A walk through the Projective Plane

RedPig
May 31, 2021

c 2021 RedPig. All rights reserved.

1
Preface: what you should know
The Method of Moving Points is a very powerful method which can be
traced back to the 90’s [1]. It has been recently generalized to cover more
complex cases in olympiad geometry( by yayup in the blog [link1], and some
follow ups by Vladyslav Zveryk [link2], Zack Chroman [link3], . However, I still
struggle a lot when reading these documents, especially when I was not familiar
enough with the projective geometry. In particular, I was hardly able to tell
whether I applied this method correctly on a given problem. I guess this is
common issue for newcomers in algebraic geometry like me. Hence I went over
several different lectures/notes in algebraic geometry to get me familiar with the
topic [5, 3, 2, 4, 6]. The purpose of this note is to explain in more details from
scratch what is the projective geometry and how the Method of Moving Points
works.
A proper definition of the projective space requires a lot of fundamental
concepts in abstract algebra, such as group [Wiki link], quotient group [Wiki
link], equivalence class [Wiki link], etc. While these notions are necessary for
understanding high dimensional projective space, we can get rid of most of
them when focusing on planar geometry, which is the case in maths olympiad.
Nevertheless, a minimum knowledge on linear algebra is still required, such
as three dimensional matrix-vector product, determinant, etc. Although not
necessary, for those who are encouraging, check out the full lecture (≈30h) of
Linear Algebra at MIT by Gilbert Strang, available on [Youtube watchlist]. I
will try to reduce at maximum the prerequisite on linear algebra in this note,
hopefully this is accessible to any high school student, providing a minimum
sense of what the projective space looks like.
Last but not least, applying this method can easily lead to errors, I suggest
taking an extreme attention when considering it. Especially, apply it only if you
are a hundred percent confident that you fully understand all the underlying
concepts. Also, I am not sure whether this method is allowed in national/inter-
national olympiads, probably not at the moment, in which case you would need
to build the theory during the writing of the solutions.

2
1 Why projective space?
Before even talking about geometry, let us start with a simple algebraic fact that
the equation
x2 + 1 = 0,
admits no solution in R, i.e. there does not exist any real number x ∈ R such
that x2 = −1.
In contrast, amazingly, the same equation is soluble in complex space, with
solution x = ±i. In other words, the complex space extends the real space
by appropriately introducing additional elements to complete the space. This
extension allows us to achieve the fundamental theorem of algebra:
every non-constant polynomial with complex coefficients has at least
one complex root.
which is not true if we only restrict ourselves to real numbers, i.e.
every non-constant polynomial with real coefficients does not neces-
sarily has one real root.

Hence, by extending to the complex space, we achieve a universal way to

factorize polynomials ! Such unification is extremely powerful and we can hardly
go further without such fundamental object.1
The reason that we start with this seemingly unrelated topic is to motivate
the introduction of projective plane, denoted as P2R , which can be viewed as an
extension of the ordinary Euclidean plane2 . One important unification
that projective plane provides is that
in projective plane, any two distinct lines intersect in one and only
one point.
while in the ordinary Euclidean plane,

in Eucldiean plane, any two distinct lines are either parallel or

intersect in one point.
This unification simplifies the statement of many geometrical properties, and
together with some other important characteristics, the projective space plays
a fundamental role in the modern development of algebraic geometry. In the
following, we start formally defining the projective plane P2R .

1 A proper definition of completeness and vector space extension requires much deeper tools

in abstract algebra and it is beyond the discussion in this note.

2 Formally called projective completion.

3
2 Defining the projective plane P2R
We start with an informal description of the projective plane P2R , as the extension
of the Euclidean plane with points at infinity.

Definition 2.1 (Informal Definition). The projective plane P2R is an

extensiona of the Euclidean plane by adding points at infinity such that
there is one and only one point at infinity along each direction in the
plane.
a Here extension means all the points in the Euclidean plane are preserved in the

projective plane and moreover it includes some additional points

Conceptually, theses points at infinity serves as the intersection of parallel

lines. In particular, all the parallel lines along the same direction intersect at an
unique point at infinity. In order to make this intuition concrete, we introduce
the canonical representation of the projective plane.

Definition 2.2 (Canonical representation). The points in the projective

plane P2R are represented in a three coordinate system [x : y : z], with
bracket and colons separating the coordinates in order to differentiate
from ordinary points in the space. The canonical representation of the
points in the projective plane are given by


 [x : y : 1] represents the point (x, y) in the Euclidean plane

 [x : 1 : 0] represents the point at infinity along direction (x, 1), which


 includes all directions not parallel to x-axis.
[1 : 0 : 0] represents the point at infinity along x-axis


In particular, all the points at infinity has its third coordinate equals to
zero.

The canonical representation provides an explicit representation to associate

each point in the Euclidean plane to a point in the projective plane. Moreover,
it adds in the points at infinity, which in total gives
projective plane P2R = Euclidean plane + points at infinity
One may wonder why do we call something like [x : 1 : 0] as a point at infinity
while every coordinate is finite. To answer it in an informal way, one should
regard the last coordinate as the denominator in a quotient and division by
zero gives infinity. There is indeed an analogy between the projective plane and
the rational numbers. To make this connection clearer, we introduce a more
general way to represent points in the projective plane through the homogeneous
representation.

4
 Definition 2.3 (Homogeneous representation). For any (x, y, z) ∈
R3 \(0, 0, 0), i.e. at least one of x, y, z is non-zero, we define the relation

[x : y : z] = [λx : λy : λz] for any λ ∈ R∗ .

In particular, for any (x, y, z) ∈ R3 \(0, 0, 0), there is a one and only
one point A in the canonical representation such that A = [x : y : z].
Therefore, every homogeneous representation [x : y : z] is associated to a
point in the projective plane, while a single point in the projective plane
can be represented in multiple forms in the homogeneous representation.

Proof. If z 6= 0, then A = [ xz : yz : 1]; if z = 0 and y 6= 0, then A = [ xy : 1 : 0];

if z = 0 and y = 0 then A = [1 : 0 : 0]. Conversely, if we have A and B in the
canonical representations such that A = [x : y : z] = B, then there is λ such that
A = λB. However, by enumerating different forms in the canonical representation,
this happens if and only if λ = 1. Hence the uniqueness follows.

4
! [0 : 0 : 0] is NOT a point in the projective plane.

The idea that different representation can refer to the same object is similar to
the representation of rational numbers. More precisely, the number 12 can be
represented by 24 or 2ππ
. While 12 , 24 or 2π
π
all look different at their appearances,
they indeed associate to the same number. In other words, if we use an extremely
unusual notation [x : y] to refer to rational number xy , then [1 : 2] = [2 : 4] = [π :
2π] = [λ : 2λ] for any λ 6= 0. In such a way, [1 : 0] = 01 is the infinity.
In our case, projective points are given in a three-coordinates system, and by
definition [x : y : z] = [λx : λy : λz]. Namely, two representations refers to the
same point if

“one can multiply all its coordinates by a same number λ to get the other”.

One can indeed properly define a notion of quotient in such system, –quotient of
group, again this notion requires sufficient prerequisites in abstract algebra, and
it is beyond the discussion of this note.
To summarize, on one hand, projective plane is an extension of the Euclidean
plane, where we artificially give multiple representations to the same point; on
the other hand, projective plane contains a brunch of infinite points, which
does not exist in the Euclidean plane. It is always useful to keep in mind the
analogy of rational numbers, where, multiple representations can refer to
the same point in the projective plane P2R .

5
Example 2.4. Let us go through a few examples to get a better idea of the
concept on homogeneous representation.
• Take A = [1 : 1 : 1] in the projective plane, which is in its canonical represen-
tation. We can also use [2 : 2 : 2], [e : e : e] or [π : π : π] to denote the same
point A in the homogeneous representation system. Moreover, it refers to the
point (1, 1) in the Euclidean plane.
• Take B = [1 : 5 : 0] in the homogeneous representation. Then it refers to the
point [ 15 : 1 : 0] in the canonical representation, which corresponds to the point
at infinity along the direction d~ = ( 15 , 1).

4
! We need to reduce to the canonical form before associating a projective
point back to the Euclidean plane. In particular, [2 : 2 : 2] does not refer to the
point (2, 2), it refers to the point (1, 1). The point (2, 2) in the Euclidean plane is
associated to the canonical representation [2 : 2 : 1] or more generally [2λ : 2λ : λ].

Excercise 2.5. Transform between different representations of the points.

(a) Represent the projective point [1 : 2 : 3] in canonical representation.

(b) What is the point in Euclidean plane corresponding to the projective point
[0 : 4 : 1]? What about [0 : 4 : 2] or [0 : 4 : 4] ?
(c) Represent the projective point [13 : 17 : 0] in its canonical representation.
What about [481 : 629 : 0]?

(d) Represent the projective point [e : 0 : 0] in its canonical representation.

What about [100 : 0 : 0] or [π : 0 : 0]?
From now on, we usually use the homogeneous representation to represent
projective points since they are easier to manipulate algebraically. To see that
the projective plane actually encodes the notion of “plane”, we now define the
projective curves.

6
3 Projective Curves
The curves in the Euclidean plane are usually represented as the solution of a two-
variable polynomial Q(x, y) = 0, for instance Q(x, y) = ax + by + c defining the
lines and Q(x, y) = x2 + y 2 − 1 defines the unit circle. Similarly, a natural way to
define curves in the projective plane is via a three-variable polynomial P (x, y, z),
as the projective plane is defined in a three-coordinate system. However, the
special rule that [x : y : z] defines the same point as [λx : λy : λz] imposes a
special constraint on the form of the polynomial :

it must satisfy P (x, y, z) = 0 ⇔ P (λx, λy, λz) = 0.

To satisfy this constraint, we introduce the notion of homogeneous polynomials.

Definition 3.1. A polynomial P (x, y, z) is homogeneous of degree d

if
P (λx, λy, λz) = λd P (x, y, z), for any x, y, z ∈ R3 , λ ∈ R∗ .

In particular, if we take x = y = z = 0 then P (0, 0, 0) = λd P (0, 0, 0) for any

λ, implying that P (0, 0, 0) = 0. Hence P does not have any constant term. To
illustrate some examples 3
• P (x, y, z) = x + y + z is homogeneous of degree 1.
• P (x, y, z) = x2 + y 2 − z 2 is homogeneous of degree 2.
• P (x, y, z) = x2 − yz is homogeneous of degree 2.
• P (x, y, z) = x − xyz is not homogeneous.
It is easy to see that if P is homogeneous then P (x, y, z) = 0 ⇔ P (λx, λy, λz) = 0.
Hence the solution of P is consistent with the homogeneous coordinates. This
leads to the definition of projective curves:

Definition 3.2. A projective curve is the set of projective points

[x : y : z] satisfying P (x, y, z) = 0, where P (x, y, z) is a homogeneous
polynomial.

