z3 − s(z1 − z3 )|
computes the perimeter of a triangle whose vertices have coordinates z1 +s(z2 −z3 ),
z2 +t (z3 −z1 ), and z3 +u(z1 −z3 ), which are the coordinates of some arbitrary points
on AB, BC, CA, respectively. Note that the cube [0, 1]×[0, 1]×[0, 1] is closed and
bounded in R3 , and S is continuous. Hence, S has a minimum in this cube, showing
that a triangle with minimal perimeter exists. We prove that the minimum is attained
for the orthic triangle.
The minimum corresponds to a point that lies either in the interior or on the
boundary of the cube. Let us rule out the second possibility, in which case at least
one of the vertices of the inscribed triangle lies at a vertex of the original triangle,
say at A. Since another vertex of the inscribed triangle must be on BC, the perimeter
of this inscribed triangle must be at least twice the altitude from A. Let us show that
9 Isometries 317
the altitudes of an acute triangle are greater than the semiperimeter of the orthic
triangle.
For this, let D, E, F be the feet of the altitudes from A, B, C, and let the lengths
of BC, CA, AB be a, b, c, respectively. We want to show that DE + EF + F E is
less than twice the altitude from A (Fig. 9.47). Because AE = c cos A, by applying
the Law of Sines in the triangle AEF , we obtain
c
EF = sin A cos A = 2R sin A cos A,
sin C
where R is the circumradius. Similarly, DF = 2R sin B cos B and DE =
2R sin C cos C. Also, in the triangle ABD, AD = c sin B = 2R sin C sin B. The
inequality DE + EF + F E < 2AD reduces to
sin A cos A + sin B cos B + sin C cos C < 2 sin B sin C.
Rewrite this as
sin 2A + sin 2B + sin 2C < 4 sin B sin C,
and then notice that since the triangle is acute, π/2 < B + C < π , so
sin 2A = − sin 2(B + C) = − sin 2B cos 2C − sin 2C cos 2B.
Thus, we have to show that
sin 2B(1 − cos 2C) + sin 2C(1 − cos 2B) < 4 sin B sin C,
or
sin B cos B sin2 C + sin C cos C sin2 B < sin B sin C.
Dividing by sin B sin C, we transform this into the equivalent inequality

sin C cos B + sin B cos C < 1, which is obvious because the left-hand side is
sin(B + C) (and B + C > π/2).
Fig. 9.47 The orthic triangle A
E
F
B D C
318 9 Isometries
Fig. 9.48 Proof of the C

minimum forms a billiard
path
U V
A W W1 B
So the minimum of the function S is not attained on the boundary of the cube,
namely, for a degenerate triangle. Could the minimum be a triangle U V W (U ∈
BC, V ∈ AC, W ∈ AB) different from the orthic triangle? Notice that the minimum
should be a billiard path, meaning that
BU W = CU V , CV U = AV W, AW V = BW U.
Indeed, if one of these inequalities does not hold, say the first, then reflect V across
A to V (Fig. 9.48), and let W1 be the intersection of the segments AB and U V
(they intersect because the triangle is acute). Then the triangle inequality in U W V
implies
U W1 + W1 V = U W1 + W1 V = U V < W V + W U = W V + W U,
so the triangle U V W1 has a smaller perimeter, showing that U V W is not a

minimum. But, as we have seen above, the orthic triangle is the only triangle that
forms a billiard path, so it has to be the minimum.
Remark Note that both arguments fail for obtuse triangles, and indeed, in such a
triangle, the shortest altitude is less than the semiperimeter of the orthic triangle. Do
you see why the proofs fail?
In the second argument, we have used a method for finding the minimum which
is referred to as Sturm’s Principle: given a function f that has a minimum on a
certain domain, if there is a point x0 such that every point x = x0 is not a minimum
for f , then x0 is the minimum of f . It is very important that you check first that f
has a minimum on the given domain, for example, here we could not just work with
the interior of the cube, since it is not true that a continuous function on the interior
of the cube has a minimum. This property is, however, true for the closed cube, as
the Heine-Borel Theorem states that a set in Rn is compact if and only if is closed
9 Isometries 319
and bounded, and a function defined on a compact set has a both a maximum and a
minimum.
Source The first proof of Fagnano’s Theorem was found by Hermann Schwarz (and
was communicated to us by Cosmin Pohoaţă).
44 We argue on Fig. 9.49. Let Q be the isogonal conjugate of P with respect to
the triangle AED. Then, by using the definition of the isogonal conjugate as well as
inscribed angles in ω and , we obtain
(QA, AD) = (EA, AP ) = (EB, BA)
and
(QD, DA) = (ED, DP ) = (EC, CD).
Note that P E tangent to is equivalent to (P E, DE) = (EC, DC), which, via

the isogonal conjugate, translates to (AE, QE) = (EC, DC), and the latter is
equivalent to QECD.
So the problem reduces to showing that QE is parallel to CD. To prove this,
consider the degenerate triangle whose sides are the segment BC and the rays |BA
and |CD, and let R be the isogonal conjugate of E with respect to this triangle.
Then
Fig. 9.49 Isogonal ω

conjugates of P and E with
respect to AED and A B
BC∞
R E Q
D C
Ω
320 9 Isometries
(RB, BC) = (AB, BE) = (P A, AE) = (DA, AQ),
where the first equality follows from the fact that R and E are isogonal conjugates,
the third follows from the fact that P and Q are isogonal conjugates, and the second
equality follows from inscribed angles in circle ω. So RBC = DAQ. Similarly,
RCB = ADQ, which implies that R and Q coincide under a reflection over the
common perpendicular bisector of AB and CD (which is the symmetry axis of the
trapezoid). Since E is on the line that is parallel to BA and CD and at equal distance
from the two, so is its isogonal conjugate R. But then the reflection Q of R over the
perpendicular bisector of AB is also on this line, so EQ is parallel to CD, and the
problem is solved.
Remark The reader should observe that key step of the solution is not the use of
an explicit isometry, but a construction based on isometries: the construction of the
isogonal conjugate.
Source Romanian Master of Mathematics, 2019, proposed by Jakob Jurij Snoj.
45 Denote the lengths of the sides BC, AC, AB of the triangle, respectively, by
a, b, c. We reason on Fig. 9.50.
Let AB C be the reflection of the triangle ABC over the angle bisector of
BAC. Construct the parallelograms AP NB , AC MP , B NMC . Now we are
almost in the setting of the Pappus’ Area Theorem (see Problem 18), except that
two of the parallelograms are constructed internally. Nevertheless, it is still true that
the sum of the areas of the first two parallelograms is equal to the third. This can be
shown as follows.
Translate AP along its line of support to A P where A ∈ B C . Then the
parallelograms AP NB and A P NB have the same area because they have the
same base (AP = A P ) and the same height. Similarly, the parallelograms
N N
B B
C C
P P A
M M
A B A B
C C
Fig. 9.50 Proof of the Erdős-Mordell Theorem

9 Isometries 321
AC MP and A C MP have the same area. But the sum of the areas of A P NB
and A C MP is the area of B NMC , which proves our claim.
Returning the problem, note that the area of AP NB is db · c and the area of
AC MP is dc · b, while the parallelogram B NMC has sides BC and AP and so
its area is less than or equal to BC · AP . We thus have the inequality
db · c + dc · b ≤ a · AP .
Dividing by a, we obtain
c b
db + dc ≤ AP .
a a
Similar arguments yield
a c b a
dc + da ≤ BP and da + db ≤ CP .
b b c c
Adding the three inequalities, we obtain
a
c b c b a
+ da + + db + + dc ≤ AP + BP + CP .
b c c a a b
Using the fact that for x > 0, x + x1 ≥ 2, we obtain that the left-hand side is greater
than or equal to 2(da + db + dc ), and we are done.
Remark The key step of the proof is an inequality for areas proved by shearing,
exactly like in the case of the Pappus’ Area Theorem.
46 The broken tool makes constructions very elaborate. We have to divide the work
into small steps; the steps are presented in the guise of two lemmas.
Lemma 1 Given the points A, B, and C in the plane and a compass with fixed
opening, one can construct a point D such that ABCD is a parallelogram.
−→
Proof We are supposed to translate the point A by the vector BC. Let a be the width
of the opening of the compass. The construction is very simple in the particular case
where AB = BC = a. Indeed, if we construct the two circles of radius a centered
at A and C, one of their intersections is B and the other is the desired point D. Our
intention is to reduce the general case to this particular one.
−→
Let us show how to translate A by the arbitrary vector BC when AB = a. While
moving toward C, by using the compass, we can construct a sequence of points
P1 , P2 , . . . , Pn such that BP1 = P1 P2 = · · · = Pn−2 Pn−1 = a a and Pn−1 is
within a from C. Intersect the circles of radii a centered at C and Pn−1 to obtain Pn
such that Pn−1 Pn = Pn C = a. This is illustrated in Fig. 9.51.
−−→
The point D is obtained from A by a translation of vector BP1 , followed by a
−−→ −−→ −−→
translation of vector P1 P2 , then P2 P3 , . . . , and finally Pn C.
322 9 Isometries
Fig. 9.51 Reduction of the translation of a segment to the simplest case
A A A D
C C
C
B B B
Fig. 9.52 Construction of D such that ABCD is a parallelogram
If AB has arbitrary length, construct Q1 , Q2 , . . . Qn such that BQ1 = Q1 Q2 =

−→
· · · = Qn A = a. Translate Q1 by BC to R1 , so that BQ1 R1 C is a parallelogram,
then Q2 to R2 so that Q1 Q2 R2 R1 is a parallelogram, and so on. The translate of A
will be the desired point D. This entire process is illustrated in Fig. 9.52.

Lemma 2 Given the points A and B in the plane and a compass with fixed opening,
one can construct a point C such that the triangle ABC is equilateral.
Proof Let the width of the opening of the compass be a. We start with an
observation. Given two points M, N in the plane, let N be the rotation of N
−−→
around M by α. Let also τ and ρ be the translation by NN and the rotation
around N by α; then ρ ◦ τ is some rotation by α (Theorem 1.9). If M = τ (M),
then MM N N is a parallelogram. In this parallelogram MN = MN = M N
and M N M = N MN = α, because N is the rotation of N about M by α.
Consequently, ρ ◦ τ (M) = ρ(M ) = M; thus, M is the center of the rotation.
Therefore, ρ ◦ τ is the rotation around M by α. This means that to construct the
image of a point P through the rotation around M by α, you first translate P by
−−→
N N and then rotate around N by α.
9 Isometries 323
B B
B
A
A A
Fig. 9.53 Construction of the equilateral triangle
Fig. 9.54 Construction of D D

such that AB⊥CD
A B
C
Our goal is to rotate B around A by 60◦ . Choose the points P0 = A, P1 , . . .,

Pn = B such that P1 P2 = P2 P3 = · · · = Pn−1 Pn = a. One can easily construct
an equilateral triangle P0 P1 Q1 , with Q1 being the rotation of P1 around P0 = A
by 60◦ . Indeed, you just intersect the circles of radius a centered at P0 and P1 .
Then translate the broken line P1 P2 . . . Pn to a broken line that starts at Q1 . By the
observation made at the beginning of the proof, C is the rotate of the end of this
broken line around Q1 by 60◦ . We have reduced the problem from a broken line
with n segments to a broken line with n − 1 segments. Repeating we produce in the
end the point C. These steps are shown schematically in Fig. 9.53.

Now we solve the actual problem. The construction can be followed on Fig. 9.54.
Using Lemma 2 construct M and N such that AMB and ANB are equilateral. Then
MN is perpendicular to AB. Next translate MN to CD using Lemma 1. The line
CD is parallel to MN so it is perpendicular to AB, and we are done.
324 9 Isometries
Fig. 9.55 The case where the U

contact points are
diametrically opposite
Remark In other words, D is obtained as the image of B under a composition of a

−→
60◦ rotation about A followed by a translation by AC.
Note that with the rusty compass, you cannot construct the 90◦ rotation of B
about A.
Source Lemma 1 was given at the Bundeswettbewerb Mathematik in 1977.
47 Let us first understand what “making it impossible to move P” means, in the
language of isometries. Let P1 , P2 , . . . , Pk be the points where the nails have been
placed. Then the impossibility of moving the polygon P means that any small
movement f forces f (P) to contain at least one of the points Pi . Here “small
movement” means any orientation-preserving isometry with small parameter (small
vector in the case of a translation, small angle in the case of a rotation).
To gain insight, we examine some particular cases. It is clear that we need three
nails for a triangle. What about a square? Were there less than four nails, one of the
sides would have no nails at all, and we could translate the square in the direction
normal to this side, so we need at least four nails. And four nails at the midpoints of
the sides fix the square.
The idea that solves the general case is the use of inscribed circles. That is, to
consider a maximal circle ω contained in P and investigate the points at which ω
touches the sides of P; these points will be the candidates for nails. With this in
mind, let H be the convex hull of these tangency points.
First consider the case where two vertices of H, say, U and V , are such that
U V is a diameter of ω, as in Fig. 9.55. Place two nails at U and V . The sides that
contain U and V are parallel (both perpendicular to U V ), so the only movement
these nails allow is a translation perpendicular to U V . Indeed, any small translation
not perpendicular to U V would make f (P) cover either U or V : just consider the
component parallel to U V . Also, any small clockwise rotation with center to the left
−→
of U V would make f (P) cover V and so on. Two additional nails, one to the left of
P and one to the right of P, prevent P from moving to the left or to the right.
What if this is not the case? Let O be the center of ω. Arguing by contradiction,
suppose that O is not inside H. Let P Q be the side that separates O from H and
9 Isometries 325
Fig. 9.56 The case where the

contact points are not
diametrically opposite V
U
O
A A
B C C B
Fig. 9.57 Cutting the cake
let the tangents to ω at P and Q meet at K. A homothety (contraction) with center

K and ratio slightly larger than 1 keeps ω inside P, a contradiction. Hence, O is
inside H. Dissect H into triangles, and let U V W be the triangle that contains O, as
shown in Fig. 9.56. Placing nails at U , V , and W we can see that they fix the triangle
formed by the tangents to ω at these points (any small movement but a rotation with
center O makes ω cover one of these points, and a small rotation with center O
makes the polygon P cover all three points). We conclude that three nails fix the
polygon P.
Therefore, the answer to the problem is 4.
Remark Here we think of geometric transformations as “geometry in motion.”
Source 10th Sharygin Geometry Olympiad, 2014, proposed by N. Beluhov and
S. Gerdgikov, Russia.
48 Cut the cake along the perpendiculars from the incenter to the sides into pieces
A, B, C arranged in counterclockwise order. Now place these slices in clockwise
order around the incenter, as shown in Fig. 9.57. The result is the mirror image of
the triangle.
Remark Recall the Wallace-Bolyai-Gerwien Theorem which says that the problem
has solution if we do not limit the number of slices. The problem is made interesting
by requiring the use of exactly three slices.
326 9 Isometries
Fig. 9.58 Invariance under

rotation
Source V.G. Boltyanskii, Hilbert’s Third Problem (transl.R. Silverman), V.H. Win-
ston, Washington, DC, 1978.
49 The blue square can be mapped to the red square by a rotation about the center
of the blue square that brings it to the correct orientation, followed by a translation
by the vector defined by the centers of the two squares.
We check first that the sum of the red sides is equal to the sum of blue sides in the
case where only the rotation about the center is performed. In our figures, we draw
the red square with continuous line and the blue square with dotted line. Examining
Fig. 9.58, we see that four equal right triangles of the red square are left outside of
the octagon. They are similar (because of equal angles) to the four triangles of the
blue square that are left outside of the octagon. But since the squares have equal
areas, the areas of the triangles cut out from the red square are equal to the areas of
the triangles cut out from the blue square. So all eight triangles are equal, and hence
they have equal hypotenuses. Thus, the sum of the four red hypotenuses equals the
sum of the four blue hypotenuses, proving the equality of the sums of the sides of
each color in this case.
Next we will show that the sum of the red sides and the sum of the blue sides of
the octagon are both invariant when a translation is performed. In fact we only have
to check the invariance of the sum of the blue sides, because the two squares are
translated one with respect to the other, so the situation is symmetric.
Observe that the translation can be written as the composition of two translations,
each in the direction of a pair of opposite sides of the red square (horizontal and
vertical in Fig. 9.59). So it suffices to check invariance under translations parallel to
the sides (see Fig. 9.59).
It is important that after each translation, the two squares still overlap to form
an octagon. And because a pair of parallel lines crossed by another pair of parallel
lines determine equal segments (because the segments form parallelograms), we see
that the segments that are lost on one side are gained on the other (examine the pairs
of boldface parallel segments in Fig. 9.59!). Hence, the sum of the segments of one
color is preserved under translations, and the problem is solved.
9 Isometries 327
Fig. 9.59 Invariance under

translation
Fig. 9.60 A red triangle

congruent to a black triangle
Remark Here we think about geometric transformations as “geometry in motion.”

We first check an easy case, then transform the easy case into the general case by
a translation, and check invariance of the sum of red (respectively blue) segments
under the translation. To this end, we decompose an arbitrary translation into simpler
translations and check invariance under the simpler translations. We use the fact that
the group of translations is generated by horizontal and vertical translations (with
the “horizontal” and “vertical” chosen at our discretion).
An idea worth remembering: to check invariance under a group of transforma-
tions, it suffices to check invariance under the group generators.
Source Kvant (Quantum), proposed by Vyacheslav Viktorovich Proizvolov.
50 An example of a coloring, with the black vertices indicated by black dots, is
shown in Fig. 9.60.
Consider the rotations ρk , k = 1, 2, . . . , n − 1, of the polygon by the angles
2π k/n. Let ai be the number of red vertices that are mapped to black vertices by
the rotation ρi . To solve the problem, it suffices to show that there is i such that
ai ≥ p/2 + 1. Because when performing the n − 1 rotations each red vertex
overlaps with exactly n − p black vertices (because that is how many black vertices
there are), we must have
a1 + a2 + · · · + an = p(n − p).
328 9 Isometries
But n − p ≥ p implies n ≥ 2p, so n > 2p − 1. A little algebra shows that this

inequality implies
p(n − p) p
> .
n−1 2
Thus, the average of the ai ’s is larger than p/2, proving that some ai is greater than
or equal to p/2+1. But wait, we are not yet done! We must also show that we have
indeed a polygon, and not just a segment. We do have a polygon if p/2 + 1 ≥ 3.
However, it is also possible that p/2 + 1 < 3, which happens precisely when
p = 3. In that case, when n ≥ 8 we automatically have
3(n − 3)
> 2.
n−1
The cases n = 6, p = 3, and n = 7, p = 3 should be examined case-by-case (there

are only seven cases to check and they are left to the reader).
Remark Note the use of the Pigeonhole Principle.
Source Romanian Mathematical Olympiad, 1995.
51 (a) We denote the set by P and its area by S. Place the square so that one of
its corners is at the origin and two sides are aligned along the positive x- and y-
axes. Let P1 and P2 be the translates of P by two vectors of length 0.001, the first
of which pointing in the positive direction of the x-axis, and the other making a
60◦ angle with the first, as sketched in Fig. 9.61. Then P, P1 , and P2 are mutually
disjoint, while P1 and P2 are also mapped into each other by a translation whose
vector has length .001 (observe the equilateral triangle formed by the translation
vectors). All three surfaces lie inside a square of side 1.001; hence, 3S < 1.0012 .
We obtain S < 0.335.
(b) We have to improve the estimate from (a), and we do this using the four vectors
from Fig. 9.61. The vectors −→
v3 and −→
v4 make an angle of 2 arcsin √1
and have length
√ −
→
2 3
0.001 × 3, with v pointing in the positive direction of the x-axis. Let P and P
3 3 4
Fig. 9.61 The set P, made of

a triangle, quadrilateral, and
pentagon, and its translates
9 Isometries 329
Fig. 9.62 The four vectors

by which we translate the set v4
P
v5
v3
v6
be the translates
√ P by these vectors. The two vectors form a triangle with sides
of √
.001 × 3, 0.001× 3, 0.001, so P4 can be obtained from P3 by a translation by a
vector of length 0.001. It follows that P3 and P4 do not intersect (Fig. 9.62).
Of P3 and P4 , choose the one whose overlap with P has the smaller area; let
this be P3 . Then the area of the overlap of P and P3 is at most S/2, so the total
area covered by P and P3 is at least 3S/2. Now consider the translates P5 and P6
of P in the directions of the vectors − →
v5 and −→
v6 which have lengths equal to 0.001
◦ −
→
and make angles of 30 with v3 . Then P5 and P6 are also mapped into each other
by a translation whose vector has length 0.001 (notice the equilateral triangle). So
P, P5 , P6 do not intersect each other. Moreover, both P5 and P6 translate to P3 by
vectors of length 0.001, so they do not intersect P3 . It follows that P, P3 , P5 , P6
altogether cover a total area that is at least S√+ S/2 + S + S = 7S/2. These sets lie
inside a square of side-length 1 + 0.001 × 3 < 1.0018. Therefore,
2
S< (1.0018)2 < 0.287,
7
and the inequality is proved.
Remark The given condition means that every translation P of P by a vector of
length 0.001 does not intersect P. So, in (a), we use three copies of P each of which
is a translation by a vector of length 0.001 of the others. Of course, the best we can
do in the plane is using an equilateral triangle.
To improve that, in (b) we try to add a new translation, forming two equilateral
1
triangles (hence, the angle 2 arcsin √ ). But with that, we introduce some overlaps
2 3
between translates. This is not desirable, since we do not have much control of the
overlaps. Having two nonoverlapping options P3 and P4 gives us some control, as
P would intersect the least with one of them.
A natural question is what the optimal set P is. We can construct an example
to obtain a lower bound by tessellating the plane with copies of a regular hexagon
H with side-length 0.001/2 − (and thus diameter 0.001 − 2) surrounded by
strips of width 0.001/2 + (assuring that points from different hexagons are at least
0.001 + 2 apart) and taking P to be the union of the hexagons H in the tessellation
within the square. This gives us an area arbitrarily close to
330 9 Isometries
√ 2
3/2 √
√ = 21 − 12 3 > 0.215.
1 + 3/2
So the optimal set P has area between 0.215 and 0.287.

Source Kvant (Quantum), proposed by G.V. Rozenblium.
52 Let us show that there is a monochromatic set (i.e., a set colored by one color)
that is mapped to itself by some reflection over a point. Arguing by contradiction,
assume there is no such infinite set. Then every reflection over a point changes the
colors of all but finitely points (maybe none); call these points “exceptional.”
Now consider two arbitrary reflections, σ1 and σ2 , over the points O1 and O2 ,
respectively. Consider two lines 1 and 2 parallel to O1 O2 such that σ1 (1 ) =
σ2 (1 ) = 2 and such that all the exceptional points of σ1 and σ2 are between 1
and 2 , as shown in Fig. 9.63.
−−−→
The composition σ1 ◦ σ2 is a translation by 2O1 O2 , and it transforms every point
in the exterior of the band bounded by 1 and 2 into a point of the same color. If
−→ −−−→
X is such a point, then all points Y with XY = 2k O1 O2 , with k ∈ Z, are of the
same color, because they are obtained by reflecting X and even number of times.
But then the set of Y ’s is infinite and monochromatic and has a center of symmetry
(which can be any of the Y ’s). Hence, our assumption was false, and so there exists
an infinite, monochromatic set with a center of symmetry.
Remark Can you adapt this argument to show that in the same conditions there is a
monochromatic set having an axis of symmetry?
Source Moscow Mathematical Olympiad, 1996-1997, proposed by V. Protasov.
53 Let the disks be Di , 1 ≤ i ≤ n, and let D be their union. We argue on Fig. 9.64.
Construct first a tessellation of the plane by congruent regular hexagons Hj , j ≥
1, chosen so that the distance between the centers of two neighboring hexagons is
slightly greater than 4, say 4 + 2. Let τj be the translation mapping H1 to Hj ,
j ≥ 1.
Fig. 9.63 The region 1

between 1 and 2 containing
all exceptional points
O1 O2
exceptional
2
9 Isometries 331
τj τj (x)
Di j
4 + 2a x
Fig. 9.64 How to pick the disks using the tessellation of the plane by hexagons
Given a point x in the plane, if its translates τj (x) and τk (x) belong to disks
Dij and Dik from the family, then Dij and Dik must be disjoint due to the fact that
the distance between τj (x) and τk (x) is greater than 4, while the disks Dj have
diameters equal to 2. Thus, for a point x, we could choose all disks to which the
translates of x belong, and these disks would then be pairwise disjoint. Our aim is
to pick an x that maximizes the number of such disks, and hence their total area.
To this end, let Mj = D ∩ Hj and let Fj = τj−1 (Mj ). The Fj ’s, j ≥ 1, lie all
inside H1 and so there should be a lot of overlap (unless there are very few disks in
the family, in which case the problem is trivial). Let n(x) be the number of sets Fj
that cover point x and let n = maxx n(x). Then
√
S= Area[Mj ] = Area[Fj ] ≤ nArea(H ) = 2n 3(2 + )2 .
j j
It follows that
S
n≥ √ .
2 3(2 + )2
If we choose an x that is covered by n sets Fj , then the disks that contain

translates of x have total area equal to nπ , and the above inequality implies
Sπ
nπ ≥ √ .
2 3(2 + a)2
√
All we need to do is choose a sufficiently small so that π/2 3(2 + a)2 is greater
than 2/9. This is possible because, when a = 0,
π 2
√ > .
8 3 9
332 9 Isometries
Note that this last inequality can be proved easily by squaring both sides and then
observing that π 2 > 9.6 > 256/27.
Remark The key idea is to look at the problem “modulo H1 .” The points P and
Q are “congruent modulo H1 ” when Q = τi (P ) for some i. In this sense, since
disks that contain different translates of x are disjoint, we transform a problem of
nonoverlapping disks into a problem of maximum overlap of translates; we reduce
the problem “modulo H1 .” Now it becomes a matter of computing overlapping
areas, which is essentially averaging.
A related problem is Minkowski’s Theorem: every convex set in Rn that is
symmetric with respect to the origin and has volume greater than 2n contains a
point of integer coordinates other than the origin. Try to prove it!
Yugoslavia.
54 Assume that the sum of the lengths of the chords is greater than or equal to kπ .
Then the sum of the lengths of the arcs subtended by these chords is greater than
kπ . The key idea is to add to the picture the reflections of these arcs over the center
of the circle. The sum of the lengths of all arcs is now greater than 2kπ , and so
there exists a point covered by at least k + 1 arcs. The diameter through that point
intersects at least k + 1 chords, contradicting our assumption. Hence the conclusion.
Remark Adding the reflections of the chords over the center of the circle was
essential for being able to apply the Pigeonhole Principle. Note that a diameter
intersects an arc if and only if it intersects the reflection of the arc over the center of
the circle (see Fig. 9.65).
Source Kvant (Quantum), proposed by A.T. Kolotov.
55 Call the people who are not enemies friends. Choose an arbitrary assignment
of places in which two neighbors, A and B, are enemies. Assume that A sits
to the right of B. A has at least n friends; choose n of them: A1 , A2 , . . . , An .
Each of A1 , A2 , . . . , An has one person sitting to its left; denote these people by
B1 , B2 , . . . , Bn (some Bk ’s can overlap with Ak ’s). A certain Bk is necessarily a
friend of B (because B has at most n − 1 enemies).
Fig. 9.65 A diameter

intersects an arc if and only if
it intersects its reflection
9 Isometries 333
A Ak
k
Bk A
B A B B
k

Fig. 9.66 The reflection of the arc ABk

In this case, take the arc ABk of the round table that lies inside the arc BAk , and
reflect it, so that the order of people lying in that arc is reversed (see Fig. 9.66). Now
Bk sits next to B and A sits next to Ak , so we have reduced the number of enemy
pairs by 1. Because there are only finitely many enemy pairs, this move cannot be
repeated forever. So after repeating this move sufficiently many times, we arrive at
an arrangement where all neighbors are friends.
56 Consider the isometries of the cylinder that are the compositions of a translation
in the direction of the axis of the cylinder and a rotation of the cylinder around its
axis. Imagine instead that the figure is fixed and the points move on the cylinder, all
rigidly linked to each other. Let P be one of the n points; when another point traces
S, P itself traces a figure congruent to S. So after all points traced S, P alone traced
a surface F of area strictly less than n.
On the other hand, if we rotate P around the cylinder or translate it back and
forth by 4πn r , we trace a surface of area exactly equal to n. Choose on this surface
a point P that does not lie in F , and consider the isometry that maps P to P . The
fact that P is not in F means that at this moment none of the points lies in S. This
transformation, therefore, satisfies the required condition.
Remark All the tracing described in the first paragraph describes the forbidden
positions of S, using P as a reference; in fact, by tracing S with each point, we are
doing exactly what we do not want S to do. This is useful since it greatly organizes
and simplifies the reasoning; after all, we are trying to find a possible position for S,
and using one of the points can be useful.
Since we only want to prove that the transformation is possible, we do not
actually need to find it explicitly. So we resort to areas, again using P as a reference.
Because the surface we obtain has area greater than the forbidden positions, we find
a suitable position for S. This is very similar to applying the Pigeonhole Principle,
with area; notice that this principle is useful to prove that something exists, but not
to find an explicit example.
334 9 Isometries
Although this problem is not exactly about isometries of the plane, it is almost,
since the cylinder can be produced by rolling the plane the way you roll a carpet,
and the transformations that we use come from translations of the plane. Again we
think about geometric transformations as “geometry in motion.”
Source M. Pimsner, S. Popa, Probleme de geometrie elementară (Problems in
elementary geometry), Ed. Didactică şi Pedagogică, Bucharest, 1979.
57 We claim that the number tn is equal to k 2 − k if n = 2k and k 2 if n = 2k + 1.
In fact

n−1
tn = tn−1 + .
2
Let us first check that

n−1
tn ≥ tn−1 + .
2
To this end, place the guests at the vertices of a regular n-gon A1 A2 . . . An . The
final configuration is obtained from the initial configuration by an orientation-
reversing isometry; from Theorem 1.10, we deduce that it is the reflection of the
initial configuration over some axis . Without loss of generality, we may assume
that An is one of the vertices at greatest distance from , and let Aj be its image
under the reflection over . Between An and Aj , there are k = n−1 2 sides of
the polygon, so to bring An to the location of Aj , at least k seat exchanges are
required, and these all happen between the guest An and some other guest. But the
guests A1 , A2 , . . . , An−1 need themselves tn−1 swaps to reverse their own order.
This proves the inequality.
To show that equality is actually achieved, start by exchanging the locations
of An and An−k by swapping successively An with An−1 , An−2 , . . . , An−k and
then swapping successively An−k with An−1 , An−2 , . . . , An−k+1 . Using the same
reflection axis but ignoring An and An−k , perform the same moves on the remaining
polygon (with vertices labeled in the same order). Notice that the moves do not
require crossing the locations of An and An−k . Repeat inductively until all vertices
are in the desired position.
An easy induction shows that
n

j −1
2
j =1
is equal to k 2 − k if n = 2k and to k 2 if n = 2k + 1.
Remark Another way to prove the inequality is to observe that in order for An and
Aj to switch places, they need at least k + (k − 1) jumps, while the other guests
9 Isometries 335
Fig. 9.67 Partitioning of the

triangle Si
Ai
Ti
Bi Ci
must realize the same reflection but this time with fewer guests. Thus, we obtain an
inequality of the form tn ≥ tn−2 + 2k − 1, which yields the same conclusion.
Source Kvant (Quantum), proposed by V.B. Alexeev.
58 Partition each triangle Si into four congruent triangular regions, Ai , Bi , Ci , and
Ti , i = 1, 2, . . . , n, arranged as in Fig. 9.67. Define A = (A1 ∪ A2 ∪ . . . ∪ An ) \ T ,
and define B and C similarly. Notice that S = A ∪ B ∪ C ∪ T . We will prove that
area(A) ≤ area(T ).
Let i be the line that contains the common segment of Ai and Ti . We can
suppose, without loss of generality, that the horizontal lines i are ordered from
top to bottom, dividing the plane into n + 1 strips.
Partition A into sets A(1), A(2), . . . , A(n) according to the strip (or Ai ) they are
in. More precisely, A(1) = A1 and A(i) is the set of points from Ai that are not in
any of the triangles A1 , A2 , . . . , Ai−1 .
Let T (i) be the image of A(i) under the reflection over i . It is clear that T (i) ⊆
Ti and that Area[T (i)] = Area[A(i)]. We now prove that T (i) and T (j ) are disjoint
whenever i = j . Suppose without loss of generality that i < j and, for the sake of
contradiction, suppose that there is a common point X of T (i) and T (j ). Let Xi and
Xj be the reflections of X over i and j , respectively. Notice that X, Xi , and Xj
are collinear, because they belong to a line perpendicular to all lines k . From the
definitions of T (i) and T (j ), we have that Xi ∈ A(i) and Xj ∈ A(j ). Since i < j ,
Xj is not in Ai .
Also, X is below j and so Xj lies between X and Xi . Notice that Ai and Ti cover
the segment XXi (since X ∈ Ti and Xi ∈ Ai ), so Xj is in Ti . But Xj ∈ A(j ) ⊆ A,
and so, by its definition, Xj cannot be in Ti , a contradiction.
Therefore, T (1), T (2), . . . , T (n) are disjoint, and hence
Area[A] = Area[A(1)] + Area[A(2)] + · · · + Area[A(n)]

= Area[T (1)] + Area[T (2)] + · · · + Area[T (n)]
= Area[T (1) ∪ T (2) ∪ . . . ∪ T (n)] ≤ Area[T ].
One proves similarly that Area[B] ≤ Area[T ] and Area[C] ≤ Area[T ]. We

obtain
336 9 Isometries
Area[S] = Area[T ] + Area[A] + Area[B] + Area[C] ≤ 4Area[T ],
as desired.
Source Shortlist of the Iberoamerican Mathematical Olympiad, 2010, proposed
by Argentina, used in a Brazilian Team Selection Test for the International
Mathematical Olympiad in 2011.
59 (a) The proof is based on the following result:
Lemma Let be a polygonal line in the plane with endpoints A and B, and
consider a number α ∈ (0, 1). Then among all segments with endpoints in that
are parallel to AB, there is either one of length αAB or one of length (1 − α)AB.
Proof (a) Let us choose the x-axis to be parallel to AB and let us assume that
AB = 1. For β > 0 we denote by τβ the translation of the plane to the right by β
units. The problem reduces to showing that cannot be disjoint from both τα and
τ1−α . Rephrasing, we have to show that τα cannot be disjoint from both and
τ1 . Suppose it were. We argue with the aid of Fig. 9.68.
Choose p, q ∈ τα having respectively maximal and minimal y-coordinates. Let
L+ be a vertical ray extending upward from p and let L− be a vertical ray extending
downward from q. Let L be the (infinite) polygonal line consisting of L+ , L− and
the polygonal line that is part of τα and runs between p and q, i.e.,
L = L+ ∪ (pq) ∪ L− .
Then L separates the plane into two regions. Moreover, lies to the left of L, while
τ1 lies to the right. But ∩ τ1 = ∅, since B is obviously in this intersection. We
reached the contradiction that shows that our assumption was false. This proves the
lemma.
Fig. 9.68 Proof of the Chord

Theorem
L+
Γ Tα Γ
T1 Γ
L−
9 Isometries 337
Returning to the original problem, we conclude that there is a parallel segment

of length n1 AB or one of length n−1
n AB.
If the second case holds true, then by repeating the argument for the segment
n−1 1
n AB and the number n−1 , we conclude that there is either a segment of length
1 n−1 1
· AB = AB,
n−1 n n
or one of length
(n − 1) − 1 n − 1 n−2
· AB = AB.
n−1 n n
If we continue the argument, we eventually reach a segment equal to n1 AB and

parallel to AB.
(b) Let us show that for every α ∈ (0, 1) not of the form n1 with n an integer, there
is a polygonal line joining points A = B such that there is no segment parallel to
AB, with endpoints on , and equal to αAB.
Choose n such that α ∈ ( n+1 1
, n1 ). Consider the segment AB, and consider points
P0 = A, P1 , P2 , . . . , Pn = B on AB such that for all j = 0, 1, . . . , n − 1, the
segment Pj Pj +1 has length n1 AB and points Q0 = A, Q1 , . . . , Qn = B such that
for all j = 0, 1, . . . , n, the segment Qj Qj +1 has length n+1 1
. Now draw a family
of parallel lines 0 , 1 , . . . , n through P0 , P1 . . . , Pn , respectively, and another
family of parallel lines, L1 , L2 , . . . , Ln through Q1 , Q2 , . . . , Qn−1 , respectively,
so that the two families are not parallel to each other. Let Cj be the intersection of
j −1 and Lj , and let Dj be the intersection of j and Lj , j = 1, 2, . . . , n. Then
AC1 D1 C2 D2 . . . Dn B is a polygonal line with the desired property (prove it!). The
example for n = 2 is shown in Fig. 9.69.
Remark The easiest way to prove facts about parallel segments is usually by
performing translations of those segments, and that is exactly what we do here. It is
natural to try to relate the translation by αAB and its “complementary” translation
by (1 − α)AB; this is the subject of the lemma. Also, the key idea in the lemma
is shifting αAB units to the left and (1 − α)AB units to the right, so we force a
translation by AB, which naturally takes A to B.
Finally, the proof of (a) is based on the lemma as well as on the identity
1 1 2 3 n−2 n−1
= · · ··· · .
n 2 3 4 n−1 n
Fig. 9.69 Counterexample

for the second part of the
Chord Theorem
338 9 Isometries
A2
A1
A3
a B4
B3 B2 B1
A4
b A5
Fig. 9.70 A possible trajectory of the flea
Fig. 9.71 The vectors of the

jumps determine the
trajectory
a Bj
Aj v
b
Source This result is well-known (see, e.g., D. Rolfsen, Knots and Links, AMS
Chelsea Publ. 2003) and is true for any curve, not just polygonal lines.
60 Figure 9.70 illustrates a possible trajectory of the flea.
−−−→ −−−→ −−−→
Looking at the sequence of vectors that describe the jumps, A1 B1 , B1 A2 , A2 B2 ,
. . ., we realize that by knowing this sequence, we can recover the path of the flea.
−−−→
For if we know the vector − →v = Aj Bj , then the point Bj can only be the intersection
of the line b with the translate of the line a by −
→
v , and the point Aj is the preimage
of Bj under this translation (see Fig. 9.71).
Thus, for the purpose of bookkeeping, we can translate all the vectors to originate
−−→ −−→ −−→ −−→
at some point O, and we let these translates be, in order, OC1 , OD1 , OC2 , OD2 ,
. . . This procedure is shown in Fig. 9.72. We can say something about the positions
of these vectors.
Lemma Assume that the line a transforms into the line b by a counterclockwise
−−→ −−−−→
rotation of angle α. Then, for all j , the angle between OCj and OCj +1 and the
−−→ −−−−→
angle between ODj and ODj +1 are both equal to −2α.
Proof The main difficulty of the proof is that there are many possible configura-
tions. We will argue in a way that covers all cases simultaneously, reasoning on
Fig. 9.73.
9 Isometries 339
D1
A2
D2
A1
A3 C4
O
a B4 D3
B3 B2 B1 C3
A4 C1
D4 C2
b A5
Fig. 9.72 The construction of the vectors OCj , ODj , j ≥ 1
Bj Cj
Bj
b
a O b
C j+1
A j+1 Aj a
A j+1
B j+1
Dj Dj
Fig. 9.73 Proof of the lemma
Let a and b be the translates of a and b that pass through O. Then because
−−→ −−→
the triangle Bj Aj Aj +1 is isosceles, OCj and ODj are symmetric with respect to
−−→ −−−−→
the line a . Similarly, ODj and OCj +1 are symmetric with respect to the line b .
Hence, Cj is mapped to Cj +1 by the composition of two reflections over lines. This
composition is a clockwise rotation by an angle that is twice the angle between the
two lines. The lemma is proved.

Returning to the problem, the path of the flea is periodic if and only if the
sequences C1 , C2 , . . . , Cn , . . . and D1 , D2 , . . . , Dn , . . . are both periodic. By the
lemma, this happens if and only if there is n such that 2nα is a multiple of 360◦ .
Remark The construction that associates to the trajectory of the flea the points
C1 , D1 , C2 , D2 , . . . on the circle is a particular example of the spherical image of
a curve. In general, if t → γ (t) is a smooth curve, its spherical image is obtained
by normalizing all velocity vectors γ (t) to have length 1 (i.e., by considering the
vectors γ (t)/γ (t)), and then mapping these vectors to have the same origin. The
result is a curve on the unit sphere. The usefulness of the construction comes from
the fact that the length of the spherical image equals the total curvature of the curve.
61 The second player has a strategy. We call an almost-circuit a polygonal line that
is missing one segment in order to become closed. Here is the strategy of the second
player: choose one diagonal, and then for whatever segment the first player marks,
340 9 Isometries
mark the reflection of that segment over the diagonal. Continue until you notice the
presence of an almost-circuit, and then close up the polygonal line.
Remark An example of a game is shown in Fig. 9.74, where the second player
reflects over the diagonal marked at the beginning. On an n × n table with n odd,
the second player can also use the reflection with respect to the center.
Source This is the first half of a problem published in Kvant (Quantum) by I. Vetrov
and A. Kogan
62 We can assume that p and q are coprime, otherwise shrink the size of the
chessboard by the greatest common divisor of p and q as the knight is confined
to a lattice of square-size equal to this greatest common divisor.
Consider the Klein 4-group, which is the group of symmetries of a (nonsquare)
rectangle discussed in Sect. 1.1.5, with e the identity, a and b the reflections over
the perpendicular bisectors of the sides, and c the reflection over the center of the
rectangle. Color the chessboard as in Fig. 9.75.
If p and q are both odd, then at each jump, the color of the location of the knight
is multiplied by c. Thus, after n jumps, the knight is on a square colored by cn . The
initial square was colored by e, and the equality cn = e is only possible if n is even.
If one of p and q is even and the other is odd, then at each jump, the color of the
square is multiplied by either a or b. After n jumps, the color will be a k bn−k , for
some k. The equality a k bn−k = e implies a k = bn−k so both k and n − k have to be
even. So n itself has to be even, and we are done.
Remark This problem shows how to distinguish configurations using invariants. We
distinguish squares where the knight can land from those where it cannot by their
“color” (which “color” is an element of the Klein 4-group).
Source Bundeswettbewerbe Mathematik.
63 In both cases, the first player has a winning strategy, and it works on an m × n
chocolate bar with m = 2k, k ∈ N.
Fig. 9.74 An example of a game

9 Isometries 341
Fig. 9.75 Coloring of the c

chessboard by elements of the b c b c b
Klein 4-group. a e a e a e
c b c b c b
a e a e a e
c b c b c b
a e a e a e
Fig. 9.76 In each of the two

figures, the shaded rectangles
have the same number of a s,
b s, c s
(a) The winning strategy for the first player is to first cut the bar into two equal
parts of size k × n, and then play symmetrically.
(b) The first player can still play symmetrically until the second produces a 1 ×
bar, > 1, and then cut out a 1 × 1 square.
Remark Note the similarity with Problem 61. The “symmetrization strategy” is
commonly used in games.
Source Kvant (Quantum), proposed by S.V. Fomin.
64 We assume that such a coloring exists, and try to understand its structure.
Working in some region away from the boundary, we can reason under the
assumption that the board is infinite. We start with two key observations, illustrated
in Fig. 9.76:
(1) If two 1 × 4 rectangles coincide by a horizontal or vertical translation by 3 units
in the direction of their shorter side, then they must contain an equal number of
a’s, b’s, and c’s.
(2) If two 1 × 3 rectangles coincide by a horizontal or vertical translation by 4 units
in the direction of their shorter side, then they must contain an equal number of
a’s, b’s, and c’s.
This is because in both cases, the rectangles can be added to the same
rectangle to create a 3 × 4 rectangle. The solution consists of a repeated
application of these “translations” in order to determine the possible locations
of the a, b, c’s. We state, then prove, a series of claims.
Claim 1 In every 1 × 3 rectangle, there is at most one a.

For simplicity, we argue inside the rectangle [0, 4] × [0, 5] shown in Fig. 9.77,
with [0, 3] × [0, 1] being the rectangle in question (by flipping the figure, we can
adapt this argument to vertical rectangles as well as those close to the upper side
342 9 Isometries
Fig. 9.77 Why a 1 × 3

rectangle contains at most
one a
Fig. 9.78 Why a 1 × 4

one a
a a
a a
of the 100 × 100 square). If the 1 × 3 rectangle has two or more a’s, then after we
complete the rectangle to [0, 4] × [0, 1] and use the first observation, we deduce
that the rectangle [0, 4] × [3, 4] contains at least two a’s. By using the second
observation, we deduce that the rectangle [0, 3] × [4, 5] also contains at least two
a’s, and so the rectangle [0, 4] × [3, 5] contains at least 4 a’s, impossible.
Claim 2 In every 1 × 4 rectangle, there is at most one a.
To prove this, we argue on the rectangle [0, 4] × [0, 6] shown in Fig. 9.78. If
in the rectangle [0, 4] × [1, 2] there are more than one a’s, then there are exactly
two, by Claim 1 applied to [0, 3] × [1, 2] and [1, 4] × [1, 2], and they must be in
the squares [0, 1] × [1, 2] and [3, 4] × [1, 2]. Then the rectangle [0, 4] × [4, 5] has
the same property with the a’s in the corresponding locations. But then the rectangle
[0, 4]×[2, 4] contains exactly one a, so the rectangles [0, 3]×[0, 1] and [0, 3]×[5, 6]
have an a each. But then the same is true for their translates by 3 units [0, 3] × [3, 4]
and [0, 3] × [2, 3]. This forces [0, 4] × [1, 4] to have four a’s.
Claim 3 In every 1 × 4 rectangle, there is exactly one a.
We have seen that there is at most one a. Arguing on the same [0, 4] × [0, 6]
rectangle, we see that if [0, 4] × [1, 2] and [0, 4] × [4, 5] have no a’s, then [0, 4] ×
[2, 3] and [0, 4] × [3, 4] have together three a’s, which is impossible, since each has
at most one a.
9 Isometries 343
Fig. 9.79 Why a 1 × 4

two b’s
Claim 4 In every 1 × 4 rectangle, there are at most two b’s.

In the rectangle [0, 4] × [0, 6] (Fig. 9.79), if [0, 4] × [0, 1] contains three b’s,
then so does [0, 4] × [3, 4]. The second “translation” property forces the rectangle
[0, 3] × [5, 6] to have at least two b’s, so [0, 4] × [3, 6] has at least 5 b’s, which is
not allowed.
Claim 5 In every 1 × 4 rectangle, there are at most two c’s.
We argue on [0, 4]×[0, 5]. If in [0, 4]×[0, 1] there are three or more c’s, then this
is also true for [0, 4] × [3, 4]. Moreover, there are at least two c’s in [0, 3] × [4, 5].
In [0, 4] × [2, 3], there is at least one c because there is at most one a and two b’s.
So the rectangle [0, 4] × [1, 4] has at least six c’s, a contradiction.
Claim 6 In every 1 × 3 rectangle, there is at most one b.
Indeed, if in the rectangle [0, 3] × [0, 1] there are two or more b’s, then so are in
the rectangles [0, 4] × [3, 4] and [0, 3] × [4, 5]. But then the rectangle [2, 5] × [0, 4]
has five b’s, because by Claims 3 and 5, the rectangle [0, 4] × [2, 3] has at most two
c’s and one a, so it must contain a b as well.
Claim 7 In every 1 × 3 rectangle, there is exactly one b.
If the rectangle [0, 3] × [0, 1] has no b, then in each of [0, 3] × [k, k + 1], k =
1, 2, 3 there is at most one b by Claim 6, so [0, 3] × [0, 4] has at most three b’s,
which is not allowed.
With these observations at hand, let us analyze a possible configuration. There
must be many squares containing an a, so let us start with one of these somewhere
in the middle of the table. Either to the left or to the right of it, there is a b, because
every 1 × 3 rectangle contains exactly one b. Let this be to the right (Fig. 9.80).
Then the next two on that row must be c’s, so we have the configuration abcc.
Immediately after the second c should follow an a by Claim 3 and at the same time
a b by Claim 7. This is impossible! Hence, there is no way we can produce the
desired coloring.
Remark The condition from the statement allows us to see patterns, structures that
repeat periodically. The translation invariant pattern “equally spaced 1 × 3 and 1 ×
344 9 Isometries
Fig. 9.80 Explanation for

why such a configuration a b c c ?
does not exist
Fig. 9.81 Coloring by five colors
4 rectangles that contain an equal number of a, b, c’s” completely identifies the

configuration, but leads to contradictory constraints.
Source Kvant (Quantum), proposed by V.E. Lapitski, solution by Yu.I. Yonin.
65 Let us consider a coloring as described in (a), such as the one with five colors
depicted on the left of Fig. 9.81.
Each color forms a square lattice, and the square lattices can be translated to
overlap with one another. Reasoning on Fig. 9.81, we take the two orthogonal
vectors, v and w, that determine any of these lattices, and define an equivalence
relation on the points of the plane, declaring two points to be equivalent if one is
translated into the other by some vector of the form r v + s w,
with r and s integers.
The equivalence relation identifies points of the same color, and all squares of the
same color are identified with one another. Moreover, the equivalence classes are
parametrized by the points in one square S formed by v and w, meaning that every
point in the plane is identified with one and only one point in S. Well, not quite, the
points on opposite sides of this square are identified under the translations by v and
But if we glue the opposite sides, so as to obtain a torus1 (Fig. 9.82), then the
w.
points of the torus are in one-to-one correspondence with the equivalence classes.
So the torus parametrizes the equivalence classes.
This torus contains precisely one unit square for each color, so its √ area is n, the
number of colors. Consequently the side-length of the square S is n. Examining
the original lattice, we see that the side of S can be placed in a right triangle with
integer sides, by centering the vertices of S at the centers of the colored squares. In
this configuration, let us say that there are m units on the vertical and k units on the
horizontal. By the Pythagorean Theorem,
n = m2 + k 2 .
1A torus is the surface that bounds a donut.

9 Isometries 345
Fig. 9.82 The torus that parametrizes the equivalence classes
Fig. 9.83 Tessellation of the

plane by hexagons of three
and four colors
This gives a necessary condition for n: to be the sum of two perfect squares.
Conversely, let us show that every positive integer n of the form m2 + k 2 has an
associated coloring. Color one unit square by the first color, and then move m units
to the right and k units up. Color this square by the same color. Then complete a
square lattice containing these two squares. Next, choose an uncolored unit square,
color it by the second color, and repeat the above construction. Repeating with each
of n = m2 + k 2 colors, we obtain the desired coloring.
Part (b) is similar to (a) in the sense that it uses translations parallel to the sides
of the equilateral triangular lattice. In this case, the role of the square S is played by
a rhombus R with a 60◦ angle, which again parametrizes the equivalence relations,
namely, each point in the plane is parametrized by one and only one point on this
rhombus, provided that again we identify the opposite sides, to obtain again a torus
(you have to stretch the torus, and your imagination, to see this happening). Let the
distance between √ the center of two neighboring hexagons be 1, so that√the area of
one hexagon is 3/2. The torus contains √ n hexagons; thus, its area is n 3/2. If the
side of the rhombus is a, its area is a 2 3/2; thus, a 2 = n.
Examining the configurations from Fig. 9.83, we notice that to move one hexagon
of the tessellation to another, you have to translate it some m “steps” vertically
and k “steps” up to the left or right at an angle of 60◦ with the horizontal, with
m, k nonnegative integers. The length of the total translation vector, which is a, is,
therefore, the third side in a triangle whose other two sides have lengths m and k
and make an angle of 120◦ . Using the Law of Cosines, we obtain that
346 9 Isometries
n = a 2 = m2 + mk + k 2 .
This is a necessary condition for the coloring to exist, but it is also sufficient, because
this reasoning also discloses the algorithm for the coloring: color one hexagon by
some color, go m steps vertically and k steps to the left and up by 60◦ , and then
color the hexagon where you land by the same color. Complete the lattice, and then
repeat for every other color.
Remark What if we allow lattices of different sizes? And if we do not impose that
two lattices have parallel sides?
Source Kvant (Quantum), proposed by Andrey Nikolaevich Kolmogorov.
66 The situation is described in Fig. 9.84. We will label the trees counterclockwise
by the numbers from 0 to n − 1 (these are the residue classes modulo n). Let the
trees be equally spaced around the unit circle in the complex plane, so that the 0th
2π ik
tree is at 1. Then the kth tree is at the kth root of unity: e n (Fig. 9.85).
Denote by ρ the counterclockwise rotation by 2πn i . The flight of a bird from the
kth tree to the mth tree corresponds to a rotation by 2π i(m−k)
n ; it is realized by ρ m−k .
Assume that at the beginning all birds are on the tree number 0. To bring one
bird on the tree number k, we must act upon it by ρ k . Now for every configuration
of birds, let us consider the composition, in the group of rotations about the origin,
of the rotations that have brought these birds from the tree number 0 to this
configuration. For example, for the initial configuration from our problem, this
product is equal to
n(n−1)
ρ 0 ◦ ρ 1 ◦ ρ 2 ◦ · · · ◦ ρ n−1 = ρ 2 .
In general, this product is
Fig. 9.84 Birds on trees

2
1
3
0
4
5 n− 1
·· ·
6
9 Isometries 347
Fig. 9.85 Modeling the flight 2

of birds using rotations
1
ρ
0
n− 1
5

n−1
N0 N1 Nn−1 kNk

ρ 0
◦ ρ 1
◦ ··· ◦ ρ n−1
=ρ k=0 ,
where Nk is the number of birds on kth tree.

