Plane Geometry

A Perspective Field looks like a constellation of stars at heaven. All these

points are like stars. Each star moves in its own way. Some stars adjourn
together in a flow. Just like in astronomy we just watch and sometimes we
understand why.

Perspective Fields
Chris van Tienhoven

Li4
[email protected]
Contents

Thanks To . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

I The Basics 1

0 Lengths and Angles - Hidden/Completed 2

0.1 Directed Angles and Line Arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

0.2 Trig and Vector Bashing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

0.3 Length Bashing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

0.4 Power of a Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1 Geometric Transformations - Hidden/Completed 40

1.1 Homothety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

1.2 Spiral Similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

1.3 Isogonal Conjugation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

1.4 Simson and Steiner lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

1.5 Isotomic Conjugation and Trilinear Polars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

1.6 Morley’s Equilateral Triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

2 The Cross Ratio - Hidden/Completed 75

2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

2.2 Harmonic Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

2.3 A Few Projective Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

1
AoPS Contents

3 Inversion and Polarity - Hidden/Completed 99

3.1 Basic Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

3.2 Basic Polarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

3.3 Cross Ratios under Inversion and Polarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

3.4 Apollonian Circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

3.5 Apollonius’s Circle Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

4 Complete Quadrilaterals - Hidden/Completed 126

4.1 Top 10 Points In a Complete Quadrilateral . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

4.2 Complete Cyclic Quadrilaterals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

4.3 Complete Tangential Quadrilaterals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

5 The Best Of Xn - Hidden/Completed 138

5.1 X1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

5.2 X2 through X5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

5.3 X6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

5.4 X7 through X10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

5.5 X11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

5.6 Xn , n < 99 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

5.7 Xn , n ≥ 99 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

5.8 Others . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

II The Deep End 165

6 Basic Conic Theory 166

6.1 Definitions and Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

6.2 Cross-Ratio on Conics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

6.A A Journey to The Hidden Circle Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

2
AoPS Contents

7 Projective Space 186

7.1 Conic Polarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

7.2 Involutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

7.3 Isogonal Conjugation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

7.4 General Isoconjugations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

7.A Revisiting the Cross Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

7.B Moving Points / The Polynomial Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

8 Circumrectangular Hyperbolas - Hidden/Completed 220

8.1 Special Conics and The Poncelet Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

8.2 Feuerbach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

8.3 Kiepert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237

8.4 Jerabek . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

9 Perfect Six-Point Sets and The Isoptic Cubic 251

9.1 Perfect Six-Point Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251

9.2 Perfection of Isogonal Conjugation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

9.3 Oblique Strophoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

9.4 Duality - The Isohaptic Locus of Isotomic Conjugation . . . . . . . . . . . . . . . . . . . . . . 270

10 Secrets of the Complete Quadrilateral 274

10.1 Steiner’s Hidden Deltoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274

10.2 Deltoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275

10.3 The Kantor-Hervey Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283

10.4 Morley’s Beautiful Cardioids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284

11 Basic Cubic Theory with Liang and Zelich 290

11.1 The Group Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290

11.2 The Path of Polarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298

11.3 Self-isogonal Isocubics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

11.4 Liang-Zelich . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309

3
AoPS Contents

12 Special Lines 317

12.1 Conjugate Conics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317

12.2 Zhang Zhihuan’s Permutation Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321

12.3 The Barycentric Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324

13 The Worst Of Xn 333

13.1 X1 - related points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

13.2 X1 - unrelated points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351

4
AoPS Contents

Thanks To

This translation was started on June 27th, 2024 and published on [todo].

Thanks to all of these people for contributing to the project (in no particular order)!

• Brandon Wang for letting us • [email protected] • [email protected]

use Overleaf Premium
• [email protected] • [email protected]
• [email protected]
• [email protected] • [email protected]
• [email protected]
• [email protected] • [email protected]
• [email protected]
• [email protected] • [email protected]
• [email protected]
• [email protected] • [email protected]
• [email protected]
• 39Geo • [email protected]
• [email protected]
• [email protected] • [email protected]
• [email protected]

• [email protected] • [email protected] • [email protected]

• [email protected] • [email protected] • [email protected]

• [email protected] • [email protected] • [email protected]

• [email protected] • [email protected] • [email protected]

If you have errata for this translation, please contact [email protected] or [email protected]
specifically, or submit a pull request. For any questions about the content of this book, Li4 is available to be
DM’ed on AoPS.

Conventions

We adopt some general conventions throughout this project. It’s best not to read these until you are confused
later on; this section should serve as a reference for strange notation.

Numbering is done with respect to the original book, and any additional remarks are not numbered and
done by the translators.

We typically take the default reference triangle △ABC and have a = BC, b = CA, c = AB be the three
sides respectively. We let α = ∡BAC, β = ∡CBA, γ = ∡ACB.

i
AoPS Contents

Regarding triangle style config geo specifically.

• I, I a , I b , I c respectively represent the incenter and the three excenters (IA and analogous may also be
used for the excenters); We refer to extraversion as the fact that properties true for I typically remain
true for I a , I b , I c as well.

• ω, ω a , ω b , ω c respectively represent the incircle and the three excircles, r, ra , rb , rc respectively represent
their radii;

• △DEF, △Da E a F a , △Db E b F b , △Dc E c F c respectively represent the intouch triangle and the A, B,
C-extouch triangles;

• △Ma Mb Mc represents the medial triangle;

• Ω represents the circumcircle, R represents the circumradius, A∗ , B ∗ , C ∗ represent the antipodes of

A, B, C;

• E can be the Euler line.

• ϵ represents the nine-point circle;

• △Ha Hb Hc represents the orthic triangle;

• △Na Nb Nc represents the circumcevian triangle of I, aka the triangle formed by arc midpoints,
Na ∗ , Nb ∗ , Nc ∗ respectively represent the antipodes of Na , Nb , Nc .

• L∞ is used for the line at infinity. We apologize for references to L∞ and points at infinity in Chapters
0 and 1, but keep them to remain faithful to the original textbook and for clarity.

• △(AB)(CD)(EF ) refers to the triangle with vertices at the pairwise intersections of the three lines.

• ΓA,B
r = {P | P A = r · P B} as the r-Apollonius circle.

• I and J are the circle points (1 : i : 0), (i : 1 : 0) in CP2 .

• △P Q means the circumconic of ABC through P, Q.

• (P Q)∗ means the circumconic of ABC through P ∗ , Q∗ (notation is motivated by the isogonal conjugate
of line P Q

• K a , K b , K c refer to the vertices of the tangential triangle, and Ga Gb Gc refer to the vertices of the
anticomplementary triangle.

Along with the original book, we use the following notation for transformations.

ii
AoPS Contents

• hO,k represents a homothety with center O and ratio k.

• h⃗v represents a translation of ⃗v .

• sO represents a reflection about O.

• rO,θ represents a rotation about O of θ.

• JΩ represents an inversion about a circle Ω.

• JO,k represents an inversion about O with power k.

• SO,k,θ = rO,θ ◦ hO,k represents a spiral similarity with ratio k and rotation θ.

• P ∁ = hG,− 12 (P ) represents the complement of a point P wrt. △ABC. P ∁,△ABC will be used if the
reference triangle is unclear.
∁ ∁
• P = hG,−2 (P ) represents the anticomplement of a point P wrt. △ABC. P ,△ABC
will be used if the
reference triangle is unclear.

• SP is the Steiner line of P on the circumcircle.

• t(P ) is the trilinear polar of P wrt. △ABC.

• P ′ can denote the image under of P under some transformation, or under isotomic conjugation.

• P ∗ can be used to refer the image of P under isogonal conjugation or the antipode with respect to a
circle.

• For some known A, B by context, and C on AB, we define C ∨ as the harmonic conjugate of C with
respect to {A, B}.

Regarding angles,

• We use ∠ for undirected angles, taken with config issues.

• We use ∡ABC = ∡(AB, BC) for directed angles, taken (mod 180◦ ). ∡(AB, CD) refers to the angle
between AB, CD.
−−→
• We use ∡AB for line arguments (mod 180◦ ). We use ⊥ AB = ∡AB + 90◦ . We use AB when things
are (mod 360◦ )

• We use directed points on a circle where (A + B)Ω = ∡AB and (A − C)Ω = ∡AXC for X ∈ Ω.

• We use degrees for the most part but use radians sometimes.

iii
AoPS Contents

Regarding conics, we define (where Γ is an arbitrary conic)

• TΓ as the set of all tangents of Γ; In the degenerate case where Γ is a point P , TP is the set of lines
through P .

• TP Γ as the tangent line to Γ at P , where P ∈ Γ;

• Tℓ Γ as the point of tangency of ℓ and Γ, where ℓ ∈ TΓ;

• pΓ (A) as the polar of a point A wrt. Γ (for higher degree curve K, pnK (P ) will represent the nth polar
of P wrt. K.)

• c(P ) is the circumconic of △ABC with P as perspector.

• H generally refers to a hyperbola.

• HF e , HK , HJ is the Feuerbach, Kiepert, Jerabek hyperbola.

• PerΓP (X) and LiΓP,Q (X) refer to the P -permutation line and P, Q-Li conjugate of X wrt. Γ. If Γ is
omitted, assume it to be the circumcircle.

For complete quadrilaterals, we define

• (AC)(BD) refers to the quadrilateral consisting of the four lines AB, BC, CD, DA

• Q as the complete quadrilateral (BC, CE, EF, F B)

• The Miquel point as MQ

• The Newton line as τQ

• The Steiner line as SQ .

• Aij = ℓi ∩ ℓj

Regarding barycentrics and point operations, where A and B are arbitrary points in a triangle, (note that
many of these terms are newly coined in this book; Chapter 12 may be a better reference if unfamiliar with
vocab.)

• A × B refers to the (termwise) barycentric product of the two points. Similarily, A ÷ B refers to the
barycentric quotient.

• A ∩ B refers to the common points on both curves.

• The crosspoint of A and B is A ⋔ B

iv
AoPS Contents

• the cevapoint is A ⋆ B.
⋔
• The cross conjugate is A B

• The ceva conjugate is A/B.

• ϕG and ϕK are isogonal and isotomic conjugation respectively.

• P D is the difference point of P .

• P s is the square transformation, and P r is the radical transformation.

Here’s a short but important exposition on notation and definitions of projective geometry used in this
book; for more detail and rigor check the Wikipedia pages. (for experts, only recommended to read after
Part I).

• RP1 is the real projective line, which is roughly represented with [x : y] in homogeneous coordinates up
to scaling (excluding [0 : 0]). and RP2 is the real projective plane (three variable variant). These can
roughly be thought of as {R} ∪ {P∞ } and {R2 } ∪ {ℓ∞ }. Analogously, CP1 and CP2 are the same over
complex numbers.

• When we refer to the point at infinity in P1 , we will typically use [x : 0] for some x ̸= 0. Similarily,
when we refer to the line at infinity in P2 , we will typically use [x : y : 0] for some (x, y) ̸= (0, 0).

• CP1 is called the complex projective line, and has the structure of having one point at infinity, and
inversion is done over it. Important: Note that we do not call this a plane.

• CP2 is called the complex projective plane, with coordinates [x : y : z]. Despite similarily to the
term “complex plane”, these are two very different objects! The “complex plane” typically
refers to C1 , which is technically a one-dimensional space (it’s only two-dimensional if you split the
complex numbers into real and imaginary parts), while this is two-dimensional (four, if you split into
imaginary and real parts). We apologize for the confusion.

• When we don’t care about the underlying scalar field, we will use Pn .

• The term “projective map” is used for an automorphism between P1 → P1 which preserves cross-ratio.

• A homography is an automorphism between Pn → Pn which preserves cross-ratio and collinearity/inci-

dence.

• A collineation is some transformation from Pn → Pn that preserves collinearity of points. In RP2 (but
not CP2 ), these transformations also preserve the cross-ratio and are thus homographies.

v
AoPS Contents

• In this book, “projective transformation” is used for an automorphism between P2 → P2 which preserves
cross-ratio and collinearity/concurrence. However, we will only use “projective transformation” when
the distinction between homography and collineation does not matter, such as when working in the real
projective plane.

• The circle points are the two complex points in CP2 of [1 : i : 0], [1 : −i : 0] that every circle passes
through. These will be denoted as I and J usually, unless this is confusing in context (incenters), in
which case we will use ∞i and ∞−i .

• A Möbius transformation / fractional linear transformation is a homography specific to CP1 ; these are
the combination of inversions and (orientation-preserving) spiral similarities.

A small amount of group theory knowledge will be assumed for this book, but it is not technically necessary.

Online Encyclopedias of points, lines, and figures mentioned in this text are as follows:

• Kimberling’s Encyclopedia of Triangle Centers here.

• Bernard Gilbert’s Catalogue of Triangle Cubics here.

• van Tienhoven’s Encyclopedia of Quadri-figures here.

vi
AoPS Contents

Part I: The Basics Part II: The Deep End

Ch 0

Ch 1 Ch 2 Ch 6

Ch 4 Ch 3 Ch 7

Ch 5 Ch 8 Ch 9

Ch 11 Ch 10

Ch 12 Ch 13

vii
 Part I

The Basics

1
Chapter 0

Lengths and Angles -

Hidden/Completed

Two points define a length between them, and two lines define an angle between them. Here, we assume that
lengths and angles are preserved under rotations, translations, and reflections.

0.1 Directed Angles and Line Arguments

♠: In this section, assume that all lines are not the line at infinity L∞ .

One geometric diagram can be drawn in many ways by different people. Sometimes a point can lie inside
another angle in one diagram but outside in another. These are what are popularly called “config issues.”
For example:

Example 0.1.1. If ∠BAC = 20◦ , ∠BAD = 40◦ , what is angle ∠CAD?

To this question, some will answer 20◦ while others will respond 60◦ . This is because we don’t know if
point C lies inside ∠BAD. This motivates a way to resolve this ambiguity. The common way to do this is
through directed angles.

Definition 0.1.2 (Directed Angles). For two rays / directed lines ℓ⃗1 , ℓ⃗2 (where ℓ1 , ℓ2 represent, respectively,
the undirected lines), we define the directed angle ∡(ℓ1 , ℓ2 ) between ℓ⃗1 , ℓ⃗2 as

1
(i) If ℓ1 , ℓ2 aren’t parallel, or ℓ1 ∩ ℓ2 ̸∈ L∞ , then it is the amount needed to rotate ℓ1 counterclockwise
to ℓ2 .
1 Here, we use ℓ1 ∩ ℓ2 as the intersection point of ℓ1 and ℓ2 .

2
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

(ii) If they are parallel, then it is 0◦ if they face the same direction and else 180◦

Relative to two vectors v⃗1 , v⃗2 , we define the angle ∡(v1 , v2 ) between v⃗1 , v⃗2 by the directed angle formed
by taking the vectors’ corresponding directed lines. To use the above definition, we have to take everything
modulo 180◦ = π to direct the lines. In other words, ∡(ℓ1 , ℓ2 ) ≡ ∡(ℓ⃗1 , ℓ⃗2 ) (mod π).

Definition 0.1.3 (Parallelism and Perpendicularity). Let ℓ1 , ℓ2 be two lines.

(i) If ∡(ℓ1 , ℓ2 ) = 0◦ , then we denote ℓ1 , ℓ2 as parallel, and we represent this as ℓ1 ||ℓ2 .

(ii) If ∡(ℓ1 , ℓ2 ) = 90◦ , then we denote ℓ1 , ℓ2 as perpendicular, and we represent this as ℓ1 ⊥ ℓ2 .

Proposition 0.1.4. Given any three directed lines ℓ⃗1 , ℓ⃗2 , ℓ⃗3 , we have that

∡(ℓ⃗1 , ℓ⃗2 ) + ∡(ℓ⃗2 , ℓ⃗3 ) = ∡(ℓ⃗1 , ℓ⃗3 ).

Proof. Check the original book for a more in depth proof, but the idea is case work based off which lines are
parallel.

Definition 0.1.5. Given any three points A, B, O, define

−→ −−→
∡AOB = ∡(OA, OB)

Now, let us revisit our beginning example with directed angles.

Example 0.1.6. If ∡BAC = 20◦ , ∡BAD = 40◦ , what is angle ∡CAD?

We can answer this now as

∡CAD = ∡BAD − ∡BAC = 20◦ .

Now, let us reformulate all our previous angle facts for undirected angles in terms of directed angles.

̸ P . Then
Proposition 0.1.7. Let P be a point and let A, B, C be points on a line such that A =
∡P AB = ∡P AC.

Proof. Follows since AB ∥ AC.

Proposition 0.1.8. If three points A, B, C satisfy CA = AB, then we have

∡CBA = ∡ACB (mod 360◦ )

and A lies on the perpendicular bisector of BC.

3
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Proof. See original proof for more details but this is effectively just isosceles triangle and considering angle
bisector.

(This proposition has a converse, namely A lying on the perpendicular bisector on BC / angle condition
with nonzero angle imply the other two.)

Proposition 0.1.9. Given a fixed △ABC, we consider the three perpendicular bisectors ℓA , ℓB , ℓC of BC,
CA, AB. These lines then concur at a point called the circumcenter O of △ABC. O is also the center of
the unique circle through A, B, and C, called the circumcircle.

Proof. Note that no three of the perpendicular bisectors which are parallel. Let O be the intersection of ℓB
and ℓC , then OC = OA and OA = OB, so OB = OC by the transitive property. As such, by Proposition 0.1.8
to get that O ∈ ℓA , and thus the three perpendicular bisectors of a triangle concur.

Since we have that OA = OB = OC, this implies that there is a unique circle centered at O through A,
B, and C. Conversely, the center of a circle through A, B, and C lies on ℓA , ℓB , and ℓC , and thus must be
O.

Proposition 0.1.10. Let O be the circumcenter of △ABC. Then ∡BAC + ∡CBO = 90◦ .

Proof. By Proposition 0.1.8, we have that

2 · ∡BAC = ∡BAC + ∡BAO + ∡OAC = ∡BAC + ∡OBA + ∡ACO

= ∡OBC + ∡BCO + 180◦ = 2 · ∡OBC + 180◦ (mod 360◦ )

Dividing by two finishes.

Example 0.1.11. Let O be the circumcenter of △ABC. Let ℓ = A∞⊥BC be the line going through A
perpendicular to line BC. Then we have

∡(AB, ℓ) + ∡(AC, AO) = ∡BAO + ∡(AC, ℓ) = (90◦ − ∡ACB) + (90◦ + ∡ACB) = 0◦ ,

so we have that ∡(AB, ℓ) = ∡(AO, AC).

In general, call any pair of lines (ℓ1 , ℓ2 ) isogonal lines in ∠A if ∡(AB, ℓ1 ) = ∡(ℓ2 , AC).

Proposition 0.1.12. Given any four points A, B, P , and Q, there is a circle passing through all four points
A, B, P , Q if and only if
∡AP B = ∡AQB.

2
We call such a quadrilateral cyclic. We call the points A, B, P , and Q as being concyclic.
2We will see later on in 6.A that this is equivalent to A, B, P , and Q being conconic with two fixed points at infinity.

4
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Proof. Assume no three of {A, B, P, Q} lie on one line. We seek to prove that △P AB and △QAB have
the same circumcenters, denote these respectively as OP and OQ . Note that both of these points lie on
the perpendicular bisector of AB. Thus our condition is equivalent to proving ∡BAOP = ∡BAOQ . By
Proposition 0.1.10, we get

∡AP B + ∡BAOP = 90◦ = ∡AQB + ∡BAOQ ,

so our statement is equivalent to ∡AP B = ∡AQB.

The above proposition tells us that given two points A, B and a fixed angle θ, then the locus of points
P such that ∡AP B = θ lie on a circle. In other words, {P | ∡AP B = θ) = Γ ∩ {A, B} for some circle Γ
through A and B.

When θ = 90◦ , we get by Proposition 0.1.10 that ∡BAO = 90◦ − ∡AP B = 0◦ , so O lies on AB. We call
this circle (AB), where AB is a diameter of said circle.

Often when angle-chasing in a cyclic quadrilateral configuration, we use the concept of antiparallel lines.

Definition 0.1.13. We say that a pair of lines are antiparallel (L1 , L2 ) with respect to (ℓ1 , ℓ2 ) if

∡(L1 , ℓ1 ) + ∡(L2 , ℓ2 ) = 0◦ .

This is equivalent to L1 ∩ ℓ1 , L1 ∩ ℓ2 , L2 ∩ ℓ1 , L2 ∩ ℓ2 lying on a circle.

We can check that this concept is symmetric in (L1 , L2 ), and also forms a equivalence relation (L1 , L2 ) ≻
(ℓ1 , ℓ2 ) on pairs of lines.

(i) (ℓ1 , ℓ1 ) ≻ (ℓ1 , ℓ2 ) follows immediately.

(ii) L ≻ K ⇐⇒ K ≻ L follows by symmetry.

(iii) A ≻ B, B ≻ C =⇒ A ≻ C follows by angle chasing.

Example 0.1.14 (Reim’s theorem). Let A1 , A2 , B1 , and B2 be concyclic, let C1 be a point on the line
A1 B1 . Prove that (B1 B2 C1 ), A2 B2 , and the line through C1 parallel to A1 A2 are concurrent.

Proof. Define C2 on A2 B2 such that C1 C2 ∥ A1 A2 . Then since (A1 A2 , B1 B2 ) is anti parallel with
(A1 B1 , A2 B2 ), it follows that (C1 C2 , B1 B2 ) and (B1 C1 , B2 C2 ) are antiparallel, so B1 , B2 , C1 , C2 are con-
cyclic.

Example 0.1.15. Let Γ be the circumcircle of an acute-angled triangle △ABC. Draw the angle bisector of
∠ABC, and let this line intersect segment AC at point B1 , let this line intersect arc AC at point P . Draw

5
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

the perpendicular line to BC from B1 and let this line intersect arc BC at point K. Draw the perpendicular
line to AK through B, let this intersect AC at point L. Prove that K, L, and P collinear.

A
P

B1

B C

Proof. Instead of proving that K, L, P collinear, we instead prove that ∡AKL = ∡AKP . Furthermore, we
also know that
∡AKP = ∡ABP = ∡P BC = ∡B1 BC = 90◦ − ∡KB1 B

and also ∡AKL = 90◦ − ∡KLB. Thus we just need to prove ∡KLB = ∡KB1 B, or just B, K, L, B1 concyclic.

Proving B, K, L, B1 concyclic is equivalent to proving ∡KB1 L = ∡KBL, so we just calculate

∡KB1 L = 90◦ − ∡ACB = 90◦ − ∡AKB = ∡KBL,

and we are done.

Theorem 0.1.16 (Triangle Miquel (known in Chinese as “three circle theorem”)). Given a fixed triangle
△ABC, let E, F be points on segments CA, AB. Let D be an arbitrary point. Then D lies on segment BC
iff (EAF ), (F BD), (DCE) are concurrent. (Furthermore, given a triangle △DEF with vertices on the three
sides of ABC, denote this point of concurrency as the Miquel point.)

Proof. Let P be the intersection point of (F BD) and (DCE) that’s not D. We want to prove that P being

6
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

located on (EAF ) is equivalent to point D lying on BC. We proceed with directed angles, we want to show

∡EAF = ∡EP F ⇐⇒ ∡BDC = 0◦ .

Expand out and get

∡BDC = ∡BDP + ∡P DC = ∡BF P + ∡P EC = ∡F AE + ∡EF P ,

thus,
∡EAF = ∡EP F ⇐⇒ ∡BDC = ∡F AE + ∡EF P = 0◦ .

Example 0.1.17. Let △Ma Mb Mc be the medial triangle of △ABC (Ma , Mb , Mc are the midpoints of their
respective sides). Prove that the Miquel point of △Ma Mb Mc is the circumcenter of △ABC.

Note: typically it’s hard to characterize the Miquel point of DEF wrt. ABC. It’s very hard to characterize
the angle ∡BAM given △DEF and △ABC.

Proof. It’s equivalent to show that O lies on (AMA MB ) which finishes by symmetry. Then note that
∡AMA O = ∡AMB O = 90◦ which finishes.

Now consider the above theorem when D, E, F are collinear and lie on AB, BC, CA. Then we get that
(AEF ), (DEC), (BDF ) have a Miquel point. However, since B, D, C are also collinear and lie on AF, F E, AE,
it follows that (ABC) also goes through this point. We get the following.

Theorem 0.1.18 (Miquel’s Theorem). Let ℓ1 , ℓ2 , ℓ3 , ℓ4 be four lines such that no 3 are concurrent. Then the
circumcircles of the four triangles formed by the intersections of these four lines are concurrent at a point
called the Miquel point with respect to the four lines / that quadrilateral.

7
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

A B Q

The Miquel point is the most important point of a quadrilateral, and we will further address this in
Chapter 4.

Definition 0.1.19. Given two triangles △A1 B1 C1 and △A2 B2 C2 ,

(i) We call △A1 B1 C1 and △A2 B2 C2 to be directly similar if all the corresponding directed angles are
equal, and inversely similar if the corresponding directed angles have opposite sign.

Denote this as
+ −
△A1 B1 C1 ∼ △A2 B2 C2 , △A1 B1 C1 ∼ △A2 B2 C2

respectively.

(ii) We call △A1 B1 C1 and △A2 B2 C2 directly congruent if they are directly similar and corresponding side
lengths are equal, and inversely congruent if they are inversely similar and have equal side lengths.

Denote this as
+ −
△A1 B1 C1 ∼
= △A2 B2 C2 , △A1 B1 C1 ∼
= △A2 B2 C2

respectively.

(iii) Finally, we can just call △A1 B1 C1 and △A2 B2 C2 to be similar or congruent if we don’t care about
directed angles, denoted as ∼ and ∼
= respectively.

Proposition 0.1.20. If △A1 B1 C1 and △A2 B2 C2 are similar, we have

B1 C1 C1 A1 A1 B 1
= = .
B2 C2 C2 A2 A2 B 2
8
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

To prove this we will first prove a sub-lemma:

Lemma 0.1.21. If A1 , B1 , C1 and A2 , B2 , C2 lie on two different lines ℓ1 and ℓ2 respectively, and
A1 A2 ||B1 B2 ||C1 C2 , then
A1 B 1 A2 B 2
= .
B1 C 1 B2 C 2

Note that in this lemma we can exchange some of these parallelisms with concurrencies: for example if
A1 = A2 and B1 B2 ||C1 C2 then the lemma stll holds.

Proof. Since
A1 B 1 A2 B 2 A1 C 1 A2 B 2
= ⇐⇒ =
B1 C1 B2 C 2 C 1 B1 B2 C2

we can “swap the sequence” of A, B, C.

By the fact that A1 A2 ||B1 B2 , we get

[△A1 B1 B2 ] = [△A2 B1 B2 ],

(brackets represent the (signed) area of the triangle). Similarly, we also have that

[△C1 B1 B2 ] = [△C2 B1 B2 ],

and thus
A1 B 1 [△A1 B1 B2 ] [△A2 B1 B2 ] A2 B 2
= = =
B1 C1 [△B1 C1 B2 ] [△B1 C2 B2 ] B2 C2

Proof of (0.1.20). Suppose △A1 B1 C1 and △A2 B2 C2 were inversely similar. Then just reflect A2 over B2 C2
to get △A1 B1 C1 and △A′2 B2 C2 are directly similar.

Thus we can assume WLOG that △A1 B1 C1 and △A2 B2 C2 are directly similar.

Let us first look at the translation sending A2 to A1 . Let this translation send △A2 B2 C2 to △A1 B2′ C2′ .
(Now points A2 , A1 are the same so we call them both A). Now take a rotation of AB2′ C2′ around A such
that AB2′ becomes AB1 . Since they are directly similar we have that ∡B1 AC1 = ∡B2 AC2 , so this rotation
also sends A1 C2′ to A1 C1 . Thus we now have A, B1 , B2′ collinear and A, C1 .C2′ collinear. Also note that

∡(B1 C1 , B2′ C2′ ) = ∡C1 B1 A − ∡C2′ B2′ A = 0◦

9
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

so we know that B1 C1 ||B2′ C2′ . Now we can use our previous lemma! We get

C1 A B1 A AB1
= ′ =
C2′ A B2 A AB2′

Doing this for each other vertex finishes.

0.1.1 Further Analysis Into Line Arguments

When we calculate angles, we’re actually just calculating the difference between the “arguments” of two lines
(arc-tangent of slopes). So sometimes when we have an “inaccessible” angle in a problem, we can instead just
look at the two lines that form the angle. If we know the arguments of these lines, then we can just subtract
them to get the “inaccessible” angle.

(Note that this also implies usage of line arguments is equivalent to drawing a bunch of parallel lines in
your problem.)

Here’s some examples of what I mean:

(i) Instead of ∡(ℓ1 , ℓ2 ), we can express this in line arguments as ∡ℓ2 − ∡ℓ1 .

(ii) Instead of ∡(ℓ1 , ℓ2 ) + ∡(ℓ2 , ℓ3 ) = ∡(ℓ1 , ℓ3 ), this becomes (∡ℓ2 − ∡ℓ1 ) + (∡ℓ3 − ∡ℓ2 ) = ∡ℓ3 − ∡ℓ1 .

(iii) Two lines are parallel iff they have the same line arguments, or ℓ1 ∥ ℓ2 ⇐⇒ ∡ℓ1 = ∡ℓ2 .

(iv) Two lines (L1 , L2 ) are antiparallel with respect to (ℓ1 , ℓ2 ) iff.

(∡ℓ1 + ∡ℓ2 ) − (∡L1 + ∡L2 ) = ∡(L1 , ℓ1 ) + ∡(L2 , ℓ2 ) = 0.

or equivalently ∡L1 + ∡L2 = ∡ℓ1 + ∡ℓ2 . (This also will give a condition for finding cyclic quadrilaterals,
which we will address later.)

This lets us move away from the world of geometry and enter into the world of addition and subtraction
(modulo 180◦ , note that ∡ℓ = ∡ℓ + 180◦ ).
→
−
We use the shorthand ⊥ ℓ = ∡ℓ + 90◦ as well. We also use ℓ when we want to do things mod 360◦ .

We can now expand our definition for directed angles, by defining

∡(L1 + · · · + Ln , ℓ1 + · · · + ℓn ) := (ℓ1 + · · · + ℓn ) − (L1 + · · · + Ln )

= ∡(L1 , ℓ1 ) + · · · + ∡(Ln , ℓn )

10
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

and noting that swapping ℓi , ℓj and swapping Li , Lj does not change this directed angle.

Now, the most powerful use of line arguments is in cyclic quadrilaterals. We know that if A, B, P, Q are
cyclic, then ∡AP B = ∡AQB. Symbolically, this is just

P B − P A = QB − QA.

(This can also be shown from antiparallelism).

Now, by repeating this in fact we actually also have

∡AP + ∡BQ = ∡AQ + ∡BP

∡AP + ∡BQ = ∡AB + ∡P Q

by further exchanging.

By merging this all into one, we have that

∡AB + ∡P Q = ∡AP + ∡BQ = ∡AQ + ∡BP

This is the power of line arguments. We have three equalities from just a simple cyclic quadrilateral! In
fact, proving any of these equalities is equivalent to proving A, B, P, Q cyclic.

We use the notation (A + B + P + Q)Γ to represent how every pair has the same sum. In general,
(A + C)Γ = ∡AC and (A − C)Γ = AΓ − CΓ . These will be used somewhat irregularly but the above guarantees
that all the operations are well defined.

Let’s give another example. Let’s convert the proof of Miquel’s theorem to a proof with line arguments.
We want to prove that D lies on BC iff (EAF ), (F BD), (DCE) concyclic, where E, F are known to lie on
CA and AB respectively.

Let P ̸= D be the second intersection of (F BD) and (DCE). Then we have

∡P E = ∡P D + ∡CE − ∡CD = ∡P D + ∡AE − ∡CD,

∡P F = ∡P D + ∡BF − ∡BD = ∡P D + ∡AF − ∡BD.

And D lying on BC is equivalent to the argument condition of BD = CD. So this lets us just subtract
the two above equations and get that

∡P E − ∡P F = ∡P D − ∡P D + ∡AE − ∡AF − ∡CD + ∡BD = ∡AE − ∡AF

11
 AoPS Chapter 0. Lengths and Angles - Hidden/Completed

so A, P, E, F are cyclic, we are done. The converse is similar.

Example 0.1.22. Let O1 , O2 respectively be the two centers of circles c1 and c2 , and let c1 and c2 intersect
at points A and B. Additionally, let O2 be outside c1 and similarly for O1 . Let line O1 A intersect c2 at point
P2 ̸= A, let line O2 A intersect c1 at point P1 ̸= A. Prove that O1 , O2 , P1 , P2 , B are concyclic.

P2
d

P1 e

A
c

O2

O1

Solution. We use line arguments. Note that since △O1 AP1 and △O2 AP2 are isosceles,

∡O1 P1 + ∡O2 P2 = (2∡AP1 − ∡O1 A) + (2∡AP2 − ∡O2 A) = ∡O1 P2 + ∡O2 P1

so O1 , O2 , P1 , P2 are cyclic.

Now, by Proposition 0.1.10, we get that

∡BP1 − ∡BP2 = (90◦ − (∡AB + ∡AP1 − ∡AO1 )) − (90◦ − (∡AB + ∡AP2 − ∡AO2 ))

= (2∡AP2 − ∡AO1 ) − ∡O1 P2 = ∡O1 P1 − ∡O1 P2

12
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

so O1 , B, P1 , P2 is cyclic.

Another nice formula about line arguments is Steiner’s theorem, in (TODO 1.4.2).

The rest of this section’s contents will be a rigorous definition of line arguments, feel free to skip it. We
will look at the addition group

k
( )
M X
A= Zℓ := ni ℓi k ∈ N ∪ {0}, ni ∈ Z, ℓi is a line
ℓ i=1

with its addition defined as the natural

k
X k
X k
X
ni ℓ i + mi ℓi = (ni + mi )ℓi .
i=1 i=1 i=1

Given a line ℓ0 , we define the homomorphism ϵ

ϵ: A −−−−−→ Z

k
X k
X
ni ℓi 7−−→ ni
i=1 i=1
P P
and this homomorphism has a kernel K = ker ϵ = { ni ℓi | ni = 0} spanned by ℓ − ℓ0 , as such

X X M
ni ℓi = ni (ℓi − ℓ0 ) ∈ K = Z(ℓ − ℓ0 )
ℓ̸=ℓ0

And now we can define directed angles as taking a mapping mod 180.

◦
∡: K −−−−−−→ R ⧸180◦

k
X k
X
ni (ℓi − ℓ0 ) 7→ ni · ∡(ℓ0 , ℓi ).
i=1 i=1

By Proposition 0.1.4 we know that

∡
(ℓ2 − ℓ1 ) = (ℓ2 − ℓ0 ) − (ℓ1 − ℓ0 ) 7→ ∡(ℓ0 , ℓ2 ) − ∡(ℓ0 , ℓ1 ) = ∡(ℓ1 , ℓ2 ),

or in other words, “addition” is preserved. We can then do our math completely in K and then convert to
actual angles to finish.

13
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

As such, ∡ is a homomorphism as well. This allows further abuse notation such as the following.

∡(ℓ1 + · · · + ℓk , L1 + · · · + Lk ) = ∡((L1 + · · · + Lk ) − (ℓ1 + · · · + ℓk ))

= ∡(ℓ1 , L1 ) + · · · + ∡(ℓk + Lk ).

As seen previously, we can swap any two lines in this and it won’t change the directed angles, so we can
write the following “equality”:

L1 + · · · + Lk − (ℓ1 + · · · + ℓk ) = ∡(ℓ1 + · · · + ℓk , L1 + · · · + Lk )

and if ∡(ℓ1 + · · · + ℓk , L1 + · · · + Lk ) = 0, we can define this equivalence relation:

L1 + · · · + Lk = ℓ1 + · · · + ℓk .

Note that all of this stuff holds for directed lines (i.e. rays) if we work in R◦ /360◦ instead of R◦ /180◦ .

Practice Problems

Problem 1 (Orthocenter). Let D be the foot from A to BC and defne E and F analogously.

(i) Prove that B, C, E, and F are concyclic. Then similarly, C, A, F , D and A, B, D, E are concyclic.

(ii) Let H be the Miquel Point of △ABC with respect to △DEF . Prove that A, H, D are collinear.
Similarly, B, H, E and C, H, F are collinear.

Hence the three altitudes AD, BE, CF share a point H. We call H the orthocenter of △ABC, and △DEF
the orthic triangle of △ABC. Note that A, B, C are also the orthocenters of △BHC, △CHA, △AHB
respectively, so we can also define the four points (A, B, C, H) to be an orthocentric system.

(iii) Prove that ∠BHC = 180◦ − ∠BAC. Thus if HA is the reflection of H about BC, then HA lies on
(ABC).

Problem 2. Let A, B, C, D be four concyclic points. Define A′ , C ′ as the foots of the altitudes from A, C
onto BD, respectively. Similarly, let B ′ , D′ be the foots of the altitudes from B and D onto AC, respectively.
Prove that the points A′ , B ′ , C ′ , D′ are concyclic.

Problem 3 (Miquel’s Pentagon Theorem, Five Circle Theorem in Chinese). Let ABCDE be a convex
pentagon, with A′ = BC ∩ DE, and B ′ , C ′ , D′ , E ′ defined analogously. Let A′′ be the intersection of
(EAC ′ ) and (ABD′ ) other than A, and define B ′′ , C ′′ , D′′ , E ′′ analogously. Prove that A′′ , B ′′ , C ′′ , D′′ , E ′′
are concyclic.

14
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Problem 4. Let E and F be two points on the side BC of the convex quadrilateral ABCD such that B,
E, F , C are on line BC in that order. Given that ∠BAE = ∠F DC and ∠EAF = ∠EDF , prove that
∠F AC = ∠BDE.

Problem 5 (APMO 2010/1). Let ABC be a triangle with ∠BAC ̸= 90◦ . Let O be the circumcenter of the
triangle ABC and Γ be the circumcircle of the triangle BOC. Suppose that Γ intersects the line segment
AB at P different from B, and the line segment AC at Q different from C. Let ON be the diameter of the
circle Γ. Prove that the quadrilateral AP N Q is a parallelogram.

Problem 6 (ISL 2007 G2). Let M be the midpoint of side BC in an isosceles triangle △ABC with AC = AB.
Let X be a point on minor arc M A of (AM B). Let T be a point inside of ∠BM A such that ∠T M X = 90◦
and T X = BX. Prove that ∠M T B − ∠CT M does not depend on the choice of X.

Problem 7. Let △ABC be an acute triangle where AB < AC. Let D, E respectively be the midpoints of
segments CA, AB, and let P be the second intersection of (ADE) and (BCD), similarily let Q be the second
intersection of (ADE) and (BCE). Prove AP = AQ.

Problem 8. Let ABC be a triangle with circumcenter O and circumcircle ω. Let D be a point on segment
BC such that ∠BAD = ∠OAC. Let E = AD ∩ ω. Let M , N , and P be the midpoints of BE, OD, AC
respectively. Prove that M , N , P are collinear.

0.2 Trig and Vector Bashing

Definition 0.2.1. Given an angle θ ∈ (0◦ , 90◦ ), consider △ABC with ∠ABC = 90◦ and ∠BAC = θ
(mod 360◦ ). We define
BC AB
sin θ = and cos θ = .
CA CA
Note that this definition is independent of the choice of △ABC (by Proposition 0.1.20). We extend this to
any angle θ (mod 360◦ ) by the following relations:

cos 0◦ = sin(90◦ ) = 1,

sin θ = sin(180◦ − θ) = − sin(−θ),

cos θ = − cos 180◦ − θ = cos(−θ).

In addition, define
sin θ
tan θ = ∈ R ∪ {∞}.
cos θ

Under this definition, | sin θ| is well-defined for an undirected angle, and tan θ is well defined for directed

15
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

angles (mod 180◦ ). We can also check that

sin θ = cos(90◦ − θ).

Proposition 0.2.2 (Law of Sines). If the circumradius of ABC is R, then

BC CA AB
= = = 2R.
| sin ∠BAC| | sin ∠CBA| | sin ∠ACB|

Proof. Let B ∗ be the antipode of B on (ABC) - then BB ∗ = 2R and

2R · | sin ∠BAC| = 2R · | sin ∠BB ∗ C| = BC.

Proposition 0.2.3 (Law of Cosines). For a triangle △ABC, let a = BC and analogously define b, c. Then

b2 + c2 − a2
cos ∡BAC = .
2bc

Proof. Let D be the foot from A onto BC. Let α = ∡BAC (mod 360◦ ) and define β and γ analogously.
Then
a = |BD + DC| = c cos β + b cos γ =⇒ a2 = ca cos β + ab cos γ.

Similarly, we have that

b2 = ab cos γ + bc cos α, c2 = bc cos α + ca cos β.

Thus, combining these three equations gives the result.

Corollary 0.2.4 (Pythagorean Theorem). If △ABC satisfies ∠BAD = 90◦ , then

BC 2 = CA2 + AB 2 .

We also have the identity sin2 θ + cos2 θ = 1.

In addition, we have

Corollary 0.2.5 (Triangle Inequality). For any △ABC,

BC < CA + AB.

16
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Proof. Let a = BC and analogously for b, c. Note that

cos2 α = 1 − sin2 α < 1

so
b2 + c2 − a2
−1 ≤ cos α = =⇒ −2bc < b2 + c2 − a2 =⇒ a2 < (b + c)2 .
2bc

Taking the square root gives that BC = a < b + c = CA + AB.

Since cos : (0◦ , 180◦ ) → R is injective (see Problem 2 of this section), we can combine this with the Law
of Cosines to obtain the converse of Proposition 0.1.20 - namely that if

B1 C 1 C1 A1 A1 B1
= =
B2 C 2 C2 A2 A2 B2

then △A1 B1 C1 ∼ △A2 B2 C2 . Using Law of Sines as well, we obtain thesimilarity rules AA, SAS, RHS, and
we can generalized AA and SAS to directed angles.

Proposition 0.2.6. Given point P on circle Γ with center O, then a line L passing through P is tangent to
Γ if and only if L ⊥ OP .

Proof. For any point Q ∈ L, by Pythagoras,

OQ2 = OP 2 + P Q2 ≥ OP 2 =⇒ OQ ≥ OP,

where equality occurs if and only if P = Q. Therefore, L is tangent to Γ.

Using this allows us to relate the angle of the tangent line with an angle of a chord in the circle:

Proposition 0.2.7. Given △ABC, a line L passing through A is tangent to (ABC) if and only if

∡(L, CA) = ∡ABC.

Proof. Let O be the circumcenter of △ABC. Then by Proposition 0.2.6, L is tangent to (ABC) if and only
if L ⊥ OA. From Proposition 0.1.10,
∡OAC = ∡ABC + 90◦

. Therefore, L ⊥ OA if and only if ∡(L, CA) = ∡ABC.

17
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Remark. This can also be expressed as the limiting case of antiparallelism: If L is a line through A on
(ABC), then L is tangent iff (L, BC) = (CA, AB). Using line arguments, this writes as

L + BC = “AA′′ + BC = AB + AC.

Let’s see some applications.

Example 0.2.8 (Argentina 2011/3). Let ABCD be a trapezoid with bases BC ∥ AD, where AD > BC,
and non-parallel legs AB and CD. Let M = AC ∩ BD. Let Γ1 be a circle that passes through M and is
tangent to AD at point A; let Γ2 be a circle that passes through M and is tangent to AD at point D. Let
S = AB ∩ CD, X ̸= A be Γ1 ∩ AS, Y ̸= D be Γ2 ∩ DS, and O be the circumcenter of triangle ASD. Show
that SO ⊥ XY .

Proof. From BC ∥ AD,

∡M XB = ∡M XA = ∡M AD = ∡M CB

so X lies on (BCM ). Similarly, Y lies on (BCM ). Now (BC, XY ) are antiparallel about (SA, SD). By
Example 0.1.11, (SO, S∞⊥AD ) are antiparallel about (SA, SD). Hence (BC, XY ) are antiparallel about
(SO, S∞⊥AD ). By using BC ∥ AD again,

∡(SO, XY ) = ∡(BC, S∞⊥AD ) = 90◦ .

Definition 0.2.9. For two vectors v⃗1 and v⃗2 , we define the dot product as

v⃗1 · v⃗2 = |v⃗1 | · |v⃗2 | · cos ∡(v⃗1 , v⃗2 )

We also define the outer product as

v⃗1 × v⃗2 = |v⃗1 | · |v⃗2 | · sin ∡(v⃗1 , v⃗2 ).

This is equivalent to the magnitude z component to the normal dot product and cross product of two
vectors in the XY plane. We can also show that the following hold:

v⃗1 · (av⃗2 ) = (av⃗1 ) · v⃗2 = a(v⃗1 · v⃗2 ),

v⃗1 × (av⃗2 ) = (av⃗1 ) × v⃗2 = a(v⃗1 × v⃗2 ),

18
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

as well as the fact that v⃗1 · v⃗2 = v⃗2 · v⃗1 , v⃗1 × v⃗2 = −v⃗2 × v⃗1 . Finally, by definition the dot product of
perpendicular vectors and the cross product of parallel vectors are both 0.

Proposition 0.2.10. The dot and cross products are distributive, meaning for any vectors v⃗1 , v⃗2 , v⃗3 , we
have

v⃗1 · (v⃗2 + v⃗3 ) = v⃗1 · v⃗2 + v⃗1 · v⃗3

v⃗1 × (v⃗2 + v⃗3 ) = v⃗1 × v⃗2 + v⃗1 × v⃗3

Proof. Feel free to check the original chinese version but the idea is just to write everything out and deal
with config issues. (Alternatively, this is just immediate based off the definitions over R3 ).

Proposition 0.2.11 (Perpendicularity Criterion). For four points A, B, C, D, AB ⊥ CD holds iff AC 2 −

AD2 = BC 2 − BD2 .

Proof. Note that

(AC 2 − AD2 ) − (BC 2 − BD2 ) = (AC + AD) · (AC − AD) − (BC + BD) · BC − BD)

= −(AC + AD − BC − BD) · −2AB · CD = −2AB · CD

which implies the result.

Proposition 0.2.12 (Sine Area Formula). In any triangle △ABC, the directed area is

1 −−→ −→ 1
[ABC] = AB × AC = AB · AC · sin ∡BAC.
2 2

Proof. The core idea is that it just reduces to the right angle case, which is immediate.

Formally, if D is the foot from B to AC, then

−−→ −→ −−→ −−→ −−→ −−→ −−→ −−→ −−→ −−→

AB × AC = AB × AD + AB × DC = DA × DB + DB × DC

which reduces to the cases of △DAB, △DBC. Both of these cases can be easily checked.

Notably, the directed area of a triangle is positive if it is oriented counterclockwise and else negative.

Corollary 0.2.13 (Heron’s Formula). For any △ABC, let a = BC, b = CA, c = AB. Let s = 12 (a + b + c)
be the semiperimeter. Then
p
[ABC] = s(s − a)(s − b)(s − c).

19
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Proof. By Sine Area Formula, Law of Cosines, Pythagorean Theorem, we get that
s  2 2
bc bc p 2
bc b + c2 − a2
|[△ABC]| = · |sin ∡BAC| | = 1 − cos ∡BAC = · 1−
2 2 2 2bc
1 p 1 p
= · (2bc)2 − (b2 + c2 − a2 )2 = · ((b + c)2 − a2 )(a2 − (b − c)2 )
4 4
p
= s(s − a)(s − b)(s − c).

Practice Problems

Problem 1 (Sum and Difference Formulas). Show that for any two angles θ1 , θ2 , we have that

(i) sin(θ1 + θ2 ) = sin θ1 cos θ2 + cos θ1 sin θ2

(ii) sin(θ1 − θ2 ) = sin θ1 cos θ2 − cos θ1 sin θ2

(iii) cos(θ1 + θ2 ) = cos θ1 cos θ2 − sin θ1 sin θ2

(iv) cos(θ1 − θ2 ) = cos θ1 cos θ2 + sin θ1 sin θ2 .

Problem 2. Use Problem 1 to show that cos : [0◦ , 90◦ ] → [0, 1] is strictly decreasing and hence bijective over
[0◦ , 180◦ ].

Problem 3 (Sum and Product Formulas). Show that for any two angles θ1 , θ2 , we have that

(i) sin θ1 ± sin θ2 = 2 sin( θ21 ± θ2 θ1

2 ) cos( 2 ∓ θ2
2 ).

(ii) cos θ1 + cos θ2 = 2 cos( θ21 + θ2 θ1

2 ) cos( 2 − θ2
2 ).

(iii) cos θ1 − cos θ2 = −2 sin( θ21 + θ2 θ1

2 ) sin( 2 − θ2
2 )

(iv) sin θ1 · sin θ2 = 12 (cos(θ1 − θ2 ) − cos(θ1 + θ2 ))

(v) sin θ1 · cos θ2 = 12 (sin(θ1 + θ2 ) + sin(θ1 − θ2 ))

(vi) cos θ1 · cos θ2 = 12 (cos(θ1 − θ2 ) + cos(θ1 + θ2 ))

Problem 4. Let H be the orthocenter of △ABC. Show that △ABC, △HBC, △HAB, △HAC have the
same circumradius.

Problem 5 (Austria Federal 2019/2/5). Let ABC be an acute-angled triangle. Let D and E be the feet of
the altitudes on the sides BC and AC, respectively. Points F and G are located on the lines AD and BE so
AF BG
that FD = GE . The line passing through C and F intersects BE at point H and the line passing through C
and G intersects AD at point I. Prove that points F , G, H and I lie on a circle.

20
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

0.3 Length Bashing

Counting lengths is a powerful tool (compared to calculating angles). Many questions boil down to calculating
lengths. Now, let’s see what things we can cook.

Theorem 0.3.1 (Menelaus). Given any △ABC with points D, E, F on lines BC, CA, AB, respectively, we
have that points D, E, F are collinear if and only if

BD CE AF
· · = −1.
DC EA F B

AB
(here, the ratio BC is negative if B lies outside AC)

Proof. Let D′ = EF ∩ BC and let X be on BC such that AX ∥ EF .

Then we get that

CE AF CD′ D′ X CD′
· =− ′ = ′ =−
EA F B DX DB BD′
BD BD ′
As such, D = D′ only holds iff DC = D′ C , which is equivalent to their product being −1.

Theorem 0.3.2 (Ceva). Given any △ABC with points D, E, F on lines BC, CA, AB, respectively, we have
that AD, BE, CF , concur if and only if

BD CE AF
· · = 1.
DC EA F B

Proof. Let P = BE ∩ CF and D′ = AP ∩ BC. Then by Menelaus,

CB D′ P AP D′ C D′ C
   
CE AF
· = − ′
· · − ′
· = .
EA F B BD P A P D CB BD′

BD BD ′
We have that D = D′ iff DC = D′ C , which implies the result.

Example 0.3.3. Let △Ma Mb Mc be the medial triangle of △ABC. Then

BMa CMb AMb

· · = 1 · 1 · 1 = 1,
Ma C Mb A Mb B

so we get that AMa , BMb , CMc concur. We call this concurrence point the centroid of △ABC.

Proposition 0.3.4 (Ratio Lemma). Pick a point D on side BC of △ABC, then

BD AB · sin ∡BAD
= .
DC CA · sin ∡DAC

21
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Proof. By the Sine Area Formula, we get

1
BD [△ABD] 2 · DA · AB · sin ∡BAD AB · sin ∡BAD
= = 1 = .
DC [△ADC] 2 · CA · AD · sin ∡DAC CA · sin ∡DAC

Note that the ratio

sin ∡BAD
sin ∡DAC
−−→
is the same regardless of the sign of the directed length AD. So we can write the above ratio as

−−→
sin ∡(AB, AD)
−→ .
sin ∡(AD, AC)

With this property, we can rewrite Menelaus’s theorem in terms of angles.

Theorem 0.3.5 (Trig Menelaus). Given a △ABC and lines d, e, f on BC, CA, AB respectively, and points
D = BC ∩ d,E = BC ∩ e,F = AB ∩ F , D, E, F are collinear iff

(i) It holds that

−−→ −−→ −→
sin ∡(AB, d) sin ∡(BC, e) sin ∡(CA, f )
−→ · −−→ · −−→ = −1.
sin ∡(d, AC) sin ∡(e, BA) sin ∡(f, CB)

(ii) For some fixed point P not on AB, BC, CA, we have that

sin ∡BP D sin ∡CP E sin ∡AP F

· · = −1.
sin ∡DP C sin ∡EP A sin ∡F P B

Proof. We reduce this to normal Trig Menelaus. Using Ratio Lemma, we get

Y BD Y AB · sin ∡(− −→
AB, d) Y sin ∡(AB, d
= −→ = .
cyc
DC cyc CA · sin ∡(d, AC) cyc
sin ∡(d, AC)

Similarly, we have that

Y BD Y P B · sin ∡BP D Y sin ∡BP D

= = .
cyc
DC cyc
CP · sin ∡DP C cyc
sin ∡DP C

Remark 0.3.6. You may have noticed that the contribution of the sign of AB, BA is irrelevant due to
cancelling out.

22
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

So this really could also be written as

Y sin ∡(AB, d) sin ∡(BC, e) sin ∡(CA, f )

· · = −1.
sin ∡(d, CA) sin ∡(e, AB) sin ∡(f, BC)

Example 0.3.7. Given △ABC, let the tangents to (ABC) at A, B, C intersect BC, CA, AB at X, Y, Z,
respectively. Prove that X, Y, Z are collinear.

Solution. We investigate the quantity

−−→
sin ∡(AB, AX)
−→
sin ∡(AX, AC
which finishes. Note that this quantity is always negative because AX lies outside AB, AC.

As such, using tangent, angle chasing, we get that

−−→
sin ∡(AB, AX) sin ∡BAX sin ∡BCA
−→ = − sin ∡XAC = − sin ∡ABC
sin ∡(AX, AC)

so the cyclic product of the above cancels out to −1.

Ceva also has a trig version.

Theorem 0.3.8 (Trig Ceva). Given a △ABC and lines d, e, f through A, B, C respectively, then d, e, f are
concurrent iff

(i) It holds that

−−→ −−→ −→
sin ∡(AB, d) sin ∡(BC, e) sin ∡(CA, f )
−→ · −−→ · −−→ = 1.
sin ∡(d, AC) sin ∡(e, BA) sin ∡(f, CB)

(ii) For some fixed point P not on AB, BC, CA, and points D = BC ∩ d,E = BC ∩ e,F = AB ∩ F ,

sin ∡BP D sin ∡CP E sin ∡AP F

· · = 1.
sin ∡DP C sin ∡EP A sin F P B

Proof. It’s the same as for Trig Menelaus, except you do it for Ceva. Apply ratio lemma again.

Remark. Note that in the above proof of Menelaus we can see that we really only needed P, Q, R to satisfy

BP CQ AR
· · = 1.
P C QA RB

This lets us see a certain duality in Ceva and Menelaus’s theorems, (which we will formalize later, in
chapter 3), but for now we can notice that D, E, F collinear and AD, BE, CF concurrent are both conditions
on
sin ∡BP D sin ∡CP E sin ∡AP F
· · = ±1
sin ∡DP C sin ∡EP A sin ∡F P B
23
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

just with a change of sign.

Example 0.3.9 (Ceva for Circles). Let AF BDCE be a convex hexagon inscribed in a circle. Then AD,
BE, CF concur if and only if
AF · BD · CE = EA · F B · DC.

Solution. By Trig Ceva, AD, BE, and CF concur if and only if

−−→ −−→ −→
sin ∡(AB, AD) sin ∡(BC, BE) sin ∡(CA, CF )
−→ · −−→ · −−→ = 1.
sin ∡(AD, AC) sin ∡(BE, BA) sin ∡(CF , CB)

Because AF BDCE is convex, we know that each of these sines is positive, so by the Law of Sines we have

−−→
sin ∡(AB, AD) sin ∡BAD BD
−→ = sin ∡BAC = DC .
sin ∡(AD, AC)

Then AD, BE, and CF concur if and only if

BD CE AF
· · = 1,
DC EA F B

which is equivalent to the theorem statement. This theorem is relevant when we discuss perfect hexagons in
(TODO Chapter 9).

−
Example 0.3.10 (Incenters and Excenters). Let ℓ+
A , ℓA be the interior and exterior angle bisectors of ∠BAC,

i.e. the lines that satisfy

−−→ −
→ −
→− −→ −−→ −
→ −
→− −→ ◦
∡(AB, ℓ+
A ) = ∡(ℓA , AC), ∡(AB, ℓ+
A ) = ∡(ℓA , AC) + 180 (mod 360◦ ).

Then these lines satisfy

−−→ −−→
sin ∡(AB, ℓ+ −A ) sin ∡(AB, ℓ−A)
−→ = 1, = −1.
− −→
sin ∡(ℓ+A , AC) sin ∡(ℓA , AC)
− −
If we define ℓ+ +
B , ℓB , ℓC , ℓC in a similar fashion for the angles ∠CBA and ∠ACB, respectively, then we have

that −−→ −−→ + −→ +

sin ∡(AB, ℓ+ A ) sin ∡(BC, ℓB ) sin ∡(CA, ℓC )
−→ · · −→ = 1 · 1 · 1 = 1,
+ −−→ + −
sin ∡(ℓ+A , AC) sin ∡(ℓB , BA) sin ∡(ℓC , CB)

so ℓ+ + +
A , ℓB , and ℓC concur at a point I, which we call the incenter of △ABC. Similarly, we obtain that
− − − − − −
the triples of lines (ℓ+ + +
A , ℓB , ℓC ), (ℓA , ℓB , ℓC ), (ℓA , ℓB , ℓC ) concur at points IA , IB , IC respectively, which we

call the A-excenter, B-excenter, and C-excenter respectively.

If we define D, E, and F to be the feet of the altitudes from I to BC, CA, and AB respectively, then

24
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

AEIF is cyclic, and hence

∡IF E = ∡IAE = ∡F AI = ∡F EI

∡AF E = ∡AIE = 90◦ − ∡EAI = 90◦ − ∡IAF = ∡F IA = ∡F EA,

so IE = IF and AE = AF . Similarly, we obtain ID = IE = IF , BF = BD, and CD = CE.

If we define ω = (DEF ), then ID ⊥ BC, and by Proposition 0.2.6, ω is tangent to BC at D. Similarly,

ω is tangent to CA, AB at E, F respectively. We define ω to be the incircle of △ABC, and we let △DEF
be the contact triangle or intouch triangle of △ABC.

Since D, E, F lie on BC, CA, AB respectively, we must have

 


AE = AF, 

 BD + DC = a := BC,

 

BF = BD,  CE + EA = b := CA,

 


CD = CE, 
AF + F B = c := AB,

so we obtain by solving

b+c−a c+a−b a+b−c

AE = AF = , BF = BD = , CD = CE = .
2 2 2

Similarly, if DA , EA , FA are the feet of the A-excenter to BC, CA, AB respectively, then ωA = (DA EA FA )
can also be found to be tangent to all the sides of △ABC, and we call ωA the A-excircle of △ABC. We also
call △DA EA FA the A-extouch triangle of △ABC. We can find that, similarly to the incircle,

a+b+c a+b−c c+a−b

AEA = AFA = , BFA = BDA = , CDA = CEA = .
2 2 2

Note that BD = DA C and BDA = DC. Thus D and DA are reflections about the midpoint MA of BC.

In a similar fashion, we can define the B-excircle and the C-excircle as (DB EB FB ) and (DC EC FC ),
respectively. Note that DB and DC have the same reflective properties about MB and MC , respectively.

Having introduced the incircle and excircles, let’s look at a related example:

Example 0.3.11 (19 Czech-Slovak MO P4). Let △ABC be an acute-angled triangle, let P lie on BC
satisfying AB = BP , and B is between P and C. Let Q be on line BC such that AC = CQ, and C is
between Q and B. Suppose the A-excircle (J) touches AB and AC at D and E. Let DP intersect EQ at F .
Prove that AF ⊥ F J.

Proof. We need to show that F is on the circle with diameter AJ, or (ADJE). Let D′ be the point where the

25
AoPS Chapter 0. Lengths and Angles - Hidden/Completed
− −
A-excircle touches BC. Then BD′ = BD and CD′ = CE, so △ABD′ ∼ △P BD, △ACD′ ∼ △QCE. Hence

∡DF E = ∡P DB + ∡DAE + ∡CEQ

= ∡BD′ A + ∡DAE + ∡AD′ C = ∡DAE,

so F ∈ (ADE), which is what we wanted.

Theorem 0.3.12 (Stewart’s Theorem). Given three collinear points A, B, C and another point P not on
that line, we have
−−→ −→ −−→ −−→ −→ −−→
P A2 · BC + P B 2 · CA + P C 2 · AB + BC · CA · AB = 0.

Proof. WLOG assume ∡BAC = 180◦ (mod 360◦ ). By the Law of Cosines, we have

P A2 + CA2 − P C 2 P A2 + AB 2 − P B 2
+ = cos ∡P AC + cos ∡P AB = 0,
2 · P A · CA 2 · P A · AB

so we have
(P A2 − CA · AB)(AB − CA) + P B 2 · CA − P C 2 + AB = 0.

We obtain the directed lengths version by using our assumption that ∡BAC = 0◦ (mod 360◦ ), which gives us

−−→ −→ −−→ −−→ −→ −−→

P A2 · BC + P B 2 · CA + P C 2 · AB + BC · CA · AB = 180.

Example 0.3.13 (Apollonius’s Theorem). Let AM be the A-median of △ABC. Then we have

AB 2 AC 2 BC 2
AM 2 = + − .
2 2 4

Solution. This is a simple corollary of Stewart’s Theorem.

The same method can be used to calculate the length of the angle bisector - namely, if AD is the angle
bisector with D on BC, then
AD2 = AB · AC − BD · DC.

Theorem 0.3.14 (Subtended Angle Theorem). Pick four distinct points A, B, C, P . Then A, B, C are
collinear iff.
sin ∡BP C sin ∡CP A sin ∡AP B
+ + = 0.
PA PB PC

26
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Proof. By Sine Area Formula, we have

X1
[P BC] + [P CA] + [P AB] = · P B · P C · sin ∡BP C.
cyc
2

Iff A, B, C are collinear then

[ABC] = [P BC] + [P CA] + [P AB] = 0,

which is equivalent to
sin ∡BP C sin ∡CP A sin ∡AP B
+ + = 0.
PA PB PC

Practice Problems

Problem 1 (Albania MO 2012/5). Let ABC be a scalene triangle. Let P be the foot of the altitude from C
to AB. Let H be the orthocenter and O be the circumcenter of triangle ABC. Let D = OC ∩ AB. Let E be
the midpoint of CD. Prove that EP bisects OH.

Problem 2 (Incenter-Excenter Lemma). Let I, IA be the incenter and A-excenter of △ABC. Let NA =
AI ∩ (ABC)(̸= A). Prove that NA is the circumcenter of the circle (BICIA ).

Remark. This theorem is also known at “Fact 5” in America or “Chicken Feet Theorem” in Chinese speaking
places. The name “Fact 5” is from an old handout detailing certain geometry lemmas. The name “Chicken
Feet Theorem” can be found by observing that the figure consisting of the four segments NA B, NA I, NA C,
NA IA looks like a chicken foot.

Problem 3. Let ABCD be a cyclic quadrilateral. Let ID and IA be the incenters of △ABC and △BCD,
respectively, let JB be the A-excenter of △CDA, and let JC be the D-excenter of △DAB. Prove that IA ,
JB , JC , and ID are collinear.

Problem 4. Let the contact triangle of △ABC be △DEF . Prove that AD, BE, and CF concur. This
concurrency point is called the Gergonne point of △ABC.

Problem 5. Let the A-excircle, B-excircle, and C-excircle touch BC, CA, and AB at D, E, and F
respectively (this triangle is called the extouch triangle of △ABC). Prove that AD, BE, and CF concur.
The concurrency point is called the Nagel Point of △ABC.

Problem 6 (Kariya’s Theorem). Let ABC be a triangle with incenter I and intouch triangle DEF . Select
−→ −→ −→
X, Y , Z on rays ID, IE, IF such that IX = IY = IZ. Show that lines AX, BY , and CZ concur.

27
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Problem 7. Let △DEF be the intouch triangle of △ABC. Select points X, Y , Z on the incircle of △ABC.
Prove that AX, BY , and CZ concur if and only if DX, EY , F Z concur or DX ∩ EF , EY ∩ F D, F Z ∩ DE
are collinear.

Remark. This is a special case of (TODO 1.1.1).

Problem 8. Let (I) be the incircle of triangle ABC. Let △DEF be the medial triangle of △ABC. Draw
the tangents to (I) through the points D, E, and F that are not the sides of the triangle ABC. These three
tangents respectively intersect EF , F D, and DE at X, Y , Z respectively. Prove that X, Y , and Z are
collinear.

Problem 9. In triangle △ABC, let AD be the internal angle bisector of ∠BAC, such that ∠ADC = 60◦ .
−−→
Construct point E on AD such that DE = DB. Let the ray CE intersect AB at F . Prove that

AF × AB + CD × CB = AC 2 .

Problem 10 (Taiwan MO 2021/4). Let I be the incenter of triangle ABC and let D the foot of altitude from
I to BC. Suppose the reflection D′ of D with respect to I satisfies AD′ = ID′ . Let Γ be the circle centered
at D′ that passes through A and I, and let X, Y ̸= A be the intersection of Γ and AB, AC, respectively.
Suppose Z is a point on Γ so that AZ is perpendicular to BC.

Prove that lines AD, D′ Z, XY are concurrent.

0.4 Power of a Point

Definition 0.4.1. Given a fixed circle Γ and a point P , we can define a function denoted as the power of
point P defined as such:
PowΓ (P ) := OP 2 − R2

where O and R represent the center and radius of Γ.

Proposition 0.4.2. Construct a line going through P that intersects Γ at points A, B. Then

−→ −−→
P A · P B = PowΓ (P ).

Proof. Let O be the center of Γ, and let M be the midpoint of AB. Then we know that OM ⊥ AB. So by
(0.2.11), we get

−→ −−→ −−→ −−→ −−→ −−→

P A · P B = (P M + M A) · (P M + M B) = P M 2 − M A2

= P O2 − OA2 = PowΓ (P )

28
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Corollary 0.4.3. Draw two lines ℓ1 , ℓ2 through a point P , let us pick two points Ai , Bi on ℓi such that

−−→ −−→ −−→ −−→

P A1 · P B1 = P A2 · P B2

Then A1 , B1 , A2 , B2 are concyclic. (The converse of this also holds).

Proof. Let Γ = (A1 B1 A2 ) and let B2′ = Γ ∩ ℓ2 . Now

−−→ −−→ −−→ −−→ −−→ −−→

P A2 · P B2 = P A1 · P B1 = PowΓ (P ) = P A2 · P B2′

so B2 = B2′ as desired.

Example 0.4.4 (Azerbaijan TST 2017/2/1). Let ABC be an acute angled triangle. Points E and F are
chosen on the sides AC and AB, respectively, such that

BC 2 = BA · BF + CE · CA.

Prove that for all such E and F , the circumcircle of the triangle AEF passes through a fixed point different
from A.

Solution. Let M be the midpoint of BC - we compute Pow(AEF ) (M ). If it is fixed, then we are done because
then AM ∩ (AEF ) is the desired fixed point. We compute

1
1
Pow(AEF ) (M ) = OM 2 − r2 = OB 2 − r2 + OC 2 − r2 − BC 2
2 4
1
1 2
= Pow(AEF ) (B) + PowAEF (C) − BC
2 4
1 1 2
= (BA · BF + CE · CA) − BC
2 4
1 2
= BC
4

and we are done.

Proposition 0.4.5. A line through point P is tangent to a circle Γ at T . Then

PowΓ (P ) = P T 2 .

This can be thought of as the limiting case of Proposition 0.4.2.

29
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Proof. By Proposition 0.2.6, OT ⊥ P T . Hence by the Pythagorean Theorem,

P T 2 = OT 2 − OP 2 = PowΓ (P )

Corollary 0.4.6. Two lines ℓ1 and ℓ2 are drawn through a point P . Let A, B ∈ ℓ1 and T ∈ ℓ2 Then (ABT )
is tangent to ℓ2 at T if
P A · P B = P T 2.

Example 0.4.7 (CWMO 2019/5). Let O, H be the circumcenter and orthocenter of acute △ABC respectively.
The line passing through H and parallel to AB intersects line AC at M , and the line passing through H
and parallel to AC intersects line AB at N . L is the reflection of the point H in M N . Line OL and AH
intersect at K. Prove that K, M , L, N are concyclic.

Solution. We first begin by trying to prove that OM = ON , or that Pow(ABC) (M ) = Pow(ABC) (N ). Note
that
∡M CH = 90 − ∡BAC = ∡HBN, ∡HM C = ∡BAC = ∡BN H.

−
so it follows that △CM H ∼ △BN H Now

Pow(ABC) (M ) = M A · M C = HN · M C

= HM · N B = N A · N B = Pow(ABC) (N )

as desired. Now ALM N is an isosceles trapezoid and hence cyclic. Hence △OAM ∼
= △OLN . We conclude
that (AO, AH) is antiparallel with respect to (AM, AN ), so

∡N LK = ∡OAM = ∡N LK

and AKLN is cyclic - combined with ALM N cyclic gives the desired.

Definition 0.4.8. For any two circles Γ1 , Γ2 , we define the radical axis of the two circles as

{P | PowΓ1 (P ) = PowΓ2 (P ).}

Proposition 0.4.9. The radical axis of Γ1 and Γ2 is a straight line perpendicular to the line through the
centers of Γ1 and Γ2 .

Proof. Let O1 and r1 be the center and radius of Γ1 , respectively, and analogously define O2 and r2 . Note
that
PowΓ1 (P ) = PowΓ2 (P ) ⇐⇒ O1 P 2 − r12 = O2 P 2 − r22 =⇒ O1 P 2 − O2 P 2 = r12 − r22

30
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

so by the Perpendicularity Criterion the locus of P is a straight line perpendicular to O1 O2 .

From this proof, we see that any line perpendicular to O1 O2 is the set of points defined by

{P | PowΓ1 (P ) − PowΓ2 (P ) = k}

for some k ∈ R.

Proposition 0.4.10. The radical axis of two circles Γ1 and Γ2 intersecting at A and B is the line AB.

Proof. Note that

PowΓ1 (A) = PowΓ2 (A) = PowΓ1 (B) = PowΓ2 (B) = 0.

so both A, B lie on the radical axis.

Theorem 0.4.11 (Radical Axis Theorem). For any three circles ω1 , ω2 , ω3 , let ℓ1 , ℓ2 , and ℓ3 be the three
radical axis of the pairs of circles (ω2 , ω3 ), (ω3 , ω1 ), and (ω1 , ω2 ) respectively. Then ℓ1 , ℓ2 , and ℓ3 concur, are
all parallel, or are all the same line.

Proof. If ℓ1 ∩ ℓ2 = P ̸∈ L∞ , then

Powω1 (P ) = Powω2 (P ) = Powω3 (P )

by the definition of radical axis. The edge cases (all parallel or all the same line) can easily be checked.

Definition 0.4.12. For any three circles ω1 , ω2 , ω3 , let ℓ1 , ℓ2 , and ℓ3 be the three radical axis of the pairs of
circles (ω2 , ω3 ), (ω3 , ω1 ), and (ω1 , ω2 ) respectively. Then:

• If ℓ1 = ℓ2 = ℓ3 , then ω1 , ω2 , and ω3 are said to be coaxial.

• Otherwise, ℓ1 , ℓ2 , ℓ3 concur at the radical center of ω1 , ω2 , and ω3 , or they are all parallel.

Example 0.4.13. Let △DEF be the orthic triangle of △ABC. Then BCEF , CAF D, and ABDE are all
cyclic, and their radical axis is the intersection of AD, BE, and CF , which is the orthocenter H.

Example 0.4.14 (ISL 2009 G3). Let ABC be a triangle. The incircle of ABC touches the sides AB and
AC at the points Z and Y , respectively. Let G be the point where the lines BY and CZ meet, and let R
and S be points such that the two quadrilaterals BCY R and BCSZ are parallelograms.

Prove that GR = GS.

31
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Solution. Let ωa be the A-excircle, and let it be tangent to BC, CA at X ′ and Y ′ respectively. Then

Powωa (B) = BX ′2 = CY 2 = BR2 = PowR (B)

Powωa (Y ) = Y Y ′2 = BC 2 = RY 2 = PowR (Y )

where PowR (P ) denotes the power of a point with respect to a circle with radius 0 at R (we call this the
point circle at R). Then BY is the radical axis of the point circle at R and ωa . Similarly, CZ is the radical
axis of the point circle at S and ωa . Hence G is the radical center of the point circles at R and S and ωa . It
follows that GR = GS as desired.

Corollary 0.4.15. Let △ABC be a triangle. Suppose D1 , D2 ∈ BC, E1 , E2 ∈ CA, and F1 , F2 ∈ AB such
that D1 D2 E1 E2 , E1 E2 F1 F2 , and F1 F2 D1 D2 are all cyclic. Then the hexagon D1 D2 E1 E2 F1 F2 is cyclic.

Proof. Assume otherwise. Let Γ1 = (D1 D2 E1 E2 ) and analogously define Γ2 and Γ3 . Then the radical axis of
Γ1 and Γ2 is CA, and similarly the radical axis of Γ2 and Γ3 is AB and the radical axis of Γ3 and Γ1 is BC.
If these circles do not coincide, then by Radical Axis Theorem their radical axis should either be all-parallel,
coincide, or concur, but none of these three things are true, contradiction.

Proposition 0.4.16. If (O1 ), (O2 ), and (O3 ) are three coaxial circles, then O1 , O2 , O3 are collinear. This
line is perpendicular to the common radical axis of the three circles.

Proof. O1 O2 , O2 O3 , and O3 O1 are all perpendicular to the common radical axis ℓ, which is enough.

Example 0.4.17. Let AB be a chord of (O) and let M be the midpoint of the minor arc AB. Let C be a
point outside (O), and suppose that the two tangents from C touch O at S and T , respectively. Let M S and
M T intersect AB at E, F respectively. The line perpendicular to AB through E and F intersect OS, OT at
X, Y respectively. A line through C intersects (O) at P and Q. Let R be the intersection of M P and AB.
Finally, let Z be the center of (P QR). Prove that X, Y , and Z are collinear.

32
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

O
Y
T

F
A B
E R

Solution. Since EX ⊥ AB ⊥ M O and F Y ⊥ AB ⊥ M O,

+ +
△ESX ∼ △M SO, △F T Y ∼ M T O.

Hence XE = XS and Y F = Y T . Let ωX and ωY be the circles centered at X and Y passing through E and
F respectively. Then ωX and ωY are tangent to (O) at S and T and tangent to AB at E and F , respectively.
Since ∡M SA = ∡M BA = ∡EAM , M A is tangent to (AES). Hence

M A2 = M E · M S

and similarly M A2 = M F · M T = M P · M R so

PowωX (M ) = PowωY (M ) = Pow(P QR) (M ).

Also,
PowωX (C) = CS 2 , PowωY (C) = CT 2 , Pow(P QR) (C) = CP · CQ.

Hence CM is the (common) radical axis of ωX , ωY , and (P QR). By Proposition 0.4.16, we are done.

33
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

For problems requiring proofs of concyclicity, we often use the following theorem:

Theorem 0.4.18 (Ptolemy’s Inequality). For any four points A, B, C, D,

BC · AD + CA · BD + AB · CD ≥ 2 max{BC · AD, CA · BD, AB · CD}

with equality if and only if A, B, C, D are collinear or concyclic. More commonly, Ptolemy’s inequality is
written for convex quadrilaterals ABCD, in which

AB · CD + AD · BC ≥ AC · BD.

There are many great proofs of this theorem, and we will present one that is completely based on the
properties outlined in this chapter. Before we prove this theorem, we need the following lemma, which is the
basis of (TODO section 1.2):

+ +
Lemma 0.4.19 (Spiral Similarity). If △ABC ∼ △AB ′ C ′ , then △ABB ′ ∼ △ACC ′ .

+
Proof. △ABC ∼ △AB ′ C ′ means

AB AB ′
∡BAC = ∡B ′ AC ′ (mod 360◦ ), =
AC AC ′

which is equivalent to
AB AC
∡BAB ′ = ∡CAC ′ (mod 360◦ ), =
AB ′ AC ′
+
which implies △ABB ′ ∼ △ACC ′ .

Proof of 0.4.18. WLOG assume that BC · AD is maximal among the four points and that A, B, C, D don’t
+ +
lie on a line. Let E be a point such that △DEB ∼ △DCA, so by Lemma 0.4.19, △DCE ∼ △DAB. Now

CA · BD = BE · AD, AB · CD = EC · AD.

Thus
CA · BD + AB · CD + BC · AD − 2BC · AD = (BE + EC − BC)AD ≥ 0

by the Triangle Inequality. Equality holds when E lies on BC, which can be checked to be equivalent to A,
B, C, D are concyclic.

Theorem 0.4.20 (Casey’s Theorem). Given four non-intersecting circles ΓA , ΓB , ΓC , ΓD , define d+

IJ to be

the length of the external common tangents of ΓI and ΓJ , and define d−

IJ to be the length of the internal

34
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

common tangents of ΓI and ΓJ . There exists a circle Ω tangent to the four circles ΓA , ΓB , ΓC , ΓD iff.

dBC dAD ± dCA dBD ± dAB dCD = 0,

where


d+ , if ΓI , ΓJ are both tangent on the same side of Ω,

IJ
dIJ =
d− , if ΓI , ΓJ are tangent on opposite sides of Ω.

IJ

(This weird definition lets us avoid writing sixteen forms of this theorem. Here, ± refers to any of the 4
possible equalities holding).

Proof. We first prove the “if” side of the “iff”. (The proof of the other side will be left until (TODO 3.3)
when we define inversion.)

Suppose Ω is tangent to ΓA , ΓB , ΓC , ΓD , with O, R respectively being the center of Ω and its radius.
Define OI , rI similarily for ΓI .

We first prove that

IJ p
dIJ = (R ± rI )(R ± rJ ),
R

where the R ± rn is positive if Γn is externally tangent, and negative if it’s internally tangent. By the Law of
Cosines we get
(R ± rI )2 + (R ± rJ )2 − OI OJ2 R2 + R2 − IJ 2
= cos(∡IOJ) = ,
2(R ± rI )(R ± rJ ) 2R2

and thus

d2IJ = OI OJ2 − (rI ± rJ )2

IJ 2
 
= (R ± rI )2 + (R ± rJ )2 − 2(R ± rI )(R ± rJ ) 1 − − (rI ± rJ )2
2R2
IJ 2
= 2 (R ± rI )(R ± rJ ).
R

Taking the square root of both sides, we get our desired expression, and we proved this small lemma.

Back to the proof of Casey’s; we proceed by Ptolemy’s theorem to get that

p
(R ± rA )(R ± rB )(R ± rC )(R ± rD )
dBC dAD ± dCA dBD ± dAB dCD = · (BC · AD ± CA · BD ± AB · CD)
R2

and then we one-shot by our previous lemma, to get this whole nasty expression is just 0.

35
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Corollary 0.4.21 (Three Chords Theorem). For any four concyclic points A, B, C, D, we have

DA · sin ∡BDC + DB · sin ∡CDA + DC · sin ∡ADB = 0.

+
Proof. Similarly to the proof of Ptolemy’s Inequality, take a point E such that △DEB ∼ △DCA - thus B,
E, C are collinear. Now

DA · sin ∡BDC + DB · sin ∡CDA + DC · sin ∡ADB

DB · DC
= sin ∡BDC + DB · sin ∡EDB + DC · sin ∡CDE
DE
2
= ([DBC] + [DEB] + [DCE]) = 0
DE

since B, E, and C are collinear.

Remark. If you know what inversion is, this is also the inverted version of Theorem 0.3.14 from earlier.

Example 0.4.22 (ISL 2015 G4). Let ABC be an acute triangle and let M be the midpoint of BC. A circle
ω passing through A and M meets the sides AB and AC at points P and Q respectively. Let T be the point
such that AP T Q is a parallelogram. Suppose that T lies on the circumcircle of ABC. Determine all possible
AT
values of AM .

Solution. By the Three Chords Theorem, AP M Q cyclic means

AP · sin ∡BAM + AQ · sin ∡M AC = AM · sin ∡BAC

and ABT C cyclic means

AB · sin ∡T AP + AC · sin ∡QAT = AT · sin ∡QAP.

By Ratio Lemma, we have

sin ∡BAM : sin ∡M AC : ∡BAC :: AC : AB : 2AM

sin ∡T AP : sin ∡QAT : ∡QAP :: AQ : AP : AT.

36
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

Hence the two length equations can be written as

AP · AC + AQ · AB = AM · 2AM

AB · AQ + AC · AP = AT · AT.

AT
√
This gives AT 2 = 2AM 2 , or equivalently AM = 2.

Example 0.4.23 (APMO 2017/2). Let ABC be a triangle with AB < AC. Let D be the intersection point
of the internal bisector of angle BAC and the circumcircle of ABC. Let Z be the intersection point of the
perpendicular bisector of AC with the external bisector of angle ∠BAC. Prove that the midpoint of the
segment AB lies on the circumcircle of triangle ADZ.

Solution. By the Three Chords Theorem, it suffices to show that

AM · sin ∠DAZ + AZ · sin ∠M AD = AD · sin M AZ.

Define a = BC and analogously define b and c, and define α = ∠BAC and analogously define β and γ. Then
Then the left-hand side of the above equation is

c b 1 b+c
+ · sin α =
2 2 sin 12 α 2 2

and the right-hand side (from the Law of Sines) is

    
b 1 ◦ 1
· sin β + α · sin 90 + α
sin β 2 2
b 1 b+c
= · (cos(90◦ − β) − cos(90◦ + α + β)) =
sin β 2 2

where the last equality is because

b c
· cos(90◦ + α + β) = · (− sin γ) = −c.
sin β sin γ

Practice Problems

Problem 1 (Euler’s Theorem). Let O and I be the circumcenter and incenter of △ABC with circumradius
R and inradius r.

37
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

(i) Let F be the foot of I to AB, Na be the intersection of AI and (ABC), and let N ∗a be the antipode
+
of Na on (ABC). Prove that △AF I ∼ △Na∗ BNa .

(ii) By the Incenter-Excenter Lemma, show that OI 2 = R2 − 2Rr.

Problem 2 (Taiwan IMOC 2021). Let the A-midline (line connecting midpoints of sides AB and AC) of
△ABC intersect (ABC) at two points P , Q. Let the tangent to (ABC) at A intersect BC at T . Prove that
∡BT Q = ∡P T A.

Problem 3. Let triangle ABC have centroid G. Let the medians AG, BG, CG intersect (ABC) at points
A′ , B ′ , C ′ . Prove that
AG BG CG
+ + = 3.
GA′ GB ′ GC ′

Problem 4 (Taiwan IMOC 2021). Let BE and CF be the altitudes of △ABC, and let D be the antipode
of A on (ABC). The lines DE and DF intersect(ABC) again at Y and Z, respectively. Prove that Y Z, EF ,
and BC concur.

Problem 5 (Iran TST 2011/1). In acute triangle ABC, ∠B > ∠C. Let M be the midpoint of BC. Let D
and E be the feet of the altitude from C and B, respectively. Let K and L be the midpoints of M E and
M D respectively. If KL intersects the line through A parallel to BC in T , prove that T A = T M .

Problem 6. Let circles O1 and O2 intersect at the two points X and Y . Draw a line ℓ1 through the center
of O1 that intersects O2 at two points P, Q, Draw another line ℓ2 through the center of O2 that intersects O1
at two points R, S. Prove that if P, Q, R, S are concyclic, then their circumcenter lies on XY .

Problem 7. Let ABC be a triangle with incenter I and circumcenter O for which BC < AB < AC.
Let Y and X be points in the interiors of sides AB and AC, respectively, of a triangle ABC, such that
CX = BC = Y B. Prove that XY ⊥ IO.

Problem 8 (2019 Estonia TST/2). In an acute-angled triangle ABC, the altitudes intersect at point H,
and point K is the foot of the altitude drawn from the vertex A. Circle c passing through points A and K
intersects sides AB and AC at points M and N , respectively. The line passing through point A and parallel
to line BC intersects the circumcircles of triangles AHM and AHN for the second time, respectively, at
points X and Y . Prove that |XY | = |BC|.

Problem 9 (2017 ISL G4). In triangle ABC, let ω be the excircle opposite to A. Let D, E and F be the
points where ω is tangent to BC, CA, and AB, respectively. The circle AEF intersects line BC at P and Q.
Let M be the midpoint of AD. Prove that the circle M P Q is tangent to ω.

Problem 10 (Canada 2016/5). Let △ABC be an acute-angled triangle with altitudes AD and BE meeting
at H. Let M be the midpoint of segment AB, and suppose that the circumcircles of △DEM and △ABH

38
AoPS Chapter 0. Lengths and Angles - Hidden/Completed

meet at points P and Q with P on the same side of CH as A. Prove that the lines ED, P H, and M Q all
pass through a single point on the circumcircle of △ABC.

Problem 11 (Taiwan 2015/2J/I1-2). Let the incircle of △ABC be ω, and suppose ω meets BC at D. Let
AD ∩ ω = L(̸= D), and let the A-excenter of ABC be IA . Let M be the midpoint of BC and let N be the
midpoint of IA M . Prove that BCN L is cyclic.

39
Chapter 1

Geometric Transformations -
Hidden/Completed

1.1 Homothety

Theorem 1.1.1 (Desargues’s). Given two triangles △A1 B1 C1 , △A2 B2 C2 , we have that B1 C1 ∩B2 C2 , C1 A1 ∩
C2 A2 , A1 B1 ∩ A2 B2 are collinear iff. A1 A2 , B1 B2 , C1 C2 are concurrent.

Proof. Let X = B1 C1 ∩ B2 C2 , Y = C1 A1 ∩ C2 A2 , and Z = A1 B1 ∩ A2 B2 .

First suppose that A1 A2 , B1 B2 , C1 C2 concur at some point P .

Then by Menelaus’s theorem, we have that

B1 X C1 C2 P B2
· · = −1,
XC1 C2 P B2 B1
P C2 C1 Y A1 A2
· · = −1,
C2 C1 Y A1 A2 P
B1 B2 P A2 A1 Z
· · = −1
B2 P A2 A1 ZB1

Multiplying all of these three together gives that

B 1 X C 1 Y A1 Z
· · = −1
XC1 Y A1 ZB1

so X, Y, Z lie on a line.

40
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

C1

Z
A1

C2 B1
A2

B2
Y

Figure 1.1: Note that this is a symmetric relation between triangles.

Now suppose that X, Y, Z lie on a line. Consider the triangles △B1 B2 Z, △C1 C2 Y . We have that
B1 C1 , B2 C3 , Y Z concur at X. As such, A2 = B2 Z ∩ C2 Y, A1 = ZB1 ∩ Y C1 , B1 B2 ∩ C1 C2 lie on a line which
finishes.

If this is the case, we say that △1 = △A1 B1 C1 and △2 = △A2 B2 C2 are perspective. The line through
B1 C1 ∩ B2 C2 , C1 A1 ∩ C2 A2 , A1 B1 ∩ A2 B2 is their perspectrix and the point that A1 A2 , B1 B2 , C1 C2 concur
at is the perspector.

A specific case of this is when B1 C1 ∥ B2 C2 , C1 A1 ∥ C2 A2 , A1 B1 ∥ A2 B2 , then B1 C1 ∩ B2 C2 , C1 A1 ∩

c2 A2 , A1 B1 ∩ A2 B2 all lie on the line at infinity L∞ . We can mimic the proof of Desargues (or take a
projective transformation, defined in Chapter 2); this means that the two triangles are perspective, and we
call them homothetic. The perspective center is called the homothetic center. If O is the homothetic center,
then by the parallel sides we have that

OA2 OB2 OC2

= = = k. (♠)
OA1 OB1 OC1

We call k the homothetic ratio of the homothety. If we are given △1 , O, k for finite O, then we can use (♠)
to construct △2 . Removing △1 , we get the following definition.

Definition 1.1.2. Given a point O ̸∈ L∞ and k ̸= 0, we define the homothety hO,k as a mapping on points

41
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

P 7→ Q for P ̸∈ L∞ such that Q ∈ OP and

OQ
= k.
OP

The line at infinity is fixed under this operation.

(This is not standard notation in places like America).

We also define for a vector ⃗v the translation P 7→ P + ⃗v with the mapping h⃗v . We consider this a
homothety with ratio 1 taken at the point ∞⃗v at infinity.

Note that a homothetic transformation h maps lines to lines. Furthermore, we have that

−−−−−−→ −−→
h(A)h(B) = k · AB

so homotheties also preserve circles (consider the center and radius).

If k = −1, then hO,−1 (P ) is the reflection of P about O. In that case we use the notation sO = hO,−1 .

Example 1.1.3 (Euler line). Let △Ma Mb Mc be the medial triangle of △ABC. We know that

Ma Mb ∥ AB, Mb Mc ∥ BC, Mc Ma ∥ CA.

Hence, △ABC is homothetic to △Ma Mb Mc . The center of homothety of the two triangles is G = AMa ∩
BMB ∩ CMc , and the ratio of the homothety is

GMa 1
=−
GA 2

As such, we know that the circumcenter O of △ABC is the orthocenter of Ma Mb Mc (since Ma O ⊥ BC ∥

Mb Mc ). Thus, if H is the orthocenter of △ABC, then O, G, H lie on a line and

GO 1
=− .
GH 2

This line is the Euler line of △ABC.

△
! : If △ABC is an equilateral triangle, then O, G, H (and more generally most triangle centers) are the
same, so no Euler line exists.

Example 1.1.4 (2021 2J I1-G). Let ABCD be a convex quadrilateral with all distinct side lengths and
AC ⊥ BD. Let O1 , O2 be the circumcenters of △ABD, △CBD, respectively. Prove that AO2 , CO1 , and the
Euler lines of △ABC and △ADC concur.

Solution. By symmetry, we only need to prove that AO2 , CO1 , and the Euler line of △ABC are concurrent.
Let O, H be the circumcenter and orthocenter of △ABC. Then we only need to show △CAH and △O1 O2 O

42
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

are homothetic, so we can just prove that AH ∩ O2 O, HC ∩ OO2 , CA ∩ O1 O2 are collinear. Now note that
since AH ⊥ BC ⊥ O2 O, HC ⊥ AB ⊥ OO1 , CA ⊥ BD ⊥ O1 O2 , we are done.

D
H B
O2
O

O1

Example 1.1.5 (Nine-Point Circle). Let H be the orthocenter of △ABC, △Ma Mb Mc , △Ha Hb Hc be the
medial and orthic triangles of △ABC, respectively. Let Ea , Eb , Ec be the midpoints of AH, BH, CH,
respectively. Then Ma , Mb , Mc , Ha , Hb , Hc , Ea , Eb , Ec all lie on a common circle ϵ, called the nine-point
circle of △ABC.

Solution. We show that the nine-point circle is the image of (ABC) under hH, 12 .

Then we get that hH, 12 (A) = Ea and so forth. Now, by (TODO HaReflection) we get that hH,2,( Ha ) = HA
and so forth. It remains to show that hH,2,( Ma ) ∈ (ABC). Let A∗ be the A antipode on (ABC).

Then we have that BH ⊥ CA ⊥ A∗ C, HC ⊥ BA∗ so BHCA∗ is a parallelogram. This implies the result.
Thus, the nine-point center N9 is the image of hH, 21 (O), where O is the circumcenter. As such, N9 is the
midpoint of OH, and is thus on the Euler line. In summary,

(N ) = (Ma Mb Mc ) = hG,− 21 (ABC).

Given △ABC, we define the complement of a point P as P ∁ = hG,− 12 . If the reference triangle needs
specification, P ∁,△ABC will be used. Then the complement of (ABC) is ε and A∁ = Ma , H ∁ = O.
∁
Likewise, we define the anti-complement of P as P is the inverse of that operation. We define the
∁
anticomplementary triangle as the image (△ABC) .

43
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Ea Hb
Mb
Mc
O
Hc N9
H G

Eb Ec
B Ha Ma C

A∗

Figure 1.2: The full nine-point circle configuration. We will revisit this in Chapters 5 and 6.

Example 1.1.6 (Hong Kong TST 2018/6). Let the orthocenter and circumcenter of △ABC be H and O,
respectively, and let the circumcenters of △HBC, △HCA, △HAB be Oa , Ob , Oc , respectively. Prove that
AOa , BOb , COc , and OH concur.

Solution. Let △Ha Hb Hc be the orthic triangle with respect to △ABC.

Notably, △HBC, △HCA, △HAB all share the same orthic triangle. As such, △ABC, △HBC, △HCA, △HAB
share a nine point circle and thus the same nine point center N9 .

Since AOa , BOb , COc all have N9 as a midpoint, N9 lies on all such lines.

Proposition 1.1.7. The composition of two homotheties is itself a homothety whose ratio is the product of
the ratios of the earlier two homotheties.

Proof. The original proof does a load of casework involving vectors so we present an alternative proof.
Consider points in R2 .

Note that a homothety is effectively a mapping from P to k(P − O) + O, which simplifies as P 7→ kP + C

for some constant C.

Then the image of two mappings with ratios K1 , k2 is of the form P 7→ k1 k2 P + C.

Proposition 1.1.8. For any two circles, there exist two homotheties h+ , h− of positive and negative ratio with
the same magnitude respectively, that map Γ1 to Γ2 . We call h+ the exsimilicenter and h− the insimilicenter.