A natural question that one may come to mind would be how to extend a curve
in the Euclidean plane to a curve in the projective plane. To do so, we perform
the homogenization on the polynomial. More precisely, given a two variable
polynomial Q(x, y) with highest degree d, we define
x y
P (x, y, z) = z d Q , .
z z
3 Check out the wikipedia page as well for more explanation [Wiki link]

7
For instance, if Q(x, y) = x + y − 1, then P (x, y, z) = x + y − z; if Q(x, y) = x2 +
y 2 − 1 then P (x, y, z) = x2 + y 2 − z 2 ; if Q(x, y) = x2 − y then P (x, y, z) = x2 − yz,
etc. As lines and circles are the main focus in the olympiad geometry, we consider
polynomials with degree not greater than 2 in the following.

3.1 Projective Lines

A projective line is given by a degree 1 homogeneous polynomial, i.e.

Definition 3.3. A projective line in the projective plane P2R has the
equation

`[a:b:c] : ax + by + cz = 0, where (a, b, c) ∈ R3 \(0, 0, 0)

In other words, a point A = [xA : yA : zA ] lies on `[a:b:c] if and only if

axA + byA + czA = 0.

When there is no ambiguity, we abbreiviate [a : b : c] as the line coordinates of

`[a:b:c] , as opposite to the point coordinates. To understand how this definition
extends the notion of lines in the ordinary sense, we illustrate with an example.

Example 3.4. Consider the projective line `[1:2:3] : x + 2y + 3z = 0, then it

contains all the points [x : y : 1] such that x + 2y + 3 = 0. In other words, all
the points lie on the line x + 2y + 3 = 0 (this is a line in the Euclidean plane)
also lies on `[1:2:3] . But, `[1:2:3] also contains one and only one point at infinity
[−2 : 1 : 0]. Hence the projective line `[1:2:3] extends the line x + 2y + 3 = 0 by
adding the point at infinity [−2 : 1 : 0].

More generally, if a or b is non-zero, then the projective line `[a:b:c] can be viewed
as an ordinary line plus a point at infinity:

`[a:b:c] = {[x : y : 1] | ax + by + c = 0} ∪ [−b, a, 0]

| {z } | {z }
an ordinary line in the Euclidean plane a point at infinity

This justifies the name projective line as it extends the ordinary line.
Example 3.5. Besides all the ordinary lines, the projective plane also includes
an important line: the line at infinity, which has equation z = 0. This line
includes all the point at infinity, and moreover, none of the points in the Eucldiean
plane lies on this line. This line is “fictive” in the sense that we are not able to
visualize it. Nevertheless, it does exist in the projective plane, characterized by
its equation.
Excercise 3.6. With the notation of projective lines `[a:b:c] : ax + by + cz = 0.

(a) Find a point lies both on `[1:2:3] and the line `[3:2:1] ? Is this point unique?
(b) Find a point lies both on `[1:2:3] and the line `[−1:−2:4] . Is this point unique?

8
 (c) Find a point lies both on `[1:2:3] and the line at infinity. Is this point
unique?
(d) Find a point lies both on `[1:2:3] and the line `[−2:−4:−6] ? Is this point
unique?

With the notion of projective lines in hand, we are now ready to present one
important property in the projective plane: two distinct lines always intersect.

Theorem 3.7. Given two distinct projective lines `1 and `2 in the pro-
jective plane, there is one and only one point A lies simultaneously on `1
and `2 . In other words, any two distinct projective lines intersect at one
point.

Proof. Assume that `1 is given by a1 x + b1 y + c1 z = 0 and `2 is given by

a2 x + b2 y + c2 z = 0. Since `1 and `2 are distinct, the vector (a1 , b1 , c1 ) and
(a2 , b2 , c2 ) are not aligned. Let
       
xP a1 a2 b1 c2 − b2 c1
 yP  =  b1  ×  b2  = c1 a2 − a1 c2  (1)
zP c1 c2 a1 b2 − a2 b1

be the cross product of the above two vectors. Therefore xP , yP , zP are not all
zero, which defines a point in the projective plane P = [xP : yP : zP ]. Moreover,

a1 xP + b1 yP + c1 zP = 0, and a2 xP + b2 yP + c2 zP = 0

Hence P ∈ `1 ∩ `2 , which shows the existence of the intersection point.

The uniqueness follows from the fact that the vectors (a1 , b1 , c1 ) and (a2 , b2 , c2 )
forms a plane in R3 . If a point Q = [xQ : yQ : zQ ] lies both on `1 and `2 then the
vector (xQ , yQ , zQ ) is orthogonal to this plane, which is necessarily parallel to the
cross product, hence there is λ such that xQ = λxP , yQ = λyP and zQ = λzP ,
implying that [xQ : yQ : zQ ] = [xP : yP : zP ].4

In other words, we get rid of the notion of parallelism and every two lines
intersect, either in the ordinary plane, or at a point at infinity. A natural question
that follows is in what situation three pairwise distinct lines `1 , `2 and `3 are
concurrent, i.e. passing through a common point.
4A more rigorous proof requires the notion of basis and dimension from linear algebra.

9
 Theorem 3.8. Given three pairwise distinct projective lines `1 , `2 and
`3 in the projective plane, assume that `i has equation ai x + bi y + ci z = 0
for any i ∈ [1, 3]. Then the lines are concurrent if and only if
 
a1 a2 a3
det  b1 b2 b3  = 0, (2)
c1 c2 c3

where by definition
 
a1 a2 a3
det  b1 b2 b3  = a1 b2 c3 + a2 b3 c1 + a3 b1 c2 − a1 b3 c2 − a2 b1 c3 − a3 b2 c1 .
c1 c2 c3

Proof. The lines are concurrent if and only if `3 pass through the intersection of
`1 and `2 . Namely, their intersection point (xP , yP , zP ), given by Equation 1,
lies on `3 if and only if
a3 xP + b3 yP + c3 zP = a3 (b1 c2 − b2 c1 ) + b3 (c1 a2 − a1 c2 ) + c3 (a1 b2 − a2 b1 ) = 0,
re-organizing the terms recovers the expression of determinant.
Excercise 3.9. Determine the point of intersection between any pair of lines,
then determine whether they are concurrent. Justify the answer using the value
of determinant given in Equation 2.
• `[1:0:0] , `[0:1:0] , `[0:0:1] .
• `[1:2:1] , `[1:2:2] , `[1:2:3] .
• `[1:2:1] , `[0:0:1] , `[3:6:4] .
Now we have a characterization for concurrent lines, it is natural to ask what
about collinear points. Before that, we start with a seemingly evident result.

Theorem 3.10. Given two distinct points A,B in the projective plane
P2R , there is one and only one projective line pass through A and B.

Proof. The proof is essentially the same as the proof of the Theorem 3.7. Let
A = [xA : yA : zA ] and B = [xB : yB : zB ], let
     
a xA xB
 b  =  yA  ×  yB  (3)
c zA zB
then the line `[a:b:c] defined by the equation ax + by + cz = 0 passes through
the two points A and B. The uniqueness follows similarly as in the proof of
Theorem 3.7.

10
 Theorem 3.11. Given three distinct points A,B, C in the projective
plane P2R , they are collinear if and only if
 
xA xB xC
det  yA yB yC  = 0, (4)
zA zB zC

Excercise 3.12. Determine the equation of lines AB, BC and CA, then deter-
mine whether they are collinear. Justify the answer using the value of determinant
given in Equation 2.
• A = [1 : 0 : 0], B = [0 : 1 : 0], C = [0 : 0 : 1].
• A = [1 : 2 : 1], B = [1 : 2 : 2], C = [1 : 2 : 3].
• A = [1 : 2 : 1], B = [0 : 0 : 1], C = [3 : 6 : 4].
It is worth noticing that the properties of lines and the property of points are
somehow determined by the same equation/formula. Such similarity is called
duality between the points and lines in the projective plane. We will make
it clearer later in Section 5. Briefly speaking, there is a one-to-one (bijective)
mapping

φ : point in P2R → line in P2R

[a : b : c] 7→ `[a:b:c] .

between the point coordinates and the line coordinates. The line pass through
A and B is given by φ(A × B),5 and, the intersection of φ(X) and φ(Y ) is
X × Y . This allows us to use the same three-coordinate system to represent
and manipulate projective lines. All we need is to evaluate the cross product or
determinant. When there is no ambiguity, we say that a line AB has coordinates
[a : b : c] to refer the line `[a:b:c] .
Note that the notion of parallelism no longer present in P2R , as every pair
of lines in the projective plan intersect. Nevertheless, we do have the notion of
parallel lines in the Euclidean plane, which we could naturally extend into the
projective language:
Proposition 3.13 (Parallelism and orthogonality). Let `[a:b:c] be a line in the
projective space with a 6= 0 or b 6= 0.
• All the lines parallel to `[a:b:c] passes through P∞ = [b : −a : 0].
• All the lines perpendicular to `[a:b:c] passes through Q∞ = [a : b : 0].
In particular, if A = [xA : yA : zA ] be a (finite) point, then
• The line through A parallel to `[a:b:c] is the line AP∞ , given by `[azA :bzA :−axA −byA ] .
5 Here we abbreviate A × B as the cross product of their homogeneous coordinates

11
 • The line through A perpendicular to `[a:b:c] is the line AQ∞ , given by `[bzA :−azA :bxA −ayA ] .
Proof. The restriction of the line `[ a : b : c] in the Euclidean plane is ax+by +c =
0. Hence the result follows from the fact that all the parallel to it has the equation
ax+by+d = 0 and all the perpendicular to it has the equation bx−ay+d = 0.

4
! The notion of parallelism and orthogonality are properties from the Eu-
clidean plane, it is necessary that the lines we consider are finite. It make no
sense to talk about parallelism and orthogonality regarding the line at infinity.

While defining the distance (or metric) is possible on the projective plane,
it has a quite different nature compared to the usual distance in the Euclidean
plane. Hence, instead defining points based on the underlying distance, it is
more convenient to define points as image of linear transformations, such as
homothety or rotation. For instance, if we want to introduce the midpoint M of
a segment AB, then we can state M as the image of B under the homothety of
center A with ratio 1/2.
Proposition 3.14 (Homothety). Let A = [xA : yA : zA ] be a finite point. The
homothety of center A with ratio λ 6= 0 is given by the following formula
      
xM λzA 0 (1 − λ)xA xM (1 − λ)xA zM + λzA xM
H =  yM  →  0 λzA (1 − λ)yA   yM  =  (1 − λ)yA zM + λzA yM 
zM 0 0 zA zM zA zM
In particular, λ = −1 gives the formula for reflexion with respect to a given
point.
Similarly, the rotation with a finite center is given by the following transformation:

Proposition 3.15 (Rotation). Let A = [xA : yA : zA ] be a finite point. The

rotation of center A with angle θ is given by the following formula
    
xM zA cos θ −zA sin θ −xA cos θ + yA sin θ + xA xM
R =  yM  →  zA sin θ zA cos θ −xA sin θ − yA cos θ + yA   yM 
zM 0 0 zA zM
These basic computations provide us a straightforward method (based on
linear transformation) to bash geometry problems that only involve lines, in a
way similar to cartesian coordinates. Usually, computing all the coordinates in a
problem is very tedious and time-consuming. Even though it is a painful task,
let us perform such expensive computation once in our life, which will facilitate
and motivate our later discussion.
Problem 3.16 (USA TST 2012 P1). In acute triangle ABC, ∠A < ∠B and
∠A < ∠C. Let P be a variable point on side BC. Points D and E lie on sides
AB and AC, respectively, such that BP = P D and CP = P E. Prove that as P
moves along side BC, the circumcircle of triangle ADE passes through a fixed
point other than A.