Note that the product of a counterclockwise rotation and a clockwise rotation
by 2πn i is just ρ ◦ ρ −1 , which is the identity map. So the move described in the
statement, where two birds change their location, the first by an action of ρ and the
second by ρ −1 , does not change the product. So this product is an invariant, which
does not change under the move.
When the birds accumulate on the kth tree, the product is
k
ρ kn = ρ n = 1k = 1,
where 1 denotes the identity map. So the birds can accumulate on a single tree only
if
n(n−1)
ρ 2 = 1.
This is equivalent to n(n−1)

2 ≡ 0(mod n), which happens if and only if n is odd.
Thus, the birds cannot accumulate on the same tree when n is even.
And indeed, when n is odd, that is, n = 2m + 1, the birds can accumulate on
the 0th tree in the following way: for each k = 1, 2, . . . , m, the birds on the kth and
m − kth tree fly toward each other until that arrive on the 0th tree.
Remark We can ask a more general question, given two configurations of birds
on trees, when can they be transformed into each other by the move specified in
the statement? Clearly, for this to happen, the above invariant should have the same
value for the two configurations. We will show that this is also a sufficient condition.
Let the value of the invariant be ρ r , with r one of the numbers 0, 1, 2, . . . , n. We
348 9 Isometries
will show that any configuration having this invariant can be transformed into the
one that has all birds on the 0th tree except for one which is on the rth tree. Indeed,
pick a bird, which we call the “flyer.” Pair each bird with the flyer, and let the two
fly in opposite directions until that bird lands on the 0th tree. After all birds are on
the 0th tree, the flyer must be on some tree, and given that the invariant has the value
ρ r , this must be the rth tree.
We are in the presence of what is called a complete invariant: two configurations
can be transformed into each other if and only if they have the same value of the
invariant. In mathematics complete invariants are rare, but they are very desirable.
You should note the similarity with the problem about balls in a box discussed in
the introduction.
Source Kvant (Quantum), this problem was given at the university entrance exam
at the Moscow State University.
67 Translate A to the left by k = max A (i.e., transform A → A − {k}) and B
to the left by = min B (B → B − {}) so as to obtain the sets A and B with
max A = min B = 0. All sums in A + B are translated by k + , so |A + B | =
|A + B|, and, of course, |A| = |A | and |B| = |B |. And we can argue on A + B
instead, which is significantly easier! This is because all elements of A and B are
in A + B , as 0 ∈ A and 0 ∈ B , and also A ∩ B = {0}. We therefore have
|A + B | ≥ |A ∪ B | = |A | + |B | − 1, proving the inequality.
For the equality case, let t = 0 be the element from A ∪ B of minimal absolute
value; suppose without loss of generality that t ∈ B, so that t > 0. Notice that
because we are in the equality case, A + B = A ∪ B . Consider the translation
A → A + {t}. Since max(A \ {0}) + t < 0, A + {t} consists of A ∪ {t} from which
we exclude min A . Thus, A consists of the elements of an arithmetic progression
with common difference t. Now that we have established that the maximal nonzero
element of A is −t. One can also deduce that the elements of B are in arithmetic
progression as well, by noticing that B + {−t} is B ∪ {−t} from which we exclude
max B . Reversing the translation, we have the answer |A + B| = |A| + |B| − 1
if and only if the elements of each of sets A and B are in arithmetic progressions
with same common difference (the arithmetic progressions do not need to overlap,
though).
Remark Translating A and B does not alter the cardinality of A + B, so we can shift
them in a convenient manner, so as to have A ∩ B = {0}.
68 Consider the translates A = A + {−a} and B = B + {b}. Then condition
(ii) is the same as A ∪ B ⊆ A ∪ B . But we have the following inequality for
cardinalities: |A ∪ B | ≤ |A | + |B | = |A| + |B|, and, since A and B are disjoint,
|A| + |B| = |A ∪ B|. Thus,
|A ∪ B| ≤ |A ∪ B | ≤ |A | + |B | = |A| + |B| = |A ∪ B|.
This means that |A ∪ B| = |A ∪ B | and that A and B are disjoint. It follows
that A∪B = A ∪B . So we have four sets A, B, A , B so that A∩B = A ∩B = ∅
9 Isometries 349
and A ∪ B = A ∪ B . If s(X) is the sum of the elements of X, then summing over

A ∪ B and A ∪ B , which are equal, we have
s(A) + s(B) = s(A ) + s(B ).
This is equivalent to
s(A) + s(B) = s(A + {−a}) + s(B + {b}),
and this latter equality is equivalent to
s(A) + s(B) = s(A) − a|A| + s(B) + b|B|.
This yields a|A| = b|B|, and we are done.

Remark The trick of the solution was to compare the elements of A and B and their
union to the elements of their translates A + {−a} and B + {b} and their union.
Source Asian Pacific Mathematical Olympiad, 2013.
69 The number c has the property that no integer −k can be found in the translates
by c of any of the intervals [ai , bi ]. The inequality from the hypothesis of the
problem can be written as

n
(bi − ai ) < 1,
i=1
meaning that the sum of the lengths of the intervals [ai , bi ] is less than 1.
So the problem asks us to show that the union S of finitely many nondegenerate
intervals with total length less than 1 may be translated so as not to intersect the set
Z of integer numbers. For k ∈ Z, consider the sets Sk = S ∩ [k, k + 1) and their
translates Sk − {k} ⊂ [0, 1). The total length of the union ∪k∈Z (Sk − {k}) is at most
equal to the total length of S (due to overlaps, the total length of the first might be
smaller). So the length of ∪k (Sk − {k}) is less than 1. If c is a number in [0, 1) that
does not lie in this union, then S + c does not intersect Z.
Remark The proof generalizes to n dimensions to show that a union of several solid
bodies of total volume less than 1 can be translated so that it does not contain any
point of integer coordinates.
Source Romanian Team Selection Test for the International Mathematical
Olympiad, 2008.
70 We will prove that k = 2. Let f be a permutation, which we want to write as
the product of two involutions. We know that f can be written as the composition
of disjoint cycles, so let
350 9 Isometries
f = g1 ◦ g2 ◦ · · · ◦ gr ,
where gj are the disjoint cycles, 1 ≤ j ≤ r.

Identify a cycle of length m with the rotation ρ of the regular m-gon
A0 A1 . . . Am−1 by 2πm about its center. Then this rotation is the composition of two
reflections: first the reflection over the perpendicular bisector of A0 A1 , followed
by the reflection over the perpendicular bisector of A0 A2 . The two reflections are
themselves permutations (as they map the polygon to itself), but what is more
important is that they are involutions.
Thus, we can write gj = σj ◦ σj , where σj and σj are involutions. Setting
σ = σ1 ◦ σ2 ◦ · · · σr , σ = σ1 ◦ σ2 · · · σr ,
we have that σ and σ are involutions (because they are products of involutions on
disjoint sets) and f = σ ◦ σ .
Remark This is an application of Theorem 1.8, but it is important to notice that the
given rotation can be written as a composition of two reflections in many ways. But
we need to choose reflections that map the regular polygon to itself, because only
these correspond to permutations.
71 First solution. Let us assume that this is possible. Denote by A the set that
contains 1. The other sets are obtained from A by translations by positive integers.
We group the integers that define translations in a set B, to which set we add the
element 0. Then every positive integer can be represented uniquely as a + b with
a ∈ A and b ∈ B.
Conversely if A, B are infinite sets of nonnegative integers such that 1 ∈ A and
0 ∈ B and if every integer is represented uniquely as a + b with a ∈ A and b ∈ B,
then the translates of A by elements of B determine a partition of positive integers
into sets congruent to each other.
To construct an example, let us change slightly the problem by working with
nonnegative integers instead of positive integers. To this end, we translate to the
left by 1 unit, so A must now contain the number 0. We then define A to be the
set of nonnegative integers whose binary expansion has 1s only in odd positions,
including the number 0, and let B be the set of nonnegative integers whose binary
expansion has 1s only in even positions, including 0. The sets A and B have the
desired property, so the answer to the question is affirmative.
Second solution. Quite analogous to the above construction, we can start with 1 and
then construct an infinite array as follows: at the first step, write 2 to the right of 1, 3
below 1 and 4 to the right of 3. In general at the nth step, translate the 2n−1 × 2n−1
array obtained so far to the right, down, and diagonally as to obtain a 2n × 2n array,
and then add 4n to the elements in the 2n−1 × 2n−1 block on the left, 2 × 4n to
the elements in the 2n−1 × 2n−1 block below, and 3 × 4n to the elements in the
2n−1 × 2n−1 block diagonally opposite to the original block. The second step is
shown below:
9 Isometries 351
1 2 5 6
12 3 4 7 8
→ .
34 9 10 13 14
11 12 15 16
The columns of the array obtained at the end of the process give the required
partition. The rows of the array give another partition.
Remark The second solution becomes even more transparent if you write the
numbers in binary form.
Source Kvant (Quantum), proposed by A. Fedorov, solutions by A. Fedorov and
S. Slosman.
72 We begin the solution by describing a combinatorial device that encodes both
the key-string and the resulting string of characters.
Each string of key presses corresponds to a rooted planar tree with letters labeling
the non-root vertices. The algorithm is as follows. First, add to the plane a coordinate
system so that we can talk about up-down and left-right. Given a key string, start
at the root, which will be an unlabeled vertex and serves now as the current vertex.
When pressing a key α on the keyboard, draw a vertex directly underneath the root,
connected to the root by an edge, and then label this vertex by the letter α. This
now becomes the current vertex and pressing a key would produce a new vertex
underneath this one. Pressing the key ← moves us up one vertex on the tree, thus
moving the current vertex one step up. From this moment on, the pressing of a
literal key would produce a vertex below the current vertex, connected to it by an
edge and positioned to the right of all other vertices connected by an edge to the
current vertex, and this new vertex becomes the current vertex. An illustration of
the graph produced by the instances of key pressings
a, ab, ab ← cd, ab ← cd ← e, ab ← cd ← e ←← f,
which gives rise to the word faecdb is shown in Fig. 9.86.

Given such a tree, one constructs the associated key-string by doing a left-to-
right depth-first search. This means that you start at the root, go down all the way
on the leftmost branch, writing the vertices in the order they are encountered from
left to right in the string (in our case ab). Then backtrack until reaching a vertex
from which more branches bifurcate downward, where at each upward step, an ←
is written to the right of the string (in our case ab ←). Then travel on the new
leftmost branch downward, writing to the letters encountered in the order from
left to right (in our case ab ← cd). Continue until all vertices are exhausted. The
backtracking can be easily realized by drawing a contour around the tree and then
traveling counterclockwise around this contour starting at the root (see Fig. 9.87)
while skipping all vertices that were already recorded.
352 9 Isometries
root root root root root
a a f
a a a
b c e
b c c e
b b
d
d d
Fig. 9.86 The encoding of a sequence of key strokes
Fig. 9.87 Contour around a root

tree
a f
b c e
Lemma The string produced by a key-string is obtained from the corresponding

rooted tree via a right-to-left depth-first search (meaning that we travel on the same
contour but this time counterclockwise).
Proof We induct on the number of vertices of the tree (i.e., the number of letters in
the key-string). The base case is trivial, so let us prove the induction step.
Let us assume that the property holds true for all smaller trees and let us prove
it for the tree in question. Deleting the root, we obtain a family of sub-trees
A1 , A2 , . . . , Ak rooted at the “children” of the root (where the numbering is from
left to right). If k = 1, then both the string and the key-string start with the letter
at the unique “child” of the root, and we can apply the inductive hypothesis to the
sub-tree A1 .
If k > 1, the path that defines both the left-to-right and the right-to-left depth-
first searches decomposes into the paths corresponding to the trees A1 , A2 , . . . , Ak ,
and, after surrounding each of the Aj ’s, the cursor is again at the beginning in the
case of the key-string, or at the end, in the case of the string. So the concatenation of
paths corresponds to the concatenation of both the key-strings and the strings, with
the caveat that at the beginning of each of the subpaths, we have to add the letter
corresponding to its root as well (both to the key-string and to the corresponding
string). This completes the induction, and the lemma is proved.

9 Isometries 353
We conclude that to every pair of words (A, B) with B reachable from A, we can
associate a tree T such that
• the left-to-right depth-first search starting with the root yields A, and
• the right-to-left depth-first search yields B.
The key element that solves the problem, and which is the reason why this problem
is included in a book on geometric transformations, is the reflective symmetry
between the left-to-right and right-to-left depth-first searches. So if B is reachable
from A and the tree for the pair (A, B) is T , and if we let T be the reflection of T
(over the y-axis), then T corresponds to the pair (B, A) showing that A is reachable
from B. Problem solved.
Remark Depth-first searches are widely used in combinatorics, for example, this is
the strategy for solving a maze.
Source United States of America Team Selection Test for the International Mathe-
matical Olympiad, 2014, proposed by Linus Hamilton, solution by contestant James
Tao.
73 Look at all possible colorings of the vertices of a regular p-gon by n colors (see
Fig. 9.88). This set has np elements.
We call two colorings equivalent if they can be obtained from each other by
a rotation of the polygon. Except for monochromatic colorings, each coloring is
equivalent to exactly p − 1 others. It is here where we use the fact that p is prime,
because if a rotation maps a coloring into itself, then all rotations map the coloring
into itself (this is because any nontrivial rotation of the regular p-gon generates the
group of all rotations).
Thus, if a coloring overlaps with itself after a nontrivial rotation, then by repeat-
edly applying that rotation, we find that coloring coincides with all its rotations and
that can only happen if it is monochromatic. Therefore, the colorings that are not
monochromatic can be partitioned into equivalence classes of p elements, which
implies that their total number, which is np − n, must be divisible by p.
Fig. 9.88 A coloring of the b

vertices of the regular
heptagon by three colors
a
a
c
a
b c
354 9 Isometries
Fig. 9.89 A configuration for

p = 5, a = 12 and b = 8
Remark Note the similarity with the proof of Wilson’s Theorem. It is important to
observe that the proof fails for p not prime, because there exist non-monochromatic
colorings that are invariant under some nontrivial rotations (which are they?).
Source This proof to the Fermat’s Little Theorem was given by Julius Petersen.
74 We start by writing a = a0 + α1 p and b = b0 + β1 p with 0 ≤ a0 , b0 < p and
α1 , β1 some nonnegative integers that are not necessarily less than p. Consider the
vertices of α1 regular p-gons and, in addition to that, a0 points that are not among
these vertices. Altogether we have a0 + α1 p points, of which we mark b0 + β1 p
points. An example is shown in Fig. 9.89, with the marked points specified by the
larger dots.
We allow the α1 polygons to rotate independently about their centers so that each
polygon is mapped into itself, but we keep the other a0 points fixed. Let us count
the number of configurations with b0 + β1 p marked points that are invariant under
these transformations. The only way a configuration is invariant is if on any regular
polygon either all vertices are marked or no vertex is marked (see the solution to
the previous problem). This can only happen if we markall vertices of β1 polygons,
and then choose b0 of the remaining a0 points. So of all ab configurations, there are
α1 a0
β1 b0 that are fixed by the transformations.
It follows that the number of configurations that are not fixed is

a a0 α1
− .
b b0 β1
But when on a given circle we have marked some but not all of the points, then the
configuration of those points and its p − 1 rotates are all distinct configurations.
It follows that the configurations that are not fixed can be grouped into families of
configurations that are mapped into one another by the given transformations, and
the number of members of each family is a multiple of p. Hence,

a a0 α1
−
b b0 β1
is a multiple of p. Note that if b0 > a0 , then there are no fixed configurations, and
also the second term is zero because there is no way you can choose b0 objects out
of a0 .
9 Isometries 355
Now we can prove by induction on k that if a = a0 + a1 p + a2 p2 + · · · +

ak−1 pk−1 + αk pk , b = b0 + b1 p + b2 p2 + · · · + bk−1 pk−1 + βk pk with 0 ≤
aj , bj < p, j = 0, 1, . . . , k − 1 and αk , βk ≥ 0, then

a a0 a1 αk
≡ ··· (mod p).
b b0 b1 βk
The base case for k = 1 was proved above. For the induction step, assume the
property is true for k and let us prove it for k + 1. Write a = a0 + a1 p + a2 p2 +
· · · + ak−1 pk−1 + αk pk , b = b0 + b1 p + b2 p2 + · · · + bk−1 pk−1 + βk pk with
0 ≤ aj , bj < p, j = 0, 1, . . . , k − 1, and write αk = αk+1 p + ak , βk = βk+1 p + bk .
Then, by using the induction hypothesis and the base case, we have

a a0 a1 αk a0 a1 ak αk+1
≡ ··· ≡ ··· (mod p),
b b0 b1 βk b0 b1 bk βk+1

Remark The proof can be done in one step by arranging the a0 +a1 p +a2 p2 +· · ·+
ak pk points as follows: first put the a0 points somewhere. Then consider the vertices
of a1 regular p-gons. At each of the vertices of a2 regular p-gons, place (formally)
one tiny regular p-gon. In this configuration, we have rotations that move the tiny
regular polygons by the rotations of the regular polygon that has them as vertices,
and rotations of the tiny regular polygons themselves, and we look at configurations
that are invariant under all these transformations. For a3 we iterate the process with
regular polygons placed at the vertices of regular polygons that are placed at the
vertices of regular polygons, and so on.
75 We will concentrate on the points of integer coordinates in the plane and among
those we mark the ones that have both coordinates nonnegative. For every positive
integer n, we let dn be the line of equation
ax + by = n.
Next, we mark the lines that contain at least one marked point. As an illustration,
Fig. 9.90 shows a particular case with marked lines shown in solid line and the
unmarked lines shown in dotted line.
Now we focus on two families of isometries of the plane that map the lines dn
into one another:
• translations by (p, q): (x, y) → (x + p, y + q) and
• reflections over the points (p/2, q/2): (x, y) → (p − x, q − y),
where p and q range among integers.
356 9 Isometries
Fig. 9.90 The case of the

family of equations
2x + 5y = n
Fig. 9.91 The translation of

2x + 5y = 0 by the vector
(1, 2)
Lemma If (x0 , y0 ) is a point of integer coordinates on the line dn , then the points of
integer coordinates on dn that are closest to it are (x0 −b, y0 +a) and (x0 +b, y0 −a).
Proof Consider the line d0 , which passes through (0, 0). Let (−b1 , a1 ) be the
integer point on this line that is the closest to (0, 0) such that a1 , b1 > 0.
Translating by the vector (x0 , y0 ), the segment from (0, 0) to (−b1 , a1 ) on d0
becomes the segment from (x0 , y0 ) to (x0 − b1 , y0 + a1 ) on dn (see Fig. 9.91).
Then (x0 − b1 , y0 + a1 ) must be the closest to (x0 , y0 ) in the upward direction (or
else we can translate the closest point backward to the line d0 and contradict the
minimality of (−b1 , a1 )).
Similarly, by translating by the vector (x0 + b1 , y0 − a1 ), the segment from (0, 0)
to (−b1 , a1 ) becomes the segment from (x0 + b1 , y0 − a1 ) to (x0 , y0 ). And again we
must have minimality for a downward segment starting at (x0 , y0 ). We have reduced
the problem to the line d0 and the point (0, 0). From here we deduce that the points
of integer coordinates on any of the lines dn are equally spaced, with the distance
between consecutive points independent of n. In other words, the distance between
two consecutive integer points on dn is a1 units on the y-axis and b1 units on the
x-axis.
Since (−b, a) is on d0 , it follows that b = tb1 and a = ta1 . But a and b are
coprime, so t = 1, that is, a1 = a, b1 = b. The lemma is proved.

9 Isometries 357
Fig. 9.92 The slice

0≤x≤4
Returning to the proof of the theorem, from the lemma, it follows that each dn
passes through exactly one point in the slice
S = {(x, y) | 0 ≤ x ≤ b − 1}.
If the line is marked, then necessarily a marked point on this line must be in this
slice. So we have a bijection between lines and their associated points in S, and
a line is marked if its associated point is marked. The reader can follow this on
Fig. 9.92.
2 , −2 :
1
The reflection σ over b−1
(x, y) → (b − 1 − x, −1 − y)
maps the slice S to itself, and it maps the marked points in this slice to unmarked
points and vice versa. Moreover, because
a(b − 1 − x) + b(−1 − y) = ab − a − b − ax − by,
we have
σ (dn ) = dab−a−b−n .
It follows that of the lines dn and dab−a−b−n , exactly one is marked. This proves
that for every n, exactly one of the numbers n and ab − a − b − n can be represented
as ax + by with x, y nonnegative integers.
Finally, it is clear that the lowest marked line is d0 , and so the highest unmarked
line must be dab−a−b . This proves that ab − a − b is the largest positive integer that
cannot be represented as ax + by.
Remark The geometric proof might suggest why solving the same problem for more
variables is considerably more difficult, as repeating the argument in a space of three
or more dimensions is impossible.
358 9 Isometries
Source This proof to Sylvester’s Theorem was published by Nikolai Borisovich

Vassiliev in Kvant (Quantum).
76 We prove the inequality by induction on the cardinality |A| of the set A. The
base case |A| = 1 is obvious. Now let us assume that the inequality holds for all
pairs (A, B) with |A| < n, and let us prove it when |A| = n.
Place the residue classes modulo p at the vertices of a regular polygon. We want
to rotate A until it intersects B, but in such a way that A does not lie entirely inside
B. Assuming that this is not possible, then the rotates of A are either disjoint of B
or lie inside B. Because the rotates of A cover the circle, some will be disjoint from
B and others will lie entirely inside B.
Because A has at least two elements, we can find x, y in B that are rotates of
elements in A. Let ρ be the rotation that maps x to y. Then ρ(x) = y is in B,
so ρ(y) = ρ 2 (x) must be in B too (because there is some rotation that maps two
elements of A to y and ρ(y)). Inductively ρ n (x) ∈ B for all n, and since
{ρ n (x) | n > 0} = Zp ,
it follows that B = Zp . In this case, A + B = Zp , and the inequality is satisfied

trivially.
Now let us assume that we are in the other case, where a rotation of A intersects
B but does not lie entirely in B. Then we replace the pair (A, B) by (A , B ) =
(A ∩ B, A ∪ B). In the process, we have transferred the elements of A\(A ∩ B) from
A to B. So |A| + |B| = |A | + |B |. But
A + B = [A + (A ∩ B)] ∪ [B + (A ∩ B)]
and this lies inside A + B. Thus, |A + B | ≤ |A + B|. Now we use the induction
hypothesis to conclude that
|A + B| ≥ |A + B | ≥ min(|A | + |B | − 1, p) = min(|A| + |B| − 1, p).
Remark Compare the Cauchy-Davenport Theorem to Problem 67.

77 We will construct g and h that fulfill the desired equality f = g + h in such a
way that g is an odd function (having the graph symmetric with respect to the origin)
and h is a function whose graph is symmetric with respect to the point (1, 0).
Let g be any odd function on the interval [−1, 1] for which g(1) = f (1). Define
h(x) = f (x) − g(x), x ∈ [−1, 1]. Now proceed inductively as follows (Fig. 9.93).
For n ≥ 1, let h(x) = −h(2 − x) and g(x) = f (x) − h(x) for x ∈ (2n − 1, 2n + 1],
and then extend these functions such that g(x) = −g(−x) and h(x) = f (x) − g(x)
for x ∈ [−2n − 1, −2n + 1). By construction g and h satisfy the required condition,
and we are done.
Remark Certainly, the centers of symmetry of the graphs of g and h have to be
distinct, or else the graph of f = g + h would have the same center of symmetry.
9 Isometries 359
-5 -3 -1 -1 1 3 5 7 9
Fig. 9.93 The inductive procedure for constructing f and g
Fig. 9.94 Graph of a

function that is invariant
under a 90◦ rotation about the
origin
Once we decide that the centers of symmetry are distinct, then the alternating
inductive construction is dictated by the reflective property of the graphs.
Source Kvant (Quantum).
78 The function f : R → R, f (0) = 0 and
x
f (x) = − (−1)−|x| x, x = 0,
|x|
where · is the greatest integer function, satisfies the required property. Because
the algebraic expression is rather difficult to visualize, we have drawn the graph of
the function in Fig. 9.94.
Note that because the graph is invariant under a 90◦ rotation about the origin, it
is also invariant under the square of this transformation, which is the 180◦ rotation
about the origin. This means that the graph is invariant under the reflection over the
origin, so the function f is odd. Hence, f (0) = 0.
360 9 Isometries
Remark The 90◦ rotation switches the x and y coordinates. Playing with the
translates and rotates of f (x) = x becomes natural.
Source Kvant (Quantum), proposed by A. Karinskii.
79 As required by the problem, we will show that f (A) = 0 for every point A
in the plane. To this end, we consider two finite groups of rotations: the group G

of rotations ρk of center A and angles 2kπ n , k = 0, 1, . . . , n − 1, and group G

of rotations ρj of center some point B different from A and the same angles n , 2j π
j = 0, 1, . . . , n − 1. Define Akj = ρk (ρj (A)), k, j = 0, 1, . . . , n − 1. Note that

Ak0 = A for all k.
There are two types of regular n-gons that arise in the configuration: the regular
n-gon obtained by acting on A with the elements of G and the rotates of this n-gon
by elements of G (shown with continuous line in Fig. 9.95) and the regular n-gons
that arise from making G act on some ρj (A) (shown with dotted line in Fig. 9.95).
The key observation is that the sum of the values of f at the vertices of each of these
regular n-gons is zero, and because A is fixed by G, it will follow that f (A) = 0.
Here are the details:

For each k, Ak0 Ak1 . . . Ak,n−1 is a regular n-gon, so n−1 j =0 f (Akj ) = 0, and
hence

n−1
n−1
f (Akj ) = 0.
k=0 j =0
This can be rewritten as

n−1
n−1
nf (A) + f (Akj ) = 0.
k=0 j =1
We can change the order of summation:
Fig. 9.95 Regular n-gons

that arise after rotations
9 Isometries 361

n−1
n−1
n−1
n−1
nf (A) + f (Akj ) = nf (A) + f (Akj ).
k=0 j =1 j =1 k=0
Now observe that for every j ,
A0j A1j . . . An−1,j = ρ0 (ρj (A))ρ1 (ρj (A)) . . . ρn−1 (ρj (A))
n−1
is a regular n-gon, so k=0 f (Akj ) = 0, and then

n−1
n−1
f (Akj ) = 0.
j =1 k=0
We deduce that nf (A) = 0, so f (A) = 0. Since A was arbitrary, f is identically

equal to zero.
Remark Rephrasing in complex coordinates, the problem states that if

n−1 2π i

f a + be n j = 0,
j =0
then f (z) = 0 for all z ∈ C.

Olympiad, 1996, proposed by Geffry Barad.
80 Assume first that the axes of symmetry are parallel, and let m be their common
slope (which is either a nonzero number or infinite when the axes are vertical).
Then the graph of f is invariant under the composition of the two reflections, which
is a translation in the direction perpendicular to the axes of symmetry of length
twice the distance between the lines (see Fig. 9.96). So the graph is invariant under
the translation in the direction of a line of slope −1/m. Because f is continuous,
f (x) − (−1/m)x must be bounded, so
f (x) (−1/m)x 1
lim = lim =− .
x→∞ x x→∞ x m
If the two axes of symmetry are not parallel, then the composition of the
reflections over the two axes is a rotation ρ by twice the angle between them. Let
us first consider the case where the axes are not orthogonal. Then the angle α of
rotation is different from 180◦ . Let us consider the group
G = {ρ k | k ∈ Z}
362 9 Isometries
Fig. 9.96 A graph that is translation invariant
Fig. 9.97 A graph that is

rotation invariant
generated by ρ. It is either infinite or cyclic, depending on whether α is rational or

irrational. Consider a point p on the graph of f . Then ρ k (p) must be on the graph
of f for every k.
If G is infinite, then ρ k (p) is dense in a circle, and since f is continuous, this
would mean that the graph of f contains a circle, which is absurd.
If G is cyclic, say of order n, then ρ k (p), k = 0, 1, . . . , n, is a regular n-gon.
Because f is continuous, the graph of f contains an arc of a curve from p = ρ 0 (p)
to some ρ k (p) that contains no other vertices of the n-gon. Then the rotates of this
arc are also on the graph (Fig. 9.97). Consequently, the graph contains a cycle, which
again is impossible.
We are left with the situation where the two symmetry axes are orthogonal. If
there is a point p on the graph of f that does not belong to either of the axes,
then by reflecting it over the symmetry axes and over the intersection point of the
symmetry axes, we obtain three more points that are on the graph of f . On the graph
of f , there is an arc that connects p to one of these three points such that neither of
the remaining two points is on the arc. If we take this arc together with its reflections
over the symmetry axes and over their intersection point, we produce a cycle on the
graph of f , which is impossible.
So the only possibility is that the points on the graph of f lie on the union of
the two symmetry axes. The graph is either V -shaped (which is ruled out by the
9 Isometries 363
symmetry requirement) or it is a line. In the latter case, f is a linear function, and

hence it satisfies the desired property.
Remark In Problem 78, we have seen that the graph of a function can be invariant
under the group of rotations by 0◦ , 90◦ , 180◦ , 270◦ , but if we impose continuity, an
example can no longer be constructed, as explained above.
The fact that a subgroup of the rotation group about a point is either finite and
cyclic or it is given by a set of rotation angles that are dense in [0, 2π ) is a direct
consequence of Kronecker’s Theorem, which was proved in the solution to the
example from the introduction.
Source Răzvan Gelca.
81 With the substitution x = cos t, the equation from the statement reads
√ √ π
P ( 2 cos t) = P 2 cos t − .
4
√
After a new substitution z + 1z = 2 cos t, we turn P ( 2 cos t) into a Laurent
polynomial in z (i.e., a polynomial in z and z−1 ),

1 1
Q(z) = P √ z+ .
2 z
The functional equation implies that Q(z) is invariant under the group generated by
πi
the 45◦ rotation z → e 4 z, namely, that
πi
Q(z) = Q e 4 z , k = 0, 1, . . . , 7.
And we have the following result.

Lemma Let p, q ∈ Z, q = 0, gcd(p, q) = 1. If the Laurent polynomial function
1
R(z) = a−m + · · · + a0 + a1 z + · · · + an zn
zm
pπ
is invariant under rotations of z by the angle q , then R is a Laurent polynomial in
z2q .
pπ i
Proof This means that R(e q z) = R(z), so the coefficients of R do not change
pπ i
when the variable is multiplied by e q . Since the coefficient of zk is multiplied by
kpπ i
e q under this transformation is applied, k must be a multiple of 2q.

Applying the Lemma for p = 1 and q = 4, we obtain that Q is a Laurent
polynomial in z8 . Note also that Q is invariant under the transformation z → 1/z,
consequently only terms in z8k , k ∈ Z, appear and the coefficient of z8k equals that
of z−8k . In other words,
364 9 Isometries

1 1
P √ z+
2 z
is a polynomial in z8 + 1
z8
. But we know that

1 1
z + 8 = T8 z +
8
,
z z
where T8 (y) = y 2 (y 2 − 2)2 (y 2 − 4) + 2 is a rescaling of the eighth Chebyshev

polynomial, and since y = 2x, we obtain that

1
P √ x = P1 (T8 (2x)),
2
√
for some polynomial P1 . Consequently, P (x) = P1 (T8 (2 2x)).
This means that P (x) is a polynomial in 8x 2 (8x 2 − 1)2 (8x 2 − 2) + 2. So the
answer to the problem is

m
P (x) = aj [8x 2 (8x 2 − 1)2 (8x 2 − 2) + 2]j
j =1
for some nonnegative integer m and real coefficients aj , j = 0, 1, . . . , m.

Remark How does the problem change if we want the variable to be invariant under
rotations by π3 ? What is the corresponding functional equation and what is the
answer to the problem?
Source United States of America Team Preselection Test for the International
Mathematical Olympiad, 2014, proposed by Răzvan Gelca.
82 The answer to (a) is negative, a counterexample being the greatest integer
function f (x) = x. √
For (b) the answer √ is positive. We begin by showing that if |z − w| = 3, then
|f (z) − f (w)|
√ = 3. For this, consider two arbitrary points z and w in C such that
|z−w| = 3, and construct the configuration consisting of two equilateral triangles
zz1 z2 and z1 z2 w of side 1 sharing the side z1 z2 . Then, because of the hypothesis,
f (z)f (z1 )f (z2 ) and f (z1 )f (z2 )f (w) are also equilateral triangles of side 1, which
either are distinct and share one side or are the same triangle and then f (w) = f (z).
Let us rule out the second possibility.
Assume therefore that f (z) = f (w), and consider another configuration of
equilateral triangles of side 1, zz1 z2 and z1 z2 w such that |w − w | = 1, as shown
in Fig. 9.98. Then as before, there are two possible situations: f (w ) = f (z), or
f (z)f (z1 )f (z2 ) and f (z1 )f (z2 )f (w ) are distinct equilateral triangles sharing the
side f (z1 )f (z2 ). But we should also have
9 Isometries 365
Fig. 9.98 Equilateral z1

triangles zz1 z2 , wz1 z2 , zz1 z2 ,
and wz1 z2
w
z1
z
z2
w
z2
|f (z) − f (w )| = |f (w) − f (w )| = 1,
√ happen in either situation. Hence, f (z) = f (w), and therefore

and this does cannot
|f (z) − f (w)| = 3, as claimed.
From here we deduce that f maps the vertices of any tessellation of the plane
by equilateral triangles of side 1 into the vertices
√ of a tessellation of the same√ type.
Additionally, we deduce that if |z − w| = 3, then |f (z) − f (w)| = 3, and a
similar argument then shows that f maps √ the vertices of any tessellation of the plane
by equilateral triangles of side-length 3 is mapped into a tessellation of the√same
type. Hence, |f (z) − f (w)| = |z − w| whenever |z − w| = n or |z − w| = n 3 for
any positive integer n.
Now let us concentrate on a line l that passes through some fixed point z0 , and
on this line, let us consider the sets√S and T of points at distance equal to an integer,
respectively an integer multiple of 3, from z0 . We will prove that f (S ∪ T ) lies on
a line, and this proof can be followed on Fig. 9.99. Because f preserves distances in
S and it also preserves distances in T (so it is an isometry when restricted to either
of these sets), f (S) = ρ ◦ τ (S) and f (T ) = ρ ◦ τ (T ) where τ is the translation
from z0 to f (z0 ) and ρ, ρ are rotations
√ about f (z0 ).
The numbers of the form m+n 3 form a dense set in R. If ρ = ρ , then by using
this density property, we can find z ∈ S, w ∈ T with |z − w| < 1 and |f (z) − f (w)|
arbitrarily large. But then there is u ∈ C with |z − u| = |z − w| = 1, and so
|f (z) − f (w)| < |f (z) − f (u)| + |f (u) − f (w)| = 2,
a contradiction. Hence, ρ = ρ ; thus, f |(S ∪ T ) is an isometry.

√The end result of
this discussion is that if z and w are such that |z − w| = |m + n 3| with m, n ∈ Z,
then |f (z) − f (w)| = |z − w|. It is important that, by Kronecker’s Theorem, there
are arbitrarily small distances of this form.
366 9 Isometries
f (w)
u f
z w f (u)
f (z)
√
Fig. 9.99 Proof that points of the form m + n 3 on a line are mapped to a line
Now let z, w be two arbitrary points. We will show that for every > 0,
|z − w| − 3 < |f (z) − f (w)| < |z − w| + .
Choose points z1 , z2 , . . . , zn such that

√
|z − z1 |, |z1 − z2 |, . . . , |zn − w| ∈ {m + n 3 | m, n ∈ Z}
and
|z − z1 | + |z1 − z2 | + · · · + |zn−1 − zn | + |zn − w| < |z − w| + .
Basically we use a broken line of segments of nice lengths to approximate the

segment joining z and w. Then, by using the triangle inequality, we obtain
|f (z) − f (w)| ≤ |f (z) − f (z1 )| + |f (z1 ) − f (z2 )| + · · · + |f (zn ) − f (w)|

= |z − z1 | + |z1 − z2 | + · · · + |zn−1 − zn | + |zn − w| < |z − w| + .
This proves the inequality on the right.

For the other inequality, choose w1 , w2 , . . . , wm on the line zw such that
√
|z − w1 |, |w1 − w2 |, . . . , |wn−1 − wn | ∈ {m + n 3 | m, n ∈ Z},
and such that
|w − w1 | > |w − z2 | > · · · > |w − wn | and |w − wn | < .
Then by the triangle inequality,
|f (z) − f (w)| > |f (z) − f (wn )| − |f (wn ) − f (w)|

= |z − wn | − |f (wn ) − f (w)|.
9 Isometries 367
But we have shown that |f (wn ) − f (w)| ≤ |wn − w| + (the inequality on the
right), so the right-hand side is greater than
|z − wn | − |wn − w| − > |z − wn | − 3,
which proves the inequality on the left. Since was arbitrary, by letting → 0, we
obtain
|f (z) − f (w)| = |z − w| for all z, w ∈ C,
as desired.
Source Kvant (Quantum), proposed by A. Tyshka.
Chapter 10
83 First solution. Let us ask a different question: how many transformations from
the group of translations and homotheties map a circle ω1 to a circle ω2 ? If f1 = f2
are two such transformations, then f2−1 ◦ f1 is a translation or a homothety that
maps ω1 to itself. It cannot be a translation, so it is a homothety. The center of the
homothety must be the center of the circle, since the center is mapped into itself, so
it is the fixed point of the homothety. As the radius of the circle equals the radius
of the image, the homothety ratio must be −1 (it cannot be 1 because we ruled out
the case where f2−1 ◦ f1 is the identity map). Then f2−1 ◦ f1 = σ , where σ is the
reflection over the center of the first circle. It follows that f2 = f1 ◦ σ . So we can
have at most two such transformations, f1 and f1 ◦ σ .
Case 1. The original circles have different radii. Let us show that there is a
homothety mapping ω1 to ω2 . Consider a circle ω that is exterior to both. Take the
internal tangents of ω1 and ω, and let A be there intersection (Fig. 10.1). Then there
is an inverse homothety h1 of center A that maps ω1 to ω. Now take the internal
tangents of ω2 and ω, and let B be their intersection. Again there is an inverse
homothety h2 of center B that maps ω to ω2 . Then h2 ◦ h1 is a homothety that maps
ω1 to ω2 . And h2 ◦ h1 ◦ σ is the other homothety (which is not a translation because
the circles have different radii).
Case 2. The original circles have equal radii. One such transformation is the
translation by the vector with endpoints the centers of the circles. In this case there
is at most one homothety. Perform the construction from Case 1. If the result is a
translation τ , take τ ◦ σ , which now is necessarily a homothety.
Second solution. Let the circles be |z − c1 | = R1 and |z − c2 | = R2 . Then a
homothety f (z) = kz + (1 − k)a must map c1 to c2 , so kc1 + (1 − k)a = c2 . We
deduce that
c2 − kc1
a= .
1−k
Fig. 10.1 Finding the

homothety that maps a circle ω1
into another
A ω
B
ω2
Fig. 10.2 The center of homothety mapping one circle into another
But we know that |k| = R2 /R1 . There are two situations k = R2 /R1 or k =
−R2 /R1 . We have therefore two homotheties if R2 = R1 and one homothety if
R2 = R1 .
Remark A first observation: a function f maps a γ1 into γ2 if when substituting the
variable in the equation of γ2 by f one obtains the equation for γ1 . So if γ1 is given
by g1 (z) = 0 and γ2 is given by g2 (z) = 0, then γ1 is mapped to γ2 if and only
if g2 (f (z)) = 0 is the same curve as g1 (z) = 0. In our situation, if the circles are
|z − c1 | = R1 and |z − c2 | = R2 , then the homothety f (z) = kz + (1 − k)a maps
the first circle into |kz + (1 − k)a − c1 | = R1 , and this should be the equation of the
second circle.
By examining Fig. 10.2, we see that if the circles are exterior, then the centers
of homothety are the intersection of the external tangents and the intersection of the
internal tangents. If the circles intersect, one of the centers of homothety is at the
intersection of the external tangents, and the other is interior to both circles. If one
of the circles is inside the other, both centers of homothety are inside both circles,
with the special case where the circles are concentric, in which case the centers of
homothety coincide and are the common center of the circles.
For two circles of different radii, when two homotheties exist, the centers of
these homotheties are collinear with the centers of the circles and form with them a
harmonic ratio. This can be shown easily using complex coordinates: the centers of
the homotheties be
c2 − kc1 c2 + kc1
c3 = , c4 = .
1−k 1+k
Then
c2 − kc1 − c1 + kc1 c2 + kc1 − c1 − kc1
c3 − c1 c4 − c1 1−k 1+k
: = :
c3 − c2 c4 − c2 c2 − kc1 − c2 + kc2 c2 + kc1 − c2 − kc2
1−k 1+k
c2 − c1 c2 − c1
= : = −1.
k(c2 − c1 ) k(c1 − c2 )
84 First solution. We prove directly the general case of polygons. Let A1 A2 . . . An

and A1 A2 . . . An be two polygons such that Aj Ak is parallel to Aj Ak for every
j, k. Consider a translation τ that maps A1 to A1 . Then τ (A2 ) is on the line
A1 A2 . Consider a homothety h0 of center A1 that maps τ (A2 ) to A2 . Then h =
h0 ◦ τ is a homothety that maps A1 , A2 to A1 , A2 , respectively. For j = 1, 2,
A1 Aj ||A1 Aj , and A2 Aj ||A2 Aj . Because h preserves parallelism, A1 h(Aj )||A1 Aj
and A2 h(Aj )||A2 Aj . This can only happen if h(Aj ) = Aj . So h maps the first
polygon into the second.
Second solution. (a) Let ABC and A B C be the two triangles, with AB||A B ,
AC||A C , and BC||B C . Having parallel sides, they have equal angles, and so they
are similar. At least two of the lines AA , BB , and CC must be distinct, say AA
and BB . They are not parallel because the triangles are not congruent. Assume also
that CC does not coincide with AA .
Arguing on Fig. 10.3, let O be the intersection of AA and OC. Then from
similarity
A C OA A B A C
= = = .
AC OA AB AC
Hence C = C . It follows that AA , BB , CC are concurrent and
A C
A
A B
O
O C C =C A
B
B
B
C =C
Fig. 10.3 Triangles with parallel sides are homothetic

OA OB OC
= = .
OA OB OC
There are two cases: if A is between O and A , then AB separates A from O, so

it must separate B from O; thus B is between O and B . Similarly, C is between

O and C . Thus there is a direct homothety with center A and ratio OA OA that maps
ABC to A B C . If A is between O and A , there is an inverse homothety with ratio
the negative of this quotient.
For (b) proceed by induction on the number of sides, with the base case being
the triangle. Let A1 A2 . . . An and A1 A2 . . . An be two polygons such that Aj Ak is
parallel to Aj Ak for every j, k. We may assume (why?) that A1 , A1 , An−1 , An−1
are not collinear. By the induction hypothesis and the base case A1 A2 . . . An−1 and
A1 A2 . . . An−1 are homothetic and An−1 An A1 and An−1 An A1 are homothetic. The
center of each homothety is the intersection O of A1 A1 and An−1 An−1 , and the
homothety ratio is OA1 /OA1 (we work with directed segments to take into account
a possible inverse homothety). So they are the same homothety and we are done.
Remark The homothety is unique. If the polygons are congruent, then they are
mapped into each other by either a translation or an inverse homothety of ratio −1.
Part (a) is a particular case of Desargues’ Theorem (Problem 115).
85 It is possible for two such homotheties to exist, for example, the square with
vertices 1, i, −1, −i is mapped to the square with vertices 2, 2i, −2, −2i by both
the homothety of center 0 and ratio 2 and the homothety of center 0 and ratio −2.
But three homotheties cannot exist, for if h1 , h2 , h3 are distinct homotheties of
ratios k1 , k2 , k3 that map two polygons into each other, then |k1 | = |k2 | = |k3 |
because they are equal to the similarity ratios of the polygons. So two of the ratios
must be equal, say k1 and k2 . Then τ = h−1 2 ◦ h1 is an isometry obtained by
composing two homotheties, so it is a translation. But this translation maps the
first polygon into itself, which is impossible, since a repeated application of this
translation would render the polygon unbounded. Hence we cannot have more than
two homotheties.
Remark Two homotheties might exist only if the polygons have centers of symme-
try, which happens when they have an even number of sides and the opposite sides
are parallel and of equal length.
86 (a) Let h12 , h13 , and h23 be the direct homotheties of the pairs (ω1 , ω2 ),
(ω1 , ω3 ), and (ω2 , ω3 ), respectively. From Problem 83, we know that they exist and
are unique. Because h12 ◦ h23 is a homothety that maps ω1 to ω3 , h13 = h23 ◦ h12 .
But then the center of h13 is on the line determined by the centers of h12 and h23 , as
it has been shown in Theorem 2.6, and we are done.
(b) Let the inverse homotheties of the pairs (ω1 , ω2 ), (ω1 , ω3 ), and (ω2 , ω3 ) be
h12 , h13 , and h23 , respectively. Then, since the composition of either two inverse
homotheties or two direct homotheties is a direct homothety, and the composition
of an inverse homothety with a direct homothety is an inverse homothety, we have
h12 ◦ h23 = h13 , h23 ◦ h12 = h13 , h23 ◦ h12 = h13 , h23 ◦ h12 = h23 .
Here again we used the uniqueness of the homothety that was proved in Problem 83.
By the same Theorem 2.6, the centers of the homotheties in each equality are
collinear, and they determine the four lines.
Remark If the circles are pairwise exterior, the centers of direct homotheties are
the intersection points of the external common tangents, and the centers of inverse
homotheties are the intersection points of the internal common tangents. Monge’s
Theorem implies that the three points of intersection of common external tangents
are collinear, while d’Alembert’s Theorem implies that the six intersection points of
the common tangents lie on four lines (Fig. 10.4).
87 Consider the homothety of center A that maps ω2 to ω1 , and let M , N be the

images of M, N, respectively (Fig. 10.5). Then the line M N is the image of the

line MN , so it is parallel to it. It follows that in the circle ω1 , the arcs P M and QN
are equal. We deduce that the angles P AM and QAN are equal, and these are
the same as the angles from the statement.
Remark If M = N, that is, if the line is actually tangent to ω2 , then the line AM is
the angle bisector of P AQ. This property can be rephrased as follows: if ABC is
a triangle, and if a circle ω is tangent to BC at D and to the circumcircle at A, then
AD is the angle bisector of BAC.
Fig. 10.4 The Theorems of Monge and d’Alembert

Fig. 10.5 ω2 is mapped into A

ω1 by a homothety
ω1
Q
ω2
N
P
M N
Fig. 10.6 The projections of A

a vertex of a triangle onto
angle bisectors are collinear
E B C
Source V.A. Gusev, V.N. Litvinenko, A.G. Mordkovich, Praktikum po Resheniyu

Matematicheskih Zadach, Geometria (Practice for Solving Mathematics Problems,
Geometry), Prokveshcheniye, Moscow, 1985.
88 Let D be the projection of A onto one of the bisectors of B, say the exterior
bisector as in Fig. 10.6. The reflection E of A over BD satisfies EBD = ABD,
so E lies on BC. As D is the midpoint of AE, the homothety of center A and ratio
1/2 maps E to D. It follows that D is on the midline of the sides AB, AC. The same
is true for the other projections, so they are all collinear.
Remark An angle chasing argument is possible and natural, but it is far less elegant
than this solution.
89 The solution can be followed on Fig. 10.7. Consider the diameter DM of the
incircle. We know that the homothety of center B that maps the incircle to the
excircle maps M to the point K where the excircle is tangent to the side. And we
have determined in Sect. 2.2.1 II that AD = CK = a+b−c 2 , so B1 is the midpoint
of DK. Thus I B1 is midline in the triangle DMK. It follows that I B1 is parallel to
the line MK = BK, and so it is midline in the triangle DBK, and we are done.
90 We can guess the answer by looking closely at Fig. 10.8. Because AB and MN
are diameters, they intersect at their midpoints, so ANBM is a parallelogram, and
so BN is parallel to AM. Thus the triangle AKP is mapped to the triangle BKM
by an inverse homothety of center K and ratio KB/KA. Consequently the locus is
the circle that is the image of through this homothety, shown with a dotted line in
the figure.
I
B1
K
M
B C
Fig. 10.7 Proof that B1 I passes through the midpoint of BD
Fig. 10.8 Finding the locus N

of P
Ω
P
A B
K
91 The solution can be followed on Fig. 10.9. The triangles A B C and A B C
are homothetic, and the triangles ABC and A B C are homothetic, too. So these
three triangles are pairwise homothetic, and their centers of homothety are collinear
(Theorem 2.6). Since the center of the (inverse) homothety that maps ABC to
A B C is the centroid G, and the center of the (direct) homothety that maps
A B C to A B C is the circumcenter O, it follows that the center of the
(inverse) homothety that maps ABC to A B C , which is the point where the lines
AA , BB , CC intersect, is on Euler’s line OG.
Remark The locus of the points of intersection of AA , BB , CC , when

A , B , C vary on those rays, is the segment that connects the circumcenter
with the orthocenter.
Fig. 10.9 Proof of Boutin’s A

Theorem
C B
C B
A
B C
Fig. 10.10 In a trapezoid, the N

line determined by the
midpoints of the parallel sides
passes through the
intersection of the diagonals D C
A B
92 Let ABCD be the trapezoid (AB||CD) and let M be the intersection of AC

and BD (See Fig. 10.10). Then the triangles MAB and MCD are homothetic with
respect to an inverse homothety of center M. The midpoints of AB and CD are
mapped into each other by this homothety, so they are collinear with the center of
the homothety M.
Extend the sides AC and BD until they intersect at some point N . The triangles
N AB and NDC are directly homothetic with respect to a homothety of center N .
So N and the midpoints of AB and CD are collinear. It follows that NM is median
in the triangle N AB, and since by the hypothesis M is at equal distance from AD
and BC, NM is also the angle bisector of ANB. It follows that the triangle NAB
is isosceles, and consequently the trapezoid ABCD is isosceles, too.
Remark The property that in a quadrilateral the line joining the midpoints of
two opposite sides passes through the intersection of the diagonals characterizes
trapezoids (prove it!).
93 First solution. Let a1 , a2 , . . . , an be the complex coordinates of the vertices.
Then the centroid G of the polygon has the coordinate g = a1 +a2 +···+a
n
n
, while the
centroids Gj have the coordinates
a1 + a2 + · · · + an − aj ng − aj 1
= =− aj
n−1 n−1 n−1

1
+ 1+ g, j = 1, 2, . . . n.
n−1
Hence Gj is the image of Aj through the homothety of center G and ratio − n−1 1
.
Second solution. Let a1 , a2 , . . . , an be the complex coordinates of the vertices.
Since Gj Gk = n−11
(ak − aj ), the two polygons have parallel sides and diagonals.
In view of Problem 84, they are homothetic.
Remark Note that for n = 3 we obtain the well-known fact that a triangle is
homothetic to the triangle formed by the midpoints of the sides under a homothety
of ratio 1/2 whose center is the centroid.
94 Because the sides of the polygons are parallel, and the polygons are similar,
the diagonals must be parallel, too (prove it!). By Problem 84 the polygons are
homothetic. Since one polygon lies inside the other, the center of homothety lies
inside both.
Let u and v be the distances from the center O of homothety to two sides of the
original polygon. Then the distances from O to the corresponding sides of the new
polygon are u + 1 and v + 1. The fact that the homothety maps one pair of sides to
the other implies that
u v
= ,
u+1 v+1
hence u = v. Thus all sides of the original polygon are at the same distance from
O, showing that there is a circle of center O that is tangent to all sides.
95 First solution. Let X, Y, Z, W be the intersections of the parallels through A
and B, B and C, C and D, and D and A, as shown in Fig. 10.11. A first observation
is that MN ||AC||XY , NP ||BD||Y Z, P Q||AC||ZW , and QM||BD||W X, so
MNP Q is a parallelogram, and the parallelograms MNP Q and XY ZW have
parallel sides. It is a good guess that they are homothetic.
To see why MNP Q and XY ZW are homothetic, note that AXBI is a
parallelogram, so M is not only the midpoint of AB but also the midpoint of I X.
Similarly, N, P , Q are the midpoints of I Y, I Z, I W , respectively. Thus MNP Q is
mapped into XY ZW by the homothety of center I and ratio 2. The centers J and
K of the two parallelograms are mapped by this homothety into each other, so they
must be collinear with the center I of the homothety.
Second solution. Let the complex coordinates of A, B, C, D be a, b, c, d,
respectively. The equations of the two diagonals are
(ā − c̄)z − (a − c)z̄ + a c̄ − cā = 0

(b̄ − d̄)z − (b − d)z̄ + b̄d − bd̄ = 0.
Fig. 10.11 Two homothetic W D Z

parallelograms
Q P
J K
I
A C
M N
X B Y
The coordinate z1 of their intersection point satisfies both equations. The lines
parallel to AC and BD through B and A, respectively, have the equations
a−c
z−b = (z̄ − b̄)
ā − c̄
b−d
z−a = (z̄ − ā).
b̄ − d̄
The coordinate of their intersection point X is the unique solution z2 to the system
of equations
(ā − c̄)z − (a − c)z̄ − āb + c̄b + a b̄ − cb̄ = 0

(b̄ − d̄)z − (b − d)z̄ − a b̄ + a d̄ + bā − d ā = 0.
Then z1 + z2 is the unique solution to
(ā − c̄)z − (a − c)z̄ + a c̄ − cā − āb + c̄b + a b̄ − cb̄ = 0

(b̄ − d̄)z − (b − d)z̄ + b̄d − bd̄ − a b̄ + a d̄ + bā − d ā = 0.
This system can be rewritten as
(ā − c̄)z − (a − c)z̄ + (a − c)(ā + b̄) − (ā − c̄)(a + b) = 0