44
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Proof. If the two circles are concentric, then taking the two homotheties with negative signs at their center
suffices. If the two circles have the same radius, taking a translation and reflection suffice.
O+ O2
̸ O2 and radii r1 ̸= r2 , then define O+ , O− such that
Else, let the circles have centers O1 = O+ O1 =
r2 O1 O2
r1 , O1 O1 = − rr21 . Then define h+ = hh,O+ ,r2 /r1 and h− = hh,O− ,−r2 /r1 .

Example 1.1.9. The homotheties that map the nine-point circle ε to the circumcircle are hH,2 and hG,−2
respectively.

Proposition 1.1.10. The intersection of the external tangents (if they exist) of two circles Γ1 and Γ2 is
the exsimilicenter of Γ1 and Γ2 . Similarly, the intersection of the internal tangents of Γ1 and Γ2 is the
insimilicenter of the two circles.

Proof. We will only prove one statement - the other is analogous. Suppose the two external tangents to the
two circles are K+ and L+ that meet at O+ . Let P+,1 and P+,2 be the intersections of K+ with Γ1 and Γ2 ,
respectively, and define Q+,1 and Q+,2 for L+ similarly. Finally, let O1 and O2 be the centers of Γ1 and Γ2 ,
respectively. Note that O1 O2 = ℓ is the internal angle bisector of ∡(K+ , L+ ), so P+,1 , Q+,1 ∈ (O+ O1 ) and
P+,2 , Q+,2 ∈ (O+ O2 ). Hence we have

∡O+ P1,+ Q1,+ = ∡O+ O1 Q1,+ = 90◦ = ∡(L+ , ℓ) = O+ O2 Q2,+ , = ∡O+ P2,+ Q2,+

and similarly ∡P1,+ Q1,+ O+ = ∡P2,+ Q2,+ O+ so △O+ P1,+ Q1,+ and △O+ P1,+ Q1,+ are homothetic about O+
by a homothety h = hO+ ,k . Then it is easily checked that h(O1 ) = O2 , and since

OP2,+ = k · OP1,+ ,

h(Γ1 ) = Γ2 , as desired.

This allows us to prove the following common lemma:

Example 1.1.11. Let I be the incenter of △ABC, D be the A-excircle touchpoint on BC, and M be the
midpoint of BC. Prove that AD ∥ IM .

Solution. Note that CA, AB are the common tangents between ω and the A-excircle. Thus we get that A
is the exsimilicenter of ω and ωa . Let us look at the homothety h from A which sends ωa to ω. Then we
know that h(BC) is tangent to the incircle (and obviously it’s parallel to BC. Thus h(BC) is tangent to ω
and parallel to BC, so it must be parallel to the antipode of D, which we will call D∗ . Further, we get that
D∗ = h(D), so A, D∗ , D lie on a line. Since I and M are respectively the midpoints of D∗ D and DD′ , we
have IM ∥ D∗ D′ = AD′ .

45
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

D∗

D′ M D C
B

Figure 1.3: Also, AD′ ∥ Ia M .

Ia
Theorem 1.1.12 (Monge). For any three circles Γ1 , Γ2 , Γ3 , let Oij,+ , Oij,− be the respective exsimilicenter
and insimilicenters of Γi , Γj . Then when there are an odd number of plus signs, we get that O12,± , O23,± , O31,±
are collinear.

Proof. We assume different radii for simplicity. Let Oi , ri be the center and radii of Γi . Then we have by
Proposition 1.1.8 that
Oi Oij,± ri
=∓ .
Oij,± Oj rj

When there are an odd number of plus signs, we get that

Y O1 O12,± Y r1
= ∓ = −1.
cyc
O12,±O2 cyc
r2

so by Menelaus’s theorem the result follows.

Remark. Similarly, by Ceva’s theorem we have that when the number of plus signs is even, the lines
O1 O23,± , O2 O31,± , O3 O12,± are concurrent. (This is another example of the duality we referred to earlier.)

However this is usually useless.

In reality, Monge is better stated as a special case of the following theorem:

Theorem 1.1.13. Let △1 = △A1 B1 C1 , △2 = △A2 B2 C2 , △3 = △A3 B3 C3 , be three triangles such that they
are pairwise perspective. Let Pij , ℓij be the perspector and perspectrix of △i and △j respectively. Then

46
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

O12,+
O1
O2

O3

O23,+

O13,+

Figure 1.4: An alternate proof of this theorem is that ”compositions of homotheties are still homotheties”.

(i) If ℓ23 = ℓ31 = ℓ12 , we have P23 , P31 , P12 collinear.

(ii) If P23 = P31 = P23 , we have ℓ23 , ℓ31 , ℓ12 are concurrent.

Proof. For (i), we first consider the two triangles △b = △B1 B2 B3 , △c = △C1 C2 C3 . Suppose ℓ23 = ℓ31 = ℓ12 .
Then we have that B1 C1 , B2 C2 , B3 C3 are concurrent, and thus these two triangles are perspective, so by
Desargues P23 = B2 B3 ∩ C2 C3 , P31 = B3 B1 ∩ C3 C1 , P12 = B1 B2 ∩ C1 C2 are collinear.

For (ii), consider the two triangles △b = △(C1 A1 )(C2 A2 )(C3 A3 ) and △c = △(A1 B1 )(A2 B2 )(A3 B3 ).
Suppose P23 = P31 = P12 . Then we have that A1 A2 A3 is the perspectrix △b and △c , Let bi = Ci Ai and
ci = Ai Bi . Then l23 = (b2 ∩ b3 )(c2 ∩ c3 ), l31 , l12 concur by Desargues.

(Note the similarity of the proof of (ii) to the proof of (i).)

Remark. Monge is a special case of this when you take three equilateral triangles oriented the same way in
each circle.

Practice Problems

Problem 1. Let ABCD be a quadrilateral, and M, N be the midpoints of sides AB, BC, respectively. Let
P, Q be two points on CD, DA, respectively, such that P Q∥M N . Prove that P N , QM , and BD either concur
or are parallel.

47
AoPS Chapter 1. Geometric Transformations - Hidden/Completed
1
Problem 2. For any point P in △ABC, let △DEF be the cevian triangle of P wrt △ABC.

Let X = EF ∩ BC, Y = F D ∩ CA, Z = DE ∩ AB. Show that X, Y, Z are collinear. This line is called
the trilinear polar of P , and is notated as t(P ).

Problem 3. Take △ABC. Let D, E, F be chosen on AB, BC, CA such that △ABC, △DEF are perspective
with center P . Let X, Y, Z be chosen on DE, EF, F D such that △ABC, △XY Z are perspective with center
Q. Show that △XY Z, △DEF are perspective with center R. We call R the crosspoint of P and Q, which is
denote with a P ⋔ Q.

In fact, the crosspoint is symmetric: P ⋔ Q = Q ⋔ P .

Problem 4. Let △Ma Mb Mc be the medial triangle of △ABC. Construct a line ℓ which intersects
BC, CA, AB at D, E, F respectively. Let D′ be the reflection of D across Ma , and define E ′ and F ′
respectively. Prove that

(i) D∗ , E ∗ , F ∗ lie on a line ℓ∗ .

(ii) The midpoints of AD, BE, CF lie on τℓ = (ℓ∗ )∁ .

We call τℓ as the Newton line of △ABC and ℓ. (More information about this line in (TODO 4.1)).

Problem 5. Let H be the orthocenter of △ABC, and P an arbitrary point on the circumcircle of △ABC.
2
Let E be the foot from B to CA. Pick points Q, R such that P AQB and P ARC are both parallelograms.
Let AQ and HR intersect at X. Prove that EX ∥ AP .

Problem 6 (Brazil 2012/2). △ABC is a non-isosceles triangle. Let △TA TB TC be the intouch triangle.
Let IA , IB , IC be the excenters. Let XA be the midpoint of IB IC and define XB , XC similarly. Show that
XA TA ,XB TB ,XC TC , OI concur at a point, where O is the circumcenter and I is the incenter.

(This problem is in the book with XA TA replaced with IA TA , but it remains true).

Problem 7. Prove line IO is the Euler line of the intouch triangle.

Problem 8 (Taiwan TST 2015/1J/I2-2). Given a triangle △ABC, let H be the orthocenter and G the
centroid. Show that (HG), the circumcircle, and the nine-point circle are coaxial.

Problem 9. Let I be the incenter of △ABC. Let D be an arbitrary point on BC, and let ωB and ωC
respectively be the incircles of △ABD and △ACD. Let ωB and ωC touch BC at E and F . Let P be the
intersection of AD and the line connecting the centers of ωB and ωC . Let X be the intersection of BI and
CP , and let Y be the intersection of CI and BP . Prove that EX and F Y intersect on the incircle of △ABC.
1Where D = AP ∩ BC, E = BP ∩ AC, F = CP ∩ AB.
2Q is sometimes called the “parallelogram point” of △P AB with respect to P , and similarily for R.

48
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

1.2 Spiral Similarity

Given a fixed point O and a fixed angle θ, we can define a rotation as the mapping rO,θ which maps a point
P to the point Q such that
OP = OQ, ∡P OQ = θ (mod 360◦ )

We then have that

rO,θ1 ◦ rO,θ2 = rO,θ2 ◦ rO,θ1 = rO,θ1 +θ2

In fact, the composition of two rotations is itself a rotation. We also note that a rotation preserves angles
and lengths, ie
r(A)r(B) = AB, ∡r(A)r(B)r(C) = ∡ABC (mod 360◦ )
+
As such, for a triangle △ABC we have that r(△ABC) = △r(A)r(B)r(C) ∼
= △ABC.

rO,180◦ = sO = hO,−1 is a reflection about O. Now, we merge a rotation with a composition to get a spiral
similarity SO,k,θ = rO,θ ◦ hO,k = hO,k ◦ rO,θ . Then we have that

(i) SO,k1 ,θ1 ◦ SO,k2 ,θ2 = SO,k1 k2 ,θ1 +θ2

(ii) S−1
O,k1 ,θ = SO,k−1 ,−θ
1

(iii) SO,−k,θ = SO,k,θ+180◦

Notably, we can assume that spiral similarities have positive ratios. For completeness, translations can also
be considered spiral similarities with center at infinity with k = 1, θ = 0.
T
Earlier we defined the Miquel point for four lines ℓ1 , ℓ2 , ℓ3 , ℓ4 as (ℓi ℓi+1 ℓi+2 ), let us now expand on this
definition.

Definition 1.2.1. For four points A1 , A2 , B1 , B2 satisfying A1 ̸= A2 , B1 ̸= B2 , A1 ̸= B1 , A2 ̸= B2 (notably

we allow A1 = B2 or A2 = B1 ), we define the Miquel point of the quadrilateral A1 B1 B2 A2 to be the Miquel
point of the four lines (A1 B1 , B1 B2 , B2 A2 , A2 A1 ).

Here are some degenerate cases.

• If any three of these points are collinear (WLOG here A1 , B1 , B2 ), then the circumcircle of the
degenerate triangle △(A1 B1 )(B1 B2 )(B2 A2 ) is the circle passing through B1 , B2 that is tangent to
B2 A2 .

• If any two lines are parallel (WLOG A1 B1 ∥ B2 A2 ), then we take the circumcircle of △(A1 B1 )(B1 B2 )(B2 A2 )
to be B1 B2 .

49
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

• If all four points lie on a line, then M is also on the line such that

M B1 M B2
= .
M A1 M A2

It is easy to show that such a point M still exists in these degenerate cases.

(The book often writes the quadrilateral A1 B1 B2 A2 as (A1 B2 )(B1 A2 ) to make the definitions more
symmetrical, which will be used in the case of complete quadrilaterals. Here, A1 B2 , B2 A1 are the 42 − 2

lines not in the aforementioned four lines). Then the major property of the Miquel point is the following:

Proposition 1.2.2. If O is the center of the spiral similarity that maps A1 A2 to B1 B2 , then M is the Miquel
point of a complete quadrilateral (A1 B2 )(B1 A2 ).

Proof. We will only prove the result in the non-degenerate case outlined above. If SO,k,θ is the transformation,
then take k as positive. We then have that

OB1 OB2
k= = , θ = ∡A1 OB1 = ∡A2 OB2 (mod 360◦ ),
OA1 OA2

+ +
and that △OA1 B1 ∼ OA2 B2 . By (TODO 0.4.19), we also have that △OA1 A2 ∼ △OB1 B2 . As such, if
C = A1 B1 ∩ B2 A2 , D = B1 B2 ∩ A2 A1 , then

∡A1 CA2 = ∡A1 B1 O + ∡B1 OB2 + ∡OB2 A2

= (∡A1 B1 O − ∡A2 B2 O) + ∡A1 OA2 = ∡A1 OA2 ,

∡A1 DB1 = ∡A1 A2 O + ∡A2 OB2 + ∡OB2 B1

= (∡A1 A2 O − ∡B1 B2 O) + ∡A1 OB1 = ∡A1 OB1 .

As such, O lies on (A1 A2 C) and (A1 B1 D). Similarly, O lies on (B1 B2 C) and (A2 B2 D) which finishes.

Proposition 1.2.3. For any four points A1 , A2 , B1 , B2 that satisfy A1 ̸= A2 , B1 ̸= B2 , there exists a unique
spiral similarity S such that S(A1 ) = B1 , S(A2 ) = B2 .

Proof. If A1 ̸= B1 , A2 ̸= B2 , then let O be the Miquel point of (A1 B2 )(B1 A2 ). To check all (possibly
degenerate) cases.

If there are no degeneracies and O ̸= ∞, then set C = A1 B1 ∩ B2 A2 . We have


∡OA1 B1 = ∡OA1 C = ∡OA2 C = ∡OA2 B2 ,

+
△OA1 B1 ∼ △OA2 B2
∡OB1 A1 = ∡OB1 C = ∡OB2 C = ∡OB2 A2 ,


50
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

so if we set
OB2
k= , θ = ∡A1 OB1 = ∡A2 OB2 (mod 360◦ )
OA2

then the spiral similarity S = SO,k,θ works.

There’s some other casework for degenerate cases, check the original book if you care.

The most notable degenerate case is when A1 A2 B1 B2 is a parallelogram, in which case the spiral similarity
becomes a translation.

Proposition 1.2.4. The composition of two spiral similarities is another spiral similarity. Further, the angle
rotated in the composition of two spiral similarities is the sum of the angles rotated in the two individual
spiral similarities, and the scale factor in the composition is the product of the scale factors in the two
individual spiral similarities.

Proof. The original book’s proof is casework.

Alternatively, we can note that over C, a spiral similarity consists of the mapping z 7→ keiθ · z + C for
some complex C. Then applying this twice makes the result immediate.

Practice Problems

Problem 1. Let circle (P ) and circle (Q) intersect in two points. Draw a pair of perpendicular lines
through one of these intersection points and let one of these lines intersect P Q, (P ), and (Q) at A, B,
and C respectively. The other line intersects P Q, (P ), and (Q) at D, E, and F respectively. Prove that
AB : AC = DE : DF .

Problem 2. For an arbitrary triangle △ABC, let D1 and D2 be points on BC, let E1 and E2 be arbitrary
points on CA, let F1 and F2 be arbitrary points on AB. Let M1 and M2 respectively be the Miquel points of
+
△D1 E1 F1 and △D2 E2 F2 wrt. △ABC. Prove that △D1 E1 F1 ∼ △D2 E2 F2 if and only if M1 = M2 .

Problem 3 (ISL 2014 G4). Consider a fixed circle Γ with three fixed points A, B, and C on it. Also, let us
fix a real number λ ∈ (0, 1). For a variable point P ̸∈ {A, B, C} on Γ, let M be the point on the segment
CP such that CM = λ · CP . Let Q be the second point of intersection of the circumcircles of the triangles
AM P and BM C. Prove that as P varies, the point Q lies on a fixed circle.

1.3 Isogonal Conjugation

Isogonal conjugation is one of the most important concepts in plane geometry, and is a commonly used tool
in olympiad problems.

51
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Definition 1.3.1. Given two lines ℓ1 , ℓ2 that intersect at P , call two lines L1 , L2 passing through P isogonal
lines with respect to ∠(ℓ1 , ℓ2 ) if and only if L1 , L2 are antiparallel to ℓ1 , ℓ2 .

Proposition 1.3.2 (Isogonal Ratios). Given a fixed triangle △ABC, if D, D∗ are on line BC, then AD, AD∗
being isogonal lines with respect to ∠BAC implies that

2
BD BD∗

AB
· =
DC D∗ C CA

Proof. By Proposition 0.3.4, we have that

−−→ −−→
BC/DC BD∗ /D∗ C sin ∡(AB, AD) sin ∡(AB, AD∗ )
· = −→ · −→ = 1.
AB/CA AB/CA sin ∡(AD, AC) sin ∡(AD∗ , AC)

We can further define an important object, the complete n-gon, written as

N = (ℓ1 , ℓ2 , . . . , ℓn )

n


representing the n-gon formed by the 2 intersections of ℓi , ℓj , such that no three lines are concurrent.

Definition 1.3.3. Given a complete n-gon N = (ℓ1 , ℓ2 , . . . , ℓn ), we define Aij = ℓi ∩ ℓj . If for some point P ,
there exists a point P ∗ such that for all i, j, Aij P, Aij P ∗ , are isogonal lines with respect to ∠(ℓi , ℓj ), then we
say that P and P ∗ are isogonal conjugates with respect to N . Notably, (P ∗ )∗ = P .

Example 1.3.4. The orthocenter H and circumcenter O are isogonal conjugates wrt △ABC, and I, I a , I b , I c
are fixed points under isogonal conjugation.

Proposition 1.3.5. For a triangle △ABC and point P , a isogonal conjugate P ∗ of P always exists, potentially
at infinity.

Proof. One way to do this is through Ceva’s:

Y sin ∡(−
−→ −→
AB, A∞AB+AC−AP ) Y sin ∡(AP, AC)
−→ = −−→ =1
sin ∡(A∞AB+AC−AP , AC) sin ∡(AB, AP )

Here’s a “synthetic” approach to showing existence. First suppose that P ̸∈ (ABC), L∞ . Define △Pa Pb Pc
as the circumcevian triangle of P (Pa = AP ∩ (ABC) and so forth).

Let D, E, F be the feet from P to Pb Pc , Pc Pa , Pa Pb , respectively. Then

∡EDF = ∡EDP + ∡P DF = ∡Pa Pc C + ∡BPb Pc = ∡BAC,

52
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

so similarly ∡F ED = ∡CBA, ∡DF E = ∡ACB. As such, △ABC ∼ △DEF . Let P ∗ be the point such that
+
△ABC ∪ P ∗ ∼ △DEF ∪ P . Then it follows that

∡(AP + AP ∗ , AB + AC) = ∡Pa AB + ∡P DF = ∡Pa Pb B + ∡P Pb F = 0◦

so P ∗ is the desired point.

Now suppose that P is on (ABC) or the line at infinity. Let ℓA , ℓB , ℓC be the isogonal conjugates of
AP, BP, CP wrt ∠A, ∠B < ∠C respectively. Then

∡(ℓB , ℓC ) = ∡(ℓB , BC) + ∡(BC, ℓC ) = ∡ABP + ∡P CA


0◦ ,

P ∈ (ABC)
=
∡BAC, P ∈ L∞


As such, if P ∈ (ABC) then ℓA ∥ ℓB ∥ ℓC converge at infinity. Else, if P ∈ L∞ then we get that

ℓB ∩ ℓC ∈ (ABC), so ℓA ∩ ℓB ∩ ℓC converge on (ABC).

Proposition 1.3.6. Given a triangle △ABC, points P, Q are isogonal conjugates if and only if

∡CP A + ∡CQA = ∡CBA, ∡AP B + ∡AQB = ∡ACB.

Proof. First suppose they are isogonal conjugates. Then

∡CP A + ∡CQA = ∡P CB + ∡CBA + ∡BAP + ∡CQA

= ∡ACQ + ∡CBA + ∡QAC + ∡CQA = ∡CBA

The other angle condition follows by symmetry.

For the other direction, note that the forward direction gives ∡CQA = ∡CBA − ∡CP A = ∡CP ∗ A, so
Q ∈ (CAP ∗ ). As such, by symmetry

Q ∈ (CAP ∗ ) ∩ (ABP ∗ ) = {A, P ∗ }

so Q = P ∗ .

Example 1.3.7 (Generalized Six Point Circle). Suppose the circle Γ intersects the sides BC, CA, and
AB at D1 and D2 , E1 and E2 , and F1 and F2 respectively. Let P1 = (F1 BD1 ) ∩ (D1 CE! ) ̸= D1 , and let
P2 = (F2 BD2 ) ∩ (D2 CE2 ) ̸= D2 . Prove that ∡BAP1 = ∡P2 AC.

53
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Proof. By the Triangle Miquel, P1 ∈ (E1 AF1 ) and P2 ∈ (E2 F A2 ). Now

∡BP1 C + ∡BP2 C = ∡BP1 D1 + ∡D1 P1 C + ∡BP2 D2 + ∡D2 P2 C

= ∡AF1 D1 + ∡D1 E1 A + ∡F1 F2 D2 + ∡D2 E2 E1

= ∡BAC + ∡E1 D1 F1 + ∡F1 D1 D2 + ∡D2 D1 E1

= ∡BAC.

We can get analogous equations for all three vertices of the triangle, so by (TODO 1.3.6) P1 and P2 are
isogonal conjugates in △ABC - in particular ∡BAP1 = ∡P2 AC.

E1

F1 P1

E2
F2
P2
C
B D2 D1

−
Since △AE1 F1 ∪ P1 ∼ △AF2 E2 ∪ P2 , we also have

∡(P1 D1 , BC) = ∡(P1 E1 , CA) = ∡(P1 F1 , AB)

= −∡(P2 D2 , BC) = −∡(P2 E2 , CA) = −∡(P2 F2 , AB)

which generalizes to n-gons as follows:

Proposition 1.3.8 (Generalized Pedal Circles). Given a complete-n gon N = (ℓ1 , ℓ2 , . . . , ℓn ), a point P ,
and an angle α ̸= 0◦ , take a point Pi on ℓi such that ∡(ℓi , P Pi ) = α. Then P has an isogonal conjugate with

54
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

respect to N if and only if P1 , P2 , . . . , Pn are concyclic.

In this case, let P ∗ be this isogonal conjugate. Define Pi∗ as the point on ℓi so that ∡(ℓi , P ∗ Pi∗ ) = −α.
Then Pi∗ ∈ (P1 P2 . . . Pn ) = Γα . for all i. (This is a generalization of the six-point circle theorem.) Additionally,
the center of Γα , or Oα , is on the perpendicular bisector of P P ∗ and satisfies ∠Oα P P ∗ = 90◦ − α.

Proof. Define Aij = ℓi ∩ ℓj . For the forward direction, let P ∗ be the isogonal conjugate of P with respect to
−
N , and define a Pi∗ for every ℓi such that ∡(ℓi , P ∗ Pi∗ ) = −α. Then from △Aij Pi Pj ∪ P ∼ △Aij Pj∗ Pi∗ ∪ P ∗
we know that Pi , Pj , Pi∗ , Pj∗ are concyclic. Since ℓi , ℓj , ℓk are not concurrent, we can use (TODO 0.4.15) to
get that Pi , Pj , Pk , Pi∗ , Pj∗ , Pk∗ are concyclic, and thus P1 , P2 , . . . , Pn , P1∗ , P2∗ , . . . Pn∗ are concyclic.

When α = 90◦ , we have that the circumcenter (let’s call it O90◦ ) lies on all of the perpendicular bisectors
̸ 90◦ , we can consider the spiral similarity with center P , SP,(sin α)−1 ,α−90◦ , and we get
of Pi Pi∗ . Given α =
that Γa = S(Γ90◦ ), therefore Oα = S(O90◦ ) lies on the perpendicular bisector of P P ∗ and satisfies

∡Oα P P ∗ = ∡Oα P O90◦ = 90◦ − α,

which proves this half.

For the backwards direction, let Pi∗ be the second intersection of (P1 P2 P3 . . . Pn ) with ℓi . For any
△Ajk Aki Aij , take lines
∡(ℓi , L∗i ) = ∡(ℓj , L∗j ) = ∡(ℓk , L∗k ) = −α

such that L∗i , ℓi , and (P1 P2 . . . Pn ) occur.

Then by Example 1.3.7, L∗i , L∗j , L∗k concur at the isogonal conjugation P ∗ of P with respect to △Ajk Aki Aij .
Thus L∗1 , L∗2 , · · · , L∗n are concurrent, suppose at point P ∗ , thus P ∗ is the isogonal conjugate of P in N .

So we can say that it’s similar to the situation for a triangle, in which we can say that P is the Miquel
point of (P1 , . . . , Pn ) in N = (ℓ1 , . . . , ℓn ).

When α = 90◦ , we can get rid of the directed angles.

Proposition 1.3.9. Given a complete n-gon N (ℓ1 , ℓ2 , . . . , ℓn ) and a point P , let Pi be the foot from P to ℓi .
Define P ∗ as the isogonal conjugate of P and define Pi∗ similarly. Then P1 , P2 , . . . , Pn , P1∗ , P2∗ , . . . , Pn∗ are
concyclic, with circumcenter at the midpoint of P P ∗ .

We call this circle the pedal circle of P . Additionally, if N = △ABC, we call the triangle △Pa Pb Pc made
by the feet of P to BC, CA, AB the pedal triangle of P . These concepts lead to a nice way to prove two
points are isogonal conjugates without resorting to cevian angles.

55
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

P∗
O P

Example 1.3.10. I is its own isogonal conjugate.

We prove this by noting the pedal triangle of I is simply the intouch triangle, and its pedal circle is just
the incircle. However, the incircle is tangent to the three sides of the triangle, so I = I ∗ .

Example 1.3.11. Prove O and H are isogonal conjugates.

Simply note that the pedal triangles of O and H are the medial and orthic triangles respectively, and
these both lie on the nine-point circle.

Proposition 1.3.12. Let △Pa Pb Pc be the pedal triangle of an arbitrary point P inside △ABC, and let
+
△PA PB PC be the circumcevian triangle of P . Then △Pa Pb Pc ∼ △PA PB PC .

Proof. The proof is a simple angle-chase, using all the cyclic quads that pedal triangles give us.

∡Pb Pa Pc = ∡Pb Pa P + ∡P Pa Pc = ∡ACP + ∡P BA

= ∡APA PC + ∡PB PA A = ∡PB PA PC ,

and similarly for the other three angles.

Proposition 1.3.13. The area of the pedal triangle △Pa Pb Pc is given by

Pow(ABC) (P )
[△Pa Pb Pc ] = · [△ABC],
4R2

where R is the circumradius of △ABC.

Proof. By the area formula in (TODO 0.2.12), we have that

1
[△Pa Pb Pc ] = · Pc Pa · Pa Pb · sin ∡Pb Pa Pc
2
1
= · (BP · | sin ∡CBA|) · (CP · | sin ∡ACB|) · sin ∡PB PA PC ,
2

56
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

PC
PB

Pc Pb

B Pa C

PA

Figure 1.5: This immediately proves Simson Line as well.

where △PA PB PC is the circumcevian triangle of P . By the Law of Sines (TODO 0.2.2), we have that

CP · sin ∡PB PA PC = CP · sin ∡PB CP

= P PB · sin ∡BPB C = P PB · sin ∡BAC,

and therefore

1
[△Pa Pb Pc ] = · (BP · P PB ) · sin ∡BAC · sin ∡CBA · sin ∡ACB
2
1 [△ABC]
= · Pow(ABC) (P ) · .
2 2R2

By (TODO 1.3.9) where we set N to be a complete quadrilateral Q, we get that:

Proposition 1.3.14. Given any quadrilateral (AC)(BD), and any point P , P has an isogonal conjugate in
this quadrilateral Q if and only if (P A, P C) is antiparallel to (P B, P D).

Proof. Let P ’s feet onto AB, BC, CD, DA be W, X, Y, Z respectively. Note that P has an isogonal conjugate

57
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

in Q iff. W, X, Y, Z are concyclic.

∡W XY + ∡Y ZW = 0◦ ⇐⇒ ∡W XP + ∡P XY + ∡Y ZP + ∡P ZW = 0◦

⇐⇒ ∡ABP + ∡P CD + ∡CDP + ∡P AB = 0◦

⇐⇒ ∡AP B + ∡CP D = 0◦ .

Note that this way to find all points with an isogonal conjugate gives a mysterious locus of points with
isogonal conjugates. We will fully hunt down and defeat this locus in section (TODO 9.2), but for now don’t
worry too much about it.

For deeper investigation, we will explore another way to characterize isogonal conjugates.

Definition 1.3.15. Given two triangles △ABC and △DEF , we say they are orthologic if the perpendicular
A∞EF from A to EF , B∞F D , and C∞DE concur at a point P . We call P the center of orthology.

A
E

B C
D

Suppose such a P exists. By (TODO perpcr), we have that

AE 2 − AF 2 = P E 2 − P F 2

BF 2 − BD2 = P F 2 − P D2

CD2 − CE 2 = P D2 − P E 2 .

Adding these three equations, we get

AE 2 − AF 2 + BF 2 − BD2 + CD2 − CE 2 = 0.

However now note that the roles of △ABC and △DEF are symmetric!

58
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

This gives us that orthology is symmetric!!

Obviously, the pedal triangle △Pa Pb Pc of any point P not on L∞ is orthologic to △ABC (the perpen-
diculars concur at P ). Thus, by symmetry of orthology, the three lines A∞⊥Pb Pc , B∞⊥Pc Pa , C∞⊥Pc Pa are
concurrent. Let this concurrency point be Q. Since P is the antipode of A on (APb Pc ), we have that the two
lines AQ = A∞⊥Pb Pc and AP are isogonal in ∠A. Similarily, BQ, CQ and BP, CP are pairwise isogonal.
Thus Q is the isogonal conjugate of P !

Pb

Pc
P
Q

B Pa C

By this orthogonality characterization, we can derive some more properties.

Let us first define the antipedal triangle of P as the triangle for which △ABC is the P -pedal triangle, so
just the triangle formed by the intersections of the perpendiculars to AP, BP, CP at A, B, C.

Proposition 1.3.16. Given △ABC, let O be the circumcenter of △ABC, and let (P, P ∗ ) be a pair of
isogonal conjugates in △ABC. Let (P ∗ )′ be the reflection of P ∗ in O. Let △Pa Pb Pc be the pedal triangle of
P , and let △Pa∗ Pb∗ Pc∗ be the antipedal triangle of P ∗ . Then we have:

• (P ∗ , (P ∗ )′ ) are isogonal in △Pa∗ Pb∗ Pc∗ ,

+
• △Pa Pb Pc ∪ P ∼ △Pa∗ Pb∗ Pc∗ ∪ (P ∗ )′ ,

• Let Q be the isogonal conjugate of P in △Pa Pb Pc , then P Q ∥ OP ∗ .

Proof. We prove these sequentially.

• Note that the pedal circle of P ∗ with respect to △Pa∗ Pb∗ Pc∗ is just (ABC), so the isogonal conjugate of
P ∗ wrt. △Pa∗ Pb∗ Pc∗ is just the reflection of P ∗ over O.

59
AoPS Chapter 1. Geometric Transformations - Hidden/Completed
+
• By definition, we can use two perpendicularities to get that △P Pb Pc ∼ △(P ∗ )′ Pb∗ Pc∗ , are homothetic,
and similarly we get the two other similarities

+ +
△P Pc Pa ∼ △(P ∗ )′ Pc∗ Pa∗ , △P Pa Pb ∼ △(P ∗ )′ Pa∗ Pb∗ ,

+
so we get that △Pa Pb Pc ∪ P ∼ △Pa∗ Pb∗ Pc∗ ∪ (P ∗ )′ .

• Since
+
△Pa Pb Pc ∪ P ∪ Q ∼ △Pa∗ Pb∗ Pc∗ ∪ (P ∗ )′ ∪ P ∗ ,

are homothetic, we get that P Q ∥ (P ∗ )′ P ∗ = OP ∗ .

Pb∗

Pc∗

Pb
P∗
O
Pc Q
P (P ∗ )′

B Pa C

Pa∗

Practice Problems

Problem 1. In △ABC, let the reflections of BC with respect to AB and AC meet at K. Prove that AK
passes through the circumcenter of △ABC.

Problem 2. Let M be the midpoint of BC in △ABC, and let I1 and I2 be the incenters of △ABM and
△ACM , respectively, Prove that (AI1 I2 ) passes through the midpoint of the arc BAC.
’

60
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Problem 3. △ABC is an equilateral triangle with circumcircle Γ and let P be a point in the plane not on
Γ. Let P A, P B, and P C meet Γ at A1 , B1 , and C1 , respectively. Prove that there are only two points P
such that △A1 B1 C1 is a (positive?) triangle, and they are collinear with the circumcenter of △ABC.

Problem 4. Let (P, P ∗ ) be a pair of isogonal conjugates with respect to △ABC. AP and AP ∗ intersect
(ABC) at U and B respectively, and AP ∩ BC = T . Prove that

PT AP ∗
= .
TU P ∗V

Problem 5. Let I be the incenter of △ABC, let M be the midpoint of AI, and let BI and CI intersect
(ABC) at E and F respectively. Take points X, Y on AE, AF respectively so that ∡XBC = ∡ABM and
∡Y CB = ∡ACM . Prove that I, X, and M are collinear.

Problem 6 (USA TST 2010/7). Let (P, Q) be a pair of isogonal conjugates with respect to △ABC, and let
D be a point on BC. Prove that ∡AP B + ∡DP C = 180◦ if and only if ∡AQC + ∡DQB = 180◦ .

Problem 7 (ISL 2007 G2). The diagonals of a trapezoid ABCD intersect at point P . Point Q lies between
the parallel lines BC and AD such that ∠AQD = ∠CQB, and points P and Q are on opposite sides of line
CD. Prove that ∠BQP = ∠DAQ.

Problem 8. Let N9 be the center of the nine-point circle of △ABC. Let B ′ and C ′ be the reflections of B
and C about CA and AB. Prove that B ′ C ′ is perpendicular to the reflection of AN about the angle bisector
of ∠BAC.

Problem 9 (APMO 2010/4). Let ABC be an acute angled triangle with AB > BC and AC > BC. Denote
by O and H the circumcenter and orthocenter, respectively, of △ABC. Suppose that (AHC) ∩ AB = M =
̸ A
and (AHC) ∩ AC = N ̸= A. Prove that the circumcenter of △M N H lies on OH.

Problem 10. Given △ABC and a line ℓ, let D, E, and F be the feet of the altitudes from A, B, and C to ℓ
respectively. Prove that the altitude from D to BC, the altitude from E to CA, and the altitude from F to
AB concur. This point is called the orthopole of ℓ with respect to △ABC.

Problem 11 (Taiwan APMOC 2015/5). In △ABC, points L, M , and N lie on the sides BC, CA, AB
respectively so that △AN M , △BLN , and △CM L are all acute. Let HA , HB , and HC be the orthocenters
of these three triangles (respectively). Suppose AHA , BHB , and CHC are concurrent. Prove that LHA ,
M HB , and N HC are also concurrent.

Problem 12 (Taiwan TST 2021/3J/6). Let ABCD be a rhombus with center O. Select point P lying on
segment AB. Let I, J, and L be the incenters of △P CD, △P AD, and △P BC respectively. Let H and K
be the orthocenters of △P LB and △P JA, respectively. Prove that OI ⊥ HK.

61
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Problem 13. Scalene △ABC has intouch triangle △DEF and incenter I. Let Y and Z be the midpoints of
DF and DE, respectively. Let J be the isogonal conjugate of I in △AZY . Prove that IJ ⊥ AS.

1.4 Simson and Steiner lines

In the previous section, we investigated isogonal conjugates and pedal circles. We know that the isogonal
conjugate of the circumcircle is the line at infinity, but it’s slightly hard to define pedal triangles and pedal
circles for points on the line at infinity. (We will do this in a later chapter!) So we first will instead just look
at the pedal circles of points on the circumcircle.

Theorem 1.4.1 (Simson Line). The pedal triangle of any point P on the circumcircle (ABC) is actually
just a line.

Proof. Let Pa , Pb , Pc be the feet from P to BC, CA, AB. By Miquel’s theorem, we get that Pa lies on Pb Pc
if and only if (BAC), (CPb Pa ), (Pa Pc B) concur. The circles concur at P , so we are done.

(Another immediate proof is by usage of (TODO 1.3.13), to get that the area of the pedal triangle is
zero.)

We call the line formed by Pa Pb Pc the Simson line of P with respect to △ABC. Additionally, the line
PA PB PC formed by the reflections of P over BC, CA, AB (which is just hP,2 (Pa Pb Pc )) is known as the
Steiner line of P .

Theorem 1.4.2 (Steiner’s Theorem). The Steiner line SP of any point P on (ABC) will always pass through
the orthocenter H.

Proof. Let Ω be the circumcircle.

Let PB , PC respectively be the reflection of P across CA and AB. Let HB , HC respectively be the
reflections of H across CA and AB. By 1, HB and HC lie on (ABC).

∡HPB − ∡HPC = (2∡CA − ∡HB P ) − (2∡AB − ∡HC P )

= (2C − HB − 2B + H)C)Ω

= 2C− ⊥ (C + A − B) − 2B+ ⊥ (A + B − C) = 0◦

So H lies on PB PC = SP .

Since the Simson line gets sent to the Steiner line under hP,2 , we get a simple corollary:

Theorem 1.4.3. Let H be the orthocenter of △ABC. The Simson line of any point P will bisect HP .

62
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Example 1.4.4 (2009 ISL G8). Let quadrilateral ABCD have an incircle. Define g to be a line passing
through A, which intersects the segments BC at M and CD at N . Let I1 , I2 , I3 be the incenters of
△ABM, △M N C, △N DA, respectively. Prove that the orthocenter of △I1 I2 I3 lies on line g.

Solution. Instead of proving g passes through the orthocenter of △I1 I2 I3 , we can instead show that g is the
Steiner line for some point P , or that the 3 reflections of g over I1 I2 , I2 I3 , I3 I1 concur at a point P ∈ (I1 I2 I3 ).
Note that the reflection of g over I2 I3 is line CD, the reflection of g over I1 I2 is BC, so we only have to
prove that the reflection of g over I3 I1 (let it be g ′ ) passes through C. (Then from converse Simson we get
that C ∈ (I1 I2 I3 ).

Since g is one of the common internal tangents to (I3 ), (I1 ), g ′ is the other internal tangent. Let T3 , T1 be
g’s touchpoints with (I3 ), (I1 ) respectively, then we just have to prove the length of the tangent from C to
(I3 ) intersection with (I1 ) is just T1 T3 . We get a length of

(T1 M + M C) − (T3 N − CN ) = M C + CN − T1 T2 − N M.

We can re-express T1 T3 too, as

2 · T3 T1 = 2 · AT1 − 2 · AT3 = (M A + AB − BM ) − (DA + AN − N D)

= (AB − DA) − N M + N D − BM

= (BC − CD) − N M + N D − BM since (AC)(BD) is tangential quadrilateral

= CM + N C − N M.

So the two lengths are equal and we’re done.

Proposition 1.4.5. Given a fixed point P on (ABC), let SP be either the Simson or Steiner line of P . Then
the isogonal conjugate of the point at infinity on SP is the antipode of P in (ABC).

Proof. Since ∞SP lies on the line at infinity, the isogonal conjugate of it will lie on (ABC). So we only need to
prove that A∞SP ∥ Pb Pc , and from this we only need to prove that (Pb Pc , AP ∗ ) is antiparallel with (CA, AB).
We consider the triangle △APb Pc . We know that AP, A∞⊥Pb Pc are isogonal in ∠Pb APc , and combined with
Pb Pc ⊥ A∞⊥Pb Pc and AP ∗ ⊥ AP, we get that (Pb Pc , AP ∗ ) is antiparallel with (Pb A, APc ) = (CA, AB), so
we are done.

Typically we can also directly angle-chase with the line argument of the Simson/Steiner line. Since

63
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

∡(SP ) = ∡(HPA ), where PA is the reflection of P over BC, we get

∡SP + ∡AP = 2∡BC − ∡Ha P + ∡AP

= 2∡BC − ∡Ha A + (∡AB + ∡AC − ∡BC)

= ∡AB + ∡AC + 90◦

which implies that (SP , AP ∗ ) is antiparallel with (AB, AC). This can also be written as

(A + B + C − P )(ABC) =⊥ SP

If we have two points P, Q ∈ (ABC), then A∞SP , AP ∗ and A∞SQ , AQ∗ are pairs of isogonal lines in
∠BAC. Therefore from AP ⊥ AP ∗ , AQ ⊥ AQ∗ , we can get:

Corollary 1.4.6. In △ABC and two points P, Q on (ABC), the angle between their Steiner/Simson lines
SP , SQ is just
∡(SP , SQ ) = ∡QAP.

This can also be rephrased as for a fixed Q and a moving P on (ABC), ∡(SP ) + ∡AP is constant.

Example 1.4.7. Let P , Q be two antipodal points on the circumcircle of △ABC. Show that the Simson
lines of P , Q intersect at a point on the nine-point circle of △ABC.

Solution. Let H be the orthocenter of △ABC, and let MP , MQ be the midpoints of HP and HQ. Note that
there is a homothety hH, 12 which sends (ABC) to the nine-point circle, so we have that MP , MQ lie on the
nine-point circle. From (TODO 1.4.3) we know that MP ∈ SP , MQ ∈ SQ .

Since P Q passes through the circumcenter of △ABC, we get that MP MQ is the diameter of the nine-point
circle. So from (TODO 1.4.6) we get that

∡(SP , SQ ) = 90◦ ,

and therefore SP , SQ intersect on the nine-point circle. (We can also use Fontene’s second theorem, introduced
in Chapter 8.)

64
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

P
A

MP

N9
O
H

B C
MQ

Practice Problems

Problem 1. Let H be the orthocenter of △ABC, let M be the midpoint of BC, and let N be the midpoint
of BAC
’ on (ABC). Let D be a point on CA so that M D bisects HN . Prove that CD = DA + AB.

Problem 2 (IMO 2007/2). Consider five points A, B, C, D and E such that ABCD is a parallelogram and
BCED is a cyclic quadrilateral. Let ℓ be a line passing through A. Suppose that ℓ intersects the interior of
the segment DC at F and intersects line BC at G. Suppose also that EF = EG = EC. Prove that ℓ is the
angle bisector of ∠DAB.

Problem 3. Given △ABC, let P1 , P2 , P3 ∈ (ABC). Prove that the Simson lines of P1 , P2 , and P3 concur
if and only if
∡ABP1 + ∡BCP2 + ∡CAP3 = 0◦ .

Problem 4. Let H be the orthocenter of △ABC and let P ∈ (ABC). Suppose HP ∩ CA = Q and
HP ∩ AB = R. Prove that AP QR is cyclic.

Problem 5 (Taiwan TST 2020/3J/5). Let O and H be the circumcenter and the orthocenter, respectively,
of an acute triangle ABC. Points D and E are chosen from sides AB and AC, respectively, such that A, D,
O, E are concyclic. Let P be a point on the circumcircle of triangle ABC. The line passing P and parallel
to OD intersects AB at point X, while the line passing P and parallel to OE intersects AC at Y . Suppose
that the perpendicular bisector of HP does not coincide with XY , but intersect XY at Q, and that points
A, Q lies on the different sides of DE. Prove that ∠EQD = ∠BAC.

65
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

1.5 Isotomic Conjugation and Trilinear Polars

We now define the evil twin of isogonal conjugation — isotomic conjugation.

Definition 1.5.1. Given two points P1 , P2 , we can say that two points A1 , A2 on P1 P2 are isotomic points
if the midpoint of A1 A2 is also the midpoint of P1 P2 .

And now we will extend this definition to complete n-gons, which will be represented as (P1 , P2 , . . . , Pn ),
meaning the polygon formed by these n points and the n2 lines connecting them, where no three points are

collinear.

Definition 1.5.2. Given any complete n-gon (P1 , P2 , . . . , Pn ), define Lij = Pi Pj . Given any line ℓ, if there
exists a line ℓ∗ , such that for distinct i, j, Lij ∩ ℓ and Lij ∩ ℓ∗ are isotomic points in Pi Pj , then we call ℓ and
ℓ∗ isotomic lines in (P1 , P2 , . . . , Pn ).

Example 1.5.3. For an arbitrary triangle △ABC, the sides of the medial triangle △Ma Mb Mc are isotomic
lines with themselves in △ABC. Also, the line at infinity L∞ is its own isotomic line.

From Problem 4 in (TODO 1.1), we know that:

Proposition 1.5.4. Any line ℓ that is not the three sides of △ABC has an isotomic line ℓ∗ .

Note that this problem also tells us that ℓ∗ is the complement of the Newton line of △ABC ∪ ℓ. Back in
(TODO 1.3.14) we got a condition for when a point has an isogonal conjugate in a complete quadrilateral,
and now we will consider when a line has an isotomic line in a complete quadrangle (a set of four points and
six lines).

(Note the “duality” showing up again!)

Proposition 1.5.5. Given a complete quadrangle ⨿ = (A, B, C, D) with no points at infinity, a line ℓ ̸= L∞ ,
let PXY = XY ∩ ℓ. Then ℓ has an isotomic line if and only if the midpoint of PCA PBD is also the midpoint
of PAB PCD .

Proof. Let PXY ∗ be the isotomic conjugate with respect to XY and let MXY be the midpoint of XY . Then

−−−−−−−→ −−−−−−−→ −−−−−−→

PCA∗ PAB∗ = 2 · MCA MAB − PCA PAB
−−−−−−−→ −−−−−−−→ −−−−−−→
PBD∗ PCD∗ = 2 · MBD MCD − PBD PCD .

−−−−−−−→ −−−−−−−→
Note that (MCA MBD )(MAB MCD ) is a parallelogram so MCA MAB = −MBD MCD . These two equations
hence add up to
−−−−−−−→ −−−−−−−→ −−−−−−→ −−−−−−→
PCA∗ PAB∗ + PBD∗ PCD∗ = −(PCA PAB + PBD PCD ) (ã)

66
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

If the midpoints of PCA PBD and PAB PCD coincide, then the RHS of our above equation is 0, so PCA∗ PAB∗ ∥
PBD∗ PCD∗ (or the lines coincide). Note that PBC∗ , PCA∗ , and PAB∗ are collinear, and PBC∗ , PBD∗ , PCD∗
are collinear - since PBC∗ lies on both lines, in fact these lines coincide on a line. Since PCA∗ lies on line
PAD∗ PDC∗ , PCA∗ also lies on this line, so this line is the isotomic line of ℓ.

Conversely, suppose ℓ has an isotomic line ℓ∗ , or that PXY ∗ are collinear for all choices of X, Y ∈
{A, B, C, D}, then (ã) tells us that

−−−−−−→ −−−−−−→
PCA PAB + PBD PCD = 0 or ℓ ∥ ℓ∗ .

−−−−−−−→
In the former case, the conclusion is clearly valid. In the other case, we get that MAB MCD is parallel to
−−−−−−−→
MAC MBD and so forth so either AD ∥ BC or A+D 2 + B+C
2 and thus ⨿ is a parallelogram which implies the
result.

Now we want to expand the idea of isotomic conjugation from lines to points, using trilinear poles (see
Exercise 2 in section 1.1). Formally, we have:

Definition 1.5.6 (Trilinear Polar). Reference triangle △ABC, as usual.

• For a point P ̸= A, B, C, let △Pa Pb Pc be the cevian triangle of P (As a reminder, the triangle with
points AP ∩ BC, BP ∩ AC, CP ∩ AB. We define the trilinear polar of P wrt. △ABC as the perspectrix
of △ABC and △Pa Pb Pc , notated as t(P ).

• For a line ℓ ̸= BC, CA, AB, let △ℓa ℓb ℓc be the cevian triangle of ℓ (defined as the triangle formed by
the lines ℓa = A(BC ∩ ℓ), ℓb = B(CA ∩ ℓ), ℓc = C(AB ∩ ℓ)). Then we define the trilinear pole of ℓ as
the perspector of △ℓa ℓb ℓc and △ABC, also notated as t(P ).

(These both exist because of Desargues’s theorem (TODO 1.1.1)).

Now that we have defined the cevian triangles, we will defined anticevian triangles (which will be used
much later in the future).

Definition (Anticevian Triangle). The anticevian triangle of a point or line with respect to a triangle △ABC
is the triangle △DEF such that the cevian triangle wrt △DEF is △ABC.

The proof for existence is omitted here, but a homography suffices.

We can then use more Desargues’s to show that this operation “t” is an involution, since t(t(P )) = P and
t(t(ℓ)) = ℓ.

67
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Example 1.5.7. The cevian triangle of L∞ with respect to △ABC is

△(A∞BC )(B∞CA )(C∞AB ),

∁ ∁ ∁
the anticomplementary triangle of △ABC, △A B C . Thus the trilinear pole of L∞ is the centroid G.

Example 1.5.8. Let H be the orthocenter of △ABC. Then the trilinear polar of H wrt. △ABC is the
radical axis of the circumcircle (ABC) and the nine-point circle ϵ (this is known as the orthic axis in English).

Definition 1.5.9. We define the isotomic conjugate of a point P as

P ′ = t(t(P )∗ )

(the trilinear pole of the isotomic line of P ’s trilinear polar).

This might sound like a very arbitrary definition, but we can actually characterize P ′ much easier.

Proposition 1.5.10. Given a triangle △ABC and P ̸= A, B, C, let P ′ be the isotomic conjugate of P . Then
we have that AP ∩ BC, AP ′ ∩ BC are isotomic points in segment BC. Similarly, BP ∩ CA, BP ′ ∩ CA are
isotomic in CA, and CP ∩ BC, CP ′ ∩ BC are isotomic in segment AB.

Y′

Z′
A
t(P )∗
Pc′ Pb′

P′
Pb
Pc
P X′
X B Pa Pa′ C
Z

t(P )

Proof. Let ℓ, ℓ′ respectively be the trilinear polars of P, P ′ wrt. △ABC, and let △Pa Pb Pc and △Pa′ Pb′ Pc′
be the cevian triangles of P and P ′ . Then we have that BC, ℓ, Pb Pc are concurrent by Desargues, let this
concurrency point b/e Qa . Similarly define Q′a as the concurrency point of BC, ℓ′ , Pb′ Pc′ . By Menelaus and
Ceva (or harmonics, later defined in (TODO 2.2.8)), we get

68
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

BPa BPc APb BQa CQ′ CP ′ AP ′ CP ′

= · =− = − ′ a = ′ b · ′ c = ′ a,
Pa C Pc A Pb C Qa C Qa B Pb A P c B Pa B

and thus Pa = AP ∩ BC, Pa′ = AP ′ ∩ BC are isotomic in BC.

For the rest of the section we will use this definition of isotomic conjugation.

Example 1.5.11. From (TODO 1.5.3) and (TODO 1.5.7), we can get that for △ABC, the vertices of its
∁ ∁ ∁
anticomplementary triangle △A B C are self-isotomic in △ABC. Also G is self-isotomic.

Another example of isotomic conjugates are the intouch and extouch points. From (TODO 0.3.9) we have
that
BD = Da C, CE = Eb A, AF = Fc B,

which gets us that the Gergonne point Ge = AD ∩ BE ∩ CF and the Nagel point N a = ADa ∩ BEb ∩ CFc
are isotomic conjugates.

(Note the similarity between the roles of the anticomplementary triangle + centroid in isotomic conjugation,
and the excentral triangle + incenter in isogonal conjugation. This is not a coincidence!)

Remark. We know that isogonal conjugation sends the line at infinity to the circumcircle (ABC). Isotomic
conjugation sends the line at infinity to an ellipse passing through A, B, C, called the Steiner circumellipse.
We will explore this further in (TODO 7.3) and (TODO 7.4).

We present another application of isotomic conjugation.

Proposition 1.5.12. Given an arbitrary triangle △ABC and a point P ̸= A, B, C, let △Pa Pb Pc be the
cevian triangle of P . Let Qa , Qb , Qc be the second intersections of (Pa Pb Pc ) with sides BC, CA, AB. Then
AQa , BQb , CQc concur at a point Q, known as the cyclocevian conjugate of P .

Proof. This is direct by an application of Carnot’s theorem (TODO 6.3.3), but we will present an elementary
proof by Ceva’s theorem. First,
BPa CPb APc
· · = 1.
Pa C Pb A Pc B

By Power of a Point,

APc · AQc = APb · AQb , BPa · BQa = BPc · BQc , CPb · CQb = CPa · AQa ,

69
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

so
   
BQa CQb AQc BPa CPb APc BQa CQb AQc
· · = · · · · ·
Qa C Qb A Qc B Pa C Pb A Pc B Qa C Qb A Qc B
BPa · BQa CPb · CQb APc · AQc
= · · = 1.
Pa C · Qa C Pb A · Qb A Pc B · Qc B

Finally, we combine isogonal conjugation, isotomic conjugation, and complementing into a monstrous way
to characterize cyclocevian conjugates.

Theorem 1.5.13 (Grinberg). Let P and Q be cyclocevian conjugates. Then

(P ′ )∁ , (Q′ )∁

are isogonal conjugates. In other words, if we let ϕG , ϕK represent the maps [P → P ′ ], [P → P ∗ ] (isotomic
and isogonal conjugation) respectively. (This notation will make more sense in chapter 7.), then the map
sending P to its cyclocevian conjugate is just

∁
ϕG ◦ (P ) ◦ ϕK ◦ (P )∁ ◦ ϕG .

For the proof of this ridiculous theorem, we first introduce a small lemma.

Lemma 1.5.14. For points P and (P ′ )∁ , their crosspoint P ⋔ (P ′ )∁ is the centroid of the cevian triangle
△Pa Pb Pc of P . Denote this point as GP , then we also have that the medial triangle of △Pa Pb Pc and △ABC
are perspective, with perspector (P ′ )∁ .

Proof. We wish to prove that A(P ′ )∁ bisects Pb Pc . Let △Ma Mb Mc be the medial triangle of △ABC, and
let MPb and MPc be the midpoints of BPb and CPc respectively. Then from Mb MPb ∥ BP ′ we get that
Mb MPb = (BP ′ )∁ , so (P ′ )∁ ∈ Mb MPb . By the same logic we get that (P ′ )∁ ∈ Mc MPc . Let △Na NPb NPc be
the medial triangle of APb Pc . Then Na = MPb NPc ∩ MPc NPb . By Desargues’s (TODO 1.1.1), A, Na , (P ′ )∁ are
collinear iff. MPb MPc , Mb Mc , NPc NPb are concurrent. By Menelaus we convert this into the length condition

Mb NPc ANPb Mb MPc Ma MPb

· = · .
NPc A NPb Mc MPc Ma MPb Mc

However the LHS and RHS are both equal to

CPb APc
· ,
Pb A Pc B

so we are done.

70
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

Proof of (1.5.13). Let △Pa Pb Pc , △Qa Qb Qc respectively be the cevian triangles of P and Q. Let MP and
− −
MQ be the midpoints of Pb Pc and Qb Qc . From △APb Pc ∼ △AQc Qb , we can get that △APb Pc ∪ MP ∼
△AQc Qb ∪ MQ . Therefore from lemma (TODO 1.5.14), we get that

∡A(P ′ )∁ + ∡A(Q′ )∁ = ∡AMP + ∡AMQ = = ∡AB + ∡AC,

so A(P ′ )∁ and A(Q′ )∁ are isogonal in ∠BAC. By symmetry we get that (P ′ )∁ and (P ′ )∁ are isogonal
conjugates in △ABC.

Example 1.5.15. In △ABC, we know that H and G are cyclocevian conjugates (see: nine-point circle).
Then by Grinberg’s theorem, we have that

∁
((((H ′ )∁ )∗ ) )′ = G
∁
(((H ′ )∁ )∗ ) = G′ = G

((H ′ )∁ )∗ = G∁ = G

(H ′ )∁ = G∗ = K.

So we get the isotomic conjugate of the orthocenter is the symmedian point of the anticomplementary triangle,
X69 . (We will further elaborate on this point in chapter 5).

Remark. In English, (P ′ )∁ is known as the isotomcomplement of P .

Practice Problems

Problem 1. Let the orthocenter of △ABC be H. Let P be a point on (ABC) and let S be the Simson line
of P with respect to △ABC. Prove that S has an isotomic line with respect to the complete quadrilateral
△ABC ∪ H.

Problem 2. Let (P, P ′ ) be a pair of isotomic conjugates in △ABC. Let Q be the isotomic conjugate of the
∁
anticomplement of P , Q = (P )′ . Prove that P, P ′ , Q are collinear.

1.6 Morley’s Equilateral Triangle

For this section, we will consider the triangle △ABC with its labels labeled anticlockwise. We will notate the
angle trisectors of ∠BAC as ℓBi
A , with i ∈ {−1, 0, 1} such that when i = 0, the angle trisectors are internal,

71
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

and
Bi Bj ◦
∡(ℓBi Bi
A , AC) = 2 · ∡(AB, ℓA ), ∡(ℓA , ℓA ) = (j − i) · 60 .

Then ℓCi Bi Ci Ai Ai Bi
A is defined as being an isogonal line to ℓA in ∠BAC. We can similarly ℓB , ℓB , ℓC , ℓC where

the second and forth are isogonal to the first and third respectively. Finally, define

Bj
aij = ℓCi
B ∩ ℓC ,

and bij , cij similarly.

(All of this notation is for avoiding the config-issues in this theorem. It turns out that only 18 of the 27
triangles you can get are actually equilateral!)

Theorem 1.6.1 (Morley’s Trisector Theorem). For any (i, j, k) ∈ {−1, 0, 1}3 such that i + j + k ̸≡ 2 (mod 3),
△ajk bki cij is a equilateral triangle. Additionally, (bki cij , BC) and (ajk B, ajk C) are antiparallel and so forth
cyclically.

Proof. We construct △ABC from the equilateral triangle △ajk bki cij .

Let α = ∡(ℓB0 C0
A , ℓA ) and define β, γ similarly.

We then have that

X
θ := α + β + γ = (i + j + k)◦ + ∡(ℓB0 C0 ◦ ◦
A , ℓA ) = (1 + i + j + k) = ±60 .
cyc

and thus θ = ±60◦ (mod 360◦ ). Then we let △abc be an equilateral triangle so ∡bac = ∡cab = ∡acb = θ.
We then define A0 such that

∡cA0 b = α, ∡bcA0 = θ + β, ∡A0 bc = θ + γ,

which is possible because α + (θ + β) + θ + γ = 3θ = 180◦ . Define B0 , C0 similarly.

Let b′ , c′ be the reflections of b, c over C0 a, B0 a respectively. Then we get that ab′ = ab = ac = ac′ and
that △ab′ c′ is isosceles. Notice that 2∡C0 ab, 2∡caB0 are well defined (mod 360◦ ), so then we let

∡c′ a′ b′ = 360◦ − 2∡C0 ab − ∡bac − 2∡caB0 = −2(θ + β) − θ − 2(θ + γ) = 2α − θ (mod 360◦ )

so since 3θ/2 = 90◦ (mod 180◦ ), we have that

1
∡b′ c′ a = 90◦ − (2α − θ) = (−θ + α) = −∡B0 ca = ∡B0 c′ a (mod 180◦ )
2

72
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

so B0 ∈ b′ c′ . Similarly, we get C0 lies on b′ c′ . As such, we get that

∡C0 B0 a = ∡c′ B0 a = ∡aB0 c = β, ∡B0 C0 a = γ.

+
And thus, we get that △A0 B0 C0 ∪ △abc ∼ △ABC ∪ ajk bki cij , so △ajk bki cij ∼
= △abc is equilateral.

Now we angle chase, and we get that

∡bki cij = ∡ajk C + ∡bki ajk C + ∡cij bki ajk

= ajk − (θ + β) + θ

= ajk C + ∡ajk BC = ajk B + ajk C − BC

as desired.

We can prove a similar proposition.

Proposition 1.6.2 (Morley Variant). Using the same notation as above, for (i, j) ∈ {−1, 0, 1}2 ,

△aij a(i+1),(j−1) a(i−1)(j+1)

is equilateral and B, C lies on its circumcircle. Furthermore, (a(i+1)(j−1) a(i−1)(j+1) , BC) and (aij B, aij C)
are antiparallel.

Proof. For conciseness we use i±1 = i ± 1, j ±1 = j ± 1.

Then we have that

∡Baij C = ℓBj Ci B0 C0
C − ℓB = (ℓC − ℓB ) + (i + j) · 60
◦

so this only depends on i + j which is fixed for (i, j), (i+ , j − ), (i− , j + ) which implies that B, C lie on the
circumcircle.

Next, note that

+ −
◦
∡ai+ aij ai− j + = ∡ai+ j − Cai− j + = ∡(ℓBi Bi
C , ℓC ) = 60 ,

and the same holds for the other two angles, giving an equilateral triangle.

Finally, we have that

∡ai+ j − ai− j + = ai+ j − B + ai− ,j + C − BC = aij B + aij C − BC.

73
AoPS Chapter 1. Geometric Transformations - Hidden/Completed

The 18 triangles specified in Morley’s Trisector Theorem and in Morley Variant give us the 27 Morley
triangles (though in certain contexts only the first 18 count, and most normal people only know 1 Morley
triangle). In fact, we only need to find three Morley triangles due to the following proposition.

Proposition 1.6.3. For (i, j, k) ∈ {−1, 0, 1}3 such that 3 | 1 + i + j + k, the six points

a(j+1),(k−1) , a(j−1),(k+1) , b(k+1),(i−1) , b(k−1),(i+1) , c(i+1),(j−1) , c(i−1),(j+1)

are collinear.

Proof. Reuse the i± , j ± , k ± from above. By Morley triangles, we have that

△aj − k+ bk+ i− ci− j − , △ajk aj + k− aj − k+

are both equilateral triangles and

bk+ i− ci− j − + BC = aj − k+ B + aj − k+ C, aj + k− aj − k+ + BC = ajk B + ajk C

so we get that

∡aj + k− aj − k+ bk+ i− = ∡ci− j − bk+ i− aj − k+ + (ajk B + ajk C) − (aj − k+ B + aj − k+ C)

= (1 + i− + j − + k + ) · 60◦ + ∡ajk Baj − k+ + ∡ajk Caj − k+

= −60◦ + (j − − j) · 60◦ + (k − k + ) · 60◦ = 0◦

which gives that aj + k− , aj − k+ , bk+ i− are collinear. Similar angle chases gives us all 6 points.

We now call △aii bii cii for i = 0, 1, −1 the first, second, and third Morley triangle in that order.

We will revisit this theorem when we talk about cardioids later in (TODO 10.4).

74
Chapter 2

The Cross Ratio - Hidden/Completed

In the previous chapter we have defined lengths and angles, and now through some simple calculations we
will discover that it is easier to work with something called the cross-ratio. For this section, we will work in
the real projective plane, but for this chapter it’ll be good enough to imagine it as the “real plane + one line
at infinity”.

The concept of points/lines “at infinity” might be familiar to you if you have done three-point perspective
drawing before. If you have not, first think of a flat grid of lines in 2d space. Then think about ”moving
the camera” upwards, in three-dimensional space. Then, all of the previously parallel lines will appear to
concur at some points inf, which all lie on a common horizon. For part I of this book, we will call this camera
movement a projective transformation, and we call this horizon the line at infinity, L∞ . (We will completely
rigorously define projective space later on.)

The cross ratio is our main tool to work with projective transformations: as we will prove later, it is
invariant under them.

2.1 Definition

Definition 2.1.1. For four collinear points P1 , P2 , P3 , P4 ,

P1 P3 /P3 P2
(P• ) := (P1 , P2 ; P3 , P4 ) :=
P1 P4 /P4 P2

is called the cross ratio of the bundle of points P1 , P2 , P3 , P4 .

(Note: a much more beautiful way to think of this is “if you take a projective transformation sending P1
to 0, P2 to 1, P4 to infinity, where does P3 go?”)

75
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

This definition is actually a bit problematic, because we have not dealt with the case Pi ∈ L∞ , but for
any two points A, B ∈
/ L∞ , we can define
A∞AB
= −1.
∞AB B

Under this definition, it remains to handle the case where P1 , P2 , P3 , P4 are all on the line of infinity, which
we will come back to later. First let’s look at the most important basic result for cross ratios:

Theorem 2.1.2. Given four concurrent lines ℓ1 , ℓ2 , ℓ3 , ℓ4 , if a line L ̸= ℓ1 , ℓ2 , ℓ3 , ℓ4 not at infinity intersects
T
ℓi at Pi respectively, then (P• ) is fixed (does not depend on the choice of L). If ℓi ∈ / L∞ , then

sin ∡(ℓ1 , ℓ3 )/ sin ∡(ℓ3 , ℓ2 )

(P• ) = .
sin ∡(ℓ1 , ℓ4 )/ sin ∡(ℓ4 , ℓ2 )

(Note that the choice of the direction of the lines does not affect its value.)

Proof. Let L′ ̸∋ ℓi be another line which intersects ℓi at Pi′ respectively. If

T T
ℓi ∈ L∞ , then we clearly have

P1 P3 P ′ P ′ P1 P 4 P ′P ′
= 1′ 3′ , = 1′ 4′ =⇒ (P• ) = (P•′ ).
P3 P2 P3 P2 P4 P 2 P4 P2
T
If A := ℓi ∈
/ L∞ , and without loss of generality assume P1 , P2 , P3 ∈
/ L∞ , then by (TODO 0.3.4) we have

sin ∡P1 AP3 P1 P3 /P3 P2 sin ∡P1 AP4 P1 P4 /P4 P2

= , =
sin ∡P3 AP2 AP1 /AP2 sin ∡P4 AP2 AP1 /AP2

(note that the above still holds if P4 ∈ L∞ ). Combining the two equations we get

sin ∡P1 AP3 / sin ∡P3 AP2

(P• ) = .
sin ∡P1 AP4 / sin ∡P4 AP2

which does not depend on the choice of L.

So we can define the cross ratio of a bundle of lines:

Definition 2.1.3. For four concurrent lines ℓ1 , ℓ2 , ℓ3 , ℓ4 ,

(ℓ• ) := (L ∩ ℓ• )

T
(where L ̸∋ ℓi is a line not at infinity) is called the cross ratio of the bundle of lines ℓ1 , ℓ2 , ℓ3 , ℓ4 .

And the cross ratio of points on the line of infinity:

Definition 2.1.4. For four collinear points P1 , P2 , P3 , P4 ∈ L∞ ,

sin ∡P1 AP3 / sin ∡P3 AP2

(P• ) := ,
sin ∡P1 AP4 / sin ∡P4 AP2
76
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

where A ∈
/ L∞ is a point.

From this definition, Theorem 2.1.2 still holds when L = L∞ .

Example 2.1.5. Let a pencil of lines ℓ1 , ℓ2 , ℓ3 , ℓ4 intersect line K at A1 , A2 , A3 , A4 and intersect line L at
B1 , B2 , B3 , B4 . Suppose
A1 A2 = A2 A3 = A3 A4 = 1, B1 B2 = 2, B2 B3 = 3.

Calculate B3 B4 .

Solution. Let B3 B4 = x. From the cross ratio property (A• ) = (B• ), we have

2/(−1) 5/(−3)
= .
3/(−2) (5 + x)/(−3 − x)

So we just need to solve the linear equation 5(3 + x) = 4(5 + x) and we obtain x = 5.

A1 A2 A3 A4

B1
B2
B3
B4

See how this calculation is very easy, but without cross ratios you might need to bash with Menelaus.

From the definition it is easy to see that cross ratios are preserved under any spiral similarity or homothety
ϕ, in other words (ϕ(P• )) = (P• ). Having completely defined the cross ratio, here are some properties of cross
ratios you must know:

Proposition 2.1.6. For four collinear points P1 , P2 , P3 , P4 , if (P• ) = λ, then

(i) (P2 , P1 ; P3 , P4 ) = (P1 , P2 ; P3 , P4 ) = λ−1

(ii) (P1 , P3 ; P2 , P4 ) = 1 − λ.

The proof is easy:

77
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

P2 P3 /P3 P1

 −1 
 = (P2 , P1 ; P3 , P4 ),
P1 P3 /P3 P2  P2 P4 /P4 P1
λ−1 = =
P1 P4 /P4 P2 P P /P P
 1 4 4 2 = (P1 , P2 ; P4 , P3 ),


P1 P3 /P3 P2
P1 P3 /P3 P2 P1 P2 /P2 P3
λ + (P1 , P3 ; P2 , P4 ) = +
P1 P4 /P4 P2 P1 P4 /P4 P3
P3 P1 · P 4 P2 + P1 P 2 · P4 P3
= = 1,
P2 P3 · P1 P4

where the last equality is the special case of Ptolemy’s theorem (TODO 0.4.18) where the four points lie on a
straight line. (The directed version is equivalent to

a(b − c) + b(c − a) + c(a − b) = 0.)

Of course, this is also true for cross ratios of lines instead of points. Actually, from this property we can
represent all (Pσ(•) ) in terms of (P• ) where σ ∈ S4 is any permutation of {1, 2, 3, 4}. In other words, we have
(Pσ(•) ) = ρ(σ)(P• ), where ρ : S4 → Aut(P1 ) ⊂ R(λ) is defined as

(1 2), (3 4) 7→ [λ 7→ λ−1 ], (2 3) 7→ [λ 7→ 1 − λ].

Next up is the most important property of cross ratios, which is the lifeforce of all projective geometers:

Proposition 2.1.7 (Converse Principle).

(i) For collinear points P1 , P2 , P3 , P4 , P♡ , we have that P♡ = Pk if and only if (P• ) = (P•′ ),

(ii) For concurrent lines ℓ1 , ℓ2 , ℓ3 , ℓ4 , ℓ♡ , we have that ℓ♡ = ℓk if and only if (ℓ• ) = (ℓ•′ ),

where 
i, if i ̸= k
i′ =
♡, if i = k.


Note that (ii) follows from (i) and (i) can be proved from the definitions. We will do the case k = 4:

P1 P3 /P3 P2 P1 P3 /P3 P2 P1 P4 P1 P♡
(P• ) = (P•′ ) ⇐⇒ = ⇐⇒ = ⇐⇒ P4 = P♡ .
P1 P4 /P4 P2 P1 P♡ /P♡ P2 P4 P2 P♡ P2

Now let’s define some convenient notation:

Definition 2.1.8.

78
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

(i) For any five points P1 , P2 , P3 , P4 , A,

A(P• ) := A(P1 , P2 ; P3 , P4 ) := (AP1 , AP2 ; AP3 , AP4 ).

(ii) For any five lines ℓ1 , ℓ2 , ℓ3 , ℓ4 , L,

L(ℓ• ) := L(ℓ1 , ℓ2 ; ℓ3 , ℓ4 ) := (L ∩ ℓ1 , L ∩ ℓ2 ; L ∩ ℓ3 , L ∩ ℓ4 ).

This allows to express the following fact simply, which is a useful tool in proving collinearity.

Proposition 2.1.9. Given collinear points P1 , P2 , P3 ∈ ℓ and points A, B not on ℓ, then for any point P4
not on AB, we have A(P• ) = B(P• ) if and only if P4 ∈ ℓ.

P4′
P1 P3 P2

Proof. The “if” direction is easy. For the other direction, let P•′ = P• for i = 1, 2, 3 and P4′ = AP4 ∩ ℓ, then

B(P•′ ) = (P•′ ) = A(P•′ ) = A(P• ) = B(P• ),

hence P4′ ∈ BP4 , which implies P4′ = AP4 ∩ BP4 = P4 , so P4 ∈ ℓ.

S
Proposition 2.1.10 (Prism Lemma). For points P1 , P2 , P3 and points A, B ∈
/ Pi Pj ,

A(P1 , P2 ; P3 , B) = B(P1 , P2 ; P3 , A)

if and only P1 , P2 , P3 are collinear.

79
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

P1 P3 P2 P4

Proof. Once again, the “if” direction is easy. For the other direction, let P4 = AB ∩ P2 P3 , then P2 , P3 , P4
are collinear and

A(P1 , P2 ; P3 , P4 ) = A(P1 , P2 ; P3 , B) = B(P1 , P2 ; P3 , A) = B(P1 , P2 ; P3 , P4 ),

so P1 ∈ P2 P3 .

Of course, similar results hold when we swap points with lines. The proofs are similar and we will omit
them.

Proposition 2.1.11. Given lines ℓ1 , ℓ2 , ℓ3 concurrent at P and lines K, L not passing through P , then for
any line ℓ4 not passing through K ∩ L, we have K(ℓ• ) = L(ℓ• ) if and only if P ∈ ℓ4 .

Proposition 2.1.12. For points ℓ1 , ℓ2 , ℓ3 and lines K, L not passing through ℓi ∩ ℓj for any i ̸= j,

K(ℓ1 , ℓ2 ; ℓ3 , L) = L(ℓ1 , ℓ2 ; ℓ3 , K)

if and only if ℓ1 , ℓ2 , ℓ3 are concurrent.

Next, we are going to define cross ratio on a circle.

Proposition 2.1.13. For points P1 , P2 , P3 , P4 lying on circle Ω, if A ∈ Ω is different from all the Pi ’s, then
A(P• ) does not depend on the choice of A.

80
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

A
P1

P2

P4

A′
P3

Proof. Let A′ ∈ Ω be another point different from all the Pi ’s, then from circle properties we get ∡Pi APj =
∡Pi A′ Pj . So from Theorem 2.1.2 we have A(P• ) = A′ (P• ), so A(P• ) is fixed.

Remark. For the result above, if we allow A = Pi and define AA as the tangent TA Ω to Ω at A, then the
statement remains true. We will adopt this notation for the conic chapters as well.

This allows us to define the cross ratio of four points on a circle:

Definition 2.1.14. For points P1 , P2 , P3 , P4 lying on circle Ω,

(P• ) := (P1 , P2 ; P3 , P4 ) := A(P• ),

where A ∈ Ω.

Remark. If we view the (projective) plane as P1C , the complex projective line, rather than P2R , then the
formula
(z3 − z1 )/(z2 − z3 )
(z1 , z2 ; z3 , z4 ) :=
(z4 − z1 )/(z2 − z4 )

works for both concyclic and collinear points z1 , z2 , z3 , z4 . In particular, the cross ratio is real if and only if
they are either concyclic or collinear. (We will further discuss this in later chapters on the structure of CP2 .)

Example 2.1.15. Let lines BC and EF be parallel and D be a point on the interior of segment BC. BF
and CE intersect at I. It is known that K = (CDE), L = (BDF ) are tangent to EF at E and F respectively.
Let K and L intersect again at A ̸= D, DF and K intersect again at Q ̸= D, DE and L intersect again at
R ̸= D. EQ and F R intersect at M . Prove that I, A, M are collinear.

Solution. By Proposition 2.1.10, it suffices to show that

E(I, A; M, F ) = F (I, A; M, E).

81
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Noting that
EF
E(I, A; M, F ) = (C, A; Q, E) = D(C, A; Q, E) = (∞EF , AD ∩ EF ; F, E).

If we let P = AD ∩ EF , then P lies on the radical axis of K and L, so

P E 2 = PowK (P ) = PowL (P ) = P F 2 ,

so P is the midpoint of EF . This gives us

E(I, A; M, F ) = (∞EF , P ; F, E) = −1.

Similarly, we have F (I, A; M, E) = −1, so I, A, M are collinear.

B D C

R
Q

A
F P E

We would also like to define the cross ratio of four tangents to a circle (similarly to four points on a circle).
Let Γ be a circle, and define

(i) TΓ as the set of all tangents of Γ;

(ii) TP Γ as the tangent line to Γ at P , where P ∈ Γ;

(iii) Tℓ Γ as the point of tangency of ℓ and Γ, where ℓ ∈ TΓ;

(iv) In the degenerate case where Γ is a point P , TP is the set of lines through P .

To define cross ratio on TΓ, we will need:

82
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Proposition 2.1.16. Given lines ℓ1 , ℓ2 , ℓ3 , ℓ4 ∈ TΓ, then for any line L ∈ TΓ different from the ℓi ’s, L(ℓ• ) is
fixed (does not depend on the choice of L).

Proof. Let O be the center of Γ, then from O(L ∩ ℓi ) ⊥ (TL Γ)(Tℓi Γ) we get

L(ℓ• ) = TL Γ(Tℓ• Γ) = (Tℓ• Γ),

which does not depend on the choice of L, hence proved.

So we can define:

Definition 2.1.17. For lines ℓ1 , ℓ2 , ℓ3 , ℓ4 ∈ TΓ,

(ℓ• ) := (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) := L(ℓ• ),

where L is a tangent to Γ.

Remark. Similarly to the point case, if we allow L = ℓi and define L ∩ L as the tangency point TL Γ, then
the statement remains true. We will adopt this notation for the conic chapters as well.

For an application of cross ratio on a circle, let’s prove the butterfly theorem:

Theorem 2.1.18 (Butterfly Theorem). Let AB be a chord of circle Γ and M be the midpoint of AB.
Construct chords P1 P2 , Q1 Q2 through M where P1 , P2 , Q1 , Q2 ∈ Γ. Let R1 = P1 Q1 ∩ AB, R2 = P2 Q2 ∩ AB.
Then M is the midpoint of R1 R2 .

Proof. We have
P Q2
(A, B; M, R1 ) =1 (A, B; P2 , Q1 ) = (A, B; R2 , M )

so we get
AM/M B AR2 /R2 B QR1 BR2 AB BA
= =⇒ = =⇒ = ,
AR1 /R1 B AM/M B R1 B R2 A R1 B R2 A
−−→ −−→
so R1 B = AR2 . Since M is the midpoint of AB, we get that it is also the midpoint of R1 R2 .

The butterfly theorem has nice generalizations which have to do with involutions (see Section (TODO
(7.2))). Here is an example problem using the butterfly theorem:

Example 2.1.19. Let I, O be the incenter and circumcenter of △ABC respectively. The line through I
perpendicular to OI meets the external angle bisector of ∠BAC and BC at P and Q respectively. Prove
that IP = 2QI.

83
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Solution. Let the line through I perpendicular to OI intersect (ABC) at U and V , then I is the midpoint
of chord U V . Let Nb and Nc be the second intersections of BI and CI with (ABC), then by the butterfly
theorem, R = Nb Nc ∩ U V is the reflection of Q across I.

By the Incenter-Excenter Lemma, Nb , Nc both lie on the perpendicular bisector of AI, so Nb Nc is the
perpendicular bisector of AI. Noting that the external angle bisector of ∠BAC is the image of the perpedicular
bisector of AI under homothety at I with factor 2, so P is the image of R under homothety at I with factor
2. Hence IP = 2IR = 2QI.

Practice Problems

Problem 1. Let △ABC be an isosceles triangle with apex A. points P , Q satisfy

∡ABP = ∡ACQ and ∡P CA = ∡QBC.

Prove that A, P , Q are collinear.

Problem 2. Let P be a point inside △ABC, and BP , CP meet CA, AB at E, F respectively. Let EF
intersect (ABC) at B ′ , C ′ , and B ′ P , C ′ P meet BC at C ′′ , B ′′ respectively. Prove that B ′ B ′′ , C ′ C ′′ , (ABC)
concur.

Problem 3. Let △ABC be acute with altitudes AH1 , BH2 and angle bisectors AL1 , BL2 . If O and I are
the circumcenter and incenter of △ABC respectively. Prove that O lies on L1 L2 if and only if I lies on H1 H2 .

Problem 4 (Cross Ratio Equality). Let P1 , P2 , P3 , P4 be collinear points and Q1 , Q2 , Q3 , Q4 be collinear

points (on different lines). Let Ri = Pi Qi+1 ∩ Pi+1 Qi . Prove that R1 , R2 , R3 are collinear if and only if
(P• ) = (Q• ).

Problem 5. On line ℓ lies points P1 , P2 , P3 , P4 , P4′ . Prove that it is possible to construct, using ruler only,
points Q and R on ℓ such that

(P1 , P2 ; P3 , Q) = (P1 , P2 ; P3 , P4 ) + (P1 , P2 ; P3 , P4′ )

(P1 , P2 ; P3 , R) = (P1 , P2 ; P3 , P4 ) · (P1 , P2 ; P3 , P4′ ).

Problem 6. Prove the following generalization of Problem 3: Let (P, P ∗ ), (Q, Q∗ ) be pairs of isogonal
conjugate points in △ABC, and △Pa Pb Pc , △Qa Qb Qc be the Ceva triangles of P , Q with respect to △ABC.
Prove that P ∗ lies on Qb Qc if and only if Q∗ lies on Pb Pc .

Problem 7 (Taiwan 2014/2J/I2-1). Let I and O be the incenter and circumcenter of △ABC respectively.
Let L be the tangent to the incircle parallel to BC. L meets IO at X, and Y is on L such that Y I ⊥ IO.
Prove that A, X, O, Y are concyclic.

84
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Problem 8 (ISL 1996 G3). Let O and H be the circumcenter and orthocenter of acute △ABC respectively.
F is the foot of the altitude CH of △ABC. The line through F perpendicular to OF intersects AC at P .
Prove that ∠F HP = ∠BAC.

Problem 9 (Taiwan 2021/2J/P3). Let O and H be the circumcenter and orthocenter of scalene △ABC
respectively. P is a point inside △AHO such that ∠AHP = ∠P OA. M is the midpoint of OP . Let BM and
CM intersect (ABC) again at X, Y respectively. Prove that XY passes through the circumcenter of △AP O.

2.2 Harmonic Bundles

Definition 2.2.1. We say four points P1 , P2 , Q1 , Q2 are harmonic iff.

(P1 , P2 ; Q1 , Q2 ) = −1,

and we call these four points a harmonic bundle. (note that the lengths in the definition are signed).
Similarly, we define four lines K1 , K2 , L1 , L2 to be harmonic iff.

(K1 , K2 ; L1 , L2 ) = −1,

and we call these four lines a harmonic pencil.

Notably, by Proposition 2.1.6, harmonic bundles/pencils are fixed under swapping P1 , P2 , swapping Q1 , Q2 ,
and swapping P, Q. As such, these four points can be split into an unordered set {P1 , P2 }, {Q1 , Q2 }. As for
why it’s called harmonic bundles, note that

−−−→ −−−→
P1 Q1 P1 Q2
(P1 , P2 ; Q1 , Q2 ) = −1 ⇐⇒ −−−→ + −−−→ = 0
Q1 P2 Q2 P2
−−−→ −−−→
P1 P2 P1 P2
⇐⇒ −−−→ + −−−→ = 2
Q1 P2 Q2 P2
1 1 2
⇐⇒ −−−→ + −−−→ = −−−→ ,
P2 Q1 P2 Q2 P 2 P1

−−−→ −−−→ −−−→

so P2 P1 is the harmonic mean of P2 Q1 and P2 Q2 !

Example 2.2.2. Here are some easy examples of harmonic bundles.

• If M is the midpoint of AB, then A, B, M, ∞AB make a harmonic bundle since

−−→ −−→
AM /M B 1
(A, B; M, ∞) = −−→ −−→ = = −1,
A∞/∞B −1

85
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

and thus the midpoint of AB is ∞∨

AB .

• If in △ABC, ∠BAC’s internal and external bisectors intersect BC at D and D′ , then

BD/DC AB/AC
(B, C; D, D′ ) = ′ ′
= = −1.
BD /D C −AB/AC

• In △ABC, the circumcenter O, the nine-point center N , the centroid G, and the orthocenter H make
a harmonic bundle.
HN/N G 3
(H, G; N, O) = = = −1.
HO/OG −3
Example 2.2.3. Let G and O be the centroid and circumcenter of △ABC. Let the perpendicular bisectors
of GA, GB, GC pairwise intersect at A1 , B1 , C1 . Prove that O is the centroid of A1 B1 C1 .

Solution. By symmetry, we only need to prove that A1 O is the median of B1 C1 . So we just need to prove
that
(OA1 , O∞B1 C1 ; OB1 , OC1 ) = (OA1 ∩ B1 C1 , ∞B1 C1 ; B1 , C1 ) = −1.

Then note that A1 , B1 , C1 are the circumcenters of △GBC, △GCA, △GAB, so we have that OA1 ⊥
BC, OB1 ⊥ CA, OC1 ⊥ AB. So from B1 C1 ⊥ AG, we have

(OA1 , O∞B1 C1 ; OB1 , OC1 ) = (∞OA1 , ∞B1 C1 ; ∞OB1 , ∞OC1 )

=t (∞BC , ∞AG ; ∞CA , ∞AB )

= (A∞BC , AG; AC, AB) = −1,

since AG bisects BC. Here, t is a 90◦ rotation on L∞ . (This is a common trick.).

A B1

C1
O
G

B C

A1

86
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Here’s another harmonic condition.

Proposition 2.2.4. Given four distinct points P1 , P2 , Q1 , Q2 on a line and a point A not on that line, then
any of the following two conditions implies the rest of the two conditions.

(i) (P1 , P2 ; Q1 , Q2 ) = −1.

(ii) ∡P1 AP2 = 90◦

(iii) AP1 bisects ∡Q1 AQ2 (either internally or externally)

(iv) AP2 bisects ∡Q1 AQ2 .

P1 O Q1 P2 Q2

Proof. By the second bullet point of Example 2.2.2 and angle chasing we already have most of these. Here’s
the proof of the remaining few by phantom points.

• (i) and one of {(iii), (iv)} implies the other follows by converse principle. If we let AP2′ be the angle
bisector of ∠Q1 AQ2 that AP1 isn’t, then

(P1 , P2′ ; Q1 , Q2 ) = −1 = (P1 , P2 ; Q1 ; Q2 )

so AP2 = AP2′ .

• (i), (ii) implies (iii), (iv): take Q′2 so AP1 bisects ∠Q1 AQ2 . By (ii) we have that AP2 bisects ∡Q1 AQ2 ,
so it follows that
(P1 , P2 ; Q1 , Q′2 ) = −1 = (P1 , P2 ; Q1 , Q2 ),

and thus Q′2 = Q2 .

87
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Now we characterize a bunch of length-conditions as harmonics.

Proposition 2.2.5. Given four collinear points P1 , P2 , Q1 , Q2 , and let M be the midpoint of P1 P2 . Then
the following are equivalent:

• P1 , P2 , Q1 , Q2 are harmonic;

• M Q1 · M Q2 = M P12 = M P22 ;

• P1 P2 · Q1 Q2 = 2 · P1 Q1 · P2 Q2 ;

• P1 P2 · Q2 Q1 = 2 · P1 Q2 · P2 Q1 ;

• Q1 M · Q1 Q2 = Q1 P1 · Q1 P2 ;

• Q2 M · Q2 Q1 = Q2 P1 · Q2 P2 .

Proof. Clearly, you can just use a bunch of algebra to “explode” every statement by setting P1 = −1, P2 =
1, Q1 = s, Q2 = t. However let’s do the length-bash sensibly so that we learn some length characterizations
along the way.

First we prove that (i) ⇐⇒ (iii), (iv). From Proposition 2.1.6, we have

P1 P2 /P2 Q2
1 − (P1 , P2 ; Q1 , Q2 )−1 = 1 − (P1 , P2 ; Q2 , Q1 ) = (P1 , Q2 ; P2 , Q1 ) =
P1 Q1 /Q1 Q2

P1 P2 /P2 Q1
1 − (P1 , P2 ; Q1 , Q2 ) = (P1 , Q1 ; P2 , Q2 ) =
P1 Q2 /Q2 P1

Thus (P1 , P2 ; Q1 , Q2 ) = −1 iff

P1 P2 /P2 Q2
= 2 ⇐⇒ P1 P2 · Q1 Q2 = 2 · P1 Q1 · P2 Q2 ,
P1 Q1 /Q1 Q2

that is (iii), is also equivalent to

P1 P2 /P2 Q1
= 2 ⇐⇒ P1 P2 · Q2 Q1 = 2 · P1 Q2 · P2 Q1 ,
P1 Q2 /Q2 Q1

so we get (iv) as well.

We now prove (i) =⇒ (ii), (v), (vi), geometrically: from the fact that P1 Q1 /Q1 P2 , P1 Q2 /Q2 P2 have
opposite signs, we get that there has to be one of Q1 , Q2 that lies on segment P1 P2 . So we can just assume
WLOG that Q1 ∈ P1 P2 . Draw a perpendicular to P1 P2 through Q1 and let this intersect the circle ω = (P1 P2 )
at A, B. From Proposition 2.2.4, we have that AP1 , AP2 bisect ∠Q1 AQ2 , therefore we have

∡P2 AQ2 = ∡Q1 AP2 = 90◦ − ∡AP2 Q = ∡P2 P1 A,

88
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

so AQ2 is tangent to ω at A. Thus we also have by symmetry that M A is tangent to (AQ1 Q2 ), and that
AQ2 is tangent to (M AQ1 ). As such, we have that

∡(M A + Q1 Q2 , AQ1 + AQ2 ) = 90◦ + 90◦ = 0◦

∡(AQ2 + Q1 Q2 , M A + AQ1 ) = 90◦ + 90◦ = 0◦

Thus we have that

M Q1 · M Q2 = M A2 = M P1 2 = M P2 2 , Q2 M · Q2 Q1 = Q2 A2 = Q2 P1 · Q2 P2 .

Finally, our construction of B isn’t completely useless: since ∡M AQ2 = ∡M BQ2 = 90◦ , M, Q2 , A, B are
concyclic. Thus we have Q1 M · Q1 Q2 = Q1 A · Q1 B = Q1 P1 · Q1 P2 .

Example 2.2.6 (Finland 2019/3). Cyclic quadrilateral ABCD has AB as its diameter. Let AC and BD
intersect at E, and let AD and BC intersect at F . Let EF intersect the circumcircle of ABCD at G. Extend
EF to intersect AB at H. Suppose that F G = GH. Prove that GE = EH also.

Solution. We consider the reflection of G across AB, call this point G′ . Then we know that G′ lies on the
circle with diameter AB. Thus we have

HE · HF = −HA · HB = −HG · HG′ = HG2 = HG′2 ,

so E, F, G, G′ are harmonic. Thus

F G = GH =⇒ 3F G = F G′ =⇒ 3GE = EG′ =⇒ GE = EH.

Example 2.2.7 (Bulgaria 2016/5). Isosceles triangle △ABC satisfies AC = BC, and let D be on ray AC
such that C is between A, D and AC > CD. Let the angle bisector of ∠BCD intersect BD at N . Let
the midpoint of BD be M . Draw the tangent to circle (AM D) at M , and let it intersect BC at P . Prove
A, P, M, N concyclic.

Solution. This suspicious tangency, and perpendicular and angle bisectors, motivates us to find harmonic
bundles. We consider the reflection of D across AB, point D′ . Since B, C, D′ are collinear, we have A, B, D, D′
are concyclic. Thus

89
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

∡AD′ P = ∡ADM = ∡AM P,

and A, D′ , M, P are concyclic. Thus we only need to prove that A, D′ , M, N are concyclic. Let E be the
intersection of AD′ and BD, then E lies on the perpendicular bisector of AB, so it also lies on the external
angle bisector of ∠BCD. So B, D, N, E is a harmonic bundle! So by the above length conditions, we have

EN · EM = EB · ED = EA · ED′ ,

so we have that A, D′ , M, N are concyclic by PoP.

Proposition 2.2.8. Given a complete quadrilateral Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) and one diagonal Aij Akl , the two other
diagonals Aik Alj , Ail Ajk intersect the other diagonal in a harmonic bundle with its endpoints. (Aij = ℓi ∩ ℓj ,
this notation will be used more in Chapter 4.)

Proof. Let P = Aij Akl ∩ Aik Alj , and Q = Aij Akl ∩ Ail Ajk . Because P, Aik , Alj collinear, Menelaus gives

−−−→ −−−−→ −−−−→

Aij P Akl Alj Ail Aik
−−−→ · −−−−→ · −−−−→ = −1
P Akl Alj Ail Aik Aij

From Ail Q, Akl Alj , Akl Aik concurrent at Ajk , we can use Ceva to get

Aij Q Akl Alj Ail Aik

· = 1.
QAkl Alj Ail Aik Aij

Dividing these two equations we can get

Aij P/P Akl

(Aij , Akl ; P, Q) = = −1,
Aij Q/QAkl

so Aij , P, Akl , Q are harmonic.

(A cleaner proof of this will be provided with the concept of DIT in later chapters; a faster proof with
polarity is that a complete quadrilateral with these diagonals drawn is self-polar.)

Example 2.2.9. Let AHa be the height from A in △ABC. Let P be a point on AHa . Let E = BP ∩ CA,
let F = CP ∩ AB, let D = EF ∩ AP . Draw a line ℓ through D that intersects P E, AF at X, Y . Prove that
AHa is the angle bisector of ∠XHa Y .

Solution. Since AHa ⊥ BC, by Proposition 2.2.4 we can see that AHa is the angle bisector of ∠XHa Y if
and only if
(X, Y ; D, ℓ ∩ BC) = −1.

90
AoPS Chapter 2. The Cross Ratio - Hidden/Completed
B
By projecting, we see (X, Y ; D, ℓ ∩ BC) = (E, F ; D, EF ∩ BC),

However then just use the previous lemma on the complete quadrilateral (CA, AB, BE, CF ) to get that

(E, F ; D, EF ∩ BC) = (E, F ; EF ∩ AP, EF ∩ BC) = −1.

Definition 2.2.10. We now define harmonic quadrilaterals. A cyclic quadrilateral A, B, C, D is harmonic iff.

(A, B; C, D) = −1.

Proposition 2.2.11. A quadrilateral inscribed in Γ, (P1 P2 )(Q1 Q2 ) is a harmonic iff the intersection of the
tangents TP1 Γ, and TP2 Γ, Q1 , and Q2 are collinear.