12
 The first step we conduct is to translate the problem in a way that it only
involves lines and linear transformations. This step is very important in the
sense that it provides us the mechanism to generate the points in the problem
in a projective way. In this problem, we need to a) appropriately define D and
E; b) translate the statement into concurrency of lines.
• D can be obtained as follows: let PB the orthogonal projection from P to
the line AB, then D is the image of PB under the homothety at point B
with ratio 2.
• E can be defined similarly with respect to the point C.
• The concyclic property is a bit trickier. After trying some special cases,
it is not hard to figure out that the fixed point is the orthocenter H of
4ABC. Hence, the statement that “A, D, E, H are concyclic” is equivalent
to “the perpendicular bisectors of AD, AE and AH are concurrent”. In
this way, the problem can be stated without involving any circle

Proof. We provide detailed coordinates for all the points described above.
• Without loss of generality, we assume that BC is the x-axis with B = [0 : 0 : 1]
and C = [1 : 0 : 1]. The variable point P lies on BC, has coordinates
P = [t : 0 : 1].
• A = [xA : yA : 1] where yA 6= 0. (otherwise A lies on BC)
• The line AB and AC are determined by
           
xA 0 yA xA 1 yA
AB :  yA  × 0 = −xA  AC :  yA  × 0 = 1 − xA 
1 1 0 1 1 −yA

13
• The lines perpendicular to AB all pass through the point C ⊥ = [yA : −xA : 0].
• The lines perpendicular to AC all pass through the point B ⊥ = [yA : 1−xA : 0]
• The line through P perpendicular to the side AB and AC are respectively
P C ⊥ and P B ⊥ , given by
           
t yA xA t yA xA − 1
P C ⊥ : 0×−xA  =  yA  , P B ⊥ : 0×1 − xA  =  yA 
1 0 −txA 1 0 t(1 − xA )

• The orthogonal projection from P to AB and AC are respectively PB =

AB ∩ P C ⊥ and PC = AC ∩ P B ⊥ , given by
tx2A
     
yA xA
PB = AB ∩ P C ⊥ : −xA  ×  yA  =  txA yA 
0 −txA x2A + yA2

t(1 − xA )2 + yA
2
     
yA xA − 1
PC = AC ∩ P B ⊥ : 1 − xA  ×  yA  = yA (1 − xA )(1 − t)
−yA t(1 − xA ) (1 − xA )2 + yA
2

• The homothety at B of ratio 2 sends PB to D, and the homothety at C of

ratio 2 send PC to E.
2tx2A (2t − 1)(1 − xA )2 + yA
2
   

D =  2txA yA  , E =  2yA (1 − xA )(1 − t) 

x2A + yA2
(1 − xA )2 + yA
2

• The homothety at A of ratio 1/2 sends D, E to the midpoint of AD and AE,

denoted by MD and ME respectively
xA (x2A + yA
2
(1 + xA )[(1 − xA )2 + yA2
] + 2(t − 1)(1 − xA )2
   
+ 2txA )
MD =  yA (x2A + yA2
+ 2txA )  , ME =  yA [(1 − xA )2 + yA 2
] + 2yA (1 − xA )(1 − t)] 
2(x2A + yA
2
) 2[(1 − xA )2 + yA 2
]

• The perpendicular bisector of AD and AE are MD C ⊥ and ME B ⊥ , given by

xA (x2A + yA 2
   
+ 2txA ) yA
MD C ⊥ =  yA (x2A + yA 2
+ 2txA )  × −xA 
2 2
2(xA + yA ) 0
2 2
   
2xA (xA + yA ) 2xA
= 2yA (x2A + yA2
) = 2yA 
−(xA + yA + 2txA )(x2A + yA
2 2 2
) −(x2A + yA
2
+ 2txA )

(1 + xA )[(1 − xA )2 + yA2
] + 2(t − 1)(1 − xA )2
   
yA
ME B ⊥ =  yA [(1 − xA )2 + yA2
] + 2yA (1 − xA )(1 − t)]  × 1 − xA 
2[(1 − xA )2 + yA2
] 0
2 2
   
−2(1 − xA )((1 − xA ) + yA ) 2(xA − 1)
= 2yA ((1 − xA )2 + yA 2
) = 2yA 
(1 − xA − yA + 2(t − 1)(1 − xA ))((1 − xA )2 + yA
2 2 2
) 1 − x2A − yA
2
+ 2(t − 1)(1 − xA )

14
 • The altitudes BB ⊥ and CC ⊥ are given by
           
0 yA xA − 1 1 yA xA
⊥    ⊥
BB : 0 × 1 − xA  =  yA  , CC : 0×−xA  =  yA 
1 0 0 1 0 −xA

• The orthocenter H = BB ⊥ ∩ CC ⊥ ,
       
xA − 1 xA −xA yA xA yA
H = BB ⊥ ∩CC ⊥ :  yA × yA  = −xA (1 − xA ) = xA (1 − xA )
0 −xA −yA yA

• The midpoint AH is obtained the image of H under homothety at A with

ratio 1/2  
2xA yA
2
MH = xA (1 − xA ) + yA
2yA

• Since AH ⊥ BC, the perpendicular bisector of AH is given by

     
2xA yA 1 0
MH A⊥ = xA (1 − xA ) + yA
2
× 0 =  −2yA 
2
2yA 0 xA (1 − xA ) + yA

Finally, the perpendicular bisector of AD, AE and AH are concurrent if and

only if
 
2xA 2(xA − 1) 0
det  2yA 2yA −2yA  = 0,
−(x2A + yA
2
+ 2txA ) 1 − x2A − yA2 2
+ 2(t − 1)(1 − xA ) xA (1 − xA ) + yA

Expanding the expression yields a degree 2 polynomial of t, and we can check

that each coefficient is zero, which finishes the proof.
Remark 3.17. Another way to see the determinant vanishes is to remark that
(xA − 1)MD C ⊥ − xA ME B ⊥ = MH A⊥ .

15
 Important Discussion:

While the exact computation is quite tedious and pretty time consuming,
there is one important point worth highlighting here. When the moving
point P moves along the line BC, the other points/lines move accordingly,
with their coordinates are polynomials of the variable t. More
concretely, when viewing xA and yA as constants, all the coordinates
(points and lines) described above are degree 1 polynomial of t. And
the final collinearity is to check a polynomial of degree 2, say Q(t), is
identically zero. Hence, if we are able to prove the statement for three
different values of t, then from the fundamental theorem of algebra,
the polynomial Q(t) must be identically zero! This observation is quite
amazing in the sense that we only need to check few special cases to prove
a such complex statement. For this problem, it suffices to check that

• When P is the midpoint of BC, then D, E are the foot of B and C,

hence A, D, E, H are concyclic.
• When P is B, then D = B, E is the reflexion point of C with respect to
BH, let HB be the foot of B on AC, then BH E·BH A = BH C ·BH A =
BH H · BHB , implying that A, E, H, B(= D) are concyclic.

• When P is C works similarly.

With that said, the problem becomes almost trivial under this method.
The crucial step is to determine the degree of different points, also the
degree of the statement. While as in this example, the polynomials all
have degree one, which is the best scenario, mistakes and errors can be
easily drawn when circles are involved. As a suggestion, I recommend
extreme cautious and rigorous regarding this step!

3.2 Circles
In this section, we introduce the equation of circles:

Definition 3.18. A circle in the projective plane P2R has the equation

C(a,b),r : (x − az)2 + (y − bz)2 = r2 z 2 , where (a, b) ∈ R2 , r ∈ R+ .

In particular, the circle C(a,b),r does not pass any point at infinity, because if
z = 0 then necessarily x = y = 0 and [0 : 0 : 0] is not a point in P2R .6 This
6 If complex number are allowed, then any circle pass through two fix points [1 : i : 0] and

[1 : −i : 0] at infinity.

16
means C(a,b),r lies entirely in the Euclidean plane, and, it corresponds exactly to
the circle with center (a, b) and radius r, i.e. (x − a)2 + (y − b)2 = r2 . The key
proposition that allows us to deal with circles is the following result.

Proposition 3.19 (Second intersection point). Let A = [xA : yA : zA ]

be a point on the circle C(a,b),r , then for any B = [xB : yB : zB ] ∈ P2R , the
second point of intersection C between the line AB and the circle is given
by    
xA xB
C = αB  yA  − 2βAB  yB  , (5)
zA zB
where

αB = (xB − azB )2 + (yB − bzB )2 − r2 zB

2
,
βAB = (xA − azA )(xB − azB ) + (yA − bzA )(yB − bzB ) − r2 zA zB .

The proof is rather straightforward by checking the constructed point C

actually lies on the circle, which boils down to simple algebraic calculations. I
recommend you to check it out by yourself as an exercise. Extreme rigorous is
required when applying this result, as highlight below:

4
! To apply this formula, the point A must lie on the circle!

4
! In the simplest case where A is a given point on a fixed circle, the
degree of C is twice the degree of B. Hence the degrees could quickly
expode.
4! When applied in a general setting, must keep in mind that both
the center (a, b) and the radius r2 could also be polynomial of t.

17
4 Rational parametrization
In the previous section, we have defined a projective curve as the solution set of
a homogeneous polynomial, i.e. P (x, y, z) = 0. This definition is easy to use for
checking whether a point belongs to the curve: given a point [x0 : y0 : z0 ], we
just need to evaluate P (x0 , y0 , z0 ), and check whether it vanishes. However, a
down side of this definition is that it is not clear how one can draw the curve
directly from the formula: imagine that you are asked to draw the curve defined
by P (x, y, z) = x3 − xz 2 + z 3 − y 2 z, you will be in trouble if you are not familiar
with elliptic curves. This motivates the study of parametrization which provides
a way to explicitly draw the curve:

Definition 4.1 (Rational parametrization). Given (single variable)

polynomials U (t), V (t), W (t) with gcd(U, V, W ) = 1, the set of points
[U (t) : V (t) : W (t)] for any t ∈ R∪{∞} defines a rational parametrization
of a projective curve.