(b̄ − d̄)z − (b − d)z̄ + (b − d)(ā + b̄) − (b̄ − d̄)(a + b) = 0,
and we observe that it has the unique solution z1 + z2 = a + b. This shows that M,
having the coordinate equal to a+b2 , is the midpoint of I X. Similarly N, P , Q are the
midpoints of I Y, I Z, I W , respectively; thus MNP Q and XY ZW are mapped into
Fig. 10.12 A configuration

in which the nine-point
circles of A1 B1 C1 and
A2 B2 C2 are exterior
each other by a homothety of center I . Then J is mapped to K by this homothety,

and thus I, J, K are collinear.
Remark Drawing the correct figure discloses immediately that the two parallelo-
grams have parallel sides and diagonals and then, in view of Problem 84, they are
homothetic.
96 This configuration is possible, as Fig. 10.12 demonstrates, but difficult to draw.
The argument is rather simple: Because ABC and A B C are homothetic with
center of homothety the centroid G and ratio −1/2, the same is true for A1 B1 C1
and A2 B2 C2 . Hence the two nine-point circles are homothetic, with G the center
of homothety. Because the homothety has negative ratio, the interior tangents of the
two circles intersect at the center of homothety, which is G. And G is on the Euler
line of the triangle ABC.
Remark Certainly the two nine-point circles are homothetic regardless of their
relative position, but the interior tangents exist only if the circles are exterior.
Source C. Mihalescu, Geometria Elementelor Remarcabile (The Geometry of the
Remarkable Elements), Ed. Tehnică, Bucharest, 1957.
97 First solution. Examining the particular case of a square, we guess that the
homothety is centered at the centroid of the quadrilateral and has ratio equal to −3.
And indeed, if M and N are the midpoints of AB and CD, respectively, and if G is
the centroid of the quadrilateral, then G is the midpoint of MN, so the ratio between
the distances from G to AB and the reflection of AB over N is 1/3 (Fig. 10.13).
Hence the reflection of AB over N coincides with the image of AB through the
homothety of center G and ratio −3. The same is true for the other four sides, and
the problem is solved.
Second solution. In coordinates, a and b reflect to c + d − a and c + d − b over
2 of the segment with endpoints c and d. The line through c + d − a
the midpoint c+d
and c + d − b has the parametric form c + d − a + t (a − b), t ∈ R. Similarly, the line
through a and d reflects to the line whose parametric form is b + c − a + s(a − d),
s ∈ R. The two lines intersect when t = s = −1, and the intersection point is
b + c + d − 2a. So the vertices of the quadrilateral determined by the four reflected
lines are a = b + c + d − 2a, b = a + c + d − 2b, c = a + b + d − 2c,
Fig. 10.13 Reflection over B

the midpoint is the same as
homothety centered at the
centroid M C
G
A N
Fig. 10.14 The points A

A0 , B0 , C0 , and G
B
A0
C
G
C0
B0
B A C
d = a + b + c − 2d, and
3 1 3 1 3 1 3 1 a+b+c+d
a + a = b + b = c + c = d + d = ,
4 4 4 4 4 4 4 4 4
showing that the quadrilateral a b c d is obtained from abcd through a homothety

of center a+b+c+d
4 and ratio −3.
Source Mathematical Reflections, proposed by Francisco Javier García Capitán and
Juan Bosco Romero Márquez.
98 Let the midpoints of B C , C A , A B be A0 , B0 , C0 , respectively, and let G
be the centroid of the triangle A B C , as shown in Fig. 10.14. Denote the perpen-
diculars to the sides BC, CA, AB through A0 , B0 , C0 by A , B , C , respectively.
Fig. 10.15 Four circles in a A

triangle
O1
O4
I
O2 O3
B C
The homothety of center G and ratio −2 takes A0 , B0 , C0 to A , B , C ,

respectively. It also carries A , B , C to the lines A , B , C through A , B , C ,
that are perpendicular to the sides. These three lines meet at the incenter. It follows
that A , B , C meet at the preimage of the incenter through the homothety.
Remark A configuration that uses just the points A , B , C is easier to control. The
use of the homothety that maps the triangle formed by the midpoints to the triangle
itself is then automatic.
Source Romanian Team Selection Test for the International Mathematical Olympi-
ad, 1986, solution from S. Savchev, T. Andreescu, Mathematical Miniatures, MAA,
2003.
99 Let O1 , O2 , O3 , O4 be the centers of α, β, γ , δ, as shown in Fig. 10.15.
Then the triangles ABC and O1 O2 O3 have parallel sides, so they are homothetic.
Because O1 is on the angle bisector of A, O2 is on the angle bisector of B, and
O3 is on the angle bisector of C, and because of parallel sides, the triangles ABC
and O1 O2 O3 have the same incenter. Since the homothety maps the incenter of one
triangle to the incenter of the other, the center of homothety is the common incenter
of the two triangles. The center O4 of δ is also the circumcenter of O1 O2 O3 ,
because the points O1 , O2 , O3 are at distance 2ρ from O4 . Hence O4 is on the
line connecting the center of homothety with the circumcenter of ABC, and this is
the line connecting the incenter to the circumcenter of ABC. This solves (a).
For (b), let R, r be the circumradius and the inradius of the triangle ABC.
The circumradius of O1 O2 O3 is 2ρ, and so the homothety ratio is 2ρ/R. But
the homothety ratio can also be computed using the distance from the center of
homothety, I , to the sides of O1 O2 O3 and ABC as follows: the distance from I
to AB is r, and the distance from I to O1 O2 is r − ρ; thus the homothety ratio is
(r − ρ)/r. We thus have the equality
r −ρ 2ρ
= ,
r R
which we can rewrite as

1 2 1
= + .
ρ R r
Hence
Rr
ρ= .
R + 2r
Remark The circles α, β, γ , δ exist for every triangle ABC. Indeed, we can start
with three small circles of equal radii tangent to pairs of sides, consider a circle that
is exterior tangent to all three, and vary them continuously until the circle that is
tangent to all three has radius equal to the radii of the three circles.
Even if δ does not have the same radius as α, β, γ , the argument still works to
show that its center is on OI .
You can draw a parallel with Problem 94. Here the triangle O1 O2 O3 is obtained
by translating inward the sides of ABC by the same length. The two triangles have
parallel sides, so they are homothetic by Problem 84. The center of homothety is
necessarily the incenter.
Source Part (a) appeared in Kvant (Quantum), proposed by V. Yagubiants.
100 We argue on Fig. 10.16. Let the quadrilateral be ABCD; let M, N, P , Q be
the midpoints of AB, BC, CD, DA, respectively; and let I, J, H, K be the projec-
tions of the intersection T of the diagonals onto AB, BC, CD, DA, respectively.
We want to determine first the center of the circle on which the eight points
should lie. Switch to complex coordinates, and let the coordinate axes be the
diagonals AC and BD so that A, B, C, D have coordinates a, bi, c, di, respectively,
with a, b, c, d ∈ R. Then the coordinates of M, N, P , Q are m = a+bi 2 ,n =
c+bi
2 , p = c+di
2 , q = a+di
2 , respectively. Note that (m − n)/(p − n) = i(a −
Fig. 10.16 Proof that D

M, N, P , Q, I, J, H, K lie Q P
on a circle
H
Q P
K O
T
A
C
M
I J
N
M B N
J
c)/(b − d) ∈ iR, so MNP = 90◦ , and the same is true for the other angles
of the quadrilateral MNP Q; thus this quadrilateral is a rectangle. This can also
be deduced from the fact that MN, NP , P Q, QM are midlines in the triangles
ABC, BCD, DCA, ADB, so they are parallel to the bases. Therefore MNP Q
is cyclic, and its circumcenter is the midpoint of MP , which has coordinate
4 + 4 i. But the center of the circumcircle of ABCD has the coordinate
a+c b+d
2 + 2 i, because it lies at the intersection of the perpendicular bisectors of

a+c b+d
AC and BD, and these have equations y = a+c 2 and x = 2 i.

b+d
The homothety of center T and ratio 2 maps therefore M, N, P , Q, respectively,

to some points M , N , P , Q on a circle centered at O. We are left with
showing that this homothety maps I, J, H, K to four points on the circumcircle
of M N P Q . Let J be the image of J , which is therefore the reflection of T over
J . Then J is the reflection of N about the perpendicular bisector of BC (because
the triangles BJ C and CN B transform into each other by this reflection). The
perpendicular bisector passes through O, and hence OJ = ON . It follows that J
is on the circumcircle of M N P Q , and the same is true for I , H and K . This
completes the proof.
Remark Of course we could have completed the solution without the use of the
homothety, by computing the complex coordinates of I, J, H, K (as projections
onto the sides), but that is more work.
101 We solve first the case where the circles ωb and ωc are not equal; for simplicity
ωb is smaller than ωc . Let K be the intersection point of the exterior tangents 1 and
2 (Fig. 10.17). Consider the homothety of center K that takes ωb to ωc . The image
A of A through this homothety is diametrically opposite to A in the circle ωc .
C
D
B
K A A
Fig. 10.17 Finding the locus of D

We therefore have ACA = 90◦ , and so A C is parallel to AB. It follows that

C is the image of B through the homothety, and so K is on the line BC. We deduce
that D lies on the circle ω of diameter AK. But D is also inside the angle formed
by 1 and 2 , and it is not hard to see that any point on the arc of ω that lies between
1 and 2 belongs to the locus. Hence the locus is this arc.
If ωa and ωb are equal, then the locus of D is the segment on the common tangent
through A that is determined by the two common exterior tangents.
Remark A brute-force coordinate computation discloses immediately that B → C
by the direct homothety that maps ωb to ωc . Indeed, the equations of the two circles
can be written as |z − b| = b and |z + c| = c, for b, c > 0. Switching to polar
coordinates (z = reiθ ) turns these into r = b cos θ and r = −c cos θ , and if B has
coordinate b cos θ (cos θ + i sin θ ) = b(cos2 θ + i sin θ cos θ ), then C has coordinate
π i (θ+ π )
−c cos θ + e 2 = c sin θ (i cos θ − sin θ ) = c(cos2 θ + i sin θ cos θ ) − c,
2
thus they are mapped into each other by the map z → − bc z + c, which is a direct
homothety. Since the homothety maps A to the point diametrically opposite on ωc ,
the center of homothety is on the line of centers. Moreover, every point on one circle
must be mapped to a point on the other and vice versa, so the homothety must be
the one that maps the circles into each other.
102 We fix B and C and constrain A to lie above the line BC. Then A varies on an
arc of measure 2π − 2α with endpoints B and C. The centroid G of ABC lies on
the image of that arc under a homothety of center the midpoint M of BC and ratio
1/3. Call this arc, which is the locus of G, γG .
If I is the incenter, then
BI C = π − ABC/2 − ACB/2 = π/2 + BAC/2 = π/2 + α/2.
So I is constrained to an arc of measure 2π − 2(π/2 + α/2) = π − α with endpoints

B and C. Denote this arc by γI . The arcs γG and γI can be seen in Fig. 10.18.
We claim that the arc γI lies above the arc γG , as the figure suggests. This will be
true if we show that the midpoint of γI is above the midpoint of γG . The midpoints
coincide with the incenter and the centroid in the case where the triangle ABC is
isosceles. In that situation, in the triangle I BC, BI M = π +α
4 , and so
BC π +α 1 π +α
IM = cot = cot ,
2 4 2 4
while in the triangle ABC,
1 1 α
GM = AM = cot .
3 6 2
Fig. 10.18 How to minimize

the length of I G
A
γI
I
G
γG
B C
ωG
ωI
We have to show that I M > GM, which amounts to showing that
1 π +α 1 α
cot > cot .
2 4 6 2
Denoting α/4 = θ , we have to show that
π π π
tan + θ < 3 tan 2θ, for <θ < .
4 12 4
Using the addition and the double angle formulas for the tangent, we transform this
inequality into
1 + tan θ tan θ
<6 .
1 − tan θ 1 − tan2 θ
The denominators are positive, so this is further equivalent to
tan2 θ − 4 tan θ + 1 < 0.
Since I = G for the equilateral triangle, this quadratic polynomial has a zero when
θ = π/12, or better said, when tan θ = tan π/12. The product of the roots is 1, so
the other root of the quadratic is greater than 1 = tan π/4. We conclude that when
θ ∈ (π/12, π/4), the quadratic function assumes negative values, which proves the
inequality.
Fig. 10.19 How to find the

distance between two circles
So the arcs γG and γI do not intersect, and we want to solve the easier problem of
determining the minimal distance between them. Complete the arcs to two circles
ωG and ωI . Then ωG is inside ωI . We want to determine the minimal distance
between the circles and use Fig. 10.19 for our reasoning. Let M ∈ ωG and N ∈ ωI
for which the minimum is attained. The circle of diameter MN is tangent to both
ωG and ωI , for any other point of intersection with one of the two circles would
determine with either M or N a shorter distance. Then the tangents at M and N to
ωG and ωI are parallel, so N = h(M), where h is the direct homothety that maps
γG to γI . But if O is the center of this homothety and k is its ratio, then
MO + MN NO
= = k,
MO MO
so for MN to be minimal, MO has to be minimal. This means that the minimum
is attained when M and N are on the line of centers. Rephrasing in terms of our
problem, the minimum is attained for the isosceles triangle, and this minimum is
1 π +α 1 α
cot − cot .
2 4 6 2
Remark We were not required to find the minimal distance between the arcs γG
and γI ; as I and G do not vary independently on the two arcs, they both depend

on one parameter, the position of A on the arc BAC. However, the configuration
that minimizes the distance between the arcs, and hence between the two circles,
does correspond to a triangle: the isosceles triangle. It is also important to point
out that because BI C > BAC, it is the configuration formed by the incenter
and the circumcenter of the isosceles triangle that yields the minimum (and not the
configuration of the other two points on the line of centers of ωI and ωG ).
Source Vietnamese Mathematical Olympiad, 1996.
A B1
B1
C1 C1
H
P
C
B A1
A1
Fig. 10.20 The homothety with center H and ratio 2
103 Consider the homothety of center H and ratio 2, and let A1 , B1 , and C1 be
the images of A1 , B1 , and C1 , respectively (Fig. 10.20). Then on the one hand
A1 , B1 , C1 lie on the circumcircle of ABC, and on the other hand A1 H, B1 H, C1 H
are the angle bisectors of triangle A1 B1 C1 (since they are the images through the
homothety of the angle bisectors A1 H, B1 H, C1 H of triangle A1 B1 C1 ).
Because B1 is the reflection of H over AC, and C1 is the reflection of H over
AB, it follows that H B1 Q = QH B1 , so
C1 B1 Q = H B1 Q + C1 B1 H = QH B1 + C1 B1 H

= QH B1 + H B1 A1 = 90◦ .
Similarly B1 C1 P = 90◦ . Because of the same reflections AB1 = AH = AC1 , so
the perpendicular bisector of B1 C1 passes through A. But since P C1 and QB1 are
parallel to this perpendicular bisector, it also passes through the midpoint of P Q
(by Thales’ Theorem). So the line that passes through A and the midpoint of P Q is
the perpendicular bisector of B1 C1 ; hence it is perpendicular to B1 C1 , as desired.
Remark An observation that leads to a different solution is that while H is the
incenter of triangle A1 B1 C1 , A is its excenter.
Source The Danube Mathematical Competition, Romania, 2018, solution by
Mircea Fianu.
104 Figure 10.21 shows a possible configuration, on which the subsequent con-
struction is also marked. For the solution, let G be the centroid of the triangle ABC,
F H
E
D C
A
E
B
Δ
D O
F
Fig. 10.21 The images of D, E, H through h
and consider the homothety h of center G and ratio −1/2. Then h(H ) = O. With
the standard notation: X = h(X), A , B , C are the midpoints of BC, CA, AB,
respectively. Furthermore, D is the reflection of A across B C . This means that
the line AD is the image of the line BC through h, so it is parallel to BC. Similarly
BE is parallel to AC, and CF is parallel to AB, and so the lines AD , BE , CF
form a triangle whose sides are parallel to those of triangle ABC and pass through
the vertices of this triangle.
We should also note that OA is perpendicular to BC and thus O, A , D
are collinear and OD is perpendicular to AD . Consequently D , E , F are the
projections of O onto the sides of the triangle . The fact that they are collinear
(which is the same as D, E, F being collinear), is equivalent, by the Simson-
Wallace Theorem (Theorem 3.34), to the fact that O lies on the circumcircle of
. But note that is the preimage of ABC through h, so the circumcenter of
is H = h−1 (O) and its circumradius is 2R. So the last condition is equivalent to
OH = 2R, and we are done.
Remark The observation that A is the orthocenter of H BC and that the circumcircle
of H BC is the reflection of the circumcircle of ABC over BC gives a clue on how
to construct such a configuration.
Source Short list, 39th International Mathematical Olympiad, 1998, proposed by
France, solution from D. Djukić, V. Janković, I. Matić, N. Petrović, The IMO Com-
pendium, A Collection of Problems Suggested for the International Mathematical
Olympiads: 1959-2004, Springer, 2006.
B2
C2 B1
C1
B A1 C
A2
Fig. 10.22 The triangles A1 B1 C1 and A2 B2 C2 are homothetic
105 The configuration is shown in Fig. 10.22. Note that because BCB1 C1 is cyclic,

AB1 C1 = ABC. Also, ABC =AC /2 = B2 AC. It follows that AB1 C1 =
B2 AC, so B1 C1 and B2 C2 are parallel. Similarly A1 C1 is parallel to A2 C2 and
A1 B1 is parallel to A2 B2 . Therefore the triangles A1 B1 C1 and A2 B2 C2 have parallel
sides, and hence there is a homothety h such that h(A1 ) = A2 , h(B1 ) = B2 , and
h(C1 ) = C2 .
The incenter of A2 B2 C2 is O, the circumcenter of ABC, while the incenter of
A1 B1 C1 is the orthocenter H of ABC. Then h(H ) = O. Since the line that connects
a point with its image through h passes through the center of h, it follows that A1 A2 ,
B1 B2 , C1 C2 , and H O are concurrent. But OH is the Euler line of the triangle ABC,
and so the center of homothety, which is the intersection of A1 A2 , B1 B2 , C1 C2 , is
on this line.
106 Consider a homothety of center A and ratio 1/2, and let B , C , and D be
the images of B, C, and D, respectively (see Fig. 10.23). Then B and C are the
midpoints of the sides AB and AC, respectively. The triangles NB C and D B C
are both isosceles (NB = NC and D B = D C ), thus either N = D or the line
N D is the perpendicular bisector of B C .
Fig. 10.23 The homothety of A

center A and ratio 1/2
B C
N
B C
D
Now A, D, D are collinear, so A, D, N are collinear if and only if A, D , N are

collinear. This is impossible if N = D , for in this case A would have to be on the
perpendicular bisector of B C , and hence the triangle AB C would be isosceles,
and so would be the triangle ABC, contradicting the hypothesis of the problem.
Thus A, D, N are collinear if and only if D = N.
But N is the reflection over B C of the circumcenter Q of AB C , so B NC =
B QC = 2 A. Also, because D B and D C are tangent to the circumcircle of
triangle AB C , B D C = 180◦ − 2 A. So A, D, N are collinear if and only if
2 A = 180◦ − 2 A, that is if and only if A = 45◦ , and we are done.
Source Brazilian Mathematical Olympiad, 2015.
107 Using inscribed angles in the two circles (Fig. 10.24), we obtain
EF C = ECA = EDB.
It follows that the lines CF and AB are parallel.

Let G be the intersection of BK and CF , and let G be the intersection of AE
and CF . Now we use the fact that D is the midpoint of AB. The direct homothety
that maps the triangle KAB to KCG maps D to F , and so F is the midpoint of
CG. The inverse homothety that maps the triangle EAB to EG C maps D to F , so
F is the midpoint of CG . It follows that G = G , and the problem is solved.
Source St. Petersburg Mathematical Olympiad, 1998.
K F
D
E
G=G
Fig. 10.24 Proof that AE, BK, DF are concurrent
108 Let us first notice that the circumcircles of the triangles ABC and DBC are
congruent, being symmetric with respect to the line BC. Let R be their common
radius. Then, using the well-known metric relation that computes the distance from
the vertices to the orthocenter, we have
AH1 = 2R cos A = 2R cos D = DH2 .
Since AH1 and DH2 are also parallel, it follows that AH1 DH2 is a parallelogram,
so the segments AD and H1 H2 intersect at their midpoint P , as shown in Fig. 10.25.
On the other hand, if P is the intersection of H1 H2 and EF , by using the Law
of Sines in the triangles H1 P E and H2 P F , we obtain
H1 P H1 E sin BEF AH1 cos AH1 E sin BCF

= =
H2 P
H2 F sin CF E DH2 cos DH2 F sin CBE
2R cos A cos ACB cos CBD
= = 1.
2R cos D cos CBD cos ACB
So P = P , proving that lines AD, H1 H2 , and EF intersect at P .

Since the nine-point circle is the image of the circumcircle by the homothety
centered at the orthocenter and of ratio 1/2, and since the orthocenters H1 and H2
are on the circumcircles of DBC and ABC, we obtain that P , being both the image
of H2 under the homothety of center H1 and ratio 1/2 and the image of H1 under
the homothety of center H2 and ratio 1/2, is on both nine-point circles.
Remark The information that the common point of the three lines should be on the
nine-point circle of the triangle ABC gives away the information that this point is
Fig. 10.25 AD, EF , H1 H2 A

intersect on the nine-point
circles of ABC and BCD
H1
P
B C
H2
D
C
P
Y A M B X
Fig. 10.26 M is the midpoint of XY
the image of H2 (who is on the circumcircle) under the homothety of center H1 and
ratio 1/2, so it tells us that the common point is the midpoint of the segment H1 H2 .
Source Romanian Team Selection Test for the Junior Balkan Mathematical
Olympiad, 2013.
109 The situation is described in Fig. 10.26. Consider the direct homothety hQ of
center Q that maps A to C and the inverse homothety hP of center P that maps C
to B. By Theorem 2.6, the composition hQ ◦ hP is a homothety whose center is on
P Q. Also hQ ◦ hP maps A to B, and so its center is on AB. Hence hQ ◦ hP is a
homothety with center AB ∩ P Q; this center is therefore M. But M is the midpoint
of AB, so the homothety hQ ◦ hP has ratio −1; it is the reflection over M.
We also have that hQ (Y ) = D and hP (D) = X, so hQ ◦ hP (Y ) = X. Thus the
reflection over M maps Y to X, showing that M is the midpoint of XY .
Remark The transformation hQ ◦ hP cannot be a translation because the ratios of

the two homotheties have opposite signs, so they cannot be one the reciprocal of the
other.
Source A. Engel, Problem-Solving Strategies, Springer, 1998.
110 Assume that the triangle P QR has been constructed (Fig. 10.27). Let S be
the tangency point of and ω, T the reflection of S over M, and U the point
diametrically opposite to S in ω. The line P U intersects at the point where the
excircle is tangent to RQ, because, as we have explained in the introduction, the
homothety that maps the incircle to the excircle maps U to this point. But, as we
know, this point is T , the reflection of S over M. It follows that P lies on the open
ray of the line U T that is opposite to |U T .
Conversely, for every P on this ray, we can construct the triangle P QR, and then
T must be the tangency point of the excircle, and so T , U, P are collinear, and S is
the tangency point of the incircle with . The problem is solved.
Source International Mathematical Olympiad, 1992, proposed by France.
111 The point Z is on the radical axis of the two circles, so power-of-a-point gives
ZA · ZC = ZB · ZD and we can set
ZC ZB
k= = .
ZD ZA
Fig. 10.27 The triangle P

P QR
Q ω R
T M S
ω
Fig. 10.28 The images of

BN and CM through the
Q
homothety h
M R
X
N
A B Z C D
Consider the homothety h of center Z and ratio k. Then h(B) = A, so h(BN) is a

line through A that is parallel to BN and is therefore perpendicular to DN; similarly
h(CM) is a line through D that is perpendicular to AM (Fig. 10.28).
Let Q be the intersection of AM and DN and let R be the intersection of
h(BN) and h(CM). Then Q is the orthocenter of the triangle RAD, and so RQ
is perpendicular to AD. On the other hand since P is the intersection of BN and
CM, R = h(P ), so Z, P , R are collinear. This means that R is on XY and since RQ
is perpendicular to AD, so is Q. Therefore the lines AM, DN , and XY intersect at
Q, and the problem is solved.
Source International Mathematical Olympiad, 1995, proposed by Bulgaria.

112 The solution can be followed on Fig. 10.29. Let O and G be the circumcenter
and the centroid of the triangle ABC, respectively. The points O, G, O9 , H
determine Euler’s line, and O9 O = O9 H = 3O9 G.
The nine-point circles of the triangles ABC, BH C, CH A, and AH B coincide,
because A, B, C are the orthocenters of BH C, CH A, AH B, respectively. Thus O9
is the center of the nine-point circle in each of these triangles. So AO9 , BO9 , and
CO9 are the respective Euler lines of these three triangles.
The segment H A is a median in the triangle BH C, so its intersection with the
Euler line of this triangle is the centroid of BH C. Hence A1 is the centroid of BH C,
and similarly B1 and C1 are the centroids of the triangles CH A respectively AH B,
respectively. We obtain
H A1 H B1 H C1 2
= = = .
H A H B H C 3
Consequently the triangles A B C and A1 B1 C1 are homothetic, with the center of

homothety being H and the ratio 2/3.
Fig. 10.29 The pairwise A

homothetic triangles
ABC, A B C , A1 B1 C1
C B
C1 B1
O9
H G
A1
B A C
Fig. 10.30 Square inside a ω

circle
C
D
ωc
ωd
D C
γ
A B
ωa
ωb
A
B
Also, the triangles ABC and A B C are homothetic, with the center of homoth-
ety being G and the ratio −1/2. By Theorem 2.6, the triangles A1 B1 C1 and ABC
are homothetic, with ratio −1/3 and the center of homothety on GH , dividing this
segment in the ratio 1/3. But the point with this property is O9 , the center of the
nine-point circle of ABC. And because the Euler line of the triangle ABC passes
through the center of homothety, it is mapped into itself by the homothety, so it is
the Euler line of triangle A1 B1 C1 , as desired.
Source N. Grunbaum, Gazeta Matematică (Mathematics Gazette, Bucharest),
solution by E. Drăgănescu and B.M. Barbalatt, published in C. Mihalescu,
Geometria Elementelor Remarcabile (The Geometry of the Remarkable Elements),
Ed. Tehnică, Bucharest, 1957.
113 The configuration is shown in Fig. 10.30. Let γ be the incircle of the square
ABCD. Consider the inverse homothety ha of center A that maps ωa to γ and the
direct homothety ha of center A that maps ω to ωa . Then ha ◦ ha is the unique
inverse homothety that maps ω to γ . Let T be its center. By Theorem 2.6 T is on
Fig. 10.31 The homothety of

center A A
B K =Q C
B M N C
AA . A similar argument shows that T is on BB , CC , and DD and the problem

is solved.
Olympiad, 2003.
114 The case AB = AC is obvious. Otherwise, without loss of generality, we
may assume that AB > AC. Let D be the point where the angle bisector AN
intersects the circumcircle for the second time. Then D is the midpoint of the arc

BC, so DB = DC and DM is the perpendicular bisector of BC. Consider now the
homothety of center A that maps D to R (Fig. 10.31). Let B ∈ AB and C ∈ AC
be the images of B and C. Then B C is parallel to BC, and, because DB = DC,
we have that RB = RC .
Let K be the point where B C intersects P N. Then
RB K = DBC = DAC = DAB = 90◦ − ARP = RP K.
So the points R, B , P , K are concyclic. It follows that B KR = B P R = 90◦

and, consequently, K is the image of M through the homothety. This means that
K ∈ AM, and therefore K = Q. We deduce that RQ is the perpendicular bisector
of B C , so it is perpendicular to B C and hence to BC, as desired.
Source Asian Pacific Mathematical Olympiad, 2000, solution by Bobby Poon.
115 Let M be the intersection of A1 A2 , B1 B2 , C1 C2 ; let also P , Q, R be the
intersections of the pairs (B1 C1 , B2 C2 ), (A1 C1 , A2 C2 ), and (A1 B1 , A2 B2 ), respec-
tively, as in Fig. 10.32. We apply repeatedly the trick behind the proof of Menelaus’
Theorem (Sect. 2.2.1). We introduce six homotheties:
Fig. 10.32 Desargues’ M

Theorem
A1
C1
Q
B1
R
P
B2 C2
A2
ha of center A1 that maps A2 to M, hb of center B1 that maps B2 to M,

hc of center C1 that maps C2 to M, hp of center P that maps B2 to C2 ,
hq of center Q that maps C2 to A2 , hr of center R that maps A2 to B2 .
Arguing like in the proof of Menelaus’ Theorem for the triangle MA2 B2 and the line
of the points A1 , B1 , R, the triangle MA2 C2 and the line of the points A1 , C1 , Q,
and the triangle MB2 C2 and the line of the points B1 , C1 , P , we obtain
−1
hb ◦ hr ◦ h−1 −1
a = 1, ha ◦ hq ◦ hc = 1, hc ◦ hp ◦ hb = 1.
Composing we obtain
−1 −1
hb ◦ hr ◦ h−1 −1
a ◦ ha ◦ hq ◦ hc ◦ hc ◦ hp ◦ hb = hb ◦ hr ◦ hq ◦ hp ◦ hb = 1.
Consequently hr ◦ hq ◦ hp = 1. So h−1r = hq ◦ hp , and by Theorem 2.6, the centers

of these homotheties, P , Q, R, are collinear.
Once the direct implication is proved, the converse is immediate. Let M be the
intersection of A1 A2 and B1 B2 , and let C1 be the intersection of A1 C1 and MA2 .
Let also P be the intersection of B1 C1 and B2 C2 . Then P , Q, R are collinear by
hypothesis, and P , Q, R are collinear by the direct implication. So both P and P
are at the intersection of RQ and B2 C2 . This can only happen if P = P and we are
done.
Remark Of course, we could have invoked Menelaus’ in the proof and do the
algebra. We leave it to the reader to solve the cases where one or more of the points
M, P , Q, R are at infinity.
116 Assume that AA1 , BB1 , and CC1 intersect at M (Fig. 10.33). Consider the
homotheties
Fig. 10.33 Ceva’s Theorem C

B1
A1
M
A C1 B
ha of center A that maps B1 to C, ha1 of center A1 that maps C to B,

hc1 of center C1 that maps B to A, and hc of center C that maps A to B1 .
The composition ha1 ◦ha maps B1 to B and is not a translation because the image
of A lies on AA1 , which intersects BB1 . So ha1 ◦ ha is a homothety, and its center
lies on the line of centers AA1 (by Theorem 2.6), but also on the segment BB1
because B1 is the image of B. Hence the center of ha1 ◦ ha is M. A similar argument
shows that hc ◦ hc1 is a homothety of center M. But hc ◦ hc1 ◦ ha1 ◦ ha (B1 ) = B1 , so
this composition of homotheties has two fixed points, B1 and M. It must therefore
be the identity map.
Thus the product of the ratios of the four homotheties is 1, meaning that
b1 − c a − c1 b − a1 c − a
· · · = 1,
a − c b − c1 c − a1 b1 − a
which is equivalent to the identity from the statement.

For the converse, let M be the intersection of BB1 and CC1 , and let A1 be the
intersection of AM and BC. Then A1 and A1 are the centers of homotheties of equal
ratios that map B to C, so they coincide.
Remark Of course, this is just a translate in the language of homotheties of the proof
of Ceva’s Theorem by Menelaus’ Theorem.
117 By just examining a carefully drawn figure (see Fig. 10.34), we discover some
collinearities: K, M and P ; K, N, and Q; and M, L, and N. To prove that the first
two triples consist indeed of collinear points, consider the direct homothety that
maps ω to . Because homothety preserves tangencies, the image of AB is a line
that is tangent to at the image of M. But this tangent is also parallel to AB, so it

meets at the midpoint P of AB, and hence P is the image of M. Similarly, Q is
the image of N, which proves the first two collinearities.
For the third collinearity, we use inscribed angles. From the fact that the
quadrilaterals BP ML and BQNL are cyclic, we obtain that the angles MLB
and MP B, as well as the angles NLB and NQB, are supplementary. But in
the quadrilateral KP BQ, MP B = KP B, and NQB = KQB are supple-
Fig. 10.34 The points in the C K

triples (K, M, P ),
(K, N, Q), (M, L, N ) are
collinear
N
Q
L
A
M
B
mentary, so MLB and NLB are supplementary, proving the third collinearity.
Using this, the fact that the triangle BMN is isosceles (because BM and BN are
the tangents from B to ω), and the fact that BLMP and BLNQ are cyclic, we
obtain
BP L = BML = BNL = BQL.
Also
1
LP B = LMB = NMB = MN= MKN = P KQ
2
= 180◦ − P BQ.
We conclude that the quadrilateral BP LQ has two opposite equal angles and two
adjacent supplementary angles, so it is a parallelogram.
Source Russian Mathematical Olympiad, 2000.
118 First solution. We argue on Fig. 10.35 and denote by (I ), (J ), (E), (F )
the circles specified in the statement of the problem with centers I, J, E, F ,
respectively. Let X be the intersection point of I J and BC. Then X is the center
of the homothety that maps the circle (I ) to the circle (J ), and A is the center of the
homothety that maps the circle (I ) to the circle (E). It follows that the center Y of
the homothety that maps (J ) to (E) is on AX, by the Monge-d’Alembert Theorem
(Problem 86).
For the same reason, the centers of the direct homotheties that map (E) to (F ),
(E) to (J ), and (J ) to (F ) are collinear. Let Z be the center of the first, while the
centers of the other two are A and Y . It follows that Z is on AY . But Z is also the
intersection of the exterior tangent BC of the circles (E) and (F ) with the line of
J
X=Z I C
B D
Fig. 10.35 Four circles connected by homotheties
the centers EF . We already know that the line AY = AX intersects BC at X, so

Z = X. Thus X is the intersection point of I J , EF , and BC, and we are done.
Second solution. This problem can be solved using the fact that the triangles BI E
and CJ F are perspective. Note that I E and J F pass through A, being the angle
bisectors of BAD and DAC, respectively. Also, BI and CJ are bisectors of the
interior angles at B and C, respectively, so they intersect at the incenter of ABC, and
BE and CF are angle bisectors of the exterior angles at B and C, respectively, so
they intersect at the center of the excircle of triangle ABC tangent to BC. The two
centers are collinear with A, since they lie on the angle bisector of BAC. So the
intersection points of (I E, J F ), (BE, CF ), and (BI, CJ ) are collinear. Desargues’
Theorem (Problem 115) implies that the triangles BI E and CJ F are perspective,
as claimed, so I J, BC, EF intersect at one point.
Remark As a by-product of this problem, we obtain that two excircles (E) and (F )
are congruent if and only if the two incircles (I ) and (J ) are congruent.
119 The solution can be followed on Fig. 10.36. Let O1 , O2 , O3 be the centers of
ω1 , ω2 , ω3 , respectively. The key idea is that the intersection point of the common
tangents of two exterior circles is the center of the direct homothety that maps the
circles into each other. If A is the intersection of the common tangents of ω1 and
O1
O2
O3
O
B
Fig. 10.36 Two pairs of common tangents for a circumscribable quadrilateral
ω3 , and B is the intersection of the common tangents of ω1 and ω2 , then A ∈ O1 O3

and B ∈ O1 O2 , and moreover, the ratio of the homothety that maps ω2 to ω1 is
r1 BO1
= ,
r2 BO2
and the ratio of the homothety that maps ω3 to ω1 is
r1 AO1
= .
r3 AO3
The center of the incircle of the quadrilateral formed by the tangents from B to
ω3 and from A to ω2 should be at the intersection of AO2 and BO3 , and we denote
this intersection by O. We have to show that there is a circle ω centered at O that
is mapped to ω2 by a homothety of center A and to ω3 by a homothety of center B.
This amounts to showing that there is a number r > 0 such that
r AO r BO
= and = .
r2 AO2 r3 BO3
And this reduces to showing that
AO BO
r2 = r3 .
AO2 BO3
Writing Menelaus’ Theorem (and ignoring the orientation of the segments) in the
triangle BO1 O3 traversed by the line through A, O, O2 we obtain
BO O3 A O1 O2
· · = 1,
OO3 AO1 O2 B
hence
BO AO1 O2 B r1 O2 B r1 r2
= · = · = .
OO3 O3 A O1 O2 r 3 O1 B − O2 B r3 r1 − r3 r2
Consequently
BO BO r1 r2
= = .
BO3 BO + OO3 r1 r2 + r1 r3 − r2 r3
Similarly
AO r1 r3
= .
AO2 r1 r2 + r1 r3 − r2 r3
We deduce that
AO BO r1 r2 r3
r2 = r3 = ,
AO2 BO3 r1 r2 + r1 r3 − r2 r3
and the value of this expression is equal to r, the radius of the circle that is inscribed
in the quadrilateral.
Remark Note that ω2 and ω3 are mapped into each other by a homothety of center A
followed by a homothety of center B, but also by a homothety of center B followed
by a homothety of center A. By the Monge-d’Alembert Theorem, the intersection
of the common exterior tangents of ω and ω1 and the intersection of the common
exterior tangents of ω2 and ω3 lie both on the line AB.
Source All Soviet Union Mathematical Olympiad, 1984, proposed by L.P. Kuptsov.
120 (a) Consider the homothety of center A that maps D and E to M and N ,
respectively (Fig. 10.37). The square BCDE is mapped to a square of side MN,
whose other two vertices are on AB and AC. This square must be MNP Q.
(b) The three Lucas circles are shown in Fig. 10.38. For simplicity let a = BC, b =
AC, c = AB. The homothety specified above maps the circumcircle of ABC to
a . The radius Ra and of a and the radius R of the circumcircle satisfy Ra /R =
P Q/a. On the other hand, using the extended Law of Sines, we obtain
QM PQ PQ
sin B = = = ,
QB c − AQ c − 2Ra sin C

MN P Q and BCDE are
similar. Q P
B M N C
E D
Fig. 10.38 Lucas circles
A
Oa
Oc
B Ob C
and also sin B = b

2R and sin C = c
2R . An easy computation yields
Rbc
Ra =
bc + 2aR
If Oa and Ob are the centers of the circles corresponding to the vertices A and B,
respectively, then
2aR 2 2aR
OOa = R − Ra = = Ra
bc + 2aR bc
and
2bR 2 2bR
OOb = R − Rb = = Rb ,
ac + 2bR ac
where Rb is the radius of the Lucas circle at B. Because Oa OOb = AOB =

2 C, using the Laws of Cosines and Sines, we obtain
Oa Ob2 = OOa2 + OOb2 − 2OOa · OOb cos 2C

= (R − Ra )2 + (R − Rb )2 − 2(R − Ra )(R − Rb )(1 − 2 sin2 C)
= [(R − Ra − (R − Rb )]2 + 4(R − Ra )(R − Rb ) sin2 C
2aR 2bR
= (Ra − Rb )2 + 4 Ra · Rb sin2 C
bc ac
4R 2 sin2 C
= (Ra − Rb )2 + 4Ra Rb = (Ra − Rb )2 + 4Ra Rb
c2
= (Ra + Rb )2 ,
hence a and b are tangent. And c is tangent to these two circles for a similar
reason. The problem is solved.
Remark The Lucas circles and the circumcircle form a family of four circles in
which any one is tangent to the other three.
Source This proof that the Lucas circles are pairwise tangent has appeared in
Antreas P. Hatzipolakis and Paul Yiu, The Lucas circles of a triangle, in The
American Mathematical Monthly, Vol. 108, No. 5, 2001.
121 Let A1 A2 and A3 A4 meet at Q and let A1 A4 and A2 A3 meet at P , as shown
in Fig. 10.39. We will assume that P , T1 , T3 are collinear and prove that Q, T2 , T4
are collinear.
Construct the circle ω that is tangent to A1 A2 at T1 and is also tangent to the line
A3 A4 . Then T1 is the center of the inverse homothety that maps ω to ω1 , and P is
the center of the direct homothety that maps ω1 to ω3 . By the Monge-d’Alembert
Theorem, proved in Problem 86, the center of the inverse homothety that maps ω to
ω3 is on P T1 . Furthermore, this center lies on the common tangent A3 A4 of these
two circles. But P T1 intersects A3 A4 at T3 , because this is our assumption. Thus T3
is the center of the inverse homothety that maps ω to ω3 , and hence ω is tangent to
A3 A4 at T3 . Thus QT1 = QT3 , being the common tangents from Q to ω.
Using the fact that QT1 = QT3 and the fact that the two tangents from Q to
ω2 are equal, we obtain that the common internal tangents of the pair of circles ω1
and ω2 are equal to the common internal tangents of the pair of circles ω3 and ω2 .
Consequently, T2 is the midpoint of the segment determined by the tangency points
of A2 A3 to ω1 and ω3 . Similarly, T4 is the midpoint of the segment determined by
Q
ω4
T4 A4
A1
P
ω1
ω
T1 ω3
T3
A2
T2
A3
ω
ω2
Fig. 10.39 Proof that P , T1 , T3 collinear implies Q, T2 , T4 collinear
the tangency points of A1 A4 to ω1 and ω3 . Using this, and the fact that the common
exterior tangents from P to ω1 are equal and the common exterior tangents from
P to ω3 are equal, we deduce that P T2 = P T4 . So there is a circle ω tangent to
P T2 and P T4 at T2 and T4 , respectively. Applying this time the Monge-d’Alembert
Theorem in reverse to the circles ω , ω2 , ω4 , we deduce that Q, T2 , T4 are collinear.
Remark Note the similarity between this problem and Problem 113.
Source Romanian Master of Mathematics, 2010, proposed by Pavel Kozhevnikov,
solution from T. Andreescu, C. Pohoaţă, 110 Geometry Problems for the Interna-
tional Mathematical Olympiad, XYZ Press, 2014.
122 The argument repeats the pattern of the solution to Problem 121: metric
relations that arise from equal tangents combined with applications of the Monge-
d’Alembert Theorem. The four lines AB, BC, CD, and DA form a complete
quadrilateral, and the circle ω is tangent to the four lines. This suggests that our
quadrilateral might have properties analogous to those of a quadrilateral that admits
an inscribed circle. It is a well-known fact that a quadrilateral admits an inscribed
circle if and only if the sum of one pair of opposite sides is equal to the sum of
the other pair of opposite sides, a result known as Pitot’s Theorem. The analogous
property for the complete quadrilateral in this problem states that
AB + AD = CB + CD.
Fig. 10.40 The analogue of

Pitot’s Theorem
M
A N
D
B Q
C P
ω2
ω2
A
S=T
L
D
ω1 K
B ω1
C
ω
Fig. 10.41 Proof that the common exterior tangents intersect on ω
To prove the direct implication, which we need for this problem, let us denote
by M, N, P , Q the points where AB, BC, CD, DA are tangent to ω, respectively
(Fig. 10.40). Using the fact that the two tangents from a point to a circle are equal,
we can write
AB + AD = BM − AM + AQ − DQ = BP − AQ + AQ − DN = BP − DN
= BP − CP + CP − DN = BP − CP + CN − DN = BC + CD,
and the claim is proved.

We now continue the solution with the aid of Fig. 10.41. There is a surprising
consequence of the above Pitot-like relation: if K and L are the points where AC is
tangent to the circles ω1 and ω2 , respectively, then
AB + AC − BC CD − AD + AC
AL = = = CK.
2 2
Here we have used the computations from Sect. 2.2.1 II. From the observations
made in Sect. 2.2.1 II, we deduce that L is also the point of tangency of the B-
excircle ω1 of the triangle ABC to the side AC and K is also the point of tangency
of the D-excircle ω2 of the triangle ACD to the side AC.
Once these facts have been established, we turn to geometric transformations and
employ several homotheties.
Let S be the intersection of the common exterior tangents of ω1 and ω2 . The
homothety of center S that takes ω2 into ω1 is the composition of the homothety of
center L that takes ω2 in ω1 and the homothety of center B that takes ω1 to ω1 . By
Theorem 2.6, S, L, B are collinear. Writing the homothety of center S that maps ω2
to ω1 as the composition of the homothety of center D that maps ω2 to ω2 and the
homothety of center K that maps ω2 to ω1 , we deduce that S, D, K are collinear. It
follows that S is the intersection of the lines BL and DK.
Next, let K be the point that is diametrically opposite to K in ω1 and L the
point that is diametrically opposite to L in ω2 . Let T be the image of K through
the direct homothety that maps ω1 to ω. Then B, K , T are collinear. Note that
the tangents at K , L , T to the respective circles are parallel to AC, which means
that T is the image of L through the homothety of center D that maps ω2 to ω.
Consequently D, L , T are collinear. But B, K , L are also collinear because K
is mapped to L by the homothety that maps ω1 to ω1 . Thus B, K , L, S, T are
collinear. Similarly D, K, L , S, T are collinear. It follows that S = T , being the
unique point of intersection of these two lines. So the two tangents intersect on ω,
as desired.
Remark Dan Schwarz has pointed out to us that the common interior tangents of ω1
and ω2 intersect at the intersection point U of the diagonals AC and BD. Obviously
U is on AC, because AC is one of these common tangents. On the other hand, U is
the center of the inverse homothety that maps ω1 to ω2 , which you can write as the
composition of the homothety of center B that maps ω1 to ω, and the homothety of
center D that maps ω to ω2 . Hence U, B, D are collinear.
The restriction AB = BC from the statement is superfluous; the result holds true
in the equality case as well.
Source International Mathematical Olympiad, 2008.
123 This problem is similar to the previous. We argue on Fig. 10.42, in which γ is
the incircle of the triangle CP Q, is the incircle of the triangle AP Q, and is the
A-excircle of the triangle AP Q.
Let γ touch P Q at U and let touch P Q at V . If ω touches AB, BC, CD,
and DA, respectively, at K, L, M, and N, then, by repeating the argument from the
previous problem,
CP + P Q − CQ CP + CM + P Q − CQ − CL
PU = =
2 2
P M + P Q − QL
=
2
Fig. 10.42 An extension of A

the 2008 International
Mathematical Olympiad
problem
ω
D
B T
C Γ
I γ
P Q
U R V
Ω
P K + P Q − QN P K + AK + P Q − QN − AN
= =
2 2
AP + P Q − AQ
= ,
2
so also touches P Q at U . Since P U = QV , the C-excircle of CP Q also touches
P Q at V . In some sense, AP Q and CP Q are sister triangles!
Now let us take advantage of the several circles and tangencies in order to apply
a series of homotheties. Place the figure such that the line P Q is horizontal, so the
concepts of points being up/top or down/bottom make sense. The point U is the
bottom point of both γ and , and T is the bottom point of ω. Since there is a
homothety with center A that takes ω to , A, T , and U are collinear.
We are particularly interested in the center T of the direct homothety that takes
γ to . From the observations made in Sect. 2.2.1 II applied in the triangle CP Q, if
U is the point diametrically opposite to U in γ , then, since V is also the tangency
point of the C-excircle of CP Q in P Q, it follows that the points C, U , and V are
collinear. Because U and V are top points in γ and , T lies on CV as well. On
the other hand, the inverse homothety that takes ω to γ has center C, so the bottom
point V from ω is taken to the top point U ; that is, T lies on CV , too.
Now we add the circle ω back into the equation and consider the direct
homotheties between γ , , and ω. Their centers are T (ω to γ ), A (ω to ),
and O (γ to ω). These three points are also collinear (Monge’s Theorem proved
in Problem 86), meaning that T lies on AO. If instead we consider ω, γ , and , the
corresponding centers are O, U (γ to ), and A (ω to ). Then U , A, and O are
also collinear for the same reason, and the point U also lies on AO. This means that
T lies on the line AU , which also passes through T .
We has just proved that AU and CV intersect at both T and T . Let us rule out
the case where T lies on AC, for then the configuration is symmetric with respect
to AC, making the problem immediate. Thus we can conclude that T = T .
Finally, we play with the circles ω, γ , and once more, but now using inverse
homotheties. The center of the inverse homothety that takes γ to ω is C, and
the center of the direct homothety that takes ω to is A, so the center of the
inverse homothety that takes γ to lies on the line AC (d’Alembert’s Theorem;
see Problem 86). It also lies on the common tangent P Q, so it is the intersection of
AC and P Q, which is R.
Since R and T are the centers of the inverse and direct homotheties that take γ
to , both circle centers, one of which is the incenter I of CP Q, lie on the line RT ,
and we are done.
Source Brazilian Mathematical Olympiad, 2017, proposed by Géza Kós.
124 Place the original pentagon so that A1 is at the origin of a system of complex
coordinates, and let a2 , a3 , a4 , a5 be the respective coordinates of the other vertices.
−−−→
Let Aj be the translate of Aj by A1 Aj , j = 2, 3, 4, 5. These points have the
coordinates 2a2 , 2a3 , 2a4 , 2a5 ; thus the pentagon P = A1 A2 A3 A4 A5 is the image
of P1 = A1 A2 A3 A4 A5 through a homothety of center A1 and ratio 2. Because the
center of the homothety is on the perimeter of P1 and P1 is convex, P1 lies inside of
P . Similar homotheties exist for P and each of P2 , P3 , P4 , P5 (see Fig. 10.43), so
P1 , P2 , P3 , P4 , P5 lie all inside P . Since the similarity ratio between P and any
of these polygons is 2, the ratio of their area is 4. But the sum of the areas of
P1 , P2 , P3 , P4 , P5 is 5 times the area of any one of them. So they must overlap
inside P , and the problem is solved.
Fig. 10.43 A pentagon and A5

its translates
A5
A1 A4
A4
A2 A3
A3
A2
x1 x3
F1(x)
x2
1
3
Fig. 10.44 The definition of F1 and the points x1 , x2 , x3 for the case of three lines
Remark Compare this to Problems 51, 53, 56. The idea is always the same: map
your surface by isometries into some given region, then use the area of the region
as an upper bound for the sum of the areas of the images of your surface to either
bound the original area from above or to discover overlaps.
125 We study what the points xk are supposed to be. Identify each k with R, and
let fk : k+1 → k be the orthogonal projection, 1 ≤ k ≤ n. Define
F1 = f1 ◦ f2 ◦ · · · ◦ fn , Fk = fk ◦ · · · ◦ fn ◦ f1 ◦ · · · ◦ fk−2 ◦ fk−1 , k > 1.
The map F1 is shown in Fig. 10.44 in the case of three lines. The points xk from the
statement are fixed points of Fk , k ≥ 1. How many fixed points do these maps have?
For each k, Fk : R → R is a nonconstant linear transformation (because no
two lines are perpendicular), hence is of the form Fk (x) = ak x + bk . Note also
that ak < 1 (because not all lines are parallel). It follows that Fk is a homothety of
a line into itself. This homothety has exactly one fixed point because the equation
Fk (x) = x has the unique solution xk = bk /(1 − ak ). But do these points fit into
a cycle? All you have to do is find the fixed point x1 of F1 and then project it
successively to n , n−1 , . . . , 2 . If you project the last point to 1 , you must get
back x1 ; hence you are indeed in a cycle, and the problem is solved.
Remark Figure 10.44 shows the points x1 , x2 , x3 in the case of three lines.
126 Note first that for every point P inside the triangle, there is a vertex, say A,
such that P A ≤ 9h/10, where h is the altitude of the triangle. Indeed, the circles
centered at the vertices and of radii 2h/3 cover the interior, and 2/3 < 9/10.
Fig. 10.45 The strategy of

the grasshopper
ω1
ω0
Let the flower be at C0 and consider the disk ω0 of center C0 and some small
radius . Let X be a vertex whose distance to the flower is less than 9h/10. Consider
the homothety of center X and ratio 10/9, and let ω1 the image of ω0 . The center
C1 of ω1 is still inside the triangle, and if the grasshopper is inside ω1 , then when it
jumps toward X, it lands inside ω0 .
Replace ω0 by ω1 and C0 by C1 and repeat the process to obtain the disk ω2
centered at C2 . Inductively, obtain a sequence of disks ω1 , ω2 , ω3 , . . . whose centers
are inside the triangle and whose radii grow in an unbounded fashion. For some n,
the nth disk will cover the triangle and will therefore contain the grasshopper inside.
Now the grasshopper can jump
ωn → ωn−1 → · · · → ω1 → ω0 ,
landing within distance from the flower. The last step of grasshopper’s strategy is
illustrated in Fig. 10.45.
Remark It is worth mentioning that nobody solved the problem during the contest;
the information that a homothety-based solution exists really helps.
Source 4th Mathematical Olympiad of the Former Soviet Countries, 2006.
127 In a system of complex coordinates, we let the positions at the start and the
velocities of the turtles be ai , bi ∈ C, respectively, i = 1, 2, . . . n, where |bi | = r
for all i. At time t, the homothety centered at the origin and with ratio rt1 maps the
ith turtle to the point arti + bri , which can be made arbitrarily close to the point bri on
the unit circle by choosing t sufficiently large. The points bri , i = 1, 2, . . . , n, being
on the unit circle, are the vertices of a convex polygon. If t is sufficiently large, the
images of the turtles are very close to these points, and so they are still the vertices
of a convex polygon. The turtles themselves will be at the vertices of the inverse
image of this polygon through the homothety.
Remark In this problem we change the scale, we look from afar, and the turtles
seem to start at the same point. The problem is not true if the turtles have different
Fig. 10.46 Strategy for

finding the squares
velocities, and homothety can be used to construct a counterexample. (Do you see
how?)
Source Kvant (Quantum), proposed by V. Prasolov.
128 If we compare the numbers 1/5 and 4/5, we notice the factor 4, which
suggests immediately the idea of scaling: if a square K has too little black, make it
four times smaller.
Take n sufficiently large so that all black squares lie in some 2n × 2n square and
such that their area is at most 1/5 of the area of this square. Cut this square into four
equal squares. The black region of each of these squares is at most 4/5 of its area.
Those squares whose black region is more than 1/5 of their surface are retained,
those that have no black squares are discarded, and for the others we repeat the
process. After finitely many steps, one obtains the desired family of squares. This
strategy is illustrated in Fig. 10.46, where the bold lines are the cuts and the squares
that contain no black are discarded.
Remark The problem can be rephrased in space. What should the numbers 1/5 and
4/5 be replaced with?
Source Kvant (Quantum), proposed by G.A. Rosenblium.
129 The solution uses the following result:
Lemma (I.M. Yaglom) Every convex polygon that is not a parallelogram has three
sides such that the polygon itself lies inside the triangle formed by the lines of
support of these sides.
Proof Let P be the polygon. If P is a triangle, there is nothing to prove. Otherwise
P has two nonparallel sides that do not share a vertex. Extend these sides until they
intersect, thus producing a polygon P with fewer sides than P (see Fig. 10.47).
We keep repeating this procedure until we are stuck. This can only happen if
we obtain a triangle, in which case we are done, or if we obtain a parallelogram.
In the latter situation we go one step back and adjust our procedure so as not to
Fig. 10.47 Proof of

Yaglom’s Lemma
Fig. 10.48 Proof of Yaglom’s Lemma
A
h
F
E
B D C
Fig. 10.49 Proof of the Gelfand-Markus Theorem
obtain a parallelogram, using a different pair of sides (see Fig. 10.48). This proves
the lemma.