Proof. Let A = P1 P2 ∩ Q1 Q2 , B1 = TP1 Γ ∩ Q1 Q2 , B2 = TP2 Γ ∩ Q1 Q2 . Then,

(Q1 , Q2 ; A, B1 ) = P1 (Q1 , Q2 ; P2 , P1 ) = (Q1 , Q2 ; P2 P1 ),

(Q1 , Q2 ; A, B2 ) = P2 (Q1 , Q2 ; P2 , P1 ) = (Q1 , Q2 ; P1 P2 ).

Therefore, the intersection of the tangents TP1 Γ, and TP2 Γ, Q1 , and Q2 are collinear is equivalent to B1 = B2
that is equivalent to
(Q1 , Q2 ; P2 , P1 ) = (Q1 , Q2 ; P1 , P2 ) = (Q1 , Q2 ; P2 , P1 )−1

which implies that the cross ratio is indeed −1, because (Q2 , Q1 ; P2 , P1 ) ̸= 1 as P2 =
̸ P1 , and Q2 ̸= Q1 . This
implies that (P1 P2 )(Q1 Q2 ) is harmonic.

Example 2.2.12 (Bosnia Grade 9 2018/5). Let H be the orthocenter of △ABC, and let the foot from H to
the internal angle bisector of ∠BAC be D. Let the foot from H to the external angle bisector of ∠BAC be
E. Let the midpoint of BC be M . Prove M, D, E are collinear.

Proof. We consider the circle (AH). Let the feet of the altitudes from B to CA and C to AB be Hb and Hc .
Then we have that M is the circumcenter of (BCHb Hc ). We also have

∡M Hb A = ∡M Hb C = ∡ACB = ∡Hb HA,

and thus M Hb is tangent to (AH). Similarily we have M Hc tangent to (AH). Since AD and AE are the
angle bisectors of ∠BAC, we have

91
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

−1 = (AB, AC : AD, AE) = (Hc , Hb ; D, E),

so by the above characterization we know that DE goes through Hb M ∩ Hc M = M .

Example 2.2.13 (2013 APMO P5). Let ABCD be a quadrilateral inscribed in circle ω, and let P be a
point on the extension of AC such that P B and P D aret tangent to ω. The tangent at C intersects P D at
Q and the line AD at R. Let E be the second point of intersection between AQ and ω. Prove that B, E, R
are collinear.

Proof. Because P B and P D are tangent to ω, ABCD is harmonic. We also get that ADCE is harmonic
because of the way it is constructed (AE intersects the tangents at C and D.) By the definitions of harmonic
quadrilaterals we get that (A, C; B, D) = −1. Because E lies on ω proving that (EA, EC; ER, ED) = −1
solves as it means we can project point B to point R through point E, or that they are collinear.

E(A, C; R, D) = (Q, C; R, DE ∩ CQ) = D(Q, C; R, E) = (D, C; A, E) = −1

Thus we are done.

Another commonly used idea is:

Proposition 2.2.14. Let P1 P2 Q1 Q2 be a harmonic quadrilateral, and let M be the midpoint of Q1 Q2 . Then
we have that P1 M, P1 P2 are isogonal lines in ∡Q1 P1 Q2 . (We call P1 P2 the P1 -symmedian of △Q1 P1 Q2 .)

Proof. Let P1′ be the reflection of P1 across the perpendicular bisector of Q1 Q2 . Then using the fact that
P1 Q1 Q2 P1′ is a isosceles trapezoid, we get

−1 = (P1 , P2 ; Q1 , Q2 ) = (P1′ P1 , P1′ P2 ; P1′ Q1 , P1′ Q2 ) = (∞Q1 Q2 , P1′ P2 ∩ Q1 Q2 ; Q1 , Q2 ),

so P1′ P2 passes through the midpoint of Q1 Q2 , point M . Since

∡(P1 M + P1 P2 , P1 Q1 + P1 Q2 ) = ∡M P1 Q1 + ∡P2 P1′ Q2

= ∡M P1 Q1 + ∡M P1′ Q2 = 0◦ ,

thus P1 M, P1 P2 are isogonal lines in ∡Q1 P1 Q2 .

Example 2.2.15. Let ABCD be a harmonic quadrilateral. Then the midpoint of AC and BD are isogonal
in ACBD.

92
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Example 2.2.16. Let Ω be the circumcircle of △ABC and let TA , TB , TC be the tangents to Ω at A, B, C,
and let D, E, F be TB ∩ TC , TC ∩ TA , TA ∩ TB respectively. Let X be the other intersection of AD and Ω. Then
(AX)(BC) is a harmonic quadrilateral. Hence ∡AD = ∡AX is the isogonal (reflection over the angle bisector)
of the median ∡AMa with respect to ∠BAC. Similarly, BE, CF are the isogonals of BMb , CMc with respect
to ∠CBA, ∠ACB. Hence AD, BE, CF is the isogonal conjugate of the centroid G = AMa ∩ BMb ∩ CM ac
of △ABC. We call this point K the symmedian point of △ABC.

Example 2.2.17. In acute triangle △ABC, let M be the midpoint of BC. Let P be a point inside △ABC
that satisfies ∡BAM = ∡P AC. Let the circumcenters of △ABC, △ABP, △AP C be O, O1 , O2 respectively.
Prove that AO bisects O1 O2 .

Solution. Let D be the intersection point of AP and (ABC). Then ABDC is harmonic. Since O1 O2 , O2 O, OO1
are perpendicular to AP = AD, CA, AB, we can get

(OA, O∞O1 O2 , OO1 , OO2 ) = (∞AO , ∞O1 O2 ; ∞OO1 , ∞O2 O )

= (∞⊥AO , ∞AD ; ∞AB , ∞AC )

= (AA, AD, AB, AC) = −1,

where the second equality is due to rotation on L∞ by 90◦ . Thus OA bisects O1 O2 .

Practice Problems

Problem 1 (09 Costa Rica Final round P6). Let ∆ABC with incircle Γ, let D, E and F the tangency points
of Γ with sides BC, AC and AB, respectively, and let P the intersection point of AD with Γ. Prove that,
BC, EF and the straight line tangent to Γ for P concur at a point A′ . Now, define B ′ and C ′ in an analogous
way than A′ . Prove that A′ , B ′ , C ′ are collinear.

Problem 2. Let △ABC with incircle ω tangent to CA, and AB at E, F respectively. Let P = BC ∩ EF .
Draw the line parallel to BC that is tangent to ω, and let this intersect CA and AB at Y, Z. Prove that the
tangent from P to ω that is not BC bisects segment Y Z.

Problem 3. Let △ABC with circumcircle Γ, and K be the symmedian point. Prove that for any point X
on Γ, the trilinear polar of X, t(X) pass through K.

(This is proven in Chapter 5).

Problem 4. Let Ω be the A-excircle of △ABC. Let this excircle touch BC, CA, AB at D, E, F . Pick two
points P, Q on Ω such that EP and F Q are both parallel to the line connecting D and the midpoint of

93
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

segment EF . Let X be the intersection point of BP and CQ. Prove that AM is the internal angle bisector
of ∠XAD.

Problem 5 (2018 Taiwan P6). Let P be a point inside △ABC that satisfies ∠BAC + ∠BP C = 180◦ and
AB PB
AC = PC . Prove that
∠AP B − ∠ACB = ∠AP C − ∠ABC.

Problem 6 (Taiwan TST 1 Day 3 P2). Given a triangle △ABC, A′ , B ′ , C ′ are the midpoints of BC, AC, AB,
respectively. B ∗ , C ∗ lie in AC, AB, respectively, such that BB ∗ , CC ∗ are the altitudes of the triangle ABC.
Let B # , C # be the midpoints of BB ∗ , CC ∗ , respectively. B ′ B # and C ′ C # meet at K, and AK and BC
meet at L. Prove that ∠BAL = ∠CAA′

Problem 7 (2019 Taiwan P4). Let Γ be a circle, and choose A outside of Γ, and draw the two tangents to Γ
from A. Let these tangents touch Γ at B and C. Choose D on Γ such that BC = BD. Connect AD, let it
intersect Γ again at E. Prove DE = 2 · CE.

Problem 8. Let a △ABC with incircle ω tangent to sides BC, AC, AB at points D, E, F , let X be any
point on ω with XB, XC intersecting ω again at Y , Z. Prove that DX, EY , F Z concur.

Problem 9 (APMOC 2011/1). Let the incircle of △ABC touch BC, CA, AB at P, Q, R. Let O, I be the
circumcenter and incenter respectively, and suppose the orthocenter H of △ABC lies on QR. Let N be the
A-extouch point on BC. Prove

• P H ⊥ QR.

• I, O, N are collinear.

Problem 10. Let △ABC be rectangle at A with circumcircle Γ. Then line tangent to Γ through A intersects
BC at D, and let E be the reflection of A with respect to BC, and X is the projection of A to BC. The
midpoint of AX is Y and the second intersection of BY with Γ is Z. Prove that BD is tangent to the
circumcircle of ADZ.

Problem 11 (2015 Day 2 P4). Let a △ABC with incircle ω and circumcircle Γ. Let D be the tangency
point of BC and ω, and let M be the midpoint of ID and A′ be the antipode of A in Γ. Let X be the other
intersection of A′ M with Γ. Prove that the circumcircle of △AXD is tangent to BC

Problem 12 (Buffed 2014 ISL G6). Let acute triangle △ABC have points E and F on CA and AB. Let M
be the midpoint of EF . Let the perpendicular bisector of EF intersect line BC at K. Let the perpendicular
bisector of M K intersect CA, AB at S, T . Suppose KSAT is concyclic. Prove ∠KEF = ∠KF E = ∠A.

94
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

2.3 A Few Projective Theorems

The first theorem is Desargues’s theorem (TODO 1.1.1), which we have already looked at:

Theorem 2.3.1 (Desargues). Given two triangles △A1 B1 C1 , △A2 B2 C2 , P = B1 C1 ∩ B2 C2 , Q = C1 A1 ∩

C2 A2 , R = A1 B1 ∩ A2 B2 lie on a line iff A1 A2 , B1 B2 , C1 C2 are concurrent.

Proof. Let X = B1 B2 ∩ C1 C2 , Z1 = A1 A2 ∩ B1 C1 , Z2 = A1 A2 ∩ B2 C2 Notice that X ∈ A1 A2 ⇐⇒ X ∈ Z1 Z2

holds if and only if
X
(P, C1 ; B1 , Z1 ) = (P, C2 ; B2 , Z2 ).

By Proposition 2.1.10 P, Q, R collinear if and only if

A1 (P, Q; R, A2 ) = A2 (P, Q; R, A1 );

(Note that Ai ∈
/ QR ∪ RP ∪ P Q). Thus,

A1 (P, Q; R, A2 ) = (P, C1 ; B1 , Z1 ), A2 (P, Q; R, A2 ) = (P, C2 ; B2 , Z2 ).

Thus the 2 conditions are equivalent.

Theorem 2.3.2 (Pappus). Let P1 , P2 , P3 , and Q1 , Q2 , Q3 lie on lines K and L respectively. Then

P2 Q3 ∩ P3 Q2 , P3 Q1 ∩ P1 Q3 , P1 Q2 ∩ P2 Q1

are collinear.

P3
P2
P1

R2 R1
R3

Q1
Q2 Q3
U

95
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Proof. Let Ri = Pi+1 Qi−1 ∩ Pi−1 Qi+1 , U = P1 R1 ∩ K, V = Q1 R1 ∩ L, thus

R
P1 (R1 , R2 ; R3 , Q1 ) = (U, Q3 ; Q2 , Q1 ) =1 (P1 , P2 ; P3 , V )

= (V, P3 ; P2 , P1 ) = Q1 (R1 , R2 ; R3 , P1 ).

Thus by Proposition 2.1.10 R1 , R2 , R3 are collinear.

Theorem 2.3.3 (Pascal’s Theorem). Let the points P1 , P2 , P3 , P4 , P5 , P6 lie on a circle. Then

P1 P2 ∩ P4 P5 , P2 P3 ∩ P5 P6 , P3 P4 ∩ P6 P1

are collinear.

P5
P1

P3

Q3 Q2
Q1

P6
P4

P2

Proof. Let Q1 = P1 P2 ∩ P4 P5 , Q2 = P2 P3 ∩ P5 P6 , Q3 = P3 P4 ∩ P6 P1 . Thus

Q1 (P1 , P3 ; P4 , Q3 ) = (P1 Q1 ∩ P3 P4 , P3 ; P4 , Q3 )

= P1 (P2 , P3 ; P4 , P6 ) = P5 (P2 , P3 ; P4 , P6 )

= (P2 , P3 ; P4 Q1 ∩ P2 P3 , Q2 )

= Q1 (P1 , P3 ; P4 , Q2 )

Thus Q1 , Q2 , Q3 are collinear.

Example 2.3.4 (Belarus TST 2019/6/P1). Two circles Ω, ω are tangent at A. Let BC be a chord of Ω
that is tangent to ω at L. Let AB, BC intersect ω at M, N respectively. Reflect M, N about AL to obtain

96
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

M1 , N1 , and reflect M, N about BC to obtain M2 , N2 . Let K = M1 M2 ∩ N1 N2 . Prove that AK ⊥ BC.

Proof. Let l be the line through A tangent to both circles. Then

∡BAL = ∡(BA, l) + ∡(l, AL) = ∡BCA + ∡ALB = ∡LAC

That is, AL is bisects ∠BAC. Thus, M1 , N1 lie on CA, AB respectively and L is the midpoint of arc
M
¯ N (not passing through A). Notice that M, N, M1 , N1 are concyclic, call this circle Γ. The center of this
circle is the intersection of the perpendicular bisectors M N and M M1 and AL, or L. Because

LM2 = LM = LN = LN2

We find that M2 , N2 also lie on Γ. Now we have that the 6 points M N1 N2 N M1 M2 are concyclic on Γ.
By Pascal’s Theorem, we get that A, K, ∞⊥BC = N2 N ∩ M2 M are collinear. Thus, AK ⊥ BC.

Note the “duality” between the following theorem and Pascal’s:

Theorem 2.3.5 (Brianchon’s Theorem). Let l1 , l2 , l3 , l4 , l5 , l6 are all tangent to a common circle. Then

(l1 ∩ l2 )(l4 ∩ l5 ), (l2 ∩ l3 )(l5 ∩ l6 ), (l3 ∩ l4 )(l6 ∩ l1 )

are concurrent.

Proof. We present a proof similar to Pascal’s Theorem. Let L1 = (l1 ∩ l2 )(l4 ∩ l5 ), L2 = (l2 ∩ l3 )(l5 ∩ l6 ), L3 =
(l3 ∩ l4 )(l6 ∩ l1 ). Then

L1 (l1 , l3 ; l4 , L3 ) = ((l1 ∩ L1 )(l3 ∩ l4 ), l3 ; l4 , L3 )

= l1 (l2 , l3 ; l4 , l6 ) = l5 (l2 , l3 ; l4 , l6 )

= (l2 , l3 ; (l4 ∩ L1 )(l2 ∩ l3 ), L2 )

= L1 (l1 , l3 ; l4 , L2 )

Consequently, L1 , L2 , L3 are concurrent.

97
AoPS Chapter 2. The Cross Ratio - Hidden/Completed

Remark. The two theorems above, Pascal’s Theorem and Brianchon’s Theorem, actually allow for 2 points
(or lines) to coincide. In this case, the line joining the 2 points (or the intersection of the 2 lines) becomes a
tangent (point) to the line, and vice versa.

Example 2.3.6 (APMO 2016/3). Let AB and AC be two distinct rays not lying on the same line, and let ω
be a circle with center O that is tangent to ray AC at E and ray AB at F . Let R be a point on segment EF .
The line through O parallel to EF intersects line AB at P . Let N be the intersection of lines P R and AC
and let M be the intersection of line AB and the line through R parallel to AC. Prove that M N is tangent
to ω

Proof. Let U be the point at infinity along the line AC. Let L be the tangent to ω from M which isn’t
AB. Let L intersect AC at N ′ . Note that P U is tangent to ω because P U ∥ AC. Consider the hexagon
U EN ′ M F P with incircle ω. By Brianchon’s Theorem, U M, EF, N ′ P concur. Since R = U M ∩ EF it follows
that N ′ , R, P are collinear. Thus N = N ′ and M N is tangent to ω

Practice Problems

Problem 1. Let ABC be a triangle with circumcenter O, and let P, Q be points on CA and AB respectively
such that P, O, Q collinear. Let M, N be the midpoints of BP, CQ respectively. Prove that ∠BAC = ∠M ON .

Problem 2. Let ω be the incircle of △ABC and let D, E, F be the intouch points. Let K be the second
intersection of AD with ω. Let the line tangent to ω at K intersect F D, DE at Y, Z respectively. Prove that
AD, BZ and CY concur.

Problem 3. Let ABC be the triangle with incircle ω and l be a tangent to ω. Let A′ , B ′ , C ′ be points other
than the intouch points lying on BC, CA, AB respectively. The tangent from A′ (other than BC) intersects l
at A∗ , B ∗ and C ∗ are defined similarly. Prove that AA∗ , BB ∗ , CC ∗ are concurrent.

98
Chapter 3

Inversion and Polarity -

Hidden/Completed

3.1 Basic Inversion

In previous sections we looked at transformations that preserved similarities. Now let’s look at a transformation
that doesn’t even preserve shapes.

Remark. Like what we did implicitly in (TODO 1.2), we will also work in CP1 , the complex projective line
or the inversive plane. Imagine this as the complex plane, represented by z := a + bi, and exactly one point
at infinity P∞ . (This is called a line because it is defined by exactly one complex number, so it ends up being
dimension 1 over the complex numbers.)

Definition 3.1.1. Given a point O and a constant k ̸= 0 (potentially negative), we can define an inversion
JO,k at O (the center of inversion) with power k as such:

• If P isn’t O or P∞ , then P J = JO,k (P ) is the point on OP such that such that

−−→ −−→
OP · OP J = k.

• O swaps with P∞ , under this transformation.

Notably, JO,k = sO ◦ JO,−k . When k < 0, this is called a negative inversion.

It is more natural to phrase inversion with circles. For a circle Γ with center O and radius r, we can
define the inversion JΓ as the inversion centered at O with power k = r2 . (This is also why r is called the
power of the inversion, as for P ∈ Γ, PowΓ (P ) = r2 .)

99
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Note that JΓ is an involution, that is JΓ (JΓ (P )) = P. Furthermore, Γ is fixed under JΓ .

In this section, by default the notation P J will refer to the image of P under the inversion J which is
known by context. We define the image of a line ℓ under inversion as ℓJ = {P J | P ∈ ℓ}.

Note for any two points P, Q, that aren’t O or P∞ , we have that

−−→ −−→ −−→ −−→

OP · OP J = k = OQ · OQJ ,

so P, Q, P J , QJ are concyclic by power of a point.

Proposition 3.1.2. Given any two points P, Q ̸= O, ∞, we have that

|k|
∡OP J QJ = ∡P QO, P J QJ = · P Q.
OP · OQ

In line arguments, we can rephrase the first formula as ∡P J QJ + ∡P Q = ∡OP + ∡OQ.

(The formula for P J QJ is commonly known as the inversion distance formula.)

Proof. Since P, Q, P J , QJ are concyclic, we get that

∡OP J QJ = ∡P P J QJ = ∡P QQJ = ∡P QO.

−
This gives us the similarity △OP Q ∼ △OQJ P J , therefore

OQJ k
P J QJ = · PQ = · P Q.
OP OP · OQ

Note that JO,k1 = hO, k1 ◦ JO,k2 , so if we don’t overlay the pre-inversion and post-inversion diagrams, it
k2

doesn’t matter what power we choose, as the inverted diagram will be the same up to scaling. In these cases,
we will say “invert about P ” which means an inversion with arbitrary power.

We will now characterize what inversion does to lines and circles.

Proposition 3.1.3. Given an inversion centered at O, we have that:

(i) If ℓ is a line passing through O, then ℓJ = ℓ;

(ii) If ℓ is a line not passing through O, then ℓJ is a circle passing through O;

(iii) If Ω is a circle passing through O, then ΩJ is a straight line not passing through O;

100
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

(iv) If Ω is a circle not passing through O, then ΩJ is a circle not passing through O.

Proof. Part (i) is obvious, since O, P, P J are collinear. For (ii), given any two points P , Q on ℓ, we have

∡OQJ P J = ∡QP O = ∡(ℓ, OP )

. Varying Q on ℓ or QJ on (OP J QJ ) gives that ℓJ = (OP J QJ ).. Part (iii) follows by the above logic in
reverse.

Finally, for part (iv), note that for any three points P, Q, R ∈ Ω, we have

∡QJ P J RJ = ∡QJ P J O + ∡OP J RJ = ∡OQP + ∡P RO = ∡QOR + ∡RP Q

is fixed as P varies, so ΩJ ⊆ (P J QJ RJ ).

Since the image of O is P∞ , and Ω does not pass through the point at infinity, its image will not pass
through O, either.

Remark. Note that inversion does not preserve circle centers. This is a very common mistake.

Given △ABC, consider the involution IA resulting from first applying an inversion JA with radius
√
AB · AC followed by a reflection about the angle bisector of ∠BAC. This swaps B and C, AB and AC,
√
and BC and (ABC). We will use the term bc inversion with respect to △ABC as this involution. We will
also use JA to just refer to the inversion part.

(The original book uses “Reflect-Invert about A”, but this is more clear and more widely used.)

To get a handle as to why this is useful, let’s look at some examples:

Example 3.1.4 (Croatia MO 2010/7). Given △ABC, let B ′ be the reflection of B about AC and let C ′
be the reflection of C about AB. Let (ABB ′ ) ∩ (ACC ′ ) = P ̸= A. Prove that AP passes through the
circumcenter O of △ABC.

√
Solution. We bc-invert the problem. Let us first find where everything goes. Note that AB ′ = AB, AC ′ =
AC and that AB ′ , AC ′ are isogonal, so we have the following paiings:

AB ↔ AC, B′ ↔ C ′, (ABB ′ ) ↔ CC ′ ,

(ACC ′ ) ↔ BB ′ , P ↔ P IA := BB ′ ∩ CC ′ .

As such, showing that O ∈ AP ′ is thus equivalent to showing that the orthocenter H lies on AP IA . However,
P IA = BB ′ ∩ CC ′ is just the orthocenter.

101
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Proposition 3.1.5. Given △ABC, if (P, Q) are isogonal conjugates with respect to △ABC, then IA (P )
√
and IA (Q) are also isogonal conjugates with respect to △ABC after a bc inversion.

+ +
Proof. Using the fact that △ABP ∼ △AIA (P )C, △ABQ ∼ △AIA (Q)C, we get that

∡(CIA (P ) + CIA (Q), CA + CB) = ∡BP A + ∡BQA + ∡ACB = 0◦

where the last equality follows from (TODO 1.3.6). As such, CIA (P ), CIA (Q) are isogonal. Since
BIA (P ), BIA (Q) are also isogonal by symmetry, the result follows.

Example 3.1.6. In △ABC, let I a be the A-excenter, let J be the reflection of I a about BC, let H be the
orthocenter of △BI a C, and let O be the circumcenter of △BI a C. Prove that AJ ∥ OH.

Solution. Since

∡BA + ∡BJ = (2∡BI a − ∡BC) + (2∡BC − ∡BI a ) = ∡BI a + ∡BC

∡CA + ∡CJ = (2∡CI a − ∡BC) + (2∡BC − ∡CI a ) = ∡CI a + ∡BC

√
we have that (A, J) are isogonal conjugates in △BI a C. Now, the bc inversion at I a wrt. △BI a C sends O
to J. However since (O, H) are isogonal conjugates in △BI a C, we get that this also maps H to A.

Thus, by ∡I a AJ = −∡HOI a = ∡I a OH, we get AJ ∥ OH.

As a matter of fact, inversion also preserves “angles” in a sense, or is a conformal mapping:

Definition 3.1.7. Let γ1 and γ2 be two smooth curves (in our case, this will be circles or lines for the
foreseeable future). Let P be an intersection of γ1 and γ2 . Then the angle between γ1 and γ2 at P is the
angle between the tangent line TP (γ1 ) and TP (γ2 ), and its value is ∡(TP (γ1 ), TP (γ2 )) = ∡P (γ1 , γ2 ).

If this angle is 90◦ , then the two curves are said to be orthogonal which is a symmetric relation.

Proposition 3.1.8. Let γ1 , γ2 be two smooth curves with an intersection point at P . Then the angle
between γ1 and γ2 at P is equal and opposite to the angle between γ1J and γ2J at P J .

Proof. Let ℓ1 , ℓ2 be the tangents at P to γ1 , γ2 respectively. Because γiJ and ℓJi are tangent at P J , we can
just consider ℓJ1 , ℓJ2 . By (TODO 3.1.3), we get that ℓJi are both circles through P and the center of inversion
O.

As such, both ℓJ1 and ℓJ2 are symmetric about the perpendicular bisector of OP J , so

∡P J (ℓJ2 , ℓJ1 ) = ∡O (ℓJ1 , ℓJ2 ).

102
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Let Li be the tangent at O to ℓJi . By (TODO 3.1.3), LJi = Li . Since O ∈ ℓJi ∩ Li it follows that P∞ ∈ ℓi ∩ Li ,
so ℓi ∥ Li . As such,
∡O (ℓJ1 , ℓJ2 = ∡(L1 , L2 ) = ∡(ℓ1 , ℓ2 ) = ∡P (γ1 , γ2 )

Example 3.1.9 (Buffed IMO 2015/3). Let △ABC be an acute triangle with AB < AC, and let Γ be the
circumcircle and let H be the orthocenter as usual. Let F be the foot from A. Let M be the midpoint of BC,
and let Q be a point on Γ that satisfies ∠HQA = 90◦ . Let K be the point on Γ that satisfies ∠HKQ = 90◦ .
It is given that the points A, Q, K, B, C lie on Γ in that order.

Prove that the circumcircle of △KQH is tangent to the circumcircle of △F KM .

(Note: Point Q is known as the Queue point in English.)

K H

B F M C

√
Proof. Consider the negative inversion J with center H and radius HA · HF . This sends the vertices of
△ABC to their feet on the opposite side, so Γ = ϵ is the nine-point circle of △ABC. Hence QJ lies on ε
J

with ∡QJ F H = 90◦ , so QJ is the midpoint M of BC.

Note that K J ∈ ϵ satisfies ∡K J M H = 90◦ , and (KQH) being tangent to (F KM ) is equivalent to K J M

being tangent to (AK J Q). Let MA and MQ be the midpoints of HA and HQ, respectively, so that both
midpoints lie on ϵ. Since M MA is a diameter of ε, we have

∡MA MQ = ∡AQ = ∡ ⊥ QM = ∡M K J =⇒ (MQ − K J )Γ = (M − MA )Γ ,

so K J MQ is also a diameter of ε.

103
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Because K J MA ⊥ MA MQ ∥ AQ, MA lies on the median AQ, so K J also lies on this median. Therefore

∡QK J M = 90◦ − ∡M QK J = ∡K J QA = ∡QAK J ,

which implies the conclusion.

Note that if Γ1 , Γ2 intersect in two points P, Q, then

∡P (Γ1 , Γ2 ) = −∡Q (Γ1 , Γ2 ).

So if two circles are orthogonal at P , then they are also orthogonal at Q. In this case, we call Γ1 orthogonal
to Γ2 . Further, (TODO 3.1.8) tells us that orthogonality is preserved under inversion.

Proposition 3.1.10. Let Γ1 , Γ2 be two circles with O1 , O2 as centers and r1 , r2 as radii. Then the following
are equivalent:

• Γ1 and Γ2 are orthogonal;

• O1 O2 2 = r12 + r22 ;

• PowΓ2 (O1 ) = r12 .

Proof. Pythagorean Theorem on △O1 P O2 where P is one of their intersection points.

Corollary 3.1.11. If we define J = JΓ , then for any point P and circle Ω ∋ P , Ω is orthogonal to Γ if and
only if P J ∈ Ω.

Proof. Let Γ, Ω have centers O1 , O2 and radii r1 , r2 . By power of a point, we have that P J ∈ Ω iff
r12 = OP · OP J = O1 O22 − r22 so the result follow by the above.

This gives us a key property of orthogonal circles: inverting one around the other maps it to itself.

Let us go back to the concept of “negative inversion”. In this case, we don’t have actual real points for
calculating angles however, by (TODO 3.1.10) we can still define orthogonality. Consider a circle E centered
√
at O with k as its “radius”. Set J = JE .

• When k > 0, E is a normal, real circle.

• When k = 0, E is the point circle we encountered in (TODO 0.4).

• When k < 0, E is a circle with imaginary radius.

104
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Remark. To rigorously define circles with imaginary radius, we need to go back to the complex projective
plane CP2 , as the only way to give CP1 a two-dimensional structure is by representing it as (a, b), a, b ∈ R.
We will rigorously interpret this in later chapters, don’t worry about it right now.
√ √
Definition 3.1.12. Let E1 , E2 be circles centered at O1 , O2 with radius k1 , k2 respectively. We say that
Γ1 , Γ2 are orthogonal if
O1 O22 = k1 + k2 .

In this setting, we can replace Γ in (TODO 3.1.11) with any circle E as follows:

Proposition 3.1.13. If J = JE , then for any point P and a circle Ω ∋ P , Ω and E are orthogonal if and only
if P J ∈ Ω.

We can also define a radical axis as follows

√ √
Definition 3.1.14. Let E1 , E2 be circles centered at O1 , O2 with radius k1 , k2 respectively. We define
the radical axis of E1 , E2 to be
{P | PowE1 (P ) = PowE2 (P ).}

(Power remains defined for circles with imaginary radii).

By a similar proof as (TODO 0.4.9), the radical axis of E1 and E2 is a straight line perpendicular to
O1 O2 . Then we can derive the Radical Axis theorem for any circles. Now, we return to our discussion of
orthogonality with the following theorem:

Proposition 3.1.15. Given two circles E1 and E2 , any circle E that is orthogonal to E1 and E2 has its center
O on the radical axis of E1 and E2 . Conversely, for any point O on the radical axis of E1 and E2 , there exists
exactly one circle E centered at O and orthogonal to E1 and E2 .

√ √
Proof. Let O1 , O2 respectively be the centers of E1 and E2 , and let k1 and k2 respectively be the radii of
E1 and E2 .
√
Let E be a circle orthogonal to both of these circles, and let k be the radius of this circle. Then by the
orthogonality length condition we get that

PowE1 (O) − PowE2 (O) = k − k = 0,

so O ∈ ℓ.

To prove the converse, let O be an arbitrary point on ℓ, let k = PowE1 (O) = PowE2 (O), then we have
√
the circle with radius k is orthogonal to both E1 and E2 by the same logic as before.

105
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Proposition 3.1.16. If the centers of two inversions J1 and J2 do not coincide, then their composition
J2 ◦ J1 is an inversion J coupled with a reflection about some line s. In fact, if we set J1 = JE1 = JO1 ,k1 and
J2 = JE2 = JO2 ,k2 , then J has center O = O2J1 and s is the reflection about the radical axis ℓ of E1 and E2 .

Proof. We show that J := s ◦ J2 ◦ J1 is an inversion with center O.

For a point P , let P1 = P J1 , P2 = P1J2 , P3 = s(P2 ). By (TODO 3.1.13), we have that (P P1 P2 ) is

orthogonal with E1 E2 , so its center lies on ℓ. As such, it follows that P3 also lies on (P P1 P2 ), and that
P2 P3 ⊥ ℓ ⊥ O1 O2 . Furthermore, P, P1 , O, O2 are cyclic (because P1 = J1 (P ), O = J1 (O2 )). As such, we get
that
∡P1 P O = ∡P1 O2 O = ∡P1 P2 P3 = ∡P1 P P3

so O, P, P3 are collinear (This is also Reim’s (TODO 0.1.14)).

We now find OP · OP2′ . Let O3 = s(O2 ), so we then have that

∡O1 P P3′ = ∡P1 P O = ∡P1 O2 O = ∡P2 O2 O3 = ∡O2 O3 P3 = ∡O1 O3 P3

so O1 , O3 , P, P3 are concyclic. As such, OP · OP3 = OO1 · OO3 is fixed, which finishes.

Proposition 3.1.17. Given an inversion J centered at O, for any circle Ω, the circle ΩJ has center J(JΩ (O)).

Proof. Let OΩ be the center of Ω, let OΩJ be the center of ΩJ . Let A, B be the two intersection points of OOΩ
with Ω. Then since AB is orthogonal to Ω we get that AJ B J is orthogonal to ΩJ . Therefore ΩJ = (AJ B J ).
From
OAJ OB
=
OB J OA
OAJ
we can get that ΩJ , Ω are homothetic with scale factor OB . Thus

OB OOΩ
OJΩ (O) · OOΩ = PowΩ (O) = OA · OB = k · =k· ,
OAJ OOΩJ

which gets us that OJΩ (O) · OOΩJ = k, and thus J(JΩ (O)) = OΩJ .

Proposition 3.1.18. Given an inversion J centered at O with radius k, for any circle Ω not centered at O,
ΩJ has radius
rk
PowΩ (O)

where r is the radius of Ω.

Proof. Let A, B be the intersection points of OOΩ and Ω. Then the radius of ΩJ is

106
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

1 1 1 k k
· AJ B J = · OB J − OAJ = · −
2 2 2 OB OA
|k| OA − OB rk
= · = .
2 OA · OB PowΩ (O)

Practice Problems

Problem 1. In △ABC with ∡A = 90◦ , let D be a point on side AB. Two circles are tangent to BC at B
and C and intersect at D and E. Prove that ∠CBA = ∠DEA.

Problem 2 (Taiwan TST 2015/3J/I3-2). In a scalene triangle ABC with incenter I, the incircle is tangent
to sides CA and AB at points E and F . The tangents to the circumcircle of triangle AEF at E and F
meet at S. Lines EF and BC intersect at T . Prove that the circle with diameter ST is orthogonal to the
nine-point circle of triangle BIC.

Problem 3 (Taiwan TST 2019/1J/M6). Given △ABC, denote its incenter and orthocenter by I and H,
respectively. If there is a point K with

AH + AK = BH + BK = CH + CK,

show that H, I, and K are collinear.

3.2 Basic Polarity

Polarity is a way to exchange the roles of points and lines. This is the explanation for “duality” we saw in
previous sections.

In this section, we are back to working in CP2 (So imagine a line at infinity again.).

Definition 3.2.1. Take a circle E and let O be its center.

• For an arbitrary point P ̸= O, P ∈

/ L∞ , we define the polar pE (P ) of P wrt. E as the image of the
circle (OP ) under JE . (The polar of O is the line at infinity L∞ .)

• For P on the line at infinity, the polar of P is the line O∞⊥P .

107
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

From this definition we can see that pE (P ) always passes through JE (P ) and is always perpendicular to
OP (every line is perpendicular to the line at infinity, we will elaborate on this later).

When P lies outside E, also note that the polar of P is the line through the touch-points of the two
tangent lines to E from P. (The tangency points lie on both E and (OP ) and are fixed under inversion).

Let M be the midpoint of P and P J . for P ∈

/ L. We then have that

PowE (M ) = OM 2 − OP · OP J

= (OP 2 + 2 · OP · P M + P M 2 ) − OP · OP J = P M 2 = PowP (M ),

and thus M lies on the radical axis of E and the point circle P .

Proposition 3.2.2. For a point P ∈

/ L∞ and a circle E, the polar of P across E is the image of the radical
axis of P, E under a homothety of scale factor 2 from P .
√
We can redefine polarity in a nicer way: if the radius of E is k, then

−−→ −−→
pE (P ) = {Q | OP · OQ = k} ∪ {∞⊥OP }.

From this, we get the most important principle of polarity.

Proposition 3.2.3 (La Hire’s Theorem). Given two points P and Q, and a circle E, we have that P ∈
pE (Q) ⇐⇒ Q ∈ pE (P ).

Proof. Let P, Q not be O or L∞ which can be checked manually. Then we have that

−−→ −−→
P ∈ pE (Q) ⇐⇒ OP · OQ = k ⇐⇒ Q ∈ pE (P ).

If P = O or P ∈ L∞ , then Q lies on L∞ or ⊥ OP respectively, and the result follows by definitions.

This is very useful.

Definition 3.2.4. Let E be a circle, and let P and Q be two points such that P lies on the polar of Q (by
La Hire’s, we also have vice-versa). Then we call P, Q conjugates in E.

Now, suppose that three points P, P1 , P2 are collinear, then the intersection Q of pE (P1 ) and pE (P1 ),
satisfies that pE (Q) = P1 P2 . Since P ∈ P1 P2 , it follows that Q ∈ pE (P ). Varying P , this tells us that for a
line K and a point P ∈ K, we have that pE (P ) passes through a fixed point regardless of P .

Definition 3.2.5. Let E be a circle and K be a line. For a varying point P ∈ K, define the fixed point that
pE (P ) passes through to be the pole of line K wrt. E.

108
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

From this definition, we get the following relations:

pE (P ) ∩ pE (Q) = pE (P Q), pE (K)pE (L) = pE (K ∩ L).

Additionally, we get that the function pE is an involution (we have pE (pE (K)) = K.) By (TODO 3.2.3)
we can get that:

Proposition 3.2.6 (Dual of La Hire’s). Let E be a circle and let K, L be two lines in the plane. Then

pE (K) ∈ L ⇐⇒ pE (L) ∈ K.

Definition 3.2.7. Let E be a circle, we call two lines K, L conjugate wrt. E if pE (K) ∈ L, and vice versa.

Remark. This is the “duality” we have been alluding to in previous chapters!! Given a configuration of
points and lines, we can take the polar dual of the entire diagram, which swaps lines to points, and swaps
concurrencies into collinearities. We call this kind of transformation incidence-preserving. For example,
the polar dual of Ceva’s theorem is Menelaus’s theorem, and the polar dual of Desargues’s theorem is the
converse of Desargues’s theorem.

Example 3.2.8. Let non-equilateral triangle △ABC have intouch triangle △DEF . Let X be the intersection
point of EF and BC, and let P be the second intersection point of AD with the incircle. Prove that XP is
tangent to the incircle.

Solution. Note that X = EF ∩ BC = pω (A) ∩ pω (D), so

P ∈ AD = pω (X) =⇒ X ∈ pω (P ).

Since P lies on the incircle, we know P ∈ pω (P ) and that pω (P ) is the tangent to the incircle at P . Thus
XP is tangent to the incircle.

Example 3.2.9 (Taiwan 2014/1J/P3). Let △ABC have incenter I, and let the incircle touch CA, AB at
E, F . Let the reflections of E, F across I be G, H. Let Q be the intersection of GH and BC. Let M be the
midpoint of BC. Prove IQ ⊥ IM .

Solution. Let the incircle be ω. Let A′ be the reflection of A across I. Since AE and AF are tangent to ω,
we know that A′ G and A′ H are tangent to ω, so GH = pω (A′ ). If we let D be the intouch point on BC,
then BC = pω (D). Thus Q = pω (A′ ) ∩ pω (D) = pω (A′ D). So we have IQ ⊥ A′ D, so all we need to prove to
finish the problem is that A′ D ∥ IM .

109
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

If we let D∗ be the antipode of D in ω, and let D′ be the reflection of D across M (the A-extouch
point), then we can get that A′ D ∥ AD∗ and IM ∥ D∗ D′ (Since I, M are the midpoints of DD∗ and DD′
respectively). Therefore we only need to prove that A, D∗ , D′ are collinear, but this is obvious by homothety
at A sending the incircle to the excircle.

Proposition 3.2.10. Given a circle E, then two points P and Q are conjugate if and only if the circle with
diameter P Q is orthogonal to E.

Proof. By (TODO 3.1.13), we get that (P Q) is orthogonal to E iff. P J ∈ (P Q). But this just means that
P J Q ⊥ OP, so Q ∈ pE (P ).

Proposition 3.2.11. Given a circle E, and two conjugate points P, Q such that P Q intersects E (at real
points) A, B, then
(P, Q; A, B) = −1.

Proof. If P and Q are conjugates in E, then if we let M be the midpoint of AB, then from ∡OM Q = 90◦ =
∡OP J Q we know that O, P J , Q, M are concyclic. Thus

P A · P B = PowE (P ) = P P J · P O = P Q · P M,

so (P, Q; A, B) = −1, by fractions.

Definition 3.2.12. Let E be a circle such that △ABC has A, B, C pairwise conjugate in E (which is the
same thing as BC, CA, AB pairwise conjugate). Then we call △ABC a self-conjugate triangle wrt. E.

√
Let the center of E be O and let E have radius k. Then we have OA ⊥ BC, OB ⊥ CA, OC ⊥ AB, so O
is the orthocenter of △ABC and

−→ −−→ −−→ −−→ −−→ −−→

OA · OD = OB · OE = OC · OF = k,

where △DEF is the orthic triangle of △ABC.

Switching our frame of reference, given any triangle △ABC, there exists a circle E centered at H (possibly
with imaginary radius) such that △ABC is self-conjugate. We call this circle the polar circle of △ABC.

Remark. If H lies inside △ABC, then this circle has imaginary radius, as k is negative. You can just
√
think of polarity across this circle as the composition of polarity across the real circle with radius −k and
reflecting about H.

We now look at a new way to find self-conjugate triangles.

110
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Proposition 3.2.13 (Brokard’s Theorem). Let E be a circle and let P1 , P2 , P3 , P4 be four points on this
circle. Set
X = P2 P3 ∩ P1 P4 , Y = P3 P1 ∩ P2 P4 , Z = P 1 P2 ∩ P3 P 4 ,

then △XY Z is self-conjugate wrt. E.

Proof. Let Y Z intersect P2 P3 and P1 P4 at R and S. Then by (TODO 2.2.8) we have that

(X, R; P2 , P3 ) = (X, S; P1 , P4 ) = −1.

and from (TODO 3.2.11) we know that R, S ∈ pE (X), so pE (X) = RS = Y Z.

A quick corollary of this is that the orthocenter of △XY Z is the center of E, which is very useful.

Example 3.2.14. Let cyclic quadrilateral ABCD be inscribed in Ω. Let the diagonals AC and BD intersect
at E, let AB and CD intersect at F , let AD and BC intersect at G, and let M be the midpoint of F G. Let
segment EM intersect Ω at T . Prove that (F GT ) is tangent to Ω.

Solution. Note that if AC and EM coincide, we win immediately: By symmetry, WLOG let T = C. Then
from M being the midpoint of F G we can use (TODO 2.2.8) (or Menelaus) to get that BD ∥ F G. Thus from

∡(TC (F GC), TC Ω) = ∡(TC (F GC), CG) + ∡(BC, TC Ω)

= ∡CF G + ∡CDB = 0◦

we get that (F GT ) is tangent to Ω.

Thus it remains to show that we can pick points A, B, C, D on Ωfor fixed E, F, G that satisfy the above
statement. By (TODO 3.2.13), △EF G is self-conjugate wrt. Ω.

Now, let A′ = T and let B ′ , D′ , C ′ be the other intersections of F A′ , GA′ , F D′ with the circle respectively.
Now, by (TODO 3.2.13) again, we find that E ′ := A′ C ′ ∩ B ′ D′ , G′ := A′ D′ ∩ B ′ C ′ both lie on pω (F ) = GE.
As such,
G′ = A′ D′ ∩ B ′ C ′ = GA′ ∩ B ′ C ′ ∩ pω (F ) = GA′ ∩ GE = G.

and thus E ′ lies on pω (G) = EF , and thus E ′ = GE ∩ EF = E.

Then (A′ , B ′ , C ′ , D′ ) is our desired points.

Example 3.2.15 (2009 China TST Day 1 P1). Let D be a point on segment BC of △ABC such that
∠CAD = ∠ABC. Suppose that O is the midpoint of BD and the circle with center O through B and D

111
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

intersects AB, AD again at E and F respectively. Let G = BF ∩ DE and let M be the midpoint of AG.
Show that CM ⊥ AO.

Solution. Since ∠CAD = ∠ABC, we get that CA is parallel to (ABD), so CA2 = CB · CD and C lies on
the radical axis of (BD) and point circle A. Then, by (TODO 3.2.13) G lies on the polar of A wrt. (BD).
Now, by (TODO 3.2.3) M thus lies on the radical axis of (BD) and A as well. As such, CM is the radical
axis which finishes.

Practice Problems

Problem 1 (2014 ISL G3). Let Ω and O be the circumcircle and circumcenter of acute triangle △ABC with
̸ B. Let Γ be the circle with diameter BM .
AB > BC. The angle bisector of ∠ABC intersects Ω at M =
The angle bisectors of ∠AOB and ∠BOC intersect Γ at points P, Q, respectively. The point R is chosen on
the line P Q so that BR = M R. Prove that BR∥AC.

3.3 Cross Ratios under Inversion and Polarity

The next theorem shows that the pole-polar transformation preserves cross-ratios, coming in handy often.

Theorem 3.3.1. Let Γ be a circle, and L, l1 , l2 , l3 , l4 be five lines, and let their respective poles be A, P1 ,
P2 , P3 , P4 , then
A(P• ) = L(l• )

Proof. Let O be the center of Γ, Qi = ℓi ∩ L, Ki = APi , then Ki = pΓ (Qi ), and we want to show that
(Q• ) = (K• ). We have two scenarios:

(i) If O ∈
/ L, then OQi ⊥ Ki , and hence

∡Qi OQj = ∡(OQi , OQj ) = ∡(Ki , Kj ).

So by the angular definition of the cross-ratio,

(Q• ) = O(Q• ) = (K• ).

(ii) If O ∈ L, construct an arbitrary line ℓ that does not pass O, and let Q′i be the projection of Qi onto ℓ,
and let Ki′ be the polar of Q′i with respect to Γ. Assume Ri = Ki ∩ Ki′ , then R1 , R2 , R3 , R4 are on the
same line O∞ℓ . So we have
(Q• ) = (Q′• ) = (K•′ ) = (R• ) = (K• ).

112
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

We can then use the above theorem to prove that the inverse also preserves the cross ratio.

Proposition 3.3.2. For any collinear/concyclic points P1 , P2 , P3 , P4 ,

(P•J ) = (P1J , P2J , P3J , P4J ) = (P• )

where J is taken wrt a circle E.

Note that because P1 , P2 , P3 , P4 are collinear/concyclic, then P1J , P2J , P3J , P4J is also collinear/concyclic.

Proof. Let the center of inversion be O. We consider four cases:

(i) If P1 , P2 , P3 , P4 are collinear and O ∈ ℓ, then from (TODO 3.3.1) we have

(P• ) = (pE (P• )) = ℓ(pE (P• )) = (P•J ).

/ ℓ, then P1J , P2J , P3J , P4J are concyclic, so

(ii) If P1 , P2 , P3 , P4 are collinear and O ∈

(P•J ) = (OP•J ) = ((OP• ) = (P• ).

(iii) If P1 , P2 , P3 , P4 are concyclic on Γ and O ∈ Γ, then P1J , P2J , P3J , P4J are collinear but do not pass through
O, so by the above case we get (P• ) = (P•J ).

(iv) If P1 , P2 , P3 , P4 are concyclic on Γ and O ∈

/ Γ, then by the invariance of the cross-ratio under inversion,
̸ 0. Then it follows that PiJ ∈ Γ, and let
we can choose the power of inversion to be PowΓ =
Qij = Pi PjJ ∩ PiJ Pj . By (TODO 3.2.13), Qij lies on the polar line pΓ (O) of O with respect to Γ.
Therefore, since Q14 , Q24 , Q34 are collinear, we have

(P• ) = P4J (P• ) = (Q14 , Q23 ; Q34 , P4 P4J ∩ pΓ (O)) = P4 (P•J ) = (P•J ).

As an application of inversion preserving cross-ratio, we now prove the converse of the previously mentioned
Casey’s theorem (TODO 0.4.20):

Theorem (Casey’s Theorem, Repeated). Given four non-intersecting circles ΓA , ΓB , ΓC , ΓD , define d+

IJ to

be the length of the external common tangents of ΓI and ΓJ , and define d−

IJ to be the length of the internal

113
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

common tangents of ΓI and ΓJ . There exists a circle Ω tangent to the four circles ΓA , ΓB , ΓC , ΓD iff.

dBC dAD ± dCA dBD ± dAB dCD = 0,

where 
d+ , if ΓI , ΓJ are both tangent on the same side of Ω,

IJ
dIJ =
d− , if ΓI , ΓJ are tangent on opposite sides of Ω.

IJ

We first prove the following lemma:

Lemma 3.3.3. Let rI , rJ be the radii of ΓI , ΓJ , then the ratio

(d±
IJ )
2

rI rJ

is invariant under inversion about any point X.

Proof. Let OI and OJ be the centers of circles ΓI and ΓJ , respectively. Let PI , QI be the intersection points
of the line segment OI OJ and circle ΓI , and let PJ , QJ be the intersection points of the line segment OI OJ
and circle ΓJ . Then

2 2
d±
IJ OI OJ − (rI ± rJ )
2
=
rI rJ rI rJ



OI OJ + rI ± rJ OI OJ − rI ∓ rJ
=4·
(2rI ) (2rJ )
PI PJ · QI QJ PI QJ · QI PJ
=4· or 4·
PI QI · PJ QJ PI QI · PJ QJ
= 4 · |(PI , QJ ; PJ , QI )| or 4 · |(PI , PJ ; QJ , QI )| . (♠)

Let circle ω be orthogonal to ΓI , ΓJ and intersects ΓI , ΓJ at RI , SI , RJ , SJ . Then the center of ω lies on the
radical axis L of ΓI , ΓJ . Let one of the intersections of L and ω be point A, then there exists a inversion
centered at A that preserves ΓI , ΓJ , and sends ω to a line ℓ. Since ω is orthogonal to these two circles, ℓ is
also orthogonal, so ℓ is just OI OJ . Since inversion preserves cross ratios, we have

(PI , QJ ; PJ , QI ), (PI , PJ ; QJ , QI ) = (RI , SJ ; RJ , SI ), (RI , RJ ; SJ , SI ). (♣)

′
If Γ′I , Γ′J are the images of ΓI , ΓJ under inversion at X (similarily define rI′ , rJ′ , d±
IJ ), we know that OI OJ

will invert to a circle ω ′ that is orthogonal to Γ′I , Γ′J . Let ω ′ intersects Γ′I at PI , QI , which have images PI′ , Q′I

114
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

under inversion at X. Similarily define PJ , QJ , PJ′ , Q′J . By (♠), (♣) and (TODO 3.3.2), we have

(d+ 2
(d− )2
 
IJ )
, IJ = {4 · |(PI , QJ ; PJ , QI )|, 4 · |(PI , PJ ; QJ , QI )}
rI rJ rI rJ
= {4 · |(PI′ , Q′J ; PJ′ , Q′I ), 4 · |(PI′ , PJ′ ; Q′J , Q′I )|}
′ ′
( )
(d+IJ )
2
(d−IJ )
2
= ,
rI rJ rI rJ

(d+ 2
(d− 2
′ ′
 
− − IJ ) IJ )
Since d+ +
IJ > dIJ , dIJ > dIJ , we have that rI rJ , rI rJ is invariant and does not swap under inversion.

We now prove the converse of Casey’s theorem.

Proof. WLOG let Γd have the smallest radii rD . We now use a technique called expansion in America.

We shrink the radii of circles ΓD and other circles on the same side of the desired ω by rD , and increase
the radii of the circles on the opposite by rD .

Then in this new configuration, a circle ω ′ exists iff a circle ω existed in the original configuration, but
also shirnked / expanded by rD .

We can also check that this operations preserves both the existence and lengths of dIJ . This also preserves
the non-intersection condition for circles with internal tangencies. As such, we now reduce to the case where
D is a point circle

Invert about D, which due to the non-intersection condition maps ΓA , ΓB , ΓC to circles Γ′A , Γ′B , Γ′C . We
now wish show the existence of a line tangent to all three of the inverted images. Let ρ2 be the power of the
inversion. As such, we have that

0 = dBC dAD ± dCA dBD ± dAB dCD

     
rB rC rC rA rA rB
r r r
′ ′ ′
= dBC ′ r′ dAD ± dCA ′ r′ dBD ± dAB ′ r′ dCD
rB C rC A rA B
 s s s 
′
rA ′
rB rC′
rA rB rC  ′
r
′ ′
= ′ ′ ′ dBC dAD ± dCA dBD ± dAB dCD 
rA rB rC rA rB rC

rI′ ρ2
If rI = 0 for some I ∈ {A, B, C}, then we set rI
:= d2ID
, which occurs as a limiting case to the above lemma
ρ 2 rI
by (TODO 3.1.18), which says rI′ = d2ID
. Substituting this into the above expression, it simplifies as

s s s
′
rA ′
rB ′
rC
0= d′BC dAD ± d′CA dBD ± d′AB dCD = ρ(d′BC ± d′CA ± d′AB ).
rA rB rC

115
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Now, we once again apply expansion, and WLOG let this map Γ′A into the point circle A′ . Let QR
be a tangent to Γ′B , Γ′C such that Q ∈ Γ′B , R ∈ Γ′C , QR = d′BC . Then take P on QR such that RP =
d′CA , P Q = d′AB (which holds because one of the lengths is the sum of the two others). Then the locus
of points PowΓ′B (S) = d′AB is a circle containing A and R with center OB
′
. Similarly, the locus of points
PowΓ′C (S) = d′CA also contains A and R and has center OC
′
. Thus, the intersection of points in these loci are
{R, R′ } where R′ is the reflection of R about OB OC . Thus, A ∈ {R, R′ }, and R, R′ both lie on a common
tangent of Γ′B and Γ′C , which suffices as the desired line.

Example 3.3.4 (Feuerbach’s Theorem). For △ABC, the nine-point circle ϵ and the incircle ω are internally
tangent, and the nine-point circle and the three excircles are externally tangent. We call the incircle tangency
point the Feuerbach point of △ABC, and we call the three excircle tangency points the A,B,C-Feuerbach
points.

△
! : When △ABC is a equilateral triangle, the Feuerbach point is not defined, but we can still use the
A,B,C-Feuerbach points.

Solution. Let △Ma Mb Mc , △DEF be the medial and intouch triangles of △ABC. We use Casey’s theorem
to prove that there’s a circle through Ma , Mb , Mc that’s tangent to ω = (DEF ). By Casey we only need to
prove
Mb Mc · Ma D ± Mc Ma · Mb D ± Mc Ma · Mc D = 0.

Let a = BC, b = CA, c = AB. Then assume WLOG that a ≥ b ≥ c, then

a b−c b a−c c a−b

M b Mc · Ma D ± Mc Ma · Mb E ± Mc Ma · Mc F = · ± · ± · = 0.
2 2 2 2 2 2

Thus ϵ and ω are tangent.

There is a very simple construction of the Feuerbach point that will be very important later on.

Let I, I a respectively be the incenter and A-excenter, let X be the reflection of D across AI, then F e is
just the second intersection of Ma X and ω. The proof is relatively straightforward.

Proof. Let T be AI ∩ BC, then T X is clearly tangent to ω. We consider the inversion J at Ma with power
2
Ma D , and note that J(ω) = ω. Thus we just need to prove that J(T X) = ϵ.

Let D′ be the reflection of D in Ma , let Ha be the foot from A to BC. Then

∞⊥BC
(Ha , T ; D, D′ ) = (A, T ; I, I a ) = (BA, BC, BI, BI a ) = −1.

116
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Thus we have Ma Ha · Ma T = Ma D2 , which tells us that J(T ) = Ha , thus J(T X) is a circle passing
through Ha , Ma with the tangent at Ma parallel to T X. Thus we only need to prove that the tangent from
Ma to ϵ is also parallel to T X and we’re done.

We angle chase

∡(Ha Ma , T X) = 2 · ∡Ha T A = 2 · (∡Ha AI + 90◦ ) = ∡Ha AI + ∡IAO = ∡Ha AO,

where O is the circumcenter of △ABC. Let Ea be the second intersection of AHa and ϵ, which is just the
midpoint of A and H. We have Ea Ma ∥ AO by considering hH,2 , so

∡(Ha Ma , T X) = ∡Ha AO = ∡Ha Ea Ma ,

and T X is tangent to (Ha Ma Ea ) = ϵ as desired.

For more details on the Feuerbach point view section (TODO 8.2).

3.4 Apollonian Circles

The Apollonian circle is the locus of points with a constant ratio of distances from two other points.

Proposition 3.4.1. Given two points A, B and a positive real r ̸= 1, the set

ΓA,B
r = {P | P A = r · P B}

is a circle, and A, B are inversive images in this circle. We call ΓA,B

r the r-Apollonius circle.

Proof. If r = 1, then Γr is just the perpendicular bisector of segment AB. So assume r ̸= 1, then pick two
points on AB (call them P+ , P− ) such that

−−→ −−→
AP+ AP−
−−−→ = r, −−−→ = −r,
P+ B P− B

and it’s obvious that (A, B; P+ , P− ) = −1. We know that P ∈ ΓA,B

r if and only if P P+ is the angle bisector
of ∡AP B. However by (TODO 2.2.4), this is equivalent to ∡P− P P+ being a right angle. Thus ΓA,B
r must
be a circle with P+ P− as diameter.

Since (A, B; P+ , P− ) = 1, by properties of inversion and cross ratio (TODO 3.2.11), we get that A and B
are inversive images over ΓA,B
r .

117
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

We then have the following

• ΓA,B
r = ΓB,A
r −1 ;

• ΓA,B
r is orthogonal to (AB) by (TODO 3.1.11);

• There exists an Apollonian circle wrt. A, B for any two inverses in the circle (AB) on line AB.

By some algebraic calculations, we can get the following:

Proposition 3.4.2. Given any two points A, B and a positive real r, the circumcenter Or of the Apollonian
circle ΓA,B
r and the radius ρr satisfy

−−→
Or A 2 r
−−→ = r , ρr = |r2 − 1| · AB.
Or B

Proof. Choose point P on Apollonian circle ΓA,B

r , then since A, B are inversive images in ΓA,B
r we have
−
△Or AP ∼ △Or P B. Therefore since Or lies outside segment AB by the inversion property, we have

−−→ −−→ −−→ −→ −→

Or A Or A Or P AP AP
−−→ = −−→ · −−→ = −−→ · −−→ = r2 .
Or B Or P Or B PB PB

Combined with
AP
AOr · AB = PowΓr (A) = Or A2 − ρ2r , Or A = · Or P = r · ρr ,
PB

we get
|r2 − 1|
AB = · ρr ,
r

and we are done.

We can actually extend this as follows.

Proposition 3.4.3. Let P be a point on the r-Apollonius circle ΓA,B

r of A, B. Then ΓA,B
r is orthogonal to
(P AB).

Proposition 3.4.4. Let O be the circumcenter of △ABC. For any three positive reals r, s, t, the power of
O wrt. the three Apollonian circles ΓB,C
r , ΓC,A
s , ΓA,B
t is equal. Furthermore, ΓB,C
r , ΓC,A
s , ΓA,B
t are coaxial if
and only if rst = 1.

Proof. Let Or , Os , Ot be the three circumcenters of ΓB,C

r , ΓC,A
s , ΓA,B
t . Let Ω be the circumcircle of △ABC.
From (TODO 3.4.3) we get that Ω is orthogonal to ΓB,C
r , so we have by (TODO 3.1.10) that PowΓB,C
r
(O) = R2 ,
where R is the circumradius. Applying this symmetrically gives the result.

118
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

For the coaxal condition, we know ΓB,C

r , ΓC,A
s , ΓA,B
t are coaxal if and only if Or , Os , Ot are collinear. By
Menelaus’s theorem and (TODO 3.4.2), this is the same thing as

BOr COs AOt

−1 = · · = (−r2 ) · (−s2 ) · (−t2 ) = −(rst)2 ,
Or C Os A Ot B

or rst = 1.

Remark. When A, B, C are collinear, then the condition of ΓB,C

r , ΓC,A
s , ΓA,B
t being coaxal still holds if and
only if rst = 1. We can’t use the above proof because O is on the line at infinity and it’s too degenerate, but
simple algebra works in this case.

Now we look at Apollonian circles going through a fixed point.

Definition 3.4.5. Given two points A, B and a point P , we define the P -Apollonian circle ΓA,B
p as the
PA
unique Apollonian circle of A, B that goes through P . In this case, r is obviously just PB . Further, we
obviously have ΓA,B
p = ΓB,A
P .

Given two points A, B, with respect to two points P, Q we know that

PA QA AP BP
= ⇐⇒ = .
PB QB AQ BQ

As such, ΓA,B
p = ΓA,B
Q if and only if ΓP,Q
A = ΓP,Q A,B
B . From (TODO 3.4.3), we know that Γp is orthogonal
to (P AB).
√
Proposition 3.4.6. Given △ABC, let IA be a bc-inversion at A. Then

• IA (ΓB,C
A ) is the perpendicular bisector of BC;

• IA (ΓC,A
B ) is the circle through C with center B;

• IA (ΓA,B
C ) is the circle through B with center C.

Proof. Since ΓB,C

A is orthogonal to both BC and (ABC) by (TODO 3.4.3), and ΓB,C
A passes through A, the
image of ΓB,C
A is a line orthogonal to (ABC) and BC, which is just the perpendicular bisector of BC.

Since ΓC,A
B is orthogonal to CA and (ABC) and passes through B, IA (ΓC,A
B ) is orthogonal to IA (CA) =

AB, IA ((ABC)) = BC and passes through IA (B) = C, so it must be the circle centered at B going through
C, and same logic for ΓA,B
C .

Corollary 3.4.7. In △ABC, the three Apollonius circles ΓB,C C,A A,B
A , ΓB , ΓC all pass through two common
points S1 , S2 . Further, these points are inversive pairs under inversion over (ABC).

119
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed
√
Proof. This follows by the Angle-bisector theorem and (TODO 3.4.4). Alternatively, let IA represent bc-
inversion at A. Choose T1 , T2 such that △T1 BC, △T2 BC are equilateral triangles. By (TODO 3.4.6), we can
see that IA (ΓB,C C,A A,B B,C C,A A,B
A ), IA (ΓB ), IA (ΓC ) concur at T1 , T2 . Thus ΓA , ΓB , ΓC concur at S1 = IA (T1 ), S2 =
IA (T2 ).

Next, from (TODO 3.4.4), we get that O has equal power R2 wrt. ΓB,C C,A A,B
A , ΓB , ΓC . Thus O lies on the

radical axis S1 , S2 of the three circles, and

OS1 · OS2 = PowΓB,C (O) = R2 ,

A

so S1 , S2 are inversive pairs under inversion in (ABC).

Thus we know that S1 ̸= S2 and one of them lies inside (ABC), and one of them lies outside of (ABC).
(Note neither can lie on the circumcircle because T1 , T2 ̸∈ BC.) So assume WLOG that S1 lies inside (ABC).

Definition 3.4.8. We call S1 the first isodynamic point and we call S2 the second isodynamic point of
△ABC.

Proposition 3.4.9. Given △ABC, construct three external equilateral triangles △A1 BC, △B1 CA, △C1 AB
on sides BC, CA, AB. Similarily define A2 , B2 , C2 by constructing internal equilateral triangles. Then
AA1 , BB1 , CC1 concur at a point F1 , and AA2 , BB2 , CC2 concur at a point F2 . Further, F1 and F2 are the
isogonal conjugates of S1 and S2 respectively.

√
Proof. Since S1 is inside of (ABC), under a bc-inversion we get that T1 = IA (S1 ) is on the opposite side of
line BC as point A. Formally, note that by (TODO 3.1.2),

AB · AC AB · AC
· IA (O)T1 = OS1 < OA = .
AIA (O) · AT1 AIA (O)

Since IA (O) is the reflection of A over BC, and this implies IA (O)T1 < AT1 , the result follows. Similarily
T2 = IA (S2 ) is on the same side of BC as point A. Thus by the proof of (TODO 3.4.1), we get T1 = A2 , T2 =
A2 .

Let i ∈ {1, 2}. Then AAi and ASi are isogonal, and thus by symmetry AAi , BBi , CCi concur at the
isogonal conjugate of Si .

From
AB AC1
∡BAB1 = ∡BAC + 60◦ = ∡C1 AC (mod 360◦ ), =
B1 A CA

we get
+
△ABB1 ∼ △AC1 C.

120
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

Thus
∡BF1 C = ∡(BB1 , C1 C) = ∡(AB, AC1 ) = 120◦ .

Similarily we have the other two internal angles ∡CF1 A = ∡AF1 B = 120◦ . In a similar nature, F2 we
have ∡BF2 C = ∡CF2 A = ∡AF2 B = 60◦ .

As such, we can characterize F1 , F2 in a completely different way.

Definition 3.4.10. Given △ABC labeled counterclockwise, the points F1 , F2 where Fi satisfies ∡BFi C =
∡CFi A = ∡AFi B = −i · 60◦ are called the first Fermat point and second Fermat point respectively.

Remark. When no angle of a triangle exceeds 120◦ , F1 minimizes the sum of AX + BX + CX over points
X.

Finally, (TODO 3.4.9) lets us recharacterize the isodynamic points.

Proposition 3.4.11. The pedal triangles wrt the two isodynamic points Si and △ABC are equilateral
triangles.

Proof. Fix Si . Then let △Sa Sb Sc represent the pedal triangle of Si , then

∡Sb Sa Sc = ∡Sb Sa S + ∡SSa Sc = ∡Sb CS + ∡SBSc = ∡Fi CB + ∡CBFi = ∡(CFi , BFi ) = i · 60◦ ,

and similarily on the other corners of the pedal triangle to get that all angles are equal.

Remark. Additionally, by (TODO 3.4.6) inverting △ABC around S maps it to an equilateral triangle.

We will revisit Fermat and isodynamic points with the Neuberg cubic in later chapters.

3.5 Apollonius’s Circle Problem

Let’s now figure out how to construct a circle tangent to three other circles, Γ1 , Γ2 , Γ3 .

Suppose Ω is tangent to Γ1 , Γ2 , Γ3 . We will find some characterizations of Ω.

Proposition 3.5.1. Let R be the radical center of Γ1 , Γ2 , Γ3 , and let Ti be the touchpoint of Ω and Γi , let
Pi be the pole of line RTi across Γi . Then we have P1 , P2 , P3 are collinear. Further, let J be an inversion
centered at R with power PowΓi (R). Then P1 P2 P3 is the radical axis of Ω and ΩJ .

Proof. Since inversion preserves tangency, from ΓJi = Γi , we get that ΩJ is tangent to Γ1 , Γ2 , Γ3 as well (note
this does not necessarily imply Ω = ΩJ .) Note that the tangency point of ΩJ is just TiJ . Since Pi Ti is the

121
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

radical axis of Ω and Γi and Pi TiJ is the radical axis of ΩJ , Γi , Pi is the radical center of Ω, ΩJ , Γi , so Pi lies
on the radical axis ℓ of Ω, ΩJ .

From this characterization we can know that the pole of ℓ wrt. Γi (let it be Qi ) must lie on RTi .

Proposition 3.5.2. Extending the notation from (TODO 3.5.1), let TiJ be the touchpoint of ΩJ and Γi ,
then Ti Tj and TiJ TjJ intersect at one of the similicenters of Γi and Γj , Uij . Further Uij lies on the radical
axis ℓ of Ω and ΩJ .

Proof. Since Ti is one of the similicenters of Ω and Γi , Tj is one of the similicenters of Ω and Γj , we can use
Monge (TODO 1.1.12) to get that Ti Tj passes through one of the similicenters of Γi , Γj . Similarily we get
that TiJ TjJ passes through one of the similicenters of Γi , Γj . So for the following section we will prove that
these two similicenters have the same sign, or instead that Oi Oj , Ti Tj , TiJ TjJ are concurrent. Let Oi be the
center of circle Γi , and let O, OJ be the centers of Ω, ΩJ . Note that △OTi Tj and △OJ TiJ TjJ are perspective,
let their perspector be R. By Desargues’s theorem (TODO 1.1.1) we have

Ti Tj ∩ TiJ TjJ , Oj = Tj O ∩ TjJ OJ , Oi = OTi ∩ OJ TiJ

are collinear.

Thus Oi Oj , Ti Tj , TiJ TjJ are concurrent.

Since the radical axis of Ω and (Ti Tj TiJ TjJ ) is line Ti Tj , and the radical axis of ΩJ and (Ti Tj TiJ TjJ ) is
TiJ TjJ , we have that Uij is the radical center of Ω, ΩJ , (Ti Tj TiJ TjJ ), thus Uij also lies on the radical axis of Ω
and ΩJ .

By Monge (TODO 1.1.12), we have that the pairwise similicenters of the same sign Oij,± of Γ1 , Γ2 , Γ3
form a complete quadrilateral with sides (ℓ1 , ℓ2 , ℓ3 , ℓ4 ), where

ℓ1 = O23,+ O31,− O12,− , ℓ2 = O23,− O31,+ O12,− ,

ℓ3 = O23,− O31,− O12,+ , ℓ4 = O23,+ O31,+ O12,+ .

So by (TODO 3.5.2) we have that the radical axis of Ω and ΩJ has to be a line in {ℓ1 , ℓ2 , ℓ3 , ℓ4 }.

We already know most of the information needed to reconstruct Ω, we just need to construct Ω given
the complete quadrilateral formed by the various similicenters. Let R and Qi be defined as before. If RQi
and Γi do not intersect, then we can’t construct Ω, so assume they intersect for some i. These intersection

122
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

points will be Ti and TiJ . Now, it remains to distinguish between Ti , TiJ such that (T1 T2 T3 ), (T1J T2J T3J ) are
all tangent to Γ1 , Γ2 , Γ3 . To prove this we just need:

Proposition 3.5.3. Let R be a point on the radical axis of Γ1 , Γ2 , and let U be one of their similicenters.
Let ℓ be a line through U , and let Qi be the pole of ℓ wrt. Γi . If T1 is one of the intersection points of RQ1
and Γ1 , then U T1 , RQ2 , Γ are concurrent.

Proof. We consider the homothety h with center U sending h(Γ1 ) = Γ2 . We have

h(Q1 ) = h(pΓ1 (ℓ)) = ph(Γ1 ) (h(ℓ)) = pΓ2 (ℓ) = Q2 .

Let S = h(T1 ), let T2 be the second intersection point of ST1 and Γ2 . Let S ′ , T2′ respectively be the second
intersections of SQ2 , T2 Q2 with Γ2 . By Brokard’s theorem (TODO 3.2.13) we have ST2 ∩ S ′ T2′ is on the
polar of Q2 = SS ′ ∩ T2 T2′ , so U ∈ S ′ T2′ . By Reim’s (TODO 0.1.14) we also have that since S, T2 , S ′ , T2′ are
concyclic, so T1 , T2 , T1′ , T2′ concyclic, where T1′ = h−1 (S ′ ). Since T1 T1′ is the radical axis of (T1 T2 T1′ T2′ ) and
Γ1 , and T2 T2′ is the radical axis of (T1 T2 T1′ T2′ ) and Γ2 , we have T1 T1′ = T1 Q1 and T2 T2′ = T2 Q2 intersect at
the radical axis of Γ1 , Γ2 , so R, T2 , Q2 are collinear.

We now work on constructing ω given Γi . We first construct T1 , T2 , T3 and their inverses, repeating
definitions for clarity: Let Oi be the centers of Γi and take ℓ as one of the above four Monge line. Define
Uij = Oi Oj ∩ ℓ. Take R as the radical center of Γi , Qi as the pole of ℓ wrt Γi and let T1 , T1J arbitrarily be the
intersections of Γ1 and RQ1 . We can then construct T2 as T1 U12 ∩ Q2 T2′ , and T2J as the other intersection of
T2 Q2 with Γ2 , such that U12 = T1 T2 ∩ T1J T2J . We can do the same for T3 , T3J such that T1 T3 ∩ T1J T3J = U31 .

Now, notice then that △T1 T2 T3 and △T1J T2J T3J are perspective at R, so by Desargues’s (TODO 1.1.1),
we have
T2 T3 ∩ T2J T3J , T3 T1 ∩ T3J T1J = U31 , T1 T2 ∩ T1J T2J = U12

are collinear, so we have U23 = T2 T3 ∩ T2J T3J passes through U23 . Now, we only need to prove that
(T1 T2 T3 ), (T1J T2J T3J ) are tangent to Γ1 , Γ2 , Γ3 .

Proposition 3.5.4. Let T1 , T2 , T3 respectively lie on circles Γ1 , Γ2 , Γ3 such that Ti Tj passes through one of
the similicenters Uij of Γi , Γj , and U23 , U31 , U12 are collinear. Then the circumcircle of △T1 T2 T3 is tangent
to Γ1 , Γ2 , Γ3 .

Proof. Borrowing the previous notation, let R be the radical center of Γ1 , Γ2 , Γ3 , let Qi be the pole of
U23 U31 U12 wrt. Γi . Let Sij be the second intersection of Ti Tj and Γj , and let hij be the homothety centered
at Uij sending hij (Γi ) = Γj , and Sij = hij (Ti ), which is on Γj .

123
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed
+
Since hij (Qi ) = Qj , we have Sij Qj ∥ Qi R, and thus by spiral similarity, △RTi Tj ∼ △Qj Sij Tj . Combining
+ + +
△RT2 T1 ∼ △Q1 S21 T1 , △RT3 T1 ∼ △Q1 S31 T1 , we get △T1 T2 T3 ∼ △T1 S21 S31 , thus Ω = (T1 T2 T3 ) is tangent
to Γ1 = (T1 S21 S31 ), and similarily Ω is tangent to Γ2 , Γ3 .

Since Ω is tangent to each Γi , O = O1 T1 ∩ O2 T2 ∩ O3 T3 is the circumcenter of △T1 T2 T3 , and △O1 O2 O3

and △T1 T2 T3 have U12 U23 U31 as a perspectrix.

This can also be proven through Menelaus’s Theorem. If O is the perspector of △O1 O2 O3 and △T1 T2 T3
(which exists by Deargues’s), then by Menelaus’s

O1 T1 OT2 O1 U12 r1
· =− =±
T1 O T2 O2 U12 O2 r2

where ri is the radius of Γi . This then implies that OT1 = OT2 , which follows by symmetry.

Remark. Each of the 4 choices of ℓ gives us a different pair of circles, which gives us a total of 8 circles as
solutions to being tangent to all of Γi .

In the above, we solved the problem when Γ1 , Γ2 , Γ3 are circles. However, Apollonius’s problem also exists
for the degenerate cases where circles are instead replaced with lines and points. The proof is similar to
(TODO 3.5.4) when there’s multiple circles, and is left to reader.

Now, here’s the constructions.

(i) All three are circles: See above section.

(ii) Γ1 = L1 is a line, the rest are circles: Let R be the intersection of the radical axis of Γ2 , Γ3 with L1 , let
O12,± be the intersections of O2 ∞⊥L1 with Γ2 , define O31,± similarly. Take a monge line U23 U31 U12 .
Then let T2 = RQ2 ∩ Γ2 be one intersection, then T1 = U12 T2 ∩ L1 , T3 = U23 T2 ∩ L1 .

(iii) Γ1 is a circle, Γ2 = L2 , Γ3 = L3 are lines: Let R = L2 ∩ L3 , define O12,± , O31,± as the same as in (i).
Define O23,± as the points along infinity on the angle bisectors between L2 and L3 . Take a monge line
U23 U31 U12 , and finish the same as above.

(iv) All Γ1 = L1 , Γ2 = L2 , Γ3 = L3 are lines: Then Ω is just an incircle or excircle of △L1 L2 L3 .

(v) Γ1 = P1 is a point, Γ2 , Γ3 are circles: R is still the radical center, let O31,± and O12,± all be P1 . Take a
Monge line P1 O23,± and let U23 = O23,± . Then define T2 as an intersection of RQ2 with Γ2 , define T3
normally, and let T1 = P1 .

(vi) Γ1 is a circle, Γ2 = P2 , Γ3 = P3 are points: Define R as the radical center, and take the Monge line
P2 P3 . Define T1 = RQ1 ∩ L1 , and take T2 = P2 , T3 = P3

124
AoPS Chapter 3. Inversion and Polarity - Hidden/Completed

(vii) Γ1 = P1 , Γ2 = P2 , Γ3 = P3 are points: Then Ω is just the circumcircle of △P1 P2 P3 .

(viii) Γ1 is a circle, Γ2 = L2 is a line, Γ3 = P3 is a point: R is the intersection of the radical axis of P3 , Γ1 with
L2 . Define O12,± as in the first case, define O23,± O31,± as P3 . Take a Monge line P3 O12,± , U12 = O12,± ,
and define T1 = RQ1 ∩ L1 , T2 = U12 T1 ∩ L2 , T3 = P3 .

(ix) Γ1 = L1 is a line, Γ2 = P2 , Γ3 = P3 are points: Let U = L1 ∩ P2 P3 . Take T1 on L1 such that

2
U T1 = U P2 · U P3 , and Ω = (T1 P2 P3 )

(x) Γ1 = P1 is a point, Γ2 = L2 , Γ3 = L3 are lines: Let ℓ be a line through P1 parallel to one of the angle
bisectors of L2 , L3 . Let U2 , U3 be the intersections of ℓ with L2 , L3 respectively. Let P1′ be the reflection
of P about the midpoint of U2 U3 . Then Ω is the circle through P1 , P1′ tangent to L2 .

125
Chapter 4

Complete Quadrilaterals -
Hidden/Completed

A complete quadrilateral is just four non-concurrent lines and the six points they intersect at. Taking
just three non-concurrent lines gave us a triangle, giving us such beautiful theorems as the Euler line, the
nine-point circle, Feuerbach’s tangency, etc.

What secrets do four lines hold, and how can we apply these back to triangle geometry later on?

4.1 Top 10 Points In a Complete Quadrilateral

In this section we will define and characterize some well-known points in a complete quadrilateral (note that
many of these can be defined for a normal convex quadrilateral too, but for simplicity we just consider the
complete quadrilateral case.). Given a complete quadrilateral Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ), we define

• Six intersection points: Aij = ℓi ∩ ℓj ;

• Four triangles: △i := △ℓi+1 ℓi+2 ℓi+3 ;

• Three diagonals: Aij Akl (or (Aij , Akl ));

• Three quadrilaterals: Aij Ajk Akl Ali (diagrams please...)

• The diagonal triangle (cevian triangle): δ, formed by the extensions of the three diagonals previously
defined.

We also use notation from here below, where say QL-P1 is the Miquel Point.

126
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

Proposition 4.1.1 (QL-P1, the Miquel Point). The four circumcircles of △1 , △2 , △3 , △4 concur.

Proof. We have proved this in (TODO 0.1.18), but it is just angle chasing.

Example 4.1.2 (APMO 2015/1). Let ABC be a triangle, and let D be a point on side BC. A line through
D intersects side AB at X and ray AC at Y . The circumcircle of triangle BXD intersects the circumcircle
ω of triangle ABC again at point Z distinct from point B. The lines ZD and ZY intersect ω again at V and
W respectively. Prove that AB = V W .

Solution. [AoPS, TelvCohl] Note that Z is the Miquel point of complete quadrilateral {AB, BC, CA, XY },
so
Z ∈ (CDY ) =⇒ ∠V ZW = 180◦ − ∠ACB =⇒ AB = V W.

Example 4.1.3 (ISL 2020 G6). Let ABC be a triangle with AB < AC, incenter I, and A excenter IA . The
incircle meets BC at D. Define E = AD ∩ BIA , F = AD ∩ CIA . Show that the circumcircle of △AID and
△IA EF are tangent to each other.

Solution. Notice that if (AID) and (IA EF ) are tangent at the Miquel point of △BIA C ∩ AD, then angle
chasing in the problem becomes almost trivial. So what if we just guess that the tangency point is M ?

First we prove that M lies on (AID) (since M already lies on (IA EF ), we don’t need to care about that).
From M ∈ (BIA C), we get

∡DM I = ∡DM B + ∡BM I = ∡DEB + ∡BIA I = ∡DAI,

and thus M lies on (AID). Next, we prove tangency at M by usage of line-arguments.

TM (AID) = M I + M D − ID = M IA + ∡(BC, M D)

= M IA + ∡BEM = M IA + M E − IA E = TM (IA EF )

Despite that we literally just blindly guessed the Miquel point appearing here, the “motivation” for it
is that if the tangency point is not the Miquel point, then we have absolutely no way to chase all of these
angles. So whenever you see a tangency condition (aka: angles) in a complete quadrilateral problem, just
guess the Miquel point; you have nothing to lose.

127
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

Proposition 4.1.4 (QL-L1, the Newton Line). The three midpoints of the diagonals of the segments A23 A14 ,
A31 A24 , A12 A34 are collinear on a line τ .

Proof. Let the three midpoints be M1 , M2 , M3 respectively, and let △N1 N2 N3 be the medial triangle of
△A23 A31 A12 . Then we get that M1 , M2 , M3 lie on N2 N3 , N3 N1 , N1 N2 respectively. Then by Menelaus we
get
N2 M1 N3 M2 N1 M3 A12 A14 A23 A24 A31 A34
· · = · · = −1,
M1 N3 M2 N1 M3 N2 A14 A31 A24 A12 A34 A34

and then we do Menelaus once again on M1 , M2 , M3 and win.

Proposition 4.1.5 (QL-Ci3, the Miquel Circle). Let Oi be the circumcenter of △i . Then M, O1 , O2 , O3 , O4
are concyclic.

Proof. We prove that M lies on (O1 O2 O3 ) which finishes by symmetry. We know that the reflections of M
across O2 O3 , O3 O1 , O1 O2 are points A14 , A24 , A34 respectively. These three points all lie on ℓ4 . Thus by the
converse of Simson line (TODO 1.4.1), we get that M lies on (O1 O2 O3 ). Do this symmetrically and we’re
done.

Note that by Steiner line (TODO 1.4.2) we can get that the orthocenter Hl′ of △Oi Oj Ok lies on ℓ1 .

Proposition 4.1.6 (QL-L2, Orthocenter Line, Steiner Line). Let Hi be the orthocenter of △i . Then
H1 , H2 , H3 , H4 are collinear.

Proof. Let Mi′ be the reflection of M across ℓi . By Steiner’s theorem, we have that Hi , Mj′ , Mk′ , Ml′ are
collinear. Do this cyclically to get that H1 , H2 , H3 , H4 , M1′ , M2′ , M3′ , M4′ are collinear.

Remark. Take n ≥ 5, and a complete n-gon N (ℓ1 , . . . , ℓn ). If M lies on the circumcircle of △ℓi ℓj ℓk for all
i, j, k, then the orthocenters of △ℓi ℓj ℓk also lie on a line.

Proposition 4.1.7 (Gauss-Bodenmiller Line). The three circles

(A23 A14 ), (A31 A24 ), (A12 A34 )

are coaxial, and the Steiner line S is their common radical axis.

Proof. We prove that H1 has the same power wrt. all of these three circles. Let △DEF be the orthic triangle
of △A23 A31 A12 , then



 Pow(A23 A14 ) (H1 ) = H1 A23 · H1 D,


 Pow(A31 A24 ) (H1 ) = H1 A31 · H1 E,



(A12 A34 ) (H1 ) = H1 A12 · H1 F,
Pow

128
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

Since A31 , A12 , E, F are concyclic we know that H1 A31 · H1 E = H1 A12 · H1 F, and similarily we have

H1 A23 · H1 D = H1 A31 · H1 E = H1 A12 · H1 F.

Thus H1 lies on the radical axis, and by symmetry we have that H2 , H3 , H4 ∈ S also lie on the radical
axis.

Because the radical axis of two circle is perpendicular to the line between their centers, by (TODO 0.4.9),
it follows by the above that:

Corollary 4.1.8. The Newton line τ is perpendicular to the Steiner line S.

So in reality we could have also used radical axis properties to prove the Newton line directly, instead of
spamming Menelaus.

Example 4.1.9. Let I be the incenter of △ABC, and let H be its orthocenter. Let E, E ′ be the CA-intouch
and extouch points. Let F, F ′ be the AB-intouch and extouch points. Let I ′ be the reflection of I in EF .
Prove HI ′ ⊥ E ′ F ′ .

Solution. Note that I ′ is the orthocenter of △AEF , so HI ′ is the Steiner line of complete quadrilateral
△ABC ∪ EF = (BC, CA, AB, EF ). So proving HI ′ ⊥ E ′ F ′ is equivalent to proving E ′ F ′ is parallel to the
Newton line of (BC, CA, AB, EF ).

Let X, Y, Z respectively be the midpoints of EF , BE, CF , then

CE E′A FB AF ′
ZX = = , XY = = ,
2 2 2 2

+
so we have ZX ∥ E ′ A, XY ∥ AF ′ . Thus we get that △XY Z ∼ AF ′ E ′ . Thus Y Z (which is just the Newton
line of △ABC ∪ EF ) is parallel to E ′ F ′ , we are done.

Another easy corollary is:

Corollary 4.1.10. Given △ABC and a point P , choose D, E, F on BC, CA, AB such that

∡AP D = ∡BP E = ∡CP F = 90◦ ,

then D, E, F are collinear. We call this line the orthotransversal of P .

Proof. We proceed by phantom points. Let D′ be the intersection of EF and BC. Let Q be the complete
quadrilateral △ABC ∪ EF , then we know that the circles with diameters as the three diagonals of Q, circles

129
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

(AD′ ), (BE), (CF ) are coaxal. Note that P is one of the intersections of (BE) and (CF ), so P also lies on
(AD′ ). Thus D = D′ .

We will further expand on orthotransversals in (TODO 12.1).

Finally, we look at a super-useful theorem.

Theorem 4.1.11. The Miquel point M of a complete quadrilateral Q is the isogonal conjugate of the point
at infinity ∞τ along the Newton line.

Proof. Denote the antipode of M in △1 as M1∗ , and similarily for M2∗ , M3∗ , M4∗ . From (TODO 1.4.5) we
have that the isogonal conjugate of M1∗ in △1 is on the line at infinity, and furthermore it’s the point at
infinity along the Steiner line, ∞S . Thus the isogonal conjugate of M in △1 is just ∞⊥S = ∞τ . Similarily
the isogonal conjugate of M in every other triangle is also ∞τ , so we are done.

This theorem can be more easily expressed in line arguments as

∡(τ ) = ∡(Ajk Aki ) + ∡(Ajk Aij ) − ∡(Ajk M ).

This lets us locate the Miquel point through the Newton line, which we often combine with (TODO
problem 4 from 1.1) and (TODO 4.1.8).ss

We will show more examples of the utility of this fact.

Example 4.1.12. Let I b , I c be the B, C-excenters of △ABC. Let the perpendiculars dropped from I b , I c to
BC intersect CA, AB at E, F . Prove that the circumcenter of △ABC lies on the radical axis of (ABE) and
(ACF ).

Solution. Let M be the Miquel point of (CA, AB, BE, CF ). Then the radical axis of (ABE) and (ACF ) is
AM . As such, we need to prove that AM goes through the circumcenter of △ABC, or that the isogonal line
of AM in ∠BAC is the perpendicular to BC. By the previous theorem, we know the isogonal line of AM in
∠BAC is A∞τ , and it remains to show that τ is perpendicular to BC. Since τ is the line connecting the
midpoints of BC and EF , we just need to prove that the midpoint of EF lies on the perpendicular bisector
of BC. However I b E, I c F ⊥ BC, so it is equivalent to prove the midpoint of I b I c lies on the perpendicular
bisector of BC. But by nine-point circle on Ia Ib Ic , we know it lies on BAC,
˘ so we are done.

Example 4.1.13 (IMO 2009/2). Let O be the circumcenter of △ABC. Let P, Q be two points in segments
CA and AB. Let K, L, M respectively be the midpoints of BP, CQ, P Q. Let Γ be a circle through K, L, M .
Suppose the line P Q is tangent to Γ. Prove that OP = OQ.

130
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

Solution. Let N be the Miquel point of Q = △ABC ∪ P Q. Since the Newton line of this quadrilateral is
KL, we have

∡(AB) + ∡(AC) − ∡(AN ) = ∡(KL) = ∡(M K) + ∡(M L) − ∡(P Q) = ∡(AB) + ∡(AC) − ∡(P Q),

so AN ∥ P Q. Since A, N, P, Q are concyclic, AN ∥ P Q tells us that the perpendicular bisector of P Q is the

same as the perpendicular bisector of AN , which passes through O, so OP = OQ.

Practice Problems

Problem 1. Prove the Miquel point of a complete quadrilateral lies on the nine-point circle of the diagonal
triangle.

Problem 2 (IMO 2011/6). Let ABC be an acute triangle with circumcircle Γ. Let ℓ be a tangent line to Γ,
and let ℓa , ℓb and ℓc be the lines obtained by reflecting ℓ in the lines BC, CA and AB, respectively. Show
that the circumcircle of the triangle determined by the lines ℓa , ℓb and ℓc is tangent to the circle Γ.

Problem 3 (APMO 2014/5). Circles ω and Ω meet at points A and B. Let M be the midpoint of the arc
AB of circle ω (M lies inside Ω). A chord M P of circle ω intersects Ω at Q (Q lies inside ω). Let ℓP be the
tangent line to ω at P , and let ℓQ be the tangent line to Ω at Q. Prove that the circumcircle of the triangle
formed by the lines ℓP , ℓQ and AB is tangent to Ω.

Problem 4. Let the incenter of △ABC be I and let the circumcenter be O. Let the intouch points on
BC, CA, AB be D, E, F . Let F D, DE intersect CA, AB at Y, Z. Let K be the circumcenter of △DY Z.
Prove that ∡AIO = ∡KID.

Problem 5 (2014 Iran TST3/6). The incircle of a non-isosceles triangle ABC with the center I touches the
sides BC at D. Let X be a point on arc BC from circumcircle of triangle ABC such that if E, F are feet
of perpendicular from X on BI, CI and M is midpoint of EF , then we have that M B = M C. prove that
∠BAD = ∠XAC

Problem 6. Given acute triangle △ABC, let O be the circumcenter. Let Γ be the circumcircle of △OBC,
and let G be a point on Γ. Let the circumcircles of △ABG, △ACG intersect CA, AB respectively again at
E, F . Let K = BE ∩ CF . Prove AK, BC, OG are concurrent.

4.2 Complete Cyclic Quadrilaterals

In this section we will briefly discuss what happens to a complete quadrilateral when four of its six vertices
are concyclic. This is a very common configuration to see in problems, for example if BCEF is cyclic,

131
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

A = F B ∩ CE, D = BC ∩ EF (example: a pedal triangle), we can apply this section.

We let Q be the complete quadrilateral (BC, CE, EF, F B) and let the Miquel point be MQ , let the
Newton line be τQ , let the Steiner line be SQ .

Proposition 4.2.1. The Miquel point lies on AD.

Proof. We have (F BC), (CDM ), (M AF ) concur at E, so by triangle Miquel on △BDA and C, M, F lies on
AD.

Proposition 4.2.2. We have that BOEM and COF M are concyclic.

Proof. Since A, D, M are collinear, we can just directly chase angles to get

∡BOE = ∡BCE + ∡BF E = ∡BM D + ∡AM E = ∡BM E,

The result follows by symmetry C, O, F, M are concyclic.

So we can re-define M as the second intersection point of (BOE) and (COF ). Let P be the intersection
of the diagonals BE and CF of BCEF . Now consider inverting about the circumcircle.

Corollary 4.2.3. O, P, M are collinear, and P, M are exchanged under inverting about the circumcircle
(BCEF ).

Proof. Consider inverting about (BCEF ), which maps X 7→ X ∗ . Then we have B ∗ = B, C ∗ = C, E ∗ =

E, F ∗ = F . Thus (BOE)∗ = BE and (COF )∗ = CF . By (TODO 3.1.3) we thus have

M ∗ = ((BOE) ∩ (COF ) \ {O})∗ = BE ∩ CF = P.

By (TODO 3.2.13) we also have that △ADP is self-polar in (O), so we have OP ⊥ AD. Combining this
with the above theorem we get

Proposition 4.2.4. OM ⊥ AD.

We can also prove this with direct PoP.

Proof. Since A lies on the radical axis F B of (BCEF ), (BDF ), we have

Pow(BCEF ) (A) = Pow(BDF ) (A) = AM · AD = AM 2 − M A · M D,

132
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

and by the same logic we get that Pow(BCEF ) (D) = DM 2 − M A · M D. Therefore

OA2 − OD2 = Pow(BCEF ) (A) − Pow(BCEF ) (D)

= (AM 2 − M A · M D) − (DM 2 − M A · M D)

= AM 2 − DM 2 ,

so by (TODO 0.2.11) we have OM ⊥ AD.

Proposition 4.2.5. P lies on the Steiner line S.

Proof. We know that S is the radical axis of (AD), (BE), (CF ), and we have

Pow(BE) (P ) = P B · P E = P C · P F = Pow(CF ) (P ),

so we have P ∈ S.

Example 4.2.6 (2009 G4). Given cyclic quadrilateral ABCD, let diagonals AC and BD intersect at E,
and let AD and BC intersect at F . Let G, H respectively be the midpoints of AB and CD. Prove that EF
is tangent to (EGH) at E.

Proof. Let M be the midpoint of EF . Then we have that M, G, H is the Newton line of ABCD. Let X be
the intersection of AB and CD. Then EF is the pole of X across ABCD. Therefore, EF is the radical axis
of (OX) and (ABCD). Since E, F are inversive pairs in (ABCD), we have that the circle (EF ) is orthogonal
to (ABCD). Since M is the midpoint of EF , it is the center of (EF ), so we have

M E 2 = Pow(ABCD) (M ) = Pow(OX) (M ) = M G · M H,

so (EGH) is tangent to M E = EF .

4.3 Complete Tangential Quadrilaterals

For conciseness, in the following section we will keep using Aij = ℓi ∩ ℓj . We will also assume that
A42 A23 A31 A14 is a convex quadrilateral, A41 A12 A23 A34 is a concave quadrilateral, and A42 A31 A12 A24 is a
self-intersecting quadrilateral.

Proposition 4.3.1 (Pitot’s Theorem). For a complete quadrilateral Q = (l1 , l2 , l3 , l4 ), the following statements
are equivalent.