While as the notion is very natural and straightforward, there are multiple
subtleties that we would like to highlight for rigorous of the presentation:
Value at ∞ The definition allows taking t = ∞, which deserves a special
attention. A seemingly natural way to define the value at infinity is to take the
limit when t goes to infinity. However, the limit of a non-constant polynomial at
infinity is always infinite, which is not very helpful for defining a point. Here
is the place where homogeneity comes into play, the representation refers to
the same point when dividing all the coordinates by a same value: let d be the
maximum degree of the polynomial among P, Q and R, then we define
 
P (t) Q(t) R(t)
[P (∞) : Q(∞) : R(∞)] = lim d : lim d : lim d .
t→∞ t t→∞ t t→∞ t

Concretely, if deg(P ) < deg(S), then the limit is zero; else deg(P ) = deg(S), the
limit is the value of leading coefficients.
Example 4.2. Consider the following parametrization
• consider [P (t) : Q(t) : R(t)] = [1 : t : t + 1], when t 6= 0, we can rewrite
 
1 t+1
[1 : t : t + 1] = :1: ,
t t
in which case the coordinates are rational function of t, justifying its name
“rational parametrization”. When t → ∞, the limit point is [0 : 1 : 1].
• consider [P (t) : Q(t) : R(t)] = [t2 − 1 : 2t : t2 + 1], when t 6= 0, we can rewrite
 2
t − 1 2 t2 + 1

2 2
[t − 1 : 2t : t + 1] = : : .
t2 t t2
When t → ∞, the limit point is [1 : 0 : 1].

18
Irreducibility gcd(U, V, W ) = 1 The property of irreducibility ensures that
U, V, W does not vanish simultaneously at any value of t. Otherwise, assume that
U (t0 ) = V (t0 ) = W (t0 ) = 0, then t − t0 is a common factor of U, V, W , violating
the assumption. Hence the irreducibility guarantees that [U : V : W ] 6= [0 : 0 : 0]7 ,
which is important since [0 : 0 : 0] is not a point in the projective plane!
On the other hand, given any triple of polynomials U, V, W , we can always
turn them into an irreducible form U0 , V0 , W0 . Hence we abuse the notation
[U : V : W ] as defined by their irreducible representations [U0 : V0 : W0 ]. To
show a concrete example: it is clear that [−t : t2 : −t] can be reduced into
[−1 : t : −1]. However, if we assign t = 0 in the expression [−t : t2 : −t], all the
coordinates are zero which is not a point in the projective plane. In contrast,
t = 0 is well defined in [−1 : t : −1]. Hence we should not take the value literally
in a reducible representation, we need to first reduce it into the irreducible form.

Why [U : V : W ] defines a projective curve? By definition a projective

curve is the solution set of a homogeneous polynomial. Thus it is not trivial at
all that given any U, V, W , we can always find a non-zero polynomial P such that
P (U (t), V (t), W (t)) = 0 for any t. (Try for example U = t2 + 1, V = t2 + 2t − 2,
W = 1) Such finding process on the appropriate P is called implicitization. The
procedure is constructive is based on resultant of polynomials which we will omit
here. For whom are interested, please check chapter 17 in [7]. This means given
any U, V, W , it is always possible to find P such that P (U (t), V (t), W (t)) = 0,
which justifies the claim on projective curves.

Is any projective curve rational parametrizable? Unfortunately, this is

not true, for example elliptic curves are not rational parametrizable (non-trivial).
Hence rational curves is only a subset of projective curves:

Definition 4.3. A projective curve C (defined by homogeneous poly-

nomial) in P2R is rational (parametrizable) if there are polynomials
U (t), V (t), W (t) with gcd(U, V, W ) = 1 such that
• [U (t) : V (t) : W (t)] lies on C for any t ∈ R ∪ {∞};
• For any point A on the curve C, there is t ∈ R ∪ {∞} such that
A = [U (t) : V (t) : W (t)].
The degree of the parametrization is the maximum degree of U, V, W .

Theorem 4.4. A projective line is degree 1 rational parametrizable.

7 Indeed we also need to prove that the point at ∞ is not [0 : 0 : 0], which is clear from the

previous paragraph.

19
First proof: direct parametrization. Given the line `[a:b:c] , we provide a direct
parametrization case by case:
• If a = b = 0, then the line is the line at infinity, where [t : 1 : 0] is a
parametrization.
• If a =6 0, then the point [− ac : 0 : 1] is its intersection with the x-axis. Then
c
[− a : 0 : 1] + t[−b : a : 0] gives an appropriate parametrization.

• If b 6= 0, then the point [0 : − cb : 1] is its intersection with the y-axis. Then

[0 : − cb : 1] + t[−b : a : 0] gives an appropriate parametrization.
In all cases, we have a degree one parametrization.
Second proof by projection. The main purpose of this second proof is to introduce
a fundamental idea of projective geometry, fix a center and project from a line.
We start with the parametrization of the infinite line `∞ , given by [t : 1 : 0].
Now consider a different line `[a:b:c] (this implies that either a 6= 0 or b 6= 0).
We are going to create a bijection between the infinite line and the given line,
leading to a parametrization. To do so, we take an arbitrary point A, that does
not lie on either of the lines, as the projection center. Then for any point X
from the infinite line, draw the line AX, intersecting `[a:b:c] at one and only one
point AX ∩ `[a:b:c] . The function X → AX ∩ `[a:b:c] , viewed as a projection of
center A, gives a bijection8 of the infinite line and the line `[a:b:c] .
To provide a concrete example, let me take A = [−ca + a : −cb + b : a2 + b2 ].
(Exercise: justify that A lies on neither of the lines) Consider the following
projection

f : `∞ → `[a:b:c]
X 7→ AX ∩ `[a:b:c]

Given a moving point X = [t : 1 : 0] on the infinite line, we have

−(a2 + b2 )
     
−ca + a t
AX =  −cb + b  × 1 =  (a2 + b2 )t 
a2 + b2 0 (1 − c)(a − bt)

Therefore
−(a2 + b2 )
    2
(a c + b2 )t − (1 − c)ab
 
a
AX ∩ `[a:b:c] =  (a2 + b2 )t  ×  b  = −(1 − c)abt + (a2 + b2 c) ,
(1 − c)(a − bt) c −(a2 + b2 )(at + b)

which gives a degree 1 parametrization.

Example 4.5. Given the formula in the second proof, the x-axis y = 0 (`[0:1:0] )
is parametrized by [t : 0 : −1], which is clearly an appropriate (bijective)
parametrization. The limit when t → ∞ gives [1 : 0 : 0].
8 Exercise: prove that this is indeed a bijection

20
 Theorem 4.6. A circle is degree 2 rational parametrizable.

First proof: direct parametrization. Given a circle C(a,b),r centered at (a, b) with
radius, the parametrization [a(t2 + 1) + r(t2 − 1) : b(t2 + 1) + 2rt : t2 + 1] is an
appropriate parametrization of the circle.
Second proof by projection. We approach in a similar way as the projections of
lines, this time the projection center A needs to lie on the circle. Let us take the
point A = [a − r : b : 1] lying on the circle. Consider the following projection

f : `∞ → C(a,b),r
X 7→ second intersection of AX and C(a,b),r

which is bijective. Now take any point X = [t : 1 : 0] on the infinite line, then
using the second intersection formula in (5), we have

αB = (xB − azB )2 + (yB − bzB )2 − r2 zB

2
= t2 + 1
βAB = (xA − azA )(xB − azB ) + (yA − bzA )(yB − bzB ) − r2 zA zB = −rt.

This gives the parametrization

   2
a(t + 1) + r(t2 − 1)
  
a−r t
αB  b  − 2βAB 1 =  b(t2 + 1) + 2rt  ,
1 0 t2 + 1

which is indeed the same parametrization we provided in the direct parametriza-

tion.

Example 4.7. Given the formula in the projection proof , the unit circle C(0,0),1
is parametrized by [t2 − 1 : 2t : t2 + 1], which is clearly an appropriate (bijective)
parametrization. The limit when t → ∞ gives [1 : 0 : 1].

Definition 4.8 (Moving point). For abbreviation, we call any rational

parametrization [U (t) : V (t) : W (t)] a moving point.

The key idea underlying what we have discussed is to transfer geometry

object into algebraic object, i.e. polynomials, and, geometric properties into
algebraic identities. Hence whenever we refer to point/line, we always regard
it as a moving point/line. A fixed point simply refers to constant polynomials
U, V, W . The general idea is to start with a moving point on a line or a circle
and construct the others accordingly:

21
 Proposition 4.9. A list of useful results are given in the following:
• A fixed point (do not depend on the moving point) has degree 0.
• If A has degree dA and B has degree dB , then the line AB has degree
at most dA + dB . [Hint: apply (3)]

• If A has degree dA and B has degree dB , then the midpoint of AB

has degree at most dA + dB . [Hint: apply homothety transformation
as in Prop 3.14]
• If `1 has degree d1 and `2 has d2 , then their intersection has degree at
most d1 + d2 . [Hint: apply (1)]

• If A has degree dA on a fixed circle C(a,b),r , B has degree dB , then

the second intersection C = AB ∩ C(a,b),r has degree at most dA + 2dB .
[Hint: apply (5)]

4
! The formula of the final bullet is not symmetric in A and B !

• If A is a fix point, then C = AB ∩ C(a,b),r has degree 2dB .

• If B is a fix point, then C = AB ∩ C(a,b),r has degree dA .

While as the degree of the points/lines are usually increasing, there is one
special case where the degree can be reduced: drawing lines between points on
the circle. Let’s start with an example.
Example 4.10. Consider the point A = [1 : 0 : 1] and a moving point P on the
unit circle, parametrized by [t2 − 1 : 2t : t2 + 1], then the line AP is given by
  2   
1 t −1 −2t
AP = 0 ×  2t  =  −2 
1 t2 + 1 2t

which indeed has degree one while P has degree 2!

Theorem 4.11. Given two distinct moving points on the unit circlea
A = [P (t) : Q(t) : R(t)] and B = [U (t) : V (t) : W (t)], then the line AB
has degree at most dA +d
2
B
.
a Here we mean that for any t, the point A and B always lie on the unit circle, in

other words, P 2 (t) + Q2 (t) = R2 (t) for any t, and, U 2 (t) + V 2 (t) = W 2 (t) for any t.