Returning to the problem, consider three sides of P whose lines of support form
a triangle ABC containing P inside. Choose in the interiors of the three sides the
points D, E, F , so that D ∈ BC, E ∈ AC, F ∈ AB. Let also M be a point inside
P. The segments MD, ME, MF dissect the polygon into three polygons P1 , P2 ,
respectively, P3 (Fig. 10.49).
A homothety of center A and ratio slightly less than 1 transforms P into a
polygon still covering P1 . Analogous homotheties with centers B and C transform
P into polygons that cover P2 , respectively, P3 . Hence the three images of P
through these homotheties cover P entirely.
In the case of a parallelogram, the four vertices must be in different homothetic
copies of the parallelogram, so the number is at least 4. And, of course, 4 copies
suffice. The problem is solved.
Remark The result is true for general convex regions in the plane, as any such
region can be approximated by convex polygons. Israel Gohberg and Alexander
Markus have proposed the same problem in space. In space, what numbers should
correspond to 3 and 4?
130 Because the triangles are similar, there is a positive number k such that
AC1 /AB1 = AC2 /AB2 = AC3 /AB3 = k, and a directed angle α such that
B1 AC1 = B2 AC2 = B3 AC3 = α. The spiral similarity of center A with
angle α and scaling factor k maps the line through B1 , B2 , B3 into the line through
C1 , C2 , C3 .
Remark The triangles are mapped into each other by spiral similarities, and spiral
similarities come in pairs.
131 (a) First solution. Consider X1 , X2 ∈ ω, and let P be the intersection of the
lines X1 X1 and X2 X2 . By Theorem 2.22, the center M of the spiral similarity is
the intersection of the circumcircles of P X1 X2 and P X1 X2 . Thus P is the second
intersection point of the circumcircles of MX1 X2 and MX1 X2 , and this is the
second intersection point of ω and s(ω). If follows that the lines XX pass through
this intersection point. The situation is shown in Fig. 10.50.
Second solution. Let ω be the unit circle, and let M have coordinate equal to 1.
Let s have ratio r and angle α, so that s(z) = w(z − 1) + 1, where w = reiα . For a
point ∈ ω, the line determined by and s() has the equation

1 1 1

z w( − 1) + 1 = 0,

z w( − 1) + 1
which is the same as
(z − )(w − 1)( − 1) − (z − )(w − 1)( − 1) = 0.
Let u = (w − 1)/(w − 1), which is a number of absolute value 1. Then, using the
fact that = 1, we can write the equation of the line as
Fig. 10.50 The intersection h(ω)

point of the pencil of lines
determined by points and P
their images
M
ω
X2
X1
X1
X2
( − 1)z + ( − 1) = u( − 1)z + u( − 1).
Now we can write this equation for another point on the unit circle and solve the
system in z and z to find the intersection point of the two lines and then notice
that this intersection point does not depend on the two chosen points, or simply
notice that u itself satisfies the equation of the line, and therefore u is the common
intersection point of all such lines. Since u has absolute value 1, it is on the
unit circle, that is, on ω. But by exchanging the roles of ω and s(ω), that is, by
considering s −1 instead of s, we deduce that u must be also on s(ω), so u is the
second intersection point of these circles.
Remark This shows how to construct the image of a point X ∈ ω under the spiral
similarity: intersect the line AX with s(ω).
132 Look at Fig. 10.51. Consider a spiral similarity s(B) of center X that maps the
circle through B; call it ω to the circle through C; call it ω . Because AB is tangent
to ω = s(ω), s(B) = A and s(A) = C, as we are in the limiting case of the result
that was proved in Problem 131. So we are in the situation described in Sect. 2.4.3,
and we can invoke Theorem 2.23 to conclude that X lies on the symmedian from A
in the triangle ABC.
133 First solution. It is natural to start by locating the centers of possible spiral
similarities (see Fig. 10.52). By Theorem 2.22, the center O of such a spiral
similarity is on the circumcircle of P A1 A2 . And moreover, it should satisfy
OA2 /OA1 = k; hence it belongs to the locus of the points X with the property
that XA2 /XA1 = k, and this is the Apollonian circle in question. There are two
intersection points of the two circles, so there are at most two such spiral similarities
(because the spiral similarity is completely determined by the center and the image
of a point).
On the other hand, if O is one of the intersection points, then using inscribed
angles, we obtain (OA1 , OA2 ) = (P A1 , P A2 ) = (1 , 2 ), and once we know
that it maps A1 to A2 it must map 1 to 2 . Hence there are exactly two such spiral
similarities.
Second solution. The above geometric facts can be read immediately in the
analytic equation of the spiral similarity of center b, ratio k, and angle α:
Fig. 10.51 Spiral similarities A

and symmedians
C
B
X
Fig. 10.52 A spiral

similarity that maps a line to a
line
A2
O
P 1
A1
Fig. 10.53 A spiral M

similarity that maps a circle
to a circle
O1 O2
ω1
ω2
s(a1 ) − b
= keiα ,
a1 − b
From here we infer that the angle of the spiral similarity should be the angle between
the two lines, but that is also the angle formed by the line through b and s(a1 ) = a2
with the line through b and a1 , which implies that b, a1 , a2 , p are concyclic. On the
other hand, taking absolute values, we obtain |s(a1 ) − b|/|a1 − b| = |a2 − b|/|a1 −
b| = k, which shows that b is on the said Appolonian circle.
134 First solution. Let the centers and radii of ω1 and ω2 be O1 , r1 and O2 , r2 ,
respectively, and let the spiral similarity be of center M and scaling factor k
(Fig. 10.53). Then r2 /r1 = k, so the scaling factor is completely determined, and
since O1 is mapped to O2 , the center M of the spiral similarity is on the Apollonian
circle MO2 /MO1 = k. Also, because O2 MO1 = 90◦ , M is on the circle of
diameter O1 O2 . The two circles intersect at two points, and each of the intersection
points determines a spiral similarity.
Second solution. We want to map the circle |z−a1 | = r1 to the circle |z−a2 | = r2
by a spiral similarity whose equation is s(z)−m = ±ki(z−m). The spiral similarity
is therefore s(z) = ±kiz − (±kiz − 1), which we substitute in the equation of the
second circle.

1 r2
| ± kiz − (±kiz − 1)m − a2 | = r2 ⇐⇒ z − 1 ∓ m − a2 = .
ki k
The result should be the equation of circle. For that we should have r2 /k =
the first
r1 , meaning that k = r2 /r1 and 1 ∓ ki1 m − a2 = a1 , which then yields one
solution for each choice of signs. Hence there are exactly two spiral similarities.
Remark To emphasize the idea of the second solution, for a curve F1 (z) = 0 to
be mapped to a curve F2 (z) = 0 by s, the equations F2 (s(z)) = 0 and F1 (z) = 0
should describe the same set of points.
135 Let P be the polygon. On a side of P, choose a point M that is not a vertex
and consider a spiral similarity s with center M, angle BAC, and ratio AB/AC.
Then P and s(P) cross at M, and so they must cross at one more point (at least). Let
P be the second crossing point of P and s(P), and let N = s −1 (P). Then N is on
P. We have NMP = BAC, and MN/MP = AB/AC, so the triangles MNP
and ABC are similar. Done.
136 First solution. Consider the spiral similarity of center H , angle 90◦ , and ratio
BP /BC, which maps the triangle BH C to P H B (see Fig. 10.54; the angle of the
spiral similarity in the figure is oriented clockwise). This spiral similarity maps C
to B, and therefore it maps the line CD to a line through B that is perpendicular to
CD, and this line is BC. Since BQ/CD = BP /BC, it follows that D is mapped
to Q, and therefore DH Q = 90◦ , as desired.
Second solution. For a coordinate solution place the origin at B, and let A, C, D
have coordinates i, 1, 1 + i, respectively. Let 0 < t < 1 so that P = it, Q = t. The
line CP has the equation

1 1 1

1 it z = 0 ⇐⇒ (t − i)z + (t + i)z − 2t = 0.

1 −it z
The projection of the origin B onto P C is the point H whose coordinate is
(t − i) · 0 − (t + i) · 0 + 2t t
h= = .
2(t − i) t −i
Fig. 10.54 DH Q = 90◦ A D

P
B C
Q
−−→
The vector H D has the complex number form
t it − i + 1
1+i−h=1+i− = ,
t −i t −i
−−→
while the vector H Q has the complex number form
t t 2 − it − t it − i + 1
t −h=t − = = (−it) .
t −i t −i t −i
We recognize that the second vector is the first multiplied by −it; thus Q is the
image of D through the spiral similarity of center H , clockwise angle 90◦ , and ratio
t = BP /BC.
137 (a) The pentagons can be mapped into each other by a spiral similarity or
a translation. We may therefore invoke the Averaging Principle (Theorem 2.18)
to conclude that the midpoints of the given segments form a pentagon similar to
each of the two pentagons and hence regular. An instance of this result is shown in
Fig. 10.55.
(b) Consider the spiral similarity s of center A1 that maps the circumcircle of
A1 A2 A3 A4 A5 to the circumcircle of A1 A2 A3 A4 A5 . Then s(Aj ) = Aj , j =
2, 3, 4, 5, and the conclusion follows from Problem 131.
Remark So the Averaging Principle is true for all transformations of the form
f (z) = rz + s, r, s ∈ C, including translations.
138 Examining Fig. 10.56 we observe that the triangle ABC is mapped into
the triangle AOD by a spiral similarity of center A. The points M, N, A are the
midpoints of the segments BO, CD, and the degenerate segment AA. Applying the
Fig. 10.55 The Averaging A5 A4

Principle applied to regular
pentagons
A5
A4
A1 A3
A2 A3
A1
A2
Fig. 10.56 A spiral N

similarity hidden in a square D C
O
M
A B
Fig. 10.57 Equilateral A

triangles with two sides
sharing the midpoint
A
B
B M C
C
Averaging Principle (Theorem 2.18), we obtain that the triangle AMN is similar to
each of the triangles ABC and AOD, so it is an isosceles right triangle.
139 Let M be the common midpoint of BC and B C

√ (Fig. 10.57). Consider the
◦
spiral similarity s of center M, angle 90 , and ratio 3 that maps B to A. Then√
s(B ) = A and consequently s(BB ) = AA . We thus obtain that AA /BB = 3,
and the two segments form a 90◦ angle.
140 First solution. Let A3 , B3 , C3 be the midpoints of the segments OA, OB, OC,
respectively. Then ABC is mapped to A3 B3 C3 by a homothety h of center O and
ratio 1/2. On the other hand, the triangles ABC and A1 B1 C1 are directly similar, so
by Proposition 2.21, there is a spiral similarity s that maps A1 B1 C1 to ABC. Then
h ◦ s is a spiral similarity that maps A1 B1 C1 to A3 B3 C3 . The points A3 , B3 , C3
divide the segments A1 A2 , B1 B2 , and C1 C2 in half (as they are the centers of the
three parallelograms), so we can apply the Averaging Principle (Theorem 2.18) to
the triangles A1 B1 C1 and A3 B3 C3 and the points A2 , B2 , C2 in order to deduce
that there is a spiral similarity that maps A1 B1 C1 to A2 B2 C2 . Therefore A2 B2 C2
is similar to A1 B1 C1 , and so it is similar to ABC. The situation is described in
Fig. 10.58, with the three triangles that appear when using the Averaging Principle
drawn with dotted line.
Second solution. We can be bolder and apply the general form of the Averaging
Principle (Theorem 2.19) to the similar triangles ABC, A1 B1 C1 and the complex
Fig. 10.58 The Averaging A2

Principle applied to A1 B1 C1 ,
A3 B3 C3 , and A2 B2 C2 O
A3
A
B2
A1
C2
C3 C1
B3
B1
B C
coefficients 1 and −1. These coefficients define the points A2 , B2 , C2 if the origin
of the coordinate system is chosen at O, because a1 + a2 = a + 0, b1 + b2 = b + 0,
and c1 + c2 = c + 0. Hence A2 B2 C2 is similar to each of the triangles ABC and
A1 B1 C1 .
141 Taking a closer look at the triangle MAj Mj , we see that Mj is the image
of Aj under the spiral similarity s of center M, angle α2 − 90◦ , and ratio 2 sin α2 .
Consequently M1 M2 . . . Mn is the image of A1 A2 . . . An by the spiral similarity s,
and so its area is 4S sin2 α2 .
Remark Note the spiral similarity hidden in the rotation!
Mongolia.
142 The argument can be followed on Fig. 10.59. Using inscribed angles and the
fact that AM and AN are tangents, we obtain that
MAB = BNA and AMB = BAN.
Consider a spiral similarity of center A that maps AMB to AC N. Then the angle
between the lines C N and MB is the angle of the spiral similarity, which is
BAN = AMB. Hence the lines AM and C N form the same angle with MB, so
they are parallel.
On the other hand, as spiral similarities come in pairs (Theorem 2.25), there is a
spiral similarity that maps the triangle AMC to ABN. The angle between MC and
BN is the angle of the spiral similarity, so it is equal to MAB = BNA. Thus
MC and AN make the same angle with BN , so they are parallel. It follows that
AMC N is a parallelogram, showing that C = C .
Fig. 10.59 Proof that

AP Q = AN C A
P
M B
N
Q
C=C
Fig. 10.60 The closed path

of the grasshopper
We deduce that AMB and ACN are mapped into each other by a spiral similarity,
so we can invoke the Averaging Principle (Theorem 2.18) to conclude that there is
a spiral similarity that maps ACN to AQP . Consequently ANC = AP Q, and
we are done.
143 A situation with three circles is shown in Fig. 10.60. Let si be the spiral
similarity of center O that maps ωi to ωi+1 , (ωn+1 = ω1 ). Problem 131 shows
that si (Xi ) = Xi+1 . Also sn ◦ sn−1 ◦ · · · ◦ s1 is a spiral similarity that maps ω1 into
itself. As the center of this transformation lies on ω1 , this transformation can map ω1
into itself only if it is the identity transformation. Which implies that Xn+1 = X1 ,
144 First solution. The configuration is shown in Fig. 10.61. Let α1 = P BA =
RBC = QAC and α2 = P AB = RCB = QCA and let also k1 =
P B/AB = RB/BC = QA/CA, k2 = P A/AB = RC/BC = QC/AC. The
spiral similarity of center B, angle α1 , and ratio k1 maps A to P and C to R. The
spiral similarity of the same angle and ratio but of center A maps C to Q and keeps A
Fig. 10.61 AP RQ is a A
parallelogram Q
P
R
B C
fixed. Thus P R and AQ are images of the same segment through spiral similarities
of the same angle and the same ratio, so they are parallel and have equal length. This
shows that AP RQ is a parallelogram.
Second solution. The argument is an application of Theorem 2.25: “spiral
similarities come in pairs.” As there is a spiral similarity that maps AP B to RBC,
there is also a spiral similarity that maps P BR to ABC. Its angle is −α1 . Also there
is a spiral similarity that maps BRC to AQC, so there is a spiral similarity that maps
ABC to QRC. Its angle is α2 . Composing we find that there is a spiral similarity of
angle α2 − α1 = (BP , AP ) that maps the triangle P BR to QRC. Hence QR and
AP make the same angle with BP , so they are parallel. Moreover,
PB P B AB P B AC P B AC AQ
= · = · = · ·
QR AB QR AB QC AB AQ QC
1 PB PB
= k1 · · = .
k1 AP AP
Hence QR = AP , showing that AP RQ is a parallelogram.

Third solution. We use coordinates, with the usual convention that the lower case
letter is the coordinate of the upper case point. With the same notation as in the first
solution, P is obtained from A by a spiral similarity of center B, ratio k1 , and angle
α1 , so
p−b
= k1 eiα1 .
a−b
Therefore p = b + k1 eiα1 (a − b).

The same spiral similarity maps C to R, while a spiral similarity of center A and
same angle and ratio maps C to Q. Hence
r = b + k1 eiα1 (c − b), q = a + k1 eiα1 (c − a).
Then
a+r a + b k1 iα1 p+q

= + e (c − b) = ,
2 2 2 2
showing that P Q and AR have the same midpoint; hence AP RQ is a parallelogram.
Remark The original version of the problem was phrased with three isosceles
triangles. This more general version discloses the trick, as directly similar triangles
lead to spiral similarities.
Source Romanian Mathematical Olympiad, 2001, the particular case where the
three similar triangles are isosceles was on the short list of the International
Mathematical Olympiad in 1983, proposed by Belgium.
145 Denote by K, L, M the midpoints of the sides BC, CA, AB, respectively.
First, we verify the case where P = M and Q = L (note that in this case AMOL is
cyclic because it has two opposite right angles). Because OK, OL, OM are orthog-
onal to BC, CA, AB, respectively, they are also orthogonal to LM, MK, KL,
respectively. It follows that O is the orthocenter of KLM, and consequently K
is the orthocenter of OML, which solves the problem in this case.
In the general case, shown in Fig. 10.62, we have
AP O = 180◦ − AQO = OQC,
and so OP M = OQL, showing that the right triangles OP M and OQN are
directly similar. It follows that there is a spiral similarity s1 of center O such that
s1 (OP M) = OQN . And since spiral similarities come in pairs (Theorem 2.25),
there is a spiral similarity s2 of center O such that s(OML) = OP Q.
Let H = s2 (K). As O is the orthocenter of KLM, O is also the orthocenter of
H P Q, so H is the orthocenter of P OQ. And as s2 (OMK) = OP H and spiral
similarities come in pairs, there is a spiral similarity that maps OMP to OKH .
Hence OKH = OMP = 90◦ . This implies that the line KH is perpendicular to
OK, so it must coincide with the line BC. Therefore H is on BC, and the problem
is solved.
Fig. 10.62 The orthocenter A

of P OQ is on BC
Q
M L
P
O
B K H C
Fig. 10.63 O1 O2 passes

through N
A D
O1
M
O2
C
B
Source All-Russian Mathematical Olympiad, 2011, solution from T. Andreescu,

M. Rolínek, J. Tkadlec, Geometry Problems from the AwesomeMath Year-Round
Program, XYZ Press, 2013.
146 The solution can be followed on Fig. 10.63. By angle chasing in the
circumcircles of MAD and MBC and noticing that the points A and O1 are on
the same side of the line MD and B and O2 are separated by the line MC, we can
write
MO1 D = 2 A = 2(180◦ − B) = MO2 C.
It follows that the isosceles triangles MO1 D and MO2 C are similar and oriented
the same way. Hence there is a spiral similarity of center M that maps these triangles
into each other. In other words, there is a spiral similarity of center M that maps O1
to O2 and D to C.
By Theorem 2.25 (spiral similarities come in pairs), there is a spiral similarity
that maps O1 to D and O2 to C. Now we can apply Theorem 2.22 to determine
that the center of this last spiral similarity is the second intersection point of the
circumcircles of N O1 D and N O2 C, where N is the intersection of the lines O1 O2
and CD. But we know that the center of this spiral similarity is the point M, so M is
at the intersection of the circumcircles of N O1 D and N O2 C. Consequently N is
at the intersection of the circumcircles of MO1 D and MO2 C, and hence N = N.
It follows that N is on O1 O2 , as desired.
Remark We have proved additionally that N is on the line CD.
Source International Zhautykov Olympiad, 2013.
Fig. 10.64 Finding the A A1

common center of the spiral
similarities. M
B D
C
147 We argue on Fig. 10.64. The points A, B, C, D, E, F form a complete

quadrilateral. We will show that center of the spiral similarities in question is the
Miquel point of this complete quadrilateral.
By Theorem 2.22, the center of the spiral similarity that maps the segment EF
to the segment BC is the intersection point of the circumcircles of DF B and DEC.
This is the Miquel point M of the complete quadrilateral. Because the triangles
A1 F E and ABC are similar, M is also the center of the spiral similarity that maps
the two triangles into each other. As the composition of two spiral similarities with
the same center is a spiral similarity with the same center, M is the center of the
spiral similarities that map the four triangles into one another.

148 The situation is described in Fig. 10.65. By the result proved in Problem 131,
the spiral similarity of center S that maps ω1 to ω2 maps A to C and B to D. It also
maps the circumcenter O1 of ω1 to the circumcenter O2 of ω2 . By the Averaging
Principle (Theorem 2.18), there is a spiral similarity of center S that maps the
configuration determined by the points S, A, B, O1 to the configuration determined
by the points S, M, N, O. It follows that O is the circumcenter of SMN . But P is
the reflection of S over O1 O2 , and since O ∈ O1 O2 , it follows that SO = P O.
Hence P is on the circumcircle of SMN . We conclude that O is the circumcenter
of MNP , as desired.
Source T. Andreescu, M. Rolínek, J. Tkadlec, Geometry Problems from the

AwesomeMath Year-Round Program, XYZ Press, 2013.
Fig. 10.65 O is the C

circumcenter of MN P
B
P
M N
O1 O
O2
D
A S
Fig. 10.66 P rotates to Q C

about R
30°
30°
P
45° 45°
M
A 15° 15° B
45°
R
30°
T
149 First solution. The point P is transformed into the point Q by a spiral
similarity of center B, angle 45◦ , and ratio BC/BP , which maps P to C, followed
by a spiral similarity of center A, angle 45◦ , and ratio AQ/AC, which maps C to
Q (see Fig. 10.66). But AQ/AC = BP /BC, because the triangles ACQ and BCP
are similar. So the product of the ratios of the two spiral similarities is 1, and hence
their composition is a rotation. The angle of this rotation is 45◦ + 45◦ = 90◦ .
If we show that the center of rotation is R, then we are done. For this we have to
show that R is fixed by the composition of the two spiral similarities.
Let T be the image of R through the first spiral similarity, that is, RBT = 45◦
and BT /BR = BC/BP . We deduce that the triangles BRT and BP C are similar,
so BRT = BP C = 180◦ − 75◦ . Denoting by M the midpoint of AB, RM
is the perpendicular bisector of AB, so BRM = 75◦ . Consequently T , R, M
are collinear, and therefore T is on the perpendicular bisector of AB. But then
using the symmetry of the figure consisting of A, B, R, and T with respect to the
perpendicular bisector of AB, we deduce that the second spiral similarity maps T
back to R. The problem is solved.
Second solution. Here is the analytical solution, maybe not as elegant as the
synthetic one. With the standard convention, we will use lower case letters to denote
the coordinates of the points designated by upper case letters. We place the origin
5π i
of the coordinate system at R, so that r = 0. Then a = e 6 b.
The point Q is obtained from C by a spiral similarity of center A, angle 45◦ , and
ratio AQ/AC = BP /BC = sin π6 / sin 7π 12 . So
q −a sin π6 π i
= e4.
c−a sin 7π
12
We obtain

sin π6 πi sin π6 πi
q= e c+ 1−
4 e 4 a.
sin 7π
12 sin 7π
12
Similarly

sin π6 − π4i sin π6 − π4i
p= e c+ 1− e b.
sin 7π
12 sin 7π
12
πi
We must show that q = ip = e 2 p. Note that the coefficient of c in q does have this
5π i
property. On the other hand, using the fact that a = e 6 b, we are left to show that

7π π πi 5π i 7π π − πi
sin − sin e 4 e 6 = sin − sin e 4 i.
12 6 12 6
The solution to the problem ends with a computation based on the formulas
π √ √
12 = sin + = sin π3 cos π4 + cos π3 sin π4 = 6+ 2
sin 7π 3
π
4 4
√ √ 5π i
√
± π4i
e = 2
2
± 2
2 i, e 6 =− 2
3
+ 12 i.
Source 17th International Mathematical Olympiad, 1975, proposed by the Nether-

lands.
Fig. 10.67 D ∈ BC if and A

only if O is on the
circumcircle of AEF
M F
T O
E
B D C
150 Both the direct and the converse implication can be followed on Fig. 10.67.
The point M is the Miquel point of the complete quadrilateral determined by
the lines AC, AB, BC, and EF . As seen in the proof of Miquel’s Theorem
(Theorem 2.24), there is a spiral similarity s of center M that maps the segment
EF to the segment BC.
To prove the direct implication, assume D ∈ BC. Then T = s −1 (D) is on EF .
The triangles MT D and MEB are similar (automatic similarity). But EF is the
perpendicular bisector of the line segment MD, so the triangle MT D is isosceles.
It follows that the triangle MEB is isosceles as well, so EMB = EBM. The
Exterior Angle Theorem in this triangle then gives
AEM = 2 ABM = AOM.
This shows that O is on the circumcircle of the triangle AEF .

For the converse we use the same spiral similarity. Define T in the same way, but
now we do not know that T ∈ EF . But we know that O is on the circumcircle of
AEF and from this we obtain
2 ABM = AOM = AEM = ABM + BME.
So the triangle EMB is isosceles. This implies that the triangle T MD (which is
similar to EMB by automatic similarity) is also isosceles, and so T belongs to
the perpendicular bisector of MD, which is the line EF . But then D = s(T ) ∈
s(EF ) = BC, and we are done.
Remark The result remains valid if the circumcircles of the triangles AEF and
ABC are tangent. In this case M = A, and the solution is somewhat simpler: We
consider the homothety centered at A that transforms the circumcircle of AEF into
the circumcircle of ABC. This homothety takes EF into BC. Then D ∈ BC is
equivalent to the fact that EF is the midline of the triangle ABC, which is equivalent
to AEOF being cyclic.
Source Romanian Team Selection Test for the Junior Balkan Mathematical
Olympiad, 2018, solution by contestant Ioana Popescu.
151 The reader can examine this configuration of six intersecting circles in
Fig. 10.68. The triangles ABC and A1 B1 C1 are oriented the same way because
the order of the sides of ABC dictates the order of the vertices of A1 B1 C1 . So there
is a spiral similarity that takes the triangle A1 B1 C1 to the triangle ABC; let O be its
center. The segment AB is mapped to the segment A1 B1 , so O is also the center of
the spiral similarity that maps the segment AA1 to BB1 , as Theorem 2.25 shows that
spiral similarities come in pairs. This latter spiral similarity is centered at the second
intersection point of the circumcircles of ABC2 and A1 B1 C2 (by Theorem 2.22).
So O lies at the intersection of the circumcircles of ABC2 and A1 B1 C2 , the same
argument for the other pairs of circles.
152 Let ωB and ωC be the circumcircles of the triangles ACP and ABQ,
respectively, and let X = A be their second intersection point. The angle conditions
from the statement imply that AB is tangent to ωB and AC is tangent to ωC .
Problem 132 shows that X is on the symmedian from A of the triangle ABC.
B1
C1 C2
A2 B2
B A1 C
Fig. 10.68 Six circles intersecting at one point

U V
O
X
B Q P C
N M
Fig. 10.69 The construction of X, T , U , V , and S
Let AX intersect the circumcircle of ABC again at T . Finally, let U, V be the

midpoints of AB and AC, respectively, let O be the circumcenter of ABC, and let
S be the point where the tangents to the circumcircle of ABC at B and C meet
(Fig. 10.69). By Theorem 1.22, S is on the symmedian from A, so A, X, T , S are
collinear.
Using the similarity of the triangles ABX and CAX (which follows from
tangencies) and working with directed angles modulo 180◦ , we can write
BXC = BXA + AXC = XBA + BAX + XAC + ACX

= XAC + BAX + XAC + BAX = 2 BAC = BOC,
so B, X, O, C are concyclic. Since OBS = OCS = 90◦ , S is also on this circle,

so OXS = OBS = 90◦ . Then AXO = OXT = 90◦ , and since AO = OT ,
we have AX = XT .
Next observe that, using the tangencies of ωB and ωC to AB and AC, respec-
tively, imply that the triangles AQC and BP A are both inverse similar to the triangle
ABC and hence are directly similar to each other. But the similarity of the triangles
ABX and CAX implies that X is the center of the spiral similarity that maps the
segment BA to the segment AC. And the similarity of triangles AQC and BP A
implies that the same spiral similarity maps P to Q. Furthermore, it maps U to V
since they are the midpoints of sides BA and AC.
We conclude that X is the center of a spiral similarity mapping U P to V Q and
thus of the one mapping U V to P Q (spiral similarities come in pairs). But U V is
parallel to BC, and hence to P Q, so the triangles XU V and XP Q must actually
be homothetic. It follows that U, X, P are collinear, and the same is true about
Q, X, V . Dilating by a factor of 2 about A gives that B, T , M and, respectively,
C, T , N are collinear. So BM and CN meet at T , which is on the circumcircle, as
desired.
Source 55th International Mathematical Olympiad, 2014, proposed by Georgia,
solution by contestant Sammy Luo, United States of America.
153 Because the triangle DBC is isosceles, the Exterior Angle Theorem implies
that ADB = 2 C. So ADE = ADB/2 = C, and therefore DE||BC. We
deduce that DEB = DAE, both being equal to 180◦ − EBC. It follows that
the triangles DAE and DEB are directly similar (having two pairs of equal angles),
so they are mapped one into the other by a spiral similarity of center D. If we denote
by I and J the respective incenters of these triangles, as shown in Fig. 10.70, then
the spiral similarity maps I to J .
Fig. 10.70 A spiral A

similarity that maps DAE to
DEB I
E D
B C
A
E
B
C D
Fig. 10.71 ABC, ACD, ADE map into each other by spiral similarities
But as the spiral similarity maps the triangle DAI to DEJ and spiral similarities
come in pairs (Theorem 2.25), another spiral similarity will map the triangle DAE
to the triangle DI J . From this we infer that
(I J, AE) = (DI, DA) = I DA,
and this is further equal to C/2 = 90◦ − AEB/2 (because AEBC is cyclic). Thus
(I J, AE) is the complement of the angle formed by AE with the angle bisector of
AEB/2. And this implies that I J is orthogonal to the angle bisector of AEB, as
desired.
Olympiad, 2011.
154 The triangles ABC, ACD, ADE are directly similar, so they are mapped
into one another by spiral similarities centered at A. Reasoning on Fig. 10.71, we
continue the argument by noting that the spiral similarity that maps the triangle
ABC to ADE maps the segment BC to DE, and so, by Theorem 2.22, the center
A of this spiral similarity is the second intersection point of the circumcircles of
BCP and DEP . Said differently, P is the second intersection of the circumcircles
of ABC and ADE.
We apply the same result to the triangles ABC and ACD. Here the intersection of
BC and CD is C, and we are in the degenerate situation of Sect. 2.4.3. The line CD
intersects the circumcircle of ABC once at C and a second time at the intersection
of the segments BC and CD, which is C again. Hence CD is tangent to this circle
at C. By a similar argument, CD is tangent to the circumcircle of ADE.
We thus have a configuration of two circles that intersect at A and P with CD as
the common tangent. The line AP is the radical axis of the two circles, and it passes
through the point on CD that has equal powers with respect to the circles. And this
point is the midpoint of CD, as desired.
Remark If you do not like the limiting argument in the degenerate case, you can
also prove that CD is tangent to the two circles by an angle chasing argument.
the United States of America.
155 Let D, E, F be the feet of the altitudes from A, B, C, respectively
(Fig. 10.72).
There is a spiral similarity mapping the triangle CF Z to the triangle ADX.
Using Theorem 2.22 we deduce that the center of this spiral similarity is the
second intersection point of the circumcircles of BXZ and BDF , and by the same
argument, because spiral similarities come in pairs (Theorem 2.25), this is also
the second intersection point of the circumcircles of H DF and H AC. Hence the
circumcircle of XBZ passes through the intersection point M of the circumcircles
of BF H D and H AC (Fig. 10.73) different from H .
But M is also on BG. Indeed, if B is the point in the circumcircle of AH C
that is diametrically opposite to H , then AB ⊥CH and CB ⊥AH , showing that
AB BC and CB AB. Thus BAB C is a parallelogram and so BB passes through
the midpoint of AC; it is the median. Once we know this fact, then we can consider
the spiral similarity of center H that maps the circumcircle of BF H D to the
Fig. 10.72 Construction of Y

P and H
E
F
H P
X
B D C
Z
Fig. 10.73 Construction of Y

M
F
H P
M
X
B D C
circumcircle of AH C and notice that it must map B to B . But then B, H, B are

collinear, by Problem 131. So M is on the median from B.
Note that H M is perpendicular to BG, so M is on the circle of diameter H G that
we are interested in.
On the other hand, using the spiral similarity that maps the triangle ADX to the
triangle BEY , we obtain that the circumcircle of CXY passes through the second
intersection point N of the circumcircles of H AB and CEH D. And N is on the
circle of diameter H G, as well.
We can now rephrase our problem as follows:
Problem It is given a circle of diameter H G and the points B and C outside of
this circle. The lines BG and CG intersect the circle at M and N , respectively. The
point X is on the line BC (say C is between B and X). Prove that second point of
intersection, P , of the circumcircles of XBM and XCN is on the circle of diameter
H G.
Solution We solve this problem using Fig. 10.74. Let R be the second intersection
point of the circumcircle of XBM with the circle of diameter H G. Then XRM =
MBC (cyclic quadrilateral XBMR) and MRN = BGC (cyclic quadrilateral
RMN G). Thus
XRN + XCN = XRM + MRN + BCG

= GBC + BGC + BCG = 180◦ .
Fig. 10.74 The modified X

problem
B
R
M
G H
Fig. 10.75 There is a spiral A

similarity taking X to W and
B to C and a spiral similarity
taking B to C and W to Y M
Y
N
X H ω2
B W C
ω1
Hence XCN R is cyclic, and therefore R = P and the problem is solved.

Source Brazilian Mathematical Olympiad, 2015.
156 Because W X is a diameter (Fig. 10.75), XN W = 90◦ = BNC. Also,

NXW = NBW = NBC. It follows that the triangles NXW and NBC are
similar in the same orientation, so they are mapped one into the other by a spiral
similarity with center N. But spiral similarities come in pairs (Theorem 2.25); hence
there exists a spiral similarity sN with center at N, rotation angle 90◦ , and ratio
N C = cot B that takes X to W and B to C. Analogously, there is a spiral similarity
NB
sM with center at M, rotation angle 90◦ , and ratio BM

MC = tan C that takes B to C

and W to Y .
If we compose sN and sM , we obtain another spiral similarity, s = sM ◦ sN , such
that s(X) = sM (W ) = Y . This transformation has rotation angle 90◦ + 90◦ = 180◦ ,
that is, it is an inverse homothety. Let us find its center. Notice that
NH
= cot AH N = cot B
NA
and H NA = 90◦ , so sN (H ) = A. We can deduce in the same fashion that

sM (A) = H . So
s(H ) = sM ◦ sN (H ) = sM (A) = H.
The only fixed point of a homothety is its center, so the center of s is H . This means
that H , X, and s(X) = Y are collinear.
Source International Mathematical Olympiad, 2013, proposed by Warut Suk-
sompong and Potcharapol Suteparuk, Thailand, solution by contestant Aleddine
Sabbagh, Tunisia.
Chapter 11
Inversions
157 (a) Placing the center of inversion at the origin of the coordinate system, the
inversions have the equations χ1 = R12 /z1 and χ2 = R22 /z2 , so their composition
has the equation χ2 ◦ χ1 (z) = (R22 /R12 )z, which is the homothety of center O and
radius R22 /R12 .
(b) and (c) Let the inversion be χ (z) = R 2 /z and the homothety h(z) = kz.
Then h ◦ χ = kR 2 /z which is the inversion of ratio kR 2 with the same center and
χ ◦ h = (R 2 /k)/z which is the inversion of ration R 2 /k with the same center.
Remark The composition of two inversions of the same center is not commutative,
that is, χ2 ◦χ1 = χ1 ◦χ2 . Neither is the composition of an inversion and a homothety
with the same center commutative.
158 Take a Möbius transformation (or just an inversion, or a circular transforma-
tion) that maps the circle to a line. Using the Symmetry Principle (Theorem 3.15)
and the fact that inversion preserves angles, we reduce the problem to showing that
two points are reflections of each other over a line if and only if every circle or line
passing through the two points is orthogonal to the line. But a line is orthogonal to
a circle if and only if it passes through its center, and so the line in question must
be the locus of the centers of the circles that pass through the two points, and this
locus is the perpendicular bisector of the segment determined by the two points (see
Fig. 11.1).
Remark There is a less sophisticated way to see why if A and B reflect to each other
over a circle ω, then any circle ω passing through A and B is orthogonal to ω. As
one of A, B is inside ω and the other one is outside, ω intersects ω twice. Those
points are invariant under the inversion, and A and B are mapped into each other, so
ω must be invariant under the inversion. It is therefore orthogonal to ω.
It might be worth remembering that a circle that passes through two points that
correspond under an inversion is invariant under that inversion.
438 11 Inversions
Fig. 11.1 Line orthogonal to

circles passing through two
points
159 We can solve this problem easily using coordinates. Let a, b, p be the complex
coordinates of A, B, P , respectively, and let u be the complex coordinate of the
center of inversion, to be determined, and let R be the radius of inversion. Then the
images of the three points have complex coordinates
R2 R2 R2
+ u, + u, + u.
a−u b−u p−u
The condition that the third is the midpoint of the segment determined by the first
two is equivalent to the equality
1 1
+
= a−u b−u.
1
p−u 2
Solving for u we obtain
2ab − (a + b)p
u= .
a + b − 2p
Since p = a+b2 , as P is not the midpoint of AB, this last formula makes sense, and
it yields the complex coordinate of the center of inversion.
Remark The center of the inversion must be on the line through a, b, p, and you can
read this in the formula by choosing the coordinates a, b, p to be real (i.e., the line
through a and b is the x-axis), and then u is real, too. If we relax the requirement by
replacing inversion with Möbius transformation, then the problem becomes trivial
since Möbius transformations are parametrized by the images of three given points.
160 Let O be the center of inversion. By Theorem 3.2, OAB = OB A , hence
ABB A is a cyclic quadrilateral; let ω be its circumcircle. Then AB is the radical
11 Inversions 439
Fig. 11.2 AB and A B

intersect on the radical axis of
ω and ω ω
A A
B B ω
ω
Fig. 11.3 Circles tangent to three given pairwise tangent circles
axis of ω and ω , while A B is the radical axis of ω and ω . The two radical axes
intersect at the radical center of the three circles, which lies also on the radical axis
of ω and ω . The situation is described in Fig. 11.2.
161 Being orthogonal to ω, the two circles are invariant under inversion with
respect to ω. As inversion preserves tangencies, it maps the point of tangency of
the two circles to another point of tangency of the two circles. But there is only one
point of tangency, so this point has to be mapped into itself, proving that it lies on
the circle of inversion.
162 The three cases described in the statement can be seen in Fig. 11.3.
Let O be the tangency point of ω1 and ω2 . Consider an inversion of center O.
This maps ω1 and ω2 to two parallel lines ω1 and ω2 , with the image ω3 of ω3 being
a circle tangent to the two lines. Any circle tangent to ω1 , ω2 , and ω3 becomes a
circle tangent to the two lines and to ω3 . There are exactly two such circles, ω4 and
ω5 , one on each side of ω3 , as shown in Fig. 11.4. There are three possible cases:
Case I. O is inside one of ω4 or ω5 .
Case II. O is on one of ω4 and ω5 .
Case III. O is in the exterior of both ω4 and ω5 .
In cases I and III, the inverses of ω4 and ω5 are circles. If one of ω4 or ω5 contains
O inside, then ω1 , ω2 , ω3 are interior tangent to its inverse and are exterior tangent
440 11 Inversions
Fig. 11.4 Circles tangent to ω1

a circle and two parallel lines
ω4 ω3
ω5
ω2
A
A
D
D
B
B
Fig. 11.5 Simplification of the verification of the equality of angles using inversion
to the inverse of the other circle. If neither of the circles ω4 and ω5 contains O
inside, then ω4 and ω5 are exterior tangent to ω1 , ω2 , ω3 . Thus I corresponds to (i)
and III to (iii).
Finally, if O is on one of the circles ω4 or ω3 , then the preimage of that circle is
a line that is a common tangent of ω1 , ω2 , ω3 . Therefore II corresponds to (ii), and
we are done.
Remark A corollary of what we have just proved is that the largest number of circles
in the plane that are pairwise tangent with the tangency points being distinct is 4.
Can you reformulate this problem in space?
163 Invert about a circle centered at A. With the usual convention for the notation,
the circumcircles of ABC, ABD, and ACD are mapped to the lines B C , B D ,
and C D , respectively. And the circumcircle of BCD is mapped to the circumcircle
of B C D . Because inversion preserves angles (Theorem 3.21), we are left with
showing that the angle between the lines B C and B D is equal to the angle
between the line C D and the circumcircle of B C D (Fig. 11.5). But both angles

measure half of the arc C D , and we are done.
Remark This is a recurring theme: use a circular transformation to simplify the

figure as much as possible without altering the hypothesis and the conclusion, and
11 Inversions 441
Fig. 11.6 Proof that ω is

orthogonal to ω1
ω2
ω
then solve the problem for the simpler configuration. In Fig. 11.5 we have placed
side-by-side the noninverted and the inverted figure, so that you can contrast the two
configurations and understand the advantage of this method.
164 Let ω be a circle that is orthogonal to both ω1 and ω2 , and let be the circle of
inversion. Take an inversion centered at one of the intersection points of ω and ω1 .
The images of these two circles are the orthogonal lines ω and ω1 , and the image
ω2 of ω2 is orthogonal to ω , a configuration which is presented in Fig. 11.6. Note
that ω2 cannot pass through the center of inversion, so ω2 is a circle.
By the Symmetry Principle, ω1 and ω2 are reflections of each other over the
image of . By Theorem 3.4 the center of is on ω . Thus is orthogonal to
ω , and consequently is orthogonal to ω.
Remark If instead ω just cuts ω1 and ω2 at equal angles, the conclusion might not
be true.
165 First solution. All three parts of the problem, which comprise a famous
discovery of Apollonius, are easy to check when k = 1, for in this case we recover
the characterization of the perpendicular bisector of the segment AB as the locus of
points P satisfying P A = P B.
The general situation is a consequence the Symmetry Principle (Theorem 3.15)
and the invariance of the cross-ratio (Proposition 3.17). In detail: start with A and B
and some point P such that P A/P B = k. We want to understand the locus of the
points Q characterized by QA/QB = P A/P B, namely, for which
QA P A
: = .1
QB P B
As we have seen in Sect. 3.1.3, the cross-ratio of the points P , Q, A, B is invariant

under Möbius, even circular transformations. Now a Möbius transformation is
determined by specifying the images of three points, so let us choose such a
transformation that maps A, B, P to A , B , P , respectively, such that P is the
442 11 Inversions
midpoint of A B . The locus of the points Q such that QA/QB = P A/P B

is mapped to the locus of points Q such that Q A /Q B = P A /P B = 1.
This locus is the perpendicular bisector of A B (the particular case discussed in
the beginning). Note that A and B are the reflections of each other over this
perpendicular bisector. Then k (A, B) is the preimage of the perpendicular bisector
through the Möbius transformation, so it has to be a circle or a line. But for k = 1
it cannot be a line because it intersects the line AB twice (once inside the segment
AB and once outside). So it is a circle. Moreover, A, B are symmetric with respect
to this circle, by the Symmetry Principle (Theorem 3.15). This proves both (a) and
(b).
For (c), note that by the same considerations, we can start by assuming that is
a line, and then it is the perpendicular bisector of AB. And so it is also 1 (A, B),
and we are done.
Second solution. Here is a proof of (a) using coordinates. Without loss of
generality, we may assume 0 < k < 1 and then choose coordinates such that the
origin is at B. Let a be the complex coordinate of A. Then the locus is given by the
equation
|z − a| = k|z|,
which can be written as
(z − a)(z̄ − ā) = k 2 zz̄.
Transform this into
(1 − k 2 )zz̄ + a z̄ + zā + a ā = 0.
Now set 1 − k 2 = t 2 , and “complete the square” to obtain

a ā 1
tz − t z̄ − = 2 − 1 a ā.
t t t

Divide both sides by t 2 and set r = 1
t4
− 1
t2
a ā to obtain the equation of a circle
a 2

z − 2 = r 2 .
t
Can you verify (b) and (c) using coordinates?
Remark The key fact used in solving this problem is that a Möbius transformation
is determined by the images of three points. This allows us to map our configuration
into one where P is the midpoint of AB without changing the statement. The
11 Inversions 443
problem is now easy, and the property that Möbius transformations map lines or
circles into lines or circles solves the general case.
The circles defined in this problem are called the circles of Apollonius after
Apollonius of Perga, who, according to Pappus and other sources, has introduced
them for the first time. It is important to notice that the Apollonian circles of the
points A and B are orthogonal to any circle passing through A and B (Problem 158).
So we are in the presence of two mutually orthogonal families of circles.
166 Such a Möbius transformation
az + b
φ(z) =
cz + d
must map the real axis to itself. Let us first work in the case d = 0. For t ∈ R,
az + b (at + b)(c̄t + d̄) a c̄t 2 + (a d̄ + bc̄)t + bd̄

φ(t) = = = ∈R
cz + d (ct + d)(c̄t + d̄) |ct + d|2
Hence a c̄t 2 + (a d̄ + bc̄)t + bd̄ ∈ R for all t ∈ R. Setting t = 0 we deduce that

bd̄ ∈ R. But then [a d̄t + (a d̄ + bc̄)]t ∈ R for all t ∈ R, so a c̄t + (a d̄ + bc̄) ∈ R.
Again setting t = 0 we obtain a d̄ + bc̄, and finally a c̄ ∈ R. Since a c̄, bd̄ ∈ R, there
are s, t ∈ R such that a = sc and b = td. But then
a d̄ + bc̄ = scd̄ + t c̄d = scd̄ + tcd̄ ∈ R.
Set w = cd̄. Then sw + t w̄ should be real. This can happen if either w is real or if
s = t. But if s = t, then φ is constant, which is not allowed. So cd̄ ∈ R, meaning
that c = ud, u ∈ R. Normalizing so that d ∈ R, we see that every such Möbius
transformation is of the form
az + b
φ(z) = , a, b, c, d ∈ R.
cz + d
The case where d = 0 can be treated similarly, and the same conclusion is reached.
We thus conclude that these are precisely the Möbius transformations that map the
real axis to itself. But do they all map the upper half-plane to itself?
The upper half-plane is mapped to either the upper half-plane or the lower half-
plane {z | Im z < 0}. It is mapped to the upper half-plane when the imaginary part
of φ(i) is positive. We have
ai + b (ai + b)(−ci + d) ac + bd + (ad − bc)i

φ(i) = = = .
ci + d (ci + d)(−ci + d) |c|2 + |d|2
The imaginary part of this number is positive when ad − bc > 0. Thus the Möbius
transformations that map the upper half-plane to itself are
444 11 Inversions
az + b
z → , a, b, c, d ∈ R, ad − bc > 0.
cz + d
Remark The Möbius transformations that map the upper half-plane to itself are the
real projective special linear group P SL(2, R), that is, the group of real fractional
linear transformations. In the Poincaré model of the hyperbolic plane (i.e., the plane
of the Lobachevskian geometry in which given a point and a line not containing
it, there are infinitely many lines that pass through the point and are parallel to the
line), the elements of P SL(2, R) are the orientation preserving isometries.
167 Certainly, circular transformations have this property.
Conversely, let f be such a transformation. Choose a Möbius transformation φ
such that φ(f (∞)) = ∞. Then
g : C ∪ {∞} → C ∪ {∞}, g =φ◦f
maps lines to lines and circles to circles. By Theorem 2.28 g(z) = az + b or g(z) =
a z̄ + b, for some a, b ∈ C, a = 0. Then f = φ −1 ◦ g is a circular transformation.
Hence the conclusion.
Remark Let us embed the complex plane into the three-dimensional space by z =
x1 +ix2 → (x1 , x2 , 0), and then take the inversion of center N = (0, 0, 1) and radius
1. This inversion maps a point P to P such that P ∈ |NP and NP · NP = 1.
The complex plane together with the point at infinity is mapped onto the sphere
of center (0, 0, 1/2) and radius 1/2 (note that N is the image of ∞). The lines and
circles of C ∪ {∞} are in one-to-one correspondence with the circles on . This is
why we call C ∪ {∞} the Riemann sphere, because it can be actually realized as a
sphere.
The inversion maps the plane to the sphere and hence also the sphere to the plane.
This map from the sphere to the plane is called stereographic projection.
The result we have just obtained allows us to describe all transformations of the
sphere to itself that map circles to circles. These correspond, via the stereographic
projection, to transformations of the plane that map every line or circle to a line or
a circle, that is, to the circular transformations.
168 (a) If the polynomial has the complex zeros x1 , x2 , x3 , then the first operation
yields a polynomial whose zeros are x11 , x12 , x13 , and the second yields a polynomial
whose zeros are x1 − t, x2 − t, x3 − t. Clearly one has to take into account the case
where 0 is a zero of the polynomial, as well as that where the polynomial has degree
less than 3. For that we have to work on the Riemann sphere, thus including the point
at infinity. The successive application of the two operations amounts to changing
the zeros of the polynomial by a Möbius transformation. The first polynomial is
P1 (x) = x 3 + x 2 − 2x = x(x − 1)(x + 2), which has three distinct zeros, while
the second is P2 (x) = x 3 − 3x − 2 = (x + 1)2 (x − 2), which has only two distinct
zeros. And no Möbius transformation maps a set with three elements to a set with
two elements. Thus the answer to the question is negative.
11 Inversions 445
(b) Two zeros of the polynomial x 3 − 3x − 3 are nonreal, while all zeros of P1 (x)
are real. The two operations map polynomials with real zeros into polynomials with
real zeros, so the answer is still negative.
(c) The polynomial x 3 − 3x − 1 has three distinct real zeros. Adding the third
operation to the picture reduces the problem to one about the possibility of mapping
the sets of zeros of the polynomials into each other. The operations (i) and (ii)
correspond to the inversion and translation on the real axis, and we will now show
that by composing them we can√also obtain homothety on the real axis. Let α be a
positive number, and let β = 1/ α. Consider the sequence of transformations
1 1 1 −z
z → z + β → → − =
z+β z+β β βz + β 2
βz + β 2 β2 β2 z
→ − = −β − −
→ → − 2 = −αz.
z z z β
√
Of course by applying this twice (and working with α instead of α), we can write
the map z → αz as the composition of the complex inversion and translations.
Thus, using Theorem 3.13, we conclude that every linear fractional transformation
with real coefficients can be written as a composition of maps of the form z → 1z
and z → z + t, t ∈ R.
Now let x1 , x2 , x3 be the zeros of x 3 − 3x − 1. Consider the linear fractional
transformation that maps them into 2, 0, 1, respectively. Write this transformation
as a composition of the inversions and translations, and apply the corresponding
sequence of operations (i) and (ii) to P1 (x). In the end, you have obtained a
polynomial that is a real multiple of x 3 − 3x − 1. The third operation allows us
to transform these two polynomials into each other, and we are done.
Remark Working with complex numbers and considering for α = 0 a complex
number β such that β 2 = − α1 , by performing the sequence of transformations in
the solution to (c), we can strengthen the statement of Theorem 3.13: every Möbius
transformation can be written as the composition of translations and the complex
inversion.
Source Part (a) of the problem was given at a test at the Mathematical Olympiad
Summer Program, being proposed by Răzvan Gelca.
169 Consider an inversion of center A (Fig. 11.7). The circles ω1 and ω2 become
the lines B M and B N , respectively, where B , M , N are the images of B, M, N.
The lines AM and AN are mapped into themselves; the fact that they form 0◦ degree
angles with the circles means that the opposite sides in AM B N form 0◦ angles, so
this quadrilateral is a parallelogram. Also, because B is the midpoint of AC, C is
the midpoint of AB , so C is the center of the parallelogram. Hence C is on M N ,
that is, M , C , N are collinear. But then Theorem 3.4 implies that A, M, C, N are
concyclic, and we are done.
446 11 Inversions
A A
B C
M N M
B
C
Fig. 11.7 AMCN is cyclic
Fig. 11.8 Proof of Steiner’s A

identity
B D E C
B
C
D
E
170 First solution. Consider an inversion with center A and radius R and let
B , C , D , E be the inverses of B, C, D, E, respectively (Fig. 11.8). Then by
Theorem 3.4, A, B , C , D , E are concyclic. And since D AB = DAB =
EAC = E AC , it follows that B D = E C and B E = D C . The
formula (3.1) turns these equalities into
R2 R2 R2 R2
BD = EC and BE = DC.
AB · AD AE · AC AB · AE AD · AC
Dividing the two equalities yields
AE BD AD EC
· = · ,
AD BE AE DC
which is equivalent to the formula from the statement.
11 Inversions 447
Second solution. The coordinate solution is a variation of the synthetic one and
is based on the rewriting of Steiner’s identity as
DB AB AB EB
: = : .
DC AC AC EC
We denote the coordinates of the points by lowercase letters and let DAB =
CAE = α/2 and EAD = β/2. The problem reduces to proving the equality
of cross-ratios
(d, a, b, c) = (a, e, b, c).
(The complex conjugate is motivated by the fact that DB/DC and EB/EC are
real, while the argument of AB/AC is the negative of the argument of AC/AB.)
Consider an inversion of center A and choose the complex coordinates so that the
image of the line BC is a circle centered at the origin. Then the images of b, d, e, c
are b , d = b eαi , e = b e(α+β)i , c = b e(2α+β)i , respectively, because the arcs

B D and E C have measure α and the arc D C has measure β. Since inversion
transforms a cross-ratio to its complex conjugate, the problem reduces to showing
that
(d , ∞, b , c ) = (∞, e , b , c ),
and this is equivalent to
d − b e − c
= .
d − c e − b
This is further equivalent to
eαi − 1 e−(α+β)i − e−(2α+β)i

= .
eαi − e(2α+β)i e−(α+β)i − 1
The right-hand side is transformed into the left-hand side by multiplying both the
numerator and the denominator by e(2α+β)i , and we are done.
171 Invert with respect to the inscribed circle and let A , B , C , D be the inverses
of A, B, C, D, respectively. By Proposition 3.3, A , B , C , D are the midpoints of
MQ, MN, N P , and P Q, respectively (Fig. 11.9). In the triangles that are formed by
three of the four vertices of the quadrilateral MNP Q, A B , B C , C D , and D A
are midlines, so they are parallel to the corresponding diagonals of this quadrilateral.
Thus A B C D is a parallelogram.
On the other hand, since A, B, C, D are concyclic, their images A , B , C , D
are concyclic as well. Thus A B C D is both a cyclic quadrilateral and a parallelo-
gram; hence it is a rectangle. It follows that A B is perpendicular to B C . As NQ
448 11 Inversions
Fig. 11.9 Inversion about the D

inscribed circle
P
D
Q
C C
I
A A B
N
M
B
A4 A3
A2 A3
A4
A1
A1 A2
Fig. 11.10 Parallelogram obtained by inverting tangent circles
and MP are parallel to these two lines, NQ is perpendicular to MP , and we are

done.
172 First solution. Invert about a circle of center P and radius R. The images
of the four circles form a quadrilateral whose opposite sides are parallel since
their preimages are circles tangent at the center of inversion. This quadrilateral
is therefore a parallelogram (Fig. 11.10). If A1 , A2 , A3 , A4 are the images of
A1 , A2 , A3 , A4 , respectively, then formula (3.1) yields
R2
Aj Aj +1 = · Aj Aj +1 , j = 1, 2, 3, 4.
P Aj · P Aj +1
But because A1 A2 A3 A4 is a parallelogram, A1 A2 = A3 A4 and A2 A3 = A1 A4 , so
R2 R2
· A1 A2 = · A3 A4 ,
P A1 · P A2 P A3 · P A4
11 Inversions 449
Fig. 11.11 Two pairs of

circles intersecting at
concyclic points
R2 R2
· A2 A3 = · A1 A4 .
P A2 · P A3 P A1 · P A4
Multiply the two equalities, then divide the resulting identity by R 4 /P A1 · P A3 , to

obtain
A1 A2 · A2 A3 A3 A4 · A1 A4
= .
P A24 P A22
And this is equivalent to the identity from the statement.