133
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

(i) Q has an incircle, or is tangential;

(ii) A42 A23 + A31 A14 = A23 A31 + A14 A42

(iii) A41 A12 + A23 A34 = A12 A23 + A34 A41

(iv) A43 A31 + A12 A24 = A31 A12 + A24 A43

Proof. If Q has incircle ω, let Ti be the point of tangency of li to ω. For all permutations i, j, k, l of 1, 2, 3, 4
we get

Aij Ajk + Akl Ali = Aij Tj + Tj Ajk + Akl Tl + Tl Ali

= Aij Ti + Tk Ajk + Akl Tk + Ti Ali = Ajk Akl + Ali Aij

where lengths are directed along each line in a specific way. This implies (ii), (iii), (iv).

Next, assume condition (ii) is satisfied, we need to prove that Q has an incircle

Proposition 4.3.2. To check if Q has an excircle, the following properties are equivalent:

• Q has an excircle;

• A42 A23 − A31 A14 = A23 A31 − A14 A42

• A41 A12 − A23 A34 = A12 A23 − A34 A41

• A43 A31 − A12 A24 = A31 A12 − A24 A43

Proof. The proof is the exact same as (TODO 4.3.1), so we will omit it.

Example 4.3.3. Let ABCD be a convex quadrilateral. Let P, Q, R, S respectively be four points on
AB, BC, CD, DA. Let X be the intersection point of P R and QS. Suppose SAP X, P BQX, QCRX, RDSX
are all tangential quadrilaterals. Prove that ABCD is also tangential.

Solution. Let ωA , ωB , ωC , ωD be the incircles of SAP X, P BQX, QCRX, RDSX respectively. Let TIDC for a
tangent DC to ωI be the touch point. By (TODO 4.3.1), we want to show that AB + CD = BC + DA.

We have that

   
AB + CD = ATAA + TAA T P B + TBP B B + CTCCR + TCCR TDRD + T RD D
D
   
= ATASA + TAXS TBQX + TBBQ B + CTCQC + TCXQ TD SX T SX D ,
D
   
BC + DA = TBBQ B + TCQC TBBQ + CTCQC + TD SX D + T SA T SX + AT SA
A D A

134
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

and thus it remains to show that TAXS TBQX + TCXQ TD

SX and T QC T BQ + T SA T SX are the same value. To do
C B A D

this, note that

   
TAXS TBQX + TCXQ TD
SX = T SX X + XT QX + T XQ X + XT SX
A B C D

= TAP X X + XTBXP + TCRX X + XTD

XR

XR = T QC T BQ + T SA T SX
= TCRX TBXP + TAP X TD C B A D

For the rest of the section let I be the center of the quadrilateral’s incircle.

Proposition 4.3.4. I lies on the Newton line τ of Q.

Proof. In reality, this is Newton’s conic theorem (TODO 6.3.13) on circles. However let’s give an elementary
proof: let Ti be the tangency point of line ℓi and ω, let M2 , M3 be the midpoints of A31 A24 , A12 A34 , then we
have using (signed area)

1
[△IM2 M3 ] = ([△IA31 A12 ] + [△IA24 A12 ] + [△IA31 A34 ] + [△IA24 A34 ]])
4
1
= ([△IA31 T1 ] + [△IT1 A12 ] + [△IA24 T2 ] + [△IT2 A12 ]
4
+ [△IA31 T3 ] + [△IT3 A34 ] + [△IA24 T4 ] + [△IT4 A34 ]) = 0

with the final equality following because

[△IA31 T1 ] = [△IT3 A31 ], [△IT1 A12 ] = [△IA12 T2 ]

[△IA24 T2 ] = [△IT4 A24 ], [△IT3 A34 ] = [△IA34 T4 ].

So I ∈ τ = M2 M3 .

Proposition 4.3.5. Let M be the Miquel point of Q. Then M I is the internal angle bisector of the angles
∠A23 M A14 , ∠A31 M A24 , and ∠A12 M A34 . Further, if we let J to be the reflection of I across M , we have
(IJ)(A23 A14 ), (IJ)(A31 A24 ), (IJ)(A12 A34 ) are all cyclic harmonic quadrilaterals.

Proof. Let Ti be the tangency point of ℓi with ω. We consider the inversion J about the incircle. We have
A∗ij := J(Aij ) as the midpoint of Ti Tj , and

\
M ∗ := J(M ) = (A∗jk A∗ki A∗ij ).

135
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

Let G be the centroid of (T1 , T2 , T3 , T4 ), then G is also the common midpoint of A∗23 A∗14 , A∗31 A∗24 , A∗12 A∗34 .
We consider the reflection about G, call this transformation s. Then we have s(A∗ij ) = A∗kl , so

\ \
s(M ∗ ) = s((A∗jk A∗ki A∗ij )) = (A∗il A∗jl A∗kl ) = I.

Therefore (IM ∗ )(A23 A14 ) is a parallelogram, and thus

∡A23 M I − ∡IM A14 = ∡IA∗23 M ∗ − ∡M ∗ A∗14 I = 0◦ ,

so M I bisects ∠A23 M A14 . Similarily, M I also bisects ∠A31 M A24 , ∠A12 M A34 . Since J is the reflection of I
across M , we know that J ∗ := J(J) is the midpoint of IM , which is G. From the fact that J ∗ = G is the
common midpoint of A∗23 A∗14 , A∗31 A∗24 , A∗12 A∗34 , we get that (IJ)(A23 A14 ), (IJ)(A31 A24 ), (IJ)(A12 A34 ) are all
harmonic and cyclic.

Remark. In reality, M ∗ is the Poncelet point of (T1 , T2 , T3 , T4 ), which we will investigate in (TODO 8.1), so
it will lie on the cevian circle of (T1 , T2 , T3 , T4 ) and will lie on the pedal circle of Tl wrt. △Ti Tj Tk (which is
just the Simson line Ll ). So when we invert this question about ω, we can get that M lands on the pedal
circle of I wrt. △(A23 A14 )(A31 A24 )(A12 A34 ) and J(Lℓ ).

Example 4.3.6 (Buffed 2014 Taiwan TST 3 P3). Let M be a moving point on the circumcircle of △ABC.
Draw the two tangents from M to the incircle ω of △ABC. Let these two tangents intersect BC at X1 , X2 .
Prove that the circumcircle of M X1 X2 passes through a fixed point.

(Note: In the original problem we want to prove the fixed point is the A-mixtouch point.)

B C

136
AoPS Chapter 4. Complete Quadrilaterals - Hidden/Completed

Proof. Let Y1 , Y2 be the second intersections of M X1 , M X2 with the circumcircle of (ABC). By Poncelet’s
closure theorem, we have that Y1 Y2 is tangent to the incircle ω of △ABC. Thus Q = △M Y1 Y2 ∪ BC is a
complete tangential quadrilateral with incircle ω. Further we get the second intersection of (M X1 X2 ) and
(ABC) is the Miquel point of Q. Let N be the intersection of Y1 Y2 and BC.
∗
We now invert about the incircle. From (TODO 4.3.5) we get that IMQ , X1∗ Y2∗ , X2∗ Y1∗ , M ∗ N ∗ have a
common midpoint, let this be G. . Let s represent reflection across G. Then we have

s(Ω∗ ) = s((Y1∗ Y2∗ M ∗ )) = (X2∗ X1∗ N ∗ ) = (BC)∗ = (ID),

where Ω and D are the circumcircle and BC-intouch point respectively. This means that G is fixed as M
∗
varies, so MQ as the reflection of I over G is also fixed, and thus MQ is fixed.

If the reader is familar with mixtilinear incircles, then we can prove that MQ is the A-mixtouch point. Let
Ma be the second intersection of AI and Ω, then MQ is the A-mixtouch point if and only if ∡IMQ MA = 90◦ ,
however by inverting around ω we get this condition is equivalent to

∗
∠MQ MA∗ A∗ = ∠MQ
∗
MA∗ I = 90◦ ,

∗ is the diameter of Ω∗ . If △DEF be the intouch triangle of △ABC, then Ω∗ = (N ) is the

or that A∗ MQ I

nine-point circle of △DEF ! Let HI be the orthocenter of △DEF and note that OI = I is the circumcenter.
As such A∗ is the midpoint of EF . If N9′ is the center of (ID), then G is the midpoint of N9 N9′ so N9 IN9′ MQ
∗

∗
is a parallelogram. Taking a homothety of size 2 at I and using (TODO 1.1.5), it follows that MQ is the
midpoint of HI D, and thus it is the antipode of A∗ .

137
Chapter 5

The Best Of Xn - Hidden/Completed

For the sake of conciseness, we will use conventions a lot here, so review them if necessary. Here, Xn refers to
triangle centers from here.

5.1 X1

• X1 = I is the incenter, defined as the intersection point of the three internal angle bisectors of △ABC.
Intersecting any two external bisectors and the other internal bisector gets us three excenters, I a , I b , I c .
(Note that these are not triangle centers as they are not symmetric.)

Also recall that ω is the incircle and ωa , ωb , ωc are the A, B, C-excircles.

Proposition 5.1.1. Let D∗ be the antipode of D on ω. Then A, D∗, and Da are collinear.

−−−−−−→
Proof. Consider the homothety h at A that sends ω to ω a . Then h−1 (Da ) is on ω and satisfies h−1 (Da )I ∥
−− −→ −→
I a Da ∥ DI, so D∗ = h−1 (Da ) which is enough.

In the same vein, we have that Da , Db , Dc ’s antipodes on the three excircles, ω a , ω b , ω c will lie on AD,
ADc , ADb respectively. Also note that the midpoint of DDa is the midpoint of BC (denoted as Ma ). This
gives us that:

138
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

D∗

F
I

B D Ma Da C

Corollary 5.1.2. IMa ∥ ADa and IMa bisects AD.

Proposition 5.1.3. Let D∨ = EF ∩ BC. Then D∨ is the harmonic conjugate of D wrt B, C.

Proof. Menelaus. Specifically, we have

BD CE AF BD∨ CE AF
− · · = −1 = ∨ · ·
DC EA F B D C EA F B

so
BD/DC
(B, C : D, D∨ ) = = −1.
BD∨ /D∨ C

Alternatively, using harmonic properties on the cevian triangle of the Gergonne point suffices.

Proposition 5.1.4. Let the second intersection of the circle (AI) = (AEIF ) with Ω be S (in America, we
call this the A-Sharkydevil Point of △ABC). Then D, S, and Na are collinear.

Proof. Consider the inversion around (BIC) = (Na ). This sends the circle (AIEF ) to the circle with diameter
IX, where X = AI ∩ BC. Then the inverse of S is (IX) ∩ BC = D, so S, Na , and D are collinear.

139
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

E
Hd
F
I

B D C

Na

Proposition 5.1.5. Continuing notation from above, we have that IS goes through the foot from D to EF
(denoted as Hd ).

Proof. Let Hd′ = IS ∩ EF . Since the tangent at I to (AI) is parallel to F E and the tangent at Na is
+
parallel to BC, by spiral similarity we have that △SBC ∪ Na ∼ △SF E ∪ I. By the above claim, we get that
+ +
△SBC ∪ D ∼ △SF E ∪ Hd′ , so △SBF ∼ △SDHd′ . Thus

′
∡SNa A = ∡SBA = ∡SBF = ∡SDHD =⇒ DHd′ ∥ Na A ⊥ EF ,

which tells us that Hd′ = Hd .

a
Proposition 5.1.6. Let OI be the intersection of I b F b and I c E c (which is also the circumcenter of △II b I c
a
by angle chasing). Then A, Na∗ , Fb , Ec , OI are concyclic.

Proof. Note that △ABC is the orthic triangle of △IA IB IC , so (ABC) is the nine-point circle of △IA IB IC .
As such, it follows that Na∗ , which is the reflection of the midpoint of IIA over O, is the midpoint of IB IC .
Then,
a a
∡AFb OI = ∡AEb OI = ∡ANa∗ OIa = 90◦ .

Proposition 5.1.7. AS ∥ Fb Ec .

Proof. Using the above proof, we have that

a a
∡Fb EC + 90◦ = ∡AFb + ∡AEc − ∡AOI = 2 · IB IC − AOI = AOI

140
AoPS Chapter 5. The Best Of Xn - Hidden/Completed
a
where OI is the circumcenter of △I a I b I c (which is the reflection of OI over I b I c ).

By Proposition 5.2.4, it then follows that OI is the reflection of I over O, so AOI is parallel to A∗ I =
AS + 90◦ where A∗ is the A antipode.

Fb O I a

Ib

Ec
Na∗

Ic S
E

F
I

B D C

Na

Proposition 5.1.8. AMa , EF , and ID concur.

Proof 1. Let T be the intersection of ID and EF . Draw a line through T parallel to BC, denoted as T ∞BC .
Let this intersect sides CA and AB at CT , BT . Then the feet of I onto BT CT , CT A, ABT are collinear.
Thus by (TODO 1.4.1), we have that I lies on (ABT CT ). Further, since △ABT CT ∩ I is homothetic to
△ABC ∩ Na (with A as center of homothety), we have A, the midpoint T of BT CT , Ma are collinear.

Proof 2. Let T = ID ∩ EF , and draw a line A∞BC parallel to BC through A, and let it intersect EF at S.
Then S and T are both on the pole pω (A) of A about ω. Since IT ⊥ AS, pω (T ) = AS, so

pω
A(B, C; T, ∞BC ) = (F, E; S, T ) = −1.

Hence AT bisects BC.

141
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Ib

A
Q
E
Y
Ic Z
P I
F

B D C

Proposition 5.1.9. Let BI and CI intersect CA, AB at Y and Z respectively. Let the two intersections of
Y Z and Ω be P and Q. Then I, I b , I c , P , and Q are concyclic.

Proof. Y is on the radical axis of (ACP Q) and (ACII b ), so I, I b , P , and Q are concyclic. Similarly I c lies
on (IP Q).

Proposition 5.1.10. Let Ha be the foot from A to BC. Then

(Ha , AI ∩ BC; D, Da ) = −1.

Proof. From
(A, AI ∩ BC; I, I a ) = B(A, C; I, I a ) = −1,

we can project (A, C; I, I a ) through the point at infinity ∞⊥BC to get that

(Ha , AI ∩ BC; D, Da ) = −1.

Proposition 5.1.11. DI a bisects AHa .

142
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Proof.
(A, Ha ; ∞AHa , DI a ∩ AHa ) = D(A, AI ∩ BC; I, I a ) = −1.

In the rest of this section, let HA . HB , and HC be the orthocenters of △BIC, △CIA, and △AIB
respectively (this notation is not super standard).

Proposition 5.1.12. Point HA is the reflection of I a about M a .

Proof. We have (by line arguments) that

∡BHA = ∡CI + 90◦ = ∡CI a , ∡CHA = ∡BI + 90◦ = ∡BI a .

Kb
E
Mb
F
I

B D Ma C

Proposition 5.1.13 (Iran Lemma). Let Kb be the foot from B to CI, and similarily define Kc as the foot
from C to BI. Then Kb = EF ∩ Mc Ma , and Kc = EF ∩ Mb Ma similarly.

Proof. Since Ma is the circumcenter of (BKb Kc C), we have that

∡Ma Kb = ∡Kb B + ∡Kb C − ∡BC + 90◦ = 2 · ∡CI − ∡BC = ∡CA,

and because of this we have that Kb ∈ Mc Ma . Note that since B, I, Kb , F are concyclic, we have

∡IF Kb = ∡BIC + 90◦ = ∡IAE = ∡IF E,

143
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

and thus Kb ∈ EF . By the same logic we have that Kc = EF ∩ Ma Mb as well.

Proposition 5.1.14. We have

(D, ID ∩ EF ; I, HA ) = −1.

Proof. Let Kb DKc be the orthic triangle of △BIC. Then we have

Kb (D, ID ∩ EF ; I, HA ) = (D, D∨ ; B, C) = −1.

Combining this with the properties of cevian triangles (see (TODO 5.3.2)), we get:

Corollary 5.1.15. The cevian triangle of I with respect to △HA HB HC is the intouch triangle △DEF .

We can also characterize HA as follows:

Proposition 5.1.16. The polar pω (HA ) is the A-midline Mb Mc , or the image of HA about inversion by ω is
the foot from I to the A-midline.

Proof. Let T = ID ∩ EF as usual, then pω (T ) is A∞BC , the line parallel to BC through A. Since polarity
preserves cross-ratios, we have that

pω
(BC, A∞BC ; L∞ , pω (HA )) = (D, T ; I, HA ) = −1,

so pω (HA ) is the A-midline Mb Mc .

5.2 X2 through X5

• X2 is the centroid (TODO 0.3.3), the intersection of the three medians, typically represented as G;

• X3 is the circumcenter (TODO 0.1.9). the intersection of the three perpendicular bisectors, typically
written as O;

• X4 is the orthocenter ((TODO 0.1), problem 1), the intersection of the three altitudes, typically written
as H;

• X5 is the nine-point center, the center of ϵ, the nine-point circle (TODO 1.1.5), typically written as N9 .

The most important properties will be listed below.

144
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Proposition 5.2.1. O and H are isogonal conjugates.

Proposition 5.2.2. G, O, H, N all lie on the Euler line E, and

HG OG
= = 2, (G, H; O, N ) = −1.
GO GN

These two were proven in previous chapters.

The rest of this section will relate H and O to I, the incenter.

Proposition 5.2.3. The Euler line of the intouch triangle △DEF is OI.

Proof. This problem was given as an exercise in a previous chapter, but we will re-present the proof here.
Note that the Euler line of the excentral triangle, △I a I b I c is line OI, and furthermore note that △DEF
and △I a I b I c are homothetic, so the Euler lines of these two triangles are parallel. However, I clearly lies on
the Euler line of △DEF , so we are done.

(Note from translators: the center of homothety between the intouch and excentral triangles is X57 , the
isogonal conjugate of the Mittenpunkt.)

Proposition 5.2.4. The circumcircle Ω is the nine-point circle of all four of △I a I b I c , △II b I c , △I a II c , △I a I b I.

Proof. Note that I, I a , I b , I c make an orthocentric system and that △ABC is the orthic triangle wrt any of
the triangles.

Proposition 5.2.5. The trilinear polar of I with respect to △ABC, line t(I), is perpendicular to OI.
Further, define Y and Z as I c I a ∩ CA, I a I b ∩ AB. Then we have OI ⊥ Y Z.

Proof. From
Y I c · Y I a = Y C · Y A, ZI a · ZI b = ZA · ZB,

we have that Y Z is the radical axis of (ABC), (I a I b I c ), so it is perpendicular to the line connecting O and
the circumcenter of △I a I b I c . Further note that O is the nine-point center of △I a I b I c , so this line is just the
Euler line of the excentral triangle, which is just line OI.

Then it follows by harmonic properties that Y Z is also t(I) which finishes.

Proposition 5.2.6. Let U be the reflection of D across EF (where △DEF is the intouch triangle). Then
AU, BC, IO are concurrent.

Proof. Let D∗ be the antipode of D wrt. ω, and let HI be the orthocenter of △DEF . Let HI′ be the
reflection of HI across EF . Then note that HI′ lies on ω. Let the foot from HI to BC be point X. Then

145
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

∡XHI D = ∡D∗ DHI′ = −∡HI′ DD∗ , ∡DXHI = 90◦ = −∡D∗ HI′ D,

−
tells us that △HI XD ∼ △DHI′ D ∗ . Thus if we let Md be the midpoint of EF , by

HI X HI X DHI IMd ID
= = ∗ = = ,
HI U DHI′ D D ID∗ IA

we get that △IAD and △HI U X are homothetic. Combining (TODO 5.2.3) with this result, we get that
IHD = IO, AU, DX = BC are concurrent.

Proposition 5.2.7. I, H are isogonal conjugates in the orthic triangle △Hd He Hf of the intouch triangle
△DEF .

Proof. By symmetry, all we need to prove is that

∡(Hd I) + ∡(Hd H) = ∡(Hd He ) + ∡(Hd Hf ) = 2 · ∡EF.

By the proof of (TODO 5.1.5), we have that the complete pentagon △ABC ∪ EF ∪ DHd has a Miquel
point S. Thus by (TODO 4.1.6) we have that the orthocenters of △AEF , △ABC, and Hd are collinear.
Note that the orthocenter of △AEF is the reflection of I across EF (let this be I ′ ). Thus

∡(Hd I) + ∡(Hd H) = ∡(Hd I) + ∡(Hd I ′ ) = 2 · ∡EF.

Proposition 5.2.8. The perspector of the triangle formed by E a F a , F b Db , Dc E c and △ABC is H. Fur-
thermore, H is the circumcenter of this triangle. (We will denote this triangle as △′I .)

Proof. Note that F b Db ⊥ BI, Dc E c ⊥ CI.

By symmetry, all we need to prove is that the corresponding sides of △AF b E c and △IBC are orthologic, or
that I∞⊥F b E c , BA∗ = B∞⊥E c A , CA∗ = C∞⊥AF b are concurrent. By (TODO 5.1.7), F b E c is perpendicular
to IA∗ , so the concurrence point of the three lines listed above is just A∗.

Then it follows that the foot from A to BC passes through F b ∞⊥BI ∩ E c ∞⊥CI = F b Db ∩ Dc E c as
desired.

146
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

5.3 X6

• X6 is the symmedian point (TODO 2.2.16), the isogonal conjugate of G.

Proposition 5.3.1. AK bisects Hb Hc .

−
Proof. Let MHa be the midpoint of Hb Hc . Since △ABC ∼ △AHb Hc , we have

∡Hb AMHa = −∡BAMa = ∡CAK,

so A, MHa , K collinear.

Proposition 5.3.2. Let the three triangles △Ka Kb Kc , △KA KB KC , △K a K b K c respectively be the K-cevian
triangle, circumcevian triangle, and anticevian triangle. Then

(A, Ka ; K, KA ) = (B, Kb ; K, KB ) = (C, Kc ; K, KC ) = −2,

(A, KA ; Ka , K a ) = (B, KB ; Kb , K b ) = (C, KC ; Kc , K c ) = −1.

Proof. We have

B
(A, Ka ; K, KA ) = (A, C; KB , KA ) = (A, C; KB , B) · (A, C; B, KA ) = (−1) · 2 = −2,

and similarly for the rest. For the harmonic one we can note that △K a K b K c is the tangential triangle of
△ABC, so we get
B
(A, KA ; Ka , K a ) = (A, KA ; C, B) = −1,

and similarly for the rest.

Proposition 5.3.3. Ma K bisects segment AHa .

Proof. Let Ta be the intersection point of the tangents at B, C to (ABC), then A, K, Ta are collinear and
Ma Ta is perpendicular to BC. Thus

M B
(A, Ha ; Ma K ∩ AHa , ∞AHa ) =a (A, AK ∩ BC; K, Ta ) = (A, C; BK ∩ (ABC), B) = −1,

so Ma K bisects AHa .

Proposition 5.3.4. The trilinear polar t(X) of any point X on the circumcircle will always pass through K.

147
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Proof. Let △Xa Xb Xc be the cevian triangle of X with respect to △ABC. Construct Xb∨ and Xc∨ on lines
CA, AB such that
(C, A; Xb , Xb∨ ) = (A, B; Xc , Xc∨ ) = −1

(so just the harmonic conjugates of Xb and Xc ).

Note that t(X) is then just Xb∨ Xc∨ . From

X
(C, A; B, XXb∨ ∩ Ω = (C, A; Xb , Xb∨ ) = −1 = (C, A; B, BK ∩ Ω),

we can get KB := XXb∨ ∩ BK ∈ Ω, and similarly we get KC := XXb∨ ∩ BC ∈ Ω. Now we insta-kill by Pascal
on ABKB XKC C to get that K, Xb∨ , Xc∨ are collinear, so K ∈ t(X).

Proposition 5.3.5. Given any point P , let △PA PB PC be its circumcevian triangle, and let this triangle
have symmedian point KP . Then KP , K, P collinear.

Proof. Invert at P with power PowΩ (P ), call this transformation J. Then we have that

J J J
A 7−
→ PA , B 7−
→ PB , C 7−
→ PC .

Since inversion preserves cross-ratios, we get

(PB , PC ; PA , J(KA )) = (B, C; A, KA ) = −1.

so J(KA ) = PA KP ∩ Ω =: KPA . Let Ka be the intersection of AKA and BC, then we get that J(Ka )
is the second intersection of (P PA KPA ) and (P PB PC ). By (TODO radical axis theorem), we get that
P J(Ka ), PA KPA , PB PC concurrent, and thus Ka , P, PA KP ∩ PB PC are collinear. By ((TODO 5.3.2)), we get

P
(PA , PA KP ∩ PB PC ; P K ∩ PA KP , KPA ) = (A, Ka ; K, KA ) = −2 = (PA , PA KP ∩ PB PC ; KP , KPA ),

and thus K, KP , P are collinear.

5.4 X7 through X10

• X7 is the Gergonne point (see (TODO 0.3) problem 4), defined as the perspector of the intouch triangle
and the reference triangle, commonly denoted as Ge;

• X8 is the Nagel point (see (TODO 0.3) problem 5), defined as the perspector of the extouch triangle
and the reference triangle, commonly denoted as N a;

148
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

• X9 is the Mittenpunkt, defined as the perspector of the excentral triangle and the medial triangle,
commonly denoted as M t;

• X10 is the Spieker center, the incenter of the medial triangle, commonly denoted as Sp.

Proposition 5.4.1. The complement of the Gergonne point is the Mittenpunkt.

Proof. By an A-version of (TODO 5.1.2), we get that Ma M t = Ma I a ∥ AGe. Symmetrically we get that
+
Mb M t ∥ BGe, Mc M t ∥ CGe. Thus we get that △ABC ∩ Ge ∼ △Ma Mb Mc ∩ M t, so M t = Ge∁ .

Proposition 5.4.2. The complements of N a, I are I, Sp, respectively. Also

(G, N a; I, Sp) = −1

on the Nagel line, and HN a ∥ OI ∥ N Sp.

Proof. From (TODO 5.1.2), we get Ma I ∥ AN a. Similarly we have Mb I ∥ BN a, Mc I ∥ CN a. Thus we get

+
that △ABC ∩ N a ∩ I ∼ △Ma Mb Mc ∩ I ∩ Sp, and thus I = N a∁ and Sp = I ∁ . The harmonic is trivial by
the ratios from the definition of the complement.

Proposition 5.4.3. M t is the symmedian point of △I a I b I c .

Proof. Since △ABC is the orthic triangle of △I a I b I c , by (TODO 5.3.1), we have that the symmedian point
of the excentral triangle X satisfies

X = I a Ma ∩ I b Mb ∩ I c Mc = M t.

Proposition 5.4.4. The line HA Ge bisects segment EF .

Proof. We use the same notation as in (TODO 5.1.14). Let AD ∩ EF at point U . Let the midpoint of EF
be Md . Then we get that

E
(D, EF ∩ BC; B, C) = Md (D, U ; A, Ge) = −1 = Md (D, S; I, HA ),

so Ge, Md , HA collinear.

Proposition 5.4.5. Line AI is the complement of HA N a.

149
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Proof. Let X be the reflection of N a across Ma . Then G is the centroid of AXNa . Further, we get that I is
the midpoint of AX. Note that since the reflection of HA across Ma is I a , we can get that HA N a ∥ I a X = AI.
Then, we can then use the fact that I is the complement of N a to tell us that AI = (HA N a)∁ .

Proposition 5.4.6. I, K, M t are collinear.

Proof. By (TODO 5.3.5), we get that IK passes through the symmedian point of the arc-midpoint triangle
(circumcevian triangle of I). Let this point be KI . Then take the homothety at I sending △I a I b I c to
△Na Nb Nc . Since M t is the symmedian point of △I a I b I c , we get that I, KI , M t are collinear, and thus
I, K, M t are collinear.

150
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Proposition 5.4.7. The triangle made by lines Ea Fa , Fb Db , Dc Ec is homothetic to △I a I b I c with homothety

center M t. Furthermore, the ratio of homothety is − 2R+r
2R , where R, r represent the circum- and inradius

respectively.

D1′

Fb

Ib

Ec

Ic
E
Fc

Eb
F Mt

Db
D Da
Dc C
B
F1′

Ea

Fa

E1′ Ia

Proof. Let X = Fb Db ∩ Dc Ec = D1′ . Then all we need to prove is that X, I a , Ma are collinear, but this is
because the center of homothety between △XDb Dc and △I a BC is Ma (note that the midpoint of Db Dc is
Ma ).

To prove this we first observe that by (TODO 5.2.8), we have XHA = I a Ma ∥ AD, and thus AX ∥ DHA .

151
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Let OI be the circumcenter of △I a I b I c . Then the quadrilateral I a OI Na Na∗ is actually a parallelogram. Thus

−−→ −−→ −−→ −→ −−→ −−−→

XH = XA + AH = (ID − IHA ) + 2 · OMa
−−−−→ −−−→
= r − 2 · Na Ma + 2 · OMa

= r + 2R,
−− −→ −−−→
I a OI = Na Na∗ = −2R.

and we get that the ratio of homothety is − 2R+r

2R , by comparison of circumradii.

Proposition 5.4.8 (Mittenpunkt Line). H, M t, Sp are collinear.

Proof. See (TODO 5.6.8).

5.5 X11

• X11 is the Feuerbach point (see (TODO 3.3.4), defined as the tangency point of the incircle ω and the
nine-point circle ϵ, commonly denoted as F e.

Proposition 5.5.1. The three points I, N, F e are collinear.

Proof. Trivial.

152
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Proposition 5.5.2. Let X, Y, Z respectively be the reflections of D, E, F (intouch) across the perpendicular
bisectors of EF , F D, DE. (which are just the three angle bisectors of △ABC.) Then F e is the homothety
center of △XY Z and Ma Mb Mc .

Fe

E
F′ E′
Mb
Mc

D′
C

B D Ma

Proof. We have
Y Z = ((F + D − E) + (D + E − F ))ω = 2 · Dω = ∡Mb Mc .

By symmetry, we get that △XY Z and △Ma Mb Mc are homothetic. What is left is the proof by construction
of the existence of F e (See (TODO 3.3.4)).

+
Proposition 5.5.3. We have △AOI ∼ △F eMa D

Proof. Let X be the reflection of D over the perpendicular bisector of EF . Then we get F e, X, Ma are
−
collinear by (TODO 5.5.2), and further, by Ma D tangent to ω we get △F eMa D ∼ △DMa X. Thus our
+
original problem reduces into proving △AA∗ I ∼ △DD′ X.

Notice that Na (the intersection of XD’s perpendicular bisector with DD′ ’s perpendicular bisector) is the

153
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

circumcenter of △DD′ X, and from the fact that A∗ I intersects Na D on Ω by (TODO 5.1.4), we know that

∡XD′ D = ∡ANa D = ∡AA∗ I = −∡IA∗ A.

and also
∡D′ DX =⊥ AI − ∡BC = ∡HAI = ∡IAO = −∡A∗ AI.

+
Hence △AA∗ I ∼ △DD′ X, and we are done.

Corollary 5.5.4. F e is the exsimilicenter of ϵ and ω.

Proof. From (TODO 5.5.2) and (TODO 5.5.3), we have

 2
F eX Ma X Ma D
=1− =1−
F eMa Ma F e Ma F e

This just proves that F e is not the insimilicenter. By a similar proof this can tell us that the three
extraversions of the Feuerbach point F ea , F eb , F ec respectively are the insimilicenters of ω a , ω b , ω c and ϵ.

Corollary 5.5.5. F e is the anti-Steiner point of OI with respect to △Ma Mb Mc and △DEF .

Proof. By (TODO 5.5.3), we have that

(Ma + Mb + Mc − F e)ε = (Ma + Mb + Mc − Ha )ε + ∡Ha Ma F e

= ∡TMa ε + ∡AOI =⊥ OI,

(D + E + F − F e)ω = (E + F )ω + ∡F eDMa

=⊥ AI + ∡AIO =⊥ OI.

Thus we know that F e is the anti-Steiner point of OI with respect to both of these triangles. Further, by
(TODO 1.4.5) we have that the Steiner line of F e with respect to △DEF is parallel to OI. Also note that
OI is the Euler line of △DEF (since O, I of △ABC become H, O of △DEF ).

Alternatively, the result follows by (TODO 8.1.20) and (TODO 8.1.21).

Proposition 5.5.6. JΩ, (I)F e is parallel to the Euler Line.

154
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Proof. We length-bash.
R
OJΩ, (I) R2 R2 2 N9 F e
= 2 = 2 = R
=
OI OI R − 2Rr 2 −r N9 I

5.6 Xn , n < 99

5.6.1 X20

• X20 is the de Longchamps point, the reflection of H over O. For this section, we will call it L.
Obviously, it lies on the Euler line.

Proposition 5.6.1.
(O, H; G, L) = (N9 , O; G, H) = −1.

Proof. Use length relations between all these points.

Proposition 5.6.2. Let La be AL ∩ BC. Then OLa bisects AHa .

Proof. Harmonics!

O A
(A, Ha ; OLa ∩ AHa , ∞AHa ) = (AO ∩ BC, Ha ; La , Ma ) = (O, H; L, G) = −1

Proposition 5.6.3. The three points I, Ge, L are collinear, and have length ratios of

IGe r
=− .
GeL 4R + 2r

Proof. We take the complements of all 3 points. We have Sp = I ∁ , M t = Ge∁ , H = L∁ , which are shown to
be collinear in (TODO 5.6.8). To get the length ratio,

HM t 2R + r HSp IGe SpM t r

= , = 1 =⇒ = =− .
M tBe 2R SpBe GeL M tH 4R + 2r

155
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

5.6.2 X21

Proposition 5.6.4. The Euler lines of triangles △ABC, △IBC, △AIC, △ABI all concur in a common
point. Denote this point as Sc = X21 , the Schiffler point. Additionally, we have

GSc 2r
= ,
ScO 3R

where r and R represent the inradius and circumradius as usual.

Proof. Let E, EA respectively be the Euler lines of △ABC and △IBC. Let XA = E ∩ EA , and symmetrically
define EB , EC , XB , XC . We directly go for
GXA 2r
= ,
XA O 3R

since if we prove this we’re done by symmetry.

Let Y, Z respectively be the reflections of I, Na over BC. Since HA is the reflection of Ma over I a , we
can apply the A-extraversion of theorem (TODO 5.1.2) to get that

+ +
△HA Ma Z ∼ △I a Ma Na ∼ △ADI.

If we define P as the intersection point of the perpendicular line to BC drawn from G and the line EA ,
and let GA be the centroid of △IBC, then this tells us that

+ +
△GA P G ∼ △HA Na Z ∼ △AY I

by AA similarity. Thus,
GXA GP GA G IY 2r
= = · = .
XA O Na O AI Na O 3R

(There is also a very elegant proof of existence with Pascal on the Feuerbach hyperbola, which will be
shown in later chapters.)

Proposition 5.6.5. The cevian triangle of Sc, △Sca Scb Scc is perspective with △I a I b I c , with perspector O.

Proof. Let X = ASc ∩ OMa . By Menelaus we have

Ma X Ma A GSc r
=− · = .
XO AG ScO R

Thus,
Sc r/R DI
(Y = AI ∩ BC, AI ∩ OSca ; A, Na ) =a (Ma , O; X, Na ) = = .
Ma Na /N aO Ma Na

156
AoPS Chapter 5. The Best Of Xn - Hidden/Completed
+
We can get that △ADY ∪ I ∼ △I a Ma Y ∪ Na . Then by inverting through (IBC), it follows that

DI AY (IBC)
= a = (A, I a ; Y, ∞AI ) = (Y, I a ; A, Na )
M a Na I Y

5.6.3 X40

• X40 is the Bevan point, defined as the reflection of I over O, typically written as Be.

This sounds like a very useless point, but it shows up a lot more often than you expect. For example, we have

Proposition 5.6.6. Be is the circumcenter of the excentral triangle (△I a I b I c ).

Proof. Since Na O is perpendicular to BC, by 2x homothety at I we can get that I a Be ⊥ BC. Since the
excentral triangle is homothetic to the intouch triangle, we get that the circumcenter of the excentral triangle
OI satisfies I a OI ∥ DI ∥ I a Be. It follows then by symmetry that OI = Be.

Proposition 5.6.7. Be is the midpoint of N aL.

Proof. Since    
GN a IBe OL 2 3
· · = · (−2) · = −1,
N aI BeO LG 3 4

by Menelaus we get N a, Be, L are collinear. By Menelaus again we get that

N aBe N aI GO 1
=− · = −(−3) · = 1,
BeL IG OL 3

and we’re done.

Proposition 5.6.8. H, M t, Sp, Be are collinear, and

HM t 2R + r HSp
= , = 1.
M tBe 2R SpBe

Proof. Define △′I to be the triangle referenced in (TODO 5.4.7). By (TODO 5.2.8), we have that H is the
circumcenter of △′I . We also know that Be is the circumcenter of △I a I b I c , so by homothety we get that
HBe passes through their center of homothety, which is just M t by (TODO 5.4.7). We also have that

HM t 2R + r
= .
M tBe 2R

157
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Since
H
(O, ∞OI ; I, HSp ∩ OI) = (G, N a; I, Sp) = −1,

HSp passes through the reflection of I about O, which is just the Bevan point. Thus H, M t, Sp, Be
collinear.

5.6.4 X54

• X54 is the Kosnita point, the isogonal conjugate of the nine-point center N9 . This is typically denoted
as Ko or as N9∗ .

Proposition 5.6.9. Let OOA be the circumcenter of △OBC. Then A, Ko, OOA are collinear.

−
Proof. Note that △ABC ∼ △AHb Hc and the circumcenter of △AHb Hc is the midpoint of AHa (so it lies on
−
the nine-point circle ϵ). Thus △ABC ∪ OOA ∼ △AHb Hc ∪ N. It follows then that AOOA , AN are isogonal
which implies the result.

Proposition 5.6.10. Let B ′ , C ′ respectively be the reflections of B, C across CA and AB. Then AKo ⊥
B′C ′.

Proof. Let X be the reflection of the circumcenter O across point A. Let the tangents to the circumcircle at
B and C intersect at point D. Then all we need to prove is XD ⊥ B ′ C ′ . Note that

B′C XO C ′B
= = , ∡B ′ CD = ∡XOD = ∡C ′ BD,
CD OD BD
+
and thus D is the spiral center of similarity sending △B ′ XC ′ ∼ △COB, so we’re done.

Proposition 5.6.11. The Schiffler point Sc is the Kosnita point of △Na Nb Nc .

Proof. By symmetry, we only need to prove that △IBC’s Euler line EA goes through Ko(△Na Nb Nc ). Let
GA be the centroid of △IBC, and let G′A be the reflection of I across GA . Then we have that the condition
of EA = Na Ga passing through the Kosnita point of △Na Nb Nc is equivalent to the condition of I a G′A passing
through KoI = Ko(△I a I b I c ). Let J be the reflection of I with respect to Ma , and note that J is thus the
orthocenter of △I a BC. Then we have

(I, G′A ; Ma , J) = −1 = (Be = OI , N I ; GI , I = H I ).

Since (I a I, I a Be), (I a Ma , I a GI ), (I a J, I a I) all are pairs of isogonal lines in ∠I b I a I c , we have that

(I a G′A , I a N I ) are also isogonal lines. So we have that KoI ∈ I a G′A .

158
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Corollary 5.6.12. The Euler line of the orthic triangle △Ha Hb Hc is parallel to OKo.

5.6.5 X55 , X56

• X55 is the insimilicenter of Ω and ω,

• X56 is the exsimilicenter of Ω and ω.

Proposition 5.6.13. I, O, X55 , X56 are collinear, and

(I, O; X55 , X56 ) = −1.

Proof. See definition.

Proposition 5.6.14. X55 is the isogonal conjugate of Ge, and X56 is the isogonal conjugate of N a. (Thus
we will denote them as Ge∗ and N a∗ .)

Proof. Let X, Y, Z respectively be the reflections of D, E, F across AI, BI, CI. Then (suppose △ABC is
labeled counterclockwise), we have that

−→ −
→ −→ −−→ −→ −→
XI = 2 · AI − DI = AB + AC − ID + 180◦
−→ −→
= AO + 180◦ = −AO,

−→ −−→ −
→ −−→
Similarily we have that Y I = −BO, ZI = −CO. Thus △ABC and △XY Z are homothetic, with a negative
scale factor, which literally just means “insimilicenter.” Since (ABC) is the circumcenter and (XY Z) is the
incircle, as
∡GeAI = ∡DAI = ∡IAX = ∡IAX55

we get that X55 is the isogonal conjugate of Ge. We can do a similar argument for X56 , except we use
the antipodes of X, Y, Z.
√
Remark (A suprising connection with mixtilinear incircles). By considering a bc inversion, the A-mixtilinear
incircle is sent to the A-excircle. Let TA be the point in which the A-mixtilinear incircle touches the circumcircle
√
(ABC); then the image of TA after a bc inversion is the A-extouch point. In other words, ATA , BTB , CTC
concur at X56 . (A similar concurrency with the mixtilinear excircles is at X55 .)

Proposition 5.6.15. G, F e, Ge∗ are collinear, and H, F e, N a∗ are collinear.

Proof. Monge on incircle, circumcircle, and nine-point circle.

159
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

5.6.6 X65

• X65 is the orthocenter of the intouch triangle.

Proposition 5.6.16. X65 is the isogonal conjugate of the Schiffler point, denoted as Sc∗ .

Proof: By (TODO 5.6.5), ASc, OI a , BC are concurrent. Applying an extraverted version (TODO 5.2.6) on
A, we get A(Da )′ , OI a , BC are concurrent, where ′ represents reflection over E a F a . Thus we just need to prove
+
that A(Da )′ , AX65 are isogonal conjugates in ∠BAC. But this is because △AEF ∩ X65 ∼ △AF A E a ∩ D′ .
Thus we represent X65 as Sc∗ .

Proposition 5.6.17. Point Sc∗ lies on OI, with

ScI r
=
IO R

+ Sc∗ I ID r
Proof. We have △DEF ∪ I ∪ Sc∗ ∼ △Na Nb Nc ∩ O ∩ I, thus I, Sc∗ , O concurrent and IO = ONa = R.

Proposition 5.6.18. O is the midpoint of Ge∗ and Sc∗ , further, we have,

(I, Be; Ge∗ , Sc∗ ) = −1.

ScO r+R IO
Proof. IO = R = Ge∗ O

Proposition 5.6.19.
GeN a∗ ∩ Ge∗ N a = Sc;

GeN a ∩ Ge∗ N a∗ = Sc∗ .

Proof. Since
GSc OGe∗ IN a
 
2r R 3
· · = · · − = −1,
ScO Ge∗ I N aG 3R r 2

by Menelaus’s theorem, Sc, Ge∗ , N a are collinear. Similarly, by (TODO 5.6.3), we have

LSc ON a∗ IGe
     
12R + 6r R r
· · = − · − · − = −1,
ScO N a∗ I GeL 3R r 4R + 2r

and thus Sc, N a∗ , Ge are similarly collinear. By (TODO 7.4.5), a theorem on isoconjugation, we easily have
that Sc∗ = GeN a ∩ Ge∗ N a∗ , but we can also give an elementary proof with previous material; we already
know that OI = Ge∗ N a∗ , so by (TODO 5.6.3), (TODO 5.6.7), (TODO 5.6.17) combined, we get that

LSc ON a∗ IGe
   
12R + 6r R r
· · = − · (− ) · − = −1
ScO N a∗ I GeL 3R r 4R + 2r

160
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

and thus we get Sc, N a∗ , Ge are collinear by Menelaus.

Proposition 5.6.20. Line SpSc∗ is the complement of ISc.

Proof. Since I ∁ = Sp, we just need to prove that Sc ∗ Sc∁ is parallel to ISc. By (TODO 5.6.4) and (TODO
5.6.17),
Sc ∗ I r 3 GSc Sc∁ Sc
= = · = ,
IO R 2 ScO ScO

so Sc ∗ Sc∁ ∥ ISc.

5.6.7 X69
∁
• X69 is the anticomplement K of the symmedian point.

Proposition 5.6.21. X69 is the isotomic conjugate of H.

Proof. Let the isotomic conjugate of H be H ′ . Let M be the midpoint of AHA . Then by (TODO 5.3.3), we
∁ ∁
get that A, H ′ , M are collinear, and we can check that M lies on BC. Furthermore,
∁
BM Mb M CHa
∁ = =
M C M Mc Ha B
∁
tells us that M is the reflection of Ha across the midpoint of BC, and thus A, X69 , H ′ are collinear. Now
note that every step in our proof can be done with a different vertex, so we can apply it cyclically. We then
have
H ′ = AX69 ∩ BX69 ∩ CX69 = X69 .

Proposition 5.6.22. Ge, N a, X69 are collinear.

Proof. We take the complement. Then this collinearity is equivalent to proving M t = Ge∁ , I = N a∁ , K =
(X69 )∁ are collinear, which finishes by (TODO 5.4.6). Alternatively, we can prove that

B(N a, H ′ ; Ge, C) = C(N a, H ′ ; Ge, B).

Let B ′ be the reflection of B about Nb . Since Nb is the midpoint of II b , we have

B(N a, H ′ ; Ge, C) = (E ′ , BH ′ ∩ CA; E, C)

= (I b , B ′ ; I, ∞⊥CA C ∩ BI)

= (I, B; I b , ∞⊥CA A ∩ BI),

161
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

and similarly we get that

C(N a, H ′ ; Ge, B) = (I, C; I c , ∞⊥AB A ∩ CI).

Thus we only need to prove that BC, I b I c , (∞⊥CA A ∩ BI))(∞⊥AB A ∩ CI)) are concurrent. This is equivalent
to proving that
I b I c ∩ BC, (∞⊥CA A ∩ BI)), (∞⊥AB A ∩ CI))

are collinear, but this is exactly the orthotransversal of A with respect to △BIC!!

5.7 Xn , n ≥ 99

This will be completed later. (This is what the original book says.)

5.8 Others

5.8.1 X19

Definition 5.8.1. Let ℓa be the common internal tangent to ω and ω a that is not line BC. Symmetrically,
define ℓb and ℓc . Let ℓ′a be the common external tangent to ω b and ω c that is not line BC. Again, symmetrically
define ℓ′b and ℓ′c . We denote △ℓa ℓb ℓc as the intangents triangle(△Ta Tb Tc ) and △ℓ′a ℓ′b ℓ′c as the extangents
triangle(△Ta′ Tb′ Tc′ ).

Obviously, the intangents triangle’s three sides Tb Tc , Tc Ta , Ta Tb are the reflections of BC over II a , etc.
Similarly, the extangents triangle’s three sides Tb′ Tc′ , Tc′ Ta′ , Ta′ Tb′ are the reflections of BC over I b I c , etc.

Proposition 5.8.2. The orthic triangle △Ha Hb Hc is homothetic to both the intangents triangle △Ta Tb Tc
and the extangents triangle △Ta Tb Tc (note that the homothety centers are different).

Proof. We proceed with line arguments: we want to prove that ∡(Hb Hc ) = ∡(ℓa ) = ∡(ℓ′a ). Since

∡Hb Hc = ∡AB + ∡AC − ∡BC,

∡ℓa = 2 · ∡II a − ∡BC = ∡AB + ∡AC − ∡BC,

∡ℓ′a = 2 · ∡I b I c − ∡BC = ∡AB + ∡AC − ∡BC,

and we get the desired homotheties.

162
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Now let’s look at the two centers of homothety:

• X19 is the Clawson point, defined as the homothety center between the extangents triangle and the
orthic triangle, typically written as Cl;

• X33 is the homothety center between the intangents triangle and the orthic triangle.

Proposition 5.8.3. Ta I ⊥ BC, and Ta I bisects ∠Tb Ta Tc . Further,Ta′ I a ⊥ BC, and it bisects ∠Tb′ Ta′ Tc′ .

Proof. First we characterize Ta′ . Let Y ′ , Z ′ respectively be the intersections of Tc′ Ta′ , Ta′ Tb′ with BC. Then

Y ′ D′ = AFa = Ea A = D′ Z ′ ,

so D′ is the midpoint of Y ′ Z ′ . This tells us that I a lies on the perpendicular bisector of Y ′ Z ′ , and also

Ta′ Y ′ + Ta′ Z ′ = (2I c I a − CA) + (2I a I b − AB) = 2BC = 2Y ′ Z ′

tells us that Ta′ also lies on this perpendicular bisector. Thus we get that Ta′ I a is perpendicular to BC and
that it bisects ∠Tb′ Ta′ Tc′ = ∠Y ′ Ta′ Z ′ .

The proof for Ta is analogous.

Note that I a ∞⊥BC , I b ∞⊥CA , I c ∞⊥AB concur at Be, and thus we get

Corollary 5.8.4.
+ +
△Ha Hb Hc ∩ H ∼ △Ta Tb Tc ∩ I ∼ △Ta′ Tb′ Tc′ ∩ Be.

Notably, this gets us that the Clawson point lies on the (TODO 5.6.8) (H, M t, Sp, Cl, Be collinear), and
that H, I, X33 are collinear.

Proposition 5.8.5. The circle centered at Be tangent to the three sides of △Ta′ Tb′ Tc′ exists (call it ωBe ),
and has radius 2R + r.

Proof. Let DBe be the tangency point of Tb′ Tc′ to ωBe , and let Bea be the reflection of Be over I b I c . Then
by the A-version of (TODO 5.6.6), we have that Bea D ⊥ BC, so we can then reflect BeDBe ⊥ Tb′ Tc′ over
I b I c to get that D is the reflection of DBe over I b I c . Thus,

−−−→ −→ −−→ −→
BeDBe = Bea D = |Bea I + ID| = |2ONa + ID| = 2R + r.

−→ −−−−→ −−→
Note additionally that this also gets us that OA ∥ BeDBe ∥ −IDI , where DI is the tangency point of ω
and Tb Tc .

163
AoPS Chapter 5. The Best Of Xn - Hidden/Completed

Lemma 5.8.6. Let Y , Z, Y ′ , Z ′ respectively be the intersection points of Tc Ta , Ta Tb , Tc′ Ta′ , Ta′ Tb′ with BC.
Then the circumcenter of △AY Z ′ is I b , and the circumcenter of △AY ′ Z is I c .

Proof. This is because Z ′ A and AY have I a I b and II b respectively as perpendicular bisectors, so they concur
at I b . (Similarly for △AY ′ Z.)

Lemma 5.8.7. The intangents triangle △Ta Tb Tc , the extangents triangle △Ta′ Tb′ Tc′ , the incentral triangle
(cevian triangle of I) △Xa Xb Xc , and the tangential triangle △TA TB Tc , are all perspective. Furthermore,
their perspector is Ge∗ , the isogonal conjugate of Ge.

Proof. We have that

+
△TA TB TC ∪ O ∪ △ABC ∼ △Ta Tb Tc ∪ I ∪ △DI EI FI
+
∼ △Ta′ Tb′ Tc′ ∪ Be ∪ △DBe EBe FBe .

Additionally,
−→ −−→
Ge∗ O R OA Ge∗ I −r IDI
= = −−−−→ , = = −−−−→
Ge∗ Be 2R + r BeDBe Ge∗ Be 2R + r BeDBe
tells us that Ge∗ ∈ OI as the homothetic center of △TA TB Tc , △Ta Tb Tc , △Ta′ Tb′ Tc′ .

By (TODO 5.8.6), we have that Xa lies on the radical axis of (AY Z ′ ) and (AY ′ Z), which is just A∞⊥I b I c ,
and thus

Xa Y Xa Y ′
Xa Y · Xa Z ′ = Xa Y ′ · Xa Z =⇒ =
Xa Z Xa Z ′

+
combined with the fact that △Ta Y Z ∼ △Ta′ Y ′ Z ′ gives us that Ta , Ta′ , Xa are collinear, and thus we get
that Ge∗ is the perspector of △Ta Tb Tc , △Ta′ Tb′ Tc′ , △Xa Xb Xc .

Since X25 is the homothetic center between △TA TB TC and the orthic triangle △Ha Hb Hc , by (TODO
1.1.13) we have that

Corollary 5.8.8. Cl, X25 , X33 , Ge∗ collinear.

164
 Part II

The Deep End

165
Chapter 6

Basic Conic Theory

6.1 Definitions and Basic Properties

We first give a common definition for a conic:

Definition 6.1.1. Let Ω ∈ R3 be a circle with center O, and ℓ be the line through O perpendicular to the
plane containing Ω. For any point V ̸= O on ℓ, the surface S formed by the union of all straight lines through
V and some point M ∈ Ω is a cone. Ω is the directrix of S and V is the vertex of S.

Definition 6.1.2. On plane E, a curve C is a conic if there exists a circular conical surface S with vertex
V ∈
/ E such that C = S ∩ E.

Remark (Technical Details). This defines a conic in the Euclidean plane. If we want to define a conic on a
projective plane E instead, we need to first define the circular conical surface S in P3 := P3R which contains S
as well as the points of infinity on all possible lines V M . Then a conic on E will be S ∩ E.

In what follows, everything happens on the real plane E and in R3 (unless the space is mentioned).
Conics can be extended to spaces like CP2 , but things like foci and classical definitions do not exist nicely in
generality.

Definition 6.1.3. Let C be a conic and L∞ be the (real) line at infinity. Then C is

(i) an ellipse if |C ∩ L∞ | = 0;

(ii) a parabola if |C ∩ L∞ | = 1;

(iii) a hyperbola if |C ∩ L∞ | = 2.

Here are some alternative characterizations for each of these types of conics.

166
AoPS Chapter 6. Basic Conic Theory

Proposition 6.1.4. Let C be a conic. Then

1. C is an ellipse if and only if there exists points F1 , F2 and a > F1 F2 /2 such that

C = {P | F1 P + F2 P = 2a};

2. C is a parabola if and only if there exists point F and line L such that

C = {P | F P = d(L, P )};

3. C is a hyperbola if and only if there exists points F1 , F2 and a < F1 F2 /2 such that

C = {P | |F1 P − F2 P | = 2a},

here F1 , F2 or F are called the foci (plural of focus) of C, while L is the directrix of C (if applicable).

Proof. We will only prove the ellipse case, the other cases are similar.

By definition of conic, there exists circular conical surface S and plane E such that C = S ∩ E. Let B1 , B2
be spheres tangent to both S and E, and let ω1 = S ∩ B1 , ω2 = S ∩ B2 , F1 = E ∩ B1 , F2 = E ∩ B2 . For any
point P ∈ S, let A1 , A2 be the intersections of P V with ω1 and ω2 respectively, then clearly the length A1 A2
is fixed. Take a = A1 A2 /2, then

F1 P + F2 P = A1 P + A2 P = A1 A2 = 2a

so C ⊆ {P | F1 P + F2 P = 2a}. If P ∈ E \ C satisfies F1 P + F2 P = 2a, then if ray F1 P intersects S at P ′ ∈ C,

then
F1 P + F2 P = 2a = F1 P ′ + F2 P ′ =⇒ F2 P = F2 P ′ ± P P ′

so F2 , P, P ′ are collinear with P and P ′ on the same side of F2 , contradiction, hence C = {P | F1 P +F2 P = 2a}.

The foci are in general the touch points of B1 , B2 with the plane.

Remark. A video form of the proof can be found at here.

We will accept without proof that all (non-degenerate) conics are smooth (but we will prove it in section
(TODO 6.2.1)), and use the notations from the circle case:

(i) TC is the set of all tangents of C;

(ii) TP C is the tangent to C at P , where P ∈ C;

167
AoPS Chapter 6. Basic Conic Theory

(iii) Tℓ C is the point of tangency of C and ℓ, where ℓ ∈ TC.

We then have TC = {TP C|P ∈ C}, C = {Tℓ C|ℓ ∈ TC}.

Proposition 6.1.5. Let C be a conic and ℓ be a line. Then ℓ intersects C at exactly one point if and only if
ℓ ∈ TC.

Proof. As before, let C = S ∩ E where E is a plane and S is a circular conical surface with directrix Ω. Let
E1 be the plane containing ℓ and the vertex of S, E2 be the plane containing Ω, then

1 = |ℓ ∩ C| = |E ∩ E1 ∩ S| ⇐⇒ 1 = |(E1 ∩ E2 ) ∩ Ω| = |E1 ∩ E2 ∩ S|.

hence we may assume C is a circle, and the result clearly holds.

Now we can prove the optical property of conics:

Proposition 6.1.6. Let C be a conic (which is not a parabola) with foci F1 , F2 . Then for any P ∈ C, the
tangent TP C bisects ∠F1 P F2 .

Proof. We only prove the case where C is an ellipse, the hyperbola case is similar. Let TP′ be the external
angle bisector of ∠F1 P F2 , F2′ be the reflection of F2 across TP′ , then F1 , P, F2′ are collinear. So for any Q ̸= P
on TP′ ,
F1 P + F2 Q = F1 Q + F2′ Q > F1 Q + F2′ Q > F1 F2′ = F1 P + F2′ P = F1 P + F2 P

/ C, hence TP′ = TP C.
and so Q ∈

Proposition 6.1.7. Let P is a parabola with focus F and directrix L. Let P be a point on P. Let K be the
projection P onto L. Then TP P is the internal angle bisector of ∠F P K and also the perpendicular bisector
of F K.

Proof. Let TP′ be the internal angle bisector of ∠F P K, from F P = LP we know TP′ is the perpendicular
bisector of F K. For any point Q ̸= P on TP′ , let KQ be the projection of Q onto L, then we have

KQ Q < KQ = F Q

/ P, hence TP′ = TP P.
so Q ∈

Remark. Considering the property above, we can view the other “focus” of a parabola as the point of
infinity in the direction perpendicular to the directrix, so that the optical property still holds.

Below, we consider parabolas as a limiting case for the definitions.

168
AoPS Chapter 6. Basic Conic Theory

Definition 6.1.8. For any conic C with foci F1 , F2 , the center O of C is defined as the midpoint of F1 F2 .

From the definition of foci (TODO (6.1.4)), we clearly have:

Proposition 6.1.9. For any non-parabola conic C with center O, C is symmetric about O.

Definition 6.1.10. For conic C, we define the major axis of C as the segment whose endpoints are the
intersections of C and F1 F2 , where F1 , F2 are the foci of C.

Proposition 6.1.11. Let F be a focus of conic C, and Γ be the circle whose diameter of the major axis of C.
Then for any ℓ ∈ TC, the projection of F onto ℓ lies on Γ.

Proof. In the (limiting) case where C is a parabola, it is not hard to show that this is equivalent to (TODO
(6.1.7)): In particular, Γ is the tangent to C which is parallel to the directrix.

We will prove the result for the ellipse case; the hyperbola case is similar. Let F ′ be the other focus of C,
O be the center of C, and T = TC ℓ. By (TODO (6.1.6)), F ′ T passes through the reflection F ′′ of F across ℓ,
so the projection M of F onto ℓ is the midpoint of F F ′′ , and so

1 1
OM = · F ′ F ′′ = (F T + F ′ T )
2 2

but F T + F ′ T is the length of the major axis, so M ∈ Γ.

Practice Problems

Problem 1. Prove that C is a conic if and only if C is a circle or there exists point F , line L and e > 0 such
that
C = {P | F P = e · d(L, P )}

where

(i) if e < 1, then C is an ellipse;

(ii) if e = 1, then C is a parabola;

(iii) if e > 1, then C is a hyperbola.

The circle case can be seen as the limiting case where L = L∞ and e = 0, such that e · d(L, P ) and F is
the center of the circle.

Problem 2. Let F1 and F2 be the foci of conic C. Let ℓ1 , ℓ2 ∈ TC and A be the intersection of ℓ1 and ℓ2 .
Prove that ℓ1 and ℓ2 are isogonal in ∠F1 AF2 .

169
AoPS Chapter 6. Basic Conic Theory

6.2 Cross-Ratio on Conics

We have already defined cross ratio on a circle (see Section (TODO (2.1))), now we want to extend this
definition to conics.

Remark (Technical Details). Here, our conics are considered over RP2 exposition-wise, but everything here
naturally caries over to CP2 . This will remain true for the next few sections (In other words, we have points
at infinity).

Definition 6.2.1. Let E1 and E2 be planes in the space P3R , and let V be a point not on either plane. Define
φ : E1 → E2 as
φ(Q) = V Q ∩ E2

then φ is said to be a perspective transformation from E1 to E2 with center V .

Proposition 6.2.2. Let E1 , E2 be planes and φ : E1 → E2 be a projective transformation, then for any line
ℓ ∈ E1 , φ(ℓ) is also a straight line.

Proof. Let V be the center of φ as above, and let G be the plane containing V and ℓ. Then

φ(ℓ) = {φ(Q) | Q ∈ ℓ} = {P Q ∩ E2 | Q ∈ ℓ} = {P Q | Q ∈ ℓ} ∩ E2 = G ∩ E2 ,

which is a straight line as it is the intersection of two distinct planes.

Theorem 6.2.3. Let E1 , E2 be planes and φ : E1 → E2 be a projective transformation with center V ,

then for collinear points P1 , P2 , P3 , P4 ∈ E1 , we have that the cross ratios (P• ) := (P1 , P2 ; P3 , P4 ) and
(φ(P• )) := (φ(P1 ), φ(P2 ); φ(P3 ), φ(P4 )) are equal.

Proof. Since ℓ and φ(ℓ) are coplanar, we have

(P• ) = V (P• ) = V (φ(P• )) = (φ(P• )).

Theorem 6.2.4. Let E1 , E2 be planes and φ : E1 → E2 be a projective transformation with center V , then
for concurrent lines ℓ1 , ℓ2 , ℓ3 , ℓ4 ∈ E1 , we have (ℓ• ) = (φ(ℓ• )).

Proof. Let E3 be a plane containing V but not containing the concurrency point of ℓ• . Let Ri = ℓi ∩ E3 , then

(ℓ• ) = (R• ) = (φ(R• )) = (φ(ℓ• )).

170
AoPS Chapter 6. Basic Conic Theory

From the definition of a conic, we also get:

Proposition 6.2.5. Let E be a plane and C ⊂ E be a conic. Then there exists a plane E1 , a circle Ω ⊂ E1
and a perspective transformation φ : E → E1 such that φ(C) = Ω.

To prove the next result, we will accept without proof that five points determine a conic and five tangent
lines determine a conic (we will prove it at the end of this section). We will also use (P1 P2 P3 P4 P5 ) to denote
the conic through P1 , P2 , P3 , P4 , P5 and (ℓ1 ℓ2 ℓ3 ℓ4 ℓ5 ) to denote the conic tangent to ℓ1 , ℓ2 , ℓ3 , ℓ4 , ℓ5 when there
is no risk of ambiguity.

Theorem 6.2.6 (Fundamental Theorem of Conic Sections). Let P1 , P2 , P3 , P4 , A, A′ be points in the plane, no
three collinear. Then P1 , P2 , P3 , P4 , A, A′ lie on a common conic or are concyclic if and only if A(P• ) = A′ (P• ).

Proof. Let C = (P1 P2 P3 AA′ ) and E be the plane containing C. Then there exists plane E1 and circle Ω ∈ E1
and a perspective transformation φ : E → E1 such that φ(C) = Ω.

(⇒) Given P4 ∈ C or φ(P4 ) ∈ Ω, we have

A(P• ) = φ(A)(φ(P• )) = φ(A′ )(φ(P• )) = A′ (P• ).

(⇒) Let Pi′ = Pi for i = 1, 2, 3, P4′ = AP4 ∩ C \ {A}, then by (⇒) we have

A′ (P•′ ) = A(P• ) = A′ (P• ),

so P4′ = A′ P4 ∩ AP4 = P4 , so P1 , P2 , P3 , P4 , A, A′ are concyclic.

The dual result is:

Theorem 6.2.7. Let ℓ1 , ℓ2 , ℓ3 , ℓ4 , L, L′ be lines in the plane, no three concurrent. Then ℓ1 , ℓ2 , ℓ3 , ℓ4 , L, L′

are tangent to a common conic if and only if L(ℓ• ) = L′ (ℓ• ).

The proof is similar. The above theorem allows us to finally define cross ratio on a conic.

Definition 6.2.8. Let P1 , P2 , P3 , P4 be points on a conic C. Then

(P• )C := (P1 , P2 ; P3 , P4 )C := A(P• ),

for any A ∈ C, is the cross ratio of (P1 , P2 , P3 , P4 ) with respect to C.

171
AoPS Chapter 6. Basic Conic Theory

Definition 6.2.9. Let ℓ1 , ℓ2 , ℓ3 , ℓ4 be tangents to a conic C. Then

(ℓ• )C := (ℓ1 , ℓ2 ; ℓ3 , ℓ4 )C := L(ℓ• ),

where L ∈ TC, is the cross ratio of (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) with respect to C.

We also extend the concept of harmonic quadrilateral to a conic:

Definition 6.2.10. For points P1 , P2 , P3 , P4 on a conic C, the quadrilateral P1 P3 P2 P4 is said to be harmonic

on C if (P• )C = −1.

Similar to (TODO (2.2.11)), we have:

Proposition 6.2.11. For points P1 , P2 , P3 , P4 on a conic C, let the tangents to C at P1 and P2 met at A.
Then (P1 P2 )(P3 P4 ) is harmonic if and only if A lies on P3 P4 .

6.2.1 Conics and Degree-2 Curves

We now prove the property that we omitted the proof of earlier – the fact that five points define a conic, and
five tangent lines determine a conic. The proof will be in two parts:

(i) Any five points, no four collinear, determine a unique degree-2 curve;

(ii) Conics are the same as non-degenerate degree-2 curves.

Here, a degree-2 curve is simply a curve defined on the projective plane (with coordinates [x : y : z]) by
the homogeneous quadratic function

f (x, y, z) = Ax2 + By 2 + Cz 2 + 2Dyz + 2Ezx + 2F xy = 0

This rewrites as v T MC v = 0 for vertical vector v in x, y, z and

 
A F E
 
MC =  F B D 
 
 
E D C

For this to be solvable over the reals, we must assume that the matrix MC is not positive-definite or negative-
definite (so v T MC v isn’t always positive or negative), and is not the zero matrix. Furthermore, C is degenerate
(a double line) if and only if det MC = 0.

172
AoPS Chapter 6. Basic Conic Theory

We first show that any five points determine a degree-2 curve. We can assume that for five pairwise
distinct points P1 , P2 , P3 , P4 , P5 no four are collinear (if four of them lie on a common line, then the union of
that line and any line through the remaining point is a degree-2 curve).

Hence, WLoG assume P1 , P2 , P3 are not collinear. Through a projective transformation, we can even
assume
P1 = [1 : 0 : 0], P2 = [0 : 1 : 0], P3 = [0 : 0 : 1].

Plugging it back into the original equation, we get A = B = C = 0. So all degree-2 curves through P1 , P2 , P3
are of the form
f (x, y, z) = 2Dyz + 2Ezx + 2F xy = 0.

Now if P4 = [x1 : y1 : z1 ], P5 = [x2 : y2 : z2 ], we want to solve


Dy1 z1 + Ez1 x1 + F x1 y1 = 0,


Dy2 z2 + Ez2 x2 + F x2 y2 = 0


where D, E, F are not all zero. We know that this equation has three variables and thus must have a nonzero
solution, and we want this solution set to be one dimensional (as (D, E, F ) and (λD, λE, λF ) describe the
same line), which implies a unique cubic.

Note that (y1 z1 , z1 x1 , x1 y1 ), (y2 z2 , z2 x2 , x2 y2 ) both aren’t (0, 0, 0), or else we get that P4 , P5 are in
{P1 , P2 , P3 }. If we define Q1 = [y1 z1 : z1 x1 : x1 y1 ], Q2 = [y2 z2 : z2 x2 : x2 y2 ] which are distinct points, then
[D : E : F ] is the coefficients of the unique line Q1 Q2 which finishes. Else, suppose Q1 = Q2 which means
that
y1 z1 = λy2 z2 , z1 x1 = λz2 x2 , x1 y1 = λx2 y2 , λ ̸= 0.

If one of x2 , y2 , z2 = 0, WLOG say x2 = 0, then y2 z2 ̸= 0. This tells us then that x − 1 = 0, however, then
P2 , P3 , P4 , P5 are collinear, contradiction.

As such, we get that

y1 z1 z1 x1 x1 y1 x1 y1 z1 √
λ= · = · = · =⇒ = = =± λ
y2 z2 z2 x2 x2 y2 x2 y2 z2

or P4 = P5 , contradiction. This proves (i).

Next, let’s show that five lines determine a conic, however this is effectively the same as the five line case:
we can assign lines to points in P2

ℓi : ai x + bi y + ci z = 0 ←→ ℓ∨
i : [ai : bi : ci ].

173
AoPS Chapter 6. Basic Conic Theory

Then ℓi having no four concur is the same as no four of ℓ∨

i being collinear, so there is a unique conic C
∨

through ℓ∨ ∨ ∨
i . As such, it remains to find a map C 7→ C that maps the points on C to the tangents of C.

Proposition 6.2.12. Given a conic C : Ax2 + By 2 + Cz 2 + 2Dyz + 2Ezx + 2F xy = 0 and a point

P = [x0 : y0 : z0 ] on it, then the tangent at P is L : ax + by + cz = 0 for

[a : b : c] = [Ax0 + F y0 + Ez0 : F x0 + By0 + Dz0 : Ex0 + Dy0 + Cz0 ].

Proof. The proof requires a bit of calculus: the tangent to the contour f (x, y, z) = 0 at P = [x0 : y0 : z0 ] is
the line      
∂f ∂f ∂f
·x+ ·y+ · z = 0.
∂x P ∂x P ∂z P
 
A simpler way to find L (parameterizing (x, y) as [x : y : 1]) is to consider the line x0
z0 − cb t, yz00 + ac t , c =
−(ax0 + by0 ) (for P that isn’t the origin or at infinity), and then consider when the quadatic f (x, y, 1) := f (t)
has a double root (by considering the discriminant).

As such, it follows that C is mapped to C ∨ by the map (because our map ℓ 7→ ℓ∨ acts as the identity)

[x : y : z] 7→ [Ax + F y + Ez : F x + By + Dz : Ex + Dy + Cz] =: [a : b : c]

This means that we can explicitly describe C as g(a, b, c) = wT MC ∨ w = 0 where w is a vertical vector in
(a, b, c). By taking a change of basis, we get:

g(a, b, c) = wT MC ∨ w ⇐⇒ f (a, b, c) = v T MC MC ∨ MC v

so MC ∨ = M⌋−1 .

Since (M −1 )−1 = M holds for nonsingular matrices M , we get that (C ∨ )∨ = C, and the map C 7→ C ∨ is
bijective. If C ∨ is not degenerate, then the lines ℓi uniquely correspond to a conic C. If it is, then three lines,
WLOG ℓ1 , ℓ2 , ℓ3 concur at P . Then if Q = ℓ4 ∩ ℓ5 , then C is a double cover of the line P Q.

Finally, we need to show that quadratic curve are equivalent to conics. Take a plane E ⊂ P3 , with
coordinates [w : x : y : z], and let the plane at infinity be E∞ : z = 0. By rotating and translating, we can
WLOG assume w = 0, so E can be written as [x : y : z].

Then for a circular conic surface S : R3 with central circle O is the locus of

(w0 , x0 , y0 ) + r cos θ · e⃗1 + r sin θe⃗2

where e⃗1 , e⃗2 are two unit orthogonal vectors. If we let e⃗3 = e⃗1 × e⃗2 , then for some s ̸= 0, we have V has

174
AoPS Chapter 6. Basic Conic Theory

coordinates
(w1 , x1 , y1 ) = (w0 , x0 , y0 ) + s · e⃗3 .

w−w1

If we let U = e⃗1 e⃗2 e⃗3 , ⃗v = x−x1 , then S is the locus of points (a : b : c) satisfying
y−y1

 
a
r 2  
a2 + b2 = ·c ,  b  = U −1⃗v
 
s  
c

ew

Scaling, we can assume that s = 1. Then we get if ⃗e = e⃗3 = ex
ey
, then

2
|⃗v | = a2 + b2 + c2 = (r2 + 1)c2 = (r2 + 1)(⃗e · ⃗v )2

Shifting, we can assume x1 = y1 = 0. Rotating about the w axis, we assume ey = 0.

6.2.2 Various Conic Theorems

Now that we can deal with cross-ratio in conics, we can re-prove some old theorems. We will revisit Pascal
and Brianchon (TODO 2.3.3) (TODO 2.3.5) in their full conic forms.

Theorem 6.2.13 (Pascal’s). Given points P1 , P2 , P3 , P4 , P5 , P6 , with no three collinear, and all conconic, we
have that
P1 P2 ∩ P4 P5 , P2 P3 ∩ P5 P6 , P3 P4 ∩ P6 P1

are collinear.

Proof. We can essentially use the same proof as (TODO 2.3.3). Let Q1 = P1 P2 ∩ P4 P5 , Q2 = P2 P3 ∩ P5 P6 ,
Q3 = P3 P4 ∩ P6 P1 ; so that

P1 (P2 , P3 ; P4 , P5 ) = Q1 (P1 , P3 ; P4 ; Q3 ),

P5 (P2 , P3 ; P4 , P6 ) = Q1 (P1 , P3 ; P4 , Q2 ).

Then because P1 , P2 , P3 , P4 , P5 , P6 are conconic,

P1 (P2 , P3 ; P4 , P6 ) = P5 (P2 , P3 ; P4 , P6 )

⇐⇒ Q1 (P1 , P3 ; P4 , Q3 ) = Q1 (P1 , P3 ; P4 , Q2 ),

and the result follows.

175
AoPS Chapter 6. Basic Conic Theory

The dual of this theorem is:

Theorem 6.2.14 (Brianchon’s). Given lines ℓ1 , ℓ2 , ℓ3 , ℓ4 , ℓ5 , ℓ6 with no three concurrent, and all six tangent
to the same conic, then

(ℓ1 ∩ ℓ2 )(ℓ4 ∩ ℓ5 ), (ℓ2 ∩ ℓ3 )(ℓ5 ∩ ℓ6 ), (ℓ3 ∩ ℓ4 )(ℓ6 ∩ ℓ1 )

are concurrent.

The proof follows as the dual of Pascal’s Theorem.

Remark. Like in the case of a circle, we are allowed to let two points (lines) coincide, and then the line
passing through them (intersection point) becomes a tangent line (point), and conversly [what]

Theorem 6.2.15 (Carnot’s). Given △ABC, and points (D1 , D2 ), (E1 , E2 ), (F1 , F2 ) on segments BC, CA, AB,
the following three conditions are equivalent:

(i) D1 , D2 , E1 , E2 , F1 , F2 are conconic;

(ii) AD1 , AD2 , BE1 , BE2 , CF1 , CF2 are tangent to one conic;
BD1 BD2 CE1 CE2 AF1 AF2
(iii) · · · · · = 1.
D1 C D2 C E1 A E2 A F1 B F2 B

Proof. We will only prove (i) ⇐⇒ (iii), (ii) ⇐⇒ (iii). Let E2 F1 , F2 D1 , D2 E1 intersect BC, CA, AB at X, Y, Z
respectively. Menelaus gives

BX CE2 AF1 BD1 CY AF2 BD2 CE1 AZ

· · = · · = · · = −1.
XC E2 A F1 B D1 C Y A F2 B D2 C E1 A ZB

Then by Pascal’s, D1 , D2 , E1 , E2 , F1 , F2 is conconic if and only if AX, BY, CZ are concurrent, if and only if

BX CY AZ BD1 BD2 CE1 CE2 AF1 AF2

· · = −1 ⇐⇒ · · · · · = 1.
XC Y Z ZB D1 C D2 C E1 A E2 A F1 B F2 B

Definition 6.2.16. Given △ABC,

(i) We define the conic C as the circumconic of △ABC if A, B, C ∈ C;

(ii) We define the conic ⌋ as the inconic of △ABC if BC, CA, AB ∈ T⌋ .

Much like the circumcircle, we have for a conic:

Definition 6.2.17. (i) Given △ABC with circumconic C, and any point P , we define with P and △ABC
the C-cevian triangle with △(AP ∩ C)(BP ∩ C)(CP ∩ C).

176
AoPS Chapter 6. Basic Conic Theory

(ii) Given △ABC with inconic ⌋, and any line ℓ, we define with ℓ and △ABC the ⌋-cevian triangle with
△((BC ∩ ℓ)⌋)((CA ∩ ℓ)⌋)((AB ∩ ℓ)⌋).

The next theorem is stated as the special case when n = 3, but that is when it is most useful.

Theorem 6.2.18 (Poncelet’s Porism). Given △A1 B1 C1 and △A2 B2 C2 on the plane, △A1 B1 C1 and
△A2 B2 C2 share a circumconic if and only if they share an inconic.

Proof. Notice that

A1 (B1 , C1 ; B2 , C2 ) = (B2 C2 )(A1 B1 , C1 A1 ; A2 B2 , C2 A2 ),

A2 (B2 , C2 ; B1 C1 ) = (B1 C1 )(A2 B2 , C2 A2 ; A1 B1 , C1 A1 ),

then △A1 B1 C1 , △A2 B2 C2 share a circumconic if and only if

A1 (B1 , C1 ; B2 , C2 ) = A2 (B2 , C2 ; B1 , C1 )

⇐⇒ (B2 C2 )(A1 B1 , C1 A1 ; A2 B2 , C2 A2 ) = (B1 C1 )(A1 B1 , C1 A1 ; A2 B2 , C2 A2 )

which is true if and only if △A1 B1 C1 , △A2 B2 C2 share an inconic.

Let us recall: a complete quadrilateral q = (P1 , P2 , P3 , P4 )’s cevian triangle is the triangle with vertices

P2 P3 ∩ P1 P4 , P3 P1 ∩ P2 P4 , P1 P2 ∩ P3 P4

, so it’s actually possible to define P4 completely from △P1 P2 P3 and the cevian triangle.

Theorem 6.2.19 (Nine-Point Conic). For complete quadrilateral q = (P1 , P2 , P3 , P4 ), define Mij to be the
midpoint of Pi Pj , and let X, Y, Z be the vertices of the cevian triangle. Then

M23 , M14 , M31 , M24 , M12 , M34 , X, Y, Z

are conconic.

We can actually extend this.

Theorem 6.2.20 (Extended Nine-Point Conic). For complete quadrilateral q = (P1 , P2 , P3 , P4 ) with cevian
triangle XY Z, choose a line ℓ, and define Qij = Pi Pj ∩ ℓ. Choose Rij on Pi Pj such that

(Pi , Pj ; Qij , Rij ) = −1

177
AoPS Chapter 6. Basic Conic Theory

then
R23 , R14 , R31 , R24 , R12 , R34 , X, Y, Z

lie on one conic.

Proof. You’re mother

The dual of this theorem is also useful:

Theorem 6.2.21. For complete quadrangle Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) that has cevian triangle with sides x, y, z,
choose a point P and define lij = (ℓi ∩ ℓj )P . Choose kij in the pencil of lines through (ℓi ∩ ℓj ) such that

(ℓi , ℓj ; lij , kij ) = −1,

then k23 , k14 , k31 , k24 , k12 , k34 , x, y, z are all tangent to a common conic.

Definition 6.2.22. We call the conic in (TODO 6.3.8) the ℓ-nine-point conic of the complete quadrangle q
wrt. line ℓ. If line ℓ is not specified, then assume it’s just L∞ .

When ℓ = L∞ , then (R23 R14 )(R31 R24 )(R12 R34 ) make a hexagon with parallel sides, so the center of the
nine-point conic is just the midpoint of R23 R14 , which is just the centroid of complete quadrangle q.

We define its dual similarily.

Definition 6.2.23. We call the conic in (TODO 6.3.9) the P -nine-line conic of the complete quadrilateral
Q wrt. point P .

6.2.3 Newton’s Three Theorems

These are three big theorems about complete quadrilaterals and conic sections.

Theorem 6.2.24 (Newton I). For complete quadrilateral Q, the midpoints of the three diagonals are
collinear.

(We’ve already seen this in (TODO 4.1.4).)

Theorem 6.2.25 (Newton II). For complete quadrilateral Q, let C be a conic tangent to the four sides of Q,
then the center of C lies on the Newton line. The converse also holds; if you choose a point on the Newton
line then there is a conic centered at it tangent to all four sides of C.

Proof. Assume WLOG that ℓ1 isn’t parallel to ℓ4 . Let Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ), Pij = ℓi ∩ ℓj , R2 , R3 be the
midpoints of P31 P24 , P12 P34 respectively, and let O be the center of C. Draw ℓ′1 , ℓ′4 as the reflections of ℓ1 , ℓ4

178
AoPS Chapter 6. Basic Conic Theory
′
over O, then P14 = ℓ′1 ∩ ℓ′4 is the reflection of P14 over O. Let Q2 , Q3 respectively be the reflections of P14
over R2 , R3 , since Q and ℓ′1 , ℓ′4 are all tangent to C, we have

′
∞ℓ1 (Q2 , Q3 ; P14 , ∞ℓ4 ) = (P24 , P34 ; ℓ′1 ∩ ℓ4 , ∞ℓ4 ) = (P12 , P31 ; ∞ℓ1 , ℓ1 ∩ ℓ′4 )
′ ′
= ∞ℓ4 (Q3 , Q2 ; ∞ℓ1 , P14 ) = ∞ℓ4 (Q2 , Q3 ; P14 , ∞ℓ1 ),

′
and thus by cross ratio (TODO 2.1.9) we have that Q2 , Q3 , P14 are collinear, and by a homothety with scale
1
factor 2 from P14 we get that O ∈ R2 R3 .

Theorem 6.2.26 (Newton III, British Flag Theorem). For complete quadrilateral Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ), Pij =
ℓi ∩ ℓj , if inconic C is tangent to the four sides at Q1 , Q2 , Q3 , Q4 , then Q1 Q3 , Q2 Q4 , P12 P34 , P23 P41 are
concurrent.

Proof. Brianchon on (ℓ1 ℓ1 ℓ2 ℓ3 ℓ3 ℓ4 ) and (ℓ1 ℓ2 ℓ2 ℓ3 ℓ4 ℓ4 ) gets us that Q1 Q3 , P12 P34 , P23 P41 are concurrent and
Q2 Q4 , P12 P34 , P23 P41 are concurrent.

Practice Problems

Problem 1. Let △S1 S2 S3 be a equilateral triangle, and let P be an arbitrary point. Prove that the Euler
lines of △P S2 S3 , △P S3 S1 , △P S1 S2 concur.

Problem 2. Let I be the incenter of △ABC, and let D be the foot from I to BC, and let M be the midpoint
of BC. Prove IM bisects AD.

Problem 3. Let ABCD be a cyclic quadrilateral and let P be intersection of diagonals AC and BD. Let
MAB , MBC , MCD , MDA respectively be the midpoints of arcs AB, BC, CD, DA, let IAB , IBC , ICD , IDA re-
spectively be the incenters of △P AB, △P BC, △P CD, △P DA. Prove that MAB IAB , MBC IBC , MCD ICD , MDA IDA
are concurrent.

Problem 4. Let H be the orthocenter of △ABC, and let M be the midpoint of AH. Let E, F be the
feet from B, C to CA, AB. Choose point R on EM such that RBC = 90◦ , and choose S on F M such that
BCS = 90◦ . Prove that A, R, S are collinear.

Problem 5. Let △DEF be the arc-midpoint triangle of △ABC, and let X1 , X2 , Y1 , Y2 , Z1 , Z2 be the
intersections of △ABC and △DEF (in the sequence counterclockwise D, Z1 , Z2 , E, X1 , X2 , F, Y1 , Y2 , D). Let
Pbc = EY1 ∩ F Z2 , Pcb = F Z1 ∩ EY2 , and similarily define Pca , Pac , Pab , Pba . Prove that Pbc Pcb , Pca Pac , Pab Pba
are concurrent.

Problem 6. Choose six points on the sides of equilateral triangle △ABC (two on each side), where A1 , A2
are the points on BC, etc, such that A1 A2 B1 B2 C1 C2 is a equilateral hexagon. Prove that A1 B2 , B1 C2 , C1 A2
are concurrent.

179
AoPS Chapter 6. Basic Conic Theory

Problem 7. Let I be the incenter of △ABC, and let P be the inversive image of I in (ABC). Let line ℓ be
a tangent from P to I, and let ℓ intersect (ABC) at X, Y . Prove that ∠XIY = 120◦ .

Problem 8. Suppose there exists eight points P1 , P2 , . . . , P8 in the plane, and Qi = Pi−2 Pi−1 ∩ Pi+1 Pi+2 .
Prove that P1 , . . . , P8 are conconic if and only if Q1 , . . . , Q8 are conconic.

6.A A Journey to The Hidden Circle Points

Plane geometry is essentially the theory of the complex projective plane along with two special points I, J,
and a special line L∞ . Let’s fully generalize it.

First, let’s define angles in complex numbers. For a real line in R2 with slope t ∈ RP1 = R ∪ {∞}, after a
rotation by θ, its slope becomes
sin θ + t cos θ
.
cos θ − t sin θ

Note that if we let Θ = e2iθ , then by Euler’s theorem eiθ = cos θ + i sin θ, we have

sin θ (eiθ − e−iθ )/2i Θ−1

= iθ = ,
cos θ (e + e−iθ )/2 i(Θ + 1)

so in C2 , we can define a complex slope t ∈ CP1 = C ∪ {∞}, and then after a “rotation” by Θ ∈ C× = C\0,
we get a slope of
(Θ − 1) + it(Θ + 1)
tΘ := .
i(Θ + 1) − t(Θ − 1)

Note that we still have (tΘ1 )Θ2 = tΘ1 Θ2 as expected.

Remark (Skip this if needed). If we define S1 as Spec Z[r, s]/⟨r2 + s2 − 1⟩, there is a group action of rotations
on slopes as follows: Identify the line at infinity L∞ with RP1 via [x : y : 0] 7→ t = y/x and identify

S1 (R) = {(r, s) ∈ R2 | r2 + s2 = 1} ⊆ R2

with {Θ ∈ C | ∥Θ∥ = 1} ⊂ C× via (r, s) 7→ r + is. We then have the action

S1 (R) × RP1 RP1

sin θ+t cos θ s+t(r+1)
(r, s, t) = (e2iθ , t) cos θ−t sin θ = (r+1)−ts

The extended action on CP1 given above is the same function on C, as we can identify

S1 (C) = {(r, s) ∈ C2 | r2 + s2 = 1} ⊆ C2

180
AoPS Chapter 6. Basic Conic Theory
 
Θ+Θ−1 Θ−Θ −1
and C× ∋ Θ via Θ = r + is, (r, s) = 2 , 2i , yielding

−1 −1
s + t(r + 1) ( Θ−Θ ) + t( Θ+Θ + 1)
= 2i
−1
2
−1 = tΘ .
(r + 1) − ts Θ+Θ
( 2 Θ−Θ
+ 1) − t( 2i )

Then for any t1 , t2 ∈ CP1 not equal to ±i, tΘ

1 = t2 if and only if

t1 + i t2 − i
Θ= = (t1 , t2 ; −i, i)
t1 − i t2 + i

Now we can finally define angles in complex numbers.

∡e (t1 , t2 ) = (t1 , t2 ; −i, i)

 
1 1 e
∡(t1 , t2 ) = log ∡e (t1 , t2 ) = ∡ (t1 , t2 ) · 180◦ ∈ C/πZ] = C◦ /180◦ Z
2i 2πi

and for any two lines ℓ1 , ℓ2 , we define ∡(ℓ1 , ℓ2 ) = ∡(∞ℓ1 , ∞ℓ2 ). In particular, (t1 , t2 ; −i, i) = −1 if and only
π
if ∡(t1 , t2 ) = 2 = 90◦ .

When t = ±i, tΘ = t for all Θ ∈ C× . These t correspond to the titular circle points [1 : ±i : 0] ∈ CP2 ,
which we call I, J. (The original text uses ∞±i instead to avoid confusion with the incenter.)

These complex points are important because they are actually intersection of all circles, defined as

((x − x0 z)2 + (y − y0 z)2 − kz 2 ) = 0

for some x0 , y0 . So the circumcircle (ABC) of any triangle △ABC is actually just the circumconic (ABCIJ)
of A, B, C, I, J. Thus we can actually work with circles purely projectively now, as a family of conics through
two fixed points!

Remark. This lets us define angles with a simple cross ratio: let A, B, C be three complex points and let
I, J be the circle points. Then
1
∡BAC = log(AB, AC; AI, AJ).
2i
π
For example, if (AB, AC; AI, AJ) = −1, Euler’s formula gives us that ∡BAC = 2 = 90◦ , which gives us a
criterion for perpendicularity.

Here are some examples of how plane geometry reduces to P2 and the circle points.

Example 6.A.1. Since (ABC) is actually just the circumconic (ABCIJ), for a point P on (ABC),

1 1
∡BP C = log(P B, P C; P I, P J) = log(AB, AC; AI, AJ) = ∡BAC,
2i 2i
181
AoPS Chapter 6. Basic Conic Theory

which proves the inscribed angle theorem.

Example 6.A.2. Let X be the intersection of BC with the line at infinity L∞ := IJ, and let X ∨ be a point
on L∞ such that (I, J; X, X ∨ ) = −1, then AX ∨ is the perpendicular from A to BC. Thus we can actually
use the circle points to construct the orthocenter H of △ABC purely projectively. (See (TODO 7.1.25) for
more information).

Proposition 6.A.3. I and J are isogonal conjugates in any triangle △ABC.

Proof. Let ℓ+ , ℓ− be the internal and external bisectors of ∠BAC. Since these are perpendicular, we have

(AI, AJ; ℓ+ , ℓ− ) = −1,

so the projective line involution (see: (TODO 7.2), for right now just interpret it as ”involution that preserves
cross ratio) at A with the internal and external bisectors as fixed points has to be reflection across the angle
bisector, and thus must swap AI, AJ.

(An alternate proof of this by AoPS user popop614 is that the two circle points are the two intersections
of the line at infinity and the circumcircle, and since the line at infinity and the circumcircle swap under
isogonal conjugation, by Steiner conic theorem (all involutions on a conic are given by the second intersection
map), isogonal conjugation must switch the two intersections.)

Remark. If we think of △ABC on the complex line (C1 which is isomorphic to the Euclidean plane, not
C2 ) and define barycentric coordinates, then the barycentric coordinates of I, J are just [C − B : A − C :
B − A], [C − B : A − C : B − A]. (If you read (TODO 7.3), (TODO 7.4), note that the barycentric product
of I × J is just [a2 : b2 : c2 ], so isoconjugation about this point is just isogonal conjugation.)

If we define △ABC as three points in C2 , then we need to define a “complex coordinate system” in
A, B, C through the mapping

C2 C ⊗R C C
(x, y) 1⊗x+i⊗y x + iy

where conjugation 1 ⊗ x + i ⊗ y = 1 ⊗ x − i ⊗ y is defined with the first C.

For a conic C (defined with real coefficients), let ℓ++ , ℓ+− be the two tangents from I to C, and let
ℓ−+ , ℓ−− be the two tangents from J to C. Then ℓ−+ , ℓ−− is the complex conjugate of ℓ++ , ℓ+− .

Proposition 6.A.4. The two points F+ = ℓ++ ∩ ℓ−+ , F− = ℓ+− ∩ ℓ−− are the two foci of C.

182
AoPS Chapter 6. Basic Conic Theory

Proof. Since the complex conjugates of F± are themselves, the two points F± are real. For a point P ∈ C, let
FP′ be the reflection of F+ across the tangent TP C, then by the definition of foci, we only need to prove that
the locus of FP′ is a circle (which is just a conic through I, J.)

Let XP , YP , ZP respectively be the intersections of TP C with ℓ++ , ℓ−− , F FP′ . Then

(XP , YP ; ZP , ∞TP C ) = (F, FP′ ; ZP , ∞F FP′ ) = −1

then from the harmonic bundle in a complete quadrilateral, we get that FP′ = IYP ∩ JXP . Obviously,
XP → P → YP is a cross-ratio preserving map, so thus the locus of FP′ is a conic through I, J.

We can also define a pair of “imaginary” foci Fi = ℓ++ ∩ ℓ−− , F−i = ℓ+− ∩ ℓ−+ . Note that we can’t
distinguish between (F+ , F− ) and (Fi , F−i ) purely projectively, so let’s just define two pairs of foci for a given
conic; two are real and two are purely imaginary.

Note that all of this theory still works for two arbitrary points W = W+ , W− instead of the circle points
I, J. We define the W-points at infinity L∞ = L∞W := W+ W− . Similarily define a line at infinity L
connecting W+ , W− .

First note that we can define the W-angles ∡eW between two lines ℓ1 and ℓ2 analogously to how we did it
with the circle points, as:

∡eW (ℓ1 , ℓ2 ) := (L ∩ ℓ1 , L ∩ ℓ2 ; W− , W+ ), ℓi ∈
/ L, ℓi ∩ W = ∅

so we can define angles completely in terms of W.

We now have that ∡eW (ℓ1 , ℓ2 ) = 1 if and only if ℓ1 ∩ ℓ2 ∈ L, and we call lines ℓ1 , ℓ2 W-parallel. Similarily
define W-perpendicular lines for if the cross-ratio is harmonic.

Since affine transformations are just projective transformations that fix L, we can define W-affine
transformations as transformations that preserve L, and consider the group of W-affine transformations:

AW = {Φ : P2 → P2 | Φ(L) = L}.

Similarily, we define the group of W-spiral similarities as projective transformations that fix the points
W+ , W− :
SW = {Φ ∈ A | Φ(W± ) = W± }.

183
AoPS Chapter 6. Basic Conic Theory

Homotheties are just transformations in PW such that there is also some fixed point on L, so we define the
group of W-homotheties as:
HW = {Φ ∈ P | Φ(P ) = P, ∀P ∈ L}.