Proof. We start with the following lemma:

22
Lemma 4.12. Let P, Q, R by polynomials satisfying P 2 (t) + Q2 (t) = R2 (t) with
degree at most d and gcd(P, Q, R) = 1. Then there exist polynomial S and T
2 2 2 2
with degree at most d/2 such that P = S −T
2 , Q = ST , and R = S +T 2 .
Proof of lemma. We have Q2 = R2 − P 2 = (R − P )(R + P ). Note that gcd(R −
P, R + P ) = gcd(R, P ) = 1, hence both R − P and R + P are squares of
polynomials. Thus there exists S and T such that R − P = T 2 , R + P = S 2 and
Q = ST . As a consequence 2deg(S) ≤ d and 2deg(T ) ≤ d.
According to the lemma, there are S, T such that [P : Q : R] = [S 2 − T 2 :
2ST : S 2 + T 2 ] and X, Y such that [U : V : W ] = [X 2 − Y 2 : 2XY : X 2 + Y 2 ].
Then the line AB is given by
     2
S − T2
  2
X −Y2
  
P U 2(T X − SY )(SX − T Y )
AB = Q× V  =  2ST × 2XY  = 2(T X − SY )(T X + SY )
R W S2 + T 2 X2 + Y 2 2(SX + T Y )(SY − T X)
Note that T X − SY is not identically zero, otherwise the two moving points
are identical for any t. Removing the common factor yields AB = [SX − T Y :
SY + T X : −SX − T Y ], which has degree at most max(deg(S), deg(T )) +
max(deg(X), deg(Y )) ≤ dA +d
2
B
.
Example 4.13. Consider the point A = [3 : 4 : 5] and a moving point P =
[t2 − 1 : 2t : t2 + 1] both on the unit circle, then the line AP is given by
  2   2   
3 t −1 4t − 10t + 4 2(t − 2)(2t − 1)
AP = 4 ×  2t  =  2t2 − 8  =  2(t − 2)(t + 2) 
2
5 t +1 −4t2 + 6t + 4. −2(t − 2)(2t + 1).
In particular, when t = 2, the point P = A, which implies t − 2 is a common
factor of the coordinates. Factoring out t − 2 gives a degree one parameterization
of the line AP .

If we take t = 2 directly in the equation of AP , we get [0 : 0 : 0] which is

not a point in P2R . This happens as the expression is reducible, where t−2 is
a common factor. This shows the importance of the irreducibility.

A question raised naturally: what does AP refer to when P = A? To

answer it, we need to adapt the algebraic perspective where A and P are
functions of t. While as A is a constant function, P moves on the circle.
Thus “AP at P = A” is the limit of AP when P approaches A, which is
also known as AA, the tangent line at A.

In other words, AP not only relies on the absolute position of A and P ,

but also depend on how they move. In this simple case that A is fixed,
AP depends on the trajectory of P in the neighborhood of A. Hence, even
though P coincides with A when t = 2, the line AP is still well defined.
This is a fundamental concept that will come back again and again.

23
 Lemma 4.14 (Coincidence Lemma). Given two non-identical moving
point A = [P (t) : Q(t) : R(t)] and B = [U (t) : V (t) : W (t)], if A = B
for k different values of t, then the line AB has degree at most deg(A) +
deg(B) − k.

Proof. Let ti be one of the values such that A = B. Then AB evaluated at ti is

[0 : 0 : 0], meaning that t − ti is a common factor of the coordinates AB.
Q This is
true for any ti such that A = B, therefore AB can be factorized by (t − ti )
with degree at most deg(A) + deg(B) − k.

Remark 4.15. In some cases, we might have a root t0 with multiplicity larger
than 1, i.e. (t − t0 )n is a common factor of the coordinates AB. However, this
is in general hard to check without deriving the exact formula of AB.

Remark 4.16. This lemma is extremely useful for reducing the degree of points/-
lines. Indeed, Example 4.13 is just a special case of applying the Coincidence
Lemma, as P coincides with A at value t = 2 ([t2 − 1 : 2t : t2 + 1]t=2 = [3 : 4 : 5]).

Degree of statement As we have seen in Problem 3.16, the final step of the
proof is to check a polynomial is identically zero. We call the degree of such
polynomial as the degree of the statement.
• If A, B, C has degree dA , dB , dC , then the statement A, B, C are collinear
is equivalent to “a degree at most dA + dB + dC polynomial is identically
zero”. [Hint: apply (4)]
• If `1 , `2 , `3 has degree d1 , d2 , d3 , then the statement `1 , `2 , `3 are concurrent
is equivalent to “a degree at most d1 + d2 + d3 polynomial is identically
zero”. [Hint: apply (2)]
For statement such as “A, B, C, D are concyclic”, one way to reformulate it is
“the perpendicular bisector of AB, AC and AD are concurrent”. This turns out
to be a concurrency statement after carefully evaluating the degree of each line.
We will see a more advanced characterization for concyclic statement later based
on angle conditions in Corollary 5.14.

Lemma 4.17 (Coincidence Lemma for statement). Given three moving

points A, B and C, if A = B for k different values of t, then the statement
“A, B, C are collinear” has degree at most deg(A) + deg(B) + deg(C) − k.

24
 General routine to apply the method of moving point is as follows
1. Formulate the problem in the language of projective geometry. (For
example, use transformations to describe the problem);
2. Determine a moving point in the problem. Animate it along a line/-
circle, which define the initial parametrization (have degree one/two
respectively).
3. Carefully evaluate the degree of the points/lines encountered in the
problem;
4. Bound the degree d of the polynomial corresponding to the statement;

5. Inspect whether coincidence lemma can be applied to reduce the

degree of the statement, assume that k coincidences are found.
6. Find d − k + 1 distinct values of the parameter t ∈ R ∪ {∞} where
we can easily check the problem is true. This implies that the desired
polynomial is identically zero.

Now we are ready to apply the moving point method, let us first use it to show
some classic theorems.

Theorem 4.18 (Pappus’s Theorem). Given two distinct lines `1 and `2 . Let
A1 , B1 , C1 be arbitrary points on `1 , A2 , B2 , C2 be arbitrary points on `2 , then
the intersections D = A1 B2 ∩ B1 A2 , E = B1 C2 ∩ C1 B2 and F = C1 A2 ∩ A1 C2
are collinear.
Proof. We fix the points A1 , B1 , C1 , A2 , B2 and let C2 be a moving point on the
line `2 , which has degree 1. We start by evaluating the degrees of all the points
we care about:
• The line A1 B2 , A2 B1 , C1 B2 and C1 A2 are fixed, hence D = A1 B2 ∩ B1 A2 is
fixed.
• The line B1 C2 has degree deg(B1 ) + deg(C2 ) = 0 + 1 = 1, hence E as the
intersection of B1 C2 and C1 B2 has degree deg(B1 C2 )+deg(C1 B2 ) = 1+0 = 1.
• Similarly, A1 C2 has degree 1 implying that F is degree 1.
Therefore, the statement D, E, F are collinear has degree deg(D) + deg(E) +
deg(F ) = 0 + 1 + 1 = 2. We start by checking whether we can reduce the
statement degree via coincidence lemma. Indeed,
• When C2 = `1 ∩ `2 (which always exist in P2R ), we have E = B1 C2 ∩ C1 B2 =
`1 ∩C1 B2 = C1 , similarly F = C1 . Hence by coincidence lemma for statement,
the degree of the statement reduces by 1.
Therefore the degree of the statement is 1, it suffices to check two special cases:

25
 • When C2 = A2 , we have F = C1 A2 ∩ A1 C2 = C1 A2 ∩ A1 A2 = A2 . Note that
E = B1 C2 ∩ B2 C1 lies on B1 C2 = B1 A2 and D lies on B1 A2 by definition,
we conclude that D, E, F all lies on B1 A2 .
• When C2 = B2 , we obtain similarly that D, E, F all lies on A1 B2 .

Theorem 4.19 (Pascal’s Theorem). Given a circle ω and let ABCDEF be

arbitrary points on ω. Then the intersections X = AB ∩ DE, Y = BC ∩ EF
and Z = CD ∩ F A are collinear.
We leave the proof as an exercise to the readers, which is essentially the same as
Pappus’s theorem. The key is to apply Theorem 4.11 to reduce the degree of
lines as we are working on circles. Next, we show another classic theorem, the
Butterfly Theorem.
Theorem 4.20 (Butterfly Theorem). Let M be the midpoint of a chord P Q
of a circle Ω. Two other chords AB and CD of Ω are drawn, both passing
through M . Let AD and BC intersect chord P Q at X and Y respectively. Then
M is the midpoint of XY .
Proof. We reformulate the problem as follows. Fix the circle Ω, the chord P Q
and CD. Let A be a moving point on the circle, B be the second intersection of
AM and Ω. Let X = AD ∩ BC and X 0 be the reflexion of X with respect to
M . We show that B, X 0 , C are collinear.
We start by evaluating the degrees of different points.
• A moves on Ω, has degree 2.
• As M is fixed, the second intersection of AM and Ω, which is B, has degree
at most deg(A) + 2deg(M ) = 2.
deg(A)+deg(D) 2+0
• As A,D both lie on Ω, AD has degree 2 = 2 = 1.
• The intersection of AD and P Q, which is X, has degree deg(AD)+deg(P Q) =
1 + 0 = 1.
• The reflexion at M , which is fixed, sends X to X 0 , we have deg(X 0 ) =
deg(X) = 1.
• Finally the statement B, X 0 , C are collinear has degree 2 + 1 + 0 = 3
Before we check special cases, we first check whether we can reduce the statement
of the problem. We can easily check that

• When A = P , we have X = A = P and X 0 = Q = B.

• Similarly A = Q gives X 0 = P = B.
• When A = D, we have B = C.

26
Therefore, by coincidence lemma, the determinant of B, X 0 , C (which is a poly-
nomial of t) has degree at most deg(B) + deg(X 0 ) + deg(C) − 3 = 2 + 1 − 3 = 0.
Finally, it suffices to check 1 special case. Indeed,
• When A = C, we have B = D and X = M = X 0 , the collinearity clearly
follows, which finishes the proof.

Remark 4.21 (Important Discussion). This proof involves several

important aspects deserving further clarifications:
• The determinant of B, X 0 , C is a polynomial of t (the moving variable).
It has degree 0 means that the determinant is constant. However, the
constant might not be zero so we still need to check one special case.
• We should be careful not confusing the coincidence lemma with the
special cases, which have very different nature:
• When coincidence occurs, for example B = C, it means that the alge-
braic expression represented by the line BC is reducible (by a common
polynomial factor), hence the degree of the algebraic objects/statement
can be reduced.
• The special cases should be considered in their irreducible forms. In
other words, even though B = C (when A = D in the Butterfly
theorem), the collinearity of B, X 0 , C is non-trivial, which should be
interpreted as: the tangent line at C (also known as CC) passes
through X 0 . The reason we get the tangent line is because B moves
on Ω. As B approaches C, the limit of the line BC approaches
the tangent line at C. Therefore, the theorem takes the following
degenerated form:

Let M be the midpoint of a chord P Q. Another chord CD

passes through M . The tangent line at C and D with respect
to the circle Ω intersects P Q at X and Y respectively. Prove
that M is the midpoint of XY .
Once we prove this statement (which is not hard by drawing a parallel
line at C of DD), we can use A = D as the special case in our final
step instead of A = C. In such case, the point A = D helps us reducing
the degree twice, one for coincidence and the other as special point.
• In the general case, whenever two points S and T coincides, the line
ST refers to the tangent line to whatever the algebraic curve ST
jointly defined. However, characterizing this limit line is in general
difficult. We have a concrete example in our proof: when A = P ,
we have B = Q = X 0 . However characterizing the irreducible form

27
 of BX 0 is not trivial at all. As X 0 moves towards Q on P Q and
B moves on Ω, the limit line BX 0 (when B = X 0 ) depends on the
relative speed between B and X 0 , which is not easy to handle. Since
we are unable to interpret the irreducible form of BX 0 , we can not
use A = P as a special case. Nevertheless, it is used to reduce the
degree via coincidence lemma.
• To summarize, we should be aware that coincidence lemma
is intrinsically different than special cases, even though it is
rare that one can use one point for both purposes (as the
case A = D here).