(b) Using the same inversion, we note that A1 , A2 , A3 , A4 are concyclic if and
only if their images are. But the images of these four points are concyclic if the
parallelogram they form is a rectangle (as in Fig. 11.11), namely, if we are in
the presence of two pairs of orthogonal circles. But the orthogonality of circles
translates into the orthogonality of 1 and 2 , and we are done.
Second solution. We will prove that
A1 A2 P A2 P A2 A3 A2
: = : .
A1 A4 P A4 P A4 A3 A4
Switching to complex coordinates, we prove the stronger equality of cross-ratios
(a1 , p, a2 , a4 ) = (p, a3 , a2 , a4 ).
Use a Möbius transformation φ such that φ(p) = ∞ and the invariance of the
cross-ratio (Theorem 3.10) to reduce the problem to
(a1 , ∞, a2 , a4 ) = (∞, a3 , a2 , a4 ),

450 11 Inversions
where ai = φ(ai ), i = 1, 2, 3, 4. This is equivalent to
a2 − a1 a4 − a3

= .
a4 − a1 a2 − a3
We can interpret this is saying that if a2 is the image of a4 under a spiral similarity
of center a1 , scaling factor k, and angle α, then a4 is the image of a2 under a
spiral similarity of center a3 , the same scaling factor k, and the same angle α. And
this is true because a1 , a2 , a3 , a4 form a parallelogram, as Möbius transformations
preserve angles. This proves (a). The proof of (b) is analogous to that shown in
the first solution: Möbius transformations preserve the property of four points to be
concyclic, and the vertices of a parallelogram form a cyclic quadrilateral if and
only if they form a rectangle. But this means that the images of 1 and 2 are
perpendicular, and since Möbius transformations preserve angles, so are the lines
themselves.
Third solution. A clumsier solution is possible if we use a Möbius transformation
that maps A2 to ∞. The new configuration is shown in Fig. 11.12, with the images
of points denoted by dashes. We have seen in the second solution that the identity
from (a) is invariant under Möbius transformations, so it suffices to prove it in this
new setting. And in this setting, the terms that contain A2 cancel out because they
are equal infinities, and so the identity reads
P A4 2 P A4 A3 A4

= 1 ⇐⇒ = .
A1 A4 · A3 A4 A1 A4 P A4
And this is a consequence of the similarity of the triangles A4 P A1 and A4 A3 P ,
as you can guess by examining Fig. 11.12. To prove that the triangles are similar,
consider the spiral similarity of center A4 that maps ω3 to ω2 . The tangent to ω1 at
the image of P forms with ω2 an angle equal to (ω3 , ω2 ), so the image of P must
be A1 . Consequently the image of A3 is P , and so the triangles are mapped into
each other by the spiral similarity.
For (b), notice that the points A1 , A2 , A3 , A4 are concyclic if and only if
A1 , A4 , A3 are collinear, and because of the similarity of the two triangles, this
happens if and only if P A4 A1 = P A4 A3 = 90◦ . But then P A1 and P A3 are
diameters, so ω2 and ω3 are orthogonal, so the two pairs of circles in the original
configuration are orthogonal. Consequently the points are concyclic if and only if
1 and 2 are orthogonal.
Remark The wisdom of this problem is to recognize the conclusion as being
invariant under Möbius transformations, even under circular transformations, and
then to map everything to a convenient configuration. And as the second and third
solutions show, this convenient configuration may not be unique.
Note the similarity of the identity in (a) to Steiner’s identity from Problem 170.
Regarding the proof of (b), the original configuration contains five circles, the
four circles passing through the origin, plus the circle formed by the four pairwise
11 Inversions 451
Fig. 11.12 A clumsy ω3

solution
ω4 A3
A4
ω2
P A1
ω1
Fig. 11.13 Circle orthogonal

to another circle and to its ω B1
ω1
chord
S T
A1 A2
B2
intersection points. A transformation that moves P at infinity significantly simplifies

the problem, transforming four of the five circles into lines. It is always easier
to prove that points obtained as intersections of lines are concyclic because, for
example, angle chasing is simpler.
Source Part (a) is a short listed problem of the 2003 International Mathematical
Olympiad, proposed by Armenia.
173 You can follow the argument on Fig. 11.13. We use the inversion of center M
and radius MS = MT . This inversion maps the circle ω to the line ST , and because
ω1 is orthogonal to both ω and to ST , so is its image. But the image of ω1 is a circle
that is homothetic to ω1 , under a homothety centered at M. These conditions can be
satisfied simultaneously only by ω1 itself, so ω1 is invariant under the inversion. As
the image of the intersection is the intersection of the images, the inversion maps
A1 to B1 and A2 to B2 . So the lines A1 B1 and A2 B2 pass through the center M of
inversion, as desired.
174 First solution. Consider an inversion with center the tangency point O of 1
and 2 . Then 1 and 2 transform into two parallel lines, 1 and 2 , perpendicular
to the line of centers, as can be seen in Fig. 11.14. The circles ωn , n ≥ 0, transform
452 11 Inversions
Fig. 11.14 The shoemaker’s

knife problem
ω1
:1
ω2
ω1
ω0
ω0
:2
:1 :2
into circles ωn , n ≥ 0, that are tangent to 1 and 2 and so that the center of ω0
is on the line of centers of 1 and 2 , and ωn is tangent to 1 , 2 , and ωn−1
. The
relation from the statement clearly holds for ωn , n ≥ 1. Since the circles ωn are

homothetic to ωn with center of homothety O, the ratios are the same for ωn ’s, and
the problem is solved.
Second solution. Let us consider complex coordinates with the origin at O so that
the ray connecting O to the centers of 1 and 2 is the positive semiaxis, and let
us use the inversion from the first solution, with R being the radius of inversion. We
have seen in the analytic proof of Theorem 3.27 that the image through the inversion
of center 0 and radius R of the circle |z − x| = r, x ∈ R is the circle
2
Rx Rr
w − = ,
x2 − r 2 x2 − r 2
If we rotate the center to z0 = xeiθ , then both the circle and its inverse are rotated
by θ , turning the equation of the circle into |z − z0 | = r and the equation of its
image into
2
Rz0 Rr
w − = ,
|z0 |2 − r 2 |z0 |2 − r 2
Now if the lines 1 and 2 are z = c and z = c + 2r, respectively, then the
center of ωn is c + r + 2nri, n ≥ 1. Consequently, the coordinate of the center of
ωn is
11 Inversions 453
R(c + r + 2nri) Rc + r 2nRr

zn = = 2 +i 2 ,
c2 + 2cr + r 2 + 4n2 r 2 − r 2 c + 2cr + 4n2 r 2 c + 2cr + rn2 r 2
while its diameter is

2Rr
dn = .
c2 + 2cr + 4n2 r 2
Taking the ratio of hn = zn to the diameter dn we obtain n, and the problem is
solved.
Remark It is worth mentioning that the centers of ωn , n ≥ 1, lie on the ellipse
whose foci are the centers of 1 and 2 and which passes through O. Indeed, if
O1 , O2 , R1 , R2 are the centers and radii of 1 and 2 , respectively, with 2 inside
1 , and if Cn and ρn are the center and radius of ωn , then O1 Cn = R1 + rn and
O2 Cn = R2 − rn , so
O1 Cn + O2 Cn = R1 + rn + R2 − rn = R1 + R2 = constant.
The name shoemaker’s knife (arbelos in Greek) is motivated by the figure, more
precisely, the region inside 1 but outside of both 2 and ω0 looks like a
shoemaker’s knife.
Source Pappus of Alexandria.
175 The solution can be followed on Fig. 11.15. Invert with respect to the
circumcircle of ABC. The points A, B, C are fixed while, as a consequence of
Proposition 3.3, the points A1 , B1 , C1 are mapped to the intersections of the
tangents to the circumcircle at A, B, C, as shown in the figure. The circumcircles
of AA1 O, BB1 O, and CC1 O are mapped into lines (Theorem 3.4), and these lines
are AA1 , BB1 , CC1 . The problem reduces to showing that AA1 , BB1 , CC1 are
concurrent.
Fig. 11.15 The points A1

A1 , B1 , C1 and their inverses
A1 B
C
O C1
B1
B1 A C1
454 11 Inversions
The equality of the tangents from a point to a circle implies that A1 B = A1 C,
B1 A= B1 C, C1 A = C1 B, hence
A1 B C1 A B1 C

· · = 1,
C1 B B1 A A1 C
which by the reciprocal of Ceva’s Theorem implies that AA1 , BB1 , CC1 are
concurrent, and the problem is solved.
Remark If the triangle ABC is acute, then its circumcircle is the incircle of the
triangle A1 B1 C1 , and AA1 , BB1 , CC1 meet at the Gergonne point of this triangle.
If the triangle ABC is obtuse, then its circumcircle is an excircle of the triangle
A1 B1 C1 and the lines AA1 , BB1 , CC1 meet at an adjoint of the Gergonne point.
Using the same inversion, we can prove that, as a consequence of the fact that
the medians of the triangle ABC intersect, the circumcircles of AA1 O, BB1 O,
and CC1 O have a second common point. Said differently, if ABC is a triangle,
A1 , B1 , C1 are the points of tangency of the incircle with the sides, and I is the
incenter, then the circumcircles of AA1 I , BB1 I , CC1 I have a second common
point.
Source Romanian certification exam for high school teachers, 1984, solution by
Ion D. Ion.
176 Let R and r be the radii of and ω, respectively. Arguing on Fig. 11.16, we
notice that OD · OA = OB · OE√= OC · OF = r · R. It is natural to take
an inversion of center O and radius rR, so that A, B, C are mapped to D, E, F ,
respectively. The points A, B, C are on a circle passing through the center O of
inversion, so their images are collinear. Done.
Remark For the same reason, the converse is also true: if and ω are concentric,
with ω inside , and with common center O, and if the chord EF of is tangent
Fig. 11.16 The circle (ABC) E

is mapped into the line DEF
D
O A
F
11 Inversions 455
Fig. 11.17 The orthogonality A

of AO and EF
E
F
B C
to ω, then the intersections of ω with the radii OE and OF are on the circle of
diameter OA.
177 Let ABC be the triangle, and let BE and CF be the altitudes from the vertices
B and C, respectively (Fig. 11.17). If we showed that AO is orthogonal to EF ,
then, by symmetry, the corresponding orthogonalities would hold for the other two
vertices and the problem would be solved. Because the angles formed by BE and
AC, and by CF and AB are right, the quadrilateral BF EC is cyclic, and so by
writing the power of the point A with respect to its circumcircle, we obtain
AE · AB = AF · AC.
√
This motivates us to consider an inversion of center A and radius AF · AB =
√
AE · AC. This inversion maps the line EF to the circumcircle of ABC, and
the line that connects the center of inversion A to the circumcenter O of ABC
is orthogonal to EF , as Theorem 3.4 states. This shows that AO is orthogonal to
EF , and we are done.
Remark The fact that the perpendiculars from the vertices onto the sides of
the orthic triangle intersect can be derived from the Theorem of Orthological
Triangles, which states that if the perpendiculars from the vertices of a triangle
onto the respective sides of a second triangle meet, then the same is true for the
perpendiculars from the vertices of the second triangle onto the sides of the first.
Applying this theorem to ABC and DEF solves the problem.
The same inversion can be used to show that the altitudes of a triangle intersect.
Let H be the intersection of BE and CF . The inversion maps the side BC into the
circumcircle of AEH F , and so the diameter AH of this circle is perpendicular to
BC. In other words, the third altitude also passes through H .
456 11 Inversions
A
C
D
C P D
P B
Fig. 11.18 Proof that C P = P D
178 Denote by P the second intersection point of AP with (see Fig. 11.18).
By the construction of the symmedian (Theorem 1.22), ACP D is harmonic, and
so the complex coordinates of its vertices satisfy
a − c p − c
: = −1.
a − d p − d
Take the inversion of center A and radius AB. Then the circle is mapped to
the line t, and A, C, P , D are mapped to ∞, C , P , D . As inversion preserves
real valued cross-ratios (Theorem 3.18), the points ∞, C , D , P form a harmonic
division on t, which means that P is the midpoint of C D .
Remark It is important to know that the inversion from A that maps to t is called
the stereographic projection of the circle onto the line.
179 We invert about a circle centered at C and use our convention to mark the
image of a geometric object with a dash. The images k and k3 of k and k3 are
lines, in fact parallel lines because inversion preserves angles. Also, k1 and k2 are
circles tangent to both k and k3 and to each other, so they must be equal circles; the
tangency points are A , B , X , Y , as shown in Fig. 11.19. The distance between the
centers of k1 and k2 is equal to two radii, which is the same as the distance between
k and k3 . Hence A B Y X is a square.
On the other hand, because A ∈ CA and B ∈ CB,
A CB = ACB = 180◦ − ASB/2 = 135◦ .
Consequently
A CB + A X B = 135◦ + 45◦ = 180◦ ,
showing that the quadrilateral A CB X is cyclic. We deduce that C is on the

circumcircle of the square A B Y X , and hence
11 Inversions 457
B
C k
k2 B
Y k3
k2
X C
A
S A
k3
Y
k1
k
X k1
Fig. 11.19 The circles k, k1 , k2 , k3 and their images through the inversion
Fig. 11.20 Inverted A

configuration for the proof
that A, B, M are collinear
S B T
XCY = X CY = X A Y = 45◦ ,
as desired.
Remark It is easier to chase angles with lines than with circles, and the inversion
considerably simplifies the figure, from four circles to two circles and two lines.
Source Czech and Slovak Mathematical Olympiad, 1995, solution by Leandro
Maia.
180 (a) First solution. The configuration is shown in Fig. 11.20. Consider an
inversion χ1 of center M and radius MS = MT . Then the circle ω and the line
ST are inverses of each other, so the circle ω1 is mapped to a circle χ1 (ω1 ) that is
tangent to both ST and ω. But χ1 (ω1 ) must be homothetic to ω1 under a homothety
of center M, and this is only possible if χ1 (ω1 ) = ω1 . It follows that the point of
tangency B of ω1 and ST is mapped to the point of tangency A of ω1 and ω. And
the center of inversion M is collinear with B and A = χ1 (B), as desired.
Second solution. Consider the circular transformation φ that maps ω into itself
obtained as the composition of an inversion of center B and the reflection over B
(prove that it exists!). This transformation maps ω1 into a line φ(ω1 ) that is tangent
458 11 Inversions
Fig. 11.21 Inverted S A

configuration for the proof
that A, B, M are collinear
M T
ω
B ω1
Fig. 11.22 The case where A

the circles are tangent
B D C

to the arc SMT and is parallel to ST . The point of tangency of φ(ω1 ) and the arc is
therefore M. Since φ preserves tangencies, φ(M) = A, and therefore B, A, φ(A) =
M are collinear.
Third solution. Consider an inversion χ2 of center A and arbitrary radius. Then
ω = χ2 (ω) and ω1 = χ2 (ω1 ) are parallel lines, and χ2 (ST ) is a circle that passes
through the center A of the inversion, crosses ω at S = χ2 (S), T = χ2 (T ),

and is tangent to ω1 at B = χ2 (B), as shown in Fig. 11.21. Then S B =T B so
S AB = B AT . It follows that SAB = BAT , so SB is the angle bisector
of SAT . But AM is also the angle bisector of SAT , so the two lines coincide,
proving the collinearity of A, B, M.
(b) Consider again the inversion of center M and radius MS = MT . As seen
in the first solution to (a), ω1 and ω2 are invariant under this inversion, hence so
is ω1 ∩ ω2 . It cannot happen that both P and Q are invariant under this inversion,
since MP = MQ, so they cannot both be equal to MS. The only possibility is that
P is mapped to Q, and so P , Q, and the center M of the inversion are collinear, as
desired.
(c) With changed notation, if the circles are tangent, then we are in the particular
case of (b) where P = Q (Fig. 11.22), and then the common tangent AD to the two

circles passes through the midpoint M of the arc BC. Hence
1 1
BAD = BAM = BM= CM= CAM = CAD.
2 2
11 Inversions 459
Fig. 11.23 AB is the 1

common tangent of ω and ω1
A ω1
ω
ω2
2
B = B1
Conversely, if BAD = DAC, then the intersection M of AD with the

circumcircle is the midpoint of the arc BC that does not contain A. As seen in
(a), the circles ω1 and ω2 are fixed by the inversion with center M and radius MB.
The points where they are tangent to AD are also fixed by the inversion. But the
inversion has only one fixed point on the ray |OA, so these points coincide. Done.
Remark The power of M with respect to any circle ω1 interior tangent to ω, tangent
to ST , and on the other side of ST than M is equal to MS 2 ; thus M is on the radical
axis of any two such circles and is the radical center of any three such circles.
Homothety is the most natural approach to (a) consider the homothety h of center
A that maps ω1 to ω. This homothety maps the tangent to ω1 at B to the tangent to
ω at h(B). The two tangents must therefore be parallel, so the tangent at h(B) is
parallel to the chord ST . This implies that h(B) = M. And A, B, h(B) = M are
collinear, as desired. But the proof by inversion allows progress toward (b) and (c).
Source Part (a) is a variation of Archimedes’ Lemma. Part (c) was given at a
Romanian Team Selection Test for the International Mathematical Olympiad in
1997.
181 The configuration is shown in Fig. 11.23.
Let B1 be the intersection of the common interior tangent of ω and ω1 with
2 . Consider an inversion of center B1 and radius B1 A. Because the tangent is
orthogonal to the radius at the point of contact, the inversion circle is orthogonal
to both ω and ω1 . So ω and ω1 are invariant under the inversion (Proposition 3.25).
The line 1 is mapped to a circle that passes through B1 and is tangent to both ω
and ω1 . Moreover, because 1 and 2 are parallel, and 2 passes through the center
of inversion, the image of 1 is also tangent to 2 (as 1 and 2 form a 0◦ angle).
Thus the image of 1 is tangent to ω, ω1 , and 2 . But there is only one such circle,
namely, ω2 . But then ω2 must pass through the center B1 of inversion, because it is
the image of a line. And this can only happen if B1 = B.
Remark This is yet another example of a “position argument.”
182 Rephrasing the question, we will prove that the line √ CO passes √
through M.
To this end, consider an inversion of center C and radius CA · CD = CB · CF ,
460 11 Inversions
Fig. 11.24 The inversion G

used for proving that
O, C, M are collinear
B F
O
C
A A M
E D
and let A , B , O be the images of A, B, O, respectively (see Fig. 11.24). Note first
that CA = CD and CB = CF , because CA · CA = CA · CD and CB · CB =
CB · CF .
By Theorem 3.4, CO is perpendicular to A B , as A B is the image of the
circle under the inversion. We are now to show that CO passes through M,
which is equivalent to showing that CM is perpendicular to A B . This is a direct
consequence of Problem 15 from the first chapter, applied to the triangle CA B
with squares constructed on the sides CA and CB , since CM passes through the
fourth vertex of the parallelogram determined by C, D, F .
Remark Like in the previous problem, the radius of inversion is determined by the
length of a geometric object present in the statement, in this case the square root of
the equal areas. With this choice, C becomes the natural center of inversion.
183 We invert about a circle of center E and radius EB = EC. The image of
N through this inversion is the center O of the circle (by Proposition 3.7), and the
inverses of K and M are the points K and M which are the other intersection
points of EK and EM with the circle, respectively (Fig. 11.25). It follows that line
through K, M, N is is mapped by the inversion to the circle through E, K , M , O,
so these four points are concyclic. Therefore
OM E = OK K = OKK .
But then
M MO = MM O = OKE.
11 Inversions 461
Fig. 11.25 The inversion of M

center E
C
M
E O
N
K
B A
K
It follows that EMOK is cyclic. From here we deduce that EMK = EOK =
90◦ , and the conclusion follows.
Remark We could have also concluded the solution by noting that EM O =

EKO implies that M is the reflection of K over EO, because it is on the arc traced
by the vertex of an angle of measure EMO and this arc intersects the semicircle
that lies above the line EO in only one point, the reflection of M.
Source Tournament of the Towns, 2002, solution by Leandro Maia.
184 The point H is the incenter of the orthic triangle A1 B1 C1 , and A2 , B2 , C2
are the points of tangency of the incircle of this triangle to its sides. Thus (a) is a
consequence of the following result:
Lemma If ABC is a triangle that is not isosceles, I is its incenter, and D, E, F are
the tangency points of the incircle to the sides BC, AC, and AB, respectively, then
the circumcircles of AI D, BI E, and CI F have a second intersection point besides
I.
Proof Consider the inversion with respect to the incircle of the triangle ABC
(Fig. 11.26). Because AI , BI , and CI are angle bisectors, and hence medians,
in the isosceles triangles EAF , F BD, and DCE, respectively, A , B , C are the
midpoints of EF , F D, and DE (by Proposition 3.3). It follows that the lines DA ,
EB , and F C , which are the images of the three circles, are medians in the triangle
DEF . So the second point of intersection of the three circles is the inverse of the
centroid of DEF .

For (b) consider again the inversion with respect to the incircle of the orthic
triangle, in order to reduce the problem to showing that A A2 , B B2 , and C C2
are concurrent. As seen above, A1 , B1 , C1 are the midpoints of the segments
B2 C2 , C2 A2 , A2 B2 , and A , B , C lie on the lines B1 C1 , C1 A1 , respectively A1 B1
(because H B1 AC1 , H C1 BA1 , H A1 CB1 are cyclic). In fact A , B , C are the feet
of the altitudes of the triangle A1 B1 C1 , and in particular H A , H B , and H C are
concurrent. We are done with the solution once we prove the following result:
462 11 Inversions
Fig. 11.26 Proof that the A

circumcircles of AI D, BI E,
CI F have a second
intersection point
A E
F
I
C
B
C
B D
Fig. 11.27 Configuration A

with three concurrent triples
of lines
P X N
Z
Y
C
B M
Lemma If ABC is a triangle, and if M is on BC, N is on AC, and P is on AB

such that AM, BN, and CP are concurrent, and if moreover X is on NP , Y is on
MP , and Z is on MN such that MX, NY , and P Z are concurrent, then AX, BY ,
and P Z are also concurrent.
Proof Such a configuration is shown in Fig. 11.27.
Note that if P QR is a triangle and S is on QR (Fig. 11.28), then by applying the
Law of Sines in the triangles P QS and P SR, we obtain
PR sin P SR PQ sin P SQ
= and = .
SR
sin SP R QS sin QP S
Using the fact that supplementary angles have equal sines, we obtain
sin QP S PR PQ QS P R
= : = · .

sin SP R SR QS SR P Q
Applying this in the triangles ANP , BP M, and CMN , for the points X, Y, Z on
the respective sides, and multiplying, we obtain
11 Inversions 463
Fig. 11.28 Proof of P

trigonometric formula for a
cevian in a triangle
Q S R
sin P AX sin NCY sin MBZ P X AN NY CM MZ BP

· · = · · · · ·
sin XAN sin Y CM sin ZBP XN AP MY CN P Z BM

P X NY MZ AN CM BP
= · · · · · .
XN MY P Z AP CN BM
Each of the products in the parentheses is equal to 1, by Ceva’s Theorem in the

triangles ABC and MNP . Thus
sin P AX sin NCY sin MBZ

· · = 1.
sin XAN sin Y CM sin ZBP
Thus AX, BY, CZ are concurrent by the trigonometric form of Ceva’s Theorem.

Remark Examining the proof of the second lemma, we realize that the following
stronger statement is true:
Let ABC be a triangle, let M, N, and P be on BC, AC, and AB, respectively, and
let X, Y, Z be on NP , MP , and MN, respectively. Then the concurrency of any two
of the triples of lines (AM, BN, CP ), (MX, N Y, P Z), and (AX, BY, P Z) implies
the concurrency of the third.
Source Gazeta Matematică (Mathematics Gazette, Bucharest).
185 First solution. A possible configuration is shown in Fig. 11.29. Let us consider
the inversion whose center is the tangency point of C3 and and has the property
that maps C into itself. The circle C3 is mapped to a line C3 that is parallel to ,
while the circles C1 and C2 are mapped to the circles C1 and C2 that maintain the
same tangency pattern, as illustrated in Fig. 11.30. Of course, C1 and C2 have equal
radii.
Let O, O1 , O2 be the centers of C, C1 , C2 , respectively, and let T be the point of
contact of C1 and C2 ; let also x be the radius of C1 (and also of C2 ). The Pythagorean
Theorem in the right triangle OT O1 yields
x 2 + (x − 1)2 = (x + 1)2 ,
464 11 Inversions
Fig. 11.29 The circles

C, C1 , C2 , C3 and the line
C1
C2
C3
Fig. 11.30 The inverted C

circles C, C1 , C2 , C3 and the C3
line O
C1 x
C2
O1 T O2
that is, x 2 − 4x = 0. It follows that x = 4, and hence the distance from O to is

2x − 1 = 7.
Second solution. The proof of Theorem 3.27 works in the limiting case where
one of the circles is a line, so there is an inversion χ that maps C and into two
concentric circles. Now the circles C1 , C2 , and C3 have equal radii and lie between
the two concentric circles, being tangent to them. The center of the inversion χ lies
on the image of , so there is a rotation that maps C3 into a circle that passes through
the center of χ . Compose χ with this rotation to map C and back to their original
position, while C3 is mapped to a line C3 parallel to (the two lines are parallel
because they form a 0◦ angle), and C2 and C3 have become two equal circles. And
we are again in the configuration from Fig. 11.30, and the solution continues with
the same computation.
Remark The reader can recognize that we are in a limiting case of Steiner’s Porism
(Theorem 3.28), with the two circles being C and , and the chain of circles being C1 ,
C2 , and C3 . This explains why the distance from the center of C to only depends on
the radius of C and not on the radii of C1 , C2 , C3 . Both solutions reduce the problem
to a particular chain of three circles in which one of the three circles is a line, where
the computation is easy.
Finland, second solution from D. Djukić, V. Janković, I. Matić, N. Petrović, The
IMO Compendium, Springer, 2006.
11 Inversions 465
Fig. 11.31 The configuration C

for Problem 186 before and
after inversion
C
D
D
F
B B
E
186 Examining Fig. 11.31 we realize that the problem reduces to showing that
F CD = F EB. Consider an inversion of center F and arbitrary power, and use
the standard convention that the image of a point is denoted by the same letter,
but with a dash. Theorem 3.2 allows us to reduce the problem to showing that
F DC = F B E.
Looking again at the figure, we notice that the configuration is simplified

considerably after inversion since the circumcircles of BCF and DEF have become
the lines B C and D E , which form the same angle as the original circles,
and hence are parallel. Similarly B E and C D are parallel, and so B C D E
is a parallelogram. From here we obtain E D C = E B C , which can be
decomposed as
F DE + F DC = F B E + F B C .
On the other hand BED = ACB yields F ED + F EB = F CD + F CB,

and again from Theorem 3.2, we obtain
F DE + F B E = F DC + F B C .
Subtracting the two relations, we obtain the desired F D C = F B E .

Remark This solution relies on the behavior under inversion of both the angle
between two curves and the angle determined by two points and the center of
inversion.
466 11 Inversions
Fig. 11.32 Inversion about A

the excircle
B A1 C
B C B1
A
C1
Ia

Olympiad, 2008.
187 We consider the inversion χ about the excircle. By Proposition 3.3 χ (A) =
A , χ (B) = B , and χ (C) = C . Thus we are in the situation from Fig. 11.32.
Since χ maps the circumcircle of ABC to the circumcircle of A B C , the line a ,
which passes through the center of inversion and the circumcenter of A B C , also
passes through the circumcenter of ABC. From here we infer that a , b , and c
pass through the circumcenter of ABC, and the problem is solved.
Source Petru Braica.

188 We consider an inversion of center A and examine the configuration before
and after inversion on Fig. 11.33. With the convention of denoting the image of a
point by the same letter but with a dash, the image of the circumcircle of ABC is
the line B C , and the image of the line BC is the circumcircle of AB C .
Because AE is symmedian, ABEC is a harmonic quadrilateral (Proposi-
tion 1.23), and so the cross-ratio of the complex coordinates of A, B, E, C is
−1. As inversion preserves real valued cross-ratios, the cross-ratio of the complex
coordinates of the images of these points is also −1. But the images of these points
are ∞, B , E , C , which are collinear and therefore form a harmonic division. This
means that E is the midpoint of B C .
Because inversion preserves the angle between two curves, and because AD
and BC are orthogonal, AD and the circumcircle of AB C are orthogonal, too,
showing that A and D are diametrically opposite in this circumcircle. Also, because
AF and the circumcircle of ABC are orthogonal, so are AF and B C , with F

on the circumcircle of AB C . But then AF D =AD /2 = 90◦ , so AF is
orthogonal to both B C and D F . This implies that D F is parallel to B C , and
11 Inversions 467
A
A
C
E
D F F
B C
E B
D
Fig. 11.33 The circumcircles of ABC and DEF are tangent

collinearity of M, N, P , Q
Q
P
N
M X
B B C
so B D F C is an isosceles trapezoid. And since E is the midpoint of B C , the

circumcircle of D E F is tangent to B C . Because inversion preserves tangencies,
the circumcircle of DEF is tangent to the circumcircle of ABC, as desired.
189 We argue on Fig. 11.34. Let B be the point that is diametrically opposite to
B in C1 . Then BN B = 90◦ , so B N is parallel to AC. Hence NB B = ACB.
Now let N be the second intersection point of intersection of AN and C1 . Then
because BN B N is cyclic it follows that
BN A = BN N = BB N = ACB,
which implies that A, B, N , C lie on a circle, so N is on the circumcircle of ABC.

Similarly, if P is the second intersection point of AP and C2 , then P is on the
circumcircle of ABC as well.
468 11 Inversions
The altitude from A is the radical axis of the circles C1 and C2 . Writing the fact
that the power of the point with respect of the two circles is the same yields
AM · AB = AN · AN = AP · AP = AQ · AC.
√
Let χ be the inversion of center A and radius AM · AB. Then χ transforms the
circumcircle ω of ABC into a line, and since χ (B) = M, χ (N ) = N, χ (P ) = P ,
χ (C) = Q, the points M, N, P , Q belong to χ (ω). Hence these points are collinear.
Remark You can see here how power-of-a-point can lead to inversion. The case
where X is the foot of the altitude is a well-known problem.
Olympiad, 2014, proposed by Petru Braica.
190 There are three circles passing through A: the circumcircles of ABL, ACK,
and AKL. We can turn them into lines by taking an inversion of center A. With the
usual convention, we denote the image of a point X by X . The original figure and
the inverted one are presented side-by-side in Fig. 11.35.
The points B and C are the midpoints of the segments AK and AL ,
respectively. So P is the intersection of the medians K C and L B in the triangle
AK L . Consequently it is the centroid of this triangle. On the other hand, Q is the
intersection of the median AP with the side K L , thus AP = 23 AQ . It follows
that AP = 32 AQ, the desired equality.
Remark The idea of the solution is that, for an inversion of center O that maps
X and Y to X and Y , respectively, if OX/OY = m/n then OX /OY = n/m.
Inversion “inverts” the ratio!
Source Baltic Way, 2006, communicated by Titu Andreescu.
√
191 Let φA be the bc inversion corresponding to the vertex A in the triangle
ABC (see Fig. 11.36). Because φA maps into a line that is parallel to BC and
A A
C
K L B P
Q
L
B C Q
P K
Fig. 11.35 2AP = 3AQ is equivalent to 3AP = 2AQ

11 Inversions 469
ω1
E D
ω2
B C
Fig. 11.36 Proof of tangency of circumcircles of ABC and ADE
the circumcircle of ABC and the line BC into each other, it maps ω1 and ω2 into
two circles, ω1 and ω2 , that are exterior tangent to , BC, and the circumcircle of
ABC.
The circles ω1 and ω2 must be equal, since they are tangent to two parallel lines.
Set D = φA (D) ∈ ω1 and E = φA (E) in ω2 . Then D and E are tangency points
of ω1 and ω2 with the circumcircle of ABC. Because the configuration formed by
the three circles is symmetric with respect to the diameter of this latter circle that
is perpendicular to BC, D E is parallel to BC. But then the images of these lines
through φA are tangent circles, that is, the circumcircles of ADE and ABC are
tangent.
Remark This is a “position argument” that uses a Möbius transformation.
192 First solution. Consider an inversion of center B and radius 1. Let A , C , and
D be the images through the inversion of the points A, C, and D, respectively,
as shown in Fig. 11.37. Using Theorem 3.2 we obtain BD A = BAD and
BD C = BCD, so
C D A = BAD+ BCD=360◦ − ABC− ADC = 360◦ − 2 · 135◦ = 90◦ .
Applying formula (3.1) we obtain
CD DA AC
C D = , D A = , A C = .
BD · BC AB · BD AB · BC
The Pythagorean Theorem in the right triangle A C D yields

2 2 2
AC CD DA
= + ,
AB · BC BD · BC AB · BD
hence
470 11 Inversions
A C
C
A
D
Fig. 11.37 The image of triangle ACD through the inversion
(AC · BD)2 = (AB · CD)2 + (BC · DA)2 .
Combining this with the relation in the statement, we obtain
(AB · CD)2 + (BC · DA)2 = 2(AB · CD) · (BC · DA).
Thus we have the equality case in the AM-GM inequality, which implies that AB ·
CD = BC · DA. In other words AB/DA = BC/CD. Since ABC = ADC =
135◦ , it follows that the triangles ABC and ADC are similar. These two triangles
share the side AC, so they are congruent. So the figure is a kite, and consequently
AC and BD are orthogonal.
Second solution. Write the equality from the statement as an equality of cross-
ratios of segments
DB AB BC AC
: =2 : .
DC AC BD AD
Take an inversion of center A, and let B , C , D be the images of the vertices

B, C, D, respectively. Then AC B = AC D = 135◦ (Theorem 3.2), so
B C D = 360◦ − 2 × 135◦ = 90◦ , i.e., the triangle B C D is right. Because
inversion preserves the cross-ratio of segments, and because A is mapped to ∞,
DB B C
= 2 .
DC B D
Let M be the midpoint of B D . Using trigonometric functions in the right triangle

B C D , we translate the above equality into
11 Inversions 471
DM
= sin B D C .
DC
Then, using the Law of Sines in the isosceles triangle MC D , we transform this
equality further into
sin B D C
= sin B D C .
sin C MD
Hence sin C MD = 1. It follows that C MD = 90◦ , showing that the right

triangle C B D is isosceles. We thus have AC B = AC D and C B = C D ,
which implies that D is the reflection of B over the angle bisector of AC. And so
D is the reflection of B over the angle bisector of AC. From here we deduce that
ABCD is a kite, so its diagonals are orthogonal.
Remark The emergence of the right triangle becomes transparent when using
complex coordinates. The angle condition from the statement is translated into
c−b 3π i a−d 3π i
= re 4 and = se 4 , r, s ∈ R.
a−b c−d
Thinking in terms of cross-ratios, we can write
c−b c−d
(a, c, b, d) = : = irs ∈ iR.
a−b a−d
If we consider a circular transformation that maps b to ∞, then (using dashes for

the images) we have
c − d
= ±irs ∈ iR,
a − d
showing that a , c , d form a right triangle (with the right angle at d ).

Source United States of America Team Selection Test for the International Mathe-
matical Olympiad, proposed by Răzvan Gelca.
193 Let AA , BB , CC be the altitudes; let also A , B , C be the projections of
the orthocenter H onto the lines MA, MB, MC, respectively (Fig. 11.38). Note
that A , B , C lie on the circle of diameter MH because MA H = MB H =
MC H = 90◦ .
Let A1 be the intersection of H A and BC, B1 the intersection of H B and

AC, and C1 the intersection of H C and AB. The quadrilateral AA A A1 is cyclic,
because AA A1 = AA A1 = 90◦ . Hence H A1 · H A = H A · H A . But H A ·
H A = H B · H B = H C · H C , and so H A1 · H A = H B1 · H B = H C1 · H C .
It follows that the points A1 , B1 , C1 are the images of A , B , C with respect to
the composition of an inversion of center H with a reflection over H , also known
472 11 Inversions
Fig. 11.38 The construction B1

of the points A , B , C
B
C
A
C
H
M
B
A1 B A C
C1
as an inversion of negative ratio (we have seen this inversion before). Consequently,
they lie on the line that is the inverse of the circle of diameter MH , and this line is
perpendicular to MH .
Remark Here we have used the circular transformation that maps the nine-point
circle to the circumcircle, obtained as the composition of an inversion of center H
and the reflection over H that was discussed in Sect. 3.1.7.
Source Gh. Ţiţeica, Probleme de Geometrie (Problems in Geometry), Ed. Tehnică,
Bucharest, 1929.
194 First solution. Consider an inversion of center A and arbitrary radius. Because
inversion preserves angles, the circumcircles of ABD and ACD are mapped to
the lines B D and C D that are parallel to the lines AC and AB , respectively.
So AB D C is a parallelogram (Fig. 11.39 shows both the original figure and the
inverted one).
As B is the midpoint of AE, E is the midpoint of AB . It follows that the
triangles AF E and B D E are congruent. Hence AF = B D = AC , that is,
A is the midpoint of C F . But then A is the midpoint of CF , and the problem is
solved. √
Second solution. Let φA be the bc inversion corresponding to the vertex A
in the triangle ABC. The image of the original configuration through this Möbius
transformation is shown in Fig. 11.40. We obtain that φA (E) = E is the midpoint
of AC, because C = φA (B) and B is the midpoint of AE. Also, tangencies
yield parallelism, so ABD C is a parallelogram, where D = φA (D). And as the
11 Inversions 473
F
F
A
A
E
B C
C
D
E
D
Fig. 11.39 The triangle ABC with the circumcircles of ABD, ACD, and ADE and their images
through the inversion
Fig. 11.40 The image of the F

circumcircles of ABD,
ACD, and ADE through φA
B C
circumcircle of ADE is mapped to the line D E , F = φA (F ) ∈ D E . Now since

AF is parallel to D C and E is the midpoint of AD, AF CD is a parallelogram,
showing that AF = D C = AB. Hence AF = AC, as desired.
Remark Note that AD is a median in the triangle ABC, and hence AD is a sym-
median in this triangle, which gives a solution based on inversion to Problem 132.
Source Turkish Mathematical Olympiad, 1998.
195 Denote the circle that is tangent to AD at A and has the center on B1 C1 by ω
and its center by O1 . The configuration is shown in Fig. 11.41.
Writing the power of A with respect to the circumcircle of OBC, we obtain
474 11 Inversions
O1
A
C1
O X
B C
D
B1
ω
Fig. 11.41 The dotted circles are orthogonal
AB · AB1 = AC · AC1 .
√
It is therefore natural to consider the inversion of center A and radius AB · AB1 ,
as this inversion maps the circumcircle of ABC to the line B1 C1 . By Theorem 3.4,
AO is perpendicular to B1 C1 . Denote by X the intersection of the lines AO and
B1 C1 .
Now we consider a different inversion; this is χ of center D and radius DA,
which transforms the circle of diameter OD into a line. But note that the radii of the
circle of inversion and of ω are orthogonal at A, so the two circles are orthogonal.
Proposition 3.25 shows that χ (ω) = ω. And the problem is simplified by using this
inversion, because now we have to show that a line is orthogonal to a circle, which
is a simpler task.
All we have to do is identify the line that is the image through χ of the circle
of diameter OD. The Leg Theorem in the right triangle AO1 D in which AX is an
altitude gives DA2 = DX · DO1 and hence χ (X) = O1 . But since AXD = 90◦ ,
X is on the circle of diameter OD, so the image of this circle passes through the
image of X, which is O1 . And any line that passes through the center O1 of ω is
orthogonal to ω. The problem is solved.
Remark The idea of the solution is that it is easier to compute the angle formed by
a line and a circle than that the angle formed by two circles.
Source Mathematical Olympiad Summer Program, 1997, solution communicated
by Leandro Maia.
11 Inversions 475
T M
X C
D
Ω B
ω
Fig. 11.42 Reformulation of the problem about proving that AX is orthogonal to XC
196 We will rephrase the problem and then solve it by inversion. Tracing the
approach on Fig. 11.42, we start by observing that triangles MXA and MAB share
the angle AMB, which together with
MXA = DAC = DAB = MAB
implies that the two triangles are similar. This similarity implies that MAX =
ABX, which then shows that the line AD is tangent to the circumcircle of ABX.
The similarity also gives
AM BM
= ,
XM AM
so
MD 2 = AM 2 = MB · MX.
This shows that MD 2 is equal to the power of M with respect to the circumcircle of
BDX, which implies that AD is also tangent to this circle. We denote the line AD
by , the circumcircle of ABX by , and the circumcircle of BDX by ω.
Next, let S and T be the other intersections of the lines AC and CX with the
circle , respectively. Because
1
BAD = DAC = AS= ABS,
2
476 11 Inversions
it follows that the lines SB and AD are parallel. Hence A is the midpoint of the arc

BAS. The conclusion will then follow if we show that T is diametrically opposite
to A in the circle . Thus we are left with solving the following problem:
Two circles and ω intersect at B and X and are tangent to the line at A and
D, respectively. Let S and T be points on such that BS is parallel to and AT is
a diameter. Then the lines BD, SA, and T X intersect (at some point C).
To solve this problem, perform an inversion of center B. With the usual notation,
the line becomes the circle = A BD , the circles and ω become the lines
X A and X D tangent at A and D to the circle , S is the intersection of the line
X A with the tangent at B to the circle , and T is the point on the line X A for
which A BT is a right angle (see Fig. 11.43). We have to show that the line BD
and the circumcircles of BS A and BT X have a second common point.
To prove this, let C be the intersection of BD and the perpendicular bisector
of A D (which passes through both X and the center of the circle ). Using the
isosceles triangle C A D and the fact that S A and S B are tangents to the circle
, we obtain
A S B = 180◦ − 2 A BS = 180◦ − 2 A D B = 180◦ − 2 A D C = A C D .
This shows that C is on the circumcircle of BA S . On the other hand
BT X = 90◦ − BA S = A S B/2 = A C D /2 = D C X .
B S
D ω
X
Fig. 11.43 Inverted figure for the problem about proving that AX is orthogonal to XC
11 Inversions 477
This shows that C is also on the circumcircle of BT X . So C lies on BD and also

on the circumcircles of BA S and BT X . The problem is solved.
Remark The official solution (which we sketch here) does not use geometric
transformations. Once the similarity of the triangles MXA and MAB and its
consequences are established, one considers the foot H of the perpendicular from
A to BC. There are two analogous cases, that when H is between B and D and
that when it is not. Working in the assumption that H is between B and D, in the
right triangle AH D, AM = H M, and so XM/H M = H M/BM. It follows that
the triangles XMH and H MB are similar. Hence H XM = BH M, and so
BH M = 180◦ − MH D = 180◦ − MDH = B − A/2.
Therefore
AXH = MXH + AXM = (B + A/2) + A/2 = B + A.
So AXH + ACH = B + A + C = 180◦ , showing that AXH C is cyclic. From

here we deduce that AXC = AH C = 90◦ , as desired.
Source Italian Mathematical Olympiad, 2019.
197 Consider the line z = t + i, t ∈ R. The complex inversion z → 1z maps this
line to the circle |z + i/2| = 1/2. Consider now two rational parameters s = m/n
and t = p/q, m, n, p, q ∈ Z. The distance between the inverses of s + i and t + i is

1 1 |pn − qm|

t + i − s + i = √ 2 .
m + n2 p2 + q 2
So in order to find the desired points, it suffices to find a family of Pythagorean

triples (ak , bk , ck ), k = 1, 2, . . . , 1975 (with ak2 + bk2 = ck2 ) such that ak /bk =
aj /bj for j = k. For this let pk be the kth prime, and set ak = p2k 2 − p2
2k−1 ,
bk = 2p2k p2k−1 and ck = p2k + p2k−1 , k = 1, 2, . . . , 1975. The inverses of the
2 2
points zk = ak /bk + i satisfy the desired condition on the circle |z + i/2| = 1/2.
Now map these points by z → 2z + i to the unit circle.
Remark If we multiply by i the Möbius transformation from the line to the unit
circle that was used in the solution, and thus work with z → 2iz − 1, then the points
t + i, t ∈ R are mapped to
1 − t2 2t
+i ,
1 + t2 1 + t2
2
which yields the standard parametrization of the circle t → 1−t 1+t 2 , 2t
1+t 2 . This
parametrization arises from the writing of cos x and sin x in terms of tan x2 and
shows that the circle is what is called a rational algebraic curve.
478 11 Inversions
A
A
X=Y
D E
P C
B C
B P
Fig. 11.44 The configuration of the triangle ABC with the point P inside and its inverted version
Source International Mathematical Olympiad, 1975, proposed by the Soviet Union.

198 Let X and Y be the points where the angle bisectors BD and CE of ABP
and ACP intersect AP , respectively. We want to show that X = Y , which by the
Bisector Theorem applied to the triangles ABP and ACP amounts to showing that
BA CA
= .
BP CP
Rewrite this equality as

BA BP b − a b − p
:
= 1 ⇐⇒ : = 1.
CA CP c −a c −p
Consider an inversion χ of center A (and use the convention X = χ (X)).

Reasoning on Fig. 11.44 and using Theorem 3.2, we have
B C P = AC P − AC B = AP C − B
= AP B − C = AB P − AC B = C B P .
It follows that the triangle P B C is isosceles, so P B = P C . In other words

b − ∞ b − p

c − ∞ : c − p = 1,
which is the same as

11 Inversions 479

χ (b) − χ (a) χ (b) − χ (p)

χ (c) − χ (a) : χ (c) − χ (p) = 1,
and since inversion preserves the absolute value of the cross-ratio, we are done.
Remark The argument implies that the locus of the points P with the given property
is the circle of Apollonius defined by the points B and C and the ratio AB/AC.
To summarize the idea of the solution: transform the conditions about angles in
the hypothesis and the conclusion into metric relations (using the Bisector Theorem)
and isosceles triangles, and then use the invariance of the cross-ratio of segments
under inversion.
Source International Mathematical Olympiad, 1996, proposed by Canada.
199 We reason on Fig. 11.45. We are supposed to show that BF is the reflection √ of
the median from B over the angle bisector of B. For this we consider φB , the bc
inversion determined by the triangle ABC and the vertex B. Note that φB maps the
circle to the line AC and the angle bisector of B to itself. Consequently it maps
D and E into each other.
As DE is the diameter of ω, φB maps ω into itself (ω is invariant under both the
inversion and the reflection that define φB ). It follows that M = φB (F ) is the point
where ω intersects the side AC. Because DE is diameter, EMD = 90◦ , showing

that M is the projection of E onto AC. But E is the midpoint of the arc AEC so
the projection E onto AC is the midpoint of AC. This shows that BM is the median
and consequently φB (BM) = BF is the symmedian, as desired.
Remark This problem exhibits another procedure for constructing the symmedian.
Source Russian Mathematical Olympiad, 2009.
Fig. 11.45 Proof that BF is F

symmedian
E
A
D
M
B C
480 11 Inversions
Fig. 11.46 Construction of A

M Ω
P F
E
M O1
O1
Q B C
ω
φ X (ω)
A O1 O1 A O1
γ φ X (γ )
X X
ω
Fig. 11.47 O1 is both the reflection of O1 over γ and of A over the circle ω
200 Let M be the intersection of CE and BF (Fig. 11.46). Then, by Theorem 3.33,
QM is the polar of A with respect to the circle ω, and, consequently, QM is
perpendicular to AO1 .
√
√ Consider the inversion χ over the circle γ of center A and radius AE · AB =
AF · AC, which is the square root of the power of A with respect to ω. Then
χ (ω) = ω, so ω and γ are orthogonal, also χ () = EF , and χ (EF ) = . Since
P is the intersection of AQ with and Q is on EF , it follows that χ (Q) = P .
Let O1 = χ (O1 ). Then A and O1 are reflections of each other over the circle ω.
There are many possible ways of proving this fact; here is one of them (Fig. 11.47).
Let X be one√of the two intersection points of the orthogonal circles γ and ω.
Consider the bc inversion φX determined by the triangle XAO1 and the vertex
X. Then φX (A) = O1 , φX (O1 ) = A, while φX (γ ) and φX (ω) are the perpendicular
bisectors of XO1 and XA, respectively. Then the composition of the reflection over
φX (γ ) with the reflection over φX (ω) maps A to O1 . By the Symmetry Principle,
O1 is mapped to A by the composition of the reflection over γ with the reflection
over ω. Hence the reflection of A over ω and the reflection of O1 over γ coincide.
As QM is perpendicular to AO1 , we obtain AO1 Q = 90◦ . By Theorem 3.2,
11 Inversions 481
Fig. 11.48 The image of A

through the inversion :
ω :
P
Q A
P
B B C
AP O1 = Aχ (O1 )χ (P ) = AO1 Q = 90◦ ,
as desired.
Remark A fact worth remembering is that if ω1 and ω2 are two orthogonal circles,
of centers O1 and O2 , respectively, and if O1 is the reflection of O1 over ω2 , then
O1 is the reflection of O2 over ω1 . This fact can also be proved in an elementary
way by writing the Leg Theorem for both legs of the right triangle whose vertices
are O1 , O2 , and one of the intersection points of the two circles.
201 We are supposed to prove yet another √ property of mixtilinear circles. Mixti-
linear circles go hand√ in hand with bc inversions, but for this problem we will
change slightly the bc inversion associated with the vertex C so that ω is invariant
under the resulting Möbius transformation.
Consider first the inversion of center C that maps the circle ω into itself. If A ∈
CA and B ∈ CB are the images of A and B, respectively, under this inversion,
then the circle is mapped by the inversion to the line A B (the dotted line in
Fig. 11.48). This line is tangent to the circle ω at the point P ∈ CP that is the
inverse of P , and it is also antiparallel to the line AB since the quadrilateral AA B B
is cyclic.
Next, the reflection over the angle bisector of ACB maps line A B into (since
the image of A B should become parallel to AB and should remain tangent to circle
ω).
Hence under the composition of the inversion and the reflection, the point P is
mapped to the point Q, and so
ACP = BCQ,
as desired.
Source European Girls’ Mathematical Olympiad, 2013.
202 Let the triangle be ABC; let the feet of the altitudes from A, B, C be
A , B , C , respectively; and let M, N, P be the midpoints of BC, AC, AB,
482 11 Inversions
Fig. 11.49 H is the A

orthocenter of AA1 M
X
H
A1
C
B A M
respectively. Denote by H the orthocenter, by A1 , B1 , C1 the intersections of the

perpendiculars from H to AM, BN, CP with BC, AC, and AB, respectively,
and by X, Y, Z the intersections of these perpendiculars with AM, BN, CP ,
respectively.
Let us look at Fig. 11.49. The segments A1 X and AA are altitudes in the triangle
AA1 M, so H is the orthocenter of this triangle. Then
A1 H · H X = AH · H A .
As AH · H A = H B · H B = H C · H C , we have
A1 H · H X = B1 H · H Y = C1 H · H Z.
If we show that X, Y, Z are on a circle that passes through H and its diameter lies
on the Euler line of the triangle ABC, then the circular
√ transformation that is the
composition of the inversion of center H and power AH · H A and the reflection
over H will transform this circle into the desired line (see Sect. 3.1.7), which will
then be perpendicular to Euler’s line. But notice that
H XG = H Y G = H ZG = 90◦ ,
where G is the centroid, and we are done.