For Φ ∈ HW and three points P, Q, R, we have

L ∩ QR = Φ(L ∩ QR) = Φ(L) ∩ Φ(Q)Φ(R) = L ∩ Φ(Q)Φ(R).

So therefore QR ∩ Φ(Q)Φ(R) ∈ L. By the same logic we have RP ∩ Φ(R)Φ(P ), P Q ∩ Φ(P )Φ(Q) ∈ L. So by

Desargues’s theorem, we have P Φ(P ), QΦ(Q), RΦ(R) concur at a point O. Since P, Q, R are arbitrary, we
have that OΦ = O is the common point of all P Φ(P ), and we call this the center of Φ, the center of the
homothety.

Note that for any Φ, Ψ ∈ HW , we have OΦ , OΨ , OΦ◦Ψ .

If OΦ lies on L, we call Φ a W-translation. The group of W-translations also forms a group:

TW = {Φ ∩ HW | OΦ ∈ L}.

−−−→ − −−−→
For two vectors →
−
v1 = P1 Q1 , →
v2 = P2 Q2 , Pi , Qi ∈ /→
/ L, W± ∈ −
vi , Pi ̸= Qi , we can find a Φ ∈ P such that
→
− →
−
Φ(P ) = P , Φ(Q ) = Q (so our Φ is just Φ( v ) = v ), and now we can define
1 2 1 2 1 2

→
−
v2
→
− = Φ.
v1

However this has a subtle problem: we have not proved commutativity. Consider:

→
−
v2 −→
w −
→ →
w −
v2
2 2
→
− ◦ → ̸= ◦ .
v1 −w 1
−
→
w 1 v1→
−

So some of these Φ do not commute and are thus not translations. How do we get the translations?

Proposition 6.A.5. The group TW is the commutator subgroup of SW , i.e.

TW = [SW , SW ] := [Φ, Ψ] := ΦΨΦ−1 Ψ−1 | Φ, Ψ ∈ SW

so the abelianization of SW is the abelian group SW

ab
= SW /TW .

Remark. Consider the exact sequence

184
AoPS Chapter 6. Basic Conic Theory

1 HW /TW SW /TW SW /HW 1

∼

1 HWab SW
ab
S1 1
⃗v2
∡eW (v1 , v2 ).
⃗v1

This perfectly explains how all spiral similarities are the composition of dilations and rotations, and also
justifies the notation v⃗2 /v⃗1 (in the original case where we set W = I, J) as basically an improvement on
defining angles with the difference of line arguments. In other words, if we think of vi , wi as the lines defined
by extending v⃗i , w
⃗i , then
n
Y n
Y n
X n
X
v⃗i = w
⃗i =⇒ vi = wi .
i=1 i=1 i=1 i=1

Practice Problems

Problem 1. Prove that (tΘ1 )Θ2 = tΘ1 Θ2 .

185
Chapter 7

Projective Space

7.1 Conic Polarity

We have already seen prior that the concept of polarity for circles has some very nice projective properties.
For one thing, we know that there is a projective transformation sending conics to circles, so we can also
define something similar to polarity in general conics.

Definition 7.1.1. Given a conic C and an arbitrary point P , construct a moving line through P that
intersects C at M1 , M2 . Choose Q such that (P, Q; M1 , M2 ) = −1. Then the locus of Q is a line. We call this
line the polar pC (P ) of P with respect to C.

We can generalize a bunch of analogous properties from polarity in circles to properties in conics.

Theorem 7.1.2 (La Hire’s). Let C be a conic, and let P, Q be two points in the plane. Then

P ∈ pC (Q) ⇐⇒ Q ∈ pC (P ).

Definition 7.1.3. Let C be a conic, and suppose two points P and Q satisfy that the polar of P wrt. C goes
through Q. Then we call P, Q conjugate in C.

Definition 7.1.4. Let C be a conic, and let K be a line. Let Q be a moving point on K. Then the fixed
point that pC (Q) passes through is the pole of K wrt. C and is denoted as pC (K).

Proposition 7.1.5 (Dual of La Hire’s). Let C be a conic, and let K, L be two lines, then

pC (K) ∈ L ⇐⇒ pC (L) ∈ K.

186
AoPS Chapter 7. Projective Space

Definition 7.1.6. Let C be a conic, if two lines K, L satisfy pC (K) ∈ L, then we call these two lines K, L
conjugate in C.

Definition 7.1.7. Let C be a conic, if △ABC satisfies that A, B, C are pairwise conjugate wrt. C, we call
△ABC a self-conjugate triangle wrt. C, and we call C the diagonal conic of △ABC.

(It’s called a diagonal conic because of complete quadrilateral properties in the diagonal intersection
points; think Brokard’s theorem.)

Proposition 7.1.8. Let C be a conic, and let P1 , P2 , P3 , P4 be four points on C. Let

X = P1 P2 ∩ P3 P4 , Y = P1 P3 ∩ P4 P2 , Z = P 1 P4 ∩ P2 P 3 ,

then △XY Z is self-conjugate wrt. C.

Theorem 7.1.9 (Polarity preserves cross-ratios). Let C be a conic, and let the five lines ℓ1 , ℓ2 , ℓ3 , ℓ4 , L have
P1 , P2 , P3 , P4 , A as poles wrt. C, then
A(P• ) = L(ℓ• ).

The following results are more connected to cross-ratios.

Corollary 7.1.10. Given a conic C, then six points P1 , P2 , P3 , P4 , A, B are conconic if and only if the six
lines pC (P1 , P2 , P3 , P4 , A, B) are all tangent to one conic.

Proof. Since
A(P• ) = pC (A)(ℓ• ), B(P• ) = pC (B)(ℓ• ),

we know P1 , P2 , P3 , P4 , A, B are conconic if and only if

pC (A)(ℓ• ) = A(P• ) = B(P• ) = pC (B)(ℓ• ),

but this is equivalent to pC (P1 , P2 , P3 , P4 , A, B) being tangent to a common conic.

Definition 7.1.11. Let C be a conic, we call two conic sections C1 , C2 conjugate wrt. C (notated as
C2 = pC (C1 )) if
{pC (P ) | P ∈ C1 } = TC2 .

Through (T ODO7.1.10), we can ensure that for a conic C and a arbitrary conic C1 , there exists a C2 such
that C1 , C2 are conjugate. Further, for two conics C1 , C2 , there exists another conic C such that C1 , C2 are
conjugate in C.

187
AoPS Chapter 7. Projective Space

Proposition 7.1.12. Let C be a conic and let O be the center of this conic. Then pC (O) = L∞ .

Proof. Draw two lines ℓ1 , ℓ2 intersecting C through O. Let these intersect C respectively at (P1 , Q1 ), (P2 , Q2 ).
Then the midpoints of P1 Q1 and P2 Q2 are obviously just O. So ∞P1 Q1 , ∞P2 Q2 ∈ pC (O), so pC (O) = L∞ .

Corollary 7.1.13 (Parallel Chords Theorem). Let C be a conic, and let O be the center of C. Let P1 , P2 be
two points in the plane, then O, P1 , P2 are collinear if and only if pC (P1 ) is parallel with pC (P2 ).

Proof. Note that O, P1 , P2 are collinear if and only if pC (O), pC (P1 ), pC (P2 ) are concurrent. However pC (O)
is the line at infinity L∞ so we are done.

Corollary 7.1.14. Let C be a conic, and let O be the center of C. Let AB be a chord on C, then OpC (AB)
bisects AB.

Proof. Let T = AB ∩ OpC (AB), U = ∞AB , then OpC (AB) = pC (U ), but by definition we have (U, T ; A, B) =
−1, so T bisects AB.

Theorem 7.1.15 (Eight Point Conic). Given △ABC, let P, Q be two arbitrary points and let △P a P b P c , △Qa Qb Qc
be the anticevian triangles of P, Q wrt. △ABC. Then P, P a , P b , P c , Q, Qa , Qb , Qc are all conconic.

Proof. Let C be the conic through P, P a , P b , P c , Q, then pC (A), pC (B), pC (C) are just the three sides of
△ABC, let AQ intersect C again at (Qa )′ , then we have

(A, AQ ∩ BC; Q, (Qa )′ ) = −1 = (A, AQ ∩ BC; Q, Qa )

and thus Qa = (Qa )′ , so Qb , Qc also lie on the conic by symmetry.

There is also a nice proof of the previous theorem with Cayley-Bacharach.

The dual of the eight-point conic is the eight-line conic.

Theorem 7.1.16 (Eight Line Conic). Let K, L be two lines, let △K a K b K c , △La Lb Lc respectively be the
anticevian triangles of K, L wrt. △ABC (triangle such that perspectrix of it and △ABC is K/L). Then
K, K a , K b , K c , L, La , Lb , Lc are all tangent to a common conic.

With the definition of polarity, we define a bunch of binary operations on points.

Definition 7.1.17. Given two points P, Q and △ABC, we define the crosspoint P ⋔ Q of P and Q as the
pole of P Q in (ABCP Q).

Note that this definition is different from the definition we gave in (TODO 1.1). (However the commutativity
of crosspoint is obvious.) So we will prove that the two definitions are equivalent.

188
AoPS Chapter 7. Projective Space

Proposition 7.1.18. Let △Pa Pb Pc be the cevian triangle of P wrt. △ABC. Then Pa (P ⋔ Q), Pb Pc , AQ
are concurrent.

By symmetry, the cevian triangle of P ⋔ Q in △Pa Pb Pc is perspective with ABC, with perspector Q (see
(TODO 1.1) problem 3).

Lemma 7.1.19 (Seydewitz-Staudt). Given △ABC and a circumconic C, let P be the pole of BC wrt. C.
Draw a line through P that intersects CA, AB at U, V , then U, V are conjugate in C.

Proof of lemma. Let D be the second intersection of BU and C. We use Pascal (TODO 6.3.1) on the hexagon
ABBDCC to get that AB, CD, P U concur at V , and thus from (TODO 7.1.8), U, V are conjugates in C.

Proof of (7.1.18). Use this lemma but set △ABC = △AP Q and set C = (ABCP Q), then we know that
Pa (P ⋔ Q) ∩ AQ lies on pC (Pa ) = Pb Pc .

Proposition 7.1.20. If R = P ⋔ Q, and △Pa Pb Pc , △Qa Qb Qc are the cevian triangles of P, Q, then both of
Pc Pa ∩ Qa Qb , Pa Pb ∩ Qc Qa lie on line AR.

Proof. We consider the equivalent problem formed by taking polarity over C = (ABCP Q). We only need to
prove that the tangent at A to C, Pb Qc , Pc Qb , P Q are concurrent at pC (AR).

Note that Pb Pc = pC (Pa ) and Qb Qc = pC (Qa ) both go through the pole of BC = Pa Qa , so we have

(B, C, AU ∩ C, A)C = −1 = A(B, C; U, Pb Qc ∩ Pc Qb ),

so A(Pb Qc ∩ Pc Qb ) is tangent to C.

We want to prove Pb Qc , Pc Qb , P Q concurrent, so by Desargues’s theorem we only need to prove that

U = Pb Pc ∩ Qb Qc , CP ∩ BQ = Pc P ∩ QQb , BP ∩ CQ = P Pb ∩ Qc Q are collinear. However these three points
all lie on the polar of BC ∩ P Q so we are done.

Proposition 7.1.21. Continuing the previous statements’ notation, let Pa∨ , Pb∨ , Pc∨ respectively be the
intersection points of the trilinear polar of P wrt. BC, CA, AB, and similarily define Q∨a , Q∨ ∨
b , Qc . Then

Pb Q∨ ∨ ∨ ∨
c ∩ Pc Qb , Pb Qc ∩ Pc Qb also lie on AR.

Proof. Let △T a T b T c be the polar of △ABC wrt. (ABCP Q). Then pC (Q∨ a
a ) = Qa T , and similarily to the

proof of (T ODO7.1.20), we first need to prove that Pc Pa ∩ Qc T c , Pa Pb ∩ Qb T b , pC (AR) = Pb Qc ∩ Pc Qb are

collinear. However this is just by Desargues’s theorem since Pb Qb ∩ Pc Qc = A, T b , T c are collinear.

Proposition 7.1.22. Let R be the crosspoint P ⋔ Q. Then the perspectrix of the anticevian triangle of R,
△R and the cevian triangle of Q, △Q , is the trilinear polar of P .

189
AoPS Chapter 7. Projective Space

Proof. Let X be the intersection point of Pb∨ Pc∨ (the trilinear polar of X) and Qb Qc . Then we only need to
prove
(AB, AC; AR, AX) = (Qc , Qb ; AR ∩ Qb Qc , X) = −1.

We consider the complete quadrilateral (Pb∨ Pc∨ )(Qb Qc ), we have Pb∨ Qc ∩ Pc∨ Qb ∈ AR by (TODO 7.1.21).
Thus we have by harmonic property of complete quadrilaterals that AR, Pb∨ Pc∨ harmonically divide Qb Qc .

⋔
We define the cross conjugate of R and Q as R Q, as the trilinear pole of the perspector of the anticevian
triangle △R of R and the cevian triangle △Q of Q. Note that by the previous proposition we have

⋔
(R Q) ⋔ Q = R.

Definition 7.1.23. Let △P a P b P c , △Qa Qb Qc be the anticevian triangles of P and Q. We define the
cevapoint of P and Q as the pole of P Q in the conic (P P a P b P c QQa Qb Qc ), notated as P ⋆ Q. (This point
may be familiar to you in the context of Cevian nest theorem).

We further define the ceva conjugate of S and Q as the point P such that S = P ⋆ Q, notated as S/Q.

From the definition of crosspoints, we discover that the cevapoint P ⋆ Q is also the crosspoint of P wrt.
△P a P b P c and Q wrt. Qa Qb Qc . Further, we get the ceva conjugate S/Q is the perspector of the cevian
triangle of △S and △Q .

Proposition 7.1.24. Given △ABC and two points P, Q, let S = P ∗ Q. Then we have

(AP, AQ; AS, A(P Q ∩ BC)) = −1.

Proof. Let D be the conic (P P a P b P c QQa Qb Qc ) (called the bianticevian conic of P and Q). Since P Q∩BC’s
polar wrt. D is pD (P Q)pD (BC) = SA, we have

(AP, AQ; AS, A(P Q ∩ BC)) = (P, Q; SA ∩ P Q, P Q ∩ BC) = −1.

Example 7.1.25. Since

(I, J; ∞⊥BC , ∞BC ) = −1

(where I, J are the circle points), we have the cevapoint of I, J (suppose it’s S) has AS ⊥ BC. Doing this
symmetrically, we get the cevapoint of I and J is H, which is crazy. (So we can actually redefine H as the
cevapoint of I, J.) This further gives us that the bianticevian conic of I, J is the polar circle of △ABC, with

190
AoPS Chapter 7. Projective Space

center H. Since the pole of A wrt. the polar circle is just BC, we have that H is the center of the polar
circle, by Brokard’s.

Proposition 7.1.26. Let F be a focus of conic C. Let L be a line through F . Then F pC (L) is perpendicular
to L.

Proof. If we use the definition of a conic’s foci from (TODO 6.A.4), this is just the characterization

(F pC (L), L; F I, F J) = −1 =⇒ F pC (L) ⊥ L.

We can also define polarity with respect to a purely imaginary conic X , as the composition of polarity
across X and reflection across the center O of X (similar to how we defined inversion and polarity across a
imaginary circle back in Chapter 3.)

Practice Problems

Problem 1. Let O be the center of Γ. Prove that for any conic Ω, pΓ (Ω) is a conic with O as one of its foci.

(The polar of a conic is just the locus of the poles of its tangent lines.)

Problem 2. Given a fixed conic C and △ABC, prove that

pC (△ABC) := △pC (BC)pC (CA)pC (AB)

is perspective with △ABC.

Problem 3. Given △ABC and P as an arbitrary point, let △Pa Pb Pc be the pedal triangle of P wrt. △ABC.
Choose points D, E, F on P Pa , P Pb , P Pc such that

P Pa · P D = P Pb · P E = P Pc · P F =: k.

Prove that

• △ABC is perspective with △Pa Pb Pc ;

• When k varies, the locus of the perspector is a conic passing through A, B, C, P and the orthocenter H,

• When k varies, the envelope of the perspectrix is a parabola tangent to BC, CA, AB.

191
AoPS Chapter 7. Projective Space

Problem 4 (Extension of Newton’s 2nd Theorem). Let complete quadrilateral Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) have
inconic C. Let Aij = ℓi ∩ ℓj . If line L and Q’s diagonal Aij Akl intersect at Pij , choose Qij on Aij Akl such
that (Aij , Akl ; Pij , Qij ) = −1. Prove that

• Q14 , Q24 , Q34 are collinear;

• CQ14 Q24 Q34 ∈ L.

Problem 5. Given triangle △ABC, for a point P and a line ℓ, let C be the nine-point conic of A, B, C, P
wrt. ℓ (TODO 6.3.8). Then the trilinear pole of ℓ wrt. △ABC, the pole of ℓ wrt. C, and P are collinear.

Problem 6. Given two triangles △1 = △A1 B1 C1 , △2 = △A2 B2 C2 , A1 , B1 , C1 , A2 , B2 , C2 are conconic if

and only if there exists a conic C such that △1 , △2 are self-polar wrt. C.

Problem 7. Prove in (TODO 7.1.21) that Pb Q∨ ∨

c ∩ Pc Qb lies on Pa p(ABCP Q) (AR).

7.2 Involutions

Unfortunately, most of the deep theory in this section relies on (TODO 7.A), but you can get most of the
contest applications by interpreting ”pencil” as ”the set of lines through a point” or ”the set of points through
a line, including the point at infinity”.

Definition 7.2.1. Let X be a pencil (read (TODO 7.A)). We call ϕ ∈ Aut(X) an involution if ϕ2 = idX
and ϕ ̸= idX . In other words, ϕ has order 2 in the automorphism group.

Example 7.2.2. Let X be a line, then reflecting about a point P on X is an involution on the pencil of the
set of points on line X.

/ L∞ , then rotating 90◦ degrees around

Example 7.2.3. Let X be TP , the set of lines through a point P ∈
P is also an involution.

Proposition 7.2.4. Let X be a pencil, if ϕ is an automorphism and there exists one point A such that
ϕ(A) ̸= A, ϕ(ϕ(A)) = A, then ϕ has to be an involution.

Proof. Note that ϕ ̸= idX since A is not fixed. Then for any P ∈ X, we have

(A, ϕ(A); P, ϕ(P ))X = (ϕ(A), A; ϕ(P ), ϕ(ϕ(P )))X = (A, ϕ(A); ϕ2 (P ), ϕ(P ))X .

So ϕ2 (P ) = P , and ϕ must be an involution.

192
AoPS Chapter 7. Projective Space

Remark. Note that every involution must have two fixed points (for a proof, see chapter 7.A and prove it
for the projective line, via inversion. There is also a nice proof with harmonics with the circle points.).

Suprisingly, given two fixed points A, B of an involution is enough to uniquely determine it. Suppose
ϕ(A) = A, ϕ(B) = B, then ϕ sends element P to its harmonic conjugate in A, B.

Additionally, given two (real or complex) pairs of points that swap under an involution, the involution is
uniquely determined.

Theorem 7.2.5 (Desargues’s Involution Theorem (DIT)). Given complete quadrangle ⨿ = (P1 , P2 , P3 , P4 )
and a line ℓ not going through the vertices, let Qij = Pi Pj ∩ ℓ, then

• There exists exactly one involution on ℓ, ϕ ∈ Aut(ℓ), such that

(Q23 , Q14 ), (Q31 , Q24 ), (Q12 , Q34 )

are reciprocal pairs;

• For the pencil (set) of conics through P1 , P2 , P3 , P4 that intersect ℓ at A, B, (A, B) are interchanged by
the previously defined involution ϕ. Further, the converse of this also holds; for two points A, B on ℓ,
P1 , P2 , P3 , P4 , A, B are conconic (including degenerate) if and only if (A, B) swap under ϕ.

We will call the involution ϕ given by (i) as the DIT involution from ℓ cutting ⨿.

Proof. Trivial by cross ratio chase

Remark. Another proof of this is by first categorizing all involutions on a conic as the ”second intersection
map” wrt. a fixed point, which is provable by a projective transformation sending the conic to a circle.

Example 7.2.6 (Harmonic Bundles from Complete Quadrilaterals). Let ℓ be the line through P3 P1 ∩ P2 P4
and P1 P2 ∩ P3 P4 . Then the DIT involution from ℓ has P3 P1 ∩ P2 P4 and P1 P2 ∩ P3 P4 as fixed points, so we
immediately get the involution on ℓ is just harmonic conjugation in these two points, which gives us (TODO
2.2.8), the harmonic bundle property of a quadrilateral.

Example 7.2.7. Let H and Ω be the orthocenter and circumcircle respectively, and let Ma be the midpoint
of BC, let Xa be the second intersection of AMa with Ω, and let Ya be the intersection of ray Ma H with Ω.
Similarily define Xb , Yb , Xc , Yc . Prove that Xa Ya , Xb Yb , Xc Yc concur on the Euler line.

Solution. DIT on Euler line cutting quad A, A-antipode, Xa , Ya .

Example 7.2.8 (Pascal’s Octagrammum Mysticum). Let A, A′ , B, B ′ , C, C ′ , D, D′ be eight conconic problems.

Consider the sixteen intersections of the two complete quadrilaterals (AA′ )(BB ′ ) and (CC ′ )(DD′ ). Prove
that:

193
AoPS Chapter 7. Projective Space

• The eight intersection points with an odd number of apostrophes are conconic;

• The eight intersection points with an even number of apostrophes are also conconic.

Solution. Let’s set up some notation. Let I be an ”iterator” from {1, 2, 3, 4}, and let PI be Ai B j ∩ C k Dl ,
where royce latex this crap and j, k, l are defined cyclically but starting at 2, 3, 4 instead of 1. For example,
P134 = A′ B ∩ C ′ D′ , P12 = A′ B ′ ∩ CD. We only need to prove that P, P12 , P13 , P23 , P14 , P24 are conconic, and
the rest follows. Consider the complete quadrangle (A, A′ , B, B ′ ) and line CD across it, since by Pascal we
have A, A′ , B, B ′ , C, D are conconic, we have by DIT that

(C, D), (P1 , P2 ), (P, P12 )

are pairs under the DIT-given involution φ. Then consider complete quadrangle (P13 , P14 , P23 , P24 ) and
line CD, then by DIT we have that (C, D), (P1 , P2 ) are also pairs. Since two pairs uniquely determine an
involution, P13 , P14 , P23 , P24 , P, P12 = φ(P ) are conconic.

Remark. A really funny proof of this is by using quartic Cayley-Bacharach.

Note that (TODO 7.2.5) is completely projective, so let’s look at its dual:

Theorem 7.2.9 (Dual of Desargues’s Theorem, DDIT). Given complete quadrilateral Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 )
and a point P that doesn’t lie on any of the component lines of Q, define points Lij = (ℓi ∩ ℓj )P, then

• There exists a unique involution φ in the pencil of lines through P (TP ) such that

(L23 , L14 ), (L31 , L24 ), (L12 , L34 )

swap under φ,

• For any conic C tangent to all of ℓ1 , ℓ2 , ℓ3 , ℓ4 , the two tangents from P to this conic also swap under
φ. Similarily to DIT, the converse also holds; for two lines s, t through point P , ℓ1 , ℓ2 , ℓ3 , ℓ4 , s, t are
tangent to a common conic (possibly degenerate) if and only if (s, t) swap under φ.

We will call the involution ϕ given by (i) as the DDIT involution from point P with respect to Q.

Remember the nine-point (or nine-line) conic? (TODO 6.3.10). The involution given by (D)DIT is
actually just polarity with respect to this conic!

Proposition 7.2.10. Let φ be the involution given by DIT, and let C be the ℓ-nine point conic of complete
quadrangle ⨿. Then (A, B) swap under φ if and only if A, B are conjugate points in C.

194
AoPS Chapter 7. Projective Space

Proof. Let ϕ′ = [A → pC (A) ∩ ℓ] ∈ Aut(ℓ), let Rij be the point such that (Pi , Pj ; Qij , Rij ) = −1. Then we
know that
Q23 = R12 R31 ∩ R24 R34 , Q14 = R12 R24 ∩ R31 R34 ,

so by Brokard, etc, we have that Q23 , Q14 are conjugate in C, or ϕ′ (Q23 ) = Q14 , ϕ′ (Q14 ) = Q23 . Similarly we
have ϕ′ (Q31 ) = Q24 , and since involutions are determined by two pairs, ϕ = ϕ′ . Thus (A, B) swap under ϕ if
and only if ϕ′ (A) = B, equivalent to A, B being conjugate in C.

Proposition 7.2.11. Let φ be the involution given by DDIT, and let C be the P -nine line conic of complete
quadrangle ⨿. Then (s, t) swap under φ if and only if s, t are conjugate lines in C.

This gives us a corollary:

Corollary 7.2.12. The fixed points of ϕ (DIT) in ℓ are the two intersections of C and ℓ.

Corollary 7.2.13. The fixed lines of ϕ (DDIT) in TP are the two tangents from P to C.

This characterization of all involutions has a useful property; we can actually prove some pencil autoomor-
phisms are involutions.

Proposition 7.2.14 (Desargues’s Assistant Theorem). Let ℓ be a line, and let ϕ : ℓ 7→ ℓ be a automorphism,
then ϕ is an involution if and only if it’s some inversion centered at ℓ (possibly a negative inversion or just a
reflection, also.)

Proof. We have proved this before in Chapter 3. One way to do it is with knowledge of Mobius transformations,
but a straight length-bash to prove cross ratios being preserved under inversion is obviously possible. This is
left as an exercise to the reader.

The previously mentioned ”nice” proof with harmonics with the circle points will now be stated.

Work in CP2 , the complex projective plane. (It is highly recommended to read 6.A).

Consider the involution given by the complete quadrangle (P1 , P2 , I, J) and line ℓ, where I, J are the two
circle points,
P1 = QI ∩ RJ, P2 = Q′ J ∩ R′ I,

such that (Q, Q′ ), (R, R′ ) are two pairs that swap under involution ϕ. This tells us that for any involutive
pair (P, P ′ ), we have P1 , P2 , P, P ′ , I, J lie on one conic, which just means P1 , P2 , P, P ′ are concyclic; in other
words, ϕ is just inversion with power AP1 · AP2 at A = P1 P2 ∩ ℓ.

Proposition 7.2.15 (Involutions on a Conic). Let C be a conic, and let ϕ : C → C be a projective map from
C to itself, then ϕ is an involution if and only if there exists a point A ∈
/ C such that A, P, ϕ(P ) are collinear.

195
AoPS Chapter 7. Projective Space

Proof. We have also proved this before. Here’s a purely cross-ratio based method. [stuff]

A direct corollary of this is (TODO 3.3.2)

Corollary 7.2.16. Inversion preserves cross ratio.

Dually:

Proposition 7.2.17. Let C be a conic, and let φ : TC → TC be a transformation from the pencil (set) of
lines tangent to C. Then φ is an involution on this pencil if and only if there exists a unique line K ∈
/ TC
such that for all S ∈ TC, K, S, φ(S) are concurrent.

So if we want to prove that the three lines formed by six points on a fixed conic are concurrent, is
equivalent to proving that they’re three pairs under an involution on the conic. (dually, if we want to prove
six lines tangent to a conic that intersect at three points are collinear, we can just prove it’s an involution as
well.) This lets us use projective maps/transformations to send the six points to others on the conic, as we
will show in the following example.

Example 7.2.18. Let Ω be the circumcircle of △ABC, and let M be a point on arc BC of Ω. Let B ′ , C ′
respectively lie on CA, AB such that BB ′ , CC ′ are both parallel with AM . Let M B ′ , M C ′ respectively
intersect Ω again at P, Q. Finally, let S be the intersection of P Q and BC. Prove that AS is the tangent to
Ω at A.

Solution. We want to prove that the tangent at A, P Q, BC are concurrent. This is equivalent to proving
that there’s an involution on Ω that swaps

(A, A), (P, Q), (B, C)

, where P, Q are the second intersections of M B ′ , M C ′ with Ω. So by DDIT on complete quadrilateral

(AB, CA, BB ′ , CC ′ ) wrt. point M , we win (note that BB ′ ∩ CC ′ = ∞AM ).

Since the two circle points I, J lie on the line at infinity L∞ by definition, we can get a way to prove
P1 , P2 , P3 , P4 are concyclic.

Proposition 7.2.19 (DIT on the Line at Infinity). Four points P1 , P2 , P3 , P4 are concyclic if and only if
the involution φ given by DIT on L∞ and complete quadrangle (P1 , P2 , P3 , P4 ) has two fixed points ∞I , ∞J
such that for any finite point P , ∠IP J = 90 ◦ .

Proof. Follows by 6.A, we want to prove that P1 , P2 , P3 , P4 , I, J are conconic. Harmonic conjugation in I, J
on the line at infinity is just rotation by 90 degrees on the line at infinity, so this is our involution φ. Let

196
AoPS Chapter 7. Projective Space

us project this involution on the line at infinity onto the pencil of lines through a finite point X, for easier
visualization. Then we want to prove that

(XP1 , XP3 ), (XP2 , XP4 ), (XI, XJ)

swap by rotating 90 degrees. Since (XP1 , XI, XP3 , XJ) is harmonic and (XI, XJ) is 90 degrees, this is true
by the harmonic conjugate definition, given two fixed points.

Proposition 7.2.20. Given complete quadrangle ⨿ and line ℓ, let F1 , F2 be the fixed points of the DIT
involution φ. Then for any circumconic C of ⨿, F1 , F2 are conjugate wrt. C.

Proof.

Example 7.2.21. By setting ℓ as the line between P3 P1 ∩ P2 P4 and P1 P2 ∩ P3 P4 , we get the duality property
(La Hire’s) of polars.

Dually, we have

Proposition 7.2.22. Given complete quadrilateral Q and point P , let f1 , f2 be the fixed lines of the DDIT
involution φ. Then for any inconic C of Q, f1 , f2 are conjugate wrt. C.

Proposition 7.2.23. Let P and Q have cevapoint S in △ABC. Then P and Q are conjugate in a circumconic
C of △ABC if and only if C passes through S.

Proof. harmonic conjugates in the intersection

Practice Problems

Problem 1 (2017 China TST 2 P3). Let ABCD be a quadrilateral, and let ℓ be a line that intersects
AB, CD, BC, DA, AC, BD at X, X ′ , Y, Y ′ , Z, Z ′ . Prove that the circles with diameters XX ′ , Y Y ′ , ZZ ′ are
coaxal.

Problem 2. For △ABC and three points P, Q, R, let D, E, F respectively be the intersections of QR with
BC, CA, AB, let AP and QR intersect at A1 , and let A2 be the point on QR such that

(Q, R; F, A1 ) = (R, Q; E, A2 ),

and similarily define B2 , C2 . Prove that AA2 , BB2 , CC2 are concurrent.

Problem 3 (Circum-ex config). Let I a be the the A-excircle of triangle △ABC. Let the common tangents
from I a and (ABC) intersect BC at X, Y . Prove that AX, AY are isogonal lines in ∠A.

197
AoPS Chapter 7. Projective Space

Problem 4. Given △ABC, let D be a point on side BC, let I, I1 , I2 be the incenters of △ABC, △ABD, △ADC,
let M, N respectively be the second intersection of (ABC) with (IAI1 ), (IAI2 ). Prove that when D moves
on BC, M N passes through a fixed point.

Problem 5. Let I b , I c be the B and C excenters of triangle △ABC. Let ω be the incircle of △ABC, and let
D be the A-intouch point. Let DI b , DI c intersect ω again at X, Y , prove that AD, BX, CY are concurrent.

Problem 6 (Taiwan 2019 Test 1 I3-2). Given convex pentagon ABCDE, let A1 be the intersection point of
sides BD and CE, let B1 be the intersection point of CE and DA, and similarily define C1 , D1 , E1 . Let A2
be the second intersection of (ABD1 ) with (AEC1 ), let B2 be the second intersection of (BCE1 ), (BAD1 ),
and similarily define C2 , D2 , E2 . Prove AA2 , BB2 , CC2 , DD2 , EE2 are concurrent.

Problem 7. Let I b , I c be the B, C-excenters of acute triangle △ABC, and let D be a point on the circumcircle
(ABC) such that AD ⊥ BC. Let DI b , DI c intersect (ABC) again at Y, Z, and let X = Y Z ∩ BC. Prove
that the triangle formed by lines AX, AO, BC is isosceles.

Problem 8 (2022 EGMO P6). Let O be the circumcenter of cyclic quadrilateral ABCD, let X be the
intersection of the angle bisector of angle A and angle B, and let Y be the intersection of the angle bisector
of B and C, and define Z, W similarily. Let P be the intersection of AC and BD. Suppose X, Y, Z, W, O, P
are all distinct. Prove that O, X, Y, Z, W are concyclic if and only if P, X, Y, Z, W are concyclic.

7.3 Isogonal Conjugation

We have already talked about isogonal conjugation (see (TODO 1.3)), but this time we want to understand
the connection with isogonal conjugation and conic sections. Remember this property (see (TODO 1.3.9)):

Proposition 7.3.1. For complete n-gon N (ℓ∞ , ℓ∈ , . . . , ℓ\ ) and a point P , let Pi be the foot from P to ℓi .
Then P has an isogonal conjugate in the n-gon N if and only if P1 , P2 , . . . , Pn are concyclic.

Also remember that if P ∗ is the isogonal conjugate of P wrt. N , then the pedal polygon of P ∗ and P are
concyclic at the midpoint of P P ∗ . Additionally P ∗ is the other center of orthology as well. Here’s how we
will relate isogonal conjugation to conic sections.

Theorem 7.3.2. For complete n-gon N (ℓ∞ , ℓ∈ , . . . , ℓ\ ) and two points P, P ∗ , then (P, P ∗ ) are a pair of
isogonal conjugates in N if and only if there exists an inconic C with (P, P ∗ ) as foci, internally tangent to N .

Proof. Let P1∗ , P2∗ , . . . , Pn∗ respectively be the reflections of P ∗ across the sides ℓ1 , ℓ2 , . . . , ℓn . Then Pn∗ are all
concyclic with circumcenter at P . For the ”if” direction, let Ti be the intersection of P Pi∗ with ℓi . Then
P Ti ± P ∗ Ti = P Pi∗ is constant (with the ± depending on signed length whether P is outside △ABC or not).

198
AoPS Chapter 7. Projective Space

This is the exact definition of an ellipse/hyperbola, so now we claim the conic with P, P ∗ through all Ti is our
desired inconic. This is true because of (TODO 6.1.6); ℓi is the angle bisector of ∠P Ti P ∗ , so C is tangent to
all ℓi and is thus an inconic of N . For the ”only if” direction, let Ti = Tℓi C be the touchpoint of the inconic
on side ℓi . Then apply the same argument in the ”if” direction (circumcenter of Pi∗ is P ) to construct two
isogonal conjugates P, P ∗ .

A more morally correct proof is through the use of (TODO 6.A.4)’s definition of the foci of a conic.
Suppose we have an inconic C of n-gon N , with foci P, P ∗ . Then by DIT, there exists an involution swapping

(Aij P, Aij P ∗ ), (Aij I, Aij J), (ℓi , ℓj )

for all i, j, where I, J are the circle points. Thus ∡Aij P + ∡Aij P ∗ = ℓi + ℓj .

For the converse, consider DDIT on the inconic C of P I, P J, P ∗ I, P ∗ J, ℓ1 . Thus there’s an involution
swapping
(A1j P, A1j P ∗ ), (A1j I, A1j J), (ℓ1 , ℓj )

So ℓj is tangent to C as well, for all j.

In the triangle case of this theorem, we have:

Proposition 7.3.3. Let (P, P ∗ ) be a pair of isogonal conjugates in △ABC, and let C be an inconic with
P, P ∗ as foci. Let T be the BC-touchpoint of C. Let X be AP cap BC, then

BT BX BP 2
× =( ) .
TC XC PC

Before proving this, we need a small lemma:

Lemma 7.3.4. Let (P, P ∗ ) be a pair of isogonal conjugates in △ABC, and let Pa∗ be the reflection of P ∗
across BC. Then (A, Pa∗ ) are isogonal conjugates in △P BC.

Proof. Alternatively, DDIT at P also works.

Proof of 7.3.3. By the optical property of an ellipse / Heron’s distance problem, the ellipse with (P, P ∗ ) as
foci is tangent to BC at the intersection of P Pa∗ with BC. Additionally, P X, P T are isogonal in ∠BP C. So
by (TODO 1.3.2), we have
 2
BT BX BP
× = .
TC XC PC

199
AoPS Chapter 7. Projective Space

Since T is defined symmetrically wrt. P, P ∗ , we also have

2
BX ∗ BP ∗

BT
× ∗ = .
TC X C P ∗C

Multiplying these two expressions and using (TODO 1.3.2) we get:

Corollary 7.3.5. Let (P, P ∗ ) be a pair of isogonal conjugates in triangle △ABC, and let C be a inconic of
△ABC with (P, P ∗ ) as foci. Let T be the BC-touchpoint, then

BT CA BP BP ∗
=± × × ∗ ,
TC AB PC P C

where the sign is up to whether P, P ∗ are in ∠BAC.

Combining this with (TODO 7.1.26), we get

Proposition 7.3.6. Let P be one of the foci of inconic C of complete n-gon N = (ℓ1 , . . . , ℓn ), tangent to
each side ℓi at Ti . Choose a point Si on every ℓi such that ∡Si P Ti = 90◦ . Then S1 , . . . , Sn are collinear.

Proof. In reality, from (TODO 7.1.26), Si = pC Ti ∩ P ∞⊥P Ti lies on the polar of P wrt. C.

Proposition 7.3.7. For complete n-gon N = (ℓ1 , . . . , ℓn ), let T1 , . . . , Tn be n non-collinear points on lines
ℓ1 , . . . , ℓn . Then the following statements are equivalent:

• There exists a conic C such that C is tangent to every ℓi at Ti .

• There exists a point P such that where if we define Aij = ℓi ∩ ℓj , line P Aij bisects angle ∠Ti P Tj .

Then P is also one of the foci of the conic.

Proof. Proved analogously to the triangle case.

You may be curious about how ”isogonal conjugation” works in quadrilaterals. We will elaborate on this
in Chapter (TODO 9) with the isoptic cubic, but currently we only need this small result.

Proposition 7.3.8. The midpoint of any isogonal conjugates in a complete quadrilateral on Q lies on the
Newton line.

Proof. These are foci of an inconic, so Newton’s second theorem (TODO 6.3.13).

Through (TODO 4.1.11) we have:

200
AoPS Chapter 7. Projective Space

Theorem 7.3.9. The isogonal conjugate of the Miquel point in a complete quadrilateral is the point at
infinity along the Newton line.

Proof. Let Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ). Consider the Steiner line of this quadrilateral(the line through reflections of
Miquel point wrt. all four sides). The ”circumcenter” of these four reflected points is the point at infinity
perpendicular to the Steiner line, which is the point at infinity along the Newton line ∞τ . An alternative
(funnier) proof of this is consider the unique parabola tangent to all four sides of the quadrilateral (QL-Co1).
By the optical property of the parabola, the reflection of the focus of this parabola across all four tangent
lines must lie on the directrix. So the focus and directrix are the Miquel point and Steiner line respectively.
Now interpret this parabola as an ellipse with one foci at the Miquel point and the other foci at ∞τ .

We will provide a very simple proof of this in chapter 9 with Clawson-Schmidt conjugation (inversion at
the Miquel point then reflection over ∠AM C), which swaps isogonal conjugates in a quadrilateral.

Back to triangles. Let φK represent isogonal conjugation in triangle △ABC (this notation will be
motivated in the next section). For some object/set of points X, φK (X) will represent the curve formed by
isogonally conjugating all points in X.

Proposition 7.3.10. Let ℓ be a line and let φK represent isogonal conjugation in △ABC. Then

If ℓ passes through a vertex V , then φ(ℓ) also goes through V .

Otherwise, φ(ℓ) is a circumconic of △ABC.

Proof. xiooix

An alternate proof is with moving points (TODO 7.B): isogonal conjugation of a moving point generally
doubles degree. (For proving that the isogonal conjugate of a circumconic is a line, simply note that it passes
through 3 vertices, and by Zack’s lemma, 2 ∗ 2 − 3 = 1.)

Proposition 7.3.11. Given △ABC, let I, I a , I b , I c be the incenter and three excenters. Let

F = {D | I, I a , I b , I c ∈ D}

be the pencil of conics through I, I a , I b , I c . Then two isogonal conjugates (P, P ∗ ) will be conjugate points in
any conic D in this pencil.

Proof. Assume P ̸= P ∗ . Consider the DIT (TODO 7.2.5)-given involution on complete quadrangle
(I, I , I , I ) cut by line P P . This involution ψ ∈ Aut(P P ∗ ) swaps
a b c ∗

(II a ∩ P P ∗ , I b I c ∩ P P ∗ ), (II b ∩ P P ∗ , I c I a ∩ P P ∗ ), (II c ∩ P P ∗ , I a I b ∩ P P ∗ )

201
AoPS Chapter 7. Projective Space

Since P, P ∗ are fixed under this involution, (TODO 7.2.20) gives us that they are conjugate points in all
circumconics D of this quadrangle.

Remark. This leads to an alternate definition of isogonal conjugates as the common point of the polar of P
wrt. all conics in this pencil. We will elaborate on this in the following section, with general isoconjugations.

Practice Problems

Problem 1. Denote φ to be isogonal conjugation in △ABC. Denote ℓ to be a straight line that does not
pass through the points A, B and C. Prove that A(ℓ ∩ BC) and TA φ(ℓ) are isogonal lines in ∠BAC.

Problem 2 (2000 ISL G3). Let O be the circumcenter and H the orthocenter of an acute triangle ABC.
Show that there exist points D, E, and F on sides BC, CA, and AB respectively such that

OD + DH = OE + EH = OF + F H

and the lines AD, BE, and CF are concurrent.

Problem 3. Let (P, P ∗ ) and (Q, Q∗ ) be two pairs of isogonal conjugate points about △ABC such that
P P ∗ ∥ QQ∗ . Denote I to be the incenter and I a , I b , I c to be the three excenters of △ABC. Denote M and
N to be midpoints of P P ∗ and QQ∗ respectively. Prove that I, I a , I b , I c , M, N are conconic.

Problem 4 (Saragossa’s Theorems). For point P and triangle △ABC, let △Pa Pb Pc be its cevian triangle
and let △Pa′ Pb′ Pc′ be its circumcevian triangle.

• Let U = Pb Pc′ ∩ Pc Pb′ , V = Pc Pa′ ∩ Pa Pc′ , W = Pa Pb′ ∩ Pb Pa′ . Prove that AU, BV, CW concur at a point
Sa1 (P ), and that Sa1 (P ) lies on conic (ABCKP ), where K is the symmedian point of △ABC.

• Prove that Pa U, Pb V, Pc W concur at a point Sa2 (P ), and K, Sa1 (P ), Sa2 (P ) are collinear.

• Prove that Pa′ U, Pb′ V, Pc′ W concur at a point Sa3 (P ), and that Sa3 (P ) is the crosspoint of P, Sa1 (P )
wrt. △ABC, and also that P, Sa2 (P ), Sa3 (P ) are collinear.

This also holds for redefining the circumcevian triangle with any circumconic C, and redefining the symmedian
point as the perspector of C and △ABC.

7.4 General Isoconjugations

In this section, we will generalize isogonal conjugation, which will be notated as ϕK (P ) = P ∗ (in here, K
means symmedian-point, which we will explain). We will use the projective properties of isogonal conjugation
we developed in (TODO 7.3.11).

202
AoPS Chapter 7. Projective Space

Definition 7.4.1. Given △ABC, we can say that

ϕ : P2 \(BC ∩ CA ∩ AB) → P2 \(BC ∩ CA ∩ AB)

is an isoconjugation on points if there exists at least two elements in the pencil of conics F = Fϕ through
four points such that

• For any conic D in this pencil (a diagonal conic of the isoconjugation), △ABC is self-conjugate in D.

• For any point P and any conic D in the pencil, P, ϕ(P ) are conjugate in D.

Remark. Another interpretation of this is for four points and the pencil of conics passing through them, the
function sending P to the common point of all the polars of P in this pencil of conics is an isoconjugation.

Proposition 7.4.2 (Fundamental Theorem of Isoconjugations). Given a fixed triangle △ABC, a isoconjuga-
tion on points ϕ is uniquely determined by two distinct diagonal conics in the pencil of this isoconjugation
D0 , D∞ . (This is equivalent to saying pD0 (P ) ̸= pD∞ (P ).)

Proof. We consider the transformation Φ(P ) = pD∞ (pD0 (P )), then we have Φ(A) = A, Φ(B) = B, Φ(C) = C.
Suppose there exists a P0 ̸= BC ∩ CA ∩ AB such that Φ(P0 ) = P0 . However, this transformation can only
have three fixed points by (T ODO7.A.10), so thus if this P0 exists, we have Φ = id, so for any point P in the
plane, we have pD0 (P ) = pD∞ (P ), so D0 = D∞ = 0, contradiction.

Definition 7.4.3. Given △ABC, we say

ϕ : (P2 )∨ \(TA ∩ TB ∩ TC) ∈ (P2 )∨ \(TA ∩ TB ∩ TC)

is a isoconjugation on lines if there exists two elements of the pencil of conics F through four points such
that

• For any D ∈ F (a diagonal conic, we have that △ABC is self-conjugate in D.

• For any line ℓ ∈

/ (TA ∩ TB ∩ TC) we have that ℓ, ϕ(ℓ) are isoconjugates in D.

Similarily, any conic satisfying these properties must go through these four points as well.

Remark. By duality, we can only address theorems on isoconjugation on points.

Additionally, in barycentric coordinates, we know that isoconjugation is a transformation in the form

 
u v w
[x : y : z] → : : = [uyz : vxz : wxy]
x y z

203
AoPS Chapter 7. Projective Space

(we will prove this later). We call P = [u : v : w] the pole of its isoconjugation, and note that P = ϕ(G). For
example, isogonal conjugation is an isoconjugation with pole K, the symmedian point, which justifies our
previous notation ϕK (). Similarily, for an isoconjugation on lines, we define ϕ(L∞ ) to be the polar of this
isoconjugation. For example, isotomic line conjugation’s polar is just L∞ .

Similarily to isogonal conjugation, we have

Proposition 7.4.4. Given △ABC, let ϕ be a isoconjugation on points and let ℓ be a line.

• If ℓ goes through one of A, B, C, suppose X, then ϕ(ℓ) is a line that also passes through X.

• With respect to all the vertices A, B, C, then

ϕ
TX → TX

XP → Xϕ(P )

is an involution. (Note that this is well-defined by part (i)).

• If ℓ does not go through the three vertices, then ϕ(ℓ) is a circumconic of △ABC.

Proof. go shit yourself

Corollary 7.4.5 (Isoconjugations Give More Isoconjugations). For an arbitrary △ABC and point isoconju-
gation ϕ, for two points P , Q, let R = P Qϕ ∩ P ϕ Q, S = P Q ∩ P ϕ Qϕ . Then S = Rϕ .

Proof. By the above listed propositions we know that A(B, C), A(P, P ϕ ), A(Q, Qϕ ), A(R, Rϕ ) define an
involution on the pencil TA. By DDIT (TODO 7.2.9) we have A(P, P ∗ ), A(Q, Q∗ ), A(R, S) also define an
involution on the pencil TA, so thus ARϕ = AS. Applying this to B, C we get S = Rϕ .

Proposition 7.4.6 (Projective characterization of barycentric product). Define × as the termwise product
of the barycentric coordinates of two points P and Q. Let P ∗ , Q∗ be the isogonal conjugates of P, Q, and let
R = P Q∗ ∩ P ∗ Q. Then the following statements are equivalent:

• P × P ∗ = Q × Q∗ ;

• P, Q, R lie on a common circumconic of △ABC

• P ∗ , Q∗ , R∗ lie on a common circumconic on △ABC;

• There exists a common inconic tangent to the three sides of △ABC and the four sides of the quadrilateral
(P P ∗ )(QQ∗ ).

204
AoPS Chapter 7. Projective Space

• There is an involution on the pencil of lines TA swapping (AB, AC), (AP, AP ∗ ), (AQ, AQ∗ ) (and
similarily for B,C).;

Proof. Note that R∗ = P Q ∩ P ∗ Q∗ . The main lemma is that if two pairs of points have the same barycentric
product, then there is some isoconjugation swapping both pairs (the proof is by the bary definition of the
isoconjugations found in Chapter 12). This gives us that (i) implies (v) automatically. Then by the converse
of DDIT, this gives us that (i) implies (iv). For (ii), since (AB, AC), (AP, AP ∗ ), (AQ, AQ∗ ), (AR, AR∗ )
swap under an involution (TODO 7.4.5), and (BC, BA), (BP, BP ∗ ), (BQ, BQ∗ ), (BR, BR∗ ) swap under an
involution, we have

A(C, P ; Q, R) = A(B, P ∗ ; Q∗ , R∗ ) = B(A, P ∗ ; Q∗ , R∗ ) = B(C, P ; Q, R),

and thus P, Q, R and A, B, C are conconic. (ii) implies (iii) (and thus (i) implies (iii)) because

A(P ∗ , Q∗ ; B, C) = A(P, Q; C, B) = R(P, Q; C, B)

= R(Q∗ , P ∗ ; C, B) = R(P ∗ , Q∗ ; B, C),

so P ∗ , Q∗ , R lie on a common circumconic of △ABC. Finally, (iii) implies (i) because if we let △Ra Rb Rc be
the cevian triangle of R, then

A(P, Q∗ ; B, C) = (P, Q∗ ; Rc , Rb )ABCR = (Q, P ∗ ; Rc , Rb )ABCR = A(Q, P ∗ ; B, C).

So there’s an involution swapping (AB, AC), (AP, AP ∗ ), (AQ, AQ∗ ).

Here’s a very important result that we’ll use in (TODO 12.2):

Theorem 7.4.7. Given △ABC, let L represent the set of all lines not passing through either of A, B, C.
Let C represent the set of all circumconics of △ABC. Then we have the following perfect pairing: EVIN
TYPE THIS, TOO MANY ARROWS where φL×C satisfies φL×C (L) = C. In other words, for any line not
passing through a vertex of △ABC, we have the following one-to-one correspondence between isoconjugations
and circumconics: TYPE THIS ALSO EVIN

Remark. A perfect pairing is a bijection from the Cartesian product of two sets (i.e pairs of choosing an
element from both sets) to another set.

Proposition 7.4.8. Given △ABC and φ a point isoconjugation on △ABC, then for an arbitrary point P,
φ(P ) is the perspector of △ABC and circumconic pφ(t(P )) (△ABC), where t(P ) is the trilinear polar of P .

Proof.

205
AoPS Chapter 7. Projective Space

Choosing P as the centroid G, we get

Corollary 7.4.9. ?

Proposition 7.4.10. Consider a transformation φ from

φ : P2 \BC ∪ CA ∪ AB → P2 \BC ∪ CA ∪ AB

that has for all vertices X ∈ A, B, C,

X, P1 , P2 collinear =⇒ X, φ(P1 ), φ(P2 )collinear

and [XP → Xφ(P )] is an involution swapping (XY, XZ)), where Y, Z are the other two vertices. Then φ
has to be an isoconjugation.

Proof. Obviously we have that φ2 = id, so φ is an involution. Choose some point P0 that’s not one of the
vertices of the triangle, or a fixed point of φ. Then note that φ is completely determined by where it sends
P0 ; for any vertex X, we have

X(Y, Z; P0 , P ) = X(Z, Y ; φ(P0 ), φ(P )), φ(P ) = Bφ(P ) ∩ Cφ(P ),

where Y, Z are the two other vertices that aren’t X. To prove that this is an isoconjugation, we need to
find (at least) two conics D′ , D∞ such that △ABC is self-conjugate in both of these conics and φ(P ) is the
intersection of the polars of P in D′ , D∞ (see (TODO 7.4.2)). Since the map on the pencil of lines through
X sending XP → Xφ(P ) is a projective map, we only need to satisfy

φ(P0 ) = pD′ P0 ∩ pD∞ P0 .

So how do we find these two conics? Choose D′ to be the conic through P0 and tangent to the line P0 φ(P0 )
such that △ABC is self-conjugate in D′ , and choose D∞ to be defined similarily except passing through
φ(P0 ) instead of P0 . Since P0 ̸= φ(P0 ), these conics are distinct. Thus since

pD′ P0 ∩ pD∞ P0 = P0 φ(P0 ) ∩ TD∞ (φ(P0 )) = φ(P0 )

we are done.

Remark. In the proof, if P0 is a fixed point of the isoconjugation (P0 = φ(P0 )), then just pick any two
conics in the pencil of conics
F = {D | P0 , P0a , P0b , P0c ∈ D},

206
AoPS Chapter 7. Projective Space

where P0a , etc are the vertices of the anticevian triangle of P in △ABC. (See (TODO 7.4.16)).

Proof of 7.4.7.

Definition 7.4.11. Given △ABC and a line L not passing through the vertices of ABC, let represent the
points on L not on the sides of △ABC. Then we define

• For any isoconjugation φ, we define Lφ = φ(L).

• For any circumconic C, we define LC as the isoconjugation obtained by the previously defined bijection
between circumconic+lines and point isoconjugations.

• For a line isoconjugation φ, we define φ

as the envelope of φ(T).

• For an arbitrary inconic ⌋, we define as the line isoconjugation obtained by the previously defined
bijection between inconics + lines and line isoconjugations.

K
As an example of this notation, the circumcircle Ω of △ABC is notated as Lφ
∞ , where φ
K
represents
isogonal conjugation. For a point isoconjugation ϕ, we still have one goal we haven’t accomplished; which is
writing out all possible diagonal conics of isoconjugations. Due to the bijection from (TODO 7.4.7), we can
replace φ with a line not passing through the vertices of △ABC and a circumconic C = Lφ . By definition,
for a conic D ∈ F , we always have pD L ∈ C = Lφ . Then why can’t we simplify this construction for a point
O ̸= A, B, C on C since we can construct DO such that L is the polar of O in DO ? (Note that DO sends
A, B, C, O respectively to BC, CA, AB, L, so DO is uniquely determined.) In reality, it’s better to construct
DO like this (note that it can be possibly imaginary): Let △OA OB OC be the cevian triangle of O in △ABC.
Consider the involution φA ∈ Aut(AO) such that φA (A) = OA , φA (O) = AO ∩ L. Let FA1 , FA2 be the two
fixed points of this involution φA , and similarily define FB1 , FB2 , FC1 , FC2 .

Proposition 7.4.12. DO is the common conic through FA1 , FA2 , FB1 , FB2 , FC1 , FC2 , and △ABC is self-
conjugate in this conic with pDO (O) = L.

Proof. Take a projective transformation/homography sending L to L∞ , to make our notation easier. Then
since FA1 , FA2 are fixed points under this involution, we have

(A, OA ; FA1 , FA2 ) = (O, ∞AO ; FA1 , FA2 ) = −1.

Therefore O is the midpoint of FA1 FA2 . Further, FA2 is one of the vertices of the anticevian triangle of FA1
′
wrt. △ABC (let’s call the other vertices FA3 , FA4 ). Let FA3 be the reflection of FA3 across O, then the
′
conic DOA := (FA1 FA2 FA3 FA4 FA3 ) satisfies the condition of having △ABC as its self-conjugate triangle
′
and having the polar of O as the line at infinity L∞ (note that FA3 ̸= FA4 ). Similarily, we can define

207
AoPS Chapter 7. Projective Space

DOB , DOC which also satisfy the above characteristics, but by uniqueness of conics through six points, we
have DOA = DOB = DOC , and thus this is our desired DO .

So we can say that the pencil of diagonal conics of φ, F = Fϕ is just

{DO | O ∈ Lφ }.

You might be familiar with the concept of cross ratio on a pencil of conics. If not, read section 7.A. We define
a cross ratio on this pencil here through Lφ :

(DO• ) = (DO1 , DO2 ; DO3 , DO4 )F = (O• )Lφ .

To prove that this works as a cross ratio (i.e preserved under homography), we can prove it is invariant wrt.
the choice of L (previously defined as the line at infinity).

Proposition 7.4.13. We proceed with projective maps. (See 7.A). Let C = Lφ , then we have the projective
map EVIN TYPE HERE We want to prove that for a point P , the polar of this transformation (aka:
O → pDO P ) is also a projective map. Note that

W = pDO P ∩ L = pDO OP ,

So U = φ(W ) is the second intersection of OP and C = Lφ . Since the ”second intersection map” is a
projective map, we have that O → U, φ : L → C are all projective maps, and we get

pP ◦ D = [W → pDO = φ(P )W ] ◦ φ ◦ [O → U ]

is a sequence of projective maps, and thus is a projective map itself.

Proof.

Proposition 7.4.14.

Proof.

Example 7.4.15.

Solution.

208
AoPS Chapter 7. Projective Space

7.4.1 No Fixed Points?

What are the fixed points of isogonal conjugation? Clearly, it’s I, I a , I b , I c . Note that I a I b I c is the anticevian
triangle of I. By taking a suitable homography, we get that if an isoconjugation φ has a fixed point S, then
the vertices of its anticevian triangle S a , S b , S c are also fixed points of φ.

Another more insightful proof of this is that a point Q is fixed under φ if and only if, for the pencil of
conics F defined by φ, we have
\
Q∈ C.
C∈F

From this viewpoint we can directly get that the maximum number of (real or complex) fixed points of an
isoconjugation is 4, since two conics can only intersect at four points.

Let’s look at this from a barycentric point of view. Let Gφ be the pole of isoconjugation φ, then denote
the radical transformation of Gφ as the set of points S, S a , S b , S c such that Gφ = S × S = · · · = S c × S c
(we will revisit this in chapter (TODO 12)). Since Sa , Sb , Sc , etc are obtained from S by negating one of the
barycentric coordinates, these are the fixed points of φ as well.

Analogously to (TODO 7.3.11), we can get

Proposition 7.4.16. For point S in △ABC, let △S a S b S c be the anticevian triangle of S in △ABC, and
let F be the pencil of conics passing through S, S a , S b , S c . For another point P ∈
/ {A, B, C}, draw lines
ℓA , ℓB , ℓC through A, B, C such that AP and ℓA are harmonic conjugates in SS a , S b S c .

Then ℓA , ℓB , ℓC intersect at a point Q, and for all conics D ∈ F , P, Q are conjugate in D.

Thus, there exists an isoconjugation φ that has any point and the vertices of its anticevian triangle as
fixed points.

Proof. We will basically replicate the proof of 7.3.11 under a suitable homography.

We only need to prove that there exists a point Q satisfying

(BP, BQ; SS b , S c S a ) = (CP, CQ; SS c , S a S b ) = −1,

then (AP, AQ; SS a , S b S c ) = −1. By taking isogonal conjugation under a homography, there exists a involution
ψ ∈ Aut(P Q) if
(SS a ∩ P Q, S b S c ∩ P Q)

are exchanged under ψ. However, P, Q are fixed under ψ, and thus

(AP, AQ; SS a , S b S c ) = −1.

209
AoPS Chapter 7. Projective Space

Since this means that F = Fφ = {D | S, S a , S b , S c ∈ D}, in reality the cross ratio in the pencil of conics
F defined by the isoconjugation previously, is actually just the typical cross ratio defined as the cross ratio
of the tangents (see (TODO 7.A)).

(D• )F := (D1 , D2 ; D3 , D4 ) = (TS Di ),

since pD (S) = TS D.

Proposition 7.4.17.

Proof.

Theorem 7.4.18.

Theorem 7.4.19.

Proof.

7.A Revisiting the Cross Ratio

Right now, we have defined the cross ratio on many objects, such as the set of points on a line, the set of lines
through a point, and the set of points on a conic, or even the set of tangent lines to a conic, that are invariant
under projective transformations. In general, we call these objects pencils, and here is a rigorous definition.

For this section, the term “projective map” will be used for a bijective function sending a copy of P1 → P1 ,
and “projective transformation” will be used for a bijective function preserving cross ratio and collinearity
sending P2 → P2 . (A projective transformation in this sense is also known as a homography. A bijective
function preserving only collinearity is known as a collineation. For most of this book, the two concepts are
interchangeable. However, the distinction does matter for a few theorems in this section.)

Definition 7.A.1. A pencil (X, (−, −; −, −)X ) is a set X and a 4-ary rational function (to define a cross
ratio)

(−, −; −, −)X : X × X × X × X 99K R ∪ ∞

(P1 , P2 , P3 , P4 ) 7→ (P1 , P2 ; P3 , P4 )X =: (P• )X ,

such that there exists a bijective function ϕ : R ∪ ∞ → X that satisfies (a• )X = (ϕ(a• )), for all a1 , a2 , a3 , a4 ∈
R ∪ ∞ that there exists a cross ratio on.

210
AoPS Chapter 7. Projective Space

Example 7.A.2. Here’s an example of some pencils:

• A fixed line and the points on it,

• The set of lines through a fixed point,

• A set of coaxal circles through two fixed points,

• A conic and the points on it,

• A conic and its tangent lines,

• The set of conics through four fixed points, (this is sometimes known as the Veronese map)

• The set of conics tangent to four fixed lines.

All of these have a naturally defined cross ratio (for the conic ones, it’s just the normal cross ratio on the
tangents from another point to the four conics).

Remark (Linear combinations of curves). You can think of all pencils as embeddings of P1 → P2 . In general,
the set of degree n curves in the complex projective plane can be thought to form a n+2


2 − 1 dimensional
n+2


vector space (there are 2 coefficients in a homogeneous polynomial in 3 variables of degree n, but you can
always homogenize some coefficient to 1, so subtract 1). “Passing through a point” reduces this dimension by
1. This will be further elaborated on in Chapter (TODO 11), using the fact that the set of cubics through 8
points has dimension 1. (also “dimension” in this remark refers to codimension. The actual dimension of the
ambient space, CP2 , is 2).

Below all of the expressions (Xi , (−, −; −, −)Xi ) represent pencils. We have already defined

A(P• ) = A(P1 , P2 ; P3 , P4 ) = (AP1 , AP2 ; AP3 , AP4 ) = (AP• ),

and if ϕ : X1 → X2 is a projective map, then we define

ϕ(P• ) = ϕ(P1 , P2 ; P3 , P4 ) = (ϕ(P1 ), ϕ(P2 ); ϕ(P3 ), ϕ(P4 )) = (ϕ(P• ))X2 .

Definition 7.A.3. We define a map (function) ϕ : X1 → X2 to be a projective map if for all P1 , P2 , P3 , P4 ∈

X1 , (ϕ(P• ))X2 = (P• )X1 .

Note that projective maps ϕ must be bijective. Denote the set of projective maps between two pencils
X1 , X2 as Hom(X1 , X2 ). Then, we have Aut(X) := Hom(X, X). Also, for two projective transformations

211
AoPS Chapter 7. Projective Space

ψ, ϕ, we have
((ψ ◦ ϕ)(P• ))X3 = (ψ(ϕ(P• )))X3 = (ϕ(P◦ ))X2 = (P• )X1 ,

so we get

Proposition 7.A.4 (Composition of Projective Maps). Given pencils X1 , X2 , X3 , if

ϕ ∈ Hom(X1 , X2 ), ψ ∈ Hom(X2 , X3 ),

then ψ ◦ ϕ ∈ Hom(X1 , X3 ).

Proposition 7.A.5. Given pencils X1 , X2 and map ϕ ∈ Hom(X1 , X2 ), then there exists an inverse ϕ−1 ∈
Hom(X2 , X1 ).

Proof. Since ϕ is bijective, ϕ−1 exists. For P1 , P2 , P3 , P4 ∈ X2 ,

(ϕ−1 (P1 ), ϕ−1 (P2 ); ϕ−1 (P3 ), ϕ−1 (P4 )X1 = (ϕ(ϕ−1 (P1 )), ϕ(ϕ−1 (P2 )); ϕ(ϕ−1 (P3 )), ϕ(ϕ−1 (P4 )))X2 = (P1 , P2 ; P3 , P4 )X2

thus ϕ−1 ∈ Hom(X2 , X1 ).

Proposition 7.A.6 (Projective Maps Are Determined By Three Points). Given line ℓ and three distinct
points P1 , P2 , P3 on the line, if
ϕ : {P1 , P2 , P3 } → ℓ

such that ϕ(P1 ), ϕ(P2 ), ϕ(P3 ) are distinct, then there exists a unique one-to-one function ϕ ∈ Aut(ℓ) such
that ϕ(Pi ) = ϕ(Pi ). (In other words, maps from a complex line CP1 to another complex line are uniquely
determined by three points and their images.)

Proof. Pencils are isomorphic to P1 , so we will just prove this for CP1 . Suppose ϕ sends Pi → Qi , then define
ϕ to be
aP + b
(Q1 , Q2 ; Q3 , ϕ) = (P1 , P2 ; P3 , P ) =⇒ ϕ(P ) = ,
cP + d

where a, b, c, d satisfy ad ̸= bc, then we have ϕ(Pi ) = ϕ(Pi ). Proving (ϕ(Q• )) = (Q• ) is left to the reader.
So we have that the group of automorphisms on a pencil is isomorphic to the group of fractional linear
ap+b
transformations, {p → cp+d | ad ̸= bc} which is also known as PGL2 (C).

Proposition 7.A.7. (TODO 7.A.6) holds for two arbitrary pencils X1 , X2 and three elements P1 , P2 , P3 ∈ X1 .

Proof. The pencils X1 and X2 are isomorphic to P1 .

Proposition 7.A.8. Automorphisms on an arbitrary pencil are isomorphic to PGL2 (C) as well.

212
AoPS Chapter 7. Projective Space

Proof. Trivial.

Remember that we defined the pencil of lines through a fixed point as TP = {ℓ | P ∈ ℓ}. Let’s address
some very fundamental theorems:

Proposition 7.A.9. If a bijective function Φ : RP2 → RP2 preserves collinearities (in other words,)

P ∈ QR ⇐⇒ Φ(P ) ∈ Φ(Q)Φ(R),

then Φ is a projective transformation.

Proof. It preserves harmonics

Proposition 7.A.10 (Projective Transformations Are Determined By Four Points). If A, B, C, D are

four points with no three collinear and Φ : RP2 → RP2 is a projective transformation, then the unique
transformation fixing A, B, C, D is the identity transformation.

Proof. Trivial by dof count

Remark. Combining (TODO 7.A.9) and (TODO 7.A.10), we can prove that a transformation Φ preserves
collinearities if and only if it sends the (homogeneous coordinates)

Φ[x : y : z] = [Φ11 x + Φ12 y + Φ13 z : Φ21 x + Φ22 y + Φ23 z : Φ31 x + Φ32 y + Φ33 z],

(also, the matrix formed by all the Φij has to be invertible.) This theorem is called the Fundamental Theorem
of Projective Geometry. Further, for four points P1 , P2 , P3 , P4 and Q1 , Q2 , Q3 , Q4 (no three collinear), there
exists a projective transformation Φ that sends Pi to Qi .

In complex projective space CP2 , a collineation does not need to be determined by four points. For
example take the transformation Φ sending [x : y : z] to [x : y : z]. This preserves collinearities but does not
preserve cross ratio (it complex conjugates it). It also cannot be written as the above matrix expression. The
reason for the failure of our previous proof is because of some nontrivial automorphisms of C, such as [z → z].
In fact, all collineations in CP2 are compositions of projective transformations and reflections.

Proposition 7.A.11. Given two points P, Q, and ϕ ∈ Hom(TP, TQ), then the locus of ℓ ∩ ϕ(ℓ) is a (possibly
degenerate) conic section. So there exists a conic section C such that C = {ℓ ∩ ϕ(ℓ) | ℓ ∈ TP }.

Proof. An easy proof of this is with the methods of (TODO 7.B), by computing the degree of this point as
2.

213
AoPS Chapter 7. Projective Space

Proposition 7.A.12. Given a pencil X and two distinct elements P1 , P2 in X, the projective involution
ϕ : X → X sending element Q to R such that (P1 , P2 ; Q, R) = k for some fixed constant k is an automorphism
on X.

Proof. Again, by isomorphism with P1 we only need to prove it for the case where X is a line and P1 , P2 are
̸ X be a line through P1 , and let R1 , R2 , R3 , R4 be four points on line X such
two points on the line. Let ℓ =
that R1 = P1 and (R• ) = k. Then for any point Q, we have R2 P2 , R3 Q, R4 ϕ(Q) are concurrent, and thus
ϕ = [S → SR4 ∩ X] ◦ [Q → R3 Q ∩ R2 P2 ] is a projective map and in Aut(X).

Proposition 7.A.13. Given conic C, line ℓ, and point P on C, let moving line L through P intersect C, ℓ
respectively at Q, R. Let A be a point on L such that (P, Q; R, A) is fixed (let it equal k). Then the locus of
A is a conic (or degenerate).

Proof.

Proposition 7.A.14 (Generalized Steiner Conic). Given conic C, ϕ ∈ Aut C if and only if there exists a line
ℓ such that for all P, Q in C, P ϕ(Q) ∩ ϕ(P )Q ∈ ℓ.

Remark. Note that this proves the fact that all involutive automorphisms on a conic are given by the second
intersection map for some point. In fact, the line ℓ would just be the polar of that point in C.

Proof.

Proposition 7.A.15 (Steiner Inconic Theorem). Given conic C, let ϕ be an automorphism in C, then there
exists another conic C ′ such that for all P ∈ C, P ϕ(P ) ∈ TC ′ .

Proof.

Remark. Note when C ′ is degenerate, that this gives you DIT.

7.B Moving Points / The Polynomial Method

Typically, all objects we encounter in geometry can be represented as some analytic transformation with
respect to some other object (for example all the polynomials and birational parametrizations in analytic
geometry), so we can actually analyze the degree of these objects and bound them, and then prove things
with the Fundamental Theorem of Algebra by checking only a few cases. We will use the fact that if a
degree-d polynomial P (x) has d + 1 roots, then P (x) = 0.

214
AoPS Chapter 7. Projective Space

Definition 7.B.1. Let ι : P1 → Pn be a algebraic map (basically a parametrization of some point in Pn ),

then we can write it as
[s : t] → [x0 (s, t) : x1 (s, t) : · · · : xn (s, t)],

where xi are all homogeneous polynomials with the same degree, such that gcd(x0 , x1 , . . . , xn ) = 1, then we
define the degree of ι as
deg ι := deg x0 = deg x1 = · · · = deg xn .

Since we are generally working in the projective plane CPn , for the following section we will only choose
n = 2 and n = 1. Further, note that all lines and conics can be parametrized in terms of P1 . For two
parametrizations (i.e maps between P1 → P2 ) ιA = [xA : yA : zA ], ιB = [xB : yB : zB ] (interpret these as two
“moving points” A, B in CP2 ), we can write the line AB in terms of points A, B, so it can also be parametrized
in terms of P1 , like this:

ι∨
AB := [s : t] → [yA zB − yB zA : zA xB − zB xA : xA yB − xB yA ],

which clearly has degree at most deg A + deg B.

Sometimes, we can make a even tighter bound on the degree of line AB.

Proposition 7.B.2 (Zack’s Lemma). For two moving points A, B, we have that line AB has degree at most

deg ι∨
AB ≤ deg ιA + deg ιB − #{P | ιA (P ) = ιB (P )}.

So basically you can subtract 1 from the bound of deg A + deg B for every time A = B in the parametrization.

Proof. Factor out a monomial each time they’re equal.

Remark. For the rest of this section, if we say “a point A moves on line/conic K projectively”, it means that
there is a parametrization of a point in P2 in terms of coordinates in P1 ∼
= R ∩ ∞ that covers the line/conic
exactly once.

Note that a point covering a degree d curve n times has this parametrization degree of d × n (proof: just
take each term of the parametrization to the nth power, since it’s homogeneous). Since lines are degree 1 and
conics are degree 2, a point moving projectively on a line/conic has degree 1 and 2.

Here’s the most important ways to bound degrees of points and lines in purely projective problems.

Proposition 7.B.3 (Projective maps between lines preserve degree). For projective map ϕ sending line
ℓ1 → ℓ2 , we have for moving points moving on lines ℓ1 , ℓ2 , then deg ι1 = deg ι2 , where ιi : P1 → P2 satisfies
ιi (P1 ) = ℓi and ι2 is the image of ι1 after ϕ.

215
AoPS Chapter 7. Projective Space

Proposition 7.B.4 (Projective maps between conics preserve degree). For projective map ϕ sending conic
C1 → C2 , we have for moving points deg ι1 = deg ι2 moving on conics C1 , C2 , where ιi : P1 → P2 satisfies
ιi (P1 ) = Ci and ι2 is the image of ι1 after ϕ.

Proposition 7.B.5 (Conic doubling). For projective map ϕ sending line ℓ → conic C (example: projecting
through a fixed point on the conic or a fixed inversion), for moving point ι1 : P1 → P2 moving on line ℓ
such that ι1 (P1 ) = ℓ, deg(ϕ ◦ ι) = 2 · ι.

Similarily, a projective map from a conic to a line halves degree of moving points on the conic.

Proposition 7.B.6. If two moving points ιA , ιB : P1 → P2 move on a conic C, then the line through them
has degree
1
deg ι∨
AB = (deg ιA + deg ιB ).
2
deg ιA
Proof. Let ι : C → P1 be a projective map. Then by conic doubling, ι ◦ ιA and ι ◦ ιB have degrees 2
deg ιB
and 2 , respectively. Let these functions be [PA : QA ] and [PB : QB ]. The line has coordinates given by
functions of degree deg ιA + deg ιB , but they all vanish when PA QB − PB QA = 0 because then the moving
points coincide. Hence the degree is deg ιA + deg ιB − deg(PA QB − PB QA ) = 12 (deg ιA + deg ιB ).

Corollary 7.B.7 (Conic doubling for non-fixed projections). If ιA : P1 → P2 is a moving point on fixed conic
C, and ιL : P1 → P2 is a moving line through ιA , let ιB be the second intersection of ιL with C. Then we have

deg ιB = 2 deg ιL − deg ιA .

These are enough ways to find (efficiently) the degree of most points in purely projective problems. (Some
other useful projective lemmas is that polar reciprocation across a fixed conic preserves degree, and that in a
Poncelet triangle configuration with fixed circumconic and inconic, with moving point X on the circumconic,
the “opposite tangency point” on the inconic also has degree of X).

Theorem 7.B.8. Given three moving points ι1 , ι2 , ι3 : P1 → P2 , if

#{P ∈ P1 | ι1 (P ), ι2 (P ), ι3 (P ) collinear} ≥ deg ι1 + deg ι2 + deg ι3 + 1,

then for all P ∈ P1 , ι1 (P ), ι2 (P ), and ι3 (P ) are collinear.

Proof. If we let ιi = [xi : yi : zi ], then the collinearity condition is equivalent to

 
x y1 z1
 1 
det x2 y2 z2  = 0.
 
 
x3 y3 z3

216
AoPS Chapter 7. Projective Space

The determinant has degree deg ι1 + deg ι2 + deg ι3 , and by assumption it vanishes at deg ι1 + deg ι2 + deg ι3 + 1
values, and is hence identically zero. This also gives that the area of this triangle has degree deg ι1 + deg ι2 +
deg ι3 .

Corollary 7.B.9. Given six moving points ι1 , . . . , ι6 : P1 → P2 , if

#{P ∈ P1 | ι1 (P ), . . . , ι6 conconic} ≥ 2(deg ι1 + · · · + ι6 ) + 1,

then for all P ∈ P1 , ι1 (P ), . . . , ι6 (P ) are conconic.

Proof. Let X(P ) = ι1 (P )ι2 (P ) ∩ ι4 (P )ι5 (P ), Y (P ) = ι2 (P )ι3 (P ) ∩ ι5 (P )ι6 (P ), and Z(P ) = ι3 (P )ι4 (P ) ∩
ι6 (P )ι1 (P ). Then by Pascal’s, ι1 (P ), . . . , ι6 (P ) are conconic if and only if X(P ), Y (P ), and Z(P ) are collinear.
Note that we have

deg X ≤ deg(ι1 ι2 ) + deg(ι4 ι5 ) ≤ deg ι1 + deg ι2 + deg ι4 + deg ι5

and similar bounds hold for Y and Z. Hence, the desired result follows from (TODO 7.B.8). (Translator’s note:
it is also possible to do a direct proof along the lines of the previous theorem using a bigger determinant.)

Corollary 7.B.10. Given four moving points ι1 , . . . , ι4 : P1 → P2 , if

#{P ∈ P1 | ι1 (P ), . . . , ι4 concyclic} ≥ 2(deg ι1 + · · · + ι4 ) + 1,

then for all P ∈ P1 , ι1 (P ), . . . , ι4 (P ) are concyclic.

Proof. Apply (TODO 7.B.9) to ι1 through ι4 together with the two circle points, which have degree 0 as
they are fixed.

Proposition 7.B.11. For four moving points ι1 , . . . , ι4 : P1 → P1 , the cross ratio mapping ι(1,2;3,4) : P1 → P1
given by
ι(1,2;3,4) (P ) = (ι1 (P ), ι2 (P ); ι3 (P ), ι4 (P ))

has degree
4
X
deg ι(1,2;3,4) = (deg ιi − Ni ) + 2N,
i=1

where Ni = #{P ∈ P1 | ιj (P ) all coincide, j ̸= i} and N = #{P ∈ P1 | ιj (P ) all coincide}. Also,

deg ι(1,2;3,4) ≤ deg ι1 + deg ι2 + deg ι3 + deg ι4 .

217
AoPS Chapter 7. Projective Space

Proof. Let ιi (P ) = [xi (P ) : yi (P )]. Then the cross-ratio is given by

ι(1,2;3,4) = [(x1 y3 − x3 y1 )(x2 y4 − x4 y2 ) : (x1 y4 − x4 y1 )(x2 y3 − x3 y2 )].

Whenever all but ιi coincide, we can cancel out a linear factor from the two terms not involving ιi . However,
if all four coincide, we have cancelled out four linear factors from each coordinate, whereas in reality we can
only cancel two.

Remark. This gives us a way to compute the degrees of angles (for example, angle ∡BAC) by computing
the degree of the cross ratio (AB, AC; AI, AJ) ≤ deg B + deg C.

Remark. Another useful thing to note is that for a fixed circle ω and a moving point ι, the degree of Powω (ι)
is
deg Powω (ι) = 2 deg ι − {# of times where ι passes through either circle point}.

Proof. Let ω be given by x2 + y 2 + bx + cy + d = 0 and let ι be given by ι(P ) = [x(P ), y(P ), z(P )]. Then

Powω (ι(P )) = [x(P )2 + y(P )2 + bx(P )z(P ) + cy(P )z(P ) + dz(P )2 : z(P )2 ].

Since both expressions vanish when ι(P ) is a circle point, the desired result follows by Zack’s lemma.

Remark. Any fixed isoconjugation doubles degree in general. (We will prove this in Chapter 12).

Remark (Three useful non-projective lemmas to deal with lines).

Rotation Let θ be a angle defined as the directed angle between two moving lines ℓ1 , ℓ2 . Given a moving
lemma: point ι and a moving line ℓt through point ι, the rotation of ℓt around point ι by θ has degree

deg(A) + deg(ℓt ) + deg(ℓ1 ) + deg(ℓ2 ).

Reflection The degree of reflecting moving point ι across moving line ℓ is degree 2 · deg ℓ + deg ι. This also
lemma: holds for reflecting a line over a line. (Think of it as 2·deg(axis of reflection)+deg(thing being reflected)).

Efficient For four moving lines ℓ1 , ℓ2 , ℓ3 , ℓ4 , let ι1 be the intersection of ℓ1 , ℓ2 , and define ι2 , ι3 , ι4 cyclically.
cyclic Then ι1 , ι2 , ι3 , ι4 are always concyclic if
lemma:
#{P ∈ P1 | ι1 (P ), . . . , ι4 (P ) concyclic} ≥ deg ℓ1 + · · · + ℓ4 + 1.

218
AoPS Chapter 7. Projective Space

Proof. The rotation lemma can be proven by considering the degree of the cross ratio. In fact, both of the
first of these can actually be proved by considering how many times the reflected/rotated object covers P1
and noting that reflection and rotation preserve cross ratio. The concyclicity lemma is trivally true by the
rotation lemma.

219
Chapter 8

Circumrectangular Hyperbolas -
Hidden/Completed

Why geometry is better than life: in life, you will only meet one special person. In geometry, you will meet
multiple special conic sections.

8.1 Special Conics and The Poncelet Point

Proposition 8.1.1. Let P be a parabola with focus at F , and let ℓ be its directrix. Let △ABC have all
three sides tangent to this parabola (similar to an excircle), then F ∈ (ABC) and ℓ passes through the
orthocenter of △ABC.

Proof. Reflect F over BC, CA, AB to get Fa , Fb , Fc , and let P touch BC, CA, AB at Ta , Tb , Tc . Then by the
optical property of a parabola we can get that Fa , Fb , Fc respectively are the feet from Ta , Tb , Tc to ℓ. Thus if
Fa , Fb , Fc are collinear, then the directrix is just the Steiner line of F wrt. △ABC! So it must pass through
H.

Proposition 8.1.2. Let P be a parabola with focus at F and ℓ as directrix. Let △ABC be a self-conjugate
triangle wrt. P, then F lies on the nine-point circle of △ABC, and ℓ passes through the circumcenter of
△ABC.

Proof. Let U be the point at infinity perpendicular to ℓ, then U ∈ P (the parabola is tangent to the line at
infinity at U ) and U is the center of P. Let Ma , Mb , Mc respectively be the midpoints of BC, CA, AB, and
let AU intersect BC, Mb Mc at P, T respectively. Then we know (A, T ; P, U ) = −1 so we know T ∈ P. From

220
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Mb Mc ∥ BC and the parallel chord theorem (TODO 7.1.13) on T, P, U , we know that Mb Mc is the polar of
T wrt. P, so Mb Mc is tangent to P. As such, by symmetry we have that the whole medial triangle is tangent
to P. As such, since the circumcenter of the medial circle is the orthocenter, by the previous property on
△Ma Mb Mc , ℓ goes through the circumcenter of △ABC and F lies on the nine-point circle.

These are the most well-known theorems for parabolas. Now we will move on to hyperbolas.

Definition 8.1.3. We call a hyperbola H rectangular if its two asymptotes are perpendicular. Equivalently,
a rectangular hyperbola is a conic section that has the two circle points as conjugates (polar of one passes
through the other).

All non-degenerate rectangular hyperbolas are spirally symmetric to the solution set of xy = 1.

Proposition 8.1.4. Given a non-right triangle △ABC, a hyperbolic circumconic (called a circumhyperbola)
is rectangular iff H also goes through H.

Proof. We consider isogonal conjugation φ in △ABC. Let O be the circumcenter of △ABC. Then H ∈ H if
and only if φ(H), which is a line, goes through O. Let φ(H) intersect φ(L∞ ) = (ABC) at two points X, Y .
Then O ∈ φ(H) if and only if X, Y are antipodes in (ABC), which implies φ(X), φ(Y ) are 90◦ apart on the
line at infinity, so H is rectangular.

If we redefine H as the cevapoint (TODO 7.1.25) of the circle points I and J, then this theorem is just a
direct application of (TODO 7.2.23).

We can also quickly solve this with analytic techniques. Since all rectangular hyperbolas are spirally
symmetric to xy = 1, we can choose a system of coordinates such that A = (a, a−1 ), B = (b, b−1 ), C =
(c, c−1 ) ∈ H = {(x, y) | xy = 1}. Then the orthocenter H of △ABC is just (−(abc)−1 , −abc).

If we have four points A, B, C, D that lie on one rectangular hyperbola H, and we choose HA , HB , HC , HD
respectively such that A is the orthocenter of △Hb Hc Hd , etc, then we have

(A, B; C, D)H = (a, b; c, d)

= (−(bcd)−1 , −(cda)−1 , −(dab)−1 , −(abc)−1 )

= (HA , HB ; HC , HD )

by projecting from ∞x=0 onto the x-axis.

Example 8.1.5. Let E be the Euler line of a non-equilateral acute triangle △ABC. Prove that the Euler
line of the triangle formed by lines CA, AB, E is parallel to BC.

221
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Proof. Let O, H be the circumcenter and orthocenter of △ABC. We consider the Jerabek hyperbola, the
rectangular hyperbola H passing through A, B, C, O, H. Note that H is the image of E under isogonal
conjugation. Let X = CA ∩ E, Y = AB ∩ E, U as the fourth intersection of H and (ABC), then we have

∡XY A = ∡(E, AB) = ∡CAU = ∡CBU,

∡AXY = ∡(AC, E) = ∡U AB = ∡U CB,

+
thus △AY X ∼ △U BC. Note that the orthocenter HU of △U BC must also lie on H, so H is the isogonal
conjugate of the Euler line of △U BC wrt. △U BC. Therefore if the Euler line of △U BC intersects CU, U B
+
at R, S, we have △U SR ∼ △ABC, so the Euler line E ′ of △AY X satisfies

E ′ = (∡XY + ∡RS − ∡BC) = (∡AY + ∡BC − ∡U B) + (∡U S − ∡AB) = ∡BC.

The following proposition can be thought of as a limiting case of (TODO 8.1.4).

Proposition 8.1.6. Given any right angled triangle △ABC, let BC be the hypotenuse and let AD be the
altitude from A. Let H be a circumconic of △ABC, then AD is tangent to H iff H is a circumrectangular
hyperbola.

Proof. We consider isogonal conjugation with φ on △ABC, let O be the circumcenter of △ABC. Then we
have that AD is tangent to H if and only if the isogonal conjugate of the circumconic φ(H), AO, BC are
concurrent, which is the same thing as O ∈ φ(H) since it’s a right triangle, which finishes.

Proposition 8.1.7. Let H be a circumrectangular hyperbola centered at T . Then ∡pH (P ) + ∡T P is a

constant value. This can also be interpreted as

∡(pH (P1 ), pH (P2 )) + ∡P1 T P2 = 0◦ ,

for any two points P1 , P2 .

Proof. Let U = ∞T P , V = ∞pH (P ) , and then by the Parallel Chords Theorem (TODO 7.1.13) we know
that pH (U ) ∥ pH (P ). Let W1 , W2 be the two intersections of H with the line at infinity. Since U and V are
conjugate points in H we know that

T
−1 = (U, V ; W1 , W2 ) = (P, V ; W1 , W2 ).

222
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Since ∡W1 T W2 = 90◦ , we have that ∠Pi T Vi has angle bisectors T W1 , T W2 . Thus

∡pH (P ) + ∡T P = ∡2 · T W1 = ∡2 · T W2

is a constant.

This has another interpretation in terms of the circle points: since △T IJ is self-conjugate in H, for any
two points P, Q we have

pH
(∞pH (P ) , ∞pH (Q) ; J, I) = T (P, Q; I, J) = (∞T Q , ∞T P ; J, I).

Corollary 8.1.8. Let H be a rectangular hyperbola centered at T , let △ABC be self-conjugate wrt. H.
Then T lies on the circumcircle of ABC.

Proof. From (TODO 8.1.7) we get that ∡BT C = −∡(CA, AB) = ∡BAC.

Corollary 8.1.9. Let H be a rectangular hyperbola centered at T . Suppose A, B, C ∈ H, then T lies on the
nine-point circle of △ABC.

Proof. Let H be the orthocenter of △ABC, then H ∈ H. From (TODO 7.1.8) we know that the cevian
triangle of H (which is just the orthic triangle) is self-conjugate in H. So then by (TODO 8.1.8) we know T
lies on the circumcircle of the orthic triangle, which is just the nine-point circle.

We can also re-interpret this in (TODO 7.4.18): T lies on the nine-point conic of the complete quadrangle
(A, B, C, H).

Proposition 8.1.10. If BC is a diameter of a rectangular hyperbola H, then given any A ∈ H, the antipode
A∗ of A in (ABC) also lies on H. Also, the tangent from A to H, TA H is the A-symmedian of △BAC.

Proof. We consider isogonal conjugation φ in △ABC. Let L = φ(H), L′ = φ(TA H), then φ(L ∩ L′ ) ∈
H ∩ TA H = A, so L ∩ L′ ∈ BC. Note that the orthocenter H of △ABC is in H, so thus the reflection of H
over the midpoint of BC lies on H, which is just A∗ . So from ∞⊥BC = φ(A∗ ) ∈ L we know that L is the
perpendicular bisector of BC. Thus TA H is the isogonal conjugate of L′ = A(L ∩ BC) in ∠ABC, and thus,
TA H must be the A-symmedian.

From this characterization we can redefine hyperbolas with a common config:

223
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Example 8.1.11 (First Isogonality Lemma). Let E, F be two points on sides CA, AB of △ABC, that also
satisfy B, C, E, F concyclic. Let P = BE ∩ CF, then the isogonal conjugate of P (suppose it’s point Q)
satisfies
∡CBQ = ∡P BA = ∡EBF = ∡ECF = ∡ACP = ∡QCB

which also means that Q lies on the perpendicular bisector of BC. Therefore the locus of P is the locus of
the isogonal conjugate of the perpendicular bisector of BC, which is just a rectangular hyperbola as it goes
through H.

Note that the antipode of A in (ABC), point A∗ , also lies on Ha and (BC)(HA∗ ) is a parallelogram.
Thus the center of Ha is just the midpoint of BC, point M . This also implies that BC is a diameter of Ha .

Further, we also get that when we choose a diameter BC of an arbitrary rectangular hyperbola, and pick
a point P on that hyperbola, we know ∡P B + ∡P C is always a constant.

(This can also be understood through the circle points again: since △M IJ is self-conjugate wrt. Ha ,
from (TODO 7.1.8) we have that X = BJ ∩ CI, Y = BI ∩ CJ ∈ H. This tells us that for any point P ∈ H,
we have
B(A, P ; J, I) = (A, P ; X, Y )H = C(A, P, I, J) = C(P, A; J, I).

which finishes by the Angle Theorem).

Example 8.1.12 (Taiwan 2022 TST 3 Problem 5). Let △ABC be an acute triangle with circumcenter O
and circumcircle Ω. Construct D, E on AB, AC respectively, and let ℓ be the line through A perpendicular
to DE. Let ℓ intersect the circumcircle of ADE and Ω again at P, Q respectively. Let OQ and BC intersect
at N , and let OP and DE intersect at S. Let W be the orthocenter of △SAO.

Prove S, N, O, W are concyclic.

224
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

S
P
D O
W

B N C

Proof. Since ∡SW O = ∡OAS, by the above theorem we only need to prove ∡SN + ∡SA = ∡ON + ∡OA,
which is the same thing as S, O lying on a rectangular hyperbola with diameter AN . Since N = OP ∩ BC,
we only need to prove that for a fixed Q, that

• The locus of S is a rectangular hyperbola H;

• AN is the diameter of H.

We proceed with moving points. When Q is fixed, E = D∞⊥AQ ∩ AC, P = AQ ∩ ∞∡AQ+∡(⊥AQ)−∡AC . This
tells us that D → E, D → P is a projective map. Thus

S = OP ∩ DE = OP ∩ ∞⊥AQ D

is on a rectangular hyperbola H through O, ∞⊥AQ . When D = A, S = OA ∩ ∞⊥AQ A = A; when D = ∞AB ,

S = O∞AQ ∩ ∞⊥AQ ∞AB = ∞AQ . Thus H also passes through ∞AQ and ∞⊥AQ . Thus H is a rectangular
hyperbola.

Next, we need to prove that N ∈ H. In other words, we need to find a position for D such that
N = OP ∩ DE, so we need to have a D = AB ∩ N ∞⊥AQ , and thus it remains to prove this when P = Q.
This is equivalent to Q being the Miquel point of AB, AC, BC, DE. However we actually have B, D, Q, N
concyclic, so
∡DN Q = ∡(⊥ AQ, OQ) = (A − Q)Ω = ∡ABQ = ∡DBQ.

225
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

So finally, we can say that AN is a diameter of H, so we only need to prove that

∡(OA + ∡ON = ∡(∞AQ A) + ∡(∞AQ N ) = 2AQ,

but that is obvious.

Example 8.1.13. From these properties of hyperbolas, we can prove inversion is actually a point isoconjuga-
tion. In reality, if J is a inversion with center O and power k, and s is reflection across a line ℓ that goes
through O, then T := J ◦ s is a isoconjugation in △OIJ, where I, J are circle points.

We will only consider the case in which the power of inversion is positive. Let Γ be a circle with radius
√
k and center O. Then J is just inversion about this circle. We consider the two intersection points Γ with ℓ,
let these points be X, Y .

Define a pencil of conics as

F = {H | X, Y ∈ H, H is a rectangular hyperbola centered at O.}

Since I, J harmonically divide any two perpendicular lines’ intersections with the line at infinity, we get
that △OIJ is self-conjugate wrt. any H ∈ F . For any P , we know that P ∗ = J ◦ s(P ) satisfies (XY )(P P ∗ )
is a harmonic quadrilateral. We only need to prove that for all H ∈ F , pH (P ) passes through P ∗ .

Let A, B be the other two intersection points of Ω := (XY P P ∗ ) and H. From (TODO 8.1.10), the
tangents to H at two points A, B are concurrent with the tangents to Ω from two points X, Y . Let’s say they
concur at point T . Note that from the harmonic quadrilateral (XY )(P P ∗ ), we also get that T lies on line
L := P P ∗ .

We consider the complete quadrangle q = (A, B, X, X), where XX is the tangent from X to Ω. We now
use DIT on q and L to get an involution

(P, P ∗ ), (L ∩ AB, T ), (L ∩ AX, L ∩ BC).