Problem 4.22 (USA TST 2014 P1). Let ABC be an acute triangle, and let X
be a variable interior point on the minor arc BC of its circumcircle. Let P and
Q be the feet of the perpendiculars from X to lines CA and CB, respectively.
Let R be the intersection of line P Q and the perpendicular from B to AC. Let `
be the line through P parallel to XR. Prove that as X varies along minor arc
BC, the line ` always passes through a fixed point
Proof. Let us fix 4ABC and consider X as a moving point on its circumcircle,
which is degree 2 parametrizable. We reformulate the problem as follows.

“ Let H be the orthocenter of 4ABC and X an arbitrary point of the

circumcircle (ABC). Define A⊥ ,B ⊥ be the intersection of AH,BH and the
line at infinity, respectively. Let P = XB⊥ ∩ AC and Q = XA⊥ ∩ BC. Denote
R = P Q ∩ BH and Z = RX ∩ `∞ , show that H, P, Z are collinear.”

We now evaluate the degree of the corresponding points/lines

28
 • The lines AH, BH are fixed implying that A⊥ and B ⊥ are fix points.
• The line XB ⊥ (XA⊥ ) has degree 2.
• P = XB ⊥ ∩ AC has degree 2, also is Q = XA⊥ ∩ BC.

• The line P Q has degree 4.

• The point R = P Q ∩ BH has degree 4.
• The line RX has degree 6.
• Z = RX ∩ `∞ has degree 6

• The statement H, P, Z are collinear is a degree 0 + 2 + 6 = 8 problem

Hence it suffices to find 9 special cases.
• when X = B, then Q = B, P ∈ BH, hence R ∈ BH and finally Z ∈ BH =
P H. Similarly X = A, C also works.

• when X be the antipole of B, then Q = C, P ∈ AC, hence R = BH ∩ AC

with XP k HR and XP = HR, hence HP k RX. Similarly X be the
anti-pole of A and C also works.
• when X be AH ∩ (ABC), then Q = BH ∩ BC and XQ = QH. Together
with RH k P X implies XP HR is a parallelogram, hence HP k RX. Similarly
X be the intersection of BH, CH with (ABC) works as well.

Remark 4.23. One may wonder that we have to check these special points are
all distinct, which is not always the case. When 4ABC is acute, the points A,
B, C and their anti-poles are distinct. However, AH (ABC) might be the
same point as the anti-pole of A, in which case 4ABC is isosceles. Nevertheless,
we have shown that for any acute non-isosceles triangle, the statement holds. To
show the degenerate case, it suffices to fix X and consider A as a moving point.
On one hand, this can be formulate as a finite polynomial in the coordinates
of A. On the other hand, we have already shown infinite many solutions: for
those A such that 4ABC is acute non-isosceles. Hence this polynomial must be
identically zero.
As we have gone through some positive examples of moving point methods,
we would like to provide a negative example which helps understanding the
limitation of the moving point method.

Theorem 4.24 (Simson Line). Let Ω be the circumcircle of triangle 4ABC.

Let P be an arbitrary point on Ω and denote D, E, F the orthogonal projection
from P to BC, CA, AB respectively. Prove that D,E and F are collinear.

29
Tentative proof. We fix 4ABC and Ω, let P be the moving point, which has
degree 2. As BC is fixed, the orthogonal direction of BC is fixed. Hence the
line ` through P perpendicular to BC has degree the same as P . Therefore
D = ` ∩ BC has degree deg(`) + deg(BC) = 2 + 0 = 2. Similarly, E, F has
degree 2. Therefore, the statement D, E, F are collinear has degree 6.

Note that when P = A, E coincides with F . Similarly when P = B, D

coincides with F ; when P = C, D coincides with E. Hence by coincidence
lemma, the degree of the statement is at most 6 − 3 = 3.

Finally, we need to find 4 special cases:

• When P is the anti-pole of A, we have E = C, F = B and D lies on BC,
hence collinearity holds.
• Similarly for P being the anti-pole of B and C.
However, we still need to find another special cases. I am not able to find another
trivial case (this could be wrong if you can find one). This special case should
not involve angle chasing argument, otherwise we can simply apply angle chasing
on the original problem, without going through the moving point process.
The main reason we fail to apply moving point method in this example is
due to the lack of special cases. It is in general not easy to determine in advance
whether we have enough special points. My recommendation would be always
try to apply angle chasing to see whether some properties can be easily drawn,
in which case, the degree of the statement might be reduced and we get a better
chance to finish up with enough special cases.
A rather surprising thing is that we can prove a generalization of Simson
Line using moving point method. The generalization is as follows:
Theorem 4.25 (Generalization of Simson Line). Let Ω be the circumcircle of
triangle 4ABC. Let Q be an arbitrary point, denote the second intersection
of AQ, BQ, CQ with Ω by A1 , B1 , C1 . Let P be an arbitrary point on Ω and
denote D = P A1 ∩ BC, E = P B1 ∩ CA , F = P C1 ∩ AB. Prove that Q, D, E
and F are collinear.
A quick proof would be to apply Pascal’s theorem on AA1 P B1 BC, which
implies that Q, D, E are collinear. If you have not recognized the Pascal’s
configuration, applying moving point can also easily prove that Q, D, E are
collinear (basically copying the proof of Pascal’s theorem).
To recover the Simson Line, it suffices to take Q to the infinite line such that
the angle between OQ and OP is given by ∠P AB + ∠P BC + ∠P CA. (where
O is the center of Ω). In this case, Q is no longer an arbitrary point, it depends
on P . The surprising thing is that such coupling of P and Q makes the problem
less appealing for moving point method. This example should give you a sense
about the limitation of moving point method, which is less favorite when specific
dependency exists. In the following, we continue our journey by introducing a
way to handle angles in moving point method.

30
5 The space of lines
As you may have already noticed in section 3.1, there is a kind of symmetry
between the collinearity in points and the concurrency between lines, governed
by the same formula of determinant. We would like to clarify such symmetry
in this section, which is formally called duality. To form the duality, we first
introduce the space of lines as the set of all projective lines in P2R :

L2R = {`[a:b:c] | (a, b, c) 6= (0, 0, 0)}9 ,

where `[a:b:c] refers to the line ax + by + cz = 0. A very first thing we realize is

that L2R inherits a homogeneous coordinate system from the projective plane.
Indeed, each triple (a, b, c) determines a line10 , the line determined is unchanged
if the triple is multiplied by a non-zero scalar, i.e. the line `[λa:λb:λc] defined by

λax + λby + λcz = 0 ⇔ ax + by + cz = 0 ∀λ 6= 0.

is the same line as `[a:b:c] ! Therefore [a : b : c] may be taken to be homogeneous

coordinates of a line in the projective plane, which is the line coordinates as
opposed to point coordinates.
Therefore, the space of lines L2R shares the same three-coordinate homoge-
neous representation as the projective plane. While [x : y : z] refers to a point in
projective plane, [a : b : c] refers to a line `[a:b:c] . The fundamental elements in
P2R are points while the fundamental elements in L2R are lines. This leads to the
concept of duality in projective geometry, the principle that the roles of points
and lines can be interchanged in a theorem in projective geometry and the result
will also be a theorem.
To formalize the concept of duality, whenever a homogeneous coordinates is
given, we call that object a point in the associated space. In such terminology,
the coordinates [a : b : c], referring `[a:b:c] , is a point in L2R . Hence, a point in
L2R is a line in P2R . One way to interpret such terminology is that the two spaces
P2R and L2R operates on a different scale: the projective plane is “microscopic”,
operating on points, and, the space of lines is “macroscopic”, operating on lines.
In other words, the space of lines L2R only care about structures and relationships
between lines. Now given that we have defined the points in L2R , we can naturally
define the lines in L2R by

9 This notation L2 is not a standard notation, I introduce it in order to distinguish between

R
point coordinates and line coordinates. In the language of abstract algebra, L2R is isomorphic
to the projective plane P2R , meaning that they share the same underlying structure, see [Wiki
link]. Hence, the space of lines is often directly denoted by P2R . This could be confusing for
someone who are not familiar with isomorphism, so I explicitly stretch the difference between
point coordinates and line coordinates by introducing the notation L2R .
10 at least one of a, b and c must be non-zero

31
 Definition 5.1 (Pencil of lines). A line in L2R is given by the equation

pl[α:β:γ] = {[a : b : c] ∈ L2R such that αa + βb + γc = 0},

where (α, β, γ) 6= (0, 0, 0).

This said, a line in L2R encodes a linear structure between the set of points in
L2R (to emphasize, a point in L2R is a line in P2R ). It is then natural to ask what
pl[α:β:γ] in L2R corresponds to in the perspective of P2R :
Example 5.2. Let us take α = 1, β = 2, γ = 3. From definition, pl[1:2:3]
consists of all the lines `[a:b:c] satisfying a + 2b + 3c = 0. Note that `[a:b:c] consists
of all the points (x, y, z) satisfying ax + by + cz = 0. Therefore `[a:b:c] ∈ pl[1:2:3]
if and only if [1 : 2 : 3] lies on the line `[a:b:c] . This means
pl[1:2:3] = { all the projective lines `[a:b:c] passing through [1 : 2 : 3] },
which is also known as the pencil of lines passing through [1 : 2 : 3].
Example 5.3. Let us take α = 1, β = 0, γ = 0, which gives the pencil of lines
passing through [1 : 0 : 0]. Note that [1 : 0 : 0] is an infinite point, a (Euclidean)
line ` pass through it if and only if ` is parallel to the vector (1, 0). Therefore
pl[1:0:0] = { all the lines parallel to the x-axis } ∪ `∞ .
Readers already familiar with projective geometry may have noticed that in
this way we immediately build the connection between cross-ratio on lines are
nothing but the cross ratio (of points) in the space of line, we will come back
to this later. Usually pencil of lines are defined as the family of lines passing
through a given point. Here, we define the pencil of line as a line in the space of
lines L2R . The two definition do match each other as we show in the example.
The important thing is that the pencil of lines encodes the line structure in the
space of lines, which allows us to derive the dual formulation of all the theorems
in section 3.1. We recall Theorem 3.7:

Theorem 3.7. Given two distinct projective lines `1 and `2 in the pro-
jective plane, there is one and only one point A lies simultaneously on `1
and `2 . In other words, any two distinct projective lines intersect at one
point.