Remark We have used the same circular transformation that maps the nine-point
circle to the circumcircle, obtained as the composition of an inversion of center H
and a reflection over H , that was used in Problem 193.
Source Gh. Ţiţeica, Probleme de Geometrie (Problems in Geometry), Ed. Tehnică,
Bucharest, 1929.
203 First solution. Consider an inversion of center A and denote, as it is the
convention, by X the image of the point X. Then on the one hand, C is the
midpoint of the segment AM (inversion “inverts” ratios), and on the other hand,
it is collinear with the points B and D because the circumcircle of BCD passes
through the center of inversion (Fig. 11.50). The line AC is the bisector of BMD
if and only if AMB = AMD, which is equivalent to AB M = AD M . And
11 Inversions 483
D
D
C
A C A M
M
B
B
Fig. 11.50 Quadrilateral in which a diagonal bisects the angle formed by the segments connecting
its midpoint with the other two vertices, and its inverse
this is equivalent to the fact that AB M D is a parallelogram, which happens if

and only if C B = C D . Rewrite this equality in terms of cross-ratios of complex
coordinates:
b − c b − c b − ∞
= −1 ⇐⇒ : = −1.
d − c d − c d − ∞
Here ∞ is the image of a, and so, as real valued cross-ratios are invariant under
inversion,
b−c b−a
: = −1.
d −c d −a
This is the condition that the quadrilateral is harmonic. And this condition is
symmetrical, so, by symmetry, it is equivalent to BD bisecting ANC.
Second solution. Examining carefully the solution given to Problem 3.7 from the
introduction, we notice that if M is the midpoint of the diagonal AC of the harmonic
quadrilateral ABCD, then the triangle DAB is similar to both triangles DMC and
CMB. And this proves that in a harmonic quadrilateral, the diagonal AC is the angle
bisector of BMD. Similarly the diagonal BD is the angle bisector of ANC. Now
it is important to notice that for every configuration DAC, there is a unique point

B on the arc AC that does not contain D such that ABCD is harmonic, the point
being at the intersection of the circle of Apollonius defined by D with respect to the
points A and C. For that point B, AC bisects BMD, and for any other point B
on that arc, B MA = BMA = DMA. Hence for no other point B is AC the
bisector of B MD. We conclude that any of the two properties from the statement
holds only in a harmonic quadrilateral and in such a quadrilateral both properties
hold.
484 11 Inversions
Fig. 11.51 The tangency Ma

point and a curious inversion A
Ω
A
I
Ua
B C
ΩA
Ua
La
Remark We have seen this idea before: transform a condition about angles into a
metric relation involving cross-ratios, and then use the invariance of cross-ratios
under transformations. It is worth remembering this property that characterizes
harmonic quadrilaterals.
204 We work on Fig. 11.51. Let X denote the inverse of X with respect to the
incircle ω, and, as usual, let r be the inradius. Since ω is inside , the inversion
with center I maps to a circle homothetic to , where the homothety has
center I and negative ratio.

Consider the midpoints La and Ma of the arcs BC. From the inversion we have
I A · I A = I Ua · I Ua = r 2 , while the power of I with respect to is I A · I La =
I Ua · I Ma , and so
I Ua I A
= .
I Ma I La
This means the Ma La is mapped by the homothety to Ua A . Because Ma La is a

diameter in , Ua A a diameter in .
Remark It is worth pointing out that when inverting about the incircle, the sides
become three circles of equal radii that pass through the center of inversion. The
circumcircle is mapped to the circle that passes through the pairwise intersection
points of these three circles, and we arrive at the configuration from Tzitzeica’s
Five-Lei Coin Problem (Theorem 3.38).
205 We will find the exact locations of the points X, Y, I in a three-step process.

In Problem 180 (a) we have proved that Q is the midpoint of the arc AB that does
11 Inversions 485
R
Q
C X
I
P
S
B
Fig. 11.52 Finding the location of the points I, X, Y
not contain P . From here it follows that RQ, P Q, and SQ are the respective angle
bisectors of ARB, AP B, and ACB, so X ∈ RQ, I ∈ P Q and Y ∈ SQ.
Next, using angle chasing we will prove that A, B, X, Y, I lie on a circle of center
Q, that is
QA = QB = QX = QY = QI.
The reasoning can be followed on Fig. 11.52. We will only check that QA = QI ,
because checking QA = QX and QB = QY involves exactly the same argument.
In the triangle AP B,
AI C = 180◦ − AI P = P AB/2 + AP B/2.
On the other hand,

I AQ = I AB + BAQ = P AB/2+ QB /2

= P AB/2+ AQB /4 = P AB/2 + AP B/2.
Hence AI Q = I AQ, showing that the triangle QAI is isosceles. We deduce that
QA = QI and this completes the second step.
It is at this moment that we employ inversion, and we do it in order to show that
X and Y are the tangency points of QR and QS to ω (a fact that the reader might
have guessed by looking at the figure).
486 11 Inversions
The inversion of center Q and radius QA maps the line AB to a circle passing
through Q, and because A and B are fixed by the inversion, this circle must be .
It follows that C is mapped to P and so QC · QP = QA2 . This shows that the
power of Q with respect to ω is the square of the radius of inversion, and hence ω is
mapped to itself by the inversion. From here we deduce that the tangency points of
QR and QS to ω are fixed points of the inversion, and since X and Y are themselves
fixed points (being on the circle of inversion), they must be the tangency points.
The rest of the solution is just angle chasing:
P XI + P Y I = 360◦ − XP Y − XI P − Y I P = XI Y − XP Y

= 180◦ − XQY /2 − XP Y = 180◦ − (XP Y − XCY )/4− XCY /2

= 180◦ − (XP Y + XCY )/4 = 180◦ − 90◦ = 90◦ ,
where for the first equality we have used the sum of the angles of the triangles P XI
and P Y I . The problem is solved.
Remark The geometric transformation need not be the only trick of the solution,
but it can produce an essential breakthrough.
Olympiad, 2013.
206 Invert with respect to a circle centered at B, and use the convention that
the inverse of a point is denoted by the same letter, but with a dash. The original
configuration and the inverted one are shown in Fig. 11.53.
By Theorems 3.4 and 3.5, the points A , C , M are collinear; the points
K , M , N are collinear; and the points A , C , N , K are concyclic. Let the
circumcircle of A C N K be ω. By Theorem 3.33 the polar pB of B with respect
to ω passes through M .
But pB is passes through the points B1 and B2 where the tangents from B to
ω touch the circle. And by Proposition 3.7, the image O of O is the midpoint of
B1 B2 . Note that pB (which is the line B1 B2 ) makes with BO an angle of 90◦ .
Using Theorem 3.2 we deduce that BMO = BO M = 90◦ , and the problem is
solved.
√
207 The solution can be followed on Fig. 11.54. Let φB be the bc inversion
defined by the triangle ABC and the vertex B. By Theorem 3.4, E = φB (E) is the
point that is diametrically opposite to B in the circumcircle of ABC (said differently
BEA = BCE ). Using Proposition 3.7 (i) we obtain that O = φB (O) is the
reflection of B over AC. Because φB (O ) = O and BE is diameter in ω, again by
Theorem 3.4, ω is mapped to the line through O that is perpendicular to OB at O.
The circumcircle of AOC is mapped to the circumcircle of AO C, which is the
nine-point circle of the triangle A1 BC1 , where A1 and C1 are the reflections of B
11 Inversions 487
B
M
N
K
O
A C C
B1
A
O pB
B2
M K N
Fig. 11.53 BMO = 90◦
over A and C, respectively. So P = φB (P ) must be on this nine-point circle, it is

also on the line AC, and it cannot be the midpoint C of the segment BC1 , because
C is the image of A. The only possibility is that P is the foot of the altitude from
A1 in the triangle A1 BC1 . Similarly Q = φB (Q) is the foot of the altitude from
C1 . Consequently, the line P Q is mapped to the circle of diameter BH1 , where H1
is the orthocenter of triangle A1 BC1 .
The tangency of ω and the circumcircle of AOC is equivalent to the fact that
is tangent to the nine-point circle of A1 BC1 , which is the same as saying that
the distance from the center of the nine-point circle of A1 BC1 to is equal to
the circumradius of ABC. The fact that P , Q, E are collinear is equivalent to the
fact that E is on the circle of diameter H1 E , so it is equivalent to the condition
H1 E B = 90◦ . But since E and H1 are the images of O and of the orthocenter
H of the triangle ABC through a homothety of center B, the last statement is
equivalent to saying that the Euler line OH of the triangle ABC is perpendicular to
the radius OB. We are left to prove the equivalence of these two facts, summarized
in the following simpler question:
In the triangle ABC, let O be the circumcenter. Then the perpendicular to the
radius OB through O passes through the orthocenter H of ABC if and only if
the distance from the reflection of O over BC to this perpendicular is equal to the
circumradius R of ABC.
To prove this, let O1 be the reflection of O over BC, and let M be the projection
of O1 on (Fig. 11.55). Then O1 M = R if and only if BOO1 M is a parallelogram.
488 11 Inversions
O Q
P
A C
E
P
E
Q
H1
A1 C1
O
Fig. 11.54 The simplification of the equivalence between the tangency of ω and the circumcircle
of AOC and the collinearity of P , Q, E
Fig. 11.55 Solution to the B

simplified problem
H O
M
A C
O1
But BOO1 H is a parallelogram, so the latter happens if and only if M = H , that

is, passes through H , and we are done.
Remark The condition that two circles are tangent is simplified via an inversion
to the condition that a circle is tangent to a line. In our case it is natural to try an
inversion of center B. But which inversion? We would like to have as many elements
of the original configuration preserved, and
√ it should be easy to locate the images of
the elements that are not preserved. The bc inversion keeps the triangle ABC in
place, while the image of E is on the circumcircle of ABC, and, also very important,
the image of O is the point diametrically opposite to B in ω.
11 Inversions 489
Γ2
X
Γ1
X
C2 B
Y
A B
A C1
O O
Fig. 11.56 Inverting the diagram
208 Consider the inversion of center X that maps the circle of center O into itself.
The original configuration and the inverted one can be seen in Fig. 11.56. Here it is
how things are transformed:
• X, Y , A , B are concyclic;
• the common tangent of 1 and 2 and the line AB become two circles C1 and
C2 that are interior tangent at X;
• the circles 1 and 2 become the lines A Y and B Y , tangent to C1 ;
• the circle of center O, invariant under the inversion, is now the incircle of the
triangle A B Y ;
• the lines XA, XB, XO are the lines XA , XB , XO.
We must show that XO bisects A XB .
We recognize C1 to be a mixtilinear incircle of A B Y , with X being the
point where the mixtilinear circle touches the circumcircle. And we have seen in

Proposition 3.43 that XO passes through the midpoint of the arc A Y B . Hence
XO is the angle bisector of A XB , and we are done.
Source The Competition of the Institute of Mathematics of the Romanian Academy
(IMAR), 2008, proposed by Radu Gologan.
209 The first observation is that ADBC is harmonic. Indeed, if we invert with
respect to a circle centered at P , then AB, γ , and δ are mapped to three parallel
lines, AB, γ , and δ , while ω is mapped to a circle ω that is tangent to γ at the
image C of C and to δ at the image D of D (see Fig. 11.57). Hence A D B C
490 11 Inversions
C A B
P
A B
D
D
Fig. 11.57 ADBC is harmonic
ω
E C∗
γ
C
B
M∗
P A∗
A B
M
δ D D∗
Fig. 11.58 DB is tangent to the circumcircle BMC
is a cyclic kite, and so it is harmonic. Because real valued cross-ratios are invariant
under inversion (Theorem 3.18), ADBC is harmonic.
Now we switch to Fig. 11.58. From the fact that DE is a diameter, we obtain
that DBE = 90◦ , and the desired conclusion that the circumcenter of triangle
BMC lies on BE is equivalent to DB being tangent to the circumcircle of BMC.
Take another inversion, this time centered at B, and mark the images by a star, to
avoid a conflict with the notation from the previous inversion. Then the harmonic
quadrilateral ADBC is mapped to a harmonic division on a line, and since B is
mapped to ∞, A∗ is the midpoint of C ∗ D ∗ . Moreover, because M is the midpoint
of AB, A∗ is the midpoint of M ∗ B. So M ∗ D ∗ BC ∗ is a parallelogram, showing that
C ∗ M ∗ is parallel to BD ∗ ; they form a 0◦ angle. But then the preimage of C ∗ M ∗ ,
which is the circumcircle of BMD, is tangent to the preimage of BD ∗ , which is the
line BD. The problem is solved.
Remark Here we have learned another way to construct the symmedian in the
triangle ABC: take the circle through A that is tangent to both the line BC and
the circumcircle, consider the other circle that is tangent to BC at the same point
11 Inversions 491
Fig. 11.59 The inversion that A

maps the circumcircle to the
nine-point circle
T
Q
S
O
K
H L
B F M C
and is also tangent to the circumcircle, and let D be this latter tangency point. Then
AD is the symmedian from A.
Source Stars of Mathematics Competition, Bucharest, 2009.
210 We argue on Fig. 11.59. Consider the inversion χ of center H and negative
ratio that maps the circumcircle and the nine-point circle into each other (the ratio of
this inversion is −AH · H F ). Then F = χ (A) and M = χ (Q), the second because
the similar right triangles H QA and H F M yield QH · H M = AH · H F . Also, if
L = χ (K) then H ML = 90◦ (similar triangles H KQ and H ML). Consequently
the inversion maps the circumcircle of KH Q to the line LM and the circumcircle of
F KM to the circumcircle of ALQ. So we are left to show that line LM is tangent
to the circumcircle of ALQ.
We prove this as follows. If S and T are the midpoints of the segments H Q
and H A (and therefore are on the nine-point circle, which is a consequence of the
existence of the homothety that maps the circumcircle and the nine-point circle into
each other), then the quadrilateral LMST is a rectangle, since LS and MT are
diameters in this circle, and the center of this rectangle is the center of the nine-point
circle. The orthocenter H and the circumcenter O are reflections of each other over
the center of the nine-point circle, and since H is on the side MS of the rectangle,
O must lie on side LT . But then, since LT is perpendicular to ST and the latter
is parallel to AQ, the line LT is the perpendicular bisector of the segment AQ.
This perpendicular bisector contains the circumcenter of ALQ. And since LM is
perpendicular to LT , it is perpendicular to the diameter at L, so it is tangent to the
circumcircle of ALQ, and we are done.
Remark There are two pieces of information that hint to the inversion to be used:
the condition that H QA = 90◦ tells that Q is mapped to M, and the more subtle
condition that H QK = 90◦ , which allows a nice description of the image of K in
terms of H and M.
Source International Mathematical Olympiad, 2015, proposed by Ukraine.
492 11 Inversions
CB
CD
A
S T
B H
D
Fig. 11.60 How to locate S and T
211 We start with some constructions and observations that can be followed on
Fig. 11.60. Let CB and CD be the reflections of C over B and D, respectively. Then
in the right triangle BSCB ,
SCB B = 90◦ − CB SB = 90◦ − BSC.
Using the condition from the statement, we obtain
SCB B + SH C = 90◦ − BSC + SH C = 90◦ + 90◦ = 180◦ ,
so S, CB , C, H are concyclic. Similarly T , CD , C, H are concyclic. Thus we have

replaced the unfriendly angle conditions from the statement with an intuitive
description of the locations of S and T . We should also note that, because AH
is perpendicular to BD, AB is a diameter in the circumcircle of the triangle AH B.
And since BC is perpendicular to AB, BC is tangent to this circumcircle. By a
similar argument, BD is tangent to the circumcircle of the triangle AH D.
It is now time for a geometric transformation, and we consider an inversion of
center H . As always, we denote the image under the inversion of a point P by P .
The problem is solved if we show that S T is parallel to B D , as these two lines
are the images of the circle through S, T , H and of the line BD: the angle formed
by the circumcircle of SH T with line BD at the point of contact is zero if and only
if the angle between S T and B D is zero, that is, if these lines are parallel.
Because the original quadrilateral is cyclic ( B + D = 90◦ + 90◦ = 180◦ ),
the quadrilateral A B C D is cyclic as well. Also, because B, C, CB are collinear,
11 Inversions 493
Fig. 11.61 The configuration

A T
after inversion
S
CD
X
D
CB
H
B
B , C , CB , H are concyclic, and because D, C, CD are collinear, D , C , CD ,H
are concyclic. On the other hand, because S, CB , C, H are concyclic, it follows

that S , CB , C are collinear, and because T , CD , C, H are concyclic, T , CD , C

are collinear. Note also that S and T are on the circumcircles of A B H and
A D H , respectively, as these circles are the images of the lines AB and AD. These
properties are illustrated in Fig. 11.61.
Let X and Y be the midpoints of A B and A D , respectively. We claim that X ∈
C CB and Y ∈ C CD
. To prove this claim, we note that because B is the midpoint
of CCB , ∞, C, B, CB form a harmonic division, and since inversion preserves real-

valued cross-ratios, H C B CB is a harmonic quadrilateral. By Proposition 1.23,
C CB is a symmedian in the triangle C H B . On the other hand, the fact that BC is
tangent to the circumcircle of AH B implies that A B is tangent to the circumcircle
of H B C . Consequently XB is tangent to this circumcircle. But because X is the
midpoint of the hypotenuse A B , XH B = XB H , so XH is also tangent to the
circumcircle of H C B CB . So X is the intersection point of the tangents at B and H ,
which, by Theorem 1.22, implies that X is on the symmedian C CB of the triangle
H B C , so X ∈ C CB . We conclude that C , CB , X, S are collinear. Similarly,
C , CD , Y, T are collinear. Now we know the precise locations of all elements in
the inverted figure.

Proving that S T is parallel to B D is the same as proving that S T is parallel
to XY . And this is equivalent to
CX CY

= .
XS YT
But now notice that A , S , B , H are concyclic (because A, S, B are collinear), and
since X is the circumcenter of triangle A B H , it follows that XS = XH = XA =
XB . Similarly Y T = Y H = Y A = Y D . Using this we transform the relation we
have to prove into
494 11 Inversions
XH YH
= .
CX CY
Using the similarity of the triangles XH CB and XC H as well as the similarity of
and Y C H , we transform the last relation into
the triangles Y H CD
H CB H C

= D.
CH CH
And this is equivalent to H CB = H CD , which before inversion is H C = H C .

B D
So we have to return to the original picture and prove H CB = H CD .
The point CB maps to CD by the composition of the reflection over AB and the
reflection over AD, which is the rotation about A by twice BAD. But then CB
maps to CD by the reflection over the angle bisector of CB ACD . On the other
hand, the quadrilateral ABCD being cyclic, we have ABD = ACD, so their
complements BAH and CAD are equal as well. Similarly, H AD = CAB.
Hence
CB AH = CB AB + BAH = DAH + CD AD = H ACD .
So H is on the angle bisector of CB ACD and consequently H CB = H CD , which

completes the solution.
Source International Mathematical Olympiad, 2014, proposed by Iran, solution by
contestant Yang Liu, USA.
212 The “complicated” circle is tangent to the sides AB and BC and to the
circumcircle ω of AOC. The sides AB and BC are good references, but ω makes
locating the circle harder. So we should try an inversion that keeps AB and BC
in place,
√ meaning that the center of inversion should be B. In fact, one should try
the bc inversion determined by the triangle ABC and the vertex B.
By Proposition 3.40 (v), O is mapped to the reflection O of B across AC. Then
the circle ω is mapped to the circumcircle ω of AO C; note that if we reflect B
across A and C, obtaining A1 and C1 , respectively, then ω is the nine-point circle
of A1 BC1 . Use Fig. 11.62 to understand the transformed configuration.
Now, if the reference triangle is A1 BC1 , is tangent externally to the nine-point
circle ω and to the extensions of the sides A1 B and BC1 . Feuerbach’s Theorem,
which is proved in the next chapter (Theorem 4.2), states that in every triangle, the
nine-point circle is tangent to the incircle and the excircles. It follows that must
be the B-excircle of A1 BC1 . Moreover, the line KL is mapped to circumcircle of
BK L , in which K and L are tangency points of to BC and AB, respectively.
The symmetry of the figure with respect to the bisector of B leads us to conclude
that the midpoint M of KL lies on BI . Because of that, since BM ⊥ KL, M is
the intersection of the perpendicular lines through K and L to the sides; that is,
M is the B-excenter Ib . Reversing the inversion yields BIb · BM = AB · BC. But
11 Inversions 495
A C
A1 C1
Ib
√
Fig. 11.62 Configuration after applying the bc inversion
BIb = 2BJb , in which Jb is the B-excenter of ABC, and one quick computation of
angles show that ABI and Jb BC are similar: in fact, BAI = A/2, and BJb C =
(90◦ − C/2) − B/2 = A/2. Then
AB BI
= ,
Jb B BC
which is equivalent to BI · BJb = AB · BC. And this means that I = Jb .

For the final step, BIb · BM = AB · BC, which implies that 2BJb · BM =
BI · BJb , and this proves that BM = BI /2. So the point M, which is the midpoint
of KL, is also the midpoint of BI , as desired.
213 In order to decide how to proceed, let us look √ at Fig. 11.63. From the
experience of Problem 212 we have learned that the bc inversion determined by
the triangle ABC and the vertex A transforms nicely the circumcircle of the triangle
BOC. But after realizing that all three circles pass through C, we might be tempted
to try an inversion centered at C. Yet another possibility is to take advantage of
496 11 Inversions
T O
C
K B
Fig. 11.63 Which inversion should we use?
the fact that K lies on the radical axis BC of the circumcircles of ABC and BOC
and invert with respect to a circle centered at K, thus fixing both circles. All three
choices lead to solutions, which we present here, leaving it to the readers to choose
their favorite.
√
First solution, by inversion centered at A. We reason on Fig. 11.64. The bc
inversion determined by the triangle ABC and the vertex A swaps B and C, and
transforms O into the reflection of A across BC (Proposition 3.40 (v)). If B1 and C1
are the reflections of A across the points B and C, respectively, then the circumcircle
of BOC is mapped to the nine-point circle of AB1 C1 . The circle ω through A and
C that is tangent to AB (i.e., makes an angle of 0◦ with AB) is mapped to a line
through C = B that is parallel to AB = AC. Taking A1 BC1 as reference, this line
is the B1 -midline of this triangle.
Let us move forward to examining the point T . This point is the intersection
of the circumcircle of BOC and ω, and therefore it is mapped to the intersection
between the nine-point circle of AB1 C1 and the B1 -midline, that is, T is the
midpoint of B1 C1 .
Next, let us find the image of K under the transformation. Since O is the
reflection of A across BC, O lies on B1 C1 and on the nine-point circle of AB1 C1 .
The line OT is mapped to the circumcircle of AO T . The line BC is mapped to the
circumcircle of ABC, so K is the other intersection of the circumcircles of ABC
11 Inversions 497
A K
B C
B1 O T C1
√
Fig. 11.64 Configuration after the bc inversion determined by ABC and A
and AO T . The angle AO T is right; thus AT is a diameter of the circumcircle

of AO T , and given that AT is the median of AB1 C1 , the circumcenter of AO T
is the midpoint of BC.
Notice that the nine-point circle and the circumcircle of ABC are symmetric
with respect to BC and that BC is an axis of symmetry of the circumcircle of
AO T . Since O and T are the intersection points of the nine-point circle and the
circumcircle of AO T , the reflection across BC maps O and T to the intersection
points of the circumcircle of ABC and the circumcircle of AO T , which are A and
K . Then AK is parallel to BC, and, reversing the transformation, this means that
the angle between line AK and the circumcircle of ABC is 0◦ , that is, AK is tangent
to the circumcircle of ABC.
Second solution, by inversion centered at C. Recall that the motivation behind
using an inversion centered at C is to turn all three circles into lines. In fact, an
inversion about C maps the circles to the lines A B , O B , and one more √ line
through A . Since we will have to work with the image of O, using the bc
inversion determined by ABC and the vertex C seems to be the right idea. This
transformation, pictured in Fig. 11.65, swaps A and B and maps O to the reflection
of C across AB. The circle through A and C that is tangent to AB is mapped to the
line through A = B that is tangent to the image of AB, which is the circumcircle
of ABC. The point T is, then, the intersection of and AO , the latter being the
reflection of AC across AB.
498 11 Inversions
T A
O
K
O
C
B
√
Fig. 11.65 Configuration after the bc inversion determined by ABC in C
Now let us find the image K of K. The line OT is mapped to the circumcircle
ω of O T C, and so K is the intersection of this circumcircle and CB = CA.
Since the lines AC and O T are symmetric with respect to AB, ω has AB as axis
of symmetry, and hence K is the reflection of T across AB.
We are supposed to prove √ that AK is tangent to the circumcircle of ABC. The
line AK is mapped by the bc inversion to the circumcircle of A K C = BK C.
And so the problem translates to proving that AB is tangent to the circumcircle of
BK C. But BT is tangent to the circumcircle of ABC, and the symmetry of T and
K with respect to AB yields
ABK = ABT = ACB,
which proves the desired tangency and finishes the problem.

Third solution, by inversion centered at K. Recall that the main idea here is
to keep the circumcircles of BOC and ABC in place by inverting about a circle
centered at K and of radius the square root of the power of K with respect to the
two circles. This inversion swaps B and C and, therefore, T and O. We want to
prove that AK is tangent to the circumcircle of ABC, which is equivalent to the fact
11 Inversions 499
O1
O2
T
O
K C
B
Fig. 11.66 Configuration after inverting about K
that KA2 = KB · KC, which, in turn, is the same as proving that A is fixed by the
inversion. We work on Fig. 11.66.
It remains to investigate the image of the circle that is tangent to AB and passes
through C, which is also the circumcircle of ACT . It is mapped to the circumcircle
of A C T = A BO. So we are done if we prove that the circumcircle of ABO is this
inverse, because then A = A . One way of doing this is by using the invariance of
the angle made with the line BC, which is fixed by the inversion. More specifically,
if O1 is the circumcenter of ACT and O2 is the circumcenter of ABO, it suffices to
prove that O1 CB = O2 BC. An angle chase shows that this is the case:
O1 CB = O1 CA + ACB = CAO1 + ACB = 90◦ − BAC + ACB,
and
O2 BC = O2 BO + OBC = 90◦ − BAO + 90◦ − BAC

= 90◦ − (90◦ − ACB) + 90◦ − BAC = 90◦ − BAC + ACB.
And we are done.

Remark The first solution also proves that AT is the A-symmedian of ABC. It also
proves that (C , T , B , K ) = −1, which is the same as (C, T , B, K) = −1.
500 11 Inversions
With a little extra effort, one can prove that AT ⊥ KT . Indeed, from the third
inversion, KT · KO = KA2 , which proves that the triangles KAT and KOA
are similar; since OA ⊥ AK, which follows from the tangency, we deduce that
KT ⊥ AT .
√
214 A bc inversion √ involves isogonal cevians, so it can be useful. Therefore
we perform the bc inversion φA defined by ABC and the vertex A. The
transformation φA takes P to a point Q on the line BC with the property that the
lines AP and AQ are isogonal conjugates with respect to the vertex A in the triangle
ABC. Then A1 = φA (A1 ) is the intersection of φA (P A) = QA and φA (BC) = ω.
Since A1 and A2 are symmetric with respect to the circumcircle ω of ABC, the
Symmetry Principle (Theorem 3.15) implies that A1 = φA (A1 ) and A2 = φA (A2 )
are symmetric with respect to φA (ω), which is the line BC. The line AA2 is mapped
to the line AA2 .
The situation up to this moment is depicted in Fig. 11.67. Since BAP =
A1 AC we obtain BP = A1 C = A2 C and P C = BA1 = BA2 , which proves
that BP CA2 is a parallelogram. Now we switch to complex coordinates, with the
convention that the lowercase letter is the complex coordinate of the uppercase letter
point. We have a2 + p = b + c, hence a2 = b + c − p. So the points on the line
AA2 are of the form
ta + (1 − t)a2 = ta + (1 − t)(b + c − p), t ∈ R.
Taking t = 1/2 gives us the point T whose complex coordinate is
a+b+c−p 3g − p
t= = ,
2 2
Fig. 11.67
√ Configuration
after the bc inversion A
determined by ABC and A
A2
A1 Q
B C
P
A1
11 Inversions 501
Fig. 11.68 Which of circles A

will be transformed into a
line?
M
X
Y Z
B D C
in which g is the coordinate of the centroid G of ABC. The expression for t is

symmetric with respect to a, b, and c, so it belongs to all lines AA2 , BB2 , and CC2 ,
which are the isogonal lines of AA2 , BB2 , and CC2 , respectively. Therefore these
three lines have a common point, namely, the isogonal conjugate of T .
It remains to prove that T lies in the nine-point circle of ABC. This can be done
−→ −→
by rearranging the expression for t as t − g = − 12 (p − g), that is, GT = − 12 GP .
This equation means that P is mapped to T by a homothety with center G and ratio
− 12 . This homothety also maps the circumcircle of ABC to the nine-point circle of
ABC, and since P belongs to the circumcircle of ABC, T belongs to the nine-point
circle of ABC. The proof is complete.
215 Proving that two circles are tangent to each other is usually harder than
proving that a line is tangent to a circle, so we should try to invert so as to turn
one of the circles into a line. Looking at Fig. 11.68 it becomes natural to pick the
circle of diameter AD as the one that is transformed into a line.
Hence we invert about a circle of center D and arbitrary radius, and denote, as
usual, the image of a point P by P . The circle ω of diameter AD is mapped to a line
ω ; since ω is tangent to BC, which is fixed by the inversion, ω is parallel to BC.
We also have DB Y = DY B = 90◦ and, similarly, DC X = DXC = 90◦
(Theorem 3.2). The quadrilateral DY MX has two opposite right angles at vertices
X and Y , so it is inscribed in the circle with diameter DM. Since this circle is also
tangent to BC, its image is the line X Y , which is parallel to BC. Putting everything
together we obtain that B Y X C is a rectangle.
The point M is the projection of D onto X Y . Since DA = 2DM, it follows that
DM = 2DA (inversion “inverts” ratios), and from here we deduce that ω is the
perpendicular bisector of DM . We are required to prove that this perpendicular
bisector is tangent to the circumcircle of X Y Z . For a better intuition, let us
examine Fig. 11.69.
Let O be the center of rectangle B Y X C . Then the power of O with respect to
the circumcircles of the triangles B X Z and C Y Z is OB · OX = OC · OY ,
502 11 Inversions
M
Y X
O
A
B D C
Fig. 11.69 Inverted figure
since all four segments in this equation are equal in length. So O lies on the radical
axis of these circumcircles, that is, O lies on the line Z D. Now
Y Z O = Y Z D = Y C D = DB X = DZ X = OZ X ,
meaning that Z O bisects X Z Y . But O also lies on the perpendicular bisector

of X Y , and these two conditions can only happen if O lies on the circumcircle of
X Y Z . As the line A O being parallel to the chord X Y in this circumcircle and

as O is the midpoint of the arc X Y , it follows that the line A O is tangent to this
circumcircle. And this completes the solution, because the line A O = ω is the
image of the circle with diameter AD and the circumcircle of X Y Z is the image
of the circumcircle of XY Z.
216 We argue on Fig. 11.70. Since AP = AQ, BP T = CQT ; so we want
to prove that the triangles BP T and CQT are similar. Now, looking at the triangle
P QS, we notice that, since P A and QA are tangent to the circumcircle γ of P QS,
ST is a symmedian, so, using the formula derived at the end of Sect. 2.4.3,
PT SP 2
= ,
QT SQ2
11 Inversions 503
Fig. 11.70 Where is the A

symmedian?
I Q
T
P
B C
and this is a clue that we should try to prove that
PB SP 2
= .
QC SQ2
Now we use Propositions 3.42 and 3.44 to deduce that I is the midpoint of DE

and that SP and SQ meet the circumcircle of ABC at midpoints of the arcs AB and

AC that do not contain the other vertex. From these observations we obtain that
P SB = ASP = QSI,
where the latter equality follows from the fact that SI and ST are isogonal
conjugates in the triangle SP Q or from Proposition 3.43. Since AB is tangent to
γ at P , we have that SP B = SQI .
The two angle equalities show that the triangles SP B and SQI are similar, so
PB QI
= .
SP SQ
Analogously, the triangles SQC and SP I are similar (interchange B and C, as well
as P and Q), and
QC QI
= .
SQ SP
Dividing these two equations yields
PB SP 2
= ,
QC SQ2
as desired.
504 11 Inversions
Remark But where is the inversion? In the proofs of Propositions 3.42 and 3.44.
Source Chinese Team Selection Test for the International Mathematical Olympiad,
2005.
√
217 Consider all three bc inversions φA , φB , φC determined by ABC, and let
ωb and ωc be the B- and C-mixtilinear excircles of ABC, respectively. If ω is the
incircle of ABC, by Proposition 3.41, φA (ωb ) = φC (φB (ωb )) = φC (ω) = ωc , and
similarly φA (ωb ) = ωc . A nice corollary of this fact is that φA (AOb ) = AOc . Let ωb
touch AB at F , and let ωc touch AC at G. Then, by the inversion, φA (F ) = G, and
if Ob = φA (Ob ), AOb G = AOb F = AF Oc = 90◦ , so Ob is the orthogonal
projection of G onto AOc .
We also have that φA (D) is the tangency point of φA (ωb ) = ωc and φA (BC),
which is the circumcircle of ABC, therefore φA (D) = S.
Now focus on Fig. 11.71, paying particular attention to ωc . The triangle Oc GA
is right-angled at G, so, by the Leg Theorem, Oc Ob · Oc A = Oc G2 . This means
that Ob is also the image of A under an inversion χc about ωc . Let Dc = χc (D), the
image of D under this inversion. Recall that Oc E ⊥ BC. Then
EDc Oc = DEOc = 90◦ ,
ωc
G
Oc
Ob
Dc F
A Oc ωb
S Ob
Y R Eb
E B C D
Fig. 11.71 Inversion, inversion, inversion! Point X not shown for the sake of clarity
11 Inversions 505
that is, EDc ⊥ DDc . Moreover, since χc (A) = Ob and χc (D) = Dc , AOb Dc D is
cyclic. So Dc lies on the circumcircle of AOb D = AOb S = φA (Ob S). Also notice
that Y lies on the line Dc D.
Analogously, if χb is the inversion about ωb and Eb = χb (E), then Eb lies on
the circle φA (Oc R). We also have DEb ⊥ EEc , so DEb EDc is cyclic, and Y is
the intersection of the diagonals Dc D and EEb . By power-of-a-point, Eb Y · EY =
Db Y ·DY , which means that Y has the same power with respect to the circumcircles
of ADOb Dc = φA (Ob S) and AEOc Eb = φA (Oc R). The other intersection point
of these two circles is φA (Ob S ∩ Oc R) = φA (X) = X . So Y lies on the line
AX , which completes our solution because we know that AY and AY are isogonal
conjugates.
√
Remark A piece of knowledge acquired from this solution is that the bc inversion
corresponding to the vertex A in the triangle ABC swaps the B- and C-mixtilinear
excircles.
Source WenWuGuangHua Mathematics Workshop, China.
218 We argue on Fig. 11.72. First we prove that the point X with this property
is unique. Indeed, suppose that X is another point with this property, and suppose
without loss of generality that X is closer to B than X. Then, if E and F are the
orthogonal projections of X onto BI and CI , respectively, and if M is the midpoint
of E F , a first observation is that BE < BE and CF > CF . If we regard BC as
the x-axis of a coordinate system, and denote by xT the x-coordinate of point T , and
assume that xB < xC , we find that BE < BE implies xE < xE , and CF > CF
implies xF < xF . Hence
xE + xF xE + xF
xM = < = xM .
2 2
But P B = P C if and only if xP = xB +x C

2 , so the inequality xM < xM implies

that M and M cannot be at the same distance from B and C simultaneously. This
proves that X is unique.
Now we work backward and prove that if X is constructed as above and is chosen
such that BAD = CAX, namely, such that it is the point of tangency of the A-
mixtilinear excircle and the circumcircle, then the point M used in the construction
of X satisfies MB = MC. We will prove for later use the following result:

Lemma Let Ia be the A-excenter of ABC, let N be the midpoint of the arc BC that
contains A, and let X be the tangency point of the A-mixtilinear excircle and the
circumcircle of ABC. Then X, Ia , and N are collinear.
√
Proof The proof is very similar to that of Proposition 3.43. Let φA be the bc
inversion determined by ABC and the vertex A. If D is the tangency point of
the incircle with the side BC then φA (D) = X, and also φA (Ia ) = I . So by
Proposition 3.40, part (iii), (AX, XIa ) = (AI, DI ). Now, if L is the midpoint of
506 11 Inversions
Fig. 11.72 Point X asks: N

should I stay or should I go? A
I
E M F
B E∗ F∗ C
M∗
Ia

the arc BC that does not contain A, then LN ⊥ BC and I D ⊥ BC, so
(AI, I D) = (AIa , LN) = ALN = (AX, XN ),
and the lemma is proved.

With this lemma in tow, we can approach the actual problem. Consider triangle
BIa C and its circumcircle . The incenter I belongs to this circle. Indeed, because
I = φA (Ia ) and B = φA (C), by using Theorem 3.2, we can infer that the triangles
AI C and ABIa are similar, and from here we obtain that AIa B = ACI = C/2,
which combined with AI B = 180◦ − A/2 − B/2 implies I BIa = 90◦ .
Similarly I CIa = 90◦ , and so not only I is on , but I Ia is a diameter of .
We also have CBIa = 90◦ − 12 B, BCIa = 90◦ − 12 C, and so
1
BIa C = 180◦ − CBIa − BCIa = 90◦ − A.
2
However, NBC = NCB = 90◦ − 12 A = BIa C. So Ia N is a symmedian in

the triangle BIa C, and X is a point on it. Thus, since BIa ⊥ BI and XE ⊥ BI ,
BIa is parallel to XE and
BE d(X, Ia B) Ia B
= = ,
CF d(X, Ia C) Ia C
11 Inversions 507
where for the second equality we used a property of symmedians proved in the
solution to Problem 16.
Finally, if E ∗ and F ∗ are the orthogonal projections of E and F onto BC,
respectively, then since EBE ∗ = I BC = I Ia C, it follows that the triangles
BEE ∗ and Ia I C are similar, so
BE BE ∗
= .
Ia I Ia C
Analogously,
CF CF ∗
= .
Ia I Ia B
Dividing these two equations yields
BE BE ∗ Ia B
= · .
CF Ia C CF ∗
This together with
BE Ia B
=
CF Ia C
implies that BE ∗ = CF ∗ . The orthogonal projection M ∗ of M onto BC is the

midpoint of E ∗ F ∗ , so from BE ∗ = CF ∗ , we deduce that M ∗ is the midpoint of BC,
so the line MM ∗ is the perpendicular bisector of BC. Consequently MB = MC,
Source Iranian Team Selection Test for the International Mathematical Olympiad,
2014.
√
219 The argument should be followed on Fig. 11.73. We perform the bc
inversion determined
√ by ABC and the vertex A, and we denote by a dash the image
under this bc inversion. We also denote by a star the image under the reflection √
over the angle bisector of A, which reflection is part of the definition of this bc
inversion.
First notice that E and F are swapped by this transformation. Let γ be the
circumcircle of DEF ; we want to understand γ , the image of γ . If X ∈ γ , then
X satisfies AX · AX = AB · AC = AE · AF . On the other hand X∗ ∈ AX , and
AX∗ · AX = AX · AX . Hence AX∗ · AX = AE · AF , so X∗ ∈ γ . Consequently
γ is the circumcircle of DEX∗ , and so γ = γ ∗ . From here we deduce that the
point T is the other intersection of γ = γ ∗ , with BC. The circles γ and γ ∗ have
the same radius, which we denote by R, and from the extended Law of Sines (which
relates the circumradius to the pair angle–opposite side), we deduce that
508 11 Inversions
Fig.
√ 11.73 A perfect day for A
a bc inversion
D E T
C
B
S1
T
F
DF FT
2R = = .
sin DEF sin F ET
Hence DF = F T , which shows that the triangle DF T is isosceles, and so the

projection M of F onto BC is the midpoint of DT . However, since F is the

midpoint of BC, M is the midpoint of BC. Then D, the tangency point of the
incircle in BC, and T are symmetric about M. This means that T is the tangency
point of the A-excircle and BC (see Sect. 2.2.1).
Now we can complete the argument. The A-excircle is fixed by the reflection
across AE, while the circumcircle γ ∗ = γ of F ET is mapped to the circumcircle
γ of DEF . And T is the intersection of γ ∗ and the A-excircle, so its reflection is
the intersection of the circumcircle γ of DEF and the A-excircle, that is, one of S1
or S2 , say S1 . However, the reflection of the line AT is the line AT , so AT passes
through S1 , and we are done.
Remark Reversing the transformation yields that T is the tangency point between
the A-mixtilinear incircle and .
Source United States of America Team Selection Team for the International
Mathematical Olympiad, 2016, proposed by Evan Chen.
220 Before starting the solution, we introduce the notation κi = 1/ri , i =
1, 2, 3, 4. These numbers are called the curvatures of the circles, and the Descartes
relation is written in terms of curvatures as
2(κ12 + κ22 + κ32 + κ42 ) = (κ1 + κ2 + κ3 + κ4 )2 ,

11 Inversions 509
O
ω1
ω1
A
ω3
Q3
ω2 O B
ω4
ω4
Q4
C
ω2
ω3
Fig. 11.74 Proof of Descartes’ formula
or

4
κj2 = 2 κj κk .
j =1 j <k
The proof of Descartes’ formula can be followed on Fig. 11.74. The starting
point is clear, too many circles, so we invert about a tangency point. Let therefore
O be the tangency point of ω1 and ω2 , and let us invert about the circle of center O
√
and radius 2 r1 r2 . The lines ω1 and ω2 are parallel, with O lying between them,
and ω1 is at distance 2r2 from O, while ω2 is at distance 2r1 from O. Let A be the
projection of O onto ω2 .
The circles ω3 and ω4 are tangent to the lines ω1 and ω2 , are tangent to each
other, and their radii are equal to r1 + r2 . Let Q3 and Q4 be the centers of ω3 and
ω4 , and let B and C be their tangency points with ω2 , respectively, and assume that
Q3 is closer to O (O lies outside of both circles). Note that BC = 2(r1 + r2 ). For
convenience we will set AB = a, and for a circle ω we will denote by ρ(ω) the
power of the point O with respect to ω.
At this moment we recall that the inverse about a circle of center O and radius
R of a circle ω that does not pass through O is also the image of this circle through
the homothety of center O and radius R 2 /ρ(ω), and so the radius r of the circle and
the radius r of the inverse satisfy
r R2
= .
r ρ(ω)
Using this formula we obtain

510 11 Inversions
r3 4r1 r2
= .
r1 + r2 ρ(ω3 )
Using the formula for the power-of-a-point, the Pythagorean Theorem, and the fact
that O is at distance |r1 − r2 | from Q1 Q2 , we obtain
ρ(ω3 ) = OQ23 − (r1 + r2 )2 = (r1 − r2 )2 + AB 2 − (r1 + r2 )2

= (r1 − r2 )2 + a 2 − (r1 + r2 )2 = a 2 − 4r1 r2 .
Combining these formulas we obtain

4r1 r2 (r1 + r2 + r3 )
a= .
r3
Similar computations for ρ(ω4 ) yield
r4 4r1 r2
=
r1 + r2 ρ(ω4 )
and
ρ(ω4 ) = OQ24 − (r1 + r2 )2 = (r1 − r2 )2 + AC 2 − (r1 + r2 )2

= (r1 − r2 )2 + (a + 2(r1 + r2 ))2 − (r1 + r2 )2 = a 2 + 4a(r1 + r2 ) − 4r1 r2 .
Substituting ρ(ω4 ) from one equation into the other, we obtain
r1 r2 (r1 + r2 )
= a 2 + 4a(r1 + r2 ) − 4r1 r2 .
r4
Replace a by the formula that we have obtained above and switch to curvatures to
obtain
√
κ4 = κ1 + κ2 + κ3 + 2 κ1 κ2 + κ1 κ3 + κ2 κ3 .
Write this as
√
κ4 − κ1 − κ2 − κ3 = 2 κ 1 κ 2 + κ1 κ 3 + κ2 κ 3 ,
then square both sides to arrive at Descartes’ formula.

Remark Descartes’ relation can be viewed as a quadratic equation in κ4 , which has
two solutions. One solution is the curvature of ω4 ; the other solution (or rather its
absolute value) is the curvature of the other circle that is tangent to ω1 , ω2 , ω3 (this
circle is either exterior tangent or interior tangent to all three). Descartes’ relation
11 Inversions 511
D0
Y
E
L
F A
X
P I
T
B C
D K
Fig. 11.75 How does T relate with the other points?
is an essential tool in studying Apollonian gaskets, which are obtained by starting

with three circles and then filling in with new circles, each tangent to other three.
In differential geometry the concept of curvature is extended to any smooth
curve; it is the reciprocal of the radius of the circle that best approximates the curve
at a given point (also known as the oscullating circle, from the Latin osculor = to
kiss). The reciprocal of the radius is the right choice because the more “curved” the
curve is, the bigger its curvature is.
Source Descartes’ relation was known to François Viète, as well. Frederick Soddy
has extended it to spheres.
221 We work on Fig. 11.75. The most peculiar point in this problem is T , so
we try to work as much as we can around it. Keep in mind that we are to prove
that the line AT goes through the tangency point of the A-mixtilinear incircle
and the circumcircle√ . We have learned from Proposition 3.42, which itself is a
consequence of the bc-inversion centered at A determined by ABC, that the line
AT is the isogonal of the line AK, where K is the tangency point of the A-excircle
with BC. This information makes us a bit more confident, but the A-excircle does
not play a big of a role in this problem; the incircle is more important. Is there
anything relating the incircle and the A-excircle?
Yes, there is! Recall from Chap. 2 that the homothety with center A that takes
the A-excircle to the incircle also takes K to the point D0 , diametrically opposite
to D in the incircle. So now we have a more concrete goal: to prove that if I is
the incenter of ABC, then AI bisects T AD0 . Furthermore, since I T = I D0 , it
512 11 Inversions
suffices to prove that the points A, D0 , I , and T are concyclic, because equal chords
subtend equal arcs.
The points D0 and T are on the incircle ω of ABC, so inverting about ω seems
to be a good idea. In fact, this inversion, which we call χ , takes A to the midpoint
A of EF (Proposition 3.3), so the problem reduces to proving that χ (A) = A ,
χ (D0 ) = D0 , and χ (T ) = T are collinear. Because DD0 is a diameter in ω, this is
equivalent to showing that A T D = 90◦ .
The inversion χ has been used for proving Euler’s relation (Theorem 3.36),
and we know that it maps the circumcircle to the nine-point circle γ of DEF .
Theorem 3.26 implies that the circles ω, γ , and are coaxial. With this in mind, let
P be the intersection point of the lines DT and XY . (If DT and XY are parallel,
then DT F E and DT XY are trapezoids, and A lies on the line I O, which means that
ABC is isosceles, and the conclusion follows from symmetry.) Then P lies on the
common radical axis of ω, γ and , and by power-of-a-point, P T ·P D = P A ·P L,
where L is the projection of D on EF . So the quadrilateral DT LA is cyclic, and
DT A = DLA = DLE = 90◦ . We are done.
Source Cosmin Pohoaţă.

Chapter 12
A Synthesis
222 (a) First solution. In all solutions the triangle ABC is oriented counterclock-
wise. The first solution uses spiral similarities and can be followed on Fig. 12.1.
Let A , B , and C be the centers of the equilateral triangles A1 BC, B1 CA, and
C1 AB, respectively. The point A maps to B by the spiral similarity of center
B, angle 30◦ , and ratio BC/BA , which maps A to C, followed by the spiral
similarity of center A, angle ◦

√ 30 , and ratio AB /AC, which maps C to B . Because
AB /AC = BA /BC = 3/3, the composition of the two spiral similarities is an
isometry which rotates figures by 60◦ ; it is a rotation. The center of the rotation is
the fixed point of the composition of the two spiral similarities. We show that this
center is C . √
Because C BC1 = 30◦ and BC /BC1 = 3/3, the first √ spiral similarity maps
C to C1 . Also, because C AC1 = 30◦ and C A/C C1 = 3, the second spiral
similarity maps C1 back to C . So C is indeed the fixed point of the rotation; it is the
center of the rotation. Therefore A C B = 60◦ and C A = C B . We conclude
that the triangle A B C is equilateral, as desired.
Second solution. This solution and the next use isometries. Looking at Fig. 12.1,
we notice that the problem can be rephrased as:
Problem On the sides of a triangle ABC construct in the exterior the triangles
A BC, B CA, and C AB such that
A BC = AB C = B CA = B AC = C AB = C BA = 30◦ .
Prove that the triangle A B C is equilateral.

Consider the rotations ρa , ρb , and ρc by 120◦ about A , B , C , respectively.
Their composition ρa ◦ ρb ◦ ρc is an isometry that rotates figures by 120◦ + 120◦ +
120◦ = 360◦ . So it is either a translation or the identity map. But ρa ◦ρb ◦ρc (B) = B,
so it is the identity map, that is
514 12 A Synthesis
Fig. 12.1 Proof of C1

Napoleon’s theorem
C A
B1
B
B
30◦
30◦
C
A
A1
ρa ◦ ρb ◦ ρc = 1.
Then
ρb ◦ ρc = ρa−1 .
This implies that the fixed point of ρb ◦ ρc is A , the center of the rotation ρa−1 . But
it is not hard to see that the fixed point of ρb ◦ ρc forms with two centers of rotation
B and C of ρb and ρc an equilateral triangle (oriented counterclockwise). This is
because ρc maps this point to its reflection over B C , while ρb will bring it back to
the initial position. Hence A , the fixed point of ρb ◦ ρc = ρa−1 , forms with B and
C an equilateral triangle.
Third solution. The complex number solution is also quite simple. Let the com-
plex coordinates of the points A, B, C, A , B , C, be a, b, c, a , b , c , respectively,
and assume that the triangle ABC is oriented counterclockwise. We work with
angles measured in radians. Because the triangle A BC is isosceles with BA C =
2π 2π
3 , C rotates to B about A by 3 . So
b − a 2π i

=e 3 .
c−a
Solving for a we obtain

2π i
b − ce 3
a = 2π i
.
1−e 3
Similarly
12 A Synthesis 515
2π i 2π i
c − ae 3
a − be 3
b = 2π i
, c = 2π i
.
1−e 3 1−e 3
To prove that A B C is an equilateral triangle, it suffices to check that a rotation

by π3 around A maps B to C . In complex coordinates, we have to check that
πi
c − a = e 3 (b − a ). Ignoring the denominator, which is the same for a , b , and
c , we have to check that
2π i 2π i πi 2π i 2π i
a − be 3 − b + ce 3 = e 3 (c − ae 3 − b + ce 3 ).
And this is straightforward by using the identities

πi 4π i 2π i 4π i
e 3 = −e 3 and 1 + e 3 +e 3 = 0.
Fourth solution. The triangles ABC1 , B1 CA, CA1 B are directly similar, so
they are mapped into one another by spiral similarities. The general form of the
Averaging Principle (Theorem 2.19) implies that the centroids of the triangles
AB1 C, BCA1 , and C1 AB form a triangle that is similar to each of the three
triangles. Hence this triangle is equilateral.
(b) First solution. The spiral similarity of center C, ratio 2 cos 30◦ , and angle
30 maps B to A and A to A1 . It thus maps B A to AA1 . This spiral similarity
◦
is depicted in Fig. 12.2. The spiral similarities of the same angles and ratios but
of centers A and B map C B to BB1 and A C to CC1 , respectively. This means
that AA1 , BB1 , CC1 are of length 2 cos 30◦ times the side-length of the equilateral
triangle A B C . This proves the equality of the three segments.
But moreover, the lines of support of these three segments form the same angles
as A B , B C , C A , and these are 60◦ angles. Let P be the intersection of AA1 and
BB1 . Because (AA1 , BB1 ) = 60◦ , P is on the circumcircles of ABC1 , BCA1 ,
ACB1 . The same must be true, by symmetry, about the intersection of AA1 and
CC1 . But the three circles have at most one point of intersection (because their
centers are not collinear). The intersection point must therefore be the common
point of the three lines.
Second solution. Using the fact that A1 , B1 , C1 are obtained by the 60◦ rotations
about C, A, B of the points B, C, A, respectively, we can write
πi πi πi
a1 = e 3 (b − c) + c, b1 = e 3 (c − a) + a, c1 = e 3 (a − b) + b,
and consequently
πi πi πi πi
a − a1 = −a + e 3 b + (1 − e 3 )c, b − b1 = −b + e 3 c + (1 − e 3 )a,
πi πi
c − c1 = −c + e 3 a + (1 − e 3 )b.
516 12 A Synthesis
Fig. 12.2 The spiral C1

similarity that maps BA to
AA1
C A
B1
B
B
C
A
A1
2π i 2π i
It is easy to check that b − b1 = e 3 (a − a1 ), c − c1 = e 3 (b − b1 ); hence the
three segments have equal lengths.
The equations of the lines AA1 , BB1 , and CC1 are
(a − a1 )z + (a1 − a)z + (aa1 − a1 a) = 0

(b − b1 )z + (b1 − b)z + (bb1 − b1 b) = 0
(c − c1 )z + (c1 − c)z + (cc1 − c1 c) = 0.
The fact that the three lines intersect at one point is reformulated by saying that this
system of equations in the unknowns z and z has solution. This is equivalent to the
fact that

a − a1 a1 − a aa1 − a1 a

b − b1 b1 − b bb1 − b1 b = 0,

c − c c − c cc − c c
1 1 1 1
and checking this is a long but routine computation.