We want to prove that this involution is the involution sending [Q → L ∩ pH (Q)]. We already have that

L ∩ pH (T ) = L ∩ pH (pH (A) ∩ pH (B)) = L ∩ AB

, so we only need to prove that pH (L ∩ AX), L, BX are concurrent, or equivalently, pH (L), U := pH (AX), L ∩
BX are collinear. But this is true since

U (A, B; C, pH (L)) = (A, B; U X ∩ AB, pH (L))

226
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed
pH
= (T A, T B; T X, L) = (U A, U B; U X, U (L ∩ BX)).

Now it remains to show that F has two other fixed points. We claim that the other two common points
√
of conics in F are actually just the two points of intersection of the circle of imaginary radius −k centered
at O and the line O∞ℓ⊥ . This can be proven algebraiclly, but it also follows as these are the points fixed
under T. As such, inversions are just isoconjugations in △OIJ.

Definition 8.1.14. For a complete quadrangle q = (P1 , P2 , P3 , P4 ) that isn’t an orthocentric system, we
define H(P1 P2 P3 P4 ) as the unique rectangular hyperbola through P1 , P2 , P3 , P4 . We call the center of this
hyperbola the Poncelet point of these four points.

Theorem 8.1.15. Let T be the Poncelet point of complete quadrangle q = (P1 , P2 , P3 , P4 ). Then T lies on
the nine-point circle of every △Pi−1 Pi Pi+1 , T lies on the pedal circle of Pi wrt. △Pi+1 Pi+2 Pi+3 for all i, and
T lies on the circle through the three intersections of diagonals of the quadrangle, (which is just the cevian
circle).

Proof. Let H be a rectangular hyperbola passing through P1 , P2 , P3 , P4 , and let △XY Z be the cevian
triangle of q. Then by Brokard we know that △XY Z is self-polar wrt. H, and thus from (T ODO8.1.9) and
(T ODO8.1.8) we know that T lies on the nine-point circle of △Pi−1 Pi Pi+1 and the cevian circle.

Let △Q1 Q2 Q3 be the pedal triangle of P4 wrt. △P1 P2 P3 , and let Mij be the midpoint of Pi Pj . From T
lying on (M14 Q2 M31 ) and (M14 Q3 M12 ) we can get

∡Q2 T Q3 = ∡Q2 T M14 + ∡M14 T Q3 = ∡Q2 M31 M14 + ∡M14 M12 Q3

= ∡Q2 P3 P4 + ∡P4 P3 Q3 = ∡Q2 Q1 P4 + ∡P4 Q1 Q3 = ∡Q2 Q1 Q3 .

and thus T lies on (Q1 Q2 Q3 ), do this cyclically.

Definition 8.1.16. Given △ABC, and point P that’s not the orthocenter or a vertex, define P
“ to be the

point such that

∡BP C + ∡B P
“C = ∡CP A + ∡C P “B = 0◦ .
“A = ∡AP B + ∡AP

We call P
“ to be the antigonal conjugate of P .

The proof of uniqueness and existence is just angle chasing. Note that this is not an isoconjugation.

Proposition 8.1.17. Given △ABC, if P

“ is the antigonal conjugate of P wrt. △ABC, then the midpoint of

P and P
“ is the Poncelet point of (A, B, C, P ).

227
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Proof. We do phantom-points. Let P ∗ be the antipode of P in the rectangular hyperbola through A, B, C, P .

Then P ∗ is the reflection of P across the center of this hyperbola, so we have

∡AP + ∡(AP ∗ ) = ∡BP + ∡(BP ∗ ) = ∡CP + ∡(CP ∗ ).

This tells us that

∡BP C + ∡BP ∗ C = ∡CP A + ∡CP ∗ A = ∡AP B + ∡AP ∗ B = 0◦ ,

so by the uniqueness of antigonal conjugation we know that P ∗ = P

“, so their center is thus the center of this

hyperbola and it’s also the Poncelet point of (A, B, C, P ).

Example 8.1.18 (2018 G4). A point T is chosen inside a triangle ABC. Let A1 , B1 , and C1 be the
reflections of T in BC, CA, and AB, respectively. Let Ω be the circumcircle of the triangle A1 B1 C1 . The
lines A1 T , B1 T , and C1 T meet Ω again at A2 , B2 , and C2 , respectively. Prove that the lines AA2 , BB2 , and
CC2 are concurrent on Ω.

Proof. When T is the orthocenter of △ABC, we note that A = A2 , B = B2 , C = C2 . So the point of

concurrency isn’t defined well wrt. H. This reminds us of antigonal conjugation.

As such, let us guess that the concurrency point is the antigonal conjugate T ∗ of T . Note the midpoint
of T and T ∗ lies on T ’s pedal circle in ABC since it’s the Poncelet point of (A, B, C, T ), so the antigonal
conjugate of T lies on the circumcircle of A1 B1 C1 by homothety. It remains to show that A, A2 , T ∗ are
collinear.

Since ∡CB1 A = −∡CT A = ∡CT ∗ A, T ∗ lies on (AB1 C). Now since C is the circumcenter of △A1 B1 T ,

∡B1 T ∗ A = ∡B1 CA = ∡B1 A1 T = ∡B1 T ∗ A2 ,

so we’re done.

Here’s a “better” characterization of antigonal conjugation.

Proposition 8.1.19.
“ = φK ◦ J(ABC) ◦ φK (P ).
P

where ϕK , J represent isogonal conjugation and inversion.

Proof. Let H be the rectangular hyperbola through ABCP . Let W1 , W2 be the two points at infinity along H.
“, P are antigonal conjugates, and since TP H ∩ T “H ∈ L∞ = W1 W2 , we know that (P P
Then since P “)(W1 W2 )
P

is a harmonic quadrilateral on H.

228
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Note that this also implies W1 , W2 ’s isogonal conjugates wrt. △ABC (let them be W1∗ , W2∗ ) are the two
intersections of P ∗ P
”∗ wrt. (ABC). Thus

“; W1 , W2 )H = (P ∗ , P
(P, P ”∗ , W ∗ , W ∗ ) = −1,
1 2

so P ∗ , P
”∗ are inverses wrt. (ABC).

We will revisit antigonal conjugation in (TODO 12.1).

Proposition 8.1.20. Given △ABC, let O be the circumcenter of △ABC. Let P and Q be isogonal
conjugates in △ABC. Let T be the Poncelet point of A, B, C, P . Then T is the anti-Steiner point of line
OQ wrt. the medial triangle of △ABC.

Proof. Let △Ma Mb Mc be the medial triangle of △ABC. Let S be the Steiner line of △Ma Mb Mc wrt. T ,
and let H be the circumrectangular hyperbola of A, B, C, P . Let U be the antipode of H in H, then we know
that U lies on (ABC).

Now we consider isogonal conjugation φ on △ABC. We know that φ(U ) is just ∞OQ . Note that the
isogonal conjugate (in the medial triangle) of T ’s antipode wrt. (Ma Mb Mc ) is just the point at infinity
along the Steiner line, and since O is the orthocenter of △Ma Mb Mc , we know that the Steiner line has to be
OQ.

A similar property also holds for pedal circles, in fact:

Proposition 8.1.21. Given △ABC, let O be its circumcenter and let P, Q be two isogonal conjugates in
△ABC. Let T be the Poncelet point of (A, B, C, P ), then the Steiner line of T in the pedal circle of P is
parallel to OQ.

Proof. Let △Pa Pb Pc be the pedal triangle of P wrt. △ABC, and let △Ma Mb Mc be the medial triangle of
△ABC. From (TODO 8.1.20) and by (TODO 1.4.5) we can instead show that

∡Pa Pb + ∡Pa Pc − ∡Pa T = ∡Ma Mb + ∡Ma Mc − ∡Ma T.

Note that (Pa Ma T ) is actually just the nine-point circle of △P BC. Thus

∡Pa T Ma = ∡P B + ∡P C − 2∡BC

= (∡BC + ∡BA − (⊥ Pc Pa )) + (∡CA + ∡CB − (⊥ Pa Pb )) − 2∡BC

= ∡Ma Mb + ∡Ma Mc − ∡Pa Pb − ∡Pa Pc .

229
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Note that the nine-point circle and the pedal circle of any point P always intersect in the Poncelet point
of (A, B, C, P ). We can actually calculate this angle of intersection.

Theorem 8.1.22. Given △ABC and P , the angle between P ’s pedal circle and the nine-point circle is just

X
⊥ (∡AP − ∡BC) = 90◦ + ∡(BC + CA + AB, AP + BP + CP ).

Proof. Let △Pa Pb Pc be the pedal triangle of P . Let △Ma Mb Mc be the medial triangle of △ABC. Let T be
the Poncelet point of (A, B, C, P ). Then we know that T lies on the nine-point circles of △P CA, △P AB.

Let ℓP , ℓM respectively be the tangent from T to the pedal circle and the nine-point circle. Then we have

ℓP = ∡T Pb + ∡T Pc − ∡Pb Pc , ℓM = ∡T Mb + ∡T Mc − ∡Mb Mc .

By the proof of (TODO 8.1.21), we have

∡Pb T Mb = ∡P C + ∡P A − 2∡CA, ∡Pc T Mc = ∡P A + ∡P B − 2∡AB,

and therefore

ℓM − ℓP = ∡Pb T Mb + ∡Pc T Mc + ∡Pb Pc − ∡Mb Mc

= (2∡P A + ∡P B + ∡P C − 2∡CA − 2∡AB) + (∡AB − ∡AC − (⊥ ∡AP )) − ∡BC

=⊥ (∡P A + ∡P B + ∡P C − ∡BC − ∡CA − ∡AB).

We also have a pretty easy corollary.

Theorem 8.1.23. Given any △ABC and a circumrectangular hyperbola H, let P, Q be two points on H.
Then the angle between the pedal circles of P, Q (at the center of H is just

∡(AQ + BQ + CQ, AP + BP + CP ).

8.1.1 Fontené’s Theorems

Like Newton, Fontené also invented three theorems, but they’re more complicated.

Theorem 8.1.24 (Fontené I). Let △Ma Mb Mc be the medial triangle of △ABC. For any two isogonal
conjugates (P, Q), let T be the Poncelet point of (A, B, C, P ). Let △Qa Qb Qc be the pedal triangle of Q wrt.
△ABC. Let Ra = Qb Qc ∩ Mb Mc , etc. Then Qa Ra , Qb Rb , Qc Rc concur at T .

230
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Proof. Let T ′ , Q′a respectively be the reflections of T, Qa across Mb Mc , then T ′ ∈ (AO) and T ′ ∈ OQ
by (TODO 8.1.20). Thus we have ∡AT ′ Q = 90◦ , which also implies that T ′ is the Miquel point of
(CA, AB, Mb Mc , Qb Qc ). Thus from

∡Ra T ′ Qb = ∡Ra Mb A = ∡Q′a AQb = ∡Q′a T ′ Qb ,

we can get T ′ ∈ Ra Q′a , so by reversing the reflection we get T ∈ Qa Ra .

Corollary 8.1.25. The orthocenter of △Ra Rb Rc is the circumcenter of △Qa Qb Qc and is also the midpoint
of P Q.

Proof. Note that △Ra Rb Rc is the cevian triangle of T wrt. △Qa Qb Qc , so △Ra Rb Rc is self-conjugate wrt.
(Qa Qb Qc ). Thus by Brokard the orthocenter of △Ra Rb Rc is the circumcenter of (Qa Qb Qc ).

Theorem 8.1.26 (Fontené II). Given an arbitrary △ABC, and ℓ as a line through the circumcenter, let Q
be a moving point on ℓ. Then the pedal circle of Q passes through a fixed point on the nine-point circle of
△ABC. Additionally, this fixed point is the Poncelet point of (A, B, C, P ) where P is the isogonal conjugate
of Q.

Proof. Note that P and Q have the same pedal circle, which thus must pass through the Poncelet point of
(A, B, C, P ). But this is also the anti-Steiner point of OQ wrt. the medial triangle, so we are done.

Finally this is the third one. This has many wrong proofs on the internet.

Theorem 8.1.27 (Fontené III). Given △ABC, let P, Q be isogonal conjugates. Then the pedal circle of P
is tangent to the nine-point circle of △ABC if and only if P Q goes through the circumcenter of △ABC.

Proof. Let O, H, N respectively be the circumcenter, orthocenter, and nine-point center of △ABC. Let T1
be the Poncelet point of (A, B, C, P ), and let T2 be the Poncelet point of (A, B, C, Q).

(i) Only if: If P ’s pedal circle is tangent to the nine-point circle, then T1 = T2 . Thus P, Q lie on one
circumrectangular hyperbola of △ABC (which also goes through H), so just take its isogonal conjugate
and get a line through O.

(ii) If: Since O ∈ P Q, we have T1 = T2 also, call it just T . These two circles being tangent is equivalent to
M, N, T collinear, where M is the midpoint of P Q. Let the reflection of H over T be T ′ , and let H be
the conic through A, B, C, P, Q. Then H, T ′ ∈ H and T is the center of H. Let T ′ ’s antipode in (ABC)
be T ∗ , and let SX be the Steiner line of X in △ABC. Then by (TODO 1.4.5), ST ∗ ∥ P Q, so

(T ′ A, T ′ B; T ′ C, T ′ O) = (A, B; C, T ∗ ) = (SA , SB ; SC , ST ∗ ) = (HA, HB; HC, H∞P Q ).

231
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Thus D := H∞P Q ∩ OT ′ ∈ H. From the Parallel Chords Theorem (TODO 7.1.13) we know that the
midpoint of HD lies on T M , so when OT ′ is dilated by 1/2 around H we get N ∈ T M .

Combining this with (TODO 8.1.22), we can get a much more powerful result.

Corollary 8.1.28. Given a fixed △ABC, for an arbitrary point P , then the following statements are
equivalent:

• ∡(AP, BC) = 90◦ ;

P

• The pedal circle of P is tangent to the nine-point circle;

• If P, Q are isogonal conjugates then P Q goes through the circumcenter.

In reality, the locus of isogonal conjugates P, Q such that the circumcenter lies on P Q is a cubic called
the McCay cubic. We will revisit this later.

Practice Problems

Problem 1. Let △ABC be self-conjugate wrt. rectangular hyperbola H. Prove that I, I a , I b , I c lie on H.

8.2 Feuerbach

This is the most commonly seen named hyperbola, and can, unsuprisingly, prove Feuerbach’s theorem.

Theorem 8.2.1 (Feuerbach’s). The nine-point circle is always tangent to the incircle.

Of course, we already proved this back in (TODO 3.3.4), but here’s a direct proof of the machinery we’ve
just established.

∡(AH, BC) = 90◦ ,.

P
Proof. Follows by applying (TODO 8.1.28) to

We return to the Feuerbach point:

Definition 8.2.2. For an arbitrary △ABC, let the tangency point of the incircle and the nine-point circle be
the Feuerbach point of △ABC. Similarily, we can define A, B, C-Feuerbach points for the A, B, C-excircle
tangency points.

232
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Remark. Note that a lot of theorems that hold for the incircle also hold for the three excircles in a slightly
modified way (maybe with flipping around signed distances). We call this extraversion, and for the rest of
this section we will only prove stuff for the incircle Feuerbach point.

Let I, G, O, H, N, F e be the incenter, centroid, circumcenter, orthocenter, nine-point center, and Feuerbach
point as usual.

Since the Poncelet point of (A, B, C, P ) lies on the intersection of the pedal circle of P and the nine-point
circle, we have

Proposition 8.2.3. F e is the Poncelet point of (A, B, C, I).

Definition 8.2.4. We call the rectangular hyperbola through A, B, C, I the Feuerbach hyperbola.

Now we give some basic properties of the Feuerbach hyperbola.

Proposition 8.2.5. The Feuerbach hyperbola HF e is the isogonal conjugate of line OI.

Corollary 8.2.6. Let X104 be the antipode of H in the Feuerbach hyperbola. Then this is the fourth
intersection of HF e with Ω. Thus X104 is also the isogonal conjugate of ∞OI , so we will call it ∞∗OI .

Proposition 8.2.7. Let FI , FO be the feet from I, O to BC, let E = OI ∩ BC. Let IE , OE be the foot from
E to AI, AO. Then F e = FI IE ∩ FO OE .

Actually, we can extend this:

Proposition 8.2.8. Let P be an arbitrary point, and let Q be its isogonal conjugate. Let T be the Poncelet
point of (A, B, C, Q). Let TP , TO be the feet from P, O to BC, and let E be the intersection point of OP
and BC. Let PE , OE respectively be the feet from E to AP, AO, then

+
△AP O ∼ △T TP TO

and T = TP PE ∩ TO OE .

Proof. Let T ′ , TP′ , TO′ be the reflections of T, TP , TO over the midline parallel to BC. Then by (TODO
Fontene I’s proof), we know that T ′ ∈ OP and A, P, T ′ , TP′ and A, O, T ′ , TO′ are both concyclic. Thus

∡OP A = ∡T ′ TP′ A = ∡TO TP T, ∡AOP = ∡ATO′ T ′ = ∡T TO TP

+
so we have △AP O ∼ △T TP TO . Note that P, TP , E, PE and O, TO , E, OE are respectively concyclic, so we
have
∡T TP E = ∡AP O = ∡PE P E = ∡PE TP e

233
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

so T ∈ TP PE . Similarily T ∈ TO OE .

From the above proof we can derive some more results:

Proposition 8.2.9. Let OP intersect BC, CA, AB at D, E, F. Then

∡AT D = ∡BT E = ∡CT F = 90◦ .

Proof. Define T ′ similarily as the reflection of T across the BC-midline. Then T ′ is the foot from A to OP ,
so T ′ in (AD), and since the midpoint of AD lies on the BC-midline, we have T ∈ (AD). Similarily we have
T ∈ (BE), (CF ).

So we can actually say that OP is the orthotransversal of T wrt. △ABC. We have a pretty good
understanding of orthotransversals of points on the nine-point circle now! The proof of (TODO 8.2.9) can
also give us some more results:

Corollary 8.2.10. The line HT is the Steiner line of complete quadrilateral △ABC ∪ OP .

Proof. H is the orthocenter of △ABC and thus lies on the Steiner line, T lies on the line due to the above
proposition.

We go back to incenter configurations: from (TODO 7.1.16) we can get

Corollary 8.2.11. Let K be a line and let △K a K b K c be the anticevian triangle of K wrt. △ABC. Let
△Ma Mb Mc be the medial triangle of △ABC, then we have lines

K, L∞ , K a , K b , K c , Mb Mc , Mc Ma , Ma Mb

are all tangent to a common parabola.

Corollary 8.2.12. Let △DEF be the intouch triangle and let X, Y, Z respectively be

EF ∩ BC, F D ∩ CA, DE ∩ AB,

then the quadrilaterals formed by △DEF ∪ XY Z ∪ △Ma Mb Mc have a common Miquel point, and further,
this Miquel point is F e, and OI is the Steiner line.

Proof. First, by applying the above lemma we get a common parabola P which by (TODO 8.1.1) passes
through F e = (DEF ) ∩ (Ma Mb Mc ), which must be the foci of P. By the above tangencies we get F e is the
common Miquel point.

Then O is the orthocenter of △Ma Mb Mc and I is the orthocenter of △DEF which finishes.

234
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Let’s revisit the Feuerbach hyperbola. Note that the insimilicenter of the circumcircle Ω and the incircle
ω is X55 and the exsimilicenter is X56 , and that these are the isogonal conjugates of the Gergonne point Ge
and the Nagel point N a. Thus we have:

Proposition 8.2.13. Ge, N a ∈ HF e and

(H, I; N a, Ge)HF e = −1.

Proposition 8.2.14. OI is tangent to HF e .

We can extend this to other isoconjugations, too.

Proposition 8.2.15. Let φ be a point isoconjugation in △ABC, and let S be its fixed point. Then for any
non-degenerate conic C through A, B, C, S, then φ(C) is just the tangent from S to C.

Proof. Suppose φ(C) isn’t tangent to C, and intersects it at S, P . Then we have

φ(P ) ∈ φ(C ∩ φ(C)) ⊂ φ(C) ∩ C = {P, S}.

However S is a fixed point, so φ(P ) = P . Since C is non-degenerate, and the fixed points of φ are S and the
anticevian triangle of S, this results in a contradiction.

The dual of this theorem also holds, and we can use it to get (TODO 5.6.3) immediately.

Corollary 8.2.16. Let L be the reflection of H across O (the de Longchamps point). Then I, Ge, L are
collinear.

Proof. Note that I, N a, G are collinear (Nagel line), so we have

(IH, IO; IG, IL) = −1 = (H, I; N a, Ge)HF e = (IH, IO; IG, I(IGe ∩ E)),

so I, Ge, L are collinear.

(In fact, this is just a special case of Liang-Zelich theorem at (TODO 11.4.4)).

Example 8.2.17 (IMO 2000/6). Let AH1 , BH2 , CH3 respectively be the altitude in △ABC. Let the
intouch points on BC, CA, AB be T1 , T2 , T3 respectively. Consider the reflections of H2 H3 , H3 H1 , H1 H2
across T2 T3 , T3 T1 , T1 T2 . Prove that the triangle formed by these three lines has vertices on the incircle of
△ABC.

235
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Proof. Let △M1 M2 M3 be the median triangle where M1 is the midpoint of BC.

We consider the Feuerbach hyperbola of △ABC, denote it as HF e . Note that H2 H3 , T2 T3 are the polars
of H1 , T1 in HF e . This implies that U1 := H2 H3 ∩ T2 T3 is the pole of BC = H1 T1 wrt. HF e .

Let ℓi be the reflection of Hi+1 Hi+2 across Ti+1 Ti+2 , then we have B, C, H2 , H3 concyclic, so as such

∡(ℓ1 ) = 2∡(T2 T3 ) − ∡(H2 H3 ) = (∡AB + ∡AC) − (∡CA + ∡AB − ∡BC) = ∡BC,

and ℓ1 is parallel to BC.

Since BU1 , CU1 are both tangent to HF e , we have that the line ℓ1 := U1 ∞BC is the polar of M1 with
respect to HF e . As such, we get that ℓ2 ∩ ℓ3 is the pole of M2 M3 , let it be P1 .

So we have now identified the vertices of the triangle formed by these three lines! We prove P1 lies on the
incircle, and the rest follow by symmetry.

By Fontené I (TODO 8.1.24), T1 F e, M2 M3 , T2 T3 are concurrent. Take the polar of this and we get that
∞T2 T3 , P1 , T1 are collinear. By the Parallel Chords Theorem (TODO 7.1.13), F e, P1 , M1 are collinear. Thus
P1 is the intersection of lines T1 ∞T2 T3 and F eM1 .

Let P1′ be the second intersection of F eM1 with the incircle. Then from (TODO 8.2.8) we have

∡M1 T1 P1′ = ∡T1 F eM1 = ∡IAO = ∡(BC, ⊥ AI) = ∡(M1 T1 , T2 T3 ),

so we have P1′ = F eM1 ∩ T1 ∞T2 T3 = P1 , so P1 must lie on the incircle.

Since O × H = N a × N a∗ (bary product) and ON a∗ ∩ HN a = ∞OI , by (T ODO7.4.5) we have:

Proposition 8.2.18. ∞∗OI is ON a ∩ HN a∗ .

Practice Problems

Problem 1. Prove the four Feuerbach points are concyclic.

Problem 2. Let X be the A-mixtouch point. Let F e be the Feuerbach point. Prove AX ⊥ BC iff. A, F e, X
are collinear.

Problem 3 (2015 All-Taiwan P3). Let △DEF be the intouch triangle, let △Dt Et Ft be the image of △DEF
under some homothety from I. Prove that ADt , BEt , CFt concur at one point It , and that the locus of It is
the Feuerbach hyperbola.

Problem 4. Given two points O1 , O2 , let Ω1 , Ω2 respectively be two circles centered at O1 , O2 with radius
O1 O2 . Choose A ∈ Ω1 outside of Ω2 , and draw the two tangents from A to Ω2 , let these be lines ATb , ATc .

236
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Suppose these tangents intersect Ω1 at B, C. Let the orthocenter of △ABC be H, and let D be the reflection
of H across BC, and let OD intersect BC at E. If M, F respectively are the midpoints of BC, AH, prove
that Tb Tc is tangent to (DF M ).

Problem 5. Let Γ be the circumcircle of △ABC, and let it have circumcenter O. Let I a be the A-excenter.
Let OI a intersect Γ at K, and let the feet from K to CA, AB be E, F . Prove that EF intersects OI a on the
A−excircle.

Problem 6 (2017 CMO). Let (O), (I) respectively be the circumcircle and incircle of △ABC. Let the
tangents to (O) at B, C intersect at L. Let (I) touch BC at D. Let Y be the foot from A to BC, and let
AO and BC intersect at X. Let OI intersect (O) at P, Q. Prove that P, Q, X, Y are concyclic if and only if
A, D, L are collinear.

Problem 7. Let ℓ be the orthotransversal of P , and let H be the rectangular hyperbola through A, B, C, P .
Prove the tangent from P to H is perpendicular to ℓ.

8.3 Kiepert

Before discussing this hyperbola, we first revisit the Fermat and isodynamic points. We will also define some
notation.

Definition 8.3.1. In △ABC, define the Fermat points F1 , F2 as

∡BFi C = ∡CFi A = ∡AFi B = −i · 60◦ .

We call F1 the first Fermat point and we call F2 the second Fermat point. (These are X13 , X14 respectively.)

Definition 8.3.2. In △ABC, define the two isodynamic points S1 and S2 as the two common intersection
points of the Apollonian circles with foci B, C through A, foci C, A through B, and foci A, B through C.
(For a proof of existence, see (TODO 3.4)). Define S1 to be the point inside (ABC) and define S2 to be the
point outside (ABC). We call S1 the first isodynamic point and we call S2 the second isodynamic point.
(These are X15 , X16 .)

Let’s revisit the alternate definitions given back in (T ODO3.4.9).

Definition 8.3.3. Given △ABC, construct equilateral triangles on BC, CA, AB either all inside (2) or all out-
side (1) of the triangle, let these triangles be △A1(2) BC, △B1(2) CA, △C1(2) AB, then AA1(2) , BB1(2) , CC1(2)
intersect in the first and second Fermat points respectively.

Now time for the hyperbola.

237
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Proposition 8.3.4 (Kiepert’s hyperbola). A, B, C, F1 , F2 lie on one rectangular hyperbola HK , and F1 F2 is

a diameter of this hyperbola.

Proof. Note that ∡BF1 C + ∡BF2 C = 0◦ (cyclically). Thus F1 , F2 are antigonal conjugates. So by (TODO
8.1.17) they must lie on one rectangular hyperbola.

Of course, we call this hyperbola the Kiepert hyperbola of △ABC. Since Fi , Si are isogonal conjugates,
we have that the rectangular hyperbola HK is just the isogonal conjugate of the line S1 S2 , also called the
Brocard axis of △ABC.

(We will further investigate the Brocard axis in (TODO 8.3.1)).

Proposition 8.3.5. For any angle θ, choose Aθ , Bθ , Cθ such that

∡Aθ BC = ∡BCAθ = ∡Bθ CA = ∡CABθ = ∡Cθ AB = ∡ABCθ = θ,

we have AAθ , BBθ , CCθ concur at a point Kθ . The locus of Kθ as θ varies gives the Kiepert hyperbola HK .

Proof. Let Kθ = BBθ ∩ CCθ . When θ = ±60◦ , we have the two Fermat points; when θ = 90◦ , Kθ is just
H. So thus the locus of Kθ is a conic through B, C, F1 , F2 , H (degree can be bounded at max. 2). However
this is just the Kiepert hyperbola so we are done, and it follows by symmetry that AAθ also passes through
this.

Notably when θ = 0◦ , G is the centroid. This means that O, K lies on the Brocard Axis

Proposition 8.3.6. For any θ, we have (Kθ , K−θ ; H, G)HK = −1.

Proof. Note that A0 is the midpoint of BC, so it’s also the midpoint of Aθ A−θ , and A90◦ = ∞⊥BC , so we
have
A
(Kθ , K−θ ; H, G)HK = (Aθ , A−θ ; A0 , A90◦ ) = −1.

When we set θ = 60◦ , we get that (F1 F2 )(HG) is a harmonic quadrilateral on HK . Additionally,

Corollary 8.3.7. The line F1 F2 bisects GH.

Proof. Since (F1 F2 )(HG) is a harmonic quadrilateral on HK , we have

pHK (GH) = TG HK ∩ TH HK ∈ F1 F2

and F1 F2 passes through the center of HK , so by (TODO 7.1.14) we have F1 F2 bisects GH.

238
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

The following theorems will use Liang-Zelich (TODO 11.4.4) to prove stuff back in Chapter 5. You should
probably wait until you read that section before coming back.

Let E be the Euler line of △ABC.

Proposition 8.3.8. Let θ ∈ [−π/2, π/2), we have

1
t(Aθ ) = t(Bθ ) = t(Cθ ) = − .
2 cos 2θ

Proof. By symmetry, we only need to prove this for Aθ . Let △Oa Ob Oc be the Carnot triangle of △ABC
wrt. Aθ , let △OA OB OC be the image of △Oa Ob Oc under a homothety of −2 cos 2θ from Aθ . Then OA is
the orthocenter of △Aθ BC. Since OA OB ∥ Oa Ob ⊥ Aθ C ⊥ OA B we know that OA ∈ BOB , and similarily
we have OA ∈ COC . Thus we have AOA , BOB , COC concur at OA . Thus

1
t(Aθ ) = − .
2 cos 2θ

Let A∗θ , Bθ∗ , Cθ∗ , Kθ∗ respectively be the isogonal conjugates of Aθ , Bθ , Cθ , Kθ . By the previous theorems
we have A, Aθ , Kθ , A, Bθ , Cθ∗ , A, Bθ∗ , Cθ and A, A∗θ , Kθ∗ collinear (and same for the cyclic permutations), so
we actually have 12 collinearities!

Theorem 8.3.9. Using the same notation from (TODO 8.3.5), we have

1
t(Kθ ) = − .
2 cos 2θ

In particular, t(Fi ) = 1 (Fi is either Fermat point).

Proof. Let t0 be the claimed value, and let T ∈ E such that t(T ) = t0 , then we have T ∈ Bθ Bθ∗ ∩ Cθ Cθ∗ by
Liang Zelich. Note that the perspectrix of △BBθ∗ Cθ and △CCθ∗ Bθ is AA∗θ Kθ∗ , so by Desargues’ theorem we
know that BC, Bθ Cθ , Bθ∗ Cθ∗ are concurrent. Since B = Kθ Bθ ∩ Kθ∗ Bθ∗ , C = Cθ Kθ ∩ Cθ∗ Kθ∗ , by Desargues’s
and the fact that Bθ Cθ ∩ Bθ∗ Cθ∗ ∈ BC, we know that triangles △Kθ Bθ Cθ and △Kθ∗ Bθ∗ Cθ∗ are perspective, or
just T ∈ Kθ Kθ∗ , therefore t(Kθ ) = t0 as well.

Proposition 8.3.10. For three angles α, β, γ, Kα , Kβ , Kγ∗ are collinear if and only if

α + β + γ = 0◦ .

Proof. Let AKα∗ intersect HK at KαA , then we have αA + α = ∡BAC. By the proof of (TODO 8.3.9) we also
have Kα KαA ∩Kα∗ Kα∗ A = Kα KαA ∩OK ∈ BC. Similarily define αB , αC so it follows that Kα KαB ∩OK ∈ CA,

239
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Kα KαC ∩ OK ∈ AB. Let OK intersect BC, CA, AB respectively at D, E, F . Since Kγ∗′ = Kα Kβ ∩ OK, then

K A
(D, E; F, Kγ∗′ )OK =α (KαA , KαB ; KαC , Kβ )HK = (AαA , AαB ; AαC , Aβ )
A
= (A−∡BAC , A−∡CBA ; A−∡ACB , A−α−β ) = (A, B; C, K−α−β )HK
∗
= (D, E; F, K−α−β )

so α + β + γ ′ = 0◦ , and thus Kα , Kβ , Kγ∗ are collinear iff. α + β + γ = 0◦ .

Remark. Once you learn more about cubics, this is really just an example of the group law of Kiepert
hyperbola union Brocard Axis.

By choosing α = β = θ, γ = −2θ, we get that

∗
Corollary 8.3.11. Kθ K−2θ is tangent to HK at Kθ .

∗
Corollary 8.3.12. For any θ, we have K−2θ = pHK (Kθ Kθ+90◦ ). (For example, K, the symmedian point, is
the pole of the Euler line wrt. HK .)

In other words, we can now know the polar of any point on line OK wrt. HK ! Combining the previous
characteristics, we get this crucially important theorem.

Theorem 8.3.13 (The Essence of the Kiepert Hyperbola). Let N be the nine-point center of △ABC, let
Tθ = Kθ Kθ∗ ∩ K−θ K−θ
∗ ∗
, Pθ := Kθ K−2θ ∗
∩ K−θ K2θ , then for any angle θ, we have:

• G = Kθ K−θ
∗
∩ Kθ∗ K−θ ;

• K = Kθ K−θ ∩ Kθ∗ K−θ

∗
;

• N ∈ Kθ Kθ+90◦ ;

• Tθ ∈ E;

• Pθ ∈ E.

Proof. (i) and (ii) are trivial by applying (TODO 8.3.10) for α = 0, β = ±θ and a = ±θ, b = ∓θ. They are
also equivalent by DDIT.

For (iii), since pHK (Kθ Kθ+90◦ ) = K2θ ∈ OK, follows by the above, we have pHK (OK) ∈ Kθ Kθ+90◦ . Let
θ = 0◦ , then we have pHK (OK) ∈ E, but since O ∈ E, we have

(G, H; O, pHK (OK)) = −1,

so N = pHK (OK).

240
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

For (iv), note that

1 1
t(Kθ ) = − =− = t(K−θ ),
2 cos(2θ) 2 cos(−2θ)

so Tθ = E ∩ Kθ Kθ∗ ∈ K−θ K−θ

∗
.
∗
Finally, for (v), since K±θ K∓2θ = TK±θ HK , and Kθ HK−θ G is harmonic on HK , we have Pθ =
pHK (Kθ K−θ ) ∈ GH.

This trivializes many of our previous results!

Example 8.3.14 (Essence of the Fermat and Isodynamic points). In the previous theorem, set θ to ±60◦ .
Thus we have

• G = F1 S2 ∩ F2 S1 , K = F1 F2 ∩ S1 S2 , F1 S1 ∥ F2 S2 ∥ E;

• F1 , F2 , O, N concyclic (we call this the Lester circle);

• GK is the symmedian of △GF1 F2 .

Proof. The first line follows by the above.

For the Lester circle, by power of a point and (TODO 8.3.7) it remains to show that (GF1 F2 ) is tangent
to the Euler line, which is true by (TODO 8.1.7).

The final result then follows since K is the polar of the Euler line wrt HK .

We can also get that Fi Si is tangent to HK .

Example 8.3.15. Some other important points on the Kiepert hyperbola are the Vecten points, V1 =
K45◦ , V2 = K−45◦ . This suprisingly actually comes up, in 2001 ISL G1! We have

• O is the pole of V1 V2 wrt. HK .

• V1 V2 = N K;

• H = V1 V1∗ ∩ V2 V2∗

Note that the map sending Tθ to Pθ is an automorphism of E, so in theory we could calculate the positions
of these points.

Proposition 8.3.16. For an angle θ, we have Tθ G = 2 · GPθ , and thus

OPθ = 2 cos 2θ · Pθ N.

If we set Qθ = Kθ K−θ ∩ E, then we have [Pθ → Qθ ] ∈ Aut(E), so we have

Proposition 8.3.17. For any θ, 3 · N Qθ = 2 cos 2θ · Qθ O.

241
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

8.3.1 Brocard, Humpty, Dumpty

This is a slight tangent, but let’s touch on the Brocard points.

Proposition 8.3.18. In a triangle △ABC, there exists two unique points Br1 , Br2 such that

∡BABr1 = ∡CBBr1 = ∡ACBr1 , ∡Br2 AC = ∡Br2 BA = ∡Br2 CB.

We call these points the first Brocard point and second Brocard point. These points are isogonal conjugates
too, obviously.

Proof. Suppose there exists a Br1 satisfying these conditions. Then from ∡CBBr1 = ∡ACBr1 we know
that (BBr1 C) and CA are tangent. Similarily, we know (CBr1 A) and AB are tangent too, and (ABr1 B)
and BC are tangent. Thus, if we let ΩA1 to be a circle through B, C tangent to CA, and similarily define
ΩB1 , ΩC1 . Then if these three circles concur, Br1 is just their concurrency point. Since

(B − C)ΩA1 + (C − A)ΩB1 + (A − B)ΩC1 = ∡BCA + ∡CAB + ∡ABC = 0◦ ,

we have they concur. Similarly we can prove that Br2 also exists.

We can also re-interpret this stuff with Trig Ceva: we want to find a angle θ such that

sin θ sin θ sin θ

· · = 1,
sin(α − θ) sin(β − θ) sin(γ − θ)

where α, β, γ are the angles of the triangle. We call this angle θ the Brocard angle of △ABC.

When θ = 0◦ , the LHS is just 0. When θ = min α, β, γ, the LHS is positive infinity. Thus, by the
Intermediate Value Theorem, θ is between 0, min α, β, γ. In other words, Br1 , Br2 both lie inside △ABC.

Additionally, the circles we defined in the proof ΩAi , ΩBi , ΩCi ’s pairwise intersections have names.

Definition 8.3.19. Let PA be the second intersection point of ΩB2 , ΩC1 , and let QA be the second
intersection point of ΩB1 , ΩC2 . We call PA , QA , the A-Humpty and A-Dumpty points. Similarily, we can
define PB , PC , QB , QC for B, C too.

From
∡CPA A + ∡CQA A = ∡BCA + ∡CAB + ∡CBA

∡APA B + ∡AQA B = ∡ABC + ∡CAB = ∡ACB

by (TODO 1.3.6) we get

242
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Proposition 8.3.20. The A-Humpty point and A−Dumpty point are isogonal conjugates.

Proposition 8.3.21. Let H, G, PA be the orthocenter, centroid, and A-Humpty points. Then B, C, H, PA
are concyclic and PA is also the foot from H to the A-median.

Proof. From

(C − B)(BHC) + (A − C)ΩB2 + (B − A)ΩC1 = (−∡BAC) + ∡BCA + ∡ABC = 0◦ ,

we have PA ∈ (BHC). Let GA be a point such that (AGA )(BC) is a parallelogram, then GA lies on (BHC)
and A, G, GA are collinear. From

∡BPA GA = ∡BCGA = ∡CBA = ∡BPA A,

we get that A, PA , GA are collinear. Combining this with ∡HPA GA = ∡HBGA = 90◦ , we get that PA is the
foot from H to PA GA = AG.

Remark. Another fast proof of concyclicity is inversion across the polar circle at H swapping A and the
foot of the A-altitude.

This foot property is also why the A-Humpty point is also sometimes called the HM-point.

Proposition 8.3.22. Let O, K, QA be the circumcenter, symmedian point, and A−Dumpty point. Then
B, C, O, QA are concyclic and QA is the foot from O to AK.

Proof. It’s basically the same proof as from before. Since

(C − B)(BOC) + (A − C)ΩB1 + (B − A)ΩC2 = 2 · ∡BAC + ∡CAB + ∡CAB = 0◦ ,

we have QA ∈ (BOC). Let KA be the pole of BC. Then KA ∈ (BOC) and from A, K, KA collinear, and

∡BQA KA = ∡BCKA = ∡BAC = ∡BQA A,

we have A, QA , KA collinear. Combining this with ∡OQA KA = ∡OBKA = 90◦ , we get that QA is the foot
from O to QA KA = AK.

Remark. The Dumpty point is also called the “midpoint of symmedian chord point”, for obvious reasons.

Theorem 8.3.23. Let O, K, Br1 , Br2 be the circumcenter, symmedian point, and the two Brocard points.
Let BrA = BBr1 ∩ CBr2 , BrB = CBr1 ∩ ABr2 , BrC = ABr1 ∩ BBr2 , and let QA , QB , QC respectively be

243
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

the A, B, C-Dumpty points. Then O, K, Br1 , Br2 , BrA , BrB , BrC , QA , QB , QC are concyclic with diameter
OK. We call this the Brocard circle ΓBr and we call OK the Brocard axis.

Proof. Since ABr1 , ABr2 are isogonal lines in ∡BAC, we have

∡BrA BC = ∡Br1 BC = ∡Br1 AB = ∡CABr2 = ∡BCBr2 = ∡BCBrA ,

so BrA lies on the perpendicular bisector of BC. Similarily BrB lies on the perpendicular bisector of CA
and BrC lies on the perpendicular bisector of AB. Then since Br1 ∈ ΩC1 , Br2 ∈ ΩB2 , we get

∡BrB Br1 BrC = ∡BAC = ∡BrB OBrC = ∡CAB,

so we have O, Br1 , Br2 , BrB , BrC , BrA are all concyclic on the Brocard circle ΓBr . Since QA is the second
intersection of ΩB1 and ΩC2 we have

∡Br1 QA Br2 = ∡Br1 QA A + ∡AQA Br2 = ∡Br1 CA + ∡ABBr2

= ∡Br1 BC + ∡BCBr2 = ∡Br1 BrA Br2 ,

so QA ∈ ΓBr . Similarily QB , QC ∈ ΓBr . Finally, (TODO 8.3.22) tells us that QB , QC are just the feet from
O to BK, CK, so O, K, QB , QC are concyclic and the diameter is OK.

Corollary 8.3.24. OK is the perpendicular bisector of Br1 Br2 .

Proof. Continuing from above, since BrA lies on the perpendicular bisector of BC,

∡Br2 Br1 O = ∡Br2 BrA O = ∡(CBrA , ⊥ BC)

= ∡(⊥ BC, BBrA ) = ∡OBrA Br1 = ∡OBr2 Br1 ,

so △OBr1 Br2 is an isosceles triangle with top vertex O. Since O is the antipode of K in the Brocard circle,
OK is the perp. bisector of Br1 Br2 .

Proposition 8.3.25. Let △Br1a Br1b Br1c , △Br2a Br2b Br2c be the pedal triangles of Br1 and Br2 . Then

+ +
△Br1c Br1a Br1b ∼
= △Br2b Br2c Br2a ∼ △ABC.

Proof. We have

∡Br1b Br1a Br1c = ∡ACBr1 + ∡Br1 BA = ∡CBBr1 + ∡Br1 BA = ∡CBA.

244
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed
+
Similarly, ∡Br1c Br1b Br1a = ∡ACB, ∡Br1a Br1c Br1b = ∡BAC. which implies △Br1c Br1a Br1b ∼ △ABC,
+
and we also have △Br2b Br2c Br2a ∼ △ABC by symmetry. The proof of congruency is because the two
triangles share a circumcircle.

Proposition 8.3.26. Keeping the notation from above, let Br1∗ be the isogonal conjugate of Br1 wrt.
△Br1a Br1b Br1c , and let Br2∗ be the isogonal conjugate of Br2 in △Br2a Br2b Br2c . Then

• Br1∗ , Br2∗ lie on the circle (Br1 Br2 );

• K = Br1 Br2∗ ∩ Br2 Br1∗ .

Proof. It’s obvious that the spiral-similarity center between △Br1c Br1a Br1b and △Br2b Br2c Br2a is their
common circumcenter, OBr , which is also the midpoint of Br1 Br2 , since those two are antipodes in their
common pedal circle. Let r be this spiral similarity at OBr . Since △Br1c Br1a Br1b ’s spiral similarity
center with △ABC is just Br1 , we have that Br1∗ is the second Brocard point of △Br1c Br1a Br1b . Thus
r(Br1∗ ) = Br2 and r(Br1 ) = Br2∗ . This tells us that

OBr Br1 = OBr Br2 = OBr Br1∗ = OBr Br2∗ ,

and Br1∗ , Br2∗ lie on (Br1 Br2 ). From

∡(Br1 Br1∗ , Br1 Br2 ) = ∡(Br1b Br1c , CA) = 90◦ + ∡BABr1

= ∡OBra Br1 = 90◦ + ∡KBr2 Br1 ,

we can get Br2 Br1∗ ⊥ Br1 Br1∗ ⊥ KBr2 , so K ∈ Br2 Br1∗ . Similarily, we also have K ∈ Br1 Br2∗ .

Proposition 8.3.27. Let CBr be the inellipse of △ABC with Br1 , Br2 as foci. We call this the Brocard
inellipse of △ABC. Then CBr touches the sides of △ABC at the vertices of the cevian triangle of the
symmedian point.

Proof. We trig-bash. Let D, E, F respectively be the touchpoints of CBr with BC, CA, AB. From (TODO
7.3.5), since Br1 , Br2 lie inside △ABC, we don’t need directed lengths and we have

BD CA BBr1 BBr2
= · · .
DC AB Br1 C Br2 C

245
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

We also have

BBr1 AB · | sin ∡BABr1 / sin ∡ABr1 B|

=
Br1 C CA · | sin ∡Br1 AC/ sin ∡CBr1 A|
AB sin ∡BABr1 sin ∡CAB AB · BC sin ∡BABr1
= · · = · .
CA sin ∡Br1 AC sin ∡ABC CA2 sin ∡Br1 AC

And similarly, we have

BBr2 AB 2 sin ∡BABr2
= · .
Br2 C BC · CA sin ∡Br2 AC

Thus 2
CA AB · BC sin ∡BABr1 AB 2

BD sin ∡BABr2 AB
= · · · · = ,
DC AB CA2 sin ∡Br1 AC BC · CA sin ∡Br2 AC CA

so D = AK ∩ BC. By the same logic, we have E = BK ∩ CA, F = CK ∩ AB.

The Brocard inellipse CBr is actually the envelope of cross-ratio preserving transformations on a circle.

Proposition 8.3.28. Consider the cross-ratio preserving transformation φ : Ω = (ABC) → Ω such that
φ(A) = B, φ(B) = C, φ(C) = A, then the envelope of

{P φ(P ) | P ∈ Ω}

is the Brocard inellipse.

Proof. From (TODO 7.A.15), the envelope of {P φ(P ) | P ∈ Ω} is a conic C. By setting P = A, B, C we can
get that this is an inconic of △ABC. Let △KA KB KC be the circumcevian triangle of the symmedian point
wrt. △ABC. Then we have

(B, KB ; C, A) = −1 = (A, KA ; B, C)

= (φ(A), φ(KA ); φ(B), φ(C)) = (B, φ(KA ); C, A),

so we have φ(KA ) = KB , φ(KB ) = KC , φ(KC ) = KA .

To prove that C is the desired inellipse, we know a conic is fully determined by six tangent lines by
Brianchon’s theorem. Three of these lines are simply the sides of the triangle, so we only need to find three
more. Thus we can just prove that KB KC , KC KA , KA KB are tangent to CBr . By symmetry, we only need to
prove that KB KC is tangent to CBr . Let S, T respectively be the intersection points of KB KC and CA, AB.
Let △DEF be the C-touchpoint triangle wrt. △ABC, then we have

K K
(A, E; C, S) =B (A, B; C, KC ) = −1 = (C, A; B, KB ) =C (F, A; B, T ).

246
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Since AF, EA respectively are tangent to CBr at F, E, and BC is also obviously tangent, we have ST = KB KC
is also tangent to CBr .

Since φ3 (A) = A, φ3 (B) = B, φ3 (C) = C, we have that φ3 has four fixed points and is thus a identity
transformation. Thus for any point P ∈ Ω, CBr is tangent to the triangle △P = P φ(P )φ2 (P ). Thus from the
definition of △P we can get that φ(△P ) = △P , as it is cyclic with period 3. This tells us that CBr has to be
the inellipse with the two Brocard points as foci.

Proposition 8.3.29. Keeping the previous notation, for a point P ∈ Ω, for triangle △P = △P φ(P )φ2 (P ),
its circumcenter, symmedian point, first Brocard point, and second Brocard point all coincide with the
corresponding points of △ABC, and as P moves, the Brocard angle ∡φ(P )P Br1 is constant. At this point,
the Brocard inellipse of △P touches its three sides at the vertices of the cevian triangle of the symmedian
point.

Proof. Obviously the circumcenter of △P is just O. We define the involution ψ : Ω → Ω on the circumcircle
as ψ(Q) = QBr1 ∩ Ω, and the cross-ratio preserving transformation Φ = ψ · φ−1 . Then we have

(Φ(A) − A)Ω = (Φ(B) − B)Ω = (Φ(C) − C)Ω = θ := ∡BABr1 ,

so thus for all Q ∈ Ω we have (Φ(Q) − Q)Ω = θ. Set Q = AP = P, BP = ϕ(P ), CP = ϕ2 (P ). Then we get

∡BP AP Br1 = ∡BP AP Φ(BP ) = θ,

∡CP BP Br1 = ∡CP BP Φ(CP ) = θ,

∡AP CP Br1 = ∡AP CP Φ(AP ) = θ,

so Br1 is the first Brocard point of △P . Similarly, Br2 is the second Brocard point of △P . Finally note that
△P ’s symmmedian point is the antipode of O in (OBr1 Br2 ), and is thus also the symmedian point of the
reference triangle. So since CBr (△P ) = CBr , and the CBr -touchpoints in △P is the cevian triangle of the
symmedian point of △ABC, it is also the symmedian point of △P .

Practice Problems

Problem 1 (τ ). Choose three points from the set A, B, C, F1 , F2 . Prove the Euler line of this triangle goes
through G.

Problem 2. Let OA , OB , OC respectively be the circumcenters of △Si BC, △ASi C, △ABSi . Prove that the
lines AOA , BOB , COC are concurrent on OK.

247
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Problem 3. Let N be the nine-point circle of △ABC, and let I a , I b , I c be the three excircles. Prove that N
and the midpoints of F1 S1 , F2 S2 lie on a common circumrectangular hyperbola of △I a I b I c .

Problem 4. Let △Ha Hb Hc , △Ma Mb Mc respectively be the orthic and medial triangles. Prove that for a
point P , the isogonal conjugate of P wrt. the orthic triangle lies on the line connecting the isogonal and
isotomic conjugates of P wrt. the medial triangle.

Problem 5. Let Kθ be a point on the Kiepert hyperbola with corresponding angle θ. Let △XY Z be the
circumcevian triangle of P wrt. the pedal triangle of P wrt. △ABC. Then prove that Kθ lies on the Kiepert
hyperbola of △XY Z, with corresponding angle −θ.

Problem 6. Prove that a quadrilateral ABCD has Brocard points (that is, points P, Q such that ∡BAP =
∡CBP = ∡DCP = ∡ADP, ∡QAD = ∡QBA = ∡QCB = ∡QDC) if and only if ABCD is a harmonic cyclic
quadrilateral.

8.4 Jerabek

We already know that for △ABC and a point P , we can define some points, lines, conics etc. with relation to
P . In previous sections, we defined the orthotransversal and the trilinear polar, let’s see some nice properties
of them.

Proposition 8.4.1. Given △ABC and a point P , let OP , t(P ) respectively be the orthotransversal and
trilinear polar of P . Let SP be the polar of P wrt. its pedal circle ω (we will also call this pω (P )), then
OP , t(P ), SP are concurrent.

Proof. Let Ω be an arbitrary circle centered at P , and define A∗ = pΩ (BC). Similarily define B ∗ , C ∗ . Then
by using polarity, pΩ (OP ) is the orthocenter of △A∗ B ∗ C ∗ , and pΩ (t(P )) is the centroid of △A∗ B ∗ C ∗ . Note
that ω, Ω, (A∗ B ∗ C ∗ ) are coaxal by inverting about ω.

Let O∗ be the circumcenter o △A∗ B ∗ C ∗ and let Γ be the circumcircle, and we consider OP ’s intersections
with ω, Ω, Γ. Let the two intersections of OP with C be C1 , C2 Since inversion preserves cross-ratio, we have
that
Ω
(P, SP ∩ OP ; ω1 , ω2 ) = (∞, pΩ (SP ); Γ1 , Γ2 )

so pΩ (SP ) is the circumcenter of △A∗ B ∗ C ∗ too. Thus the three poles are collinear by the Euler line, so
OP , t(P ), SP are concurrent.

When P lies on ABC, SP is just the Steiner line of P . At this time we have:

248
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

Proposition 8.4.2. Given △ABC and P on (ABC), then the orthotransversal of P , the trilinear polar of
P , and the Steiner line of P are concurrent.

We now look at polarity wrt. a circle centered at P . There’s some nice properties about this too:

Proposition 8.4.3. Given △ABC and P on its circumcircle, let Ω be a circle centered at P . Then

−
• pΩ (△ABC) ∼ △ABC and P lies on its circumcircle.

−
• If (Q, Q∗ ) are two isogonal conjugates in △ABC, choose Q′ such that pΩ (△ABC) ∪ Q′ ∼ △ABC ∪ Q,
then Q′ ∈ pΩ (Q∗ ).

Proof. (i) follows by angle chasing. For (ii), we consider the composition of a reflection and a spiral similarity
φ such that φ(A′ ) = A, φ(B ′ ) = B, φ(C ′ ) = C. Let C be the diagonal conic centered at P (under isogonal
conjugation) through I, IA , IB , IC , which is possible because P is on the circumcircle. Now look at the
point-to-point transformation Φ := pC ◦ φ ◦ pΩ . Obviously, A, B, C, P are fixed points of Φ, and Φ is a bijective
transformation that preserves incidence relations and cross ratio. Thus from (TODO 7.A.10), and four fixed
points, we get that this transformation is just the identity transformation! Thus

φ(pΩ (Q∗ )) = pC (Q∗ )

and from properties of isogonal conjugation we know that Q ∈ pC (Q∗ ), so Q′ ∈ φ−1 (pC (Q∗ )) = pΩ (Q∗ ).

Then modifying the proof of (TODO 8.4.1), we have

Corollary 8.4.4. Given △ABC and P on its circumcircle, the orthotransversal, trilinear polar, and Steiner
line of P respectively pass through the circumcenter, symmedian point, and orthocenter of △ABC.

Let’s consider the locus of the concurrency point of these three lines as P moves.

Proposition 8.4.5. Given △ABC and moving point P on its circumcircle, the locus of the concurrency point
Q of the orthotransversal, trilinear polar, and Steiner line is a circumrectangular hyperbola. Additionally,
P Q goes through a fixed point on (ABC).

Proof. Let O, H be the circumcenter and orthocenter. For a point P ∈ (ABC), choose D on BC such that
∡AP D = 90◦ . Let A′ , P ′ be the reflections of A, P across BC. Then

(OA, OB; OC, OD) = O(A, B; C, P ) = (HA′ , HB; HC, HP ′ ),

249
AoPS Chapter 8. Circumrectangular Hyperbolas - Hidden/Completed

so Q = OD ∩ HP ′ ∈ H, where H is the hyperbola through A, B, C, O, H (by symmetry OD is the ortho-

transversal of P ). Let H ∗ be the fourth intersection point of H with (ABC). Then

(H ∗ A, H ∗ B; H ∗ C, H ∗ P ) = (HA′ , HB; HC, HP ′ ) = (HA, HB; HC, HQ) = (H ∗ A, H ∗ B; H ∗ C, H ∗ Q),

so H ∗ ∈ P Q.

Definition 8.4.6. We call this hyperbola the Jerabek hyperbola of △ABC. The fixed point is just the
fourth intersection point of the Jerabek hyperbola with (ABC).

Immediately we get the following proposition:

Proposition 8.4.7. The Jerabek hyperbola is the isogonal conjugate of the Euler line.

Example 8.4.8. The isotomic conjugate X69 of H lies on the Jerabek hyperbola.

Practice Problems

Problem 1. Let E this intersect CA, AB at E, F . Prove the Euler line of △AEF is parallel to BC.

(This is a duplicate of a problem example from earlier)

Problem 2. Given △ABC and a point P , let OP , t(P ) respectively be the orthotransversal and trilinear
polar of P . Let △Qa Qb Qc be the pedal triangle of the isogonal conjugate of P . If H ′ , G′ are the orthocenter
and centroid of △Qa Qb Qc , prove:

• QH ′ ⊥ OP ,

• QG′ ⊥ TP .

Problem 3. Let P be a point on the Euler line of △ABC. Let H be the orthocenter of △ABC, and let Q
be the isogonal conjugate of P . Prove that HQ is tangent to the conic through A, B, C, H, P .

Problem 4. Let N be the nine-point center of △ABC, and let Ko be the Kosnita point, that is, the isogonal
conjugate of N . Prove N Ko passes through the anti-Steiner point of the Euler line of △ABC (we call this
the Euler reflection point, X110 ), and prove that this point is the antipode of the fourth intersection of the
Jerabek hyperbola with (ABC).

250
Chapter 9

Perfect Six-Point Sets and The Isoptic

Cubic

9.1 Perfect Six-Point Sets

For a thorough understanding of the isoptic cubic, we first introduce a very useful tool - the perfect six-point
set. This tool was developed by Chinese geometers Ye Zhonghao and Shan Zun (it was originally named
“Perfect Hexagon”), and wrote a poem.

The garden of mathematics is vast,

Geometry is a unique branch,
In the delightful mood of spring,
Come and admire the little flowers.

Definition 9.1.1. We call (AD)(BE)(CF ) a perfect six-point set if (BE)(CF ), (CF )(AD), (AD)(BE)
share the same Miquel Point M . In this case, M is referred as the Miquel point of (AD)(BE)(CF ).

More generally, we call the 2n points (X1 Y1 ) . . . (Xn Yn ) a perfect 2n-point set if there exists a point M
such that for all i, j, the Miquel point of (Xi Yi )(Xj Yj ) is M . In particular, for any perfect six-point set with
Miquel point M , (AD)(BE)(CF )(M ∞) is a perfect eight-point set, where ∞ is any point at infinity.

Obviously, from this definition we know that a perfect six-point set (AD)(BE)(CF ) retains this property
under reflections, translations, homotheties and rotations.

If M ∈
/ L∞ , then there exists a negative inversion (in other words, the composition of an inversion
and a reflection) Φ on points of △M ∞i ∞−i sending A, B, C to D, E, F respectively. This is called

251
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

the Clawson-Schmidt transformation (the C-S transformation) on (AD)(BE)(CF ). If M ∈ L∞ , then in

reality M can be any point on L∞ , and (BE)(CF ), (CF )(AD), (AD)(BE) are all parallelograms. That is,
(AD)(BE)(CF ) is a hexagon with parallel opposite sides, and in this case we simply denote M as ∞. Then,
the C-S transformation on (AD)(BE)(CF ) is simply a reflection about their common midpoint. Thus, given
five points B, C, D, E, F , there exists a unique point A such that (AD)(BE)(CF ) is a perfect six-point set
(here, all points at infinity are viewed as the same point ∞, we are working in CP1 ).

Example 9.1.2. Let Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) be a complete quadrilateral and Aij := ℓi ∩ ℓj be the vertices of Q.
Then, (A23 A14 )(A31 A24 )(A12 A34 ) is a perfect six-point set.

Proposition 9.1.3. A six-point set (AD)(BE)(CF ) is perfect if and only if

−−→ −−→ −→
BD CE AF
−−→ ◦ −→ ◦ ⃗ = −1 ∈ S
ab
(1)
DC EA F B
−−→
XW
where −1 represents any reflection about a point. If W ∈ L∞ \ {∞i , ∞−i }, X, Y ∈
/ L∞ , we define −−→ as −1.
WY

−−→ −−→ −−→ −−→

(Note: XY represents the signed length, defined as XY = −Y X and XY = XY .)

In the special case of (TODO Proposition 9.1.2), this property is just the familiar Menelaus Theorem.

Proof. From the existence and uniqueness, we only need to prove that when (AD)(BE)(CF ) is a perfect
six-point set,
−−→ −−→ −→
BD CE AF
−−→ ◦ −→ ◦ −−→ = −1.
DC EA F B
Let M be the Miquel point of (AD)(BE)(CF ). Then,

−−→ −−→ −→ −−→ −−→ −−→

CE M C AF M A BD MB
−−→ = ,
−−→ −−→ = ,
−−→ −→ = −−→ .
BF M B CD M C AE MA

Proposition 9.1.4. The hexagon (AD)(BE)(CF ) is a perfect six-point set if and only if there exists two
points X+ , X− such that (AD)(X+ X− ), (BE)(X+ X− ), (CF )(X+ X− ) are all harmonic quadrilaterals.

Proof. For the forward direction, let M be the Miquel point and let Φ be Clawson-Schmidt conjugation. I
claim that X+ , X− are really just the two fixed points of Clawson-Schmidt conjugation! We now show that
these work by showing (AD)(X+ X− ) is a harmonic quadrilateral.

[The original book presents this without proof: a quick proof I’m aware of is with Cayley-Bacharach
theorem on the isoptic cubic: see the chapter on cubics for more information.] Let I, J be the two circle

252
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

points. Then (IA, ID; IX+ , IX− ) = −1, and vice versa for J. Additionally A, D, X+ , X− , I, J are conconic
so we’re done.

For the backwards direction, let M be the midpoint of X+ X− . Since (X+ , X− ; M, X+ X− ∩ L∞ ) = 1, we

have that X− is one of the vertices of the anticevian triangle of X+ wrt. △M IJ. Thus there exists a point
isoconjugation Φ in △M IJ such that X+ , X− are fixed. Since

(IA, IΦ(A); X+ , X− ) = −1 = (IA, ID; IX+ , IX− ),

(and similarily for J), we have Φ(A) = D and thus Φ(B) = E, Φ(C) = F.

Proposition 9.1.5. Perfect six-point sets are preserved under inversion: that is, if (AD)(BE)(CF ) is a
perfect six-point set, then for inversion J we have that (AJ DJ )(B J E J )(C J F J ) is also a perfect six-point set.

Proof. Let M, Φ respectively be the Miquel point and Clawson-Schmidt conjugation. We can view J as just
some point isoconjugation in △OIJ, where O is the center of the inversion.

To prove (AJ DJ )(B J E J )(C J F J ) is a perfect six-point set, we just need to prove there exists a point
X and a isoconjugation on △XIJ such that AJ ←
→ DJ , B J ←
→ EJ, C J ←
→ F J . This isoconjugation is just
J ◦ ϕ ◦ J, and X is just J(ϕ(O)): we note that J ◦ ϕ ◦ J is restricted by being a involution on TI, TJ, and
sending L∞ to J ◦ ϕ(IO) = IX, J ◦ ϕ(JO) = JX.

Example 9.1.6. Let ABCD be a cyclic quadrilateral with circumcenter O. Let AC and BD intersect at P .
Then (AC)(BD)(OP ) is a perfect six-point set.

Proof. We consider inversion about O. We know that P is sent to the Miquel point M and O is sent to ∞,
and A, B, C, D are fixed. However (AC)(BD)(∞M ) is obviously a perfect six-point set, so we’re done.

Example 9.1.7. Let ABCDEF be a cyclic hexagon. Then (AD)(BE)(CF ) is a perfect six-point set if and
only if AD, BE, CF are concurrent.

Proof. Let P = CF ∩ AD, P ′ = AD ∩ BE, then by the above results we know that (CF )(AD)(OP ) and
(AD)(BE)(OP ′ ) are respectively both perfect six-point sets. Thus (AD)(BE)(CF ) is a perfect six-point set
if and only if P = P ′ , or AD, BE, CF concurrent.

Proposition 9.1.8. If (AD)(BE)(CF ) is a perfect six-point set and the six points are not all concyclic, then

(DEF ), (DBC), (AEC), (ABF )

are concurrent at P , and

(ABC), (AEF ), (DBF ), (DEC)

253
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

are concurrent at Q. Further, (AD)(BE)(CF )(P Q) is a perfect eight-point set.

Proof. Let the second intersection of (AEC) and (ABF ) be point P . We consider the inversion J centered at
P . Then obviously AJ , E J , C J are collinear and AJ , B J , F J are also collinear. Let (D′ )J = B J C J ∩ E J F J ,
then (AJ (D′ )J )(B J E J )(C J F J ) is a perfect six-point set. Thus from (AD)(BE)(CF ) being a perfect six-point
set, we get that D = (D′ )J◦J = D′ , so (DEF ), (DBC) both pass through P .

By Miquel’s theorem, (AJ B J C J ), (AJ E J F J ), (DJ B J F J ), (DJ E J C J ) go through the Miquel point M of
(B J E J )(C J F J ). Thus (ABC), (AEF ), (DBF ), (DEC) are concurrent at a point Q. Therefore (AJ DJ )(B J E J )(C J F J )(M ∞)
is a perfect eight-point set, so (AD)(BE)(CF )(QP ) is also a perfect eight-point set.

We call this kind of perfect eight-point set a cyclic perfect eight-point set. By definition, (BE)(CF )(P Q)(AD)
is also a perfect eight-point set: these four points are symmetric.

Remark. If (AD)(BE) is a cyclic quadrilateral, let X, O be defined as AD ∩ BE and the circumcenter.

Then we have that X, O, P, Q are concyclic by quartic Cayley-Bacharach.

Here’s a nice example of this:

Example 9.1.9. Let PA , PB , PC be the reflections of arbitrary point P across BC, CA, AB. Then (BPA C)
(CPB A), (APC B), (PA PB PC ) are concurrent (at the antigonal conjugate of P , see (TODO 8.1.16)).

Proposition 9.1.10. If (AD)(BE)(CF )(P Q) is a cyclic perfect eight-point set, then the midpoints of
AD, BE, CF, P Q are concyclic.

Proof. Let M be the Miquel point of (AD)(BE)(CF )(P Q), and let MA , MB , MC , MP respectively be the
midpoints of segments AD, BE, CF, P Q. Since the isogonal conjugate of the Miquel point is the point at
infinity along the Newton line (TODO 7.3.9), we have

∡(MC MA ) + ∡(MB MP ) = (∡AC + ∡AF − ∡(AM )) + (∡(P B) + ∡(P E) − ∡(P M ))

= (A + B + C + P )(ABCP ) + (A + E + F + P )(AEF P ) − ∡AM − ∡(P M ).

In the above expression note that you can swap (B, E), (C, F ), so it’s also equal to ∡(MA MB )+∡(MC MP ).

Example 9.1.11. In (TODO 9.1.9), the center of the circle is N .

Proposition 9.1.12 (Reconnecting the broken mirror). Let AF BDCE be a hexagon, and let D, E, F ’s
−
reflections over BC, CA, AB be D′ , E ′ , F ′ . Then △DEF ∼ △D′ E ′ F ′ if and only if (AD)(BE)(CF ) is a
perfect six-point set.

254
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic
−
Proof. Let E ′′ , F ′′ respectively be the reflections of E ′ , F ′ across BC. Then △DE ′′ F ′′ ∼ △D′ E ′ F, and
therefore
− + +
△DEF ∼ △D′ E ′ F ′ ⇐⇒ △DEF ∼ △DE ′′ F ′′ ⇐⇒ △DEE ′′ ∼ △DF F ′′ .

Let O be the circumcenter of △ABC, then

+ +
△CEE ′′ ∼ △OAB, △BF F ′′ ∼ △OAC.

This tells us that

BA CE DF CE · AB/OA DF DF/F F ′′
· · =− · =− .
AC ED F B BF · AC/OA DE DE/EE ′′
+
Thus (AD)(BE)(CF ) is a perfect six-point set if and only if △DEE ′′ ∼ △DF F ′′ .

Proposition 9.1.13. If (AD)(BE)(CF ) is a perfect six-point set, then there exists a point P such that

+ + +
△P BC ∼ △AF E, △AP C ∼ △F BD, △ABP ∼ △EDC.

+
Proof. Let M be the Miquel point of (AD)(BE)(CF ), and choose P such that △P BC ∼ △AF E. Then
+ + + +
△M BC ∼ △M F E tells us that △P BC ∩ M ∼ △AF E ∩ M . Thus △M AP ∼ △M F B, and with △M CA ∼
+ +
△M DF, we now have △ABC ∪M ∼ △F BD ∪M. By the same logic we have △ABP ∪M ∼ △EDC ∪M .

We call P the internal point of △DEF wrt. (AD)(BE)(CF ) (and obviously we can define internal points
for the other seven triangles too.).

Practice Problems

Problem 1. Let M be the Miquel point of perfect six-point set (A1 A4 )(A2 A5 )(A3 A6 ), and let Mi be the
foot from M to Ai Ai+1 . Prove that M1 M4 , M2 M5 , M3 M6 are concurrent.

Problem 2. Let P be the internal point of △DEF wrt. (AD)(BE)(CF ), and let Q be the isogonal conjugate
of P wrt. △DEF . Let D′ , E ′ , F ′ respectively be the reflections of D, E, F across BC, CA, AB. Prove that
−
△DEF ∪ Q ∼ △D′ E ′ F ′ ∪ Q.

9.2 Perfection of Isogonal Conjugation

Why do we consider the perfect six-point set? Let’s consider complete quadrilateral Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ). Let
△ABC = △ℓ1 ℓ2 ℓ3 , and let A∗ , B ∗ , C ∗ be the intersections of ℓ1 , ℓ2 , ℓ3 with ℓ4 such that AA∗ , BB ∗ , CC ∗ are
the three diagonals of Q.

255
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Our goal is to characterize the locus of points with an isogonal conjugate in Q. Define

K(Q) = {P | P has an isogonal conjugate in Q}

= {P | there exists a P ∗ such that for all triangles in Q, P × P ∗ = I × J }.

Since (AA∗ )(BB ∗ )(CC ∗ ) is a perfect six-point set, define a Clawson-Schmidt conjugation sending A ↔
A∗ , B ↔ B ∗ , C ↔ C ∗ . Call this ΦQ .

Proposition 9.2.1. If (P, P ∗ ) are isogonal conjugates in complete quadrilateral (BB ∗ )(CC ∗ ), then (BB ∗ )(CC ∗ )(P P ∗ )
is a perfect six-point set.

Proof. Let P ′ be a point such that (BB ∗ )(CC ∗ )(P P ′ ) is a perfect six-point set. Then note that since
(AA∗ )(BB ∗ )(CC ∗ ) is a perfect six-point set, we have (AA∗ )(BB ∗ )(CC ∗ )(P P ′ ) is a perfect eight-point set.
Since P has an isogonal conjugate, we have arg(P B) + arg(P B ∗ ) = arg(P C) + arg(P C ∗ ) (TODO 1.3.14).
From
BP ′ CB ∗ P C ∗
· · = −1,
P ′C B∗P C ∗B

we get

∡BP ′ C = arg(CB ∗ ) − arg(B ∗ P ) + arg(P C ∗ ) − arg(C ∗ B) = arg(CA) − arg(AB) + arg(P B) − arg(P C)

= arg(CA) − arg(AB) + (arg(AB) + arg(BC) − arg(BP ∗ )) − (arg(BC) + arg(CA) − arg(CP ∗ )) = ∡BP ∗ C

so do this symmetrically and we have that P ′ = P ∗ .

Proposition 9.2.2. If (P, P ∗ ) are isogonal conjugates in Q, then the midpoint of P P ∗ lies on the Newton
line τ .

Proof. Let C be a inscribed conic in Q, with P, P ∗ as foci (this must exist since they are isogonal conjugates).
Then P P ∗ ’s center is just the center of this conic. However by Newton II, (TODO 6.3.13) the center of C lies
on τ .

So we have two characteristics for isogonal conjugates (P, P ∗ ). I claim that in reality, these characteristics
are also sufficient for two points to be isogonal conjugates!

Proposition 9.2.3. Let (BB ∗ )(CC ∗ ) be a quadrilateral that’s not a parallelogram. If (BB ∗ )(CC ∗ )(P P ∗ )
is a perfect six-point set, and the midpoints of BB ∗ , CC ∗ , P P ∗ are collinear, then (P, P ∗ ) are isogonal
conjugates in (BB ∗ )(CC ∗ ).

256
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Proof. Since the midpoint of P P ∗ lies on the line through the midpoints of BB ∗ , CC ∗ (which is just the Newton
line), from (TODO 7.3.2) and (TODO 7.3.8), there exists two pairs of isogonal conjugates (Q, Q∗ ), (R, R∗ )
such that P P ∗ , QQ∗ , RR∗ have a common midpoint and that I = QR∗ ∩ Q∗ R, J = QR ∩ Q∗ R∗ . At this point
we know that (BB ∗ )(CC ∗ )(QQ∗ ), (BB ∗ )(CC ∗ )(RR∗ ) are all complete six-point sets. Thus

(BB ∗ )(CC ∗ )(P P ∗ )(QQ∗ )(RR∗ )

is a complete ten-point set. We proceed by contradiction. If P, P ∗ ̸= Q, Q∗ , R, R∗ , let M ∈

/ L∞ be the
Miquel point of Q, which is just the Miquel point of (P P ∗ )(QQ∗ )(RR∗ ). Since P Q ∩ P ∗ Q∗ ∈ L∞ ̸= I, J,
∡M P Q = ∡M Q∗ P ∗ tells us that M P ∥ M Q∗ , but since M ∈
/ L∞ , this means that M ∈ P Q∗ . Similarily we
also get that M ∈ P ∗ Q, however P Q∗ ∩ P ∗ Q ∈ L∞ , contradiction.

So we can say that for a quadrilateral Q, the locus of points with isogonal conjugates K is completely
determined by a given Clawson-Schmidt conjugation and a Newton line.

Theorem 9.2.4 (QL-Cu1, the Isoptic Cubic). Let τ be the Newton line, and let Φ be a Clawson-Schmidt
conjugation switching opposite diagonal vertices (points A, A∗ , etc). (Note that this is also just force-overlay
negative inversion at the Miquel point.) Then the part of K not on the line at infinity is given by

K(Q) \ L∞ = {P | midpoint of P Φ(P ) ∈ τ }

By bashing it out, we get that this is a degree-3 condition on the position of P (since it’s the midpoint of
a deg-1 point and a deg-2 point.)

Proposition 9.2.5. K is a cubic.

We call K the isoptic cubic of Q, since any point on K “sees” the two sides of the quadrilateral with the
same angle, by (TODO 1.3.14). This definition gets us that the isoptic cubic of a complete quadrilateral
has to go through its six vertices, the Miquel point, the point at infinity along the Newton line, and the two
circle points. Thus K is actually a circular cubic, a cubic that passes through the two circle points.

Let’s notate the isoptic cubic of quadrilateral Q as K(Q).

If ABDE is a parallelogram, we get something very nice.

Proposition 9.2.6. If (BB ∗ )(CC ∗ ) is a parallelogram, then the isoptic cubic is just a rectangular hyperbola
H through its vertices and the line at infinity.

Proof. It’s obvious that the line at infinity is in the isoptic cubic since it’s a parallelogram. So we only have
to prove P ∈ K\L∞ ⇐⇒ P ∈ H\L∞ , so we just have to prove that all points on this rectangular hyperbola

257
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

satisfy
∡BP C + ∡B ∗ P C ∗ = 0◦ .

+
Choose Q such that △BCQ ∼ △C ∗ B ∗ P, then CQ ∥ B ∗ P , and

∡BP C + ∡B ∗ P C ∗ = 0◦ ⇐⇒ ∡BP C + ∡CQB = 0◦

Q ∈ (P BC)

∡CB ∗ P = ∡P QC = ∡P BC.

Then from (TODO 8.1.10) we have that P lies on a rectangular hyperbola centered at the midpoint of this
parallelogram passing through B, B ∗ , C, C ∗ .

To sum it up, we get that if (P, P ∗ ) are isogonal conjugates in (BB ∗ )(CC ∗ ), then (B, B ∗ ) are isogonal
conjugates in (CC ∗ )(P P ∗ ), and (C, C ∗ ) are isogonal conjugates in (P P ∗ )(BB ∗ ). We call these six points
(P P ∗ )(BB ∗ )(CC ∗ ) a perfect isogonal six-point set.

Definition 9.2.7. For a complete quadrilateral Q, define K(Q) to be the locus of points with an isogonal
conjugate in Q. Define τ (Q) to be the Newton line of Q, and define ΦQ to be the Clawson-Schmidt conjugation
swapping opposite vertices of the complete quadrilateral.

The following results give us a lot of small useful lemmas about isogonal conjugates.

Proposition 9.2.8. If (P, P ∗ ), (Q, Q∗ ) are isogonal conjugates in complete quadrilateral Q, then K(Q) =
K((P P ∗ )(QQ∗ ).

Proof. This is obvious when the Miquel point of Q lies on the line at infinity, so assume it’s not on the line
at infinity. Since (BB ∗ )(CC ∗ )(P P ∗ ) and (BB ∗ )(CC ∗ )(QQ∗ ) are perfect six-point sets, (P, P ∗ ), (Q, Q∗ ) are
swapped by the Clawson-Schmidt conjugation that swaps opposite vertices of Q. Since the midpoints of
P P ∗ , QQ∗ lie on the Newton line, apply (TODO 9.2.4) and win.

In other words, we can consider the perfect 2n point set (X1 Y1 )(X2 Y2 ) . . . (Xn Yn ) such that all (Xi , Yi )
is a pair of isogonal conjugates in (X1 Y1 )(X2 Y2 ), then for any i, j, k, (Xk , Yk ) are isogonal conjugates in
(Xi Yi )(Xj Yj )! Thus the roles of everything in this is symmetric and we call it a perfect 2n-point set.

Corollary 9.2.9. Let (P, P ∗ ), (Q, Q∗ ) be a pair of isogonal conjugates in △ABC. If (R, R∗ ) are isogonal
conjugates in (P P ∗ )(QQ∗ ), then (R, R∗ ) are also isogonal conjugates in △ABC.

Proof. Since (P, P ∗ ), (Q, Q∗ ) are isogonal conjugates wrt. △ABC, there exists two inconics with foci at
P, P ∗ and Q, Q∗ (call them CP , CQ ). These conics have three common tangent lines as the three sides of the

258
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

triangle, let ℓ be their fourth common tangent line. Then (P, P ∗ ), (Q, Q∗ ) are isogonal conjugates in complete
quadrilateral △ABC ∩ ℓ, so by the previous lemma we get that (R, R∗ ) are isogonal conjugates in △ABC ∩ ℓ,
so they are also isogonal conjugates in △ABC.

Remark. Taking the “fourth common tangent” is a very common trick to turn a problem about triangles to
a problem about complete quadrilaterals.

Note that this gives a fast proof of “Isoconjugations lead to more isoconjugations” (TODO 7.4.5) in the
case of isogonal conjugation. By setting (R, R∗ ) = (P Q∗ ∩ P ∗ Q, P Q ∩ P ∗ Q∗ ), we get

Corollary 9.2.10. Let (P, P ∗ ), (Q, Q∗ ) be isogonal conjugates in △ABC, then (P Q∗ ∩ P ∗ Q, P Q ∩ P ∗ Q∗ )

are also isogonal conjugates in △ABC.

Corollary 9.2.11. Let (P, P ∗ ), (Q, Q∗ ) be isogonal conjugates in △ABC. Then the Miquel point of the
complete quadrilateral Q = (P P ∗ )(QQ∗ ) lies on (ABC), and the isogonal conjugate of the Miquel point M
on △ABC is the line at infinity along the Newton line ∞τ (Q) .

Proof. In (TODO 9.2.9) just set (R, R∗ ) to be (M, ∞τ (Q) ). Since ∞τ (Q) ∈ L∞ , M ∈ (ABC).

There’s actually a really elementary proof of the fact that the Miquel point M lies on (ABC). Let X be
+ + +
a point such that M AX ∼ M P Q ∼ M Q∗ P ∗ . Then △AP Q∗ ∼ △XQP ∗ tells us that

∡QXP ∗ = ∡P AQ∗ = ∡QAP ∗ ,

so A, P ∗ , Q, X are concyclic. Similarily.

∡AX = ∡AQ + ∡(P ∗ X) − ∡(P ∗ Q).

+ +
In the same logic, choose Y such that △M BY ∼ △M P Q ∼ △M Q∗ P ∗ , we have

∡BY = ∡BQ + ∡(P ∗ Y ) − ∡(P ∗ Q).

+
From △ABQ∗ ∼ △XY P ∗ ,

∡BY − ∡AX = ∡(P ∗ Y ) − ∡(P ∗ X) + ∡AQB = ∡AQ∗ B + ∡AQB = ∡ACB,

+
so AX ∩ BY ∈ (ABC), but since △M AX ∼ △M BY this implies M ∈ (ABC).

Corollary 9.2.12. Let (P, P ∗ ) be isogonal conjugates in △ABC, and let I either be the incenter or one of
+
the excenters. Then there exists a point M ∈ (ABC) such that △M P I ∼ △M IP ∗ . Also, if we let X be the
midpoint of P P ∗ , then the isogonal conjugate of M is ∞IX .

259
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic
+
Proof. Let M be the Miquel point of (P P ∗ )(II), then △M P I ∼ △M IP ∗ . Since the Newton line is IX, the
isogonal conjugate of M is ∞IX .

Do you still remember the cyclic perfect eight-point set (TODO 9.1.8)? Let’s see if we can define nice
isogonal conjugates on it.

Proposition 9.2.13. Let (P, P ∗ ) be isogonal conjugates in (BB ∗ )(CC ∗ ), and let

Q = (P BC) ∩ (P B ∗ C ∗ ) ∩ (P ∗ BC ∗ ) ∩ (P ∗ B ∗ C), Q = (P ∗ B ∗ C ∗ ) ∩ (P ∗ BC) ∩ (P B ∗ C) ∩ (P BC ∗ ),

then (Q, Q∗ ) are isogonal conjugates in (BB ∗ )(CC ∗ ).

Proof. From (TODO 9.1.8), we have that (BB ∗ )(CC ∗ )(P P ∗ )(QQ∗ ) is a perfect eight-point set. So we just
need to prove that the midpoint of QQ∗ lies on the Newton line of (BB ∗ )(CC ∗ ). Note that the midpoints of
BB ∗ , CC ∗ , P P ∗ , QQ∗ are cyclic, but the midpoints of BB ∗ , CC ∗ , P P ∗ are collinear on the Newton line, so
the midpoint of QQ∗ also lies on the Newton line.

Example 9.2.14 (IMO 2018/6). Let ABCD be a convex quadrilateral such that AB · CD = BC · DA, and
let X be inside ABCD such that

∠XAB = ∠XCD, ∠XBC = ∠XDA.

Prove that ∠BXA + ∠DXC = 180◦ .

Proof. This problem is equivalent to proving that X lies on the isoptic cubic of (AC)(BD). Let P, Q be
points such that
P ∈ (AXB) ∩ (CXD), Q ∈ (BXC) ∩ (DXA),

and choose Y such that (AC)(BD)(XY )(P Q) is a cyclic perfect eight-point set. So now we just need to
prove that (P, Q) are isogonal conjugates in (AC)(BD). From

∡XP B = ∡XAB = ∡XCD = ∡XP D,

we know that B, P, D are collinear, and similarily A, Q, C are collinear. If we now choose two points P ′ , Q′
on BD, AC such that P ′ , Q′ are isogonal conjugates in (AC)(BD), then we have

X ′ = (AP B) ∩ (BQC) ∩ (CP D) ∩ (DQA)

which is just the original definition of X.

260
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

[yappy trigbash kys] From (TODO 9.2.13), we have

Proposition 9.2.15. Let (P, P ∗ ) be isogonal conjugates in complete quadrilateral (BB ∗ )(CC ∗ ). Let
O, OP , OB , OC respectively be the circumcenters of

△P BC, △P B ∗ C ∗ , △P ∗ BC ∗ , △P ∗ B ∗ C

and let O∗ , OP∗ , OB

∗ ∗
, OC respectively be the circumcenters of

△P ∗ B ∗ C ∗ , △P ∗ BC, △P B ∗ C, △P BC ∗

then (OO∗ )(OP OP∗ )(OB OB

∗ ∗
)(OC OC ) is a cyclic perfect eight-point set, with common Miquel point with
(BB ∗ )(CC ∗ ), and the Newton lines of the eight-point set and the quadrilateral are perpendicular.

Proof. Let (P BC), (P B ∗ C ∗ ), (P ∗ BC ∗ ), (P ∗ B ∗ C) concur at Q, and let (P ∗ B ∗ C ∗ ), (P ∗ BC), (P B ∗ C), (P BC ∗ )

concur at Q∗ . Then by a previous lemma (Q, Q∗ ) are isogonal conjugates in (BB ∗ )(CC ∗ ). First note that

∡QBO = 90◦ − ∡BP Q = 90◦ − ∡Q∗ P B ∗ = ∡B ∗ Q∗ OB

∗
,

+
∗ ∗ ∗ +
so △OBQ ∼ △OB Q B . From △M BQ ∼ △M Q∗ B ∗ we know that

∗ + +
△M OOB ∼ △M BQ∗ ∼ △M QB ∗ .

Similarily, we have
+ +
△M O∗ OB ∼ △M B ∗ Q ∼ △M BQ∗ ,

so the Miquel point of (OO∗ )(OB OB

∗
) is M . Do this symmetrically and we get that (OO∗ )(OP OP∗ )(OB OB
∗ ∗
)(OC OC )
is a perfect eight-point set, with Miquel point M. Since

∡(OB O) + ∡(OB O∗ ) = ∡(⊥ BQ) + ∡(⊥ C ∗ P ∗ ) = ∡(⊥ BP ∗ ) + ∡(⊥ C ∗ Q) = ∡(OB OP∗ ) + ∡(OB OP ),

we have that OB has an isogonal conjugate wrt. (OO∗ )(OP OP∗ ), and since (OO∗ )(OP OP∗ )(OB OB
∗
) is a perfect
∗
six-point set, we get that its isogonal conjugate is forced to be OB . Do this on OP , OC and we get that
(OO∗ )(OP OP∗ )(OB OB
∗ ∗
)(OC OC ) is actually a perfect isogonal eight-point set. By some more angle-chasing,
from

∡(OB O) + ∡(OC OP ) = ∡(⊥ BQ) + ∡(⊥ B ∗ Q) = ∡(⊥ CQ) + ∡(⊥ C ∗ Q) = ∡(OC O) + ∡(OB OP )

we get that O, OP , OB , OC are concyclic. By symmetry we get that (OO∗ )(OP OP∗ )(OB OB
∗ ∗
)(OC OC ) is also a

261
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

cyclic perfect isogonal eight-point set.

∗ + +
Finally, to prove the Newton lines τ, τO are perpendicular, from △M OOB ∼ △M BQ∗ ∼ △M QB ∗ we
have
τ = ∡BQ + ∡(BQ∗ ) − ∡BM = ∡(⊥ OOB ) + ∡(OOB
∗
) − ∡OM = τO + 90◦ ,

so τ ⊥ τO .

[I don’t know what this is saying]

The following section is not very related to perfect six-point sets, but it’s interesting.

Proposition 9.2.16. Let (P, P ∗ ) be isogonal conjugates in (BB ∗ )(CC ∗ ), then the orthocenters H, HP , HB , HC
of △P BC, △P B ∗ C ∗ , △P ∗ BC ∗ , △P ∗ B ∗ C are collinear and are perpendicular to the Newton line τ .

Proof. By symmetry, we only need to prove that HHB is perpendicular to τ . We consider the transfor-
mation C ∗ → P, and let this send P ∗ → X, B → Y . Let C ′ , X ′ respectively be the antipodes of C, X in
(P BC), (XY P ). Then note that both of (BP )(C ′ H), (BP )(X ′ HB ) are parallelograms, so C ′ X ′ ∥ HHB .
From
∡CC ′ P = ∡CBP = ∡P ∗ BC ∗ = ∡XY P = ∡XX ′ P

+ +
and ∡C ′ P C = ∡X ′ P X = 90◦ we can get △C ′ CP ∼ △X ′ XP , or in other words △CXP ∼ △C ′ X ′ P , so
CX ⊥ C ′ X ′ . Since the midpoints of CC ∗ , XC ∗ both lie on τ , we have HHB ∥ C ′ X ′ ⊥ τ .

At this point we can consider the four other triangles △P ∗ B ∗ C ∗ , △P ∗ BC, △P B ∗ C, △P BC ∗ , to get four
more points H ∗ , HP∗ , HB
∗
, HC∗ are collinear. You might wonder when these eight orthocenters are all collinear.

Proposition 9.2.17. Let (P, P ∗ ) be isogonal conjugates in (BB ∗ )(CC ∗ ), then the orthocenters of

△P BC, △P B ∗ C ∗ , △P ∗ BC ∗ , △P ∗ BC ∗ , △P ∗ B ∗ C, △P ∗ B ∗ C ∗ , △P ∗ BC, △P B ∗ C, △P BC ∗

are all collinear on line ℓ if and only if P P ∗ , BB ∗ , CC ∗ are concurrent at a point X. Further, X ∈ ℓ.

We first prove this lemma:

Lemma 9.2.18. Let P, P ∗ be isogonal conjugates in △ABC, and let H, HP∗ be the orthocenters of
△P BC, △P ∗ BC. Let the inellipse with foci at P, P ∗ touch CA, AB at E, F . Then P P ∗ , EF, HHP∗
are collinear.

Proof. Let X be the intersection point of P P ∗ and HHP∗ . Then from P H ∥ P ∗ HP∗ and P, P ∗ isogonal

262
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

conjugates wrt. (CF )(AB), we have

PX PH BC cot ∡BP C cot ∡AP F

∗
= ∗ ∗ = = .
P X P HP ∗
BC cot ∡BP C cot ∡AP ∗ F

Note that A is the the center of a circle tangent to EF, P E, P ∗ F . Let S, T be the tangency points on
P E, P ∗ F , and let U be the intersection point of P E, P ∗ F . By Newton III (TODO 6.3.14) (Brianchion),
P P ∗ , EF, ST are concurrent. Finally, from

P X P ∗T U S cot ∡AP F tan ∡P ∗ AT tan ∡U AS

∗
· · =− · · = −1
XP T U SP cot ∡AP ∗ F tan ∡T AU tan ∡SAP

by Menelaus we have that X, T, S are collinear.

Back to the original:

Proof. From (TODO 9.2.16), the eight orthocenters are collinear if and only if the orthocenter H of △P BC
and the orthocenter HP∗ of △P ∗ BC lie on a line ℓ perpendicular to the Newton line τ . Thus, when we fix
B, C, P, P ∗ (then define B ∗ , C ∗ as the envelope of conics with foci P, P ∗ that are tangent to BC), B ∗ , C ∗
under HHP∗ ⊥ τ is unique (unless ABDE is a parallelogram, however then it’s trivial). So we just need to
prove the ”if” side of the argument, and

♠Given fixed B, C, P, P ∗ , there exists B ∗ , C ∗ such that (P P ∗ )(BB ∗ )(CC ∗ ) form a perfect isogonalsix-point set and P P ∗ , B

Let’s just construct ♠. Let C be a conic with foci P, P ∗ and tangent to BC, let TB , TC be the tangents
to C from B, C that aren’t BC, and let these touch C at U, V respectively. Let X = P P ∗ ∩ U V, B ∗ =
BX ∩ Tc , C ∗ = CX ∩ TB . By Newton III (TODO 6.3.14), B ∗ C ∗ is tangent to C, so (P P ∗ )(BB ∗ )(CC ∗ ) is
a perfect isogonal six-point set and P P ∗ , BB ∗ , CC ∗ concur at X. Now for the ”if” side. We prove that
H, HP∗ , X are collinear, and the rest follows similarily. Similarily, by Newton III (TODO 6.3.14), if U, V
respectively are the tangency points of BC ∗ , CB ∗ with C, then P P ∗ , BB ∗ , CC ∗ , U V concur at X. Then by
(TODO 9.2.18), P P ∗ , U V, HHP∗ are concurrent.

9.2.1 Taming the Isoptic Dragon

Let’s return to looking at the isoptic cubic K. Following precedent set in 6.A, let’s replace the circle points
I, J with two random points W = {W+ , W− }, then we can define the W-isoconjugate cubic as:

KW = (Q) = WW+ ×W− (Q) = {P | ∃P ∗ such that on all △ ⊆ Q, P × P ∗ = W+ × W− }.

263
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

We can convert all of the above results to W-versions. Here we can define a W-Miquel point, MW , as
the common intersection point of the four conics through some 3 points of the quadrilateral, W+ , and W− .
Similarily we can define a W-Newton line τW as the QL-Tf2 conjugate (defined as the line through the
harmonic conjugates of the three intersections of line L wrt. the sides of the diagonal triangle) (for more
elaboration see chapter 10) of the W-line at infinity. Finally define W-Clawson-Schmidt conjugation as the
point isoconjugation on △MW W+ W− that switches the vertices of Q.

Dually, we define a W-line conjugate cubic on complete quadrangle ⨿ as

∥W (⨿) = ∥w+ ×w− (⨿) = {ℓ | ∃ℓ∗ such that on all △ ⊆ ⨿, ℓ × ℓ∗ = w+ × w− }.

Since we only need to know two pairs of isogonal conjugates to determine a isoptic cubic K, let’s just
throw away the whole quadrilateral Q and see what we can do. Note that we can still find a Newton line and
Miquel point, a Clawson-Schmidt conjugation, more pairs of isogonal conjugates, etc.

Definition 9.2.19. A cubic curve K is an isoptic cubic if there exists a complete quadrilateral Q such that
K = K(Q).

The first key topic we will address is inversion on the isoptic cubic:

Proposition 9.2.20. Let A be a point on the isoptic cubic K. Then the inversion of K across any point A
(call it KJ ) is still some isoptic cubic. Furthermore, any two isogonal conjugates before the inversion are still
isogonal conjugates (on the new cubic, call it J (K)) after the inversion.

Proof. Let (B, B ∗ ) be a pair of isogonal conjugates on K, let C be the intersection of AB ∗ and A∗ B, and
let C ∗ be the intersection of AB and A∗ B ∗ . Then by DDIT we have that (C, C ∗ ) are isogonal conjugates.
Further, we have

P ∈ K ⇐⇒ P B + P B ∗ = P C + P C ∗

= (AP + AB − P J B J ) + (AP + AB ∗ − P J (B ∗ )J )

= (AP + AC − P J C J ) + (AP + AC ∗ − P J (C ∗ )J

⇐⇒ P J B J + P J (B ∗ )J = P J C J + P J (C ∗ )J

⇐⇒ P J ∈ K((B J (B ∗ )J )(C J (C ∗ )J )).