Now we apply it in the space of lines:

Theorem 5.4 (Theorem 3.7 in the space of lines). Given two distinct
lines pl1 and pl2 in the space of lines L2R , there is one and only one point
P = [aP : bP : cP ] ∈ L2R lies simultaneously on pl1 and pl2 .

32
 For completeness, I provide a proof below, which is really just a copy-paste
of the proof on Theorem 3.7.
Proof. Assume that pl1 is given by α1 a + β1 b + γ1 c = 0 and pl2 is given by
α2 a + β2 b + γ2 c = 0. Since pl1 and pl2 are distinct, the vector (α1 , β1 , γ1 ) and
(α2 , β2 , γ2 ) are not aligned. Let
     
aP α1 α2
 bP  =  β1  ×  β2  (6)
cP γ1 γ2

be the cross product of the above two vectors. Therefore aP , bP , cP are not all
zero, which defines a point in the projective plane P = [aP : bP : cP ]. Moreover,

α1 aP + β1 bP + γ1 cP = 0, and α2 aP + β2 bP + γ2 cP = 0

Hence P ∈ pl1 ∩ pl2 , which shows the existence of the intersection point.
The uniqueness follows from the fact that the vectors (α1 , β1 , γ1 ) and (α2 , β2 , γ2 )
forms a plane in R3 . If a point Q = [aQ : bQ : cQ ] lies both on pl1 and pl2 then
the vector (aQ , bQ , cQ ) is orthogonal to this plane, which is necessarily parallel to
the cross product, hence there is λ such that aQ = λaP , bQ = λbP and cQ = λcP ,
implying that [aQ : bQ : cQ ] = [aP : bP : cP ].
Now, if we interpret the above Theorem 5.4 in the projective plane P2R , we
get the dual formulation of Theorem 3.7. Note that a pencil of line is uniquely
determined by its center [α : β : γ], and the point [a : b : c] ∈ pl[α:β:γ] if and only
if αa + βb + γc = 0, which can also be interpreted as the line `[a:b:c] pass through
[α : β : γ]. Therefore, Theorem 5.4 is indeed stating the following result:

Theorem 5.5 (Dual form of Theorem 3.7). Given two distinct points
[α1 : β1 : γ1 ] and [α2 : β2 : γ2 ] in P2R a , there is one and only one line
`[aP :bP :cP ] b pass through both points.
a served as centers of pencil of lines
b served as the intersection point of pl1 and pl2 in L2R .

As a point in L2R is a line in P2R , the role of points and lines get interchanged !
Hence the dual form of Theorem 3.7 is Theorem 3.10 ! The general routine to
dualize a theorem is to restate the result in the space of lines L2R , then reinterpret
it in P2R , which can also be done by simply interchanging the role between points
and lines. For instance,
Theorem 3.7: two line intersects at a point
interchange line and point yields
Theorem 3.10 two points lie (determine) on a common line.

33
Similarly, it is easy to see that the dual form of Theorem 3.8 is Theorem 3.11 !
Such connection also explains why their proofs look quite similar. If we go a bit
further, we can show that several well known theorems are in dual formulation:
• Desargues’ theorem ↔ Converse of Desargues’ theorem,

• Pascal’s theorem ↔ Brianchon’s theorem,

• Menelaus’ theorem ↔ Ceva’s theorem.
In other words, Pascal’s theorem applied on the space of lines gives Brianchon’s
theorem in the projective plane. The derivation is pretty straightforward, which
is left to the readers.
The underlying idea of duality is to take a line as an object, in contract to
points. Such abstraction allows us to manipulate and focus on the structure
between lines. Before moving on, let us recap the main concepts and results:

Space of lines L2R Projective plane P2R

a point [a : b : c] ↔ a line `[a:b:c]

a line pl[α:β:γ] ↔ pencil of lines at center [α : β : γ]
a point lies on a line ↔ a line pass through a point
two lines intersect ↔ two points determine a line
collinear points ↔ concurrent lines

The main idea underlying the space of lines is to manipulate lines as an

object. The fundamental operation between lines is rotation, which also defines
angles. More concretely, given two lines `1 and `2 intersecting in a finite point
O of the Euclidean plane. By the angle ∠(`1 , `2 ), we mean the angle by which
`1 has to be rotated counterclockwise around O until it coincides with `2 . To
properly operate on angles, let us first introduce the following terminology.

Definition 5.6 (Direction of line). Given a finite line ` ∈ P2R , we call

the point ` ∩ `∞ (∈ P2R ) the direction of ` .

First, let us justify that such definition aligns with the usual meaning of
direction. Indeed, given two parallel (Euclidean) lines `1 and `2 , they intersect
on a unique point on the line at infinity, i.e. `1 ∩ `∞ = `2 ∩ `∞ . Therefore,
according to Definition 5.6, they have the same direction, as in the common way.
Second, such definition is orientation invariant and length invariant. As
opposed to the usual definition, where the direction is defined as the unit vector
along the line, we need to impose the length(= 1) and orientation appropriately.

34
Here, due to the property on homogeneous coordinate, we can multiply the
coordinates with an arbitrary scalar λ and it still remains the same point in P2R .
We illustrate it with a concrete example:
Example 5.7. Consider the projective line `1 : x − y + z = 0, associating to the
Euclidean line x − y + 1 = 0. By Definition 5.6, its direction is given by
     
1 0 −1
     
`1 ∩ `∞ =   ×
−1 0 −1 .
  =  

1 1 0

Now consider a second line `2 : −πx+πy+3.14z = 0, associating to the Euclidean

line −πx + πy + 3.14 = 0, its direction is given by
     
−π 0 π
     
`2 ∩ `∞ =  π  × 0 = π 
    
.
3.14 1 0

As in the homogeneous coordinate system, we have [π : π : 0] = [−1 : −1 : 0], the

two lines share the same direction. The different sign and scale does not matter.
More generally, the direction of `[a:b:c] is [b : −a : 0].

With the direction of lines in hand, we are now ready to present the important
rotation lemmas:

Lemma 5.8 (Rotation lemma (fixed angle)). Given a (Euclidean)

line `[a:b:c] and a finite point A = [xA : yA : zA ], perform the rotation Rθ
of angle θ (counterclockwise) at center A, then the image of Rθ (`[a:b:c] )
has direction

Dθ = [b cos θ + a sin θ : −a cos θ + b sin θ : 0].

Moreover, if A lies on `[a:b:c] , then the image of Rθ (`[a:b:c] ) is given by

 
(a cos θ − b sin θ)zA
 
Rθ (`[a:b:c] ) = Adθ : A×dθ = 
 (a sin θ + b cos θ)zA 

(−a cos θ + b sin θ)xA − (a sin θ + b cos θ)yA

In particular, the degree of Rθ (`[a:b:c] ) is at most deg(A) + deg(`[a:b:c] ).

Proof. To derive Dθ , it suffices to remark that rotation sends `[a:b:c] ∩ `∞ to

Rθ (`[a:b:c] ), then apply the rotation formula in Proposition 3.15.

35
 4
! To apply this formula, the angle θ must be fixed (independent of t)!

4
! It is possible to derive a formula when A does not lie on ` ,
[a:b:c]
but we rarely use it since the degree of Rθ (`[a:b:c] ) get tripled, becomes
deg(A) + 3deg(`[a:b:c] ) in such case.

Problem 5.9. Let I be the incenter of 4ABC and D be the foot of I on BC.
Points P, Q lie on lines IB, IC such that ]P A1 Q = 90◦ . Then ]P AQ =
1
2 ]BAC.

Proof. Fix 4ABC and let P be the moving point on IB, which has degree 1.
The line DP , AP has degree 1. Let RD be the rotation of center D with angle
90◦ , and RA be the rotation of center A with angle 12 ]BAC. Then RD (DP )
and RA (AP ) both have degree 1. The statement is equivalent to CI, RD (DP )
and RA (AP ) are concurrent, which is a degree 2 statement. Hence it suffices to
check 3 special cases:

• P = B then Q = I and the statement holds.

• P = I then Q = C and the statement holds.
• P = BI ∩ `∞ . Then DQ ⊥ BI. Let X = AQ ∩ BI, then ]IXA =
](IB, AQ) = ]BIC − 90◦ = 12 ]BAC, which completes the proof.

Remark 5.10. Some applications of this lemma: ISL 2006 G4, All-Russian
2009 10.7,
Problem 5.11 (IMO 2013 Problem 4). Let ABC be an acute triangle with
orthocenter H, and let W be a point on the side BC, lying strictly between B and
C. The points M and N are the feet of the altitudes from B and C, respectively.
Denote by ω1 is the circumcircle of BW N , and let X be the point on ω1 such
that W X is a diameter of ω1 . Analogously, denote by ω2 the circumcircle of
triangle CW M , and let Y be the point such that W Y is a diameter of ω2 . Prove
that X, Y and H are collinear.

Proof. We fix the triangle ABC (which fix N, M, H) and consider W as a moving
point on BC, which has degree 1. We first construct X and Y . As W X is
the diameter of ω1 , we have BW ⊥ BX and N W ⊥ N X. Therefore, X is the
intersection of `1 (the perpendicular line to N W at N ) with `2 the perpendicular
line to BC at B. Note that `1 is obtained by rotating the line N W by 90◦ at
center N (which is fixed), we have deg(`1 ) is at most deg(N ) + deg(N W ) = 0 + 1.
As BC is fixed, `2 is a fixed line. Hence X = `1 ∩ `2 has degree at most 1.
Similarly, Y has degree at most 1. Finally, the collinearity of X, Y , H is a
degree 2 statement, it suffices to show it for three positions of W .

36
 • W = B we have X ∈ CN and Y = C, hence XY = CN pass through H;
• W = C similar to first case;

• W = AH ∩ BC, then H ∈ BW N , implying that ∠XHW = 90◦ . Simi-

larly, H ∈ CW M and ∠W HM = 90◦ , hence H, X, Y are collinear.