Remark If the angles of the triangle ABC are all less than 120◦ , then P is in the
interior of the triangle and is called the Fermat-Torricelli point of the triangle. It
is the point in the plane of the triangle with the sum of distances to the vertices
minimal.
Source While the result stated in (a) is usually attributed to Napoleon Bonaparte,
scholars were able to trace it back to W. Rutherford (1825). The fourth solution
of (a) is from T. Andreescu, M. Rolínek, J. Tkadlec, Geometry Problems from the
AwesomeMath Year-Round Program, XYZ Press, 2013.
12 A Synthesis 517
223 First solution. We will employ the system of coordinates motivated by homo-
thety and reflection that was introduced before Problem 4.1, with the coordinates
of the vertices A, B, C, D being 1, aω, −ka, −kω, respectively, where k, a > 0,
ωω = 1. Then
AB · CD + AD · BC = |1 − aω| · |ka − kω| + |1 + kω| · |ak + aω|

= k|ω| · |1 − aω| · |1 − aω| + a|ω| · |1 + kω| · |1 + kω|
= k|(1 − aω)(1 − aω)| + a|(1 + kω)(1 + kω)|
= k(1 − aω)(1 − aω) + a(1 + kω)(1 + kω)
= k + ka 2 − ka(ω + ω) + a + ak 2 + ka(ω + ω)
= k + ka 2 + a + ak 2 = (k + a)(1 + k) = AC · BD.
Second solution. We can write Ptolemy’s identity as
AB DB AD BD
: + : = 1,
AC DC AC BC
or, in complex coordinates, |(a, d, b, c)| + |(a, b, d, c)| = 1. We can remove the
absolute value bars because the points lie on the circle in the order A, B, C, D,
so the cross-ratios are positive (the two fractions in each cross-ratio have the same
argument, by inscribed angles). We are supposed to prove that
(a, d, b, c) + (a, b, d, c) = 1.
As the cross-ratio is mapped by a circular transformation to either itself or its

complex conjugate, if we consider a circular transformation that sends A to ∞,
then Ptolemy’s identity is equivalent to
(∞, d , b , c ) + (∞, b , d , c ) = 1,
where dashes denote the images of points. This is the same as
d − c c − b
+ = 1,
d − b d − b
and this is obviously true.

Third solution. We have seen in the second solution that the property that a
quadrilateral satisfies Ptolemy’s identity is invariant under circular transformations.
We can map A, B, C, D by a circular transformation to points on the real axis that
have the coordinates 0, x, x + y, x + y + z, x, y, z > 0. Here we use the fact
that circular transformations preserve the relative position of points on a line (that
contains ∞). Ptolemy’s identity for these four points reads
518 12 A Synthesis
Fig. 12.3 A slightly forced B P

configuration of
A, B, C, D, S to show what
A
happens when S is not the
midpoint of P Q D
S
D
A
C
(x + y)(y + z) = x(x + y + z) + yz,
and this is obviously true.

Remark A lesson to be learned from this problem as well as from many others: If
you have to prove a property, find the largest group of transformations that preserve
it. Then, for each configuration find a configuration that is equivalent to it modulo
transformations and for which the property is easy to verify.
Working with absolute values in the last two approaches, and using the triangle
inequality, we can prove that, in general, if A, B, C, D are four points in the plane,
then AC · BD ≤ AB · CD + AD · BC, with equality if and only if A, B, C, D are
on a circle in this order.
224 First solution. Consider the reflection over the line through S that is perpen-
dicular to P Q, and let A and D be the images of A and D, respectively (Fig. 12.3).
Then
A SQ = ASP = BSP
shows that A , S, B are collinear. Similarly D , S, C are collinear, so S is the

intersection of the diagonals of the quadrilateral BCA D . Then
CD A = SD A = SDA = CBS = CBA ,
which implies that BCA D is a cyclic quadrilateral. Thus both BCDA and
BCA D are cyclic.
If the circumcircles of these two quadrilaterals are distinct, then one of the points
A and D lies outside of ω, and one is inside. But then ω intersects the circumcircle

of BCA D in three points: B, C, and some point on the arc AD. This is impossible.
So the two circumcircles coincide. But then A , D are on ω, showing that ω is
12 A Synthesis 519
Fig. 12.4 The inverses of B P A

A, B, C, D
ω B A
ω
S
C D
C D
A1
T2
T3
M2
M3
S2
S3
S1
A2 T1 M1 A3
Fig. 12.5 M1 M2 M3 and S1 S2 S3 have parallel sides
invariant under the reflection. This implies that the line of reflection passes through
the midpoint of the chord P Q, so S is the midpoint of this chord.
Second solution. The triangles SAD and SCB are similar, √ hence SA · SB =
SC · SD. Consider the inversion of center S and radius R = SA · SB, let ω be the
circle that is the inverse of ω, and let A , B , C , D be the inverses of A, B, C, D,
respectively (Fig. 12.4). Then SA = SB, SB = SA, SC = SD, SD = SA,
showing that A , B , C , D are the reflections over P Q of B, A, D, C, respectively.
Hence ω is the reflection of ω over P Q, from where we infer that P and Q on both
circles. But then P and Q are invariant under the inversion, so they are on the circle
of inversion. Therefore SP = R = SQ, and the problem is solved.
Source Polish Mathematical Olympiad, 2005.
225 We will show that the triangles M1 M2 M3 and S1 S2 S3 have parallel sides, and
then by Problem 84, they are homothetic. And then M1 S1 , M2 S2 , and M3 S3 intersect
at the center of the homothety. The configuration can be seen in Fig. 12.5.
520 12 A Synthesis
Fig. 12.6 Spiral similarity B1

keeps XY Z equilateral X
A2
B2
O
A1
Z
Y
Y
B3
Z
A3 A3
B3
Of course, because A1 A2 A3 and M1 M2 M3 have parallel sides, it suffices to show

that A1 A2 A3 and S1 S2 S3 have parallel sides. Let us show that A1 A2 is parallel to
S1 S2 . For this we employ. . . reflections!
Because T2 reflects to T3 and T1 reflects to S1 over the angle bisector of A1 , the

arc T2 T1 reflects to the arc T3 S1 . Similarly, the arc T2 T1 reflects to the arc T3 S2 over

the angle bisector of A2 . So T3 S1 =T3 S2 , showing that the two arcs reflect into each
other over the diameter from T3 , and in particular, S1 reflects over this diameter to
S2 . But A2 also reflects over this diameter to A1 , since the tangent is perpendicular
to the diameter. Therefore S1 S2 is parallel to A1 A2 . The same argument can be
repeated for the other sides, and the problem is solved.
Source International Mathematical Olympiad, 1982, proposed by the Netherlands.
226 First solution. The property is true if A1 B1 A2 B2 A3 B3 is a regular hexagon.
To arrive at an arbitrary configuration, you can transform the triangles one at a time
by a spiral similarity centered at O. Let us show that if the triangle formed by the
three specified midpoints is equilateral before such a spiral similarity is applied,
then it stays equilateral afterward. Let therefore the configuration before the spiral
similarity consist of the triangles OA1 B1 , OA2 B2 , and OA3 B3 with the midpoints
of B1 A2 , B2 A3 , and B3 A1 being X, Y, Z, respectively, such that XY Z is equilateral.
Let A3 and B3 map to A3 and B3 , respectively, under a spiral similarity of center O
(Fig. 12.6). Let also Y and Z be the midpoints of B2 A3 and B3 A1 . Then Y Y and
ZZ are midlines in the triangles B2 A3 A3 and A1 B3 B3 , and so Y Y is parallel to
A3 A3 and half its length, and ZZ is parallel to B3 B3 and half its length.
Theorem 2.25 shows that spiral similarities come in pairs, and so the spiral
similarity that maps OA3 B3 to OA3 B3 comes with a spiral similarity, also of center
12 A Synthesis 521
Fig. 12.7 Proof by B1

isometries that the triangle N
formed by midpoints is A2
equilateral
M
P
B2
T
A1
O
S
A3
R
B3
O, that maps OA3 A3 to OB3 B3 . At a closer look we see that the latter is in fact a
60◦ rotation. But then A3 A3 and B3 B3 are of equal lengths and form an angle of
60◦ . The same will be true about Y Y and ZZ . And the 60◦ rotation about X that
maps Y to Z will map Y to Z . Therefore XY Z is equilateral and the problem is
solved.
Second solution. Place the origin of the coordinate system at O. If a1 , a2 , a3 are
the complex coordinates of A1 , A2 , A3 , respectively, then since B1 , B2 , B3 are the
iπ iπ iπ
60◦ rotations of these points about O, their coordinates are a1 e 3 , a2 e 3 , a3 e 3 .
Let x, y, z be the coordinates of the midpoints X, Y, Z of B1 A2 , B2 A3 , and B3 A1 ,
respectively. Then
1 πi 1 πi 1 πi
x= (a1 e 3 + a2 ), y= (a2 e 3 + a3 ), z= (a3 e 3 + a1 ).
2 2 2
We want to show that y rotates to z about x by 60◦ , which amounts to checking that
πi
2(z − x) = 2(y − z)e 3 . This equality reads
πi πi πi πi πi
a1 (1 − e 3 ) − a2 + a3 e 3 = [−a1 e 3 + a2 (e 3 − 1) + a3 ]e 3 ,
πi 2π i
and this is an easy consequence of the equality e 3 − e 3 = 1.
Third solution. We explain this solution only for the case where A1 B1 A2 B2 A3 B3
is not a skew hexagon. Let M, N, P , Q, R, S be the midpoints of A1 B1 , B1 A2 ,
A2 B2 , B2 A3 , A3 B3 , B3 A1 , respectively. We have to show that the triangle NQS is
equilateral. We argue on Fig. 12.7.
The 60◦ rotation that maps A1 to B1 and A2 to B2 maps A1 A2 to B1 B2 , so these
two segments have the same length and form an angle of 60◦ . But MN and NP are
522 12 A Synthesis
midlines in the triangles B1 A1 A2 and A2 B1 B2 , so MN = NP and MNP = 120◦ .

Similarly P Q = QR and P QR = 120◦ and RS = SM and RSM = 120◦ .
We claim that the triangles MNS, P QN, and RSQ can be reflected over sides
as to form a dissection of the triangle NQS. If this were indeed true, then each
angle of this latter triangle is half of 120◦ , namely, 60◦ , and we are done. To prove
our claim, reflect M over NS to a point T . By the reflection, T N = MN = NP ,
and T S = MS = RS. It is sufficient to show that T Q = P Q = RQ, because
then, by equality of triangles, T would also be the reflection of P and R over the
corresponding sides.
Let us assume that this is not true. If T Q < P Q = RQ, then in the triangles
P QT and QRT , T P Q < QT P , and QRT < RT Q. Since the triangles
N T P and SRT are isosceles,
N P Q = NP T + T P Q < P T N + QT P = NT Q,
and similarly QRS < QT S. Then on the one hand by using the sum of the angles
of hexagon, we have
SMN + NP Q + QRS = 4 · 180◦ − MNP − P QR − RSM

= 720◦ − 3 · 120◦ = 360◦ .
and on the other hand
SMN + NP Q + QRS < NT S + P T N + ST Q = 360◦ ,
a contradiction. A similar reasoning rules out the situation T Q > P Q = RQ. The
problem is solved.
Remark The third solution can be adapted to the general configuration where
the hexagon can be skewed by using directed angles and taking into account the
situation where T falls outside the triangle NQS.
The idea of the solution is similar to those used for solving Problem 1.5 and
Problem 49.
The Averaging Principle implies that the midpoints of A1 B1 , A2 B2 , and A3 B3
form a triangle that is similar to both A1 A2 A3 and B1 B2 B3 .
Source The first solution was posted online by Alan Cooper.
227 An inversion of center A maps 1 to a line 1 that is parallel to t1 , and t2 and

2 are mapped to the circles t2 and 2 that are tangent to t1 (at A and D ) and to 1 ,
while t3 becomes a circle t3 that intersects t1 at A and D and intersects t2 and 2 a
second time at B and C , respectively, as shown in Fig. 12.8.
The new figure is invariant under reflection over the line E F , where E and
F are the images of E and F and hence are the intersections of the circles t2 and

2 . The line B C is therefore invariant under this reflection, so it is parallel to t1 .

12 A Synthesis 523
Γ1
C t2
E
t2 E F
B C B
Γ1
t3 F
Γ2 Γ2
A D t1 t1 A D
t3
Fig. 12.8 Configuration of circles and lines and its image under inversion
Consequently, its preimage under the inversion, which is the circumcircle of ABC,
is tangent to t1 .
Source Short list of the Balkan Mathematical Olympiad, 2003.
228 Examining Fig. 12.9, we notice that the Simson line of D with respect to the
triangle ABC is mapped by a homothety of center D and ratio 2 to a line through
the points E and F . Hence P , which is the intersection of BD and EF , must be
the image of the intersection of the diagonals through the homothety, so it is the
reflection of D over AC. It follows that A is on the perpendicular bisectors of ED
and DP , showing that it is the circumcenter of the triangle DEP . From here we
deduce that AP = AQ. Working with directed angles modulo 180◦ and using the
facts that AP D is isosceles and AQBD is cyclic, we have
AQB = ADB = DP A = BP A.
This combined with AP = AQ implies that Q is the reflection of P over AB. Thus
EQ is the reflection of DP over AB. Similarly RF is the reflection of DP over
BC. Therefore EQ = DP = F R, and we are done.
Source United States of America Junior Mathematical Olympiad, 2018, proposed
by Ray Li.
229 First solution. We argue on Fig. 12.10. The spiral similarity
√ sB of center B,
clockwise oriented angle of 30◦ , and ratio BA/BR = 2/ 3 maps R to A. Then,
the spiral similarity
√ sC of center C, clockwise oriented angle of 30◦ , and ratio
CQ/CA = 3/2 maps A further to Q. The composition sC ◦ sB is a rotation
of angle 60◦ which maps R to Q. Let us show that the fixed point of this rotation is
P.
Let P be the point in the exterior of the triangle ABC with the property that
P BC = P CB = 30◦ . Then P P B and P P C are 90◦ − 60◦ − 30◦ triangles,
524 12 A Synthesis
D
B P
R
A
Fig. 12.9 EQ is equal to RF
so sB (P ) = P and sC (P ) = P . Hence sC ◦ sB (P ) = P , as claimed. We conclude

that a 60◦ rotation about P maps R to Q, so P QR is equilateral.
Second solution. The homothety hB of center B and ratio 2 maps P to C. The
spiral similarity sA of center A, angle 60◦ , and ratio 1/2 maps C to Q. Then sA ◦
hB (P ) = Q. The composition sA ◦ hB is a spiral similarity of ratio 2 × 1/2 = 1 and
angle 60◦ , so it is a 60◦ rotation. Let us find its fixed point, which is also the center
of rotation.
We note that hB (R) is the reflection D of B over the line AN. The triangle
ADN is equilateral, so sA (D) = R. It follows that R is the fixed point of sA ◦ hB .
We conclude that P R = QR and QRP = 60◦ , which shows that triangle P QR
is equilateral.
Third solution. We use complex numbers. Let the triangle ABC be oriented
counterclockwise and denote by a, b, c the coordinates of A, B, C, respectively.
Then P has the coordinate p = b+c 2 .
12 A Synthesis 525
Fig. 12.10 Proof that P QR Q M

is equilateral
N R A
B P C
P‘
π
R is the image of A through a spiral similarity of center B, angle 6, and ratio
√
3/2, so its complex coordinate r satisfies
√
r −b 3 πi
= e6.
a−b 2
We compute
√ √
3 πi 3 πi
r= e 6 a+ 1− e 6 b.
2 2
Similarly, Q is the image of A through a spiral similarity of center C, angle − π6 ,

√
and ratio 3/2. So its complex coordinate is
√ √
3 − πi 3 − πi
q= e 6 a+ 1− e 6 c.
2 2
The condition that P QR is equilateral
r −p πi
=e3
q −p
translates to
√ √ √ √
3 πi 1 3 πi c πi 3 − πi 1 3 − πi b
e6a+ − e6 b− =e3 e 6a+ − e 6 c− .
2 2 2 2 2 2 2 2
√ √
And this is a straightforward computation using e± 6 =
πi πi
2 ±2i = 12 +
3 1 3
and e 3
2 i.
Source Romanian Mathematics Competition, proposed by C. Ottescu, 1981.
526 12 A Synthesis
230 First solution. We recognize two cross-ratios in the statement. The first cross-
ratio is in the hypothesis
AD AC
: = 1,
BD BC
the other in the question:
AB AC
: .
DB DC
Looking at the hypothesis, we should realize that much more is true. The fact that
C and D are on the same side of AB and ADB = 90◦ + ACB means that,
if we choose the appropriate orientation of the plane when putting on it complex
coordinates, then
a−d a−c π
arg − arg = .
b−d b−c 2
Consequently
a−d a−c π
: = 1 · e 2 i = i.
b−d b−c
Knowing this, let us apply a Möbius transformation that maps C to ∞. The

fact that Möbius transformations preserve the cross-ratio implies that the complex
coordinates of the images satisfy
a − d
= i.
b − d
This means that the triangle D A B formed by the images is right isosceles, with the
right angle at D . So the lines A D and B D form a right angle, which means that
their preimages, which are the circumcircles of ACD and BCD, are orthogonal.
This solves half of the problem. In the right isosceles triangle D A B , we have
a − b √ −π i
= 2e 4 ,
d − b
or, equivalently,
a − b a − ∞ √ − π i
: = 2e 4 .
d − b d − ∞
Again by the invariance of the cross-ratio,

12 A Synthesis 527
Fig. 12.11 Construction of C

B
A D
B
B
a − b a − c √ −π i
: = 2e 4 .
d −b d −c
Hence
AB AC √
: = 2,
DB DC
and this is the expression (written as a cross-ratio) that we had to compute.
Second solution. After this short and elegant solution, it seems pointless to look
for more. But it might be worth to observe that this problem can be solved not only
within the context of complex projective geometry but also in the complex affine
setting. Consider a spiral similarity of center A that maps C to D and let B be the
image of B (Fig. 12.11). Then the triangle ACB is mapped to ADB , so on the one
hand
BDB = ADB − ADB = ADB − ACB = 90◦
and on the other hand
B D AD BD
= = ,
BC AC BC
where for the last equality we used the hypothesis. Hence B D = BD. It follows
that the triangle DBB is right isosceles.
As spiral similarities come in pairs (Theorem 2.25), the triangles ACD and
ABB are mapped into each other by a spiral similarity; hence they are similar.
Therefore
√
AB BB BD 2
= = ,
AC CD CD
and hence
528 12 A Synthesis
Fig. 12.12 The circumcircles C

of ACD and BCD are
orthogonal
A
D
A
B
AB · CD √
= 2.
AC · BD
Let us also check the orthogonality of the two circles. In the concave quadrilateral
ADBC,
CAD + CBD = 360◦ − ACB − (360◦ − ADB) = 90◦ .
Next, consider the spiral similarity of center C that maps the circumcircle of
ACD to the circumcircle of CDB. By Problem 131, the image A of A is the second
intersection point of AD with the circumcircle of CDB (Fig. 12.12). The angle of
the spiral similarity is
ACA = 180◦ − CAD − CA D = 180◦ − CAD − CBD

= 180◦ − 90◦ = 90◦ .
And as the tangents at C to the two circles are mapped into each other by the spiral
similarity, they form a 90◦ angle, showing that the circles are orthogonal.
Remark So the right isosceles triangle that is at the core of the solution can be
produced not just by a projective transformation (the Möbius transformation) but
also by an affine transformation (the spiral similarity).
Source International Mathematical Olympiad, 1993, proposed by the Great Britain.
231 We assume that the radius of ω is smaller than that of ω, so that we are in the
situation from Fig. 12.13. Let T be the intersection of P P and QQ . Consider the
homothety of center T that maps ω to ω , and let A be the image of A. Because O
is mapped to O and M to M , M A O = MAO.
Now consider the inversion of center T and radius OP . By power-of-a-point,
this inversion maps ω into itself. By Theorem 3.2, the inversion maps O to M
(because T P O = T M P = 90◦ ), and it also maps A to A . It follows that
AA M O is cyclic, so M A O = M AO .
Combining the two equalities, we obtain MAO = M AO , and consequently
MAM = OAO , as desired.
12 A Synthesis 529
P
ω A
A
O M O M ω T
Fig. 12.13 Proof that MAM = OAO
P
O
H T
Q
S B M C
Fig. 12.14 Proof that AT is a median in triangle ABC

232 We will prove that the circumcenter T of the triangle H P Q lies on the median
from A. Without loss of generality we assume that AB < AC. The reasoning can
be followed on Fig. 12.14.
Note that since CH ⊥AB,
1
H QP = 90◦ − QAB = 90◦ − OAB = AOB = ACB,
2
and similarly H P Q = ABC. This shows that the triangles H P Q and ABC
are similar. The triangles have opposite orientation, so there is a map λ that is the
composition of a spiral similarity and a reflection over a line such that λ(H P Q) =
ABC.
Because
530 12 A Synthesis
Fig. 12.15 The proof that E

is angle bisector
K
K
L L
D
C
A B
AH P = 90◦ − H AC = ACB = H QP ,
the line AH is tangent to the circumcircle of H P Q. This means that λ(AH ) is

tangent to the circumcircle of ABC at A. So S = λ(A) lies at the intersection of
the tangent to the circumcircle of ABC at A with the line BC. Since λ(T ) = O,
λ(A) = S, λ(P ) = B,
OAT = P AT = BSO.
Let M be the intersection point of AT and BC. Then MSO = BSO =

T AO = MAO, so the quadrilateral AOMS is cyclic. This implies that
OMS = 180◦ − OAS = 180◦ − 90◦ = 90◦ ,
so OM⊥BC. Thus M is the midpoint of the side BC, so T belongs to the median
AM.
Remark If two figures are similar but with opposite orientation, then by reflecting
one of them over a line, we obtain a figure directly similar to the other, which can
now be mapped into the other by a spiral similarity.
Ukraine.
233 (a) Consider the homothety of center C and ratio 1/2, and let A , K , and L be
the images of A, K, L, respectively, as shown in Fig. 12.15. Then A is the midpoint
of AC, so it is the intersection of the diagonals of the parallelogram, while K and
L are the midpoints of the sides CK and CL of the triangles ECK and ECL.
Because these triangles are isosceles, K and L are the projections of E onto the
corresponding sides, so K L , which is the image of , is the Simson line of E with
respect to the triangle BCD. It follows that EA is perpendicular to BD, since A
belongs to the Simson line. We conclude that E is on the perpendicular bisector of
BD.
In the right triangles EA B and EL C,
12 A Synthesis 531
Fig. 12.16 The proof that E

BCDE is cyclic
K
D L
C
A B
1
A BE = DE= L CE,
2
so the triangles are directly similar. It follows that there is a spiral similarity of
center E that maps the triangle EA B to EL C. And as spiral similarities come
in pairs (Theorem 2.25), there is a spiral similarity that maps the triangle EA L
to EBC. The angle of this spiral similarity is A EB = DEB/2 = DCB/2
(EDB is isosceles and BCED is cyclic). So A L , as well as its preimage through
the homothety, make with BC an angle of DCB/2, and consequently they make
with AB an angle of DCB/2 = DAB/2. It follows that is the angle bisector
of DAB, and we are done.
(b) The converse is proved using a rotation. Let E be the circumcenter of CKL.
To prove that E is on the circumcircle of BCD amounts to showing that BED =
BCD. Let us prove this equality, following the argument on Fig. 12.16. The fact
that AD is parallel to BK and AB is parallel to CL implies that
BKA = KAD = BAK = CLK.
From here we obtain on the one hand that AB = CD = BK and on the other that
KC = LC. These two chords are equal in the circle of center E, and hence they are
mapped into each other by rotation ρ of center E. Then ρ maps the line BK to the
line CD, so the angle of rotation is BCD. On the other hand because the segments
KB and CD are equal and ρ(K) = C, it follows that ρ(B) = D. But then BED
is equal to the angle of rotation, which is BCD, and we are done.
Source The direct implication was given at the International Mathematical
Olympiad in 2007, being proposed by Luxembourg, the converse was published in
1987 in Kvant (Quantum) being proposed by Igor Fedorovich Sharygin.
234 First solution. This argument can be followed on Fig. 12.17. Let Q1 be the

midpoint of the arc BAC and let P1 be the midpoint of the arc BAD; let also K and
L be the midpoints of P P1 and QQ1 , respectively.
532 12 A Synthesis
D
P1
A M
C Q1
K
ω2
ω1 L
P
B Q
Fig. 12.17 Proof by spiral similarity that P MQ = 90◦
Consider the spiral similarity s1 of center B that maps ω1 to ω2 . By Problem 131

(b), s1 (C) = D. And since spiral similarities preserve midpoints of arcs, s1 (P ) =
P1 and s1 (Q1 ) = Q. So on the one hand, Problem 131 (b) implies that P P1 and
QQ1 pass through A. On the other hand, s1 maps the triangle P CQ1 to P1 DQ.
Applying the Averaging Principle (Theorem 2.18), we obtain that the triangle KML
is similar to any of the triangles P CQ1 and P1 DQ. In particular KML = 90◦ and
KM/ML = P C/CQ1 .
Let us compute (P P1 , Q1 Q). For this we use the fact that spiral similarities
come in pairs (Theorem 2.25), so s1 comes with a second spiral similarity s2 , also
of center B, satisfying s2 (P ) = Q1 and s2 (P1 ) = Q. The angle of s2 is P BQ1 =
90◦ , because P Q1 is a diameter. Hence (P P1 , QQ1 ) = 90◦ . But from this second
spiral similarity, we extract more information:
KP P P1 BP CP KM
= = = = .
LQ Q1 Q BQ1 CQ1 ML
And from the fact that
(KM, ML) = 90◦ = (P P1 , QQ1 ) = (KP , LQ)
we obtain that MKP = MLQ. It follows that the triangles MKP and MLQ are
similar, in fact directly similar, so they are mapped into each other by yet another
spiral similarity. Spiral similarities come in pairs, so KML is mapped to P MQ by
a fourth spiral similarity, showing that P MQ = KML = 90◦ .

Second solution. From the fact that P and Q are the midpoints of the arcs BC

and BD it follows that AP and AQ are the angle bisectors of the supplementary
angles BAC and BAD, so P AQ = 90◦ . The problem is solved if we show that
A, M, P , Q lie on a circle.
12 A Synthesis 533
M
D
A
C Q
P
B
Fig. 12.18 Proof by inversion that P MQ = 90◦
To prove this, consider an inversion about a circle centered at A, and use the
convention to denote the image of a point by adding a dash to its letter name (you can
see the useful part of the inverted configuration in Fig. 12.18). The problem reduces
to showing that M , P , Q are collinear. As M is the midpoint of the segment CD,
the points ∞, C, M, D form a harmonic division. As inversion preserves cross-
ratios that are real (Theorem 3.18) A, C , M , D also form a harmonic division
(A is the image of ∞ and you should always think of the points as appearing in this
circular order on the projective line R ∪ {∞}).
On the other hand, if in the triangle B C D the cevians B A, C Q , and D P
intersected at one point, then the intersection of P Q with C D would form
with A, C , and D a harmonic division (see the proof of Theorem 3.33) so this
intersection would be M , showing that M , P , Q are collinear. Thus we are left
with showing that B A, C Q , and C P are concurrent.
The lines AP and AQ are the angle bisectors of B AC and B AD , so by
the Bisector Theorem,
P C AC Q B AB
= and = .
P B AB Q D AD
It follows that
AD P C Q B AD AC AB
· · = · · = 1.
AC P B Q D AC AB AD
So by Ceva’s Theorem (Problem 116) the lines B A, C Q , D P are indeed

concurrent, and we are done.
Source Second solution by Flavian Georgescu.
235 First solution. (a) Let the point be A, let the triangle be BCD, and let its
orthocenter be H (Fig. 12.19). Construct the reflections A1 and A2 of A over BC
and DC, respectively. The composition of these reflections is a rotation about C by
534 12 A Synthesis
Fig. 12.19 Proof of Steiner’s D A

theorem
H
A2
B C
A1
2 BCD, which maps A1 to A2 . Notice that A1 belongs to the circumcircle of BH C

and A2 belongs to the circumcircle of CH D. Taking into account the homothety of
center A and ratio 2, we realize that the problem is solved if we show that A1 , A2 ,
and H are collinear. And this is true because the rotation maps the circumcircle of
H BC to the circumcircle of H AC. Interpreting the rotation as a particular case of
spiral similarity, we apply Problem 131 (b) to conclude that A1 , A2 and the point
of intersection of the two circles that is not the center of rotation are collinear. This
point of intersection is H , and we are done.
Second solution. Like in the first solution, we start by considering the reflections
σCB and σCD over lines CB and CD, respectively. Using the same homothety, we
reduce the problem to showing that A1 = σCB (A), A2 = σCD (A), and H are
collinear.
We switch to complex coordinates, place the origin of the coordinate system
at the circumcenter of BCD, and let the circumcircle of BCD be the unit circle.
The coordinates of A, B, C, D, H, A1 , A2 are a, b, c, d, h, a1 , a2 . As proved in the
solution to Problem 23, h = b + c + d.
In view of the formula (1.1) derived in the proof of Theorem 1.5 we choose the
parametric equations of lines
b−c
CB : z=c+ t, t ∈ R,
|b − c|
d −c
CD : z=c+ t, t ∈ R,
|d − c|
then apply (1.1) to obtain
(b − c)2 (b − c)2 b−c b−c

a1 = σCB (a) = a+c− c= a+c− c
(b − c)(b − c) (b − c)(b − c) b−c b−c
12 A Synthesis 535
b−c b−c
= a+c− c = −abc + c + b.
c−b c−b
bc bc
We have used the fact that bb = cc = dd = 1, as they lie on the unit circle.
Similarly
a2 = −adc + c + d.
To show that a1 , a2 , h are on a line, we have to show that (h − a1 )/(h − a2 ) ∈ R.

We compute
h − a1 d + abc (d + abc)(b + adc) (d + abc)(b + adc)

= = =
h − a2 b + adc (b + adc)(b + adc) (b + adc)(b + adc)
db + ac + bd + ac 2R(db + ac)
= = .
(b + adc)(b + adc) |b + adc|2
The latter is a real number and we are done.

(b) Let the quadrilateral be ABCD, and let H1 , H2 , H3 , H4 be the orthocenters
of the triangles BCD, CDA, DAB, and ABC. As we have seen in Problem 23, the
quadrilaterals ABCD and H1 H2 H3 H4 are mapped into each other by a reflection
over a point M that is simultaneously the midpoint of the segments AH1 , BH2 ,
CH3 , and DH4 . The conclusion now follows from Steiner’s Theorem.
Remark The second solution to (a) combines a synthetic observation (the reduction
of the problem to the collinearity of A1 , A2 , H ) with a complex number computa-
tion.
236 (a) The key observation is that the triangles F AB and F DC are similar
because they have equal angles, which is a consequence of the fact that the
quadrilateral is cyclic. But they are not directly similar, so we cannot use homothety
or spiral similarity. This can be fixed, by composing with the reflection over a line.
The ratios from the statement suggest transformations that map AB and CD into
each other, and we will use two transformations: f that is the composition of the
homothety of center E and ratio AB/CD with the reflection over the angle bisector
of AEB and g that is the composition of homothety of center E and ratio CD/AB
with the same reflection. Let G = f (F ) and H = g(F ) (Fig. 12.20). Then G, H
are on the reflection of EF over the angle bisector of AEB, and

AB CD

GH = |EG − EH | = EF · − .
CD AB
The problem reduces to showing that MN = GH /2. Examining the figure, we

notice that f maps the triangle F CD to the triangle GAB, so the triangle GAB is
similar to the triangle F BA. But these two triangles share the side AB, so they are
536 12 A Synthesis
G D
M F
N H
E
C
B
Fig. 12.20 Construction of F and G
Fig. 12.21 F AD and F BC

are directly similar A
F
Q
D
M
F
N
E
P C
B
congruent. This implies that GAF B is a parallelogram, so M is the midpoint of the

diagonal F G. For a similar reason N is the midpoint of the segment F H , so MN is
midline in the triangle F GH . Hence MN = GH /2, which proves (a).
(b) First solution. We use again reflection over a line, this time composed with
a spiral similarity, and now we focus on the similar triangles F AD and F BC. The
presence of midpoints hints to the use of the Averaging Principle (Theorem 2.18).
Let F be the reflection of F over AD (Fig. 12.21). Now F AD and F BC are
directly similar, so there is a spiral similarity that maps them into each other. Note
that Q is the midpoint of F F , while M, N are the midpoints of BA and CD,
respectively. By the Averaging Principle, the triangle QMN is similar to both F AD
and F BC.
By using instead the reflection over BD, we infer that the triangle P MN is
similar to both F AD and F BC. Consequently, the triangles QMN and P MN
are similar, and since they share the side MN, they are congruent. They form a kite,
so P Q is perpendicular to MN, which proves (b).
Second solution. We consider the system of coordinates inspired by a homothety
and a reflection used in Problem 4.1 from the introduction to this chapter (with F at
the origin). The equation of the line AD is
12 A Synthesis 537

1 z z

1 1 1 = 0,

1 −kω −kω
that is
(1 + kω)z − (1 + kω)z + k(ω − ω) = 0.
The projection of the origin onto this line is obtained by intersecting this line with
the one orthogonal to it passing through the origin, whose equation is (1 + kω)z +
(1 + kω)z = 0. Requiring both equations to be satisfied simultaneously yields a
system of two equations in the unknowns z and z, with the solution the coordinate
of Q:
k(ω − ω)
q= .
2(1 + kω)
Similarly, the equation of the line BC is (k + ω)z − (k + ω)z + ka(ω − ω) = 0, and

so the projection P of F onto the line BC has the coordinate
ka(ω − ω) kaω(ω − ω)
p= = .
2(k + ω) 2(1 + kω)
The coordinates of M and N are, respectively,
1 + aω −ka − kω
m= , n= .
2 2
Then
k(ω − ω) aω + ak + 1 + kω k(ω − ω)
p−q = · = (m − n).
4 (1 + kω)(1 + kω) 2(1 + kω)(1 + kω)
In the fraction the numerator is imaginary, while the denominator is real, so the
fraction itself is imaginary. That is (p − q)/(m − n) is imaginary, proving that P Q
is perpendicular to MN .
Remark This solution presents three ways in which, in a cyclic quadrilateral, the
two similar triangles formed by diagonals and sides can be mapped into each
other:
(i) by a direct homothety with center at the intersection of two sides followed by
a reflection over the bisector of the angle formed by those two sides (solution
to (a));
(ii) by the reflection over a side followed by a spiral similarity (first solution to
(b));
538 12 A Synthesis
(iii) by an inverse homothety centered at the intersection of the diagonals followed

by a reflection over the bisector of the angle formed by the diagonals (second
solution to (b)).
Source Part (a) was given at the Bulgarian Mathematical Olympiad in 1997. Part
(b) was given at Team Selection Test in the United States of America in 2000. First
solution to (b) from T. Andreescu, M. Rolínek, J. Tkadlec, Geometry Problems from
the AwesomeMath Year-Round Program, XYZ Press, 2013.
237 The solution can be followed on Fig. 12.22. The triangle DAI rotates by 90◦
to the triangle DLC, so AI ⊥LC, and for similar reasons BK⊥F D, CE⊥H A, and
DG⊥J B. The quadrilaterals P1 Q1 R1 S1 and P2 Q2 R2 S2 have orthogonal sides; if
C
K
Q1 P2
L S2 A Q2
T
D B
R1
I P1 O
B D
S1 R2 A
H
C
Fig. 12.22 Finding the center of the rotation that maps P1 Q1 R1 S1 to P2 Q2 R2 S2

12 A Synthesis 539
they are to be equal, they must map into each other by some 90◦ rotation ρ. We want
to identify the center O of ρ.
We expect that ρ(AI ) = LC, ρ(BK) = F D, ρ(CE) = H A, and ρ(DG) =
J B. So the center of rotation must be at the intersection of the bisectors of the
angles formed by AI and LC, BK and F D, CE and H A, and DG and J B. If T is
the intersection of AI with CL, then T D is the angle bisector of AT C (as DAI
rotates to DLC), and so T , D, O have to be collinear. A similar situation happens at
each vertex of ABCD, the four bisectors are the lines that connect the intersection
points to the corresponding vertices of ABCD, and they should pass through O.
But are the four angle bisectors concurrent? This is where the orthogonality
of the diagonals of ABCD comes into play. Construct two squares AB CD and
A BC D whose orientations are opposite to that of ABCD (Fig. 12.22). Because
AC is perpendicular to BD, the squares have parallel sides, and as a consequence
of Problem 84, they are homothetic. This implies that AA , BB , CC , DD intersect
at a certain point O. As the figure suggests, O is the center of the rotation, and we
will prove it.
The quadrilateral AD CT is cyclic, having two opposite right angles, which
implies that CT D = CAD = 45◦ , and consequently the points T , D, D are
collinear. It follows that O lies on the angle bisector of AT C. The same is true for
the other angles, and so there is a rotation ρ about O by 90◦ that takes those pairs of
lines into each other, as desired. This rotation ρ maps P2 Q2 R2 S2 into P1 Q1 R1 S1 ,
and we are done.
Source Short list of the 36th International Mathematical Olympiad, 1995, proposed
by Germany, solution from D. Djukić, V. Janković, I. Matić, N. Petrović, The
IMO Compendium, A Collection of Problems Suggested for the International
Mathematical Olympiads: 1959–2004, Springer, 2006.
238 First solution. Consider a spiral similarity that takes the triangle ABO to the
triangle AOQ . We will prove that Q = Q , with the help of Fig. 12.23.
We have Q AO = OAB and Q OA = OBA. But DOA = OAB +
OBA (by the Exterior Angle Theorem), hence Q AO = Q OD. Also, because
of the spiral similarity, AQ /AO = Q O/OB. But the later equals Q O/OD,
and we conclude that the triangles AQ O and OQ D are similar, so Q DO =
Q OA = ODC. It follows that Q = Q and hence Q is the center of the spiral
similarity that maps triangle QAO to triangle QOD. In particular it is the center of
the spiral similarity that maps AO to OD.
Second solution. Let B be the reflection of B over the line AC (Fig. 12.24).
Note first that AC is midline in the triangle BDB , so AC is parallel to DB . Also,
because AB is the reflection of AB over AC, it follows that Q ∈ AB . And because
reflection preserves angles, we have that
AB O = ABO = BDC = BDQ.
Consequently, the quadrilateral DQOB is cyclic, hence

540 12 A Synthesis
Fig. 12.23 Q is the center of D

the spiral similarity, proved C
by. . . spiral similarity
Q=Q
O
A B
Fig. 12.24 Q is the center of

B
the spiral similarity, proved
by reflection D
C
Q
O
A B
DOQ = DB Q = B AC = QAO.
Applying the Exterior Angle Theorem in the triangle AOB gives AOD =
ABO + BAO, hence
AOQ = AOD − DOQ = ABO = BDQ.
It follows that the triangles QAO and QOD are similar (having two pairs of equal
angles), and hence they are mapped into each other by a spiral similarity of center
Q. This spiral similarity maps AO to OD, and we are done.
Third solution. This approach uses symmedians; thus it is logically related to
isometries. Let Q be the intersection of the circle through A tangent to BD at O
with the circle through D tangent to AC at O (Fig. 12.25). Problem 132 shows that
OQ is symmedian in the triangle OAD. Note that if M is the midpoint of AD,
then OM is parallel to AB. Thus
BAO = MOA = Q OD = Q AO,
where the last equality follows from inscribed angles. By a similar argument
CDO = MOD = Q OA = Q DO.
The equalities BAO = Q AO and CDO = Q DO imply Q = Q . The

equalities QOD = QAO and QOA = QDO imply that the triangles QAO
12 A Synthesis 541
Fig. 12.25 Proof that Q is

the center of the spiral
similarity using symmedians
D
C
Q
M
O
A B
and QOD are similar, so they are mapped into each other by a spiral similarity
centered at Q. The conclusion follows.
Remark The third solution is probably the most natural given the discussion in
Sect. 2.4.3.
Source The second and third solutions were found by the students of the Brazilian
Mathematical Olympiad Program.
239 First solution. Let N be the midpoint of AD. The problem asks us to show
that MN is perpendicular to BC. We argue on Fig. 12.26.
Let A be the point that is diametrically opposite to A in the circle of center M
and radius MA. Then, in the triangle AA D, the segment MN is midline, so it is
parallel to A D. It remains to prove that A D is perpendicular to BC.
We begin by showing that the triangles C CA and C MD are similar. We
have CC A = MC D = 90◦ , the first because AA is diameter, the second
because MC is radius and C D is tangent. Then C MD is half of C MB and

so it measures half of the arc B C . Thus C MD = BAC. Also, ABA C is
a parallelogram because M is the midpoint of both BC and AA , so C CA =
BAC. This shows that C CA = C MD. It follows that the triangles C CA and
C MD have two pairs of equal angles and so they are similar, as claimed. From here
we deduce that there is a spiral similarity of center C mapping the triangle C CA
to the triangle C MD. By Theorem 2.25 (spiral similarities come in pairs), there is
a spiral similarity of center C that maps the triangle C CM to the triangle C A D.
Because (MC , C D) = MC D = 90◦ , the angle of the spiral similarity is 90◦ .
It follows that A D is perpendicular to CM, and we are done.
542 12 A Synthesis
Fig. 12.26 Proof that A D is

perpendicular to BC C
A
M D
N
A
B
B
C
F
A
D
R M
G
E B
Fig. 12.27 Proof that G is on the perpendicular to BC
Second solution. Consider the homothety of center A and ratio 1/2. Let E, F, G
be the images of B , C , D, respectively. Note that G is the intersection of the
tangents at E and F to the circumcircle of triangle AEF , while M, the image of A ,
is on this circumcircle. It suffices to prove that G is on the perpendicular bisector of
the side BC. We argue on Fig. 12.27.
Let R be on the circumcircle of AEF such that AR is parallel to BC. Then the
pencil AB, AC, AM, AR is harmonic, because AR is parallel to BC and AM runs
12 A Synthesis 543
through the midpoint of the segment BC, so using the line BC as the secant to
this pencil, we compute MB/MC : ∞B/∞C = −1. From Lemma 4.3 it follows
that the cyclic quadrilateral EMF R is harmonic. So the tangents at E and F to the
circumcircle intersect on MR, that is, G ∈ MR (see Sect. 1.2.1). But since AM is a
diameter in the circumcircle of AEF , ARM = 90◦ . We deduce that MR = MG
is orthogonal to BC, and we are done.
Source The Competition of the “Simion Stoilow” Institute of Mathematics of the
Romanian Academy, 2014, second solution by Ştefan Rareş Tudose.
240 The centers of three circles passing through the same point I and not tangent
to one another are collinear if and only if the circles have a second common point.
It is therefore enough to show that the circumcircles of Ai Bi I have a second point
of intersection.
To simplify the problem, apply an inversion of center I . With the convention of
denoting the image of an object by adding a dash to its name, Ci is the line Bi+1
B
i+2

while the image of the line Ai+1 Ai+2 is the circumcircle of I Ai+1 Ai+2 , which is
tangent to Bi Bi+1
and Bi Bi+2
(the configuration before and after inversion is shown
in Fig. 12.28). As I is at equal distance from the three sides of A1 A2 A3 , these three
circles have equal radii. So their centers O1 , O2 , O3 form a triangle that has sides
parallel to, and therefore is homothetic to, B1 B2 B3 .
The point I is the circumcenter of O1 O2 O3 , so its reflections over the sides
of this triangle, which are A1 , A2 , A3 , form a triangle that is homothetic to the
median triangle of O1 O2 O3 and hence to O1 O2 O3 itself. Composing homotheties
we conclude that the triangles A1 A2 A3 and B1 B2 B3 are homothetic. So the lines
A1 B1 , A2 B2 , A3 B3 intersect at one point, which is the image through the inversion
of the second intersection point of the circumcircles of A1 B1 I , A2 B2 I , A3 B3 I . The
problem is solved.
Remark The inverted figure is the configuration from Tzitzeica’s Five-Lei Coin
Problem (Theorem 3.38). The observation that A1 A2 A3 is homothetic to the median
B3 B2
A1
A1
O3
O2
I
A3
B3 A2
B2
I O1
B1
A2 A3 B1
Fig. 12.28 The image of A1 A2 A3 under the inversion centered at I

544 12 A Synthesis
G1 G
O
C
B
E
F1 L D F
Fig. 12.29 Proof that CF intersects at the reflection of G over AB
triangle of O1 O2 O3 and so it is mapped into O1 O2 O3 by a homothety of ratio −1

gives another solution to Tzitzeica’s problem.
the United States of America.
241 Let O be the center of and let L be the intersection point of the lines AB and
(Fig. 12.29). Then AEF = 90◦ because AB is diameter, so AEF = ALF .
This shows that the quadrilateral AELF is cyclic.
Also OED = 90◦ because ED is tangent to , and so OED = OLD =
90◦ ,
showing that OELD is cyclic. Using the cyclic quadrilaterals OELD and
AELF , we can write
ODE = OLE = ALE = AF E.
We deduce that the right triangles EAF and EOD are similar, in fact directly
similar, so there is a spiral similarity that maps them into each other. As spiral
similarities come in pairs (Theorem 2.25), there is a spiral similarity that maps
the triangle EOA into EDF . But the first of these two triangles is isosceles
(OA = OE), so the second is isosceles, too. Hence DE = DF .
Also from the similarity of EAF and EOD, we obtain

EOD = EAF = EAG =EBG /2 = EOG/2.
Hence EOD = GOD, showing that G is the reflection of E over OD (because

is invariant under this reflection, and reflection preserves intersections). So DG
is the other tangent from D to the circle, DG = DE = DF , and hence DF G =
DGF .
12 A Synthesis 545
Let G1 be the other intersection point of the line CF with , and let F1 be the
intersection of the line AG1 with . Consider the inversion χ of center A that maps
to . Then χ (C) = D, χ (F ) = G, χ (G1 ) = F1 , so the image of the line CF
is a circle passing through A and also through the points D, G, F1 . It follows that
AF1 DG is cyclic, so AF1 D = DGF . But the latter is equal to DF G. Hence
F1 is the reflection of F over the line through A perpendicular to , and this line is
AB. As is invariant under this reflection, the intersections G and G1 of AF and
AF1 with are images of each other under the reflection over AB, and the problem
is solved.
Remark The fact that the diameter AB is perpendicular to tells that there is an
inversion of center A that maps to . The collinearity of C, F, G1 is equivalent to
the concyclicity of their images and the center of inversion.
Source Moldovan Team Selection Test for the International Mathematical
Olympiad, 2005, solution by Liubomir Chiriac.
242 The situation is described in Fig. 12.30. We have three circles passing through
Q that we are supposed to show intersect a second time. Certainly it is easier to
check that three lines intersect, so it reasonable to use an inversion of center Q. But
that inversion would alter too much of the original figure. We would be better off
with an “inversion” that preserves most√of the figure, especially one that keeps P and
R in place. For this reason we use the bc inversion φQ determined by the triangle
A
R Q
I
E B P C
Fig. 12.30 The circumcircles of QEP , QDR, and QBI

546 12 A Synthesis
Fig. 12.31 The images

√ of
B, D, E through the bc
inversion D
Q
E
P
R M
QRP and the vertex Q. By Proposition 3.40 (v), I = φQ (I ) is the reflection of Q

over P R.
Let us find B = φQ (B) with the help of Fig. 12.31. First notice that B is at the
intersection of the tangents at P and R to the circumcircle of the triangle P QR, so,
as a consequence of Theorem 1.22, QB is symmedian in this triangle. Consequently,
B is on the median QM from Q.
Also, B is at the intersection of the circles through Q that are tangent at P and
R to P R, as these circles are the images of the sides AB and BC. If the line QB
meets the circumcircle of P QR again at N, then by power-of-a-point MN · MQ =
MP ·MR = MP 2 . And by writing the power of M with respect to the circle through
Q tangent at P to P R, MP 2 = MB · MQ shows that MB = MN, that is, B
and N are mapped into each other by a reflection over M. But then B is on the
circumcircle of I P R, because this circumcircle is the reflection of the circumcircle

of P QR both over P R and over M. The arc NP of the circumcircle of P QR and

the arc B R of the circumcircle of I P R correspond through the reflection so are
equal, and by using this, the fact that QP R and I P R are mapped into each other
by a reflection over P R, and the fact that QN is median in QP R, we deduce that
I B is symmedian in I P R.
The points D = φQ (D) and E = φQ (E) are easy to locate; they are the
intersections of the circles through Q tangent to P R at P and R with the lines RQ
and P Q, respectively. If we examine carefully Fig. 12.31, we notice that the lines
12 A Synthesis 547
P D and RE seem tangent to the circumcircle of P I R. This would be the case if in

the original figure the circumcircle of P I R were tangent to the circumcircles of both
P EQ and RDQ. And this can be checked easily by angle chasing on Fig. 12.30:
the angle formed by the tangent to the circumcircle of P I R at P with P R is P I R,
so the angle it makes with P E is P I R − BP R = P I R/2. But the angle made
by the tangent to the circumcircle of P QE at P with P E is equal to P QR, which
is P I R/2. So the two tangents make the same angle with P E; thus they coincide,
and the circles are tangent. The same argument works for checking the tangency of
the circumcircles of P I R and QRD.
By Theorem 1.22, the symmedian I B passes through the intersection point of
the tangents P E and RD . The preimages of the three lines, which are the three
circles in question, have therefore a second intersection point.
Remark The property remains true if we replace the incircle with the excircle
corresponding to the vertex B and I with the center Ib of this excircle.
243 First solution. Add to the original configuration the point B where and t
intersect the second time. Then take an inversion of center R, which solves the
problem through a “position argument.” To see how this works, with the usual
convention that X is the image of X, let us examine the inverted configuration
as depicted in Fig. 12.32.
First, since in the original configuration t is tangent to , and so they form
an angle of 0◦ , in the inverted figure the line RA and the line through J , S , K
also form an angle of 0◦ , they are parallel. This means that A B S J is a trapezoid
A
R
B
A
R
J
T
S
S
K
J
K
T
Fig. 12.32 Proof by inversion that KT is tangent to

548 12 A Synthesis
(A B ||S J ), but this trapezoid is inscribed in the circle that is the image of , and
so it must be an isosceles trapezoid.
Similarly RA J K is a trapezoid (RA ||J K ), and since line the line through
A, J, K is mapped by the inversion to a circle passing through R, the trapezoid
RA J K is also cyclic, hence isosceles.
Joining the two isosceles trapezoids, we obtain the parallelogram RB S K .
Because S is the midpoint of RT , T is the midpoint of RS and thus also the
midpoint of the other diagonal, B K , of the parallelogram. This means that a
reflection over T maps the triangle T B S to the triangle T K R, and so it maps
the circumcircle of the first to the circumcircle of the second. In particular the two
circumcircles are tangent. In the original figure, the preimage of the circumcircle
of T B S is the circumcircle of T BS, which is , while the preimage of the
circumcircle of T K R is the line T K. So T K and are tangent, and we are done.
Second solution. An approach based entirely on a reflection is also possible and
can be followed on Fig. 12.33. Chasing angles in the cyclic quadrilaterals RJ SK
and AJ ST , we obtain
KRS = KJ S = AT S,
which implies that RK is parallel to AT . Now reflect A over S to obtain the point
A . Then ARA T is a parallelogram with center S, RA is parallel to AT , so K is
on the line RA .
From
A
R
Fig. 12.33 Proof by reflection that KT is tangent to

12 A Synthesis 549

ST A = SRA =SJ R /2 = SKR
we obtain that the points S, K, A , T are concyclic. This implies that
ST K = SA K = SAT ,
which shows that KT is tangent to at T .