Thus J (K) is also the isoptic cubic of (B J (B ∗ )J )(C J (C ∗ )J ). Another note: If P ∗ is the isogonal con-
jugate of P on the isoptic cubic, then (P P ∗ )(BB ∗ )(CC ∗ ) being a perfect six-point set tells us that
(P J (P ∗ )J )(B J (B ∗ )J )(C J (C ∗ )J ) is also a perfect six-point set.

264
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

In reality we can actually also define polarity on a cubic! This is slightly more involved in algebraic
geometry and we explore it more in chapter 11. There are two ways to define polarity on a cubic, one
involving a line and one involving a (polo)conic. One day, Li4 was bored and randomly decided to take the
poloconic of the Miquel point wrt. the isoptic cubic, and found this theorem.

Proposition 9.2.21. Let M be the Miquel point of the isoptic cubic K. Let line ℓ be a moving line through
M that intersects the cubic at R, S. Let Mℓ∨ be the harmonic conjugate of M in segment R, S. Then the
locus of Mℓ∨ is a circle.

Proof. Let R∗ , S ∗ respectively be the isogonal conjugates of R, S in K. Then ΦK (Mℓ∨ ) is the midpoint of
R∗ S ∗ . Since RS and R∗ S ∗ intersect at M , we have by DDIT that RS ∗ ∩ R∗ S is the point at infinity △⊔⊣⊓(K) ,
which means RS ∗ ∥ R∗ S ∥ τ = τ (K). However the midpoints of RR∗ and SS ∗ both lie on τ, so the midpoint
of R∗ S ∗ also lies on τ, so the locus of M ∨ is ΦK (τ ).

The poloconic of ∞τ is a rectangular hyperbola. This motivates the following theorem:

Theorem 9.2.22 (Telv Cohl). Let M be the Miquel point of complete quadrilateral K, let a line through M
intersect the cubic again at points R, S. Then the circles with diameter RS are coaxal.

Proof. Keeping the previous notation, we proved the midpoint (i.e the center of the circle) of RS will always
lie on τ , so all we need to prove is that there is a fixed point with constant power wrt. all of these circles.
Suppose the Miquel point lies on the Newton line; then all of these circles are concentric at M and have
common radical axis as the line at infinity. Now in the normal case, we can actually just explicitly construct
this point. Let O be the center of the Clawson-Schmidt conjugate of the Newton line Φ(τ ). Since M and M ∨
both lie on Φ(τ ), (RS) is orthogonal to it. (Alternatively, just note that the C-S conjugate of (RS) is a circle
with the center on the Newton line). So this implies the power of O wrt. (RS) is just the radius of Φ(τ ),
which is a circle, so we’re done.

Obviously this means that the common radical axis is just the perpendicular from O to the Newton line.
We call this axis of K the Telvcohl-axis of K (blame Li4 for the name, he made it up).

If all of these circles all intersect on two common points on the Telvcohl-axis (of course, on some K these
points are complex points, in general the two points are real for all K made out of two real parts), then these
two points have a property that you can probably guess:

Proposition 9.2.23. Continuing the notation of (TODO 9.2.22), let U, V be the common points of all (RS).
Then U, V are a pair of isogonal conjugates in Q.

265
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Proof. Let R∗ , S ∗ respectively be the isogonal conjugates of R, S in K. Then

∡RU S + ∡R∗ U S ∗ = 90◦ + 90◦ = 0◦ ,

so U ∈ K, and similarly V ∈ K. From

∡RU S + ∡RV S = 90◦ + 90◦ = 0◦ = ∡R∞τ S∡R∗ U S ∗ + R∗ V S ∗ = 90◦ + 90◦ = 0◦ = ∡R∗ ∞τ S ∗

we have that (RS)(R∗ S ∗ )(U V ) is a perfect six-point set, and (U, V ) are isogonal conjugates in Q.

Proposition 9.2.24. Let M, L respectively be the Miquel point and Telvcohl axis, then M ’s foot on the
Telvcohl axis V lies on the isoptic cubic K.

Proof. Let M ∞τ (K) intersect K again at V ′ , then we know that the circle with diameter V ′ ∞τ (K) is orthogonal
to all the other circles with diameter (RS), however the circle (V ′ ∞τ (K) ) is just a line through V ′ perpendicular
to the Newton line, and since the Telvcohl axis is perpendicular to the Newton line, V = V ′ ∈ K.

The point V is very important and will be revisited in Chapter 11: remember it as the vanishing point of
K. Currently we’ll just focus on one property of it.

Proposition 9.2.25. Let V be the vanishing point of isoptic cubic K, let Y be the foot from V to the Newton
line τ of K, then for any P, Q ∈ K, (P, Q) is a pair of isogonal conjugates if and only if the orthocenters of
△P V Y, △QV Y are reflections across τ .

Practice Problems

Problem 1. Let I be the incenter of triangle △ABC, let HA be the foot from A to BC, let A∗ be the
antipode of A in (ABC). Prove that ∡BIHa = ∡A∗ IC.

Problem 2 (Taiwan 2018 3J M3). Let I be the incenter of triangle ABC, and ℓ be the perpendicular bisector
of AI. Suppose that P is on the circumcircle of triangle ABC, and line AP and ℓ intersect at point Q. Point
R is on ℓ such that ∠IP R = 90◦ .Suppose that line IQ and the midsegment of ABC that is parallel to BC
intersect at M . Show that ∠AM R = 90◦ .

Problem 3 (Taiwan 2019 2J M6). Let ABC be a triangle with incenter I and A-excenter J. Let AA′ be
the diameter of the circumcircle of △ABC, and let H1 and H2 be the orthocenters of △BIA′ and △CJA′ ,
respectively.

Prove that the line H1 H2 is parallel to the line BC.

266
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Problem 4 (Tau). Let I, O, N be the incenter, circumcenter, and nine-point center. Prove that if IN ∥ BC,
then ∡AIO = 135◦ .

Problem 5 (Taiwan 2023 IJ I2-G). In isosceles trapezoid ABCD with AD parallel to BC, let Ω be its
circumcircle. Let X be the reflection of D across BC, let Q be a point on arc BC of Ω, let P be the
intersection of DQ and BC. Point E satisfies EQ parallel to P X and EQ bisects angle ∠BEC. Prove EQ
also bisects ∠AEP .

Problem 6. Let I be the incenter of △ABC, let ω be its incircle. Point P lies on ω such that its isogonal
conjugate Q lies outside of ω. Let △QEF be the triangle formed by Q′ s two tangents to ω and P ′ s one
tangent to ω. Prove that △QEF ’s Q-mixtilinear touchpoint S lies on (ABC).

9.3 Oblique Strophoids

When a complete quadrilateral has an incircle (or equivalently, an excircle), we typically have some nice
properties. Perfect six-point sets are no exception; let’s see what cool things we can have. First, since (I, I)
are a pair of isogonal conjugates in Q, we have

Proposition 9.3.1. I is a fixed point of Clawson-Schmidt conjugation.

In other words, quadrilateral Q’s Newton line has fixed point I. Through the theory of perfect six-point
sets, we can easily prove (TODO 4.3.5) which we previously spent too much effort on.

Proposition 9.3.2. Let M be the Miquel point of tangential quadrilateral (BB ∗ )(CC ∗ ), let J be the
reflection of the incenter I across M , then (IJ)(BB ∗ ), (IJ)(CC ∗ ) are all harmonic quadrilaterals.

Proof. I, J are both fixed points of Clawson-Schmidt conjugation, so from (TODO 9.1.4) we know that
(IJ)(BB ∗ ), (IJ)(CC ∗ ) are all harmonic quadrilaterals.

Proposition 9.3.3. The isoptic cubic K of a tangential quadrilateral is special: it’s a nodal cubic (self-
intersecting). Furthermore, it’s also the image of a rectangular hyperbola that was inverted around the
incircle ω. We call K a oblique strophoid, and we call I the node of this strophoid.

Proof. Let (BB ∗ )(CC ∗ ) be our tangential quadrilateral. Let J represent inversion around ω, then (B J (B ∗ )J )(C J (C ∗ )J )
is a parallelogram, so I lies in K and the isoptic cubic is the union of a rectangular hyperbola and the line at
infinity, by (TODO 9.2.20) and (TODO 9.2.6). Thus

K = (H ∩ L∞ )J = HJ ∩ I = H.

267
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Remark. Sometimes we call the Miquel point of the tangential quadrilateral the focus of the isoptic
cubic/strophoid. This is because of an alternative definition of a strophoid as the locus of points A, B given
a focus F and a node (double-point) O such that the midpoint M of A, B lies on a fixed line through O,
line A, B goes through a fixed focus F , and AM = BM = M O. (This condition is interpreted much more
naturally with a group law.)

Corollary 9.3.4. Keeping the previous notation, for a pair of isogonal conjugates (P, P ∗ ) in K, when inverted
about I, they will be antipodes on the rectangular hyperbola.

Proof. Let M be the Miquel point of K, and fix two pairs of isogonal conjugates (B, B ∗ ), (C, C ∗ ), let J be
the reflection of M across I. Then from (TODO 9.3.2), we get that (IJ)(P P ∗ ), (IJ)(BB ∗ ), (IJ)(CC ∗ ) are
harmonic quadrilaterals, so since inversion preserves cross ratios, J ∗ is the common midpoint of segments
P J (P ∗ )J , B J (B ∗ )J , C J (C ∗ )J . This is just the center of the rectangular hyperbola, so P J , (P ∗ )J are antipodes.

Corollary 9.3.5. Let (P, P ∗ ) be a pair of isogonal conjugates in △ABC, let ω be the incircle of △ABC,
and let J represent inversion about ω. Then AJ , B J , C J , P J , (P ∗ )J , I lie on a common rectangular hyperbola.

Proof. We prove that A, B, C, P, P ∗ lie on a common strophoid. Let C be a conic with P, P ∗ as foci that’s
tangent to △ABC, let ℓ be the fourth common tangent of C and ω. Then (P, P ∗ ), (I, I) are a pair of
isogonal conjugates in Q = △ABG ∪ ℓ, so the isoptic cubic of this quadrilateral K(Q) is a strophoid through
A, B, C, P, P ∗ , I with I as its node. Therefore by inverting across ω we have that AJ , B J , C J , P J , (P ∗ )J , I lie
on a common rectangular hyperbola.

By properties of rectangular hyperbolas, we can get the following result:

Corollary 9.3.6. Let M be the focus of strophoid K, let ℓ be a line through M that intersects the strophoid
again at points R, S. Then ∡RIS = 90◦ .

Proof. Obviously, the points RJ , S J on H satisfy I, M J , RJ , S J concyclic. But since I, M J are antipodes on
H, from (TODO 8.1.10) we have that RJ S J is the diameter of circle (IM J RJ S J ), so ∡RIS = 90◦ .

Corollary 9.3.7. If a pair of isogonal conjugates (P, P ∗ ) has line P P ∗ intersect K again at VP , then
IVP ⊥ P P ∗ .

Proof. Obviously the three points P J , (P ∗ )J , VPJ on H satisfy (I, P J , (P ∗ )J , VPJ ) are concyclic, and P J , (P ∗ )J
are a pair of antipodes on H, so IVPJ is a diameter of circle (IP J (P ∗ )J VPJ ), equivalent to IVP ⊥ P P ∗ .

Remember the Telvcohl-axis? (TODO 9.2.22), this is just really setting (U, V ) to (I, I). So basically, in
the strophoid case:

268
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Proposition 9.3.8. Let M, τ respectively be the Miquel point and Newton line of strophoid K. Let line
ℓ through M intersect K again at points R, S, then the circle with diameters RS are all coaxal on the
perpendicular line from I to τ , which is also the tangent from I to circle (RS).

Proof. From (TODO 9.2.22) all that we really need to prove is that this perpendicular is tangent to (RS).
Since by the definition of a strophoid, the midpoint of RS lies on the Newton line τ , we have N I ∥ τ , so it
has to be tangent.

Let T be the vanishing point of strophoid K, then it’s just the foot from I to M ∞τ . Since the isoptic
cubic is a circular cubic, and thus inverts to itself (over some point and some radius on the cubic), this is
also true about strophoids. Specifically, you just invert about the focus with radius from focus to node, then
reflect across the line connecting the focus and node. (equivalent of Clawson-Schmidt)

Proposition 9.3.9. Let A be a point on strophoid K, then the inverse of K across A is still some strophoid.

Proof. Trivial.

Since (I, I) are always a pair of isogonal conjugates for all 4 component triangles of a complete tangential
2
quadrilateral, (using the notation in subsection (TODO 9.3.3)), K(Q) is just KI (Q). Let’s generalize this to
arbitrary circle points as well.
2
Theorem 9.3.10. Complete quadrilateral Q’s W 2 -isoconjugate cubic KW (Q) is the image of (BB ∗ CC ∗ W )
under an isoconjugation on △AA∗ W that swaps B × B ∗ = C × C ∗ .

Proof. Let Q = B × B ∗ ÷ P = C × C ∗ ÷ P. Then from (TODO 7.4.6), we have

(B(W, Q; C, C ∗ ) = B(W, BQ ∩ B ∗ P ; A∗ , A) = P (W, B ∗ ; A∗ , A),

B ∗ (W, Q; C, C ∗ ) = B ∗ (W, B ∗ Q ∩ BP ; A, A∗ ) = P (W, B, A, A∗

Thus Q lies on conic (BB ∗ CC ∗ W ) if and only if P (W, W ), P (A, A∗ ), P (B, B ∗ ) define a pencil involution on
2
TP , so by the (tangent line version of (TODO 1.3.14)) this is equivalent to P ∈ KW (Q).

Practice Problems

Problem 1. Let acute triangle △ABC have circumcenter O, and incenter I. Let D, E, F respectively be
the midpoints of arcs BC, CA, AB, let S be the intersection of BO and F D, let T be the intersection of CO
and DE, let H be the orthocenter of OEF . Prove that OI, HD, ST are concurrent.

269
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Problem 2. Let strophoid K have node I, prove that the envelope of the perpendicular to the moving line
P I at I for P on the strophoid is a conic.

Problem 3. Let tangential quadrilateral ABCD have incircle ω, let AB and CD intersect at E, let AD and
BC intersect at F , let AC and BD intersect at T , let (AT P ) and (CT D) intersect again at P , let (AT D)
and (BT C) intersect again at Q. Prove that if E, F, P, Q are concyclic, then I also lies on this circle.

9.4 Duality - The Isohaptic Locus of Isotomic Conjugation

We know that the dual of isogonal conjugate points are isotomic conjugate lines (see (TODO 1.5)), so let’s
see if we can also define the dual of the isoptic cubic in a similar manner.

Given complete quadrangle ⨿ = (A, B, C, D), we define the isohaptic curve of ⨿ as the envelope of

∥(⨿) = {ℓ | ℓ has a line isotomic conjugate in ⨿}.

Note that L∞ ∈ ∥(⨿) since it is invariant under line isotomic conjugation. So (by the notation back in
∈
(TODO 9.2.1)), we represent it as ∥L∞ (q), therefore we can define it as the projective dual of a strophoid. By
Plücker’s formulas, we obtain that the isohaptic curve is a quartic with three cusps (projectively equivalent
to a deltoid and cardioid, which we will explore in the next chapter). For easier discussion, let’s just consider
the affine part of the isohaptic curve
∥◦ (⨿) = ∥(⨿) \ {L∞ }

This lets us return to (TODO 1.5.5):

Proposition 9.4.1. For a line ℓ ̸= L∞ , let PXY be XY ∩ ℓ. Then ℓ has an isotomic line conjugate wrt. ⨿ if
and only if the midpoints of PBC PAD , PCA PBD , PAB PCD coincide.

For all lines ℓ with isotomic conjugates, let Mℓ be this point. By DIT, this gives a projective involution
on quadrangle ⨿ for any line ℓ that involves reflecting across Mℓ .

Proposition 9.4.2. Mℓ lies on the nine-point conic C⨿ (TODO 6.3.8) of ⨿. The other direction also holds;
for any M on the nine-point conic, there exists a unique line ℓ ∈ ∥◦ (q) such that M = Mℓ .

Proof. Consider the cevian triangle △XA XB XC of ⨿, and consider the point isoconjugation ϕ on it such
that the vertices of ⨿ are fixed (dual of QL-Tf2). By (TODO 7.4.17), C⨿ = ϕ(L∞ ). For any circumconic C
of ⨿, (TODO 7.2.20) tells us that Mℓ , ∞ℓ are conjugate points wrt. this circumconic, so ϕ(Mℓ ) = ∞ℓ , thus
Mℓ ∈ ϕ(L∞ ) = Cq .

270
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Now for the other direction. Let M be a point on Cq , choose line ℓ such that M is the midpoint of
PBC = BC ∩ ℓ and PAD = AD ∩ ℓ (in other words, choose ℓ = M U where U is a point at infinity such that
(∞B C, ∞A D; ∞XA M , U ) = −1). By taking the polar dual of the DIT-given involution (TODO 7.2.10) across
C⨿ , M is the fixed point of the DIT-given involution defined by ℓ cutting ⨿. Thus M is also the midpoint of
PCA PBD , PAB PCD , so M = Mℓ .

Example 9.4.3. If ⨿ is an orthocentric system (example: D is the orthocenter of △ABC), then the
nine-point conic of △ABC is just the nine-point circle. For a point M on the nine-point circle, let P be the
reflection of D, the orthocenter, across M . Then M ’s corresponding ℓ is just P ’s Simson line wrt. △ABC.

Since M simultaneously is the midpoint of AD and PBC PAD , P PBC ∥ PAD D = AD, so PBC is the
intersection of the line through P parallel to AD and line BC, which is just the foot from P to BC. Thus
symmetrically we have that P ’s Simson line PBC PCA PAB = ℓ.

So in the case of an orthocentric system, the isohaptic curve is just the envelope formed by all of the
Simson lines of △ABC. Since it’s an orthocentric system, we also have that it’s the envelope formed by all
the Simson lines of △BCD, etc. We see in (TODO 10.2) that this is actually a deltoid, so we get that the
dual of a strophoid is actually equivalent to a deltoid (after some projective transformation).

Through this method, we can easily construct the isohaptic curve of any quadrangle, including non-
tangential ones (by drawing parallel lines).

Let ℓ∗ be the isotomic line conjugate of line ℓ wrt. ⨿. Define Mℓ∗ analogously.

Proposition 9.4.4. Mℓ , Mℓ∗ are antipodes on the nine-point conic C⨿ .

Proof. Since we already know that Mℓ , Mℓ∗ already lie on C⨿ , we only need to prove that the midpoint of
segment Mℓ Mℓ∗ is the center of C⨿ . By definition,

1 1 1
(Mℓ + Mℓ∗ ) = (PBC + PAD ) + (PBC∗ + PAD∗ )
2 4 4
1 1 1
= (PBC + PBC∗ ) + (PAD + PAD∗ = (MBC + MAD ),
4 4 2

where MBC , MAD are the midpoints of BC and AD, so the midpoint of MBC MAD is just the midpoint of
Mℓ Mℓ∗ , which is also the center of C⨿ (the centroid of ⨿).

In the example of an orthocentric system in (TODO 9.4.3), ℓ, ℓ∗ are just the Simson lines of antipodes on
(ABC) P, P ∗ , so we have (by a projective transformation)

Proposition 9.4.5. ℓ, ℓ∗ intersect on the nine-point conic of C⨿ .

Proof. a bunch of projective shit Take a homography and win.

271
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic

Proposition 9.4.6. A, B, C, D, ∞ℓ , ∞ℓ∗ are conconic.

Proof. By DIT, we only need to prove that (∞ℓ , ∞ℓ∗ ) is also a pair of conjugate points defined by the DIT
involution ϕ on L∞ and ⨿.

Let Oℓ ∈ Cq be the intersection of ℓ, ℓ∗ , then (∞ℓ , ∞ℓ∗ ), (∞CA , ∞BD ), (∞AB , ∞CD ) are pairs under ϕ if
and only if
Mℓ Mℓ∗ , UCA UBD , UAB UCD

concur, where UXY is the second intersection of Oℓ ∞XY with Cq .

So we only need to prove that UCA UBD , UAB UCD are both diameters of Cq (so that these three lines
concur at the center Gq of C⨿ ). Note that

∞
CA CA ∞
(UCA , MBC ; MCA , MAB ) = (Oℓ , MAB ; Y, MCA ) = (UBD , MAD ; MBD , MCD ),

since MBC MAD , MCA MBD , MAB MCD concur at Gq , we know that UCA UBD also pass through Gq . Do it
symmetrically and win.

Let this conic be Cℓ , then by DDIT we have that Cℓ is tangent to ℓ at ∞ℓ . So ℓ is an asymptote of Cℓ .

Similarily ℓ∗ is also an asymptote of Cℓ . So we get:

Proposition 9.4.7. Cℓ is a hyperbola with asymptotes ℓ and ℓ∗ .

From these properties, we get that ∥◦ (⨿ is a projective map ϕ from the nine-point conic C⨿ , defined as
ϕ : C⨿ → L∞ , where ϕ satisfies for all diameters M M ∗ of C⨿ , M ϕ(M ) ∩ M ∗ ϕ(M ∗ ) ∈ C⨿ . In the previous
example of complete quadrangle △ABC ∪ H, C⨿ is the nine-point circle of △ABC, and ϕ sends point P on
the nine-point circle to ∞⊥(A+B+C−hH,2 (P )) .

Note that ϕ is defined by two points M1 , M2 , so similarily to isoptic cubics, we can get

Proposition 9.4.8. If (K, K∗ ), (L, L∗ ) ̸= (L∞ , L∞ ) are two pairs of isogonal lines in complete quadrangle
⨿, then
∥◦ (⨿) = ∥◦ ((KK∗ )(LL∗ )).

In this case, the notation (KK∗ )(LL∗ ) means complete quadrilateral (K ∩ L, L ∩ K∗ , K∗ ∩ L∗ , L∗ ∩ K).

We can also investigate isotomic line conjugation under a projective transformation sending L∞ to some
line L, and redefining isotomic conjugation of point X on segment P1 P2 with harmonic conjugate of P1 P2 ∩ L.
Denote this new harmonic conjugation involution on P1 P2 → P1 P2 as ψℓ . Note that the asymptotes of the
isohaptic cubic become the tangents ℓ, ℓ∗ to Cℓ at the intersection of the polar of L with Cℓ .

272
AoPS Chapter 9. Perfect Six-Point Sets and The Isoptic Cubic
2
Since ∥(⨿) is the dual of a strophoid, we can find some more properties of KW (Q) (strophoid under
projective transformation). Let C be the W -nine-line conic (see Chapter 6) of Q, let Lij be the other tangent
from Aij to C that’s not Aij Akl . Then there exists a projective map ϕ : TW → TC such that ϕ(W Aij ) = Lij ,
and
2
KW (Q) = {ℓ ∩ ϕ(ℓ) | ℓ ∈ TW }

is actually a cubic tangent to C three times. When Q is the incircle or an excircle, C is a parabola with I’s
reflection over M as the focus and the Newton line τ as the directrix.

273
Chapter 10

Secrets of the Complete Quadrilateral

10.1 Steiner’s Hidden Deltoid

10.1.1 Preliminaries

Let Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) be a complete quadrilateral, as in chapter 4, we define:

• Six vertices: Aij = ℓi ∩ ℓj

• Four triangles: △i := △ℓi+1 ℓi+2 ℓi+3

• Three diagonals and the corresponding vertices: Aij Akl and (Aij , Akl ).

• Three quadrilaterals: Aij Ajk Akl Ali , including convex quadrilateral, concave quadrilateral and self-
intersecting quadrilaterals (from a graphical point of view)

• Diagonal triangle (Cevian triangle): δ, a triangle formed by the extension of three diagonal segment.

• I believe you have heard of these special quadrilateral objects.. they have appeared in previous chapters,
for example:

Proposition 10.1.1 (QL-P1, Miquel Point). The circumcircle of the triangles △1 , △2 , △3 , △4 are concurrent
at the point M = M (Q).

Proposition 10.1.2 (QL-L1, Newton line). The midpoint of the three diagonal segment A23 A14 , A31 A24 , A12 A34
are collinear. These point are on the line τ = τ (Q).

Proposition 10.1.3 (QL-Ci3, Miquel circle). Five points M, O1 , O2 , O3 , O4 are concyclic, where Oi is the
circumcenter of △i .

274
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Proposition 10.1.4 (QL-L2, orthocenters line, Steiner line). Let Hi be the orthocenter of △i , then the four
points H1 , H2 , H3 , H4 lies on a line S.

Proposition 10.1.5 (QL-Co1). Let P the the parabola with focus M and directrix S. Then P is tangent to
the four lines in Q.

Proposition 10.1.6 (QL-Tf1, Clawson–Schmidt Conjugate). There exists exactly one transformation ΦQ
which is a composition of inversion and reflection (??) such that A23 ↔ A14 , A31 ↔ A24 , A12 ↔ A34 .

Do not doubt that a transformation can also be called an.

Proposition 10.1.7 (QL-Cu1). The isoptic cubic of the complete quadrilateral Q is

K = {P |∞∨
P Φ(P ) ∈ τ (Q)}

Using what we have learned about isoconjugations, we can also define:

Proposition 10.1.8 (QL-Tf2, QL-Line Isoconjugate). Consider the line isoconjugation in the cevian triangle
δ of Q that fixes the four lines ℓi of the quadrilateral. Then the QL-Tf2 transformation of L is ϕ(L). A
non-isoconjugation way to define this transformation is by intersecting a line ℓ with all the diagonals of the
quadrilateral, then the harmonic conjugate of the three intersection points wrt. their diagonals’ endpoints
are collinear on line ϕ(ℓ). This motivates the notation L∨ .

10.2 Deltoids

In order to thoroughly understand what we are are doing, the common tool is introduced here:

Definition 10.2.1. A deltoid is a curve made by the locus of a point on a circle rolling around the inside of
a circle three times its diameter.

275
AoPS Chapter 10. Secrets of the Complete Quadrilateral

To let the reader have a clue as to what we will be doing in a bit, we will now mention an important result:
for a moving point P on (ABC), the envelope of all the Simson lines formed by P wrt. △ABC forms a
deltoid.

Proposition 10.2.2. Let △ABC be an equilateral triangle, and (I) be it’s inscribed circle. Then the
isogonal conjugate transform of (I) in △ABC is the deltoid with A, B, C as vertices.

Proof. Let P be a point on (I), let the image of P through the homothety with center I, ratio −2/3, −4/3, −2
respectively be P1 , P2 , P3 , and let the image of A, B, C through the homothety with center I, ratio 2/3 be
A′ , B ′ , C ′ . Finally let S be the Simson line of P2 with △A′ B ′ C ′ .

Claim 10.2.3. The foot P ∗ of P3 on S is the isogonal conjugate of P with △ABC.

Proof of claim. Let Q be the reflection of P through AI. Then from symmetry, we only need to proof
A, P ∗ , Q are collinear. Let the reflection of P2 through B ′ C ′ be P2′ , and let the reflection of P3 through BC
be P3′ . Then I, P2′ , P3′ are collinear and the line containing these points is the Steiner line of P2 with △A′ B ′ C ′ .
Let X, Y respectively be the intersection of P2 P2′ , P3 P3′ with S, and Z be the reflection of P3 through AI.
Then I is the centroid of △AP3′ Z, so Q is the midpoint of AP3′ .
Now, let M, N respectively be the midpoint of BC, AI. Obviously, △P3 P ∗ Y is similar to △M QN , so:

P3 Y 2P2 X MN
′ = ′ =2=
Y P3 XP2 NA

Thus we get P ∗ P3′ ∥ AQ, that is, A, P ∗ , Q are collinear.

276
AoPS Chapter 10. Secrets of the Complete Quadrilateral

A'

P Q
N

I P2' P '
3

P1
P*
B' X C'
Y
B C
M
P2

P3
Z
I'

Return to the original proposition, P ∗ lies on (P1 P3 ) (its radius is 1/3 the radius of (I)). Let I ′ be the
reflection of I through BC, then:

∡P3 P1 P ∗ = ∡P3 IA + ∡I ′ IP3′ = 3∡P3 I ′ A

Therefore P ∗ is on the deltoid with A, B, C as vertices.

By rewriting the above proof, we can obtain that every point is on the deltoid with A, B, C as vertices of the
inscribed circle.

Proposition 10.2.4. Let △ABC be an equilateral triangle, with (I) as the inscribed circle, P is a point on
(I), Q is the isogonal conjugate of P in △ABC. Let the image of A, B, C through the homothety with center
I, ratio 2/3 respectively be A′ , B ′ , C ′ , the image of P through the homothety with center I ratio −4/3 be P ′ .
Then the Simson line of Q with △A′ B ′ C ′ is the tangent line of Q to the deltoid with A, B, C as vertices.

Proof. Let the image of P through the homothety at I with ratio 2 be P ∗ , then the isogonal conjugate of
P ∗ in △ABC is on S, so the isoptic cubic of the quadrilateral formed by line S cutting △ABC is a conic C
passing through A, B, C, P, P ∗. Let the second intersection of AP, P ∗ P with (ABC) be U, V , the tangent
line of (I) going through P intersects BC at W , M is the midpoint of BC, I ′ is the reflection of I through

277
AoPS Chapter 10. Secrets of the Complete Quadrilateral

BC. Then:
∡P ∗ W P = ∡P W I = ∡P M I = ∡P ∗ I ′ A = ∡P ∗ U P

therefore P, P ∗ , U, W are concyclic, so

∡V U W = ∡V U P ∗ + ∡P ∗ U W = 180◦

which gives U, V, W are collinear. Now let AP intersects BC at X, then

P (A, B; C, W ) = (X, B; C, W ) = U (A, B; C, V ) = P ∗ (A, B; C, P )

this gives P W is the tangent line of C at P .

Corollary 10.2.5. Let ω be a circle. For any point P on ω, draw a line ℓP ∈ T P , such that

ℓP + (P )ω

is a fixed value, that is, for all P1 , P2 ∈ ω,

∡(ℓP1 , ℓP2 ) = ∡P2 AP1

where A is any point on ω. Then the envelope of {ℓP |P ∈ ω} is a deltoid.

Proof. Let Ω be the image of ω through the homothety at the center O of ω ratio 2, a point A on ω such
that ℓA = OA, A′ is the reflection of O through A, and take B ′ , C ′ ∈ Ω such that △A′ B ′ C ′ is an equilateral
triangle. For any point P ′ on Ω, let SP′ be the Simson line of P ′ with △A′ B ′ C ′ , then SP′ goes through the
midpoint P of OP ′ . Note that OA is the Simson line of A′ with △A′ B ′ C ′ l so

∡(SP′ , ℓP ) = ∡(SP′ , SA ) + ∡(ℓ′A , ℓP ) = ∡A′ B ′ P ′ + ∡P BA = 0◦

where B is the midpoint of OB ′ . That is, SP′ = ℓP , so from (TODO 10.2.3) we know that the envelope of
{ℓP |P ∈ ω} is the deltoid.

With these properties, the theorem becomes very simple.

Theorem 10.2.6 (Steiner’s deltoid theorem). Let △ABC be any triangle, and for each point P ∈ (ABC),
let SP be the Simson line of P with △ABC, then the envelope of {SP | P ∈ (ABC)} is a deltoid.

Proof. Let H be the orthocenter of △ABC, (N ) be the nine-point circle of △ABC, and for any point

278
AoPS Chapter 10. Secrets of the Complete Quadrilateral

P1 , P2 ∈ (N ), let Pi′ be the reflection of H through Pi , then Pi ∈ SPi′ and

∡(SP1′ , SP2′ ) = ∡P2′ AP1′ = −∡P1 XP2

so X ∈ (N ). Therefore, from (TODO 10.2.4) we get that the envelope of {SP |P ∈ (ABC)} is a deltoid.

Definition 10.2.7. With the same label as the previous theorem, we call the deltoid of △ABC as D, the
circumcircle of the three vertices in D as the Steiner circle.

From the proof above, we can deduce the following result:

Proposition 10.2.8. Let △ABC be any triangle, H, (N ), D respectively be the orthocenter, nine-point
circle and the deltoid of △ABC. Let P be an arbitrary point on (ABC), M be the midpoint of HP . Let the
Simson line S of P with △ABC intersects (N ) at Q ̸= M , then S is tangent to D at the point that is the
reflection of Q through M .

Proof. The point P2 in (TODO 10.2.2) is the image of M under a 3× homothety from N .

10.2.1 Rotate it!

We proved above that the envelope of the Simson line is a deltoid. In fact, we can even define the α-Simson
line (Li4 named it randomly):

Definition 10.2.9. Let △ABC be any triangle and an angle α. For any point P ∈ (ABC), let D, E, F
respectively be points on BC, CA, AB such that

∡(P D, BC) = ∡(P E, CA) = ∡(P F, AB) = α(P E, CA) = α(P F, AB) = α

Then D, E, F are collinear (TODO 12.1.11), and is called the α-Simson line of P wrt △ABC, denoted as
Sα,P .

Obviously, when P is fixed, the envelope of Sα,P is a parabola with focus P , and when α = 0◦ , Sα,P is
the infinity line. Similarly to the original Simson line, we also have the following angular relation:

Proposition 10.2.10. For any △ABC and angles α1 , α2 , then for all P1 , P2 ∈ (ABC),

∡(Sα1 ,P1 , Sα2 ,P2 ) = α1 − α2 − ∡P1 AP2

279
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Proof. Simply calculate the angle. In fact,

∡(SP1 ,α1 , SP2 ,α2 ) = ∡(SP1 ,α1 , SP1 ,α2 ) + ∡(SP1 ,α2 , SP2 ,α2 )

= (α1 − α2 ) − ∡P1 AP2

Looking at the title of this section, you will wonder if the envelope of all α-Simson line is also the deltoid.
Of course, when α = 0◦ , all α-Simson line concides to the line at infinity, so we do not consider this situation.

Theorem 10.2.11. Given a △ABC with an angle α ̸= 0◦ , then the envelope of {Sα,P |P ∈ (ABC)} is the
deltoid.

Unfortunately, we cannot directly apply (TODO 10.2.4) because we do not have yet the “α-Steiner
theorem”. So we need the following proposition:

Proposition 10.2.12. Given a △ABC with an angle α ̸= 0◦ , H is the orthocenter of △ABC, then there
exists a circle (N ′ ) and a spiral similarity transformation S centered at H such that S((ABC)) = (N ′ ) and
for any point P ∈ (ABC), S(P ) lies on the α-Simson line of P wrt △ABC.

Proof. Let O be the circumcenter of △ABC, and on BC, CA, AB respectively take RA , RB , RC such that

∡(ORA , BC) = ∡(ORB , CA) = ∡(ORC , AB) = α

+
It’s easy to get that △RA RB RC ∪ O ∼ △ABC ∪ H. Let (N ′ ) = (RA RB Rc ), and for any point P , denote
+
S(P ) as the point Q such that △HP Q ∼ △HON ′ , and the following part proves that this satisfies our
proposition.
Let OA , OB , OC respectively be the reflection of O through BC, CA, AB. Note that H is the circumcenter of
△OA OB OC and
+ + +
△OOA RA ∼ △OOB RB ∼ △OOC RC ∼ △OHN ′

+ +
so △OA OB OC ∪ H ∼ △RA RB RC ∼ N ′ , and N ′ lies on the perpendicular bisector of OH.
+
For any point P ∈ (ABC), let Q = S(P ), from △HP Q ∼ △HON ′ , we have

N ′Q HN ′ ON ′ N ′ RA
= = = .
OP HO OH HOA

And (ABC) has the same radius as (OA OB OC ), so N ′ Q = N ′ RA , that is, Q ∈ (N ′ ). Therefore we have
S((ABC)) = (N ′ ).

280
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Let Sα,P be the α-Simson line of P wrt △ABC, and D, E, F respectively be the intersection of Sα,P with
BC, CA, AB, let PA , PB , PC be the reflection of P through BC, CA, AB. From Steiner’s theorem, we know
that PA , PB , PC are collinear on a line that goes through H. Therefore,

+ + + + +
△P DPA ∼ △P EPB ∼ △P F PC ∼ △ORA OA ∼ △ON ′ H ∼ △P QH

+
Now we have PA PB PC ∪ H ∼ DEF ∪ Q, so Q ∈ Sα,P .

Remark. This proposition is essentially exercise 5 of Section 1.4.

After having this property, the proof of (TODO 10.2.10) is the same as the proof of the original theorem
(TODO 10.2.5). Now we define analogously:

Definition 10.2.13. With the same notation as (TODO 10.2.10), we denote Dα as the α-Steiner deltoid of
△ABC

All deltoids that is tangent to △ABC must be some α-Steiner deltoid.

Then our question is how to get α given any three tangent lines on the deltoid and the triangle formed by
those lines. To solve this, we have to go deeper into the proof of (TODO 10.2.11).

Proposition 10.2.14. Using the same notation as (TODO 10.2.11), let S be ∞Sα,P wrt △ABC, then

+
△ABC ∪ S ∼ △RA RB RC ∪ S(P )

and the rotation angle is 90◦ + α

Proof. Simple angle chasing. In fact, let A∗ = S−1 (RA ), B ∗ = S−1 (RB ), then

∡RB RA S(P ) = ∡B ∗ A∗ P = −∡(SB ∗ ,α , Sα,P ) = ∡(Sα,P , CA) = ∡BAS)

and
∡(RB RC , BC) = ∡(⊥ ORA , BC) = 90◦ + α.

Finally, we prove the most important theorem of this section:

Theorem 10.2.15. Given a △ABC with an angle α ̸= 0◦ , let Dα be the α-Steiner deltoid of △ABC. Let
P1 , P2 , P3 be three points on (ABC), Si = SPi ,α . Let

β = P1 + P2 + P3 − A − B − C − 2α((ABC)).

281
AoPS Chapter 10. Secrets of the Complete Quadrilateral

(i) When β = 0◦ , S∞ , S∈ , S∋ are concurrent.

(ii) When β ̸= 0◦ , S∞ , S∈ , S∋ aren’t concurrent, and Dα is the β-Steiner deltoid of the triangle formed by
S∞ , S∈ , S∋ .

Proof. With the same notation as (TODO 10.2.11), let Si be the isogonal conjugate of ∞S⟩ wrt △ABC, Ti is
the intersection of Si−1 and Si+1 , and H ′ be the orthocenter of S(△P1 P2 P3 ). From (TODO 1.4.5), we have

X X X X
β= Pi − A − 2α = SW A, α − S⟩ − 2α
i cyc cyc i
X X X
= ∡(OB OC , SP1 ,α ) + α = Qi − RA + α ((N ′ ))
i cyc

= Q1 − RA + Q2 Q3 − RB RC + α ((N ′ ))

= S1 − A + Q2 Q3 − BC + 90◦ ((ABC))(by(T ODO10.2.3))

= ∡(⊥ Q2 Q3 , S∞ )

Note that
∡Pi−1 Ti Pi+1 = ∡(Si−1 , Si+1 ) = −∡Pi−1 Pi Pi=1

Therefore S1 , S2 , S3 are concurrent if and only if T1 = T2 = T3 = H ′ , or if and only if H ′ ∈ S1 , which is

equivalent to
β = ∡(⊥ Q2 Q3 , S1 ) = 0

This proves (i), now the following proves (ii).

Assume that S1 , S2 , S3 aren’t concurrent. Let Da be the β ′ -deltoid of △T1 T2 T3 , then by the proof of (TODO
10.2.11), H ′ is the circumcenter of △T1 T2 T3 , where β ′ = ∡(H ′ Q1 , S1 ). So

β ′ = ∡(⊥ Q2 Q3 , S1 ) = β

Taking α = 90◦ gives this famous corollary (Exercise 3 of section 1.4):

Corollary 10.2.16. Given a △ABC, P1 , P2 , P3 are three points on (ABC), Si is the Simson line of Pi wrt
△ABC, then S∞ , S∈ , S∋ are concurrent if and only if

A + B + C = P1 + P2 + P3 ((ABC))

282
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Practice Problems

Problem 1. Assume the Steiner deltoid of △ABC intersects BC, CA, AB at D, E, F . Then AD, BE, CF
are concurrent at the isotomic conjugate of the orthocenter of △ABC.

Problem 2. Let S1 , S2 , S3 be the vertices of the Steiner deltoid of △ABC. Prove that: (formal sum)

3 · S1 S2 = BC + CA + AB

10.3 The Kantor-Hervey Point

We’ve finished discussing the Steiner deltoid, so let’s try to stuff a complete quadrilateral into this.

Proposition 10.3.1. Let complete quadrilateral Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) have the four circumcenters O1 , O2 , O3 , O4 ,
and let Hi′ be the orthocenter of Oi+1 Oi+2 Oi+3 , then Oi , Hi′ are conconic and Hi′ ∈ ℓi .

Proof. Let M be the Miquel point of Q, then note that ℓi is the Steiner line of M wrt. △Oi+1 Oi+2 Oi+3 ,
and thus Hi′ ∈ ℓi . Also Hi′ just lies on the rectangular hyperbola through O1 , O2 , O3 , O4 .

Corollary 10.3.2 (QL-P5, the Clawson point). The midpoints of Oi Hi′ coincide.

Proof. This is just the Poncelet point of O1 O2 O3 O4 .

Corollary 10.3.3 (QL-Ci4, the Hervey circle). The orthocenters H1′ , H2′ , H3′ , H4′ are concyclic.

Proof. Reflect circle (O1 O2 O3 O4 ) over the Poncelet point of O1 O2 O3 O4 , then we get a circle through
H1′ , H2′ , H3′ , H4′ as desired.

Theorem 10.3.4 (QL-Qu2, the Kantor-Hervey deltoid). There exists a deltoid ∆ tangent to all four sides of
Q.

We can obtain this from the α-Steiner deltoid (in reality, ∆ is the α-Steiner deltoid of △ℓj ℓk ℓl , where the
angle
α = ∡(M (ℓi ∩ ℓj ), ℓj ) = ∡(ℓi , τ ).

(τ is the Newton line) This gives us some motivation for our proof:

Proof. Consider the transformation ϕ = Φ ◦ sT , where sT is reflection across the Clawson point (TODO
10.3.2) and Φ is Clawson-Schmidt conjugation in Q. For a point P ∈ (H1′ H2′ H3′ H4′ ), define LP to be the
perpendicular from P to M ϕ(P ), then by simple angle chasing and (TODO 10.2.4) we get the envelope of
LP is a deltoid ∆. Note that ℓi = LHi′ , so ∆ is tangent to the four sides of Q.

283
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Since this deltoid is unique, we get the converse of this proof,

Corollary 10.3.5. The Kantor-Hervey deltoid △ of quadrilateral Q is tangent to the Hervey circle of Q.
θ
Proposition 10.3.6 (QL-P3, the Kantor-Hervey point). Let be the center of the Kantor-Hervey deltoid,
θ
then lies on the four perpendicular bisectors of the segment between the orthocenter and circumcenter of
the component triangles of Q.
θ
Proposition 10.3.7 (QL-P2, the Morley point). Let M o be the foot from to the Steiner line S. Let
N1 , N2 , N3 , N4 be the nine-point centers of each component triangle of the quadrilateral. Then the lines
passing through Ni perpendicular to ℓi concur at M o.

θ θ
Proof. Continuing the previous notation, note that Ni ⊥ Oi Hi , so , Hi , Ni , M o are concyclic. We have

∡(Ni M o, ℓi ) = ∡Ni M oHi + ∡(SQ , ℓi ) = ∡Ni Hi + ∡(τ, ℓi ) + 90◦

θ

= αi − αi + 90◦ = 90◦ ,

where τ is the Newton line of Q. Thus Ni M o ⊥ ℓi .

There are also proofs of all of this without deltoids.

Practice Problems

Problem 1. Let ℓ be the reflection of the Newton line of quadrilateral Q over its Miquel point. Prove that ℓ
is tangent to the Kantor-Hervey deltoid of Q.

10.4 Morley’s Beautiful Cardioids

After talking so long about deltoids, everyone should be at least a little interested in cardioids, right?

(This joke doesn’t work in English.) (The joke is that the word for “deltoid” is the same as the word for
“tricuspid valve curve”, and “cardioid” is “heart curve”, in Chinese.)

Definition 10.4.1. A cardioid K is the curve made by the locus of a point on a circle Γ′ that’s rolling
around another circle Γ. We call the center of Γ the center of the cardioid K. Additionally, each cardioid has
one cusp, where it’s not differentiable.

This definition is useless. So we first re-define the cardioid as such:

Proposition 10.4.2. A cardioid is the image of the inverse of a parabola around its focus. (Similarily,
inverting a cardioid at its cusp gets you a parabola.).

284
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Proof. When we define K as the locus of rolling Γ′ around Γ, let Y be the cusp of K. Then it’s not hard to
show that for every point P on K, the perpendicular bisector of P Y is tangent to Γ. After inverting this
property with respect to Y , we get that the new locus is just the locus of centers of circles going through Y
and tangent to the image of Γ′ (which is just a fixed line). However, this locus is just a parabola with Y as
focus and the image of Γ′ as the directrix.

Proposition 10.4.3. Let Y be the cusp of the cardioid K, formed by rolling Γ′ around Γ. Then for every
point P on K, the perpendicular bisector of P Y is tangent to Γ.

Proof. Reflect over the perpendicular bisector, since the arc lengths formed by rolling on both circles are the
same, by the definition of rolling without slipping.

Now we present a much more elegant proof of Morley’s trisector theorem (TODO 1.6.1).

Theorem 10.4.4. For any triangle ABC (labelled counterclockwisely) draw the angular trisector ℓBi
A through

A of ∠ABC, with i = −1, 0, 1, satisfying ℓB0

A is located in ∠BAC, and

Bj ◦
∡(ℓB0 Bi
A , CA) = 2∡(AB, ℓA ), ∡(ℓBi
A , ℓA ) = (j − i) · 60

Bj
ℓCi Bi Ci Ai Ai Bi Ci
A is the isogonal conjugate of ℓA in ∠BAC, define ℓB , ℓB , ℓC , ℓC similarly. Define ai j = ℓB ∩ ℓC , define

bij , cij similarly. Then for all (i, j, k) ∈ {−1, 0, 1}3 such that 3 ∤ 1 + i + j + k, △ajk bki cij is an equilateral
triangle, and
∡(bki cij , BC) = ∡ajk BC + ∡ajk CB

The rest are similar.

Now, we’ll also define the first, second and third Morley triangle as △0 = △a00 b00 c00 , △1 = △a11 b11 c11 , △2 =
△a(−1)(−1) b(−1)(−1) c(−1)(−1) . Define

[
Λ△ABC = (bii cii ∪ cii aii ∪ aii bii )

is the union of all edges of △i , and

λ△ABC
A = {aij |(i, j) ∈ {−1, 0, 1}2 }

Similarly, we also define λ△ABC

B , λ△ABC
C , λ = λA ∪ λB ∪ λC are all the vertices.

Theorem 10.4.5. Given a complete quadrilateral Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ), let four triangles in Q be △1 , △2 , △3 , △4 , λi =

△i △
λi , Aij = ℓi ∩ ℓj . If Li ⊂ λi , Lj ⊂ λj are two lines such that Li ∩ θA jk
= {Pi , Qi } and Li ∩ θAkji = {Pj , Qj }

285
AoPS Chapter 10. Secrets of the Complete Quadrilateral

wrt Akl , then

Li ∩ Lj ∈ (Ajl Pi ∩ Ail Pj )(Ajl Qi ∩ Ail Qj ) ∩ (Ajk Pi ∩ Aik Pj )(Ajl Qi ∩ Ail Qj )

T
with i, j, k, l pairwise distinct. In other words, Li ∩ Lj ∈ λm .

Proof. Observe that:

(Ajl Pi ∩ Ail Pj )(Ajl Qi ∩ Ail Qj ) ⊂ Λk

Since △Ajl Pi Qi and △Ail Pj Qj are perspective wrt Ak l, by Desargues’s theorem,

Li ∩ Lj ∈ (Ajl Pi ∩ Ail Pj )(Ajl Qi ∩ Ail Qj ) ⊂ Λk ,

Similarly
Li ∩ Lj ∈ (Ajk Pi ∩ Aik Pj )(Ajk Qi ∩ Aik Qj ) ⊂ Λl ,
T
so Li ∩ Lj ∈ Λm .
T
Corollary 10.4.6. Using the same notation with (TODO 10.4.5), | Λi | ≥ 27 (counted with multiplicity.

Proof. We know that the number of combinations of (Li , Lj) satisfying {Pi , Qi }and {Pj , Qj } wrt the Akl is
T
27 (given i, j, k, l), so | Λi | ≥ 27.

Proposition 10.4.7. If all three sides of △ABC are tangent to a cardioid K, and K and BC are tangent at
two points, then the center P of K is P ∈ λA . In the other hand, if P ∈ λA , then there exists a cardioid K
that is tangent to all three sides of △ABC, and tangent to BC at two points.

Proof. Let K tangent to BC at T1 , T2 , and tangent to AB at T , then there is a point S located at P is

the center of the circle that goes through the cusp Y of K, and △P T1 T2 is an equilateral triangle with Y
as the center. Let P be the image of P through the homothety with center S, ratio −2, then P ′ T = AB.
Let the perpendicular bisector of P P ′ intersect P ′ T at B ′ , let M, M ′ , N respectively be the midpoint of
T1 T2 , P P ′ , SP ′ . Note that

∡M P B ′ = ∡Y P S + ∡P ′ P B ′ = ∡SN T + ∡T P ′ S = ∡B ′ P P ′

−
so △B ′ P M ∼ △B ′ P ′ M ′ , this gives ∡B ′ M P = 90◦ . Therefore B ′ = B and

3 · ∡CBP = ∡CBP + ∡P BM ′ + ∡M ′ BP ′ = ∡CBA

286
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Similarly, 3 · ∡P CB = ∡ACB, this gives P ∈ λA .

If P ∈ λa , take two points T1 , T2 on BC such that △P T1 T2 is an equilateral triangle, and then write the
proof above in reverse. The uniqueness is because P and the cusp are unique.

I (Li4) haven’t found a short proof for the following theorem. I hope you can find a shorter one for me
(?). Although the proof looks long, in fact the idea is not difficult.

Theorem 10.4.8. If a cardioid K is tangent to △ABC, then the center of K lies on Λ△ABC . In the other
hand, if P ∈ Λ△ABC , then there is a unique cardioid centered at P and is tangent to all three sides of △ABC.

Proof. Let L be a line that is tangent to K and intersects K at two points. D, E, F respectively be the
intersect of L with BC, CA, AB. Then from (TODO 10.4.7) we know that P ∈ λ△AEF
A ∩ λ△BF
B
D
∩ λ△CDE
C .
Let U, V respectively be two points on (P F D), (P DE) such that ∡U DP = ∡V DP = 60◦ , then P U ⊂
Λ△BF D , P V ⊂ Λ△CDE and D, U, V are collinear, so from (TODO 10.4.5) we know that P ∈ Λ△ABC .
In the other hand, if P ∈ Λ△ABC , assume P ∈ ℓ ⊂ Λ△ABC , let

{U1 , U2 } = ℓ ∩ λ△ABC
B , {V1 , V2 } = ℓ ∩ λ△ABC
C ,

Then (AUi , AVi ) are angular trisector of ∠BAC. Take two points S1 , T1 on AU1 , AV1 respectively, such that
△P S1 T1 is an equilateral triangle and ∡P S1 T1 = ∡U1 AU2 . Let

S2 = P T1 ∩ aU2 , T2 = P S1 ∩ AV2

Then it is obvious that S2 , T2 ∈ (AS1 T1 ), so △P S2 T2 is also an equilateral triangle. Define

E = CA ∩ S1 S2 , F = AB ∩ T1 T2

It’s easy to see that the second intersection G, H of (AS1 T1 ) with CA, AB respectively are the reflection of
P wrt S1 S2 , T1 T2 . Therefore, ES1 , F T1 are the angle bisector of ∠AEP, ∠P F A respectively, so △P S1 T1 is a
Morley triangle of △AEF . Similarly, △P S2 T2 is also a Morley triangle of △AEF .
Let D be intersection of EF and BC, W = P S1 ∩ AS2 , X = P T1 ∩ AT2 , note that P ∈ Λ△ABC ∩ Λ△AEF and

P = U1 U2 ∩ S1 W = V1 V2 ∩ T1 X, A = U1 S1 ∩ U2 W = V1 T1 ∩ V2 X

So from (TODO 10.4.5) we get P ∈ λ△BF

B
D
∩ λ△CDE
C , and then from (TODO 10.4.7) we know that there
exists cardioids KA , KB , KC with center P such that KA is tangent to all three sides of △AEF and is tangent
to EF at two points, similarly with KB , KC . Since D, E, F are collinear, we get KA = KB = KC , and are
tangent to △ABC.

287
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Finally, suppose there are two cardioids K1 , K2 centered at P and tangent to all three sides of △ABC. Let
Li be the tangent line of Ki and intersects Ki at two points. Ei and Fi are the intersection points of Li with
CA and AB respectively, then
P ∈ λ△AE
A
1 F1
∩ λ△AE
A
2 F2
.

Take two points Yi , Zi on AU1 , AV1 respectively such that △AYi Zi is a Morley triangle of △AEi Fi . Note
that
+
△P Y1 Z1 ∼ △P Y2 Z2

So △P Y1 Z1 = △P Y2 Z2 , which gives E1 = E2 , F1 = F2 . This implies K1 = K2 , which proves the uniqueness.

Proposition 10.4.9 (QL-27Qu1, Morley’s Multiple Cardioids). There are at least 27 cardioids that is
tangent to four lines of a complete quadrilateral Q (counted with multiplicity).

T T
Proof. From (TODO 10.4.6), we know | Λi ≥ 27| (counted with multiplicity), Let P ∈ Λi , then from
(TODO 10.4.8) we know that there is a cardioid Ki centered at P that is tangent to all three sides of △i . For
any distinct i, j, by (TODO 10.4.8) the last part of the proof is that Ki = Kj , so there is a unique cardioid K
centered at P that is tangent to all four sides of Q.

Practice Problems

Problem 1. Let K be a cardioid with cusp Y . An arbitrary line through Y intersects K at A, B. Prove that
the locus of the midpoint of AB is a circle.

Problem 2. With the same notation as (TODO 1.6.1), prove that: For all (i, j, k) ∈ {−1, 0, 1}3 such that
3 ∤ 1 + i + j + k,

(i) △ajk bki cij is perspective with △a(j−1)(k−1) b(k−1)(i−1 c(i−1)(j−1) wrt center Pijk (Note that not all
equilateral triangle are positively similar);

(ii) The isogonal conjugate of Pijk wrt. △ABC is the perspector of △ajk bki cij and △ABC.

Problem 3. With the same notation as (TODO 1.6.1), prove that:

(i) For all i ∈ {−1, 0, 1}, A ∈ (aii a(i+1)(i+2) a(i+2)(i+1) );

(ii) (a11 , a23 , a32 ), (a22 , a31 , a13 ), (a33 a12 a21 ) are coaxial.

Problem 4 (QL-Qu1, Morley’s Mono Cardioid). Let the Miquel point of the complete quadrilateral Q be
M , and the Miquel circle be M. Consider all circles passing through M and whose center is located on M.
Prove that the envelope of these circles is a cardioid K

288
AoPS Chapter 10. Secrets of the Complete Quadrilateral

Problem 5 (QL-Cu2). Let the four triangles of the complete quadrilateral Q be △1 , △2 , △3 , △4 . Prove that:
the 27 common intersection points of Λi defined by (TODO 10.4.5) are located on the same cubic curve.

Problem 6. Prove that the Morley triangle of △ABC is similar to the vertices of its Steiner deltoid.

289
Chapter 11

Basic Cubic Theory with Liang and

Zelich

Definition 11.0.1. In the complex projective plane CP2 , the set of roots [x : y : z] (or the zero locus) of a
degree d polynomial F (x, y, z) form a degree d curve V = V(F ). We say that V is non-singular if every point
on it has a unique tangent. This is equivalent to saying

∂F ∂F ∂F
F = = = =0
∂x ∂y ∂z

has no double roots.

A classical result in algebraic geometry is:

Theorem 11.0.2 (Bezout’s). A degree d curve intersects a degree e curve at de points in CP2 (up to
multiplicity).

11.1 The Group Law

In order to talk about the group law, it’s necessary to use a theorem that is incredibly important regarding
cubic curves:

Theorem 11.1.1 (Cayley-Bacharach). If two cubics K0 , K∞ (including factorizable cubics), intersect at

points P1 , . . . , P9 and any five of P1 , . . . , P8 are not collinear, then we have that

(i) Every cubic curve K passes through the 8 points P1 , . . . , P8 can be written as K0 + tK∞ (linear
combination), particularly, P9 ∈ K;

290
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

(ii) And
P9 (P5 , P6 ; P7 , P8 ) = (P5 , P6 ; P7 , P8 )P1 P2 P3 P4

Proof. We do a bunch of casework. For (i), first suppose we have P1 , . . . , P8 such that there are four collinear
points. Assume WLOG that P1 , · · · , P4 are the collinear points on line ℓ. Then let the K0 , K∞ , K all go
through these points. Since four are collinear, we can factor a degree-1 polynomial out of these cubics, so
they all vanish on the line. This tells us that the cubic is a product of a degree-1 polynomial and a degree-2
polynomial. So let K0 = ℓ ∩ C0 , K∞ = ℓ ∩ C∞ , K = ℓ ∩ C, where C0 , C∞ , C are cubics that pass through
P5 , · · · , P8 . Thus C + C0 + tC∞ , so we have K = K0 + tK∞ . By the same logic, suppose we have seven points
in P1 , . . . , P8 that are conconic, assume WLOG that it is P1 , . . . , P7 . Then let the three cubics through these
points be K0 = ℓ0 ∪ C, K∞ = L∞ ∪ C, K = ℓ ∪ C, where ℓ0 , L∞ , ℓ are lines through P8 . Then given a t, we have

K = ℓ ∪ C = (ℓ0 + tL∞ ) ∪ C = K0 + tK∞ .

[yappy casework] Suppose P1 , . . . , P8 are in general position (No 6 points conconic, no 3 points collinear).
Suppose there exists a K through these points such that K is not in the form K0 + tK∞ . Then we can choose
s, t such that K′ = sK + K0 + tK∞ and P1 P2 lies inside K′ . This will imply that K′ = P1 P2 ∪ C, but we know
that P3 , . . . , P8 are not conconic, so contradiction.

Because it’s possible to construct points on a cubic curve based on known points, often 3 straight lines (a
degenerate cubic) is used to construct more points. A great example of this is Miquel’s Theorem:

Example 11.1.2. For a complete quadrilateral in the form Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ), let Aij = ℓi ∩ ℓj and let
∞i , ∞−i the 2 circle points at infinity and let Ki = ℓi ∪ (Akl Alj Ajk ∞i ∞−i ). Then Ki must pass through
the 8 points Aij , ∞±i . So Ki must all pass through a common ninth point we call this point M . Notice that
Akl Alj Ajk ∞i ∞−i = ⊙Akl Alj Ajk and ℓj , ℓk , ℓl already each intersect K at two points each so the circumcircles
of the four triangles in Q all intersect at M .

A corollary to this is:

Theorem 11.1.3 (Three Conics theorem). Let three non-degenerate conics (degree-2 curves) C1 , C2 , C3
share two points P, Q. If Ci and Cj intersect at two other points Aij , Bij then A23 B23 , A31 B31 , A12 B12
concur.

Proof. Let R = A31 B31 ∩ A12 B12 . Note that both C2 ∪ A31 B31 and C3 ∪ A12 B12 pass through the points
A23 , A31 , A12 , B23 , B31 , B12 , P, Q, R. Evidently C1 ∪ A23 B23 passes through eight of those points, minus R

291
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

thus by Cayley-Bacharach, we get that R ∈ C1 ∪ A23 B23 . R ∈ C1 and R ∈ A31 B31 ∩ A12 B12 we get that
C1 , C2 , C3 intersect at a point R separate from P, Q contradicting our claim, thus, R ∈ A23 B23 .

Of course, this proposition can be proved with just the cross ratio.

Definition 11.1.4. Given a non-singular cubic curve K and a point O that lies on it, we define the addition
of points on K as follows: for any two points P, Q ∈ K, let X be the third intersection of P Q and K. We
define P + Q as the third intersection of OX and K. (if there is an overlap, then the line becomes the tangent
to the curve at the overlapped point and the and the number of intersection is the number of doubled roots.)
You can also think of this as “three points are collinear if and only if they add up to O”.

The identity element of this addition is O and commutativity holds, ie. P + Q = Q + P . Let S be the
third intersection of the tangent to K at O with K, (note that because the line is a tangent, O is counted
twice). Then the additive inverse of a point P ∈ K is the third intersection of P S with K.

We will now prove that associativity holds: let P, Q, R ∈ K and let X, Y, Z be the third intersection of
P Q, QR, (P + Q)R with K respectively. Then, O, X, P +Q; O, Y, Q+R and O, Z, (P +Q)+R are respectively
collinear. Consider the nine intersection points of P Q ∪ OY ∪ (P + Q)R with K. They are

P, Q, X, O, Y, Q + R, P + Q, R, Z

Notice that OR ∪ OX ∪ (Q + R)P passes through eight of the points (except Z), so by Cayley-Bacharach,
it must go through the ninth. In other words, Z is the third intersection of (Q + R)P and K. So P + (Q + R)
is the third intersection of OZ and K. Thus P + (Q + R) = (P + Q) + R.

Note that it is necessary to assume that K is non-singular to guarantee the existence and uniqueness of
the tangent. In other words, we can only perform group operations on the non-singular part of K, we call
this part K◦

Proposition 11.1.5. For the group (K, O), let L be the third intersection of the tangent to K at O with K.
Then three points P, Q, R ∈ K are collinear iff. P + Q + R = L

Proof. For the ‘only if’ condition we assume P, Q, R are collinear, then P + Q is the third intersection of OR
with K. Thus (P + Q) + R is the third intersection of OO (the tangent to K at O) with K, or (P + Q) + R = L

For the if condition assume P + Q + R = L, then P + Q = L − R is the third intersection of OR with K,

thus P, Q, R are collinear.

292
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Proposition 11.1.6. For any six points P1 , P2 , P3 , P4 , P5 , P6 lying on a cubic curve K. P1 , P2 , P3 , P4 , P5 , P6

lie on a conic section if and only if

P1 + P2 + P3 + P4 + P5 + P6 = 2 · L

Where L is the third intersection of the tangent to K at O.

Proof. Let Q1 , Q2 , Q3 be the third intersections of P1 P4 , P2 P5 , P3 P6 with K. Thus by Cayley-Bacharach

(TODO 11.1.1), P1 , P2 , P3 , P4 , P5 , P6 lie on a common conic section iff Q1 , Q2 , Q3 . Because

P1 + P4 + Q1 = P2 + P5 + Q2 = P3 + P6 + Q3 = L

The condition Q1 , Q2 , Q3 is equivalent to

Q1 + Q2 + Q3 = 3 · L − (P1 + P2 + P3 + P4 + P5 + P6 )

= L ⇐⇒ P1 + P2 + P3 + P4 + P5 + P6 = 2 · L

Proposition 11.1.7. For any point A on the cubic curve K, define iA : K → K be the map that sends any
P to the third intersection of AP and K. Then for any point B ∈ K satisfying 2 · A = 2 · B (the tangent at A
intersects the tangent at B on the cubic) we get

(a) For P, Q ∈ K such that B, P, Q are collinear, B, iA (P ), iA (Q) are also collinear. Then we obtain the
mapping

φ
TB −−→ TB

BP 7−→ BiA (P )

(b) φ is a projective involution.

Proof. We first prove (a), since B + P + Q = L, it follows that

B + iA (P ) + iA (Q) = B + (L − A − P ) + (L − A − Q)

= 2L + (2B − 2A) − (B + P + Q) = L

293
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Thus B, iA (P ), iA (Q) are collinear. For (b), for two points P, Q ∈ K, we have

P + iA (P ) + Q + iA (Q) + 2 · B = 2 · (L − A) + 2 · A = 2 · L,

Therefore the conic C = (P iA (P )QiA (Q)B) has a common tangent with K at B. Let X be the second
intersection of line BA with C. Then (by the fact that involutions on a conic are all just the second intersection
map from a fixed point) we have a projective involution on C such that (A, X), (P, iA (P )), (Q, iA (Q)) are
pairs under this involution. This tells us that there exists a projective pencil involution on TB such that
(TB C = TB K, BA), (BP, BiA (P ) = ϕ(BP )), (BQ, BiA (Q) = ϕ(BQ)) are involutive pairs. Since these pairs
hold for all Q, ϕ must be a projective involution.

Now let’s address the isoptic cubic (see: (TODO 9.2.4)) as an example. Let M, τ, Φ be the Miquel point, the
Newton line, and Clawson-Schmidt conjugation on K. Assume that Clawson-Schmidt conjugation has no
fixed points (or else at the fixed points, there are two tangents to K at the fixed points). Let M be the zero
element of the group law, O, and let’s see what happens.

For a pair of isogonal conjugates (P, P ∗ ) on the isoptic cubic, we know that P ∞τ ∩ P ∗ M ∈ K, so we have
P + ∞τ = P ∗ = Φ(P ). In other words, on K, Φ is just equivalent to adding ∞τ . Since Φ is an involution, we
have 2 · ∞τ = O.

Let I, J ∈ K be the two circle points. Let ∞τ be the third intersection of IJ with K. Since I + J is the
vanishing point T of K(see (TODO 9.2.24)), we have that for four points P1 , P2 , P3 , P4 on K, they are cyclic
if and only if
P1 + P2 + P3 + P4 + T = 2 · S.

Because M under Clawson-Schmidt conjugation is sent to ∞τ (we pick this specific infinity point by the
tangent), we have Φ(S) = T , and thus 2 · S = 2 · (T + ∞τ ) = 2 · T. In other words, we can simplify the above
expression as
P1 + P2 + P3 + P4 = T.

A hard to prove but very essential theorem on cubics:

Theorem 11.1.8. If a non-singular cubic K is defined on C, there exists a complex number τ, ℑτ > 0, we
have (as Riemann surfaces and groups),

(K, O) ∼
= C⧸Zτ + Z.

This belongs properly in Riemann surface theory, so we will not prove it here. However, we can use this
to get a corollary.

294
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Corollary 11.1.9. We have

(K, O)[m] := {P ∈ K | m · K = O} ∼
= (Z/mZ)2 .

11.1.1 Rematch with the Dragon

Let Q = (ℓ1 , ℓ2 , ℓ3 , ℓ4 ) be a complete quadrilateral. Our goal is to address the isoptic cubic K (see: (TODO
9.2))through these tools we developed on cubics. As before, define the notation A = A23 , B = A31 , C =
A12 , A∗ = A14 , B ∗ = A24 , C ∗ = A34 such that △A∗ B ∗ C ∗ are collinear and “cut” △ABC.

This is completely projective, so we can use this to prove some old theorems. For example, (TODO
1.3.14):

Proposition 11.1.10. For a point P not on the line at infinity, P ∈ K if and only if

P B + P B∗ = P C + P C ∗.

Proof. First, by DDIT, we have that P B + P B ∗ = P C + P C ∗ if and only if P (Aij , Akl ), P (I, J) all swap
under a pencil involution.

Let P ∗ be the isogonal conjugate of P in △ABC, and let Q = P J ∩ P ∗ I. If P ∈ K, then by (T ODO7.4.6)

we have that P, Q, I lie on a common circumconic of △ABC. But (T ODO7.4.6) also tells us that on △P QI,
we have
A × A∗ = B × B ∗ = C × C ∗ .

In other words, P (Aij , Akl ), P (I, J) = P (I, Q) define a DDIT projective involution on TP .

For the other direction, if P (Aij , Akl ), P (I, J) define a DDIT projective involution on TP, consider
isogonal conjugation on △P QI, ϕ = A × A∗ . Then since A, B, C, P, Q, I are conconic, by applying (TODO
7.4.6), we get that on △AB ∗ C ∗ , P × P ∗ = I × J. So P ∈ K.

This is equivalent to
P (A12 , A13 ; J, I) = P (A24 , A34 ; J, I),

and (google the definition of circular cubic curves as the intersection of a pencil of conics, the book doesn’t
really touch on this) we can represent K as the locus of both intersection points of conics in these two pencils:

Ct := {P | P (A12 , A13 ; J, I) = t},

Ct′ := {P | P (A24 , A34 ; J, I) = t}.

295
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

where t is some constant.

To prove this is a cubic, simply set t = 1, since then the locus of these two pencils’ intersection points is
just L∞ ∪ A14 , so K is a cubic. Since K passes through I, J by this definition, it’s also a circular cubic. (You
can also prove this by Miquel’s theorem and Cayley-Bacharach.)

Theorem 11.1.11 (Fundamental Theorem of the Isoptic Cubic). For two pairs of isogonal conjugates
(P, P ∗ ), (Q, Q∗ ) in quadrilateral Q, we have

K(Q) = K((P P ∗ )(QQ∗ ))

Proof. First let’s just set one of these pairs of isogonal conjugates to be (A, A∗ ). We want to prove K(Q) =
K((AA∗ )(P P ∗ )). From
BA + BA∗ = BP + BP ∗ ,

we have that B lies on the isoptic cubic of (AA∗ )(P P ∗ ), and thus B ∗ lies on it too. Similarily, C, C ∗ all lie
on this isoptic cubic as well. This gives us that K(Q) and K((AA∗ )(P P ∗ )) all go through the ten points
A, A∗ , B, B ∗ , C, C ∗ , P, P ∗ , I, J. Thus these are the same cubic!

Note that this also gives us that Q, Q∗ ∈ K((AA∗ )(P P ∗ )). So by the exact same proof we also have

K(Q) = K((AA∗ )(P P ∗ )) = K((P P ∗ )(QQ∗ )).

This also gives a new proof for the fact that (M, ∞τ ) are isogonal conjugates.

Proposition 11.1.12. Let M ∗ be the isogonal conjugate of the Miquel point M in Q. Then M ∗ is the
intersection of L∞ and L∨
∞ = τ.

Proof. We do a very big tethered MMP (see 7.A). Fix △ABC and A∗ , and animate M on (ABCIJ).
Let △La∞ Lb∞ Lc∞ be the anticevian triangle of L∞ wrt. △ABC, and let X, Y, Z be the intersections of
AA∗ , BB ∗ , CC ∗ with La∞ , Lb∞ , Lc∞ . Then X, Y, Z all lie on L∨
∞ , by cross-ratio. Thus

M → (CA∗ IJM ) → B ∗ → Y → L∨ ∨
∞ → L∞ ∩ L∞

is a projective map (In other words, L∞ ∩L∨

∞ moves with degree 2). So by checking three cases, M = B, C, I, J,

we have that L∞ ∩ L∨ ∗
∞ is just ∞CA , ∞AB , J, I, respectively. Thus M ∈ ∞τ .

296
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Note that from (TODO 7.4.6), on △M IJ, we have that Clawson-Schmidt conjugation on Q is

Φ := A × A∗ = B × B ∗ = C × C ∗ .

We discover that for a point P on the isoptic cubic, Φ(P ) and the isogonal conjugate of P , P ∗ have the
following relation:

∡Aij Φ(P )Aik = ∡P Akl M + ∡M Ajl P = ∡(ℓj , ℓk ) − ∡Aij P Aik = ∡Aij P ∗ Aik ,

where the first equality comes from the involution given by DIT where L∞ cuts the quadrilaterals (M, Φ(P ), Akl , Aij Φ(P )∩
Akl P ) and (M, Φ(P ), Ajl , Aik Φ(P ) ∩ Ajl P ). The final equality is directly due to P, I, J, P J ∩ P ∗ I, P ∗ J ∩ P I
lying on a common circumconic of △Ajk Aki Aij .

Specifically, this tells us that for any two pairs of isogonal conjugates (P, P ∗ ), (Q, Q∗ ), (P P ∗ )(QQ∗ )’s
Clawson-Schmidt conjugation is the exact same as the Clawson-Schmidt conjugation of the original quadrilat-
eral, and thus these two quadrilaterals have the same Miquel point! (This can also be directly proven by
group-law on the common isoptic cubic.) Since these quadrilaterals and quadrilateral (AA∗ )(P P ∗ ) have a
common isoptic cubic, their third intersection with L∞ , the point at infinity, ∞τ is the same, so they have
the same Newton line as well!

Thus ∞P P ∗ ’s harmonic conjugate on P P ∗ , ∞∨ ∨ ∨

P P ∗ lies on τ = L∞ = ∞AA∗ ∞τ .

This completely gives us the characterization of K given in (9.2.4)!!

Theorem 11.1.13. The isoptic cubic K’s affine part(remove points at infinity) is just

n o
K(Q)
e := P | τ, L∞ harmonically divide P Φ(P ) .

Proof. We already know that K \ L∞ lies in this set, and since K(Q)
fl is also a (affine) cubic, it is also the

affine part of K.

Practice Problems

Problem 1. Prove the following theorems with Cayley-Bacharach:

• Pappus and Pascal’s theorems,

• Miquel’s theorem,

• (Conic) Reim’s theorem,

• Radical axis concurrence

297
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

• Brokard’s theorem,

• Humpty point, H, B, C concyclic.

Problem 2 (Four Conics Theorem). Let C1 , C2 , C3 be three conics. Let Aij , Bij , Pij , Qij be the four
intersection points of Ci , Cj . If P23 , P31 , P12 , Q23 , Q31 , Q12 are conconic, prove: A23 B23 , A31 B31 , A12 B12 are
concurrent.

Problem 3. Let K be a cubic, and let the two tangents from A ∈ K intersect K again at B, C. Let S be the
third intersection point of BC and K. Prove that A, S’s respective tangents to K intersect also on K.

11.2 The Path of Polarity

Definition 11.2.1. Given a parametrized degree d curve V = V(F ) and a point P = [p : q : r], we can define
the polar curve of V wrt. P , notated as ∂P V as the set of points [x : y : z] such that for a given F ,

∂F ∂F ∂F
P · ∇F := p +q +r =0
∂x ∂y ∂z

This will give a curve of degree d − 1, by definition.

Obviously for two distinct points P and Q, we have ∂P ∂Q V = ∂Q ∂P V. Polar curves have a very good
characterization:

Proposition 11.2.2. Given a degree d curve V, and given any line ℓ passing through P , let Q1 , . . . , Qd be
the intersections of ℓ and V. Then for any point M ∈ ℓ, M ∈ ∂P V if and only if

(i) M = P ∈ {Q1 , . . . , Qd },

(ii) M occurs twice or more in {Q1 , . . . , Qd } (i.e. is a repeated root/tangency), or

d
Pd 1
(iii) PM = i=1 Qi M .

Remark. If you don’t know multivariable calculus, interpret this as the degree d − 1 curve through all
(possibly complex) points on V such that their tangents pass through P .

Proof. If we parameterize ℓ as {Pt = [p + tu : q + tv : r + tw] | t ∈ P1 } and let ti be the coordinate of Qi , then

d
Y
f (t) = F (p + tu, q + tv, r + tw) = c (t − ti )
i=1

298
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

for some constant c. Then

  d
∂F ∂F ∂F ∂f X f (t)
u +v +w (Pt ) = (t) = .
∂x ∂y ∂z ∂t i=1
t − ti

Since F is homogeneous of degree d, by Euler’s formula

 
∂F ∂F ∂F
d · f (t) = (p + tu) + (q + tv) + (r + tw) (Pt )
∂x ∂y ∂z
  d
∂F ∂F ∂F X tf (t)
= p +q +r (Pt ) + .
∂x ∂y ∂z i=1
t − ti

Thus M ∈ ∂P V if and only if

d
X tf (t)
d · f (t) = .
i=1
t − ti

If f (t) = 0, then t = ti for some i, and hence

• When ti = 0, the above equation obviously holds, i.e. (i);

• When ti ̸= 0, this is equivalent to (t − ti )2 | f (t), i.e. (ii).

If f (t) ̸= 0, then this is equivalent to

d d
d d X 1 X 1
= = = ,
PM t i=1
t − ti i=1
Qi M

i.e. (iii).

Remark. The most interesting use of this theorem is in the case if V is a cubic and P lies on V. Then we
have for any line through P that intersects V again at Q2 , Q3 , that

(Q2 , Q3 ; P, M ) = −1.

In this case, we call V the poloconic of P .

Corollary 11.2.3. Given any projective transformation Φ, for any degree d curve V and a point P ,

∂Φ(P ) Φ(V) = Φ(∂P V).

In other words, the polar curve is projectively invariant.

299
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Proof. Cases (i) and (ii) are clearly invariant, so we only need to show that

d d
d X 1 d X 1
= =⇒ = .
PM i=1
Qi M Φ(P )Φ(M ) i=1
Φ(Qi )Φ(M )

If all the Qi are equal to P , then the left and right equalities are always true (and ℓ ⊆ ∂P V). Otherwise,
WLOG suppose Q1 ̸= P . The left side can then be rewritten as

d d
!
X P Qi P Q1 X
0= = 1+ (P, M ; Qi , Q1 )
i=1
Qi M Q1M i=2

and then
d
X d
X
(Φ(P ), Φ(M ); Φ(Qi ), Φ(Q1 )) = (P, M ; Qi , Q1 ) = −1.
i=2 i=2

Symmetrically, this can be rewritten as

d
d X 1
= .
Φ(P )Φ(M ) i=1
Φ(Qi )Φ(M )

Through (11.2.2), we find that ∂P V in the case of d = 2 is the polar of P with respect to V, so we define

Definition 11.2.4. The kth polar of a degree d curve V wrt. a point P is defined as

pkV (P ) := ∂Pd−k V = ∂P . . . ∂P V.
| {z }
d−k times

Proposition 11.2.5. For a degree d curve V and a point P , the following statements are equivalent:

• P ∈ pkV (P ) for one of 0 ≤ k ≤ d;

• P ∈ pkV (P ) for all of 0 ≤ k ≤ d.

(Note how this gives you traditional properties of polars in conics.)

Proof. We only need to prove that P ∈ pkV (P ) if and only if P ∈ pk−1

V (P ) and then we can win by induction.

Let F (k) be defined as the expression of pkV (P ) if we set P = [p : q : r]. Then the homogenized degree of F (k)
is just k, so

F (k−1) (P ) = () (P )

= () (P ) = (k · F (k) )(P ).

300
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Thus P ∈ pkV (P ) if and only if P ∈ pk−1

V (P ).

Since the zero locus of a degree-0 polynomial F is either the empty set ∅ (when F ̸= 0) or the whole
plane P2 (when F = 0), thus:

Corollary 11.2.6. For a degree d curve V and a point P ,


if P ∈
/ V,

∅
pdV =
P2

if P ∈ V.

Because there are only two cases, we can use ̸= 0 to represent the case where it is ∅ and = 0 for the case
where it is P2 . So we may write
pdV (P ) = 0 ⇐⇒ P ∈ V.

Finally, we arrive at

Corollary 11.2.7 (Polar duality; generalization of La Hire). For a degree d curve V, any two points P, Q
and an integer 0 ≤ k ≤ d,
P ∈ pkV (Q) ⇐⇒ Q ∈ pkV (P )

Proof. Both sides are equivalent to ∂Pk ∂Q

k
V = ∂Pd−k ∂Q
k
(V) = 0.

When n = 3, if a line ℓ intersects a cubic curve K at three points P, Q, R, then ℓ and p1K (P ) meet again
at M (distinct from P ), which satisfies

3 1 1 1 2 1 1
= + + ⇐⇒ = +
PM PM QM RM PM QM RM
⇐⇒ (P, M ; Q, R) = −1
2 1 1
⇐⇒ = +
QR PR MR

Thus, R ∈ ∂Q ∂P K. Equivalently, we obtain:

Proposition 11.2.8. Three collinear points P, Q, R on a cubic curve K satisfy

∂P ∂Q ∂R K = 0.

Definition 11.2.9. Given a cubic curve K. For a variable point P on some line ℓ,

301
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

• p2K (P ) passes through four fixed points S1 , S2 , S3 , S4 . The set {S1 , S2 , S3 , S4 } will be called the pole of
ℓ w.r.t K, denoted p0 K(ℓ).

• the envelope of its polar line p2K (P ) is a conic. We call this the poloconic of ℓ w.r.t K, denoted p2 K(ℓ).

This is the poloconic of a line; don’t get this mixed up with the poloconic of a point. Of course, we must
have

Proposition 11.2.10. Given a cubic curve K, then for any line K and any point P ,

P ∈ p0K (K) ⇐⇒ K = p1K (P )

Proposition 11.2.11. Given a cubic K, let K be a point on K, then there exists an isoconjugation ϕ on
some triangle △ABC such that {p2K (P ) | P ∈ K} is the pencil of diagonal conics of ϕ. Also p2K (K) = ϕ(K).

Proposition 11.2.12. Given a cubic K, let K be an arbitrary line, then p2K (K) is the nine-point conic of K
wrt. p0K (K).

11.2.1 Triangles as Degenerate Cubics

Let △ = △ABC. Then we can also view △ as a cubic curve in the barycentric coordinate system, i.e.,
xyz = 0.

Proposition 11.2.13. For any point P :

(i) p1△ (P ) is the trilinear polar of P wrt. △, t(P );

(ii) p2△ (P ) is the circumconic of △ with perspector P , c(P ).

Proof. (i) Let Pa = AP ∩ BC; let Q be the intersection of AP with the polar line p1△ (P ). Then Q divides Q
and p2△ (P ) into thirds. Therefore,

3 2 1 QP QP QA QPa
= + =⇒ 2· + =3 =⇒ 2· + = 0 =⇒ (Q, P ; Pa , A) = −2.
QP AP Pa P AP Pa P AP Pa P

Let Pb = BP ∩ CA, and let Pa∨ , Pb∨ be the intersections of t(P ) with BC and CA, respectively. Then we
have:

(AP ∩ t(P ), P ; Pa , A) = Pa∨ (Pb∨ , Pb ; C, A), ·Pb (Pa∨ , P ; Pa , A) = −1 · (1 − (Pa∨ , Pa ; B, C)) = −2.

Thus, by symmetry, p1△ (P ) = t(P ).

302
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

(ii) Let M be the second intersection of AP with p2△ (P ). Similar to the previous calculation, we have:

3 2 1
= + =⇒ (P, M ; Pa , A) = −2.
PM AM Pa M

Let PA be the second intersection of AP with c(P ), and let P a be the intersection of the tangents to c(P )
at B and C (in other words, the vertex opposite to A in the anticevian triangle of P ). Then we also have:

(P, PA ; Pa , A) = (P, P a ; Pa , A) · (P a , PA ; Pa , A) = −1 · (1 − (P a , Pa ; PA , A)) = −2.

Thus, by symmetry, p2△ (P ) = c(P ).

From this, we can rewrite (12.3.23) as:

Proposition 11.2.14. Let S = P ∗ Q be the cevapoint of P and Q. Then the following holds:

∂P ∂Q △ = ∂S2 △.

11.2.2 The Dragon’s Defeat

Let K be an isoptic cubic. We know that K is also a circular cubic. For a point P ∈ K, denote P ∗ as the
isogonal conjugate of P (also in K).

Definition 11.2.15. For a point P , we define △P = ∂P ∂P ∗ , the P -vanishing point VP as the third intersection
of P P ∗ with K.

Since the isogonal conjugate of the Miquel point is the point at infinity along the Newton line, we have
VM = V∞τ and we call this point the K-vanishing point of the isoptic cubic.

Proposition 11.2.16. For any two pairs of isogonal conjugates (Q, Q∗ ), (R, R∗ ), let P = QR∗ ∩ Q∗ R, P ∗ =
QR ∩ Q∗ R∗ . Then △P K, QQ∗ , RR∗ are concurrent.

Proof. Let CP = ∂P K, S = pCP (Q∗ R∗ ), S ∗ = pCP (QR). From (T ODO11.2.8), we have that Q, R∗ are conju-
gates in conic CP , and thus we have pCP (Q) = R∗ S ∗ . By the same logic we have pCP (R) = S ∗ Q∗ , pCP (Q∗ ) =
RS, pCP (R∗ ) = SQ. Thus the polar of △QRS is just △Q∗ R∗ S ∗ . Combining this with Problem 2 in (TODO
7.1) (or just DDIT) we have that QQ∗ , RR∗ , SS ∗ are concurrent. Further we know that

pCP (P ∗ ) = pCP (QR ∩ Q∗ R∗ ) = SS ∗ .

303
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Corollary 11.2.17. For a point P ∈ K\I, J, we have △P K = VP ∞⊥P P ∗ . For I, J, we have that △I,J K = τ.

Proof. From (T ODO11.2.8), VP already lies on △P K. Therefore we only need to prove S := QQ∗ ∩ RR∗ ’s
foot (call it V ) onto P P ∗ is just VP . Since (Q, Q∗ ; S, QQ∗ ∩ P P ∗ ) = −1 and SV ⊥ P P ∗ , we have

∡(V Q) + ∡(V Q∗ ) = 2 · ∡(P P ∗ ) = ∡(V P ) + ∡(V P ∗ ),

so V has to lie on K. For the two circle points, we can just prove for all Q, S = QQ∗ ∩ RR∗ lies on τ . This is
trivial, since from
(Q, Q∗ ; IJ ∩ QQ∗ , S) = −1,

S is the midpoint of QQ∗ , and thus lies on τ .

Corollary 11.2.18. Borrowing notation from (T ODO11.2.16), we have that △P K, △Q K, △R K concur at

the orthocenter of the diagonal triangle δQ of the complete quadrilateral Q = △P QR ∪ P ∗ Q∗ R∗ .

If the quadrilateral isn’t tangential, then the isoptic cubic isn’t singular (otherwise it has a double-point
at the incenter).

Proposition 11.2.19. Let P be a moving point on K, then VP + 2P (addition in the group law of K) is
constant. Thus,

(K, M ) −−−−−→ (K, VK )

P 7−−−−−−−→ VP

is a homomorphism.

Proof. Since for any two pairs of isogonal conjugates (P, P ∗ ), (Q, Q∗ ) we have that P + Q∗ = P ∗ + Q, so

VP + 2P = (VP + P + P ∗ ) + (P − P ∗ )

= (VQ + Q + Q∗ ) + (Q − Q∗ ) = VQ + 2Q.

This tells us that for any point V0 on the isoptic cubic, there exists 4 = 22 points P1 , P1∗ , P2 , P2∗ such that
V0 = VP1 = VP2 . In reality, they are just the intersections of the lines ℓ1 , ℓ2 through V0 such that

2∡(ℓ1 ) = 2∡(ℓ2 ) = ∡(V0 M ) + ∡(τ )

304
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

with the isoptic cubic. Since ℓ1 = P1 P1∗ ⊥ ℓ2 = P2 P2∗ , we have

Proposition 11.2.20. The line △P1 K’s three intersection points with the isoptic cubic is just V0 , P2 , P2∗ .

Proposition 11.2.21. Let V0 ’s isogonal conjugate be V0∗ , then the poloconic p2K (V0∗ ) of V0∗ wrt. K is just
(V0∗ P P ∗ QQ∗ ).

Proof. We want to prove p2K (V0∗ ) = (V0∗ P P ∗ QQ∗ ), from symmetry we only need to prove P ∈ p2K (V0∗ ). By
duality, we have that this is equivalent to V0∗ = VP∗ ∈ p1K (P ). Since P, P ∗ , VP are collinear and

P + P + VP∗ = P + P ∗ + VP ,

VP∗ lies on P ’s tangent to K, p1K (P ).

Proposition 11.2.22. For a point V0 ∈ K, let S1 , S2 respectively be the second intersections of the previously
defined ℓ1 , ℓ2 with the polar conic ∂V0 K. Then △V0 K = S1 S2 . Let N be the pole of V0∗ ∞⊥V0 V0∗ wrt. p2K (V0 ),
(which is also the midpoint of S1 S2 ). Then V0 N is perpendicular to p1K (V0 ).

Proof. too long

Practice Problems

Problem 1. Let τ be the Newton line of a complete quadrilateral Q. Prove that p2K(Q) (∞τ ) is a rectangular
hyperbola.

11.3 Self-isogonal Isocubics

Let’s address the most important cubics for all triangle geometry.

Definition 11.3.1. For a fixed isoconjugation ϕ, we call a cubic K an isocubic if ϕ(K) = K.

To prove these cubics exist, recall the barycentric definition of isoconjugations. Suppose
 
p q r
ϕ[x : y : z] = : : ,
x y z

You (as in: Wolfram-Alpha) can prove that there are two ways for a cubic K to satisfy being self-isoconjugate:

ux(ry 2 − qz 2 ) + vy(pz 2 − rx2 ) + wz(qx2 − py 2 ) = 0 (11.1)

ux(ry 2 + qz 2 ) + vy(pz 2 + rx2 ) + wz(qx2 + py 2 ) + kxyz = 0. (11.2)

305
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

We will address only the first equation right now.

Definition 11.3.2. For △ABC and an isoconjugation ϕ, then for any point P , we can define a special
isocubic with P as pivot, defined as

Kϕp (P ) = {Q | P, Q, ϕ(Q) collinear.}.

Then we call Kϕp (P ) a pivotal cubic.

Let P = [u : v : w]. Then obviously, we have:

Proposition 11.3.3. For a point P , we have that A, B, C, P, ϕ(P ) and the four fixed points S, S a , S b , S c of
the isoconjugation ϕ (S a , S b , S c are just anticevian triangle vertices of S) all lie on Kϕp .

Definition 11.3.4. Let ϕ be an isoconjugation in △ABC, let E be the Euler line of △ABC, and let
G, O, N, L ∈ E be the centroid, circumcenter, nine-point center, and de Longchamps point of △ABC. We
define the five cubics
Kϕp (∞E ), Kϕp (G), Kϕp (O), Kϕp (N ), Kϕp (L)

respectively as the Neuberg, Thomson, McCay, Napoleon-Feuerbach, Darboux cubics. (These are K001–
K005 in the Catalogue of Triangle Cubics).

(So the weird property given by Fontené III is actually just a property of the McCay cubic.)

Proposition 11.3.5. For a cubic K and four points on it, A, B, C, P, if we define a group law such that
2 · A = 2 · B = 2 · C = 2 · P, then K is an isopivotal cubic on △ABC with P as pivot.

Proof. Consider the third intersection map: the map iP : K → K, sending points Q ∈ K to the third
intersection of P Q with K. By (TODO 11.1.7), the transformation ϕA : AQ → AiP (Q) is a well-defined
projective involution. (TODO 11.1.9) tells us that the 2-torsion points,

(K, P )[2] := {Q ∈ K | 2 · Q = O} ∼
= (Z/2Z)2 ,

so since A, B, C, P are all 2-torsion, it’s isomorphic to the Klein 4-group. Thus, by properties of this group,
we have
C + A = B + P, A + B = C + P,

(which means that iP (B), iP (C) respectively lie on CA, AB.) This tells us that ϕA is a projective involution
sending AB to AC. Define ϕB , ϕC as BQ → BiP (Q), CQ → CiP (Q), then by the same logic, these are also
projective involutions exchanging their respective sides. Thus we have that iP (Q) has to be an isoconjugation!
Since by definition we have P, Q, iP (Q) collinear, K is a isopivotal cubic with pivot at P .

306
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

̸ 0, define these cubics as Kϕn,k (P ). Note that we

Now let’s consider the cases in (TODO 11.3.1) where k =
can rewrite the bary formula as

x  
y z p q r p q r k
+ + + + = 2
+ 2+ 2− ,
u v w ux vy wz u v w uvw

x y z
Also, u + v + w = 0 is the trilinear polar (see: (TODO 1.5)) of P = [u : v : w] wrt. △ABC, and
p w r
ux + vy + wz = 0 is the formula for circumconics of △ABC with perspector ϕ(P ) c(ϕ(P )) = ϕ(t(P )), so we
can define Kϕn,k (P ) as the locus of the barycentric equations of t(P ) · ϕ(t(P )) equals a constant.

Proposition 11.3.6. For a point P and its trilinear polar t(P ), its intersections with the three sides at
Pa∨ , Pb∨ , Pc∨ all lie on some Kϕn,k .

We call P the root of this cubic (note that P most likely does not lie on the cubic).

Theorem 11.3.7. For an arbitrary isocubic Kϕn,k (P ), there exists a conic C := Cϕn,k (P ) such that Q ∈ Kϕn,k
if and only if Q, ϕ(Q) are conjugate in C.

Translator’s note: The author did not provide a proof for this statement. The given one is purely
computational and terrible.

Proof. Recall from section (TODO 6.2.1) that a conic C has the equation xT MC x = 0 for a unique symmetric
matrix MC . We will also use that two points X, Y are conjugate in C iff their coordinates x, y satisfy
xT MC y = 0, which will go unproven. (For a quick sketch, La Hire’s is obviously true for this definition and
the polar of a point on the conic is then a linearization of the conic, i.e. the tangent.)

If we express the equation for Kϕn,k as above and divide by xyz on both sides, we get

   
r q  p r q p
u y· +z· +v z· +x· +w x· +y· + k = 0.
z y x z y x

Then Q = [x : y : z] and ϕ(Q) = [p/x : q/y : r/z] lying on Kϕn,k are conjugate in a conic C when

 
C1 w v
 
MC =  w C2 u
 
 
v u C3

for C1 , C2 , C3 such that C1 p + C2 q + C3 r = k.