The key here is to operates on the lines in order to construct the point X
and Y , which are initially defined as the anti-pole of W in the corresponding
circles. The rotation lemma is extremely useful when dealing with concyclic
points. In this problem, we are in the sweet spot as the angle of rotation is 90◦ ,
which is fixed. Let’s try another harder problem.
Problem 5.12. Let O be the circumcenter of 4ABC. Let Q be an arbitrary
point, denote the second intersection of AQ, BQ and CQ with O as A1 , B1 and
C1 . Let P be an arbitrary point on the line OQ, denote the orthogonal projection
from P to BC, CA, AB as A2 , B2 and C2 . Prove that the circumcircles of
4P A1 A2 , 4P B1 B2 and 4P C1 C2 have two common points. (in other words
coaxial)
Proof. We first translate the coaxial property into a statement of collinearity.
Denote PA , PB , PC as the anti-pole of P with respect to the circumcircles of
4P A1 A2 , 4P B1 B2 and 4P C1 C2 . Then these circles are coaxial iff their centers
are collinear iff PA , PB and PC are collinear. Note that P A2 ⊥ BC, hence PA
lies on BC. Moreover P A1 ⊥ A1 PA , hence PA is indeed the intersection of BC
and the line through A1 perpendicular to P A1 .
Now we are in shape to apply moving point method. We fix 4ABC and Q,
hence A1 , B1 , C1 are fixed. Let P be a moving point on OQ which has degree 1.

37
The lines P A1 has degree 1. Consider the rotation RA of center A1 and angle
90◦ . Then by rotation lemma, as A1 is fixed, RA (P A1 ) has the same degree as
P A1 , which is degree 1. Therefore PA = BC ∩ RA (P A1 ) has degree 1. Similarly,
PB , PC has degree 1. Finally, the statement PA , PB , PC are collinear has degree
3. It suffices to find 4 special cases.

• Let P be one of the intersection QO ∩ O. Denote P 0 the anti-pole of P ,

then P 0 = A1 PA ∩ B1 PB ∩ C1 PC . By Pascal’s theorem to BB1 P 0 C1 CA, we
have Q = BB1 ∩ C1 C, PB = B1 P 0 ∩ CA and PC = P 0 C1 ∩ AB are collinear.
Similarly, Q, PA , PB are collinear. Hence PA , PB , PC are collinear. (2 cases)
• Let P such that P A1 ⊥ CA1 , implying that PA = C. Denote A0 , C 0
the anti-pole of A and C. By Pascal’s theorem on A1 AA0 C1 CC 0 , we have
Q = A1 A ∩ C1 C, O = AA0 ∩ CC 0 and A0 C1 ∩ C 0 A1 are collinear. Note that
Q = OP ∩ C 0 A1 , we have Q is indeed A0 C1 ∩ C 0 A1 . Therefore, PC = A. As
PB lies on AC by definition, we have PA , PB , PC are collinear. (3 cases)

This is a rather difficult problem as the special cases are not completely
trivial, the original problem is even harder to directly prove, see source. Some
applications of this result includes Romania TST 6 2010, P2, AOPS problem. So
far, the rotation we considered all have fixed angles (usually 90◦ ), but in some
cases, the angle of rotation can be itself a variable. This would be useful when
dealing with concyclic points. Let’s see how we proceed in such case.

Lemma 5.13 (Rotation lemma (variable angle)). Let `1 : [a1 : b1 :

c1 ], `2 : [a2 : b2 : c2 ] be two (Euclidean) lines. Given another finite point
A and a line `3 : [a3 : b3 : c3 ] passing through A. Consider the rotation
Rθ at center A of angle θ = ∠`1 `2 , then the image of `3 has direction
 
(a1 b2 − a2 b1 )a3 + (a1 a2 + b1 b2 )b3
 
Dθ =  (a1 b2 − a2 b1 )b3 − (a1 a2 + b1 b2 )a3  .


Therefore Rθ (`3 ) = ADθ has degree at most deg(A) + deg(`1 ) + deg(`2 ) +

deg(`3 ).

a1 a2 +b1 b2 a1 b2 −b2 a1
Proof. It suffices to remark that cos θ = √ and sin θ = √ .
(a21 +b21 )(a22 +b22 ) (a21 +b21 )(a22 +b22 )

38
 Corollary 5.14. Given four lines `1 , `2 , `3 , `4 , let A = `1 ∩ `2 , B =
`3 ∩`4 , C = `2 ∩`4 , D = `1 ∩`3 , then the statement A, B, C, D is concyclic
has degree at most deg(`1 ) + deg(`2 ) + deg(`3 ) + deg(`4 ).

Problem 5.15 (IMO SL 2018 G2). Let ABC be a triangle with AB = AC,
and let M be the midpoint of BC. Let P be a point such that P B < P C
and P A is parallel to BC. Let X and Y be points on the lines P B and P C,
respectively, so that B lies on the segment P X, C lies on the segment P Y , and
∠P XM = ∠P Y M . Prove that the quadrilateral AP XY is cyclic.

39
Proof. Fix P and 4ABC, and let X vary on P B, which has degree 1. Con-
sider the line ` through M parallel to the angle bisector of ∠BP C, which is fixed.

Claim 1: Consider the reflexion φ w.r.t `, then φ(M X) = M Y .

Proof: Let Y 0 = φ(M X) ∩ AC. Basic angle chasing shows that

∠M Y 0 P = ∠(`, M Y ) − ∠(`, P C) = ∠(M X, `) − ∠(P B, `) = ∠M XP.

Therefore Y 0 = Y . 

Hence M Y has degree 1, also Y has degree 1. Consider `1 = P B, `2 = P C,

`3 = Y A, `4 = AX, then the concyclic of `1 ∩ `2 = P , `2 ∩ `3 = Y , `3 ∩ `4 = A,
`4 ∩ `1 = X has degree 0 + 0 + 1 + 1 which has degree 2. It suffices to check 3
special cases.

• Let X be the foot of M on P B, then P, X, M, Y, A are concyclic.

• Let X = AM ∩ P B, then ∠M XC = ∠BXM = ∠M Y C, hence M, X, C, Y
are concyclic, implying that XY ⊥ P C, hence P, X, Y, A are concyclic.
• Let X s.t. Y = AM ∩ P C works similarly as the second case.

Problem 5.16 (IMO SL 2018 G5). Let ABC be a triangle with circumcircle
Ω and incentre I. A line ` intersects the lines AI, BI, and CI at points
D, E, and F , respectively, distinct from the points A, B, C, and I. The
perpendicular bisectors `A , `B , and `C of the segments AD, BE, and CF ,
respectively determine a triangle Θ. Show that the circumcircle of the triangle Θ
is tangent to Ω.

40
Proof. Let X = `B ∩ `C , Y = `C ∩ `A , Z = `A ∩ `B . Let L, M, N be the
second intersection of AI, BI, CI with (ABC). Then LM k XY , M N k
Y Z, N L k ZX, thus LX, M Y, N Z are concurrent at their homothety center
T . It suffices to show that T lies on (ABC). We fixed A, B, C, D, let E
moving on BI has degree 1, then F = DE ∩ CI has degree 1. `A is fixe.
The midpoint of BE,CF has degree 1. As the directions of BI, CI are fixed,
their orthogonal directions are fixed as well. Hence `B , `C has degree 1. And
Y = `A ∩ `B , Z = `C ∩ `A both have degree 1. Finally the intersection
T = M Y ∩ N Z has degree 1. Finally, the statement A, N, T, M are concyclic
has degree deg(AN ) + deg(N T ) + deg(T M ) + deg(M A) = 0 + 1 + 1 + 0 = 2.
Therefore it suffices to check 3 special cases.
• Let E = I, then F = I, Y M , ZN are the perpendicular bisector of BI and
CI intersects at L ∈ (ABC).
• Let E such that DE k CI, then F = CI ∩ `∞ . The midpoint of CF is still
F . Hence `C = `∞ and Y = `A ∩ `∞ . This implies that Y M is parallel to
`A , which is orthogonal to AI, note that M N ⊥ AI as well, therefore N lies
on M Y . This means N = M Y ∩ N Z which lies on (ABC).
• Let E = BI ∩ `∞ gives the same reasoning as the second bullet.

Problem 5.17 (IMO 2015 P4). Triangle ABC has circumcircle Ω and circum-
center O. A circle Γ with center A intersects the segment BC at points D and
E, such that B, D, E, and C are all different and lie on line BC in this order.
Let F and G be the points of intersection of Γ and Ω, such that A, F , B, C,
and G lie on Ω in this order. Let K be the second point of intersection of the
circumcircle of triangle BDF and the segment AB. Let L be the second point of
intersection of the circumcircle of triangle CGE and the segment CA. Suppose
that the lines F K and GL are different and intersect at the point X. Prove that
X lies on the line AO.

Proof. We reformulate the problem as follows:

Let F, G be the intersections of two circles O, A with A ∈ O.
Let B be a point on O and D be a point on A, let C and E be the
second intersection of the line BD with O and A respectively. Let
L be the second point of intersection of the circumcircle of triangle
CGE and the line CA. Let X = GL ∩ AO and K = F X ∩ AB, prove
that B, D, F, K are concyclic.
We consider B as a fixed point and let D moves on A, which has degree 2.
• First, we remark ∠EGL = ∠ECL = ∠BCA is a fixed angle, hence GL is
obtained by rotation of center GE and angle ∠EGL, we apply the rotation
lemma with fixed angle, the degree of GL is at most deg(G) + deg(GE) =
0 + 1 = 1. (E = BD ∩ A has degree 2 and GE has degree 1, moreover G,E
belongs to the fixed circle A, hence GE has degree 2/2 = 1.)

41
 • X = EG ∩ AO has degree 1, K = F X ∩ AB has degree 1.
• We remark that if D be one of the intersections of AB ∩ A, then D = K,
hence applying the Coincidence Lemma 4.14 yields the line DK has degree
at most deg(D) + deg(K) − 2 = 2 + 1 − 2 = 1.
• Finally, we apply corollary on KB, KD, F B, F D, the concyclic statement
has degree deg(KB) + deg(KD) + deg(F B) + deg(F D) = 0 + 1 + 0 + 1 = 2.
So it suffices to check for 3 special cases of D.

• When D = G, we have C = G, GL = CL = AC, hence X = A = K and

BDF K = BCF A ∈ O are concyclic.
• When D is the second intersection of BG ∩ A. As AF = AG, we have
∠ABF = ∠DBA, implying that D is the reflexion of F w.r.t AB. Note that
we have C = E = G, as E moves on A, we have EG = GG the tangent of
A at G, which is indeed GA0 . By construction, ∠XGA0 = ∠AGB. Then as
F and G are symmetric w.r.t AA0 , we have ∠A0 F X = ∠AGB = ∠AA0 B =
90◦ − ∠BAA0 = 90◦ − ∠BF A0 , meaning that KF ⊥ F B. By symmetry,
KD ⊥ BD, hence B, D, F, K are concyclic.
• When D = F , we have C = D = F , the line F D is the tangent of A
at F , which is indeed F A0 . We need to prove that BF K is tangent to
F A0 . Note that the circle CEG = F EG = A. Hence L = F A ∩ A
yields LG ⊥ F G implying LG k AO, implying that X = AO ∩ `∞ . Hence
F K k AO. Hence ∠KF A0 = ∠AA0 F = ∠ABF , implying that A0 F is tangent
to BF K.

42
References
[1] M. Borislav. Problems in