Remark If we invert with respect to a circle centered at S instead, we obtain an
analogous version of the problem. Another version of the second solution can be
obtained by reflecting K over S. The second approach shows how a geometric
transformation can contribute to a solution without necessarily dominating it.
Source International Mathematical Olympiad, 2017, proposed by Charles Leytem,
Luxembourg.
244 First, the fact that the lines AA , BB , and CC intersect at one point is a
classical fact; they are the radical axes of the circles ωA , ωB , ωC , so they intersect
at the radical center. The difficulty lies in identifying this radical center and showing
that it lies on I O.
The solution can be followed on Fig. 12.34. Let γ be the incircle of the
triangle ABC and let A1 , B1 , C1 be its contact points with the sides BC, CA, AB,
respectively. Let Xa be the contact point of the circles γ and ωA . Let MA be the
point where XA A1 intersects the circumcircle. Then, as seen in Problem 180 (a)

and its solution, MA is the midpoint of the arc BC not containing XA , and A1

AA , BB , CC meet on OI MC
A
B XA MB
C1
C
O B1
I K
XC
XB
B A1 C
MA
550 12 A Synthesis
is mapped to MA by the inversion of center MA and radius MA B. The equality

MA B 2 = MA A1 · MA XA shows that MA is on the radical axis B of (the degenerate
circle) B and γ . A similar argument shows that MA is on the radical axis C of C
and γ .
Define the points XB , XC , MB , MC and the line A similarly. Note that A =
MB MC , B = MA MC , and C = MA MB . Also, the fact that A is the radical
axis of A and γ implies that A is perpendicular to AI and hence parallel to B1 C1 .
Similarly B is parallel to A1 C1 and C is parallel to A1 B1 . Consequently, there is
a homothety h that maps the triangle A1 B1 C1 to MA MB MC ; let K be the center of
this homothety. Then MA A1 , MB B1 , and MC C1 meet at K.
Since the points A1 , B1 , XA , XB are concyclic, A1 B1 and XA XB are antiparallel.
And because A1 B1 and MA MB are parallel, MA MB and XA XB are antiparallel; thus
XA , XB , MA , MB are concyclic. Consequently MA K ·KXA = MB K ·KXB (which
is the power of K with respect to the circle passing through these points). But, by
symmetry, these products are also equal to MC K · KXC . The equality
MA K · KXA = MB K · KXB = MC K · KXC
shows that K is the radical center of ωA , ωB , ωC . We have identified the intersection

of AA , BB , CC !
To prove that K ∈ I O, consider the circular transformation
√ χ that is the
composition of the inversion with center K and radius MA K · KXA with the
reflection over K (the inversion of negative power). The circles ωA , ωB , ωC
are invariant under this transformation, from which it follows that χ (γ ) is the
circumcircle of MA MB MC . This circle must be tangent to ωA , ωB , and ωC
because circular transformations preserve tangencies (Theorem 3.21). Let O be the
circumcenter of MA MB MC . Because this circumcircle is tangent to ωA , ωB , ωC ,
the lines OMA , OMB , OMC are perpendicular to the tangents at MA , MB , MC
to ωA , ωB , ωC , respectively, and since MA , MB , MC are the midpoints, it follows
that OMA , OMB , OMC are perpendicular bisectors of BC, AC, AB, respectively.
Consequently O = O, the circumcenter of ABC. Finally, by Theorem 3.5, the
center of inversion, K, the center I of γ and the center O of χ (γ ) are collinear (this
is true even for inversion of negative power). The problem is solved.
Source Romanian Master of Mathematics, 2012, proposed by Fedor Ivlev.
245 First solution. Let E and F be the intersections of 2 with AN and BN ,
respectively (Fig. 12.35). The point A is on the radical axis of 1 and 2 , so its power
√ to AC · AM = AE · AN.
with respect to the two circles is the same and is equal
Consider the inversion χ of center A and radius AE · AN. Then χ (M) = C,
χ (N ) = E, and since χ ( ) is a line, it must be the line CE. Moreover, χ ( 1 ) = 1
and χ ( 2 ) = 2 , and since is tangent to both 1 and 2 , so is χ ( ) = CE. Using
an inversion of center B, we conclude that DF is also tangent to both 1 and 2 .
Denote by P the intersection of the common tangents CE and DF of 1 and
2 , and let O2 be the center of 2 . An isosceles triangle arises in the configuration
12 A Synthesis 551
Fig. 12.35 Proof by

inversion that CD is tangent Γ
to 2
Γ1 A
Γ2 E
M O2
D
F
N
B P
P CD, and we claim that O2 is its incenter. Indeed, the circle 1 is tangent to P C

and P D; thus P C reflects to P D over P O2 , and so CO2 =DO2 . Using inscribed
angles in this circle, we have

P CO2 =CO2 /2 =DO2 /2 = DCO2 ,
showing that CO2 is the angle bisector of P CD. Similarly, DO2 is angle bisector
of P DC. The claim is proved. Because 2 is tangent to P C and P D and it is
centered at the incenter O2 , it follows that 2 is the incircle of the triangle P CD;
hence CD is tangent to 2 , as desired.
Second solution. The second solution can be followed on Fig. 12.36. We denote
by O, O1 , O2 the centers of , 1 , 2 , respectively, by X and Y the intersections
of AB with the two circles, with X closer to A, and by R the intersection of O1 O2
with XY . We also let r, r1 , r2 be the radii of , 1 , 2 , respectively. Also, given a
point T and a line , we denote by d(T , ) the distance from T to . It suffices to
show that the d(O2 , CD) = r2 .
Consider the homothety of center M and ratio r/r2 takes 1 to . This homothety
maps O1 to O, and so M, O, O1 are collinear. It also maps C to A and D to B, so
CD is parallel to AB and d(M, CD)/d(M, AB) = r1 /r. Consequently
r1
d(C, AB) = d(M, AB) − d(M, CD) = d(M, AB) − d(M, AB)
r
r − r1
= d(M, AB).
r
But the distance from M to AB is O1 R = O1 O2 −O2 R plus the distance from M to
the parallel through O1 to AB. The latter is OM cos OO1 O2 = r1 cos OO1 O2
(since M, O1 , O are collinear). It follows that
552 12 A Synthesis
Fig. 12.36 Proof by

homothety that CD is tangent
to 2
A
C
O
O1
X
R
O2
M D
Y N
B
r − r1
d(O2 , CD) = d(O2 , AB) + d(C, AB) = O2 R + d(M, AB)
r
r − r1
= O2 R + |O1 O2 − O2 R + r1 cos OO1 O2 |.
r
We can compute the value of the cosine that appears in this expression using the
Law of Cosines in the triangle OO1 O2 . Because OO1 = r − r1 , OO2 = r − r2 ,
and O1 O2 = r1 , we have
r12 + (r − r1 )2 − (r − r2 )2 2r 2 − 2rr1 + 2rr2 − r22

cos OO1 O2 = = 1 .
2r1 (r − r1 ) 2r1 (r − r1 )
Also, in the triangle XO1 O2 , we have O1 X = O1 O2 = r1 , and O2 X = r2 , so,

using the Law of Cosines, we can compute
r12 + r22 − r12 r2

O2 R = O2 X cos O1 O2 X = r2 = 2 .
2r1 r2 2r1
We therefore have

r2 r − r1 r22 2r13 − 2rr12 + 2rr1 r2 − r1 r22

d(O2 , CD) = 2 + r1 − +
2r1 r 2r1 2r1 (r − r1 )
r22 r − r1 2r1 r2 r − rr22 r2 2r2 r1 − r22

= + · = 2 + = r2 .
2r1 r 2r1 (r − r1 ) 2r1 2r1

Remark The first solution relies on the following fact, which can prove useful in
other situations: if ABC is an isosceles triangle, and ω is a circle tangent to AB
12 A Synthesis 553
and AC at B and C, respectively, then ω passes through the incenter of the triangle
ABC.
Source International Mathematical Olympiad, 1999, proposed by Russia, first solu-
tion communicated by Leandro Maia, second solution from D. Djukić, V. Janković,
I. Matić, N. Petrović, The IMO Compendium, Springer, 2006.
246 We should first locate O1 and O2 in a more practical way. To this end, let La

and Lb be the midpoints of the arcs BC and AC that do not contain the other vertex.
Let γa be the circle with center La that passes through B and C; define γb in the
same fashion. Then O1 is the other intersection point of XLb and γb , and O2 is the
other intersection point of XLa and γa (Fig. 12.37). Also, notice that γa and γb meet
at incenter I of ABC, and, by Theorem 2.22, the intersection T of the circumcircles
of XO1 O2 and ABC is the center of the spiral similarity that takes O1 to O2 and
Lb to La . As spiral similarities come in pairs, this is also the center of the spiral
similarity that maps O1 to Lb and O2 to La .
√There are far too many circles in the diagram; let us turn them into lines: perform
a bc inversion determined by the triangle ABC and the vertex C. Then I is taken
to the center Ic of the C-excircle, and γa , which passes through C, I , and B, is taken
to the line AIc . Therefore the center La of γa is taken to the reflection of C across the
line γa (this is a consequence of Proposition 3.7). Since La lies on the circumcircle
of ABC, La is on its image , which is the line BC (see Fig. 12.38). Similar
properties hold for Lb .
La
Lb
O1 O2
A B
T X
Fig. 12.37 Spiral similarity?

554 12 A Synthesis
D
La A B Lb
Ic
Fig. 12.38 Inverting the diagram
We infer that AIc is the perpendicular bisector of CLa and BIc is the perpendic-
ular bisector of CLb . This means that Ic is the circumcenter of the triangle CLa Lb ,
which means that its projection D onto BC satisfies D La = D Lb , in other words,
in complex coordinates,
la − d la − ∞
: = −1.
lb − d lb − ∞
If D is the preimage of D , then by using the invariance of real valued cross-ratio

under inversion, we find that
la − d la − c
: = −1,
lb − d lb − c
which can be rewritten as

la − d la − c
=− .
lb − d lb − c
By inscribed angles
CLb O1 + CLa O2 = CLb X + CLa X = 180◦ ,
and combining this with the fact that CLa = O2 La and CLb = O2 Lb , we can write
la − c la − o2
=− .
lb − c lb − o1
12 A Synthesis 555
Consequently
la − d la − o2 d − la d − lb
= ⇐⇒ = .
lb − d lb − o1 o2 − la o1 − lb
We interpret this equality as saying that the spiral similarity of center La that maps
O2 to D and the spiral similarity of center Lb that maps O1 to D have the same
ratio and angle. But this means that the triangles DLa O2 and DLb O1 are directly
similar. This implies that D is the center of the spiral similarity that takes O2 to La
and O1 to Lb . This means that D = T , and we are done, because D is fixed when
X varies.
Remark The point D = T is the tangency point between and the C-mixtilinear
incircle. The solution seems to be a bit tricky because it looks like we have guessed
the point D; however, midpoints in inversion usually lead to equal ratios, which
interact well with spiral similarities. So in this sense, the point D is quite natural.
You can try to rewrite in geometric terms our complex number manipulations, in
order to see that they are less mysterious than they look.
Russia.
247 We invert about the incircle ω, because this circle is the focus of the problem,
and use dashes for the inverses. Let ω touch the sides BC, CA, and AB at the points
D, E, and F , respectively. The proof to Euler’s Theorem (Theorem 3.36) shows that
the inverses of the vertices of the triangle ABC with respect to the incircle are the
midpoints of the sides of the triangle DEF and that the inverse of the circumcircle
of ABC is the nine-point circle of DEF (Fig. 12.39).
Notice that ω is the M-excircle of the triangle MK1 K2 . Let ω touch MK1 and
MK2 at T1 and T2 , respectively. Then, by Proposition 3.3, K1 is the midpoint of
DT1 , K2 is the midpoint of DT2 , and M is the midpoint of T1 T2 . That is, the
Fig. 12.39 Nine-point circles A
E
A
F
H I T2
K
T1 K2
B K1 D C
K M
556 12 A Synthesis
circumcircle of MK1 K2 is mapped to the nine-point circle γ of DT1 T2 . Notice that

the circumcircle of DT1 T2 is ω. It is tempting to consider the homothety between
the two circles, but we will divide this homothety into steps.
First, consider the homothety with center D and ratio 2, which maps γ to the
circle γ ∗ that passes through T1 and T2 and has the same radius as ω. This circle
does not coincide with ω because D is not on γ . So γ ∗ is the reflection of ω across
T1 T2 . This means that γ ∗ passes through the orthocenter H of DEF . So γ passes
through the midpoint K of DH , which lies on the nine-point circle of DEF , that
is, on the inverse of the circumcircle of ABC. Now, K is diametrically opposite
to A on this nine-point circle, and by Problem 204, K is the tangency point of the
circumcircle of ABC with the A-mixtilinear incircle. Done.
Source Taiwanese Team Selection Test for the International Mathematical
Olympiad, 2014.
248 Let P and Q be the intersection points of the line F G with the sides AB and
AC, respectively, and let I and J be the second intersection points of F G with the
circumcircles of the triangles BDF and CEG, respectively, as shown in Fig. 12.40.
Because AF = AG, AO is the perpendicular bisector of the segment F G. Note
also that in the triangle ABC, the line AO and the altitude from A are isogonal, and
so F G is antiparallel to BC, meaning that AP Q = ACB.
Let Y be the intersection of BC and F G. Writing the power of the point Y with
respect to the circles and , we obtain Y B ·Y C = Y F ·Y G = Y D·Y E. This leads
us to considering the inversion of center Y that maps and into themselves. This
inversion maps F, B, and D to G, C, and E, respectively, so it maps the circumcircle
of F BD to the circumcircle of GCE.
Since Y is the center of the inversion that maps the circumcircle of F BD to the
circumcircle of GCE, it is also the center of a homothety that maps these circles

into each other. Consequently the arcs BF and EJ have the same measure because
they are mapped by the homothety into each other. We obtain
Fig. 12.40 Construction of A

the points P , Q, I, J
X L
G
J
K Q
F
P I
Y O
B D E C
12 A Synthesis 557

XF G =KI /2 = AP Q− BF /2 = ACB− EJ /2 =LJ /2 = XGF.
Hence X is on the perpendicular bisector of segment F G, which is the line AO.

Remark The key step of the solution was the interplay between inversion and
homothety.
Source International Mathematical Olympiad, 2015, proposed by Greece.
249 Invert with respect to . Since no reference will be made to the original
configuration, the images will be denoted by the same letters (with no dash), except
for G, the center of , which will remain the center of inversion (and not the point
at infinity). The inversion maps lines tangent to to circles with diameters equal to
the radius of that are internally tangent to and pass through G. After inversion,
the problem becomes as follows:
Fix a circle of center G and radius r, a circle of radius r/2 which passes
through G, and a circle inside and disjoint from . The circles ω1 and ω2 of
radii r/2 pass through G and through a variable point X on and cross again at
Y and Z. Prove that as X traces , the circumcircle of XY Z is tangent to two fixed
circles.
Let’s solve it! Since ω1 and ω2 are the reflections of the circumcircle of
the triangle GY Z over the sides GY and GZ, respectively, they pass through the
orthocenter of this triangle. The second intersection point of these circles is X, so
X is the orthocenter of the triangle GY Z (Fig. 12.41). Hence the circumcircle ω3 of
the triangle XY Z is the reflection of over Y Z; in particular the radius of ω3 is r/2.
We continue our reasoning on Fig. 12.42. Let O and L be the centers of and
, respectively, and let R be the (variable) center of ω3 . The circle is mapped into
−→
ω3 by a translation of vector LR, and this translation maps G to X. Hence GLRX
−→ −→
is a parallelogram, which implies that we have the equality of vectors XR = GL,
−→
showing that XR is constant, and so R is the is image of X under a translation by
−→
the vector GL.
−→
We use this translation of vector GL to construct the point N as the translate
of O. Then XRN O is a parallelogram, so the distance RN = OX is constant.
Consequently ω3 is tangent to the fixed circles centered at N and of radii r/2 − OX
and r/2 + OX.
But we are not done yet! We have to check that the inversion maps these two
circles into circles and not lines. The latter happens only if one of the circles contains
G, the center of inversion. Let us verify that this cannot happen. Since lies inside
, OL < r/2 − OX, so
−→ −→ −→ −−→ −→ −→
N G = NG = GL + LO + ON = 2GL + LO
−→ −→
≥ 2GL − LO > r − (r/2 − OX) = 1/2 + OX.
558 12 A Synthesis
Fig. 12.41 Inverted figure
ω3
Ω X Y
Γ
ω2 G
ω1
This shows that G is necessarily outside each of the two circles.

Remark At first glance, the inverted figure does not look simpler than the original.
But the fact that the new X is the orthocenter of GY Z gives us a significant
advantage.
We should point out that the required fixed circles are also tangent to .
Source Romanian Master of Mathematics, 2018, proposed by Ivan Frolov, Russia.
250 We consider the system of coordinates inspired by a homothety and a reflection
used in Problem 4.1 from the introduction to this chapter (with E at the origin). Let
us show that the Euler lines of ABE, ADE, and BCE are concurrent (and then
applying the argument to some other triple of triangles we obtain that all Euler lines
are concurrent).
We use the formulas for the centroid and circumcenter of a triangles with vertices
z 1 , z2 , z3 :

1 1 1

z1 z2 z3

z 1 + z2 + z3 |z |2 |z |2 |z |2
1 2 3
g= , o= .
3 1 1 1

z1 z2 z3

z z z
1 2 3
12 A Synthesis 559
Fig. 12.42 Construction of

the two circles
ω3
R
N
Z
Ω Y
L X
O
Γ G
ω2
ω1
A short computation shows that the centroids and circumcenters of ABE, ADE,
BCE are, respectively,
1 + aω ω−a
g1 = , o1 =
3 ω−ω
1 − kω k+ω
g2 = , o2 =
3 ω−ω
aω − ak −kaω − a
g3 = , o3 = .
3 ω−ω
Using the equation of a line through the points z1 , z2 :

1 1 1

z z1 z2 = 0,

z z z
1 2
after algebraic manipulations, we obtain the equations of the three Euler lines to be
[−a(ω2 +2)+ω(ω2 +2)]z+[−a(ω2 + 2) + ω(ω2 +2)]z+(a 2 − 1)(ω + ω) = 0

[k(ω2 +2)+ω(ω2 +2)]z+[k(ω2 +2)+ω(ω2 + 2)]z + (k 2 − 1)(ω + ω) = 0
[(ω2 +2)+kω(ω2 +2)]z+[(ω2 +2)+kω(ω2 + 2)]z − a(k 2 − 1)(ω + ω) = 0.
560 12 A Synthesis
These lines are concurrent if this system of three equations in the unknowns z and z
has solution, and this is equivalent to the fact that one of the equations is redundant.
To see that one equation is redundant, add the third equation multiplied by a to the
first, and subtract the third equation multiplied by k from the second to obtain
(ak + 1)[ω(ω2 + 2)z + ω(ω2 + 2)z] − (ak + 1)(ak − 1)(ω + ω) = 0

(1 − k 2 )[ω(ω2 + 2)z + ω(ω2 + 2)z] − (1 − k 2 )(ak − 1)(ω + ω) = 0.
These two equations are proportional, and so in the original system of three
equations, the second equation is redundant. Thus the three lines are concurrent,
Remark The original solution communicated to us by the authors of the problem
was synthetic and quite involved. Here we have an example where the analytic
argument is shorter than the synthetic one.
The formula for the coordinate of the circumcenter can be found in T. Andreescu,
D. Andrica, Complex Numbers from A to . . . Z, second ed., Birkhäuser, 2014.
Source Petru Braica and Vlad Robu.
251 Let Tn = An Bn Cn with altitudes An An+1 , Bn Bn+1 , and Cn Cn+1 . Let also
αn = Bn An Cn , βn = An Bn Cn , γn = Bn Cn An . Then we have the recursions
αn+1 = π − 2αn , βn+1 = π − 2βn , γn+1 = π − 2γn if An Bn Cn is acute,

π
αn+1 = 2αn − π, βn+1 = 2βn , γn+1 = 2γn if αn > ,
2
π
αn+1 = 2αn , βn+1 = 2βn − π, γn+1 = 2γn if βn > ,
2
π
αn+1 = 2αn , βn+1 = 2βn , γn+1 = 2γn − π if γn > .
2
These recursions can be visualized using Fig. 12.43.
Because we only care about triangles up to a similarity, these recursions carry all
the information of interest to us. So we may view H as the map
H : → , H (αn , βn , γn ) = (αn+1 , βn+1 , γn+1 ),
where ⊂ (0, ∞)3 consists of those triples (x, y, z) satisfying x + y + z = π .

The problem asks us to find the triples (α0 , β0 , γ0 ) for which (αn , βn , γn ) is a
permutation of (α0 , β0 , γ0 ).
Let us draw an equilateral triangular region S with altitude of length π
(Fig. 12.44). To every triple (α, β, γ ) with α +β +γ = π and α > 0, β > 0, γ > 0,
we associate a unique point P inside S such that the distances from P to the sides
are, respectively, α, β, and γ . Thus the interior of S parametrizes triangles up to a
12 A Synthesis 561
An
Bn+1 Bn+1
Cn+1 Cn+1
An
Bn An+1 Cn Bn An+1 Cn
Fig. 12.43 The map H
n=2 ρ6
S1
S2
γ S3 S6
β
α S4 S5
Fig. 12.44 The parametrization of triangles using an equilateral triangle
similarity. Moreover, the segments joining the midpoints of the sides parametrize
right triangles.
Examining the geometric content of the above recursions, we deduce that
• H is not defined on the segments joining the midpoints,
• H maps the interior of each of the 4 triangles determined by segments that join
the midpoints by a homothety into the original triangle (this homothety has ratio
2 for the triangles in the corners and −2 for the triangle in the middle).
Inductively we prove that H n is not defined on the lines parallel to sides that
divide the altitudes in 2n equal parts, and on the interior of each of the 22n triangles
determined by these lines, H n is a homothety that maps that triangle to S (the case
n = 2 is shown in the middle of Fig. 12.44).
Again, because we are only interested in triangles up to similarity, we restrict
ourselves to the case where α0 ≥ β0 ≥ γ0 . The region inside S parametrizing such
triangles is one of the six congruent right triangles determined by sides and altitudes.
Denote by S1 this triangle and by S2 , S3 , S4 , S5 , S6 the other five. Let also ρj , j =
1, 2, . . . , 6, be the isometries of S that map Sj , Respectively, to S1 (Fig. 12.44). The
fact that Tn is similar to T0 is equivalent to the fact that (ρj ◦ H n )(x) = x for some
x ∈ S1 .
If each of the 22n triangles in which S is divided is further divided in six
by its altitudes, then in each of the equilateral triangles appears a triangle that
corresponds to S1 and is similar to S1 with similarity ratio 1/2n . Let these triangles
be R1 , R2 , . . . , R2n . Now extend H to the boundaries of these triangles and view
562 12 A Synthesis
them as closed regions. Note that H n : Rk → S is a homothety onto one of the Sj ,

say Sjk .
We claim that for each j = 1, 2, . . . , 2n , there is exactly one point xk in Rk such
that
(ρjk ◦ H n )(xk ) = xk .
The map ρjk ◦ H n is a spiral similarity. Then ρj−1

k
: S1 → S1 is a spiral similarity
with ratio 1/2n . We will show that ρj−1
k
has a unique fixed point.
Uniqueness is obvious, since ρj−1 k
shrinks distances. On the other hand, ρj−1 k
,
being a spiral similarity of the plane with ratio different from 1, does have a fixed
point, and this fixed point necessarily lies in S1 , because by choosing m sufficiently
large, we can make sure that (ρj−1k
)m (S1 ) contains the fixed point. The unique fixed
point of ρj−1k
is the unique fixed point of ρjk so the claim is proved.
However, not all points xk are admissible, since some might lie on the lines that
parametrize right triangles or degenerate triangles. In fact it is not hard to see that
this is the case precisely when the triangle Rk has the larger leg on the larger leg of
S1 , for in this case the fixed point is on the boundary, and hence yields a degenerate
triangle. There are 2n such triangles, which have to be removed. So the answer to
the problem is
22n − 2n .

252 (a) First solution. Suppose without loss of generality that A is the closest
vertex to the line EF , as in Fig. 12.45. Then, since ABCD, ADEM, and ABF M
are cyclic, we have the following chain of equalities of directed angles modulo 180◦ :
AME = ADE = ADC = ABC = ABF = AMF,
so M, E, and F are collinear.

Second solution. To be in tune with the rest of the book, we can work with
complex coordinates and write

e−m e−m f −m e−d f −b
arg = arg : = arg :
f −m a−m a−m a−d a−b

c−d c−b
= arg : = 0 (mod π ),
a−d a−b
where we have used again the cyclicity of ABCD, ADEM, and ABF M, and the
collinearities of A, B, C and C, D, E. Hence M, E, F are collinear.
12 A Synthesis 563
Fig. 12.45 Cyclic complete F

quadrilateral
M
A
C
√
(b) Since the bc inversion swaps A with C, and B with D, the circumcircle of
ABCD is taken to the circumcircle of A B C D = CDAB, which, of course, is
the same circle. √
(c) From (b), since the circumcircle ABCD is fixed by the bc inversion, we
deduce that its center O lies on the common bisector of AMC, BMD, and
EMF .
We will now prove that O and P are swapped by the inversion, which will finish
this part of the problem. To see why this is true, it suffices to prove that O lies on
the circumcircles of both MAC and MBD, which are the images of AC and BD
under this inversion. If MA = MC, this is true because MO is the internal bisector
of AMC, OA = OC, and the intersection of the perpendicular bisector of AC and

the internal bisector of AMC is the midpoint of the arc AC of the circumcircle of
AMC that does not contain M. Such midpoint is O. If MA = MC, the inversion
circle goes through A and C, and the circumcircle of ABCD, being fixed by this
transformation, is orthogonal to the circle of inversion. This implies OC ⊥ MC
and OA ⊥ MA, which prove the claim in this case as well. We deduce that MAOC
and MBOD are both cyclic, completing the proof.
(d) By (a) M lies on EF , so EMF = 180◦ . Hence the bisector of EMF
divides it into two angles of 90◦ ; since this common bisector is also the line OP , it
follows that EF ⊥ OP .
(e) Part (c) already establishes that MAOC is cyclic. For the other part, let ωAC
meet say, line AB again at X. Line AC divides the plane in two half-planes and will
be taken as reference: points are opposites if they are in opposite half-planes and
neighbors if they are in the same half-plane. For now, suppose that B and M are
opposites. Then either B and X are opposites (and X and M are neighbors), and
BXC = AXC = AMC,

564 12 A Synthesis
or B and X are neighbors (and X and M are opposites), so
BXC = 180◦ − AXC = AMC
In any case, since M and O are always opposites,
BXC = AMC = 180◦ − AOC = 180◦ − 2 ABC = 180◦ − 2 XBC,
which implies that triangle BXC is isosceles with BX = XC, that is, X lies in the
perpendicular bisector of BC.
The case in which B and M are neighbors is handled similarly and is left to the
reader.
Finally, notice how the pattern works: every side contains exactly one of points
A, C and another point P ∈ {B, D}; then ωAC meets the perpendicular bisector of
the side at the side through P and the other point from A, C.
253 We work in the complex plane and assign the corresponding lowercase letters
to uppercase points. We choose the coordinate system discussed in the introduction
to this section on complete quadrilaterals (Sect. 4.2), with the Miquel point at the
origin and such that c = 1/a and d = 1/b. Rotating the figure appropriately, we
can make the common bisector of AMC and BMD be the real axis. Then p is a
positive real number, that is, p̄ = p. Since P lies on AC,

p p 1

a ā 1 = 0 ⇐⇒ p = a + ā ⇐⇒ a − 1 ā −
1
=
1 − p2
1 + a ā p2
1 1 1 p p
a ā

1 1 − p2

⇐⇒ a − = .
p p
In the above formula for p, if we divide both the numerator and the denominator
by a ā, we obtain
a + ā 1/a + 1/ā
p= = ,
1 + a ā 1 + 1/(a ā)
so

1
− = 1−p
1 2
a p p
as well.
Similarly,
12 A Synthesis 565

1 1 − p 2 1 1 1 − p2
b − = and − = .
p p b p p
√
1−p2
We deduce that A, B, C, and D lie on the circle with center 1/p and radius p .
Remark We can read in this formula what has been proved in the solution to
Problem 252, namely, that the center of the circle is the inverse of P . In fact this
problem is the reciprocal of part (c) of Problem 252.
254 First solution. Let the circumcircles of AKN and CKN be ω1 and ω2 ,
respectively. Let us first consider the case where ω1 is tangent to ω and, arguing
on Fig. 12.46, prove that ω2 is tangent to ω.
Because ω and ω1 are interior tangent, there is a direct homothety that maps the
first into the second. This homothety maps BD to KN , so BD is parallel to KN.
But then there is an inverse homothety that maps CBD to CN K, and this homothety
maps ω to ω2 . This is possible only if the two circles are tangent. The reciprocal is
similar. √
Second solution. Perform the bc inversion that fixes the complete quadrilateral.
Then A and C are swapped, B and D are swapped, K and N are swapped, and ω
is fixed. Therefore the circumcircles ω1 of AKN and ω2 of CKN are mapped into
each other, and ω and ω1 are tangent if and only if ω = ω and ω1 = ω2 are tangent.
Source Russian Mathematical Olympiad (regional level), 2001, proposed by L.
Emelyanov.
255 We argue with the help of Fig. 12.47. This problem is a showcase for spiral
similarities and Miquel points as discussed in Chapter 2. In fact, the ratios
AE BF
=
ED FC
suggest that we consider the center M of the spiral similarity that takes A to B, D
to C, and, by proxy, E to F .
Fig. 12.46 ω1 is tangent to ω

K
ω B
A
D ω2
ω1 N
566 12 A Synthesis
Fig. 12.47 Spiral similarities

and Miquel points all around S
T M
A
E
B F C
We already know that if a spiral similarity takes X to X and Y to Y , then its

center is the Miquel point of XY Y X . Therefore M is the Miquel point of AEF B,
so, by Miquel’s Theorem (Theorem 2.24), M lies on the circumcircles of SAE and
SBF , and also M is the Miquel point of EDCF , so M lies on the circumcircles of
T CF and T DE. We are done.
Source United States of America Mathematical Olympiad, 2006, proposed by
Zuming Feng and Zhonghao Ye.
256 We reason on Fig. 12.48 and solve the problem in a more general case,
replacing the assumption that BC = AD and BE = DF with the only condition
that
BE DF
= .
EC FA
We are now in a situation very similar to that of the previous problem, and
it is natural to consider the spiral similarity s that maps the segment BC to
DA, because s(E) = F . Applying Theorem 2.22 successively to the pairs of
segments (BC, DA), (BE, DF ), and (EC, F D), we deduce that the center O
of s is the second intersection point of the circumcircles of P BC and P DA, the
second intersection point of the circumcircles of RBE and RF D, and the second
intersection point of the circumcircles of QAF and QEC.
Two complete quadrilaterals arise in the configuration, BEQP RC and
AP RF QD. From the above considerations, we deduce that O is the second
intersection point of the circumcircles of P BC and RBE, so it is the Miquel point
of BEQP RC, and also O is the second intersection point of the circumcircles of
QAF and RF D, so it is the Miquel point of AP RF QD. By Miquel’s Theorem
(Theorem 2.24), O is the second intersection point of the circumcircles of P QR and
12 A Synthesis 567
Fig. 12.48 Spiral similarities D

and Miquel points again! F
A R
P O
Q
B
E
C
P BC and the second intersection point of the circumcircles of P QR and P AD. So

the circumcircles of all triangles P QR pass through the second intersection point
of the circumcircles of P AD and P BC, and this point is fixed because so are the
two triangles.
Source International Mathematical Olympiad, 2005, proposed by W. Pompe,
Poland.
257 First solution. Examining Fig. 12.49, we recognize Q as the Miquel point of
the complete quadrilateral AMP NBC. So in addition to the direct and inverse
homotheties that map AMN to ABC and MNP to CBP , which are manifest in
the statement of the problem, we also have the spiral similarities that map QAB to
QCP and QAC to QBP . Switching to complex coordinates and working with the
convention that the lowercase letter is the coordinate of the uppercase point, we can
use the abovementioned transformations to write
a−n a−n m−n a−c c−b a−c a−q
= · = · = = .
p−n m−n p−n b−c p−b b−p b−q
So the triangles ANP and AQB are similar, and consequently P AN = BAQ,
as desired.
√
Second solution. Let φA be the bc inversion corresponding to the vertex A in
the triangle ABN. Recall √ that φA is obtained as a composition of the inversion χ
of center A and radius AB · AN and the reflection σ with respect to the bisector
of BAN . From the fact that AB and MN are parallel, we obtain that AB · AN =
AC · AM, so φA (C) = M. The circumcircles of ABN and ACM are therefore
images of the lines CM and BN, respectively.
Examining once more the complete quadrilateral ANP MBC from Fig. 12.49,
we notice that Q is its Miquel point, where the circumcircles of ACM, ABN, CN P ,
and BMP meet, so Q is the second intersection point of the circumcircles of ACM
and ABN. Since this point is mapped by φA to the intersection point of CM and
BN, we have φA (Q) = P . Consequently the lines AP and AQ are the images
of one another through φA , and since the lines through A are invariant under the
568 12 A Synthesis
Fig. 12.49 The complete

quadrilateral AMP N BC A
M N
P
B C
Q
inversion χ , the lines AP and AQ are mapped into one another by the reflection σ
(they are isogonals), and the equality of the two angles follows.
Remark So at the heart of the first solution lies the observation that all the
information about the triangle ABC, up to a similarity, is encoded in the expression
c−a
.
b−a
For any triangle that is similar to ABC, this expression is the same.
Source Balkan Mathematical Olympiad, 2007, proposed by Moldova.
258 We recognize the flavor of Problem 255, and inspired by that problem, we
argue as follows. Because
AP BQ
+ =1
AC BD
is equivalent to
AP DQ
= ,
AC BD
we employ the spiral similarity that takes A to D and C to B and which therefore
also takes P to Q. The center O of this spiral similarity is the Miquel point of
ACBD (it is the Miquel point of a skew quadrilateral).
Now we reason on Fig. 12.50. Recall that spiral similarities come in pairs. In fact,
there is an (automatic) spiral similarity with center O that takes B to C and Q to P ,
so O is the Miquel point of BQP C and belongs to the circumcircles of CN P and
BNQ.
12 A Synthesis 569
Fig. 12.50 More spiral B

similarities and Miquel points
all around
A
N
P
M Q
O
D C
Fig. 12.51 Asian Pacific Z

X
problem via inversion P
Q
P
Y
A D
Q
ω F M
B D
Another automatic spiral similarity, that takes A to D and P to Q, finishes the

problem, as O is also the Miquel point of AP QD and belongs to the circumcircles
of AMP and DMQ.
Source Bulgarian Team Selection Test for the International Mathematical
Olympiad, 2004.
259 First solution. Our first solution uses inversion. Let D be the Miquel point of
the complete quadrilateral determined by P , Q , AB, and MP . A good diagram
(such as Fig. 12.51) suggests that the tangency point is D. We also define X as
P ∩ AB, Y as Q ∩ AB, and Z as P ∩ Q . Also, let F be the intersection of P Q
and AB.
We start by noticing that, since AB is parallel to the tangent line to ω at M,
1 1 1
XP F = P AM= (P A + AM) = (P A + MB) = XF P ,
2 2 2
so XP = XF . Taking advantage of the fact that X lies on the radical axis of ω and
, we invert with center X and radius equal to the square root of the common power
570 12 A Synthesis
of X with respect to ω and (which radius is XP = XF ). As usual, we denote by

K the image of K under this inversion. Then P = P , F = F , ω, , and AB are
fixed, and since D is a Miquel point, it lies on the circumcircle of the triangle XP F ,
so its inverse D lies on the line P F = P F . Therefore D is the intersection of
XD and P Q, and since A and B are swapped by the inversion, A, B, D, and D are
concyclic. And by drawing a careful diagram, we can guess that the circumcircle of
ABD D is !
To see why our guess is correct, we consider the circumcircle of DD Q. Recall
that D is a Miquel point, so it lies on the circumcircle of Y QF . Because XD·XD =
XF 2 , the triangles XF D and XD F are similar, so
DD Q = XD F = DF X = DF Y = DQY.
This implies that Q is tangent to at Q. Finally, notice that XQ < XF implies

that XQ > XF > XQ so Q = Q. From XD · XD = XQ · XQ , using power-
of-a-point, we deduce that Q also lies on . But Q lies on as well, because is
fixed by the inversion. And because two distinct points of a circle together with the
tangent line at one of the points uniquely determine the circle, = , and D lies
on .
Thus we have proved that D is on both and the circumcircle γ of XY Z. It
remains to show that D is a tangency point. The inversion helps here too: γ is
mapped to the line D Y , so we need to show that this line is tangent to = .
Another angle chasing completes the solution: since XK L = XLK for any
points K, L,
DD Y = XD Y = XY D = 180◦ − DY F
1
= 180◦ − DQF = 180◦ − DQD = DD.
2
Second solution. The official solution makes use of the idea that two tangent
circles are homothetic with respect to an inverse homothety centered at the tangency
point. We use the same notation as in the previous solution. The main idea is to
construct a triangle inscribed in that is (hopefully) homothetic to our main triangle
XY Z. Let P Q meet again at R = Q, and let us define the points S and T on
such that RS AB and RT P (see Fig. 12.52.) Then RST and XY Z are
homothetic if and only if ST Y Z, namely, if ST Q .
12 A Synthesis 571
Fig. 12.52 Asian-Pacific Z

X
problem via homothety P
Q
P
Y D
A
S
Q
ω F M
Ω
B R
T
As in the previous solution F P X = P F X, so
P RT = RP X = F P X = P F X = (AB, P R) = (RS, P R) = SRP .

We infer that Q is the midpoint of the arc SQT of . Therefore, ST Q , and RST
and XY Z are homothetic. This homothety also takes the circumcircle of RST , ,
to the circumcircle of XY Z, γ .
Now we only need to prove that D, the other intersection point of the circle
with the line XR, is the center of homothety. This happens if and only if D
belongs to the line Y S. Power-of-a-point and some angle chasing will do: since
XF 2 = XP 2 = XA · XB = XD · XR, the triangles XF D and XRF are similar,
so
DF Y = DF X = F RX = QRD = DQY,
that is, DF QY is cyclic. Finally,
Y DQ = Y F Q = (AB, QR) = (SR, QR) = SRQ = 180◦ − SDQ,
so D lies on SY , and we are done.

Source Asian Pacific Mathematical Olympiad, 2014, proposed by Ilya Igorevich
Bogdanov, Russia, and Medeubek Kungozhin, Kazakhstan.
260 Let the complete quadrilateral be obtained from the convex quadrilateral
BCB C by intersecting BC with B C at A and BC with B C at A , with B on
the segment AC and B on the segment A C, as shown in Fig. 12.53. On this figure
we have marked the respective circumcenters O, Oa , Ob , Oc of the triangles ABC,
AB C , BA C , CA B , as well as the Miquel point M of the complete quadrilateral.
572 12 A Synthesis
Oa
M
B
C
O
Ob
Oc
A B C
Fig. 12.53 The complete quadrilateral with circumcenters and the Miquel point
Theorem 4.7 proves the existence of the spiral similarities s, sa , sb , sc centered

at M that map the following pairs of triangles into each other:
ABC → Oa Ob Oc , AB C → OOb Oc , A BC → Oa OOc , A B C → Oa Ob O.
Also, by Theorem 4.6 the quadrilateral OOa Ob Oc is cyclic, and its circumcircle
passes through M.
Looking at the configuration from a different angle, we denote by H , Ha , Hb , Hc
the orthocenters of the triangles Oa Ob Oc , Ob Oc O, Oc OOa , and OOa Ob ,
respectively, and then, as shown in Problem 23, the quadrilateral H Ha Hb Hc is
the image of the quadrilateral OOa Ob Oc under a reflection over a point, and
this reflection (which is a 180◦ rotation) we denote by σ . Let Om and Hm be the
circumcenters of OOa Ob Oc and H Ha Hb Hc , respectively, which correspond to
each other through this reflection.
Now, the composition s = σ ◦ s is a spiral similarity, being the composition of
two spiral similarities, whose ratio is the same as that of s and whose angle is equal
to the angle of s plus 180◦ (which is the angle of σ ). We have
s (O) = σ ◦ s(O) = σ (Om ) = Hm ,

s (H ) = σ ◦ s(H ) = σ (H ) = O.
12 A Synthesis 573
H M
O X
Hm s(X)
Fig. 12.54 Graphic aid for understanding the transformation s = σ ◦ s
It follows that s (H O) = OHm and from here we can deduce that the triangle
OH Hm is similar to all triangles of the form MXs(X), which define the spiral
similarity s (look at Fig. 12.54 to understand why this is true). In particular OH Hm
is similar to MAOa . And since the triangle MAOa is isosceles because Oa M and
Oa A are radii in the circle of center Oa , it follows that the triangle Hm OH is
isosceles, with Hm O = Hm H . But then Hm is on the perpendicular bisector of
OH , which, by Theorem 2.10, is the perpendicular to the Euler line of ABC at
the center of its nine-point circle. A similar argument shows that the other three
perpendiculars pass through the circumcenter Hm of H Ha Hb Hc , and we are done.
Source This result was proved by S. Kantor in Bulletin des Sciences Mathématiques
in 1879 and by R.J. Hervey in Educational Times, 1891, the solution is taken
from G. Mihalescu, Geometria Elementelor Remarcabile (The Geometry of the
261 We first argue on Fig. 12.55. By definition, D lies on the bisector of angle
BAC. The angles ADE and BCD feel a little “off” with each other. So we
define D as the isogonal conjugate of D in the triangle ABC. Then
ECD = ACD = DCB = EDA = EDD ,
and we discover the much friendlier cyclic quadrilateral ECD D. And for a similar
reason, the quadrilateral F BD D is cyclic as well.
Notice that A lies on the radical axis DD of the circumcircles of ECD D and
F BD D, so by power-of-a-point AF · AB = AD · AD = AE · AC, and therefore
the quadrilateral BCEF is cyclic.
Now we can drop D and complete the diagram, and we argue now on Fig. 12.56.
Out of the three lines BC, EF , and O1 O2 , the “odd one out” is O1 O2 , so let the
lines BC and EF meet at the point T . Our goal is to prove that T lies on O1 O2 .
Let M be the Miquel point of the cyclic quadrilateral BCEF . Notice that the line
AT passes through M. As a consequence of Problem 252 (e), the point X, which is
the intersection of the side EC of BCEF and the perpendicular bisector of the side
BC, lies on the circumcircle of BEM.
We take advantage of D being on both the circle with center O1 and the circle
with center O2 and invert with center D and arbitrary radius. As usual, denote the
574 12 A Synthesis
Fig. 12.55 Getting started A

with the problem
F
E
B C
A
M
O2 F
E
O1
B C T
Fig. 12.56 Miquel point helps connecting X to the rest of the diagram
inverse of P by P for any point P = D. Before we draw the inverted diagram, let
us locate the inverses by computing some angles. Recall that DP Q = DQP
for distinct points D, P , Q. Then
F B C = F B D + DB C = BF D + DCB.
12 A Synthesis 575
Some additional angle chasing yields
F B C = BF D + DCB = ADF + BAD + DCB

= DBC + DAC + DCB,
which is symmetric with respect to B and C. So F B C = E C B . Since

BCEF is cyclic, so is B C E F , which means that B C E F is an isosceles
trapezoid. Or, speaking in terms of transformations, there is a reflection σ across
a line that takes B to σ (B ) = C and E to σ (E ) = F .
We argue now on Fig. 12.57. The reflection σ makes the inverse of almost all
the other points fall into place: since P is the intersection of the lines EF and BC,
P is the second intersection of the circumcircles of DE F and DB C , which by
symmetry is σ (D), and since the Miquel point M is the intersection of, say, the line
AP and the circumcircle of AEF , M is the intersection of the circumcircles of
DA P and A E F ; again by symmetry, M = σ (A ).
Back to the problem, we need to prove that T and the circumcenters O1 and O2
of ADC and EDX, respectively, are collinear. Recall that the inverse of the center
of a circle passing the center of inversion is the reflection of the center across the
inverted circle, which is a line (Theorem 3.7). The inversion transforms the problem
into proving that the four points D, T , and the reflections of D across A C and
E X lie on a circle.
We know that T is the reflection of D across , so , A C , and E X are
perpendicular bisectors of the line segments connecting D to the other three points.
So it is sufficient to prove that , A C , and E X meet at a single point.
Finally, since E, M, X, B lie on a circle, E , M , X , B also lie on a circle α;
since X, A, E, C lie on a line, X , A , E , C lie on a circle β; and since M is a
Miquel point, M, A, B, C lie on a circle, so M , A , B , C also lie on a circle γ .
Now, the radical axis of α and β is E X , the radical axis of α and γ is M B , and
Fig. 12.57 Inversion makes

(almost) every point nicer
F E
A M
D T
B C
576 12 A Synthesis
the radical axis of β and γ is A C . These three lines meet at the radical center of
the three circles. This radical center is also the intersection point of A C and M B ,
which by symmetry lies on . We are done.
Source International Mathematical Olympiad, 2021, proposed by Mykhailo Shtan-
denko, Ukraine.
Index
A Excircle, 82, 83, 173–175, 177, 201, 204, 210,

Affine transformation, 87, 101, 112, 207, 528 212, 255–257, 264, 374, 393, 400, 407,
Apollonian circles, 36, 113, 158, 187, 190, 408, 454, 466, 494, 504, 505, 508, 511,
415, 416, 443 547, 553, 555
Averaging Principle, 101, 102, 121, 124, 125,
258, 265, 266, 418–421, 425, 515, 522,
532, 536 G
Group of transformations, 7, 77, 112, 327, 518
C H
Complete quadrilateral, 107, 108, 234–245, Harmonic division, 79, 81, 131, 145, 159, 163,
267, 405, 425, 428, 563–569, 571, 572 187, 229, 456, 466, 490, 493, 533
Complex affine transformations, 101, 102 Harmonic pencil, 228, 229
Cross-ration, 12, 13, 35, 80, 131, 140, 141, Harmonic quadrilateral, 34–36, 145, 173, 196,
143–145, 148, 173, 180, 183–187, 191, 227, 229, 466, 483, 484, 490, 493
195, 196, 227–229, 240, 241, 261, 262, Homothety, viii, 21, 71–128, 131, 136, 137,
265, 441, 447, 449, 456, 466, 470, 471, 144, 146, 158, 160, 162, 163, 176, 177,
479, 483, 484, 490, 493, 517, 526, 527, 207, 211–215, 222, 223, 236–238, 242,
554 243, 255–267, 325, 369–437, 445, 451,
452, 457, 459, 484, 487, 491, 501, 509,
511, 517, 519, 523, 524, 528, 530, 531,
D 534–538, 542–544, 550–552, 556–558,
Directed angles, 11, 12, 114, 116, 120, 132, 561, 562, 565, 567, 570, 571
163, 188–191, 221, 223, 227, 240, 298,
300, 301, 414, 430, 522, 523, 562
I
International Mathematical Olympiad, ix, 89,
123, 191, 193, 306, 310, 312, 332, 349,
E 361, 388, 396, 408, 420, 427, 431–433,
Euler’s line, 79, 89, 90, 94–96, 123, 202, 233, 436, 451, 459, 464, 466, 468, 486, 507,
245, 375, 379, 389, 394, 395, 482, 487, 530, 531, 539, 544, 545, 555–557, 567,
558, 559, 573 569
https://doi.org/10.1007/978-3-030-89117-6
578 Index
Inversion, viii, 36, 82, 129–205, 207, 212, 110–112, 121–123, 127, 129–131, 136,
214, 215, 220, 222–224, 226, 227, 229, 141–144, 151, 153, 154, 158, 166–168,
234–237, 239, 244, 261–267, 437–512, 170–172, 179–181, 187, 198, 207, 212,
519, 522, 523, 528, 533, 543, 545–548, 213, 220, 222, 227, 228, 230, 232, 233,
550, 551, 553–557, 563, 565, 567–570, 236, 249–252, 254, 262–266, 271–278,
575 284, 287–292, 294, 298–307, 315, 320,
Isogonal conjugate of a point, 36, 38, 39, 251 330, 332–335, 339, 340, 350, 353, 355,
Isometries, viii, 3–69, 77, 109, 110, 112, 135, 357, 359, 361, 362, 369, 374, 379, 380,
141–145, 149, 249–254, 266, 271–367, 383, 387, 388, 390, 392, 393, 425, 437,
372, 410, 444, 513, 521, 540, 561 441, 442, 457, 461, 471, 472, 479–482,
486, 487, 491, 492, 494, 496–498,
507, 508, 514, 517–520, 522–524, 529,
L 533–540, 543–546, 548, 550, 553,
Linear fractional transformations, viii, 138, 556–558, 567, 568, 572, 575
145–149, 183, 445 Rotation, 8–19, 21–30, 32, 39, 40, 42, 43, 45,
46, 48, 49, 58, 59, 62, 64, 65, 69, 71,
M 100, 101, 104, 109, 110., 112, 113,
Mathematical Olympiads, ix, 46, 62, 89, 120, 118, 124, 126, 144, 148, 157, 207–209,
121, 196, 213, 239, 287, 292, 299, 303, 250–254, 265, 266, 271, 273–276,
308, 310, 315, 328, 330, 336, 349, 386, 279–285, 294–297, 302–305, 308,
390, 396, 399, 409, 411, 423, 424, 435, 310–312, 322–327, 333, 338, 339, 346,
445, 457, 473, 474, 477, 479, 481, 519, 347, 350, 353–355, 358–365, 420, 426,
523, 538, 541, 565, 566, 568, 569, 571 435, 464, 494, 513–515, 521, 523, 524,
Miquel’s point, 107, 108, 234–237, 241, 243, 531, 533, 534, 538, 539, 572
244, 258, 267, 425, 428, 564–575
Mixtilinear circles, 176, 204, 481, 489
Möbius transformations, 138–145, 149, 151, S
152, 158–159, 171, 173, 180, 183–186, Simson line, 164, 231, 236, 265, 523, 530
195, 196, 198–205, 207, 226, 227, 235, Spiral similarity, 71–128, 135, 141–144, 149,
239, 240, 261, 437, 438, 441–444, 449, 171, 172, 176, 186, 189, 207, 210, 216,
450, 469, 477, 481, 526, 528 218–220, 226, 232, 234, 235, 237,
243, 255–259, 265–267, 369–436, 450,
513, 515, 516, 520, 523–525, 527–532,
N 534–537, 539–541, 544, 553, 555, 562,
Nine-point circle, 79–82, 94, 95, 153, 166, 566, 567, 569, 572
204, 210, 212, 226, 245, 256, 264, 379, Symmedian, 32–36, 53, 81, 105–107, 113,
391, 392, 394, 395, 472, 482, 486, 487, 119, 173, 201, 202, 220, 225–227, 250,
491, 494, 496, 497, 501, 512, 555, 556, 258, 264, 283, 285, 415, 429, 430, 456,
573 466, 473, 479, 490, 491, 493, 499, 502,
503, 506, 507, 540, 541, 546, 547
O
Orientation of polygons, 15, 16
T
Translation, 8–30, 39, 43, 49, 57, 63, 67, 75,
P 77, 78, 84, 86, 88, 101–104, 109, 110,
Polar of a point with respect to a circle, 191 112, 113, 131, 141–144, 146, 186, 192,
Pole of a line with respect to a circle, 159, 160 207, 209, 249–253, 255, 271–275, 277,
282, 286, 293, 297, 298, 309, 311–313,
315, 321, 322, 324, 326–330, 333, 334,
R 336–338, 341, 343–345, 348, 350, 355,
Reflection, 3, 8–21, 24–28, 30–34, 36–39, 42, 356, 361, 362, 365, 369, 371, 372, 393,
43, 49, 50, 54–56, 77, 80, 81, 94, 95, 398, 418, 445, 513, 557

(Problem Books in Mathematics) Răzvan GelcaIonuţ OnişorCarlos Yuzo Shine - Geometric Transformations-Springer, Cham (2022)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Problem Books in Mathematics) Răzvan GelcaIonuţ OnişorCarlos Yuzo Shine - Geometric Transformations-Springer, Cham (2022)

Uploaded by

Copyright:

Available Formats

Problem Books in Mathematics

More information about this series at https://link.springer.com/bookseries/714

Carlos Yuzo Shine

ISSN 0941-3502 ISSN 2197-8506 (electronic)

Mathematics Subject Classification: 97G50, 97U40, 54H15, 51M04

There is a true geometry which is not [. . . ] intended to be merely an illustrative form of

Geometric Transformations was carefully crafted by three experienced mathe-

Lubbock, TX, USA Răzvan Gelca

Bucharest, Romania Ionuţ Onişor

São Paulo, Brazil Carlos Yuzo Shine

1 Where there is matter, there is geometry.

2.2.2 Examples of Problems Solved Using Homothety . . . . . . . . . . 88

4.2 A Story of Complete Quadrilaterals

Part III Solutions

1.1 Theoretical Results About Isometries

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 3

Fig. 1.1 How figures are transformed under isometries

Fig. 1.2 Passing from synthetic to analytic geometry

1.1.1 Definition and Basic Properties

It is desirable to reach a good level of understanding of the isometries before starting

|f (z) − f (w)| = |z − w| for all z, w ∈ C.

|f (z) − f (w)| = |r(z − w)| = |r||z − w| = |z − w|;

while if f (z) = rz + a, then

|f (z) − f (w)| = |r(z − w)| = |r||z − w| = |z − w|,

so in each case f is an isometry.

so g is an isometry, too. We have g(0) = 0 and g(1) = 1.

g(z) · g(z) = z · z and (g(z) − 1) · (g(z) − 1) = (z − 1)(z − 1).

g(z) · g(z) = z · z and g(z) + g(z) = z + z.

|z − w| = |g(z) − g(w)| = |z − w|,

rta + r(1 − t)b + s = t (ra + s) + (1 − t)(rb + s).

Fig. 1.3 Inverses and compositions of isometries

1.1.2 Translations, Rotations, Reflections

In this section we introduce some particular classes of isometries.

This clarifies the geometric meaning of the transformation defined by adding a

Fig. 1.4 A translation τ A = τ (A)

Fig. 1.5 A rotation ρ. The A = (A ) B

This is because of the identity

arg(zw) = arg(z) + arg(w),

which follows from trigonometry

k1 (cos α + i sin α)k2 (cos β + i sin β)

Theorem 1.3 The rotation by α about z0 is then given by the formula

ρ(z) = r(z − z0 ) + z0 , where r = cos α + i sin α.

Proof There is nothing to prove if α = 0. If α = 0, then the point z0 is the unique

arg r(z − z0 ) = arg r + arg(z − z0 ) = α + arg(z − z0 ).

eαi = cos α + i sin α.

correspond to the multiplicative properties of the exponential function

A complex number is written in trigonometric form as

z = reiα , where r = |z| and α = arg z.

Then the counterclockwise rotation about a point w by angle α is defined by the

which gives z = eαi z − eαi w + w.

Fig. 1.6 Definition of directed angles

If a, b, c are the complex coordinates of A, B, C, then the directed angle  ABC

 BAC +  CBA +  ACB =  (AB, AC) +  (BC, AB) +  (AC, BC) = 0.

Also, if A, B, C are not collinear, then a point D is on the circumcircle of ABC

c−a c−b a−c b−c

which we denote by (a, b, c, d) and call cross-ratio, seems to be important, if

(c, d, a, b) = (d, c, b, a) = (a, b, d, c)−1 . The cross-ratio depends only on the

where r, s ∈ C, |r| = 1 and s = 0.

Fig. 1.7 A reflection σ over a line 

Hence the formula.

Fig. 1.8 Orientation of triangles

1.1.3 Isometries as Composition of Reflections

It looks like the aforementioned transformations are only particular cases of

Fig. 1.9 Case 3

Proof There is nothing to prove if α = 0. If α = 0, then the point z0 is the unique

If a, b, c are the complex coordinates of A, B, C, then the directed angle ABC

BAC + CBA + ACB = (AB, AC) + (BC, AB) + (AC, BC) = 0.

where r, s ∈ C, |r| = 1 and s = 0.

Fig. 1.7 A reflection σ over a line

(OA, OA ) = (OA, OB) + (OB, OA )

Fig. 1.12 The case M = N B

If f preserves orientation, compose f with the reflection σ over , that is,

d(σ1 (P ), σ2 ◦σ1 (P )) = 2(d −d(P , 1 )). Therefore d(P , σ2 ◦σ1 (P )) = 2d(P , 1 )+

which is a rotation of angle arg(v12 v2 2 ) = 2 (1 , 2 ) about z0 .

to be able to take advantage of the fact that σ (O1 ) = O1 . But O1 ∈ , so ρ −1 (O1 ) ∈

σ2 = σb ◦ σ1 ◦ σa = σB ◦ σ1 ◦ σA .

sin BAM sin CAT

And we have BAM + CAM = CAT + BAT = α, so by denoting the

T BP = 180◦ − ABC − CBT = 180◦ − ABC − BAC

and similarly T CQ = T QC. It follows that the triangles T P B and T QC are

QAPB = P APB − 2 P AR = RAPB − P AR

So AQ bisects PB APC . Because APB = APC , the triangle APB P C is isosceles,