Note that this C isn’t unique (by a DoF argument), so we can actually just set C to be a circle Γ (possibly
a imaginary circle or a line), which is also equivalent to all of (Qϕ(Q)) having a common radical center (at

307
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

the center of Γ). So let’s redefine our cubic Kϕn,k (P ) as Kϕn (Γ). However Γ isn’t always unique, for instance if
⊙(Qϕ(Q)) were all coaxial. So in reality we have

Proposition 11.3.8. pC (A), pC (B), pC (C) respectively intersect BC, CA, AB at Pa∨ , Pb∨ , Pc∨ .

This is because A, B, C ∈ Kϕn,k . Note that by varying the constant k in the barycentric representation
of K, we can produce a whole s*tload of isocubics passing through A, B, C, D, E, F . A easy example of
these nonpivotal cubics is the isoptic cubic (on “quadrilateral” △ABC ∪ ℓ, since it’s invariant under isogonal
conjugation but has no center. (Additionally, any circular pivotal cubic must have its pivot at the line at
infinity, which is obviously impossible for the isoptic cubic). Here, D, E, F are collinear on ℓ, and we have
that for two isogonal conjugates P, P ∗ , that the center of (P P ∗ ) lies on the Newton line τ , so the radical axis
is just ∞⊥τ , and Γ = τ . Similar to the isoptic cubic, we have:

Proposition 11.3.9. Given △ABC, an isoconjugation ϕ and a circle Γ, if P, Q are two points on Kϕn (Γ),
then R = P Q ∩ ϕ(P )ϕ(Q) ∈ Kϕn (Γ).

Proof. From (TODO 7.4.5) we know that ϕ(R) = P ϕ(Q) ∩ ϕ(P )Q, and therefore we have the three circles
ωP := (P ϕ(P )), ωQ := (Qϕ(Q)), ωR := (Rϕ(R)) are coaxal (Steiner line), and the center of Γ lies on the
radical axis of ωP , ωQ , so it also lies on the radical axis of ωP , ωR , so R ∈ Kϕn (Γ).

When ϕ is isogonal conjugation, things become nicer. We first look at a elementary theorem:

Theorem 11.3.10. Given a fixed △ABC, let (P, P ∗ ), (Q, Q∗ ) be a pair of isogonal conjugates in △ABC,
and let ωP = (P P ∗ ), ωQ = (QQ∗ ), let ΩP , ΩQ respectively be the pedal circles of P, Q wrt. △ABC. Then
the radical axis of ωP , ωQ is the image of the radical axis of ΩP , ΩQ under a 2 homothety at H.

Corollary 11.3.11. Given △ABC, and (P, P ∗ ) and (Q, Q∗ ) as a pair of isogonal conjugates, let R =
P Q ∩ P ∗ Q∗ , then P, Q, R’s pedal circles wrt. △ABC are coaxal.

Let’s reinterpret this with Kϕn,k .

Proposition 11.3.12. Given △ABC and a circle Γ ̸⊃ L∞ , let ϕ be an isoconjugation in △ABC, let Γ′ be
the 1/2 homothety of Γ at H. Then P ∈ Kϕn (Γ) if and only if P ’s pedal circle is orthogonal with Γ′ .

However from a while ago (Chapter 8?) we actually found a way to calculate the angle between a point’s
pedal circle and the nine-point circle, so:

Corollary 11.3.13. For an arbitrary △ABC and isoconjugation ϕ, then for a point P , the following
statements are equivalent:

• P ∈ Kϕn ((ABC)),

308
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

• P ’s pedal circle wrt. △ABC is orthogonal to the nine-point circle of △ABC,

• ∡(AP, BC) + ∡(BP, CA) + ∡(CP, AB) = 0◦ .

We call this isocubic Kϕn ((ABC)) the Kjp cubic of △ABC (it is K024 in the Catalogue of Triangle Cubics).
The locus of points satisfying ∡(AP, BC) + ∡(BP, CA) + ∡(CP, AB) = 90◦ is the McCay cubic. In reality,
the locus of points satisfying ∡(AP, BC) + ∡(BP, CA) + ∡(CP, AB) = θ is a cubic Kθ , and we also have
ϕ(Kθ ) = K−θ .

Practice Problems

Problem 1. For a given △ABC and a angle θ, prove that the locus of points satisfying ∡BAP + ∡ACP +
∡CBP = θ is a cubic.

11.4 Liang-Zelich

Can this theorem really allow for interstellar travel? Let’s see.

Definition 11.4.1. Given △ABC with orthocenter H and circumcenter O, let P, Q be a pair of isogonal
conjugates. Define
TO
t(P ) = t(P, △ABC) =
TH

where T = P Q ∩ OH.

There are a bunch of points for which we cannot define t: these are exactly A, B, C, I x , O, H, where I x
are either the incenter or an excenter. We will put these in a set Z, and for the subsequent proposition we’ll
assume that P and its isogonal conjugate both do not belong in Z. Then, we have that

Proposition 11.4.2. If Q is the isogonal conjugate of P in △ABC, then t(P ) = t(Q).

If ϕ denotes isoconjugation in △ABC, then for each t = t0 , we know that the loci of t(P ) = t0 is Kϕp (T ),
where T is the point on the Euler line OH satisfiying t(T ) = t0 .

Definition 11.4.3 (Generalized Euler line). Given △ABC, a point P and a constant x, define the (x, P )-
Euler line as the line connecting HhP,1/x (O), where hP,1/x is the homothety centered at P with factor
1/x.

Theorem 11.4.4 (Liang-Zelich). Given △ABC, a constant t0 ̸= 0, ∞ and any point P , let Pa , Pb , Pc be the re-
flection of P about BC, CA, AB respectively. Let Oa , Ob , Oc be the circumcenters of △BP C, △CP A, △AP B
respectively. Then the following are equivalent:

309
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

(i) t(P ) = t0 .

(ii) △Pa Pb Pc dilated from P with a factor of t0 is perspective with △ABC.

(iii) △Oa Ob Oc (also known as the Carnot triangle of △ABC) dilated from P with a factor of t−1
0 is
perspective with △ABC.

(iv) The (t0 , P )-Euler lines of △ABC, △BP C, △CP A, △AP B are concurrent.

When t0 = 0, (i), (iii) and (iv) are equivalent. When t0 = ∞, (i) and (ii) are equivalent.

Let’s postpone the proof for now. We can deduce the following facts:

Proposition 11.4.5. Given △ABC and a point P , then for the following triangles XY Z, t(P, △XY Z) =
t(P, △ABC):

(i) P is the Carnot triangle of △ABC.

(ii) P is the pedal triangle of △ABC.

(iii) P is the anti-pedal triangle of △ABC.

(iv) P is the circumcevian triangle.

Proof. We’ll prove them in order:

(i) Let t0 = t(P, △ABC). Recall that from (TODO 11.4.4) △Oa Ob Oc dilated by a factor of 1/t0 from P is
perspective with P , so P reflected about the three sides of △Oa Ob Oc is just ABC. Thus the isogonal
conjugate of P in △Oa Ob Oc is just O.

(ii) Let Q be the isogonal conjugate of P wrt. △ABC, let △Oa′ Ob′ Oc′ be the Carnot triangle of Q wrt.
+
△ABC, let △Pa Pb Pc be the pedal triangle of P wrt. △ABC. Since △Pa Pb Pc ∪ P ∼ △Oa′ Ob′ Oc′ ∪ O,
so from (i),

t(P, △Pa Pb Pc ) = t(O, △Oa′ Ob′ Oc′ ) = t(Q, △Oa′ Ob′ Oc′ )

= t(Q, △ABC) = t(P, △ABC)

(iii) Trivial from part (ii).

(iv) Let △DEF be the circumcevian triangle of P wrt △ABC, let △Pa Pb Pc be the pedal triangle of P wrt.
△ABC, let P ∗ be the isogonal conjugate of P in triangle △Pa Pb Pc , then by angle-chasing we have
+
△DEF ∪ P ∼ △Pa Pb Pc ∪ P ∗ , so from part (ii) we have

t(P, △DEF ) = t(P ∗ , △Pa Pb Pc ) = t(P, △Pa Pb Pc ) = t(P, △ABC).

310
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Remark. Normally people call just part (ii) the Liang-Zelich theorem, however this is actually just a small
part of it. Also, for circumcevian triangle we can just invert; inversion at P preserves the value of t.

This is the main theorem, but in its proof we will use some very interesting sub-theorems. Let’s see an
important one:

Theorem 11.4.6 (Strong Sondat’s Theorem). Given a line L, for two triangles △A1 B1 C1 , △A2 B2 C2 that
don’t have L as their perspectrix, the locus of P satisfying A2 (L∩A1 P ), B2 (L∩B1 P ), C2 (L∩C1 P ) concurrent
is the union of a circumconic of △A1 B1 C1 and line L.

Proof. MMP Let X1 = B1 (A2 B2 ∩ L) ∩ C1 (A2 C2 ∩ L), and cyclically define Y1 , Z1 . Since

B1 Y1 ∩ C1 Z1 , C1 X1 ∩ A1 Z1 , AY ∩ BX ∈ L,

we have by converse Pascal (TODO 6.3.1) that A1 , B1 , C1 , X1 , Y1 , Z1 lie on one conic. Let this conic be C1 .
Define the same things on the other triangle and get C2 , X2 , Y2 , Z2 .

By Pascal on A1 A1 C1 X1 B1 Z1 , we can get that if

A∗1 = TA1 C1 ∩ B1 X1 , C1∗ = A1 C1 ∩ B1 Z1 ,

then △A∗1 B1 C1∗ and △A2 B2 C2 have perspectrix as L.

Consider the projective transformation ϕ such that

A∗1 7→ A2 , B1 7→ B2 , C1∗ 7→ C2 .

From A1 C1∗ ∩ X2 C2 , A1 B1 ∩ X2 B2 ∈ L, we know that ϕ(A1 ) = X2 , so

ϕ(TA1 C1 ) = ϕ(A1 A∗1 ) = A2 X2 .

Let P be a point on C1 , let QA = A1 P ∩ L, and similarily define QB , QC . Then

A1 (P, X1 ; Y1 , Z1 ) = A2 (QA , A2 ; B2 , C2 ).

Thus
A2 (QA , A2 ; B2 , C2 ) = B2 (QB , A2 ; B2 , C2 ) = C2 (QC , A2 ; B2 , C2 ).

So A2 QA , B2 QB , C2 QC concur at Q ∈ C2 .

311
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Now we prove the converse. Let’s prove that all P satisfying the statement must lie on C1 ∪ L. We proceed
by MMP. Consider a point P ′ ∈ AP, and using the above definitions, define Q′A , Q′B , Q′C , then B2 Q′B 7→ C2 Q′C
is a projective map. Since △A1 B1 C1 and △A2 B2 C2 do not have L as their perspectrix, B2 Q′B ∩ C2 Q′C moves
on a non-degenerate conic that intersects A2 Q′A = A2 QA at two points. These two points are formed when we
choose a P ′ such that A2 Q′A , B2 Q′B , C2 Q′C are concurrent. When P ′ = QA , obviously A2 Q′A , B2 Q′B , C2 Q′C
concur at QA , so we have checked enough cases and we have that when P ′ ̸= QA , P ′ has to be P.

When we set L as the line at infinity, this theorem just becomes

Corollary 11.4.7. For two non-homothetic triangles △A1 B1 C1 , △A2 B2 C2 , the locus of points P such that
the line through A2 parallel to A1 P, the line through B2 parallel to B1 P , the line through C2 parallel to
C1 P concur is the union of a circumconic of △A1 B1 C1 and the line at infinity.

Directly this gives us:

Corollary 11.4.8 (Sondat’s Theorem). If △A1 B1 C1 , △A2 B2 C2 are orthologic and perspective, then their
perspector lies on the line connecting the two centers of orthology.

Proof. If △A1 B1 C1 , △A2 B2 C2 are homothetic, then this is obvious, so let’s assume they’re not homothetic.

Let Q be the perspector of △A1 B1 C1 and △A2 B2 C2 , and let P1 be the center of orthology of △A1 B1 C1
and △A2 B2 C2 (such that A1 P1 ⊥ B2 C2 , etc). Similarily define P2 .

By Strong Sondat’s, let Ci ∪ L∞ be the locus of points P such that A3−i ∞Ai P , B3−i ∞Bi P , C3−i ∞Ci P
concur. Let Q = C∞ ∩ C∈ . Then by the construction in the proof of Strong Sondat’s theorem, we get that
D := B1 ∞A2 B2 ∩ C1 ∞C2 A2 ∈ C1 , and also D is the orthocenter of △P1 B1 C1 . So from A1 D ∥ TA2 C2 we know

Q(A1 , B1 ; C1 , P1 ) = D(A1 , B1 ; C1 , P1 ) = A2 (A2 , B2 ; C2 , P2 ) = Q(A2 , B2 ; C2 , P2 ).

Thus QP1 = QP2 , or Q ∈ P1 P2 .

Corollary 11.4.9. Let P1 , P2 be the two centers of orthology, then P1 P2 is perpendicular to the perspectrix
of △A1 B1 C1 and △A2 B2 C2 .

Let’s return to points that lie on the pivotal isogonal cubic Kt0 := Kϕ (T ). From (TODO 11.3.3) we have
that T, ϕ(T ), the incenter and the three excenters all lie on this pivotal isogonal cubic. However, we can get
some more free points, by the Liang-Zelich theorem!

Proposition 11.4.10. Given △ABC and a constant t0 , let △DEF be the orthic triangle of triangle △ABC.
Choose a point Ha on AD such that AHa /DHa = 2t0 , similarily define Hb , Hc . Then Ha , Hb , Hc ∈ Kt0 .

312
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Proof. Let Ha ’s reflections across BC, CA, AB be Haa , Hab , Hac . Then we know that the image of Haa
under a homothety by t0 from Ha is A, so the image of △Haa Hab Hac under a homothety of t0 from Ha is
perspective to △ABC. Then from part (ii) of Liang-Zelich, Ha ∈ Kt0 . Similarily we have Hb , Hc ∈ Kt0 .

If you want to use the above two proofs in part (iii) of Liang-Zelich, you will discover a point T , where
the other two points don’t necessarily exist.

Proposition 11.4.11. Let Kϕp (L) be the Darboux cubic of △ABC, then Kϕp (L) is symmetric across O.

Proof. Note that t(L) = 1/2, then from the equivalency of parts (i) and (ii) in (TODO 11.4.4), a point
P ∈ Kϕp (L) if and only if P ’s pedal triangle wrt. △ABC is perspective to △ABC. This lets us redefine the
Darboux cubic!

Let P ′ be the reflection of P across O, for P on the Darboux cubic. Let △Pa Pb Pc , △Pa′ Pb′ Pc′ respectively
be the pedal triangles of P, P ′ wrt. △ABC. Then their corresponding vertices are reflections over the
midpoints of their sides (i.e isotomic). Thus the two perspectors of these pedal triangles wrt. △ABC are
also isotomic conjugates (isoconjugate with G as pole), so P, P ′ both lie on Kϕp (L).

11.4.1 Proof

Most of this proof is taken from Liang and Zelich’s original paper at here. Note that the t0 of (i) is unique,
so to prove that (i) and the others are equivalent, we can actually just prove that in the condition of (i),
there exists a t0 such that it holds, and also that there exists a t0 such that (ii), (iii), (iv) hold.

If we let △Pa′ Pb′ Pc′ be the image of the homothety from P of △Pa Pb Pc , then the locus of CPc′ ∩ APa′ is a
conic CB through C, A, H, P , where H is the orthocenter of △ABC. Similarily, the locus of APa′ ∩ BPb′ is a
conic CC through A, B, H, P . Since CC , CB already have three intersection points A, H, P , they will have a
fourth (real) intersection point X. Here X is actually just the perspector of some △Pa′ Pb′ Pc′ with △ABC, so
we have proved existence for (ii).

The book doesn’t have proofs for (iii) and (iv) but (iii) is trivial and (iv) is just inversion I think

Theorem 11.4.12. Given a triangle △ABC and a point P , let HA , HB , HC be the orthocenters of
△BP C, △CP A, △AP B. Consider a triangle △DEF such that △DEF is orthologic and perspective to
△ABC, and consider the center of orthology P between △ABC wrt. △DEF (order of the two centers of
orthology is defined in the previous sections). Let P ′ be the other center of orthology. Then we have

• △DEF and △HA HB HC are orthologic and perspective.

• If we define J is the perspector of △DEF and △HA HB HC and H, H ′ as the orthocenters of △ABC
and △DEF , then we have J = HP ′ ∩ H ′ P lies on circumconic of △DEF , conic (DEF HP ′ ).

313
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Proof. it’s time to spam pascal For part (i), let X be the perspector of △ABC and DEF . Then (TODO
11.4.6) tells us that X lies on the circumrectangular hyperbola H = (ABCHP ), along with the points
HA , HB , HC ∈ H. Thus △AHB HC ’s orthocenter (let it be H ∗ ) lies on H and satisfies

HB H ∗ ⊥ HC A ⊥ BP, HC H ∗ ⊥ AHB ⊥ CP.

Therefore H ∗ is both the orthocenter of △HA BHC and also the center of orthology of △HA HB HC wrt.
△DEF .

Let K be the second intersection point of line HA D with H. By Pascal on the hexagon CXAHC KHA
we have that
CX ∩ HC K, XA ∩ KHA = D, AHC ∩ HA C = ∞⊥AB = ∞D E

are collinear, so E ∈ CX and E ∈ HC K. Similarily we have F ∈ HB K, so K is the perspector of △DEF

and △HA HB HC .

For part (ii), consider the hexagon JHAXP HA . By Pascal we get that

JH ∩ XP, HA ∩ P HA = ∞⊥BC , AX ∩ HA J = D

are collinear. From Sondat’s theorem((TODO 11.4.8)), we have that D∞⊥BC ∩ XP = P ′ , so we have
J ∈ HP ′ . Let V be the second intersection point of line HA ∞AX with hyperbola H. By Pascal on the
hexagon AXBHA V HB we have that

AX ∩ HA V = ∞AX , XB ∩ V HB , BHA ∩ HB A = ∞⊥CP

are collinear, and thus BX ∥ V HB . Similarily we can get that CX ∥ V HC , which just means that
HA ∞DX , HB ∞EX , HC ∞F X are concurrent. From (TODO 11.4.6) and the fact that △DEF, △HA HB HC
are orthologic, we can get that J lies on the rectangular hyperbola H′ = (DEF XP ′ ), since

P (A, HA ; HB , H, C) = P ′ (∞⊥EF , D; E, F )

= J(P ′ ∞⊥EF ∩ H′ , D; E, F )

= J(P ′ ∞⊥EF ∩ H′ , HA ; HB , HC ),

so we get the orthocenter of △P ′ EF HD = P ′ ∞⊥EF ∩ H′ lies on line AJ. We finish by Pascal on

JH ′ DXP ′ HD , to get that

JH ′ ∩ XP, H ′ D ∩ P ′ HD = ∞⊥EF , DX ∩ HD J = A

314
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

are collinear. Then from Sondat’s theorem, we have that A∞⊥EF ∩ XP ′ = P, so J ∈ H ′ P .

Now we have enough machinery to start the actual proof of Liang-Zelich.

First we prove (i) implies (iii). Let △Oa′ Ob′ Oc′ be the Carnot triangle of P wrt. △ABC under a t−1
0

homothety at P , let OP be the circumcenter of △Oa Ob Oc , let O′ , OP′ respectively also be the images of O, OP
under a t−1 ′ ′ ′ ′ ′
0 homothety at P . Since P Q ∥ OOP ∥ O OP , we only need to prove that △ABC and △Oa Ob Oc

are perspective, to get that O′ , H, OP′ are collinear. Let’s use (TODO 11.4.12) and choose △Oa′ Ob′ Oc′ as
our △DEF . We get that △HA HB HC and △Oa′ Ob′ Oc′ are perspective. Note that the center of orthology
of △Oa′ Ob′ Oc′ wrt. △ABC is O′ , and the center of orthology of △ABC wrt. △Oa′ Ob′ Oc′ is P , so thus
J = HO′ ∩ H ′ P ∈ C = (Oa′ Ob′ Oc′ O′ H ′ ) is the perspector of △HA HB HC and △Oa′ Ob′ Oc′ , where H ′ is the
orthocenter of △Oa′ Ob′ Oc′ .

Since (P, O′ ) are isogonal conjugates in △Oa′ Ob′ Oc′ (opposite centers of orthology), we have that H ′ P ∩O′ OP′
is the isogonal conjugate of H ′ O′ ∩ P OP′ by DDIT, and thus H ′ P ∩ O′ OP′ lies on C. Thus J is simultaneously
the second intersections of O′ H and O′ OP′ with C, so O′ , H, OP′ are collinear and (i) implies (iii). (We could
have also seen in the proof that O′ a = AU ∩ P Oa , Ob′ = BU ∩ P Ob , Oc′ = CU ∩ P Oc , where U = OP ∩ HQ.)

Now let’s prove that (i) and (iii) imply (iv). Let △Oa′ Ob′ Oc′ be defined similarily to the previous proof.
Setting △DEF in (TODO 11.4.12) to △Oa′ Ob′ Oc′ gets us that the (t0 , P ) Euler line of △BP C, △CP A, △AP B
(which are just lines Oa′ HA , Ob′ HB , Oc′ Hc ) concur at a point J, and J ∈ HP . Then from (i) we have that
HP is the (t0 , P ) Euler line of △ABC, so all four (t0 , P ) Euler lines concur.

Finally, let’s prove that (i) and (iii) imply (ii). We first prove that ΓA = (AP Oa′ ), ΓB = (BP Ob′ ), ΓC =
(CP Oc′ ) are coaxal. Let A′ = OP′ Oa′ ∩ AH, B ′ = OP′ Ob′ ∩ BH, C ′ = OP′ Oc′ ∩ CH, then

∡(P Oa′ ) − ∡(A′ Oa′ ) = ∡(P Oa ) − ∡(OP Oa )

(∡(P B) + ∡(P C) − ∡(AH)) − (∡(Oa Ob ) + ∡(Oa Oc ) − ∡(AP ))

= ∡(P A) − ∡(A′ A),

so A′ ∈ ΓA . Similarily we have B ′ ∈ ΓB , C ′ ∈ ΓC . From O′ Oa′ ∥ OOa ∥ AH and O′ , H, OP′ collinear, we have

OP′ A′ O′ H
′ ′
= ′P ′ .
OP Oa OP O

Calculating this ratio symmetrically we just get that △A′ B ′ C ′ and △Oa′ Ob′ Oc′ are homothetic with center
OP′ . This tells us that
OP′ Oa′ · OP′ A′ = OP′ Ob′ · OP′ B ′ = OP′ Oc′ · OP′ C ′ ,

so OP′ is the radical center of the three circles ΓA , ΓB , ΓC . Thus they are coaxal with common rad-ax P OP′ .

315
AoPS Chapter 11. Basic Cubic Theory with Liang and Zelich

Let △OA OB OC be the Carnot triangle of P wrt. △Oa′ Ob′ Oc′ , then △OA OB OC is also obviously the
Carnot triangle of P wrt. △Oa Ob Oc under a t0 homothety at P . Note that the centers of the circles
ΓA , ΓB , ΓC respectively are Ob Oc ∩ OB OC , Oc Oa ∩ OC OA , Oa Ob ∩ OA OB , therefore since ΓA , ΓB , ΓC are
coaxal and by Desargues’ theorem, we have that △Oa Ob Oc and △OA OB OC are perspective. This tells us
that t(P, △ABC) = t(P, △Oa Ob Oc ).

Let △Ka Kb Kc be the Carnot triangle of Q wrt. △ABC, then by the same logic we have t(Q, △ABC) =
+
t(Q, △Ka Kb Kc ). Since △Pa Pb Pc ∪ P ∼ △Ka Kb Kc ∪ O, we have

t(P, △Pa Pb Pc ) = t(O, △Ka Kb Kc ) = t(Q, △Ka Kb Kc )

= t(Q, △ABC) = t(P, △ABC).

Thus P ’s Carnot triangle wrt. △Pa Pb Pc is just △ABC, and △ABC under a homothety of t−1
0 at P is
perspective with △Pa Pb Pc . In other words, △Pa Pb Pc under a homothety of t0 at P is perspective with
△ABC.

Practice Problems

Problem 1. Given △ABC, let A1 be the reflection of A across BC, and similarily define B1 , C1 . Let
A2 = BC1 ∩ CB1 . Prove that A1 A2 is parallel to the Euler line of △ABC.

Problem 2. Given △ABC, let B ∗ , C ∗ respectively be the reflections of B, C across CA, AB. Let I b , I c be
the B, C excenters. Let P = I b C ∗ ∩ I c B ∗ , A′ , B ′ , C ′ respectively be the reflections of P across BC < CA, AB.
Prove that AA′ , BB ′ , CC ′ are concurrent.

Problem 3. Let I be the incenter and let N9 be the nine-point center. Let △XY Z be the incentral triangle
(cevian triangle of I), and let ℓ be the Euler line of △XY Z. Prove that N9 lies on ℓ.

Problem 4 (Taiwan 2017 TST 3 M3). For triangle △ABC with circumcircle Γ, let A′ be the antipode of A.
Construct equilateral triangle BCd, such that A, D lie on opposite sides of BC. Let the line perpendicular
to A′ D through A intersect CA, AB at E, F . Construct isosceles triangle with base EF and top angle 30◦
ET F , such that A, T lie on opposi

Problem 5 (USA TST 2016). Let acute triangle △ABC be non-isosceles and let P be a internal point. Let
A1 , B1 , C1 be the feet from P onto the sides. Find all points P such that AA1 , BB1 , CC1 are concurrent and

∡P AB + ∡P BC + ∡P CA = 90◦ .

316
Chapter 12

Special Lines

12.1 Conjugate Conics

Our goal for this section is to generalize orthotransversals.

Proposition 12.1.1. Given a △ABC, a point P ∈

/ {A, B, C} and angles α, β and γ such that

(α, β, γ) ̸= (∡(BC, AP ), ∡(CA, BP ), ∡(AB, CP )),

♯
there exists a circumconic CP,α,β,γ such that the following property holds:

For any point Q ∈

/ L∞ , construct points D ∈ BC, E ∈ CA, F ∈ AB such that

∡(QD, AP ) = α, ∡(QE, BP ) = β, ∡(QF, CP ) = γ.

♯
Then Q ∈ CP,α,β,γ if and only if D, E, and F lie on a line L.

Proof. Construct △XY Z such that A ∈ Y Z, B ∈ ZX, C ∈ XY such that

∡(Y Z, AP ) = α, ∡(ZX, BP ) = β, ∡(XY, CP ) = γ.

Additionally, construct point U such that BXCU is a parallelogram; construct V and W similarly. Note
that since BW ∩ CV, CU ∩ AW, AV ∩ BU ∈ L∞ , by the converse of Pascal’s theorem A, B, C, U , V , W lie
on a conic.
♯
Claim. We can choose CP,α,β,γ to be this conic.

317
AoPS Chapter 12. Special Lines
♯
Proof. Choose a Q ∈ CP,α,β,γ and construct points D ∈ BC, E ∈ CA, F ∈ AB such that

∡(QD, AP ) = α, ∡(QE, BP ) = β, ∡(QF, CP ) = γ

♯
so that QD ∥ Y Z, QE ∥ ZX, QF ∥ XY . Furthermore, construct point D′ on CP,α,β,γ such that Q, D, D′
are collinear; construct E ′ and F ′ similarly. By Pascal’s Theorem (TODO 6.3.1), E, F , BE ′ ∩ CF ′ are
collinear and similarly we have the other two collinearities, so we only need to prove that AD′ , BE ′ , CF ′ are
concurrent. We proceed by tethered MMP. We know that when Q ∈ {U, V, W }, the said proposition P holds
(all of the concurrency points lie on L∞ ). Since Q 7→ D′ , E ′ , F ′ is a projective map (TODO 7.2.15), P is
♯
purely projectively defined on CP,α,β,γ , and thus since we checked the three cases U, V, W , our result holds
♯
for all Q ∈ CP,α,β,γ .

more

Definition 12.1.2. Given a triangle △ABC, a point P and angles α, β and γ satisfying (TODO 12.1.1):

♯
• We call the above defined conic CP,α,β,γ the α, β, γ-conjugate conic of P wrt. △ABC.
♯
• If Q ∈ CP,α,β,γ , we notate the above defined line L as TQ (△ABC, P, α, β, γ) and we call it the
(P, α, β, γ)-transversal of Q wrt. △ABC.
♯ ♯
• If α = β = γ, we simplify this notation CP,α,β,γ to CP,α , and we call the above conic the α-conjugate
conic of △ABC wrt. P . We will also shorten the TQ (△ABC, P, α, α, α) transversal of a point Q on
♯
CP,α as simply TQ (△ABC, P, α).

• When α = β = γ = 90◦ , define OQ (△ABC, P ) = TQ (△ABC, P, 90◦ ) as the P -orthotransversal of Q

wrt. △ABC.

• When α = β = γ = 90◦ and P = Q, define O(△ABC, P ) = OP (△ABC, P ), which is just the

orthotransversal of P wrt. △ABC.
∼
From the proof of the construction in (TODO 12.1.1), we can see that if P is the (triangle) Miquel point
∼ ∼
♯
of △XY Z with X, Y, Z on the sides of △ABC, then CP,α,β,γ = C ∼ ∼ for an angle α = ∡(Y Z, AP ).
P ,α

In addition to conjugate conics, we can also dually define anticonjugate curves.

Definition 12.1.3. Given △ABC, a point Q, and three angles α, β, γ, we define

♭ ♯
CQ,α,β,γ := {P | Q ∈ CP,α,β,γ }

as the α, β, γ-anticonjugate curve of Q wrt. △ABC.

318
AoPS Chapter 12. Special Lines

Note that we haven’t found what these curves are yet. In fact, they’re also conics!

/ {A, B, C} and an angle α, where (Q, α) ̸= (H, 90◦ ) (H is

Proposition 12.1.4. Given △ABC, a point Q ∈
orthocenter), then
♭ ♭ ♯
CQ,α = CQ,α,α,α = CQ,−α .

Proof. LONG

Claim 12.1.5. THIS IS PROPERTY OF A SNOWBALL

Remark. In the proof we can see that if we choose X, Y, Z such that

∡V AY = ∡W AZ = α, ∡W BZ = ∡U BX = β, ∡U CX = ∡V CY = γ,

♭
and A, B, C, X, Y, Z are conconic (in other words, BZ ∩ CY, CX ∩ AZ, AY ∩ BX are collinear), then CQ,α,β,γ
is also a conic. However this is not sufficient to prove, as we haven’t used the definition as the preimage of C ♯
yet.

Let’s work on characterizing C ♯ and C ♭ .

Proposition 12.1.6. For a point P ∈

/ {A, B, C, H},

♯
• For any angle α, CP,α ♭
= CP,−α ;

• Let P
“ be the antigonal conjugate of P (see (TODO 8.1.16)), then

“} = ( C ♯ ) ∩ ( C ♭ P, α);
\ \
{A, B, C, P P,α
α α

• Let F be the pencil of conics through A, B, C, P

“, then the map

R◦⧸ ◦ → F
180
♯
α 7→ CP,α

from all angles α to the pencil of conics is a projective map.

Proof.

Corollary 12.1.7 (Conic Reim’s). THIS IS PROPERTY OF A SNOWBALL

Proposition 12.1.8.

Proof.

319
AoPS Chapter 12. Special Lines

Claim 12.1.9.

Proof.

Proposition 12.1.10.

Proof.

Proposition 12.1.11. THIS IS PROPERTY OF A SNOWBALL

Corollary 12.1.12.

Proposition 12.1.13.

Corollary 12.1.14.

Proposition 12.1.15.

Proof.

Corollary 12.1.16. THIS IS PROPERTY OF A SNOWBALL

12.1.1 Orthologic Conjugates

Definition 12.1.17.

Proposition 12.1.18. THIS IS PROPERTY OF A SNOWBALL

Proposition 12.1.19.

(i) If P ◦ ∈ BC ∪ CA ∪ AB, assume P ◦ ∈ BC, then the trilinear polar of P ◦ is the tangent T to BC. But
this implies BP ⊥ CH and CP ⊥ BH, which means P = A.

(ii) If A, P1 , P2 are collinear, then OH (△ABC, P1 ) and OH (△ABC, P2 ) intersect at D on BC. Therefore,
P1◦ , P2◦ satisfy
A(P1◦ , D; B, C) = −1 = A(P2◦ , D; B, C),

which means A, P1 , P2◦ are collinear.

Corollary 12.1.20.

320
AoPS Chapter 12. Special Lines

12.2 Zhang Zhihuan’s Permutation Line

Zhang Zhihuan, Prince of geo, uses the permutation line to nuke concyclicity in geo problems.
The following section was co-authored by him.

Proposition 12.2.1. Let △ABC have circumconic C and point P be an arbitrary point in the plane. Let
△PA PB PC be its C-cevian triangle and let X be a point in C. Define

XP,A = XPA ∩ BC, XP,B = XPB ∩ AC, XP,C = XPC ∩ AB,

then P, XP,A , XP,B , XP,C are collinear.

Proof. By Pascal’s on BCPC XPA A, we get that BC ∩ XPA = XP,A , CPC ∩ PA A = P, PC X ∩ BA = XP,C
line on a line.

By a similar Pascal’s on CAPA XPB B, the result follows.

Proposition 12.2.2. The map PerCP (X) : C → P XP,A from C to the pencil through P is a projective map.

Proof. Follows by considering the projective map X → XP,A → P XP,A .

Proposition 12.2.3. Take △ABC with circumconic C and let P be a point in the plane. Let C ′ be another
conic through A, B, C, P , and let T be the fourth intersection point of C ′ and C. Let X be a point on C; then
PerCP (X) intersects T X on C ′ .

A PB
PC
T
D
P XP,B

XP,C

XP,A
B X C

PA

Proof. Let D be the second intersection of PerCP (X) and C ′ . Since taking a permutation line is a projective
map, we have

T (A, B; C, X) = (A, B; C, X)C = (PerCP (A), PerCP (B); PerCP (C), PerCP (X))

= (AP, BP ; CP, PerCP (X)) = (A, B; C, D)T = T (A, B; C, D)

321
AoPS Chapter 12. Special Lines

As such, T, X, D are collinear as desired.

Proposition 12.2.4. Take △ABC with circumconic C, two points P, Q in the plane, and point X ∈ C.

Then Z = PerCP (X) ∩ PerCQ (X) lies on T = (ABCP Q). We will call Z LiCP,Q (X) (the P,Q-Li conjugate
of X).

Proof. Let PA = AP ∩C, QA = AQ∩C. This is effectively combining ?? for two values of P, Q, as PerCP (X)∩T
and PerCQ (X) ∩ T both lie on T X.

The fact that Z lies on T can also be proven by noting that

P
P (A, Z; B, C) = (AP ∩ BC, XPA ∩ BC; B, C) =A (A, X; B, C)C
QA
= (AQ ∩ BC, XQA ∩ BC; B, C) = Q(A, Z; B, C).

Thus we are done by the definition of a conic in (TODO 6.2.6).

Remark 12.2.5. Here’s a bad way to remember that LiCD lies on D. The Li ttle Dipper is a constellation, so
the point lies on the dipper or lower conic, wihch is D.

Proposition 12.2.6. Let T be the fourth intersection of two circumconics C, C ′ . Then for any X ∈ C, we
have that T, X, LiCC ′ collinear.

Remark 12.2.7. This gives us another lens to look at the fact that LiCC ′ (X) lies on T X – this simply follows
because PerCT (X) = T X!

We also have more projective maps:

Proposition 12.2.8. Let △ABC and some point X have circumconic C through A, B, C, X. Then the map
from another circumconic C ′ to the pencil of lines through the C, C ′ -Li conjugate

C ′ → T(LiCC ′ (X))

P 7→ PerCP (X)

is projective.

Proof. Follows since PerCP (X) = P LiCC ′ (X).

Example 12.2.9. Let △ABC have incenter I, circumcenter O, and intouch triangle △DEF . Let HA be
the orthocenter of △BIC. Let S = (AEF ) ∩ (ABC) and let T = (AIO) ∩ (ABC) other than A. Show that
T, HA , I, S are concyclic.

322
AoPS Chapter 12. Special Lines

HA

A
S
F

E I
T

O B
C
D
Y

Proof. Let HF e = (ABCHIHA ) be the Feuerbach hyperbola of △ABC.

Then it follows that PerI (S) = ID = IHA which implies that LiHF e (S) = HA .

Let U be the isogonal conjugate of ∞OI , which is also the fourth intersection of (ABC) and HF e . It then
follows that U, S, HA collinear by (TODO 12.2.5).

As such, it follows that

∡HA ST = ∡U ST = ∡U AT = ∡U AO + ∡OAT = ∡(HA I, OI) + ∡(OIT ) = ∡HA IT

as desired.

A bit of review on isoconjugations: for a line ℓ not passing through any of the vertices, we know that
a (point) isoconjugation wrt. △ABC sends ℓ to a circumconic bijectively(see (TODO 7.4.7)) (denote this
point isoconjugation as φ). Let φ(ℓ) represent this circumconic. When φ is isogonal conjugation, then when
ℓ = L∞ , φ(L∞ ) is the circumcircle.

Proposition 12.2.10. Let φ be a point isoconjugation on △ABC, and let F be its corresponding pencil of
conics (TODO 7.4.1). For a circumconic C and a point on this circumconic X, let another conic D = DX ∈ F

323
AoPS Chapter 12. Special Lines

satisfy that the polar of X wrt D, pD X = φ(C). Then for any point P , we have

PerCP (X) = pD φ(P ).

Proof. Let PA be the second intersection of AP with C, XA = BC ∩ XPA . We will prove that XA ∈ pD φ(P ).
Note that
φ(PA ) = ℓ ∩ pD PA = pD XPA ,

and therefore
pD XA = pDBC pD XPA = Aφ(PA ) = Aφ(P ),

and therefore XA ∈ pD φ(P )

12.3 The Barycentric Product

We have previously introduced the barycentric product (quotient) at the end of (TODO 7.4): given a triangle
△ = △ABC,

[x1 : y1 : z1 ] × [x2 : y2 : z2 ] := [x1 x2 : y1 y2 : z1 z2 ],

 
x1 y1 z1
[x1 : y1 : z1 ] ÷ [x2 : y2 : z2 ] := : :
x2 y2 z2

We know that the isogonal conjugate of a point P in ∆ is K × G ÷ P , where G and K are respectively the
centroid [1 : 1 : 1] and the symmedian point [a2 : b2 : c2 ]. Notice that there is no need to multiply by G again
despite all coordinates being 1; the reason we do this is that multiplying by G again makes the left and right
side “non-homogenous” (since barycentric coordinates can be arbitrarily scaled), and leads to notational
confusion, especially when we transform triangle △ABC. As another example, for a point P its isotomic
conjugate in ∆ is G × G ÷ P .

Remark (Clarification on weight). We define the weight of a point/expression as a way to formalize this
“homogeneity”, define it analogously to the degree of a polynomial. For example, the weight of G = [1 : 1 : 1]
is 1, but the weight of the fixed ratio [1 : 1] is 0. This is so if we scale the barycentric coordinate both
sides scale by the same amount. Barycentric quotient of two points returns a ratio (weight 0 point), and
barycentric product of two points returns a “isoconjugation” (weight 2 point). A projective transformation
acts on the plane (weight 1 points): ϕ[x : y : z] = [ux : vy : wz] for some [u : v : w] in P2 . So we can extend it
to any weight of points via
ϕ([x : y : z], k) := ([uk x : v k y : wk z], k)

324
AoPS Chapter 12. Special Lines

so that barycentric quotient and product are preserved. In particular, weight 0 points are preserved under
this transformation.

More precisely, we consider the set

P2 × Z = {([x : y : z], k) | [x : y : z] ∈ P2 , k ∈ Z}.

The multiplication and division above are defined as

([x1 : y1 : z1 ], k1 ) × ([x2 : y2 : z2 ], k2 ) := ([x1 x2 : y1 y2 : z1 z2 ], k1 + k2 ),

  
x1 y1 z1
([x1 : y1 : z1 ], k1 ) ÷ ([x2 : y2 : z2 ], k2 ) := : : , k1 − k2
x2 y2 z2

and a point P in the plane has weight 1, i.e. (P, 1).

Example 12.3.1. • An isoconjugation on points ϕ is a transformation of the form

  
u v w
([x : y : z], 1) 7−→ : : ,1 ,
x y z

so if we let ϕ = ([u : v : w], 2), then we have ϕ(P ) = ϕ ÷ P . Thus an isoconjugation ϕ has weight 2.

• A projective transformation Φ fixing A, B, and C is of the form

([x : y : z], 1) 7→ ([ux : vy : wz], 1),

so if we let Φ = ([u : v : w], 0), then Φ(P ) = Φ × P . We sometimes call a point with weight 0 a
proportion. In particular, the identity transformation is multiplication by ([1 : 1 : 1], 0), which we
abbreviate as 1.

In barycentric coordinates, the complement is

[u : v : w]∁ = [v + w : w + u : u + v],

and the anticomplement is

∁
[u : v : w] = [v + w − u : w + u − v : u + v − w].

In order for these operations to map points with weight 1 to points with weight 1, we assume for this purpose

325
AoPS Chapter 12. Special Lines

that both operations preserve weight, i.e.

([u : v : w], k)∁ = ([v + w : w + u : u + v], k),

∁
([u : v : w], k) = ([v + w − u : w + u − v : u + v − w], k).

Notice that these two transformations are associated with a choice of the centroid G (or equivalently the line
at infinity L∞ ) at weight 1, so in order to make them projectively invariant we should make them weight 0.
One of the most basic collinearities comes from complementing: G, P, P ∁ collinear.

Proposition 12.3.2. For any proportion r = ([u : v : w], 0),

∁
r × r∁ = r−1 .

Proof. You know how to do algebra, right?

If we take r = P ÷ G, we have

P × P ∁ = G2 × (G ÷ P )∁ = G × (G2 ÷ P )∁ .

Similarly, if we define

[u : v : w]D = [v − w : w − u : u − v] = [w − v : u − w : v − u],

D
then r × rD = r−1 .

This D operation actually has a geometric meaning: the point P D is the pole pSt (GP ) of GP in the Steiner
circumellipse St, called the difference, and is undefined when P = G (as the pole is undefined). Since G
∁
is the center of this ellipse, every P D is on the line at infinity. Since P, P ∁ , P , G are collinear, and every
∁ ∁
difference point lies on the lie at infinity, we have P D = P ∁D = P D∁ = P D = P D . In other words, P D is the
∁
inverse of P ∁ and also the inverse of P .

Definition 12.3.3. For any point X, let △Xa Xb Xc and △X a X b X c be the cevian and anticevian triangles,
respectively. We define the X-complement transformation (−)∁X as the homography that sends A, B, C, X
∁
to Xa , Xb , Xc , X respectively, and the X-anticomplement transformation (−) X
as the homography that
sends A, B, C, X to X a , X b , X c , X respectively.

In particular, the G-complement and G-anticomplement transformations are the ordinary complement
and anticomplement, respectively.

326
AoPS Chapter 12. Special Lines

Similarly, let c(X) be the circumconic with X as the perspector wrt. ABC. Then we define the X-difference
transformation (−)DX of a point P as the pole of P X in c(X).

Note that X, △Xa Xb Xc , △X a X b X c , and c(X) play the role of G, △Ga Gb Gc , △Ga Gb Gc , and St = c(G)
under the projective transformation Φ = X ÷ G, so

(−)∁X = Φ ◦ (−)∁ ◦ Φ−1 ,

∁ ∁
(−) X
= Φ ◦ (−) ◦ Φ−1 ,

(−)DX = Φ ◦ (−)D ◦ Φ−1 .

and thus we get:

Proposition 12.3.4. Fix a point X. For any point P ,

P ∁X = X × (P ÷ X)∁ ,
∁ ∁
P X
= X × (P ÷ X) ,

P DX = X × (P ÷ X)D .

More generally, for a weight k object P,

P∁X = X k × (P ÷ X k )∁ ,
∁ ∁
P X
= X k × (P ÷ X k ) ,

PDX = X k × (P ÷ X k )D .

By (TODO 7.4.6), we have

Proposition 12.3.5. Given four points P, P ∗ , Q, Q∗ such that P × P ∗ = Q × Q∗ , P, Q, P Q∗ ∩ P ∗ Q all lie

on a common circumconic of △.

Example 12.3.6. Let G = X2 , K = X6 , Ge = X7 , M t = X9 , Ge∗ = X55 , and I ′ = X75 . (These are the
centroid, symmedian point, Gergonne point, Mittenpunkt, insimilicenter of the circumcircle and incircle, and
the isotomic conjugate of the incenter respectively). Prove: GGe∗ , KGe, and M tI ′ are concurrent.

Solution. Since I, G, and N a are collinear (Nagel line), by taking the isotomic conjugate I ′ , G, and Ge lie
on a circumconic C of △. Since G × K = Ge × Ge∗ , we get that GGe∗ and KGe intersect at a point on C. It

327
AoPS Chapter 12. Special Lines

thus suffices to show that K × I ′ = Ge × M t, which follows from

K × I ′ = K × (G2 ÷ I) = (K × G ÷ I) × G = I × G

= (G ÷ Ge)∁ × G2 = (Ge ÷ G) × G × (Ge ÷ G)∁ × G = Ge × M t.

Let us recall the definition of the crosspoint, (TODO 7.1.17):

Definition 12.3.7. For any two points P and Q, the crosspoint P ⋔ Q is the pole of line P Q in the conic
(△P Q).

Proposition 12.3.8. Let R = P ⋔ Q be the crosspoint of P and Q in △. Then

R = P × (P ÷ Q)∁ = Q × (Q ÷ P )∁ .

Proof. Apply a projective transformation sending {A, B, C, Q} 7→ {A, B, C, G}. Then it suffices to prove the
case when Q = G, which is true since
R = G × (G ÷ P )∁ .

Example 12.3.9. For any △, the symmedian point K is the crosspoint of the centroid G and orthocenter
H. Then we have

G ⋔ H = H × (H ÷ G)∁ = H × O ÷ G,

G ⋔ H = G × (G ÷ H)∁ = G × (H ′ ÷ G)∁

So K is the isogonal conjugate of G and the complement of the isotomic conjugate H ′ of H.

Example 12.3.10. X55 = I ⋔ M t:

I ⋔ M t = I × (I ÷ M t)∁ = I × (Ge ÷ G)∁

= I × (M t ÷ G) = I × (I ÷ Ge) = X55 .

Next we recall the definition of the cevapoint (TODO 7.1.23):

Definition 12.3.11. Let △P a P b P c and △Qa Qb Qc be the anticevian triangles of P and Q. Then the
cevapoint P ⋆ Q is the pole of P Q in conic (P P a P b P c QQa Qb Qc ).

We define the ceva conjugate of S and Q as the unique point P = S/Q such that S = P ∗ Q.

328
AoPS Chapter 12. Special Lines

Through the definition of crosspoints, we also get that P ⋆ Q is simultaneously the crosspoint of P and Q
wrt. both triangles △P a P b P c and △Qa Qb Qc .

Proposition 12.3.12. If S = P ∗ Q is the cevapoint of P and Q, then

S = Q ÷ (P ÷ Q)∁ = P ÷ (Q ÷ P )∁ .

Proof. Similarly to (TODO xiooix), assume that Q = G. Then S is the crosspoint of P and G with respect
∁ ∁ ∁
to △A B C . Thus,
S ∁ = P ∁ ⋔ G = G × (G ÷ P ∁ )∁ = (G2 ÷ P ∁ )∁ ,

so S = G2 ÷ P ∁ = Q ÷ (P ÷ Q)∁ .

Corollary 12.3.13. Let R and S be the crosspoint and cevapoint of P and Q, respectively. Then P × Q =
R × S.

Corollary 12.3.14. Let P = S/Q be the ceva conjugate of S and Q. Then

∁
P = Q × (Q ÷ S) .

Of course, we can also define the cross conjugate of R and Q as the unique point P satisfying R = P ⋔ Q,
then
∁
P = Q ÷ (R ÷ Q) .

⋔
We denote P by R Q.

Proposition 12.3.15. For points P and Q,

⋔
P × Q = (P/Q) × (Q P ).

A reminder that we will use t(P ), t(ℓ) as the trilinear pole/polar of P, ℓ, and c(P ) as the circumconic C
with perspector P , and c(C) as the perspector of circumconic C. Define t(P ) × c(Q) as the image of t(P )
under the isoconjugation P × Q is c(Q), and vice versa.

Since t(, )c() are projectively invariant, we have

Proposition 12.3.16.

Proposition 12.3.17. THIS IS PROPERTY OF A SNOWBALL

Corollary 12.3.18.

329
AoPS Chapter 12. Special Lines

Proof.

Example 12.3.19.

Proposition 12.3.20.

Proof.

Proposition 12.3.21. THIS IS PROPERTY OF A SNOWBALL

Proof.

Proposition 12.3.22.

Proof.

Proposition 12.3.23.

Proof.

Corollary 12.3.24.

Corollary 12.3.25. THIS IS PROPERTY OF A SNOWBALL

Proof.

Theorem 12.3.26. THIS IS PROPERTY OF A SNOWBALL

Proof.

Corollary 12.3.27.

Proof.

Proposition 12.3.28.

Proof.

Proposition 12.3.29.

Proof.

Corollary 12.3.30.

Proposition 12.3.31.

330
AoPS Chapter 12. Special Lines

Proof.

Proposition 12.3.32.

Proposition 12.3.33.

Example 12.3.34.

Example 12.3.35.

12.3.1 Permutation Line Revisited

Theorem 12.3.36.

Proof.

Example 12.3.37.

Theorem 12.3.38.

Example 12.3.39.

12.3.2 Radical and Square Transformations

In the view of analytical geometry, sometimes we see an expression for a point has every term have only
squares a2 , b2 , c2 . In this case we can rewrite it as a2 = u, b2 = v, c2 = w. So can we actually just rephrase our
whole problem in terms of u, v, w? By doing this, three points X, Y, Z collinear becomes A, B, C, X ′ , Y ′ , Z ′
conconic.

So how do we interpret this purely geometrically? Let’s define a new geometrical framework, by redefining
the two circle points Iis , J s as the two intersections of c(K 2 ÷ G) with L∞ (K here is the symmedian point.),
while fixing the line at infinity. We can reinterpret

K 2 ÷ G = c(△Iis J s )

as K = c(△IJ). Since (K 2 ÷ G) × G = K 2 , (in normal geometry), we get that the equivalent of the symmedian
point in this new geometry is the point K such that K × G = K, which is just the incenter or the three
excenters (any point and its extraversions is the same under this new geometry).

This lets us define more operations in barycentric coordinates, the radical transformation (−)r . By
definition, (I s )r = I and similarly for J.

331
AoPS Chapter 12. Special Lines
∁ ∁
Example 12.3.40. Since Gr = t(L∞ ) = G, K r = I, then the Nagel point N a = I = (K )r = (H ′ )r
where H ′ is the isotomic conjugate of H. Therefore we have H r = (G2 ÷ H ′ )r = G2 ÷ N a = Ge. Further,
Or = Ge∁ = M t. This gives us the following table of points and their transformations:

Inversely, we can change a, b, c to u2 , v 2 , w2 (the geometrical interpretation of this is defining two new
circle points, I r , J r as the two intersections of c(I) with L∞ ). We call this the square transformation and we
will notate it as (−)s . s, r are inverses of each other (this requires proof but is left to the reader).

In full generality, if we want to let u = u(a, b, c), v = v(a, b, c), w = w(a, b, c), then this defines two new
circle points I new , J new as the two intersections of c(P ) and L∞ , where P = [u(a, b, c)2 : v(a, b, c)2 : w(a, b, c)2 ].

332
Chapter 13

The Worst Of Xn

The objective of this chapter is to introduce, for 1 ≤ n ≤ 100, the triangle centers Xn . Recall that:

• X1 is the incenter I, the point where the angle bisectors concur;

• X2 is the centroid G, the point where the medians concur;

• X3 is the circumcenter O, the point where the perpendicular bisectors of the sides concur;

• X4 is the orthocenter H, the point where the altitudes concur;

• X5 is the center of the nine-point circle, N ;

• X6 is the symmedian point K, the isogonal conjugate of G.

• X7 is the Gergonne point Ge, the perspective center of the reference triangle and the contact triangle;

• X8 is the Nagel point N a, the perspective center of the reference triangle and the extouch triangle;

• X9 is the Mittenpunkt M t, the perspective center of the excentral triangle and the medial triangle;

• X10 is the Spieker center Sp, the anticomplement of I;

• X11 is the Feuerbach point F e, the point where the incircle ω and the nine point circle ε are tangent;

• X13 is the first Fermat point F1 , which satisfies

∡BF1 C = ∡CF1 A = ∡AF2 B = 120◦ ;

• X14 is the second Fermat point F2 , which satisfies

∡BF2 C = ∡CF2 A = ∡AF2 B = 60◦ ;

333
AoPS Chapter 13. The Worst Of Xn

• X15 is the first isodynamic point S1 , the isogonal conjugate of F1 ;

• X16 is the second isodynamic point S2 , the isogonal conjugate of F2 ;

• X19 is the Clawson point Cℓ, the perspective center of the excentral triangle and the orthic triangle;

• X20 is the de Longchamps point L, the reflection of H over O;

• X21 is the Schiffler point Sc, the concurrency point of the Euler lines of the triangles △ABC, △IBC,
△AIC, △ABI;

• X25 is the homothetic center of the orthic and tangential triangles;

• X33 is the perspector of the orthic and intangents triangles;

• X40 is the Bevan point Be, the reflection of I over O;

• X54 is the Kosnita point Ko, the isogonal conjugate of N ;

• X55 is the insimilicenter of Ω and ω Ge∗ ;

• X56 is the exsimilicenter of Ω and ω N a∗ ;

• X65 is the orthocenter of the intouch triangle Sc∗;

• X69 is the isotomic conjugate of H, H ′ ;

• X104 is the isogonal conjugate of ∞OI , ∞∗OI .

Additionally we can define △P = △Pa Pb Pc representing the cevian triangle, △P representing the
anticevian triangle. Then we let XnP , XnP represent Xn in the triangles △P and △P . For example, OH is
the circumcenter of △H = △Ha Hb Hc , that is, N . For △H , we always select X1 , or IH , and similarly △K ,
I K = O.

First let us see a commonly used theorem:

Theorem 13.0.1. For 17 ≤ n ≤ 53, (Xn , Xn+44 ) is a pair of isogonal conjugates.

For example,

Proposition 13.0.2. The isogonal conjugate of X25 is H ′ .

Proof. By definition

∁ ∁
X25 = H/K = K × (K ÷ H) = K × (O ÷ G) = K × H ÷ G.

Therefore G × K ÷ X25 = G2 ÷ H = H ′ .

334
AoPS Chapter 13. The Worst Of Xn

So we will write (H ′ )∗ as X25 .

13.1 X1 - related points

(Warning: I’m not sure how accurate this translation is)

If we define the incenter I = X1 as the fixed point of the isoconjugation ∞i × ∞−i , then actually there
are four choices for I (the other three being the excenters). Some points depend on this selection, like the
Feuerbach point F e = X11 , while others do not, like the centroid G = X2 .

Proposition 13.1.1. The three points I, O, K lie on a (non-circum) rectangular hyperbola. This hyperbola
is called the Stammler hyperbola of △ABC, and is denoted DS .

Proof. Consider the hyperbola D through O and K, we only need to prove that D is a rectangular hyperbola.
Because O and K are the anticevian points of △K , that is, the incenter (or excenter) and Gergonne point of
the triangle T△ Ω = △(TA Ω)(TB Ω)(TC Ω), D is the Feuerbach hyperbola of the tangential triangle, and is
thus rectangular.

Because the line through the circumcenter of T△ Ω and O is the Euler line E, we have:

Corollary 13.1.2. The Euler line E is tangent to DS at O.

Lemma 13.1.3. For any point P ,

• The crosspoint of I and P , I ⋔ P , and the cevapoint I ⋆ P ∗ , are isogonal conjugates.

⋔
• The cross conjugate of P and I, P I, and the ceva conjugate P/I, are isogonal conjugates.

Proof. Because I ÷ P = P ∗ ÷ I, P ÷ I = I ÷ P ∗ , we have

(I ⋔ P ) × (I ⋆ P ∗ ) = (I × (I ÷ P )∁ ) × (I ÷ (P ∗ ÷ I)∁ ) = I 2
⋔ ∁ ∁
(P I) × (P ∗ /I) = (I ÷ (P ÷ I) ) × (I × (I ÷ P ∗ ) ) = I 2 .

Of course, if I was replaced with the fixed point of any isogonal conjugation φ, and P ∗ was replaced with
P φ , then the above proof would still be valid.

335
AoPS Chapter 13. The Worst Of Xn

13.1.1 What points are on OI and HF e ?

When n = 1, 35, 36, 40, 46, 55, 56, 57, 65, Xn ∈ OI and is related to the selection of X1 .

When n = 1, 7, 8, 9, 21, 79, 80, 84, 90, Xn ∈ HF e and is related to the selection of X1 .

Out of these points, we have already introduced n = 1, 3, 4, 7, 8, 9, 21, 40, 55, 56, 65.

Proposition 13.1.4. The quadrilateral (IM t)(N aSc) is a harmonic quadrilateral on HF e .

Proof. This is just because (by (TODO 5.6.8))

(I, M t; N a, Sc)HF e = H(I, M t; N a, Sc) = (I, Be; ∞OI , O) = −1.

Proposition 13.1.5. Under the barycentric product, I × N a = G × M t = Sp × Sc. Then I × Sp = N a × Sc∗ .

Proof. I × N a = G × M t is, under the square transformation,

equivalent to K × H ′ = G × O, which is clearly true. Under the isogonal conjugation I × N a,

(N a, I × N a ÷ Sp; M t, I) = (I, Sp; G, N a) = −1 = (N a, Sc; M t, I),

thus I × N a = Sp × Sc.

Through (TODO 5.6.5), we have

Proposition 13.1.6. The Schiffler point Sc is actually just the cevapoint I ⋆ O.

X57 , X84

• X57 is the isogonal conjugate of M t = X9 , M t∗ .

• X84 is the isogonal conjugate of Be = X40 , Be∗ .

Proposition 13.1.7. Point M t is on HF e , and

I(M t; N a, Sc)HF e = −1,

so M t∗ lies on OI and satisfies (I, M t∗ ; N a∗ , Sc∗ ) = −1.

Proof. By (TODO 5.4.5), if we let A be the orthocenter of △BIC, Na be the second intersection of AI with
Ω, then
HA (I, M t; N a, Sc) = (I, I a ; ∞AI , Na ) = −1.

336
AoPS Chapter 13. The Worst Of Xn

And this is symmetric in A, B, C, so M t ∈ (IN aScHA HB HC ) = HF e .

Given the definition of M t, we know that it is just the ceva conjugate of G and I, G/I, and so

∁
I × (I ÷ G) = I × N a = I × G ÷ Ge.

Then
∁ ∁
M t∗ = I × Ge ÷ G = I × (M t ÷ G) = I × (I ÷ Ge) = Ge/I,

which gives us:

Proposition 13.1.8. The point M t∗ is the homothetic center of the excentral triangle △I a I b I c and the
contact triangle △DEF .

Corollary 13.1.9. The points G, Ge, M t, M t∗ are collinear.

Proof. We already know that G, Ge, M t∗ are collinear ((TODO 5.4.1)), and Ge, M t, M t∗ are collinear because
△DEF , △I a I b I c are the symmedian points of the respective triangles [is this right?]

Proposition 13.1.10. We have Be = OM t∗ ∩ HM t, Be∗ = OM t ∩ HM t∗.

Proof. By (TODO 7.4.5), we just need to prove Be = OM t∗ ∩ HM t. Be ∈ OM t∗ is just (TODO 13.1.7),

Be ∈ HM t is just (TODO 5.6.8).

X35 , X36 , X79 , X80

• X36 is the inverse of I about Ω, I J ;

• X35 is the harmonic conjugate of I J on OI;

• X79 is the isogonal conjugate of X35 ;

• X80 is the isogonal conjugate of X36 .

The most famous of these four points should be X80 , because we have by (TODO 8.1.19):

ˆ
Proposition 13.1.11. The point X80 is the antigonal conjugate of I, so we denote X80 by I.

Corollary 13.1.12. We have that I, N , F e, F e∨ , Iˆ are collinear.

And our first thought upon seeing X36 should be (TODO 5.5.6), because it tells us:

Proposition 13.1.13. The three lines F e∨ X35 , F eI J , IX79 are parallel to the Euler line E, which is to say,
all pass through ∞E = X30 .

337
AoPS Chapter 13. The Worst Of Xn

Proof. We already have that F e, ∞E , I J are collinear. Thus

∞E (F e, F e∨ ; I, N ) = −1 = ∞E (I J , X35 ; I, O)

gives us that F e∨ = X13 , ∞E , X35 are also collinear.

Finally, by (TODO 13.1.11), (TODO 13.1.12),

ˆ = (I, H; X79 , I)
I(O, H; X79 , I) ˆH
Fe

= (I, O; X35 , I J ) = −1 = (H, O; ∞E , N ),

and moreover I, ∞E , X79 are collinear.

Proposition 13.1.14. We have

(I, X35 ; O, Ge∗ ) = (I, I J ; O, N a∗ ) = −1.

Proof. Because (I, O; X35 , I J ) = (I, O; Ge∗ , N a∗ ) = −1, we only need to prove that (I, I J ; O, N a∗ ) = −1.
ˆ H, N a) = −1, and this is true because the tangents at I, Iˆ to HF e are both
But this is equivalent to (I, I;
parallel to OI ∥ HN a.

Proposition 13.1.15. The Schiffler point Sc = X21 is the intersection of X35 Iˆ and I J X79 , and its isogonal
ˆ
conjugate Sc∗ = X65 is the intersection of OI = X35 I J and X79 I.

Proof. Because
ˆ H = (I, O; X35 , I J ) = −1,
(I, H; X79 , I) Fe

X79 I J passes through the polar of IH in HF e , Sc∗ , but we know already that Sc∗ ∈ OI.

Proposition 13.1.16. The points Sp, Sc, X35 , Iˆ are collinear.

Proof. We show that Sp, Sc, X35 are collinear: by (TODO 13.1.14),

Sc(I, Sp; G; N a) = −1 = Sc(I, X35 ; O, Ge∗ ),

since we only need N a, Sc, Ge∗ collinear, which is (TODO 5.6.19).

ˆ ∞∗ are collinear.
Proposition 13.1.17. The points I J , I, OI

338
AoPS Chapter 13. The Worst Of Xn

ˆ
Proof. We have (I I)(HN a) is a complete quadrilateral on HF e , so from (TODO 13.1.14) and (TODO 8.2.18),

ˆ H, N a) = −1 = (I, I J ; N a∗ , O) = ∞∗OI (I, I J ; H, N a),

∞∗OI (I, I;

ˆ ∞∗ are collinear.
and I J , I, OI

13.1.2 X46 , X90

• X46 is the ceva conjugate of H and I, H/I;

• X90 is the isogonal conjugate of H/I.

∗
Because the pole of X46 in the diagonal conic with I will pass through X90 = X46 and H, there is:

Proposition 13.1.18. The points H, X46 , X90 are collinear.

Because
∁ ∁
H/I = I × (I ÷ H) = I × (O ÷ I)

Proposition 13.1.19. The point X46 is the reflection of O over I, and notably, X46 lies on OI.

On the other hand, it turns out that:

Proposition 13.1.20. The point Sc∗ is the I-complement of O.

Proof. Because Sc∗ is the crosspoint of I and H, we have

∁ ∁
Sc∗ = I × (I ÷ H) = I × (O ÷ I) .

Of course, this could also be proven by (TODO 13.1.6), because:

Proposition 13.1.21. The points I, O harmonically divide X46 , Sc∗ , so

I, H; X90 , Sc)HF e = −1

⋔ ⋔
By (TODO 13.1.3), we have X90 = H ∗ I=O I, and the polar of O in HF e is IX90 .

ˆ ∗ , X90 = I J Be∗ ∩ IBe.

Proposition 13.1.22. We have X46 = I J Be ∩ IBe ˆ

339
AoPS Chapter 13. The Worst Of Xn

ˆ 90 . Obviously, Be ∈ OI = I J X46 .
Proof. By (TODO 7.4.5), we just need to prove that Be = I J X46 ∩ IX

Because the tangent to HF e from X90 passes through O, we have

ˆ X90 ; ∞E , O) = (I, X90 ; I,

I(I, ˆ IO
ˆ ∩ HF e )H = −1,
Fe

ˆ 90 passes through Be.

and so IX

Proposition 13.1.23. We have X46 = M tX79 ∩ M t∗ X35 , X90 = M tX35 ∩ M t∗ X79 .

Proof. By (TODO 7.4.5), we just need to prove that X35 = M tX90 ∩ M t∗ X46 . Obviously, X35 ∈ OI =
M t∗ X46 .

By (TODO 13.1.10), O, M t, Be∗ are collinear, so

(I, X90 ; M t, Be∗ ) = −1.

With (TODO 13.1.22), this tells us

X90 (I, O; M t, I J ) = (I, X90 ; M t, Be∗ )HF e = −1,

and so M t, X35 , X90 are collinear.

13.1.3 IG and (IG)∗ ?

When n = 1, 8, 10, 42, 43, 78, Xn ∈ IG depends on the choice of X1 .

When n = 1, 34, 56, 58, 86, 87, Xn ∈ (IG)∗ depends on the choice of X1 .

Proposition 13.1.24. The point Sp = I ∁ lies on the Kiepert hyperbola, HK .

Proof. Consider the diagonal conic D of I, G. Because I a , I b , I c lie on D, D is a diagonal conic; it passes
through the centroid H ∁ of the anticevian triangle of G, △A∁ B ∁ C ∁ . So D∁ is a circumconic of △ which
passes through G, H, Sp, that is, HK .

Corollary 13.1.25. The points I, G, L lie on one diagonal conic.

Proof. Because Sp = I ∁ , G = G∁ , H = L∁ share a circumconic.

• X58 is the isogonal conjugate of Sp, Sp∗ .

Because G, H, Sp lie on one circumconic, we also have:

340
AoPS Chapter 13. The Worst Of Xn

Proposition 13.1.26. The point Sp∗ lies on the Brokard line OK.

Proposition 13.1.27. The points I, Sc, Sp∗ lie on a line, which is the anticomplement of SpSc∗ .

Proof. Consider the hexagon III ′ Sc∗ Sp(I ′ )∁ on (△ISp). By Pascal’s Theorem,

TI (△ISp) ∩ Sc∗ Sp, II ′ ∩ Sp(I ′ )∁ , I ′ Sc∗ ∩ (I ′ )∁ I

∁ ∁
are collinear. Also, II ′ = (Sp(I ′ )∁ ) by (TODO 13.1.52), or I ′ Sc∗ = ((I ′ )∁ I) . So, IScSp∗ = TI (△ISp) ∥
∁ ∁
Sc∗ Sp. Because I = Sp , we have IScSp∗ = (SpSc∗ ) .

• X42 is the crosspoint of I and K, I ⋔ K;

• X43 is the ceva conjugate of K and I, K/I;

• X86 is the ceva product of I and G, I ⋆ G;

⋔
• X87 is the cross conjugate of G and I, G I.

By (TODO 13.1.3), (X42 , X86 ), (X43 , X87 ) are isogonal conjugates. Because IX42 is tangent to (△IK),
we have IX42 = (△IK)∗ = IG. Because IX43 is the trilinear polar of K in (△I X43 ), we have K ∗ = G.

13.1.4 IH and (IH)∗ ?

When n = 1, 33, 34, 73, Xn ∈ IH depends on the choice of X1 .

When n = 1, 29, 77, 78, Xn ∈ (IH)∗ depends on the choice of X1 .

• X33 is the perspective center of the intouch triangle and the orthic triangle;

• X34 is the harmonic conjugate of X33 on IH;

• X77 is the isogonal conjugate of X33 ;

• X78 is the isogonal conjugate of X34 .

Proposition 13.1.28. The ceva product of Cℓ ⋆ X33 is just H.

Proof. Notice that the homothety centered about A which sends I to I a also sends Ta to Ta′ , the reflection
over I, so
(AHa , BC; Ha Cℓ, Ha X33 ) = Ha (A, AI ∩ BC; Ta′ , Ta ) = A(Ha , I; Ta′ , Ta ) = −1.

So by symmetry Cℓ ⋆ X33 = H.

341
AoPS Chapter 13. The Worst Of Xn

Corollary 13.1.29. Under the barycentric product,

I × H = Ge × X33 = N a × X34 ,

I ÷ H = X77 ÷ Ge = X78 ÷ N a.

Proof. By (TODO 13.1.41), Cℓ = I × H ÷ G, so

∁
X3 3 = (I × H ÷ G) × (I × H ÷ G ÷ H)

= I × H ÷ X × (N a ÷ G) = I × H ÷ Ge.

Because (H, I; X33 , X34 ) = −1, I × H ÷ X34 lies on HF e and satisfies

(I, H; Ge, I × H ÷ X34 )HF e = −1,

so I × H = N a × X34 .

And the second equality follows from taking the isogonal conjugate of the first

Proposition 13.1.30. The points I, Ge, X77 , and the points I, N a, X78 are collinear.

Proof. Because I, H, Ge lie on one circumrectangular hyperbola, taking an isoconjugation oabout I × Ge

gives that Ge, X77 , I are collinear. Similarly, I, H, N a lie on one circumrectangular hyperbola, so taking an
isoconjugation about I × N a gives that N a, X78 , I are collinear.

13.1.5 IN and (IN )∗ ?

When n = 1, 11, 12, 80, Xn ∈ IN depends on the choice of X1 .

When n = 1, 36, 59, 60, Xn ∈ (IN )∗ depends on the choice of X1 .

• X12 is the harmonic conjugate of F e in IN , F e∨ .

• X59 is the isogonal conjugate of F e, F e∗ ;

• X90 is the isogonal conjugate of F e∨ , F e∨∗ .

13.1.6 IK and (IK)∗ ?

When n = 1, 37, 45, 72, Xn ∈ IK depends on the choice of X1 .

When n = 1, 28, 81, 89, Xn ∈ (IK)∗ depends on the choice of X1 .

342
AoPS Chapter 13. The Worst Of Xn

• X37 is the crosspoint of I and G, I ⋔ G;

• X81 is the isogonal conjugate of X37 .

Obviously, we have
I ⋔ G = G × (G ÷ I)∁ ,

so we will just use (I ′ )∁ instead of X37 . Also,

X81 = I 2 ÷ (I × (I ÷ G)∁ ) = I ÷ (K ÷ I)∁ = I ⋆ K.

Proposition 13.1.31. The point (I ′ )∁ lies on IK.

Proof. Because IK is tangent to (IK)∗ = (△IG), it passes through I ⋔ G = (I ′ )∁ .

Because r × r∁ = (r−1 )∁ , it follows that:

Proposition 13.1.32. The barycentric product G × (I ′ )∁ = I × Sp.

Because, for any point P , the points P , P ′ , P ∁ , (P ′ )∁ lie on one circumconic, it follows that:

Proposition 13.1.33. The points I, Sp, (I ′ )∁ , I ′ lie on one circumconic, and

(I, Sp; (I ′ )∁ , Sc∗ )C = −1.

Then, I, (I ′ )∗ , Sp∗ , I ⋆ K are collinear and

(I, Sp∗ ; I ⋆ K, Sc) = −1.

Proof. We only need to prove that (I, Sp; (I ′ )∁ , Sc∗ )C = −1. Taking an isoconjugation about I × Sp, it
follows from (TODO 13.1.32) and (TODO 13.1.5) that the proposition is equivalent to (Sp, I; G, N a) = −1,
which is obviously true.

There is another way to look at this: because ISp∗ is tangent to (△ISp), then by (TODO 13.1.43),

(I, Sp; (I ′ )∁ , Sc∗ )C = I(Sp∗ , Sp; (I ′ )∁ , Sc∗ ) = (Cℓ∗ , G; M t; M t∗ ) = −1.

Corollary 13.1.34. The point (I ′ )∁ is the harmonic conjugate of K on IM t.

343
AoPS Chapter 13. The Worst Of Xn

Proof. This is because

(I, M t; (I ′ )∁ , K) = Cℓ(I, Sp; (I ′ )∁ , M t∗ ) = −1.

Corollary 13.1.35. The points Cℓ, (H ′ )∗ , (I ′ )∗ are collinear.

Proof. Because

Cℓ(I, Sp; (I ′ )∁ , M t∗ ) = −1 = (I, Be; Ge∗ , M t∗ ) = Cℓ(I, Sp; Ge∗ , M t∗ ),

it follows that Cℓ, (H ′ )∗ , (I ′ )∁ are collinear.

13.1.7 Euler line and HJ ?

When n = 21, 27, 28, 29, Xn ∈ E depends on the choice of X1 .

When n = 65, 71, 72, 73, Xn ∈ HJ depends on the choice of X1 .

• X27 is the cevapoint of H and Cℓ, H ⋆ Cℓ;

• X28 is the barycentric product of Cℓ and (H ′ )∗ , Cℓ × (H ′ )∗ ;

• X29 is the cevapoint of I and H, I ⋆ H.

We first prove that these three points lie on E: by (TODO 13.1.41),

H ⋆ Cℓ = H ÷ ((I × H ÷ G) ÷ H)∁ = H ÷ (Sp ÷ G) = Sp◦ ,

and we know E deg = (△G◦ H ◦ ) = (△H ◦ G◦ ) = KK ∋ Sp.

Notice that

Cℓ ⋆ (H ′ )∗ = (I × H ÷ G) ÷ ((H × K ÷ G) ÷ (I × H ÷ G))∁

= I × H ÷ G ÷ (I ÷ G)∁ = I × H ÷ Sp.

Because Cℓ ⋆ (H ′ )∗ ∈ E if and only if Sp ∈ I × H ÷ E = (△ICℓ), this is just because I, Sp∗ , Cℓ∗ are collinear.

Finally, because the polar of H about the diagonal conic D of I, H, pD (H), passes through the isogonal
conjugate of H, O, and I ⋆ H, we have I ⋆ H ∈ E.

Proposition 13.1.36. The point X73 is on IH.

344
AoPS Chapter 13. The Worst Of Xn

Proof. Because X73 = I ⋔ O, the tangent to (△IO) from I, (△IO)∗ = IH, passes through X73 .

Consider the cross-ratio preserving map ϕ : E → OI, P → X73 L∗ ∩ OI. We have Oϕ = I, H ϕ = O,

Scϕ = Sc∗ . So for any point P ∈ E,

R HP r OP ϕ
· = (O, H; Sc, P ) = (O, P ; Sc∗ , ϕ(P )) = − · ϕ ,
2(R + r) P O R+r P I

which is to say
OP ϕ R HP
=− · .
P ϕI 2r P O

If P = N , we have
ON ϕ R
=− =⇒ N ϕ = I J.
N ϕI 2r

If P = ∞E , we have
O∞ϕE R
=− =⇒ ∞ϕE = X35 .
∞ϕE I 2r

If P = G, we have
OGϕ R
=− =⇒ Gϕ = N a∗ .
Gϕ I 2r

If P = L, we have
OLϕ R
ϕ
=− =⇒ Lϕ = Ge∗ .
L I 2r

To put all of these results together:

Proposition 13.1.37. We have X73 = IH ∩ I J Ko ∩ ∞E X35 ∩ KN a∗ ∩ Ge∗ L∗ .

Proposition 13.1.38. The three points I ⋔ K, sc∗ , X73 are collinear.

Proof. Because K, H, O lie on one circumconic, it follows that I ÷ K, I ÷ H, I ÷ O are collinear, and it
follows that I ⋔ K, Sc∗ = I ⋔ H, X73 = I ⋔ O are also collinear.

13.1.8 Other related points

The points which did not appear above, but which are related to X1 , are:

19, 27, 31, 38, 41, 44, 47, 48, 63, 75, 82, 85, 88, 91, 92, 100.

Recall the Clawson point:

• The Clawson point X19 , also named Cℓ, is the homothetic center of the extangent triangle △TA′ Tb′ Tc′
and the orthic triangle △Ha Hb Hc .

345
AoPS Chapter 13. The Worst Of Xn

Proposition 13.1.39. The four points K, Cℓ, X34 , Sc∗ are collinear.

Proof. If we let X = Sc∗ H ′ ∩ HM t, then by (TODO 5.8.8) we have

Sc∗ (O, H; K, H ′ ) = (O, H; K, H ′ )HJ = (H, O; G, (H ′ )∗ ),

Sc∗ (O, H; Cℓ, H ′ ) = (Be, H, Cℓ, X) = (H, Be; X, Cℓ) = Ge∗ (H, O; X, (H ′ )∗ ).

Thus K, Cℓ, Sc∗ are collinear if and only if G lies on Ge ∗ X.

By (TODO 5.6.19) and (TODO 5.6.22), Sc∗ H ′ = N aGe is the complement of IM t, so therefore X is the
reflection of M t across Sp, so the ratios given by (TODO 5.6.8) give us

HX BeGe∗ OG
 
BeM t 2R + r 1
· ∗
· = · ·
XBe Ge O GH M tH R 2
 
2R 2R + r 1
= · · = −1,
2R + r R 2

so X, Ge∗ , G are collinear.

By (TODO 5.6.18),
Cℓ(I, H, ; X33 , X34 ) = −1 = Cℓ(I, Be; Ge∗ , Sc∗ ),

so Cℓ, X34 , Sc∗ are collinear.

Proposition 13.1.40. The five points K, M t, Cℓ, Ge∗ , M t∗ lie on a common circumconic and the quadrilateral
(KCℓ)(M tM t∗ ) is a harmonic quadrilateral.

Proof. Note that the image of line GGe under isotomic conjugation is the circumconic (ABCGN a), and
since G is its own isotomic conjugate, GGe is tangent to (ABCGN a) at G. We know that GGeM tM t∗ are
collinear, so take the isogonal conjugate of this statement to get that KN a∗ is tangent to C := (ABCKGe∗ ).
So to prove that Cℓ lies on C, we only need to prove that (KM tCℓGe∗ M t∗ ) is tangent to KN a∗ , but this is
true by applying (TODO 5.6.18) and (TODO 13.1.4), with

K
Ge∗ (K, Cℓ; M t, M t∗ ) = (KGe∗ ∩ HM t, Cℓ; M t, Be) = (Ge∗ , Sc∗ ; I, Be) = −1; = K(N a∗ , Cℓ; M t, M t∗ ) = (N a∗ , Sc∗ ; I, M t∗

This also tells us that (KCℓ)(M tM t∗ ) is a harmonic quadrilateral on C.

Proposition 13.1.41. In barycentric coordinates, Cℓ × G = I × H.

f = I × H ÷ G, Sp
Proof. Suppose Cℓ f = IG ∩ H Cℓ.
f Then we have that G, H, Sp
f lie on a common circumconic

(namely, the Kiepert hyperbola HK ). This tells us that Sp

f is the second intersection of line IG with HK (by

(TODO 13.1.24) this is Sp). Thus Cl

f lies on H Sp
f = HM t. We do a similar thing for the isogonal conjugate

346
AoPS Chapter 13. The Worst Of Xn
∗
f = I × G ÷ H = Ge × M t ÷ H, so H, Ge, M t since lie on a common circumconic as well, we get that
of Cℓ
∗
f ∈ GeM t = GeM t∗ . But from (TODO 13.1.40) we know that Cℓ∗ ∈ GeM t∗ . Therefore Cℓ
Cl f is the second
∗
intersection of HM t with GeM t∗ = (ABCGe∗ M t), which is Cℓ.

Corollary 13.1.42. I, H, Cℓ lie on a common diagonal conic.

Proof. The I-complements of these three points are

I, I × (H ÷ I)∁ = I × (I ÷ O)∁ = I ⋔ O, I × (Cℓ ÷ I)∁ = I × (H ÷ G)∁ .

Therefore I, H, Cℓ lie on a common diagonal conic if and only if I, I ⋔ O, I × O ÷ G lie on a common

circumconic. Consider the isoconjugation given by the barycentric product I × O; this is equivalent to the
fact that O, Sc = I ⋆ O, G are collinear.

• X63 is the isogonal conjugate of the Clawson point, we will call it Cℓ∗ .

In barycentric coordinates,
Cℓ∗ = K × G ÷ (I × H ÷ G) = I × G ÷ H.

From (TODO 13.1.40), we directly get:

Proposition 13.1.43. Cℓ∗ lies on GGe and

(G, Cℓ∗ ; M t, M t∗ ) = −1.

Proposition 13.1.44. I, Sc, Sp∗ , Cl∗ are collinear.

Proof. Note that I = I ∗ , K = G∗ , N a∗ , Sp∗ lie on a common circumconic (ABCIK) and

(I, Sp∗ ; G, N a)(IG)∗ = (I, Sp; G, N a) = −1.

Since IG is tangent to (IG)∗ at I, from (TODO 13.1.43),

I(G, Sp∗ ; M t, M t∗ ) = I(G, Sp∗ ; K, N a∗ ) = −1 = (G, Cl∗ ; M t, M t∗ ),

giving us that I, Sp∗ , Cℓ∗ are collinear. Similarily, since IM t∗ is tangent to HF e , from (TODO 13.1.4) and
(TODO 13.1.43),

I(G, Sc; M t, M t∗ ) = (N a, Sc; M t, I)HF e ) = −1 = (G, Cℓ∗ ; M t, M t∗ ),

347
AoPS Chapter 13. The Worst Of Xn

so I, Sc, Cℓ∗ are collinear.

Proposition 13.1.45. I, G, Cℓ∗ lie on a common diagonal conic.

Proof. Note that these three points are just the three points in (TODO 13.1.42) under a projective transfor-
mation G ÷ H.

Since I × G = H × Cℓ∗ and Sc = ICl∗ ∩ GH, we have IH ∩ GCℓ∗ = I × G ÷ Sc. From (TODO 13.1.5),

I × G ÷ Sc = G × Sp ÷ N a.

In ETC this point is X226 . What is its actual characterization?

Proposition 13.1.46. X226 is actually the complement (Cℓ∗ )∁ of Cℓ∗ .

Proof. From (TODO 13.1.45), Sp = I ∁ , G = G∁ , (Cl∗ )∁ lie on a common circumconic. Since X226 =
G×Sp
G × Sp ÷ N a also lies on (△GSp) = GSp , we only need to prove that X226 , G, Cℓ∗ are collinear. Take
the isoconjugation given by I × G; this is equivalent to proving that Sc, I, H lie on a common circumconic,
however this is just the Feuerbach hyperbola.

Corollary 13.1.47. N a, L (= de Longchamps point), Cℓ∗ are collinear.

∁
Proof. Since (Cl∗ )∁ ∈ IH, we have Cl∗ ∈ IH = N aL.

Proposition 13.1.48. Sp, X46 , Cℓ∗ are collinear.

Proof. Since IH ∥ N aLCℓ∗ , we have that the line through the midpoint Sp of segment IN a and the midpoint
O of segment HL is also parallel to them. Since

(I, X46 ; O, Sc∗ ) = −1,

so we only need to prove that Sp(I, Cl∗ ; O, Sc∗ ) = −1, but

∞
Sp(I, Cℓ∗ ; O, Sc∗ ) = (G, Cℓ∗ ; GCℓ∗ ∩ OSp, (Cℓ∗ )∁ ) =IH (G, N a; Sp, I) = −1.

• X75 is the isotomic conjugate I ′ of the incenter I;

• X31 is the isogonal conjugate (I ′ )∗ of the isotomic conjugate I ′ of the incenter I.

348
AoPS Chapter 13. The Worst Of Xn

Since I, Ge, N a lie on a common circumconic HF e , by taking isotomic conjugation we get

Proposition 13.1.49. Ge, N a, I ′ are collinear.

Proposition 13.1.50. I, (I ′ )∗ , Cℓ∗ are collinear.

Proof. Since (I ′ )∗ = I × K ÷ G, we have I, (I ′ )∗ , Cℓ∗ = I × G ÷ H are collinear if and only if G, K, H ′ are

collinear, which we have proven back in (TODO 5.6.21).

Again, we can take the isotomic conjugate of the above result to get

Corollary 13.1.51. I, Cℓ, I ′ lie on a common circumconic.

Proposition 13.1.52. N a, Sc∗ , I ′ all lie on the line anticomplement of II ′∁ .

Proof. From (TODO 13.1.33),

I ′ (I, Sp; (I ′ )∁ , Sc∗ ) = −1 = I ′ (I, Sp; G, N a),

∁
∁
and therefore I ′ Sc∗ = I ′ N a. Since N aI ′ = II ′∁ , we get N aSc∗ I ′ = II ′∁ .

• X48 is the crosspoint of I and Cℓ∗ , I ⋔ Cℓ∗ ;

• X92 is the cevapoint of I and Cℓ, I ∗ Cℓ;

⋔
• X91 is the cross conjugate of X48 and X92 , X48 X92 ;

• X47 is the isogonal conjugate of X91 .

Since Cℓ = I × H ÷ G, we have

⋔
I Cℓ∗ = I × (I ÷ Cℓ∗ )∁ = I × (H ÷ G)∁ = I × O ÷ G,

I ⋆ Cℓ = I ÷ (Cℓ ÷ I)∁ = I ÷ (H ÷ G)∁ = I × G ÷ O.

This tells us that (X48 , X92 ) are a pair of isogonal conjugates, and that (X63 , X92 ) are also a pair of isogonal
conjugates.

Note that since I is its own isogonal conjugate, ICℓ is tangent to (△ICℓ∗ ), and thus:

Proposition 13.1.53. I, Cℓ, X48 are collinear.

Proposition 13.1.54. H, N a, X92 are collinear.

349
AoPS Chapter 13. The Worst Of Xn

Proof. Let X be the second intersection of line HX92 with the diagonal conic D through I, H, Cℓ. Then
(ICℓ)(HX) is harmonic on D by the tangencies. Since the Euler line E is the tangent to D at H (note that
X27 = I ⋆ H ∈ E), so

H(I, Sp; G, X92 ) = H(I, Cℓ; G, X) = −1 = (I, Sp; G, N a),

so by Prism Lemma H, N a, X92 are collinear.

From (TODO 13.1.42), IX29 , IX92 both are the tangents from I to the diagonal conic through I, H, Cℓ.
Thus we also have:

Proposition 13.1.55. I, X29 , X92 are collinear.

Similarily, CℓX27 , CℓX92 are both thhe tangents from I to the diagonal conic, so we also have

Proposition 13.1.56. Cℓ, X27 , X92 are collinear.

Proposition 13.1.57. X47 is the X92 -complement of X91 .

Proof. This is because

X47 = X48 × X92 ÷ X91 = (X91 × (X91 ÷ X92 )∁ ) × X92 ÷ X91 = X92 × (X91 ÷ X92 )∁ .

Proposition 13.1.58. X91 = I × H ÷ X24 , and therefore X47 = I × X24 ÷ H.

Proof. First, we have

∁ ∁
X91 = X92 ÷ (X48 ÷ X92 ) = I × G ÷ O ÷ (O2 ÷ G2 ) .

Thus we only need to prove

∁ ∁
H × (N ÷ H) = X24 = K × (O2 ÷ G2 ) ,

in other words
∁ ∁
(N ÷ H) = O ÷ G × (O2 ÷ G2 ) .

In reality, we prove:

Lemma 13.1.59. For all ratios (weight-0) r,

∁ ∁ ∁
(r∁ ÷ r ) = r × (r2 ) .

350
AoPS Chapter 13. The Worst Of Xn

Proof of Lemma. Let r = I ÷ G, which is just

∁
(Sp ÷ N a) = I × H ′ ÷ G2 = I ÷ H,

which is just (TODO 13.1.46) (note that Cℓ∗ = I × G ÷ H, X226 = Sp × G ÷ N a).

Choosing r = O ÷ G in the above lemma solves our problem.

Since X24 lies on the Euler line, we immediately get:

Corollary 13.1.60. X47 lies on line ICℓ∗ = I × E ÷ H.

Proposition 13.1.61. The three points Sc, X24 , X47 lie on a common circumconic, and thus Sc∗ , X68 , X91
are collinear.

Proof. We only need to prove that Sc × X47 ÷ X24 lies on ISc = ICℓ∗ = IX47 see (TODO 13.1.43). However
(TODO 13.1.58) tells us that
Sc × X47 ÷ X24 = I × Sc ÷ H.

Thus we only need to prove that I, H, Sc lie on a common circumconic, which is just HF e .

Proposition 13.1.62. X47 is the intersection of X33 X90 and X34 X46 .

13.2 X1 - unrelated points

13.2.1 Points on the Euler line

When n = 2, 3, 4, 5, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, Xn lies on the Euler line.

• X22 is the perspector of the tangential triangle and the circumcevian triangle of G.

Let △K a K b K c be the tangential triangle, let △Ka Kb Kc be the cevian triangle of K. let △Ω Ω Ω Ω
G = △Ga Gb Gc

be the circumcevian triangle of G, and let △Ω Ω Ω Ω

K = △Ka Kb Kc be the circumcevian triangle of K, then the

corresponding vertices GΩ Ω
a , Ka in the two triangles are reflections across the perpendicular bisector of BC by

isogonal conjugation. Since the perpendicular bisector of BC goes through K a , therefore K a GΩ

a intersects

BC at the isotomic conjugate of Ka (call it Ka′ ). Therefore AKa′ passes through the isotomic conjugate of
K ′ , and by symmetry we get:

Proposition 13.2.1. X22 is the ceva conjugate K ′ /K of K ′ and K.

351
AoPS Chapter 13. The Worst Of Xn

This tells us that

∁ ∁
X22 = K × (K ÷ K ′ ) = K × (K 2 ÷ G2 ) .

Proposition 13.2.2. X22 lies on the Euler line and

(G, X22 ; O, (H ′ )∗) = −1.

Proof. X22 ’s image under the radical transformation is

∁
I × (I 2 ÷ G2 ) = I × H ′ ÷ G = I × G ÷ H = Cℓ∗ ,

but E = GO under the radical transformation is GM t. Therefore from (TODO 13.1.43), X22 ∈ E and

(G, X22 ; O, K 2 ÷ O) = −1.

Finally, note that (H ′ )∗ = K × G ÷ (G2 ÷ H) = K × H ÷ G = K 2 ÷ O,

• X23 is the far-out point, defined as the inversion of G wrt. the circumcircle Ω. We will call it GJ .

• X67 is the isotomic conjugate of GJ , and is also the antipode of K on HJ . These two points are basically
useless.

• X24 is the perspector of △ABC and the orthic triangle of the orthic triangle.

• X51 is the centroid of the orthic triangle, GH .

• X52 is the orthocenter of the orthic triangle, HH .

• X53 is the symmedian point of the orthic triangle, KH .

• X68 is the isogonal conjugate of X24 .

• X95 is the isogonal conjugate G∗H of GH .

• X96 is the isogonal conjugate HH

∗
of HH .

• X97 is the isogonal conjugate KH

∗
of KH .

Proposition 13.2.3. X24 lies on the Euler line E.

352
AoPS Chapter 13. The Worst Of Xn

Proof. Consider X24 in reference to the orthic triangle; since H is its incenter (call it IH ), we have that
X24 is X46 wrt. the orthic triangle, in other words X2 4 = HH /IH . Therefore from (TODO 13.1.19),
X24 ∈ OH IH = N H = E.

(TODO 13.1.19) also tells us that X24 is the H = IH -anticomplement wrt. the orthic triangle of N =
OH . Switching our reference triangle, this means that there exists a projective transformation that sends
∁
A, B, C, H, X24 to Ha , Hb , Hc , H, N , so X24 is the H-anticomplement of N , H × (N ÷ H) . In the proof of
∁
(TODO 13.1.58) we can also see that this is equal to K × (O2 ÷ G2 ) .

Proposition 13.2.4. N, K, X24 , HH lie on a common circumconic.

Proof. If we switch our reference triangle to the orthic triangle, then this is equivalent to proving that O, M t,
∁
O’s I-anticomplement I × (O ÷ I) and H lie on a common circumconic of the excentral triangle. Take
⋔
the I-complement of this statement, this is equivalent to proving that Sc∗ , K, I O, O lie on a common
circumconic of the reference triangle, which is just HJ .

Corollary 13.2.5. We have X24 = GN ∩ KKo, X68 = GKo ∩ KN .

Proposition 13.2.6. We have X24 = X33 X35 ∩ X34 X36 .

Proof. From (I, O; X35 , I J ) = (I, H; X33 , X34 = −1, and therefore we only need to prove that X24 , X34 , X36
are collinear. This is true by Pascal on the hexagon OHX73 KoKSc∗ on the Jerabek hyperbola HJ , where

X34 = HX73 ∩ KSc∗ , X36 = X73 Ko ∩ Sc∗ O, X24 = OH ∩ KoK

are collinear.

From (TODO 5.6.8), we get O = BeH , K = M tH , HH are collinear, implying:

Proposition 13.2.7. HH lies on the Brocard axis.

From (TODO 5.4.6), we can get

Proposition 13.2.8. H = IH , KH , K = M tH are collinear.

⋔ ⋔
Obviously, we have GH = H K, HH = H X24 .

• X26 is the circumcenter OK of the tangential triangle.

By applying (TODO 5.2.3) on the tangential triangle, we get that OK lies on the Euler line.

353
AoPS Chapter 13. The Worst Of Xn

13.2.2 HJ points

n = 3, 4, 6, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74.

13.2.3 Brocard axis OK points

n = 3, 6, 15, 16, 32, 39, 50, 52, 58, 61, 62.

13.2.4 Kiepert hyperbola points

n = 2, 4, 13, 14, 76, 83, 94, 96, 98.

13.2.5 Others

We’ll do these later.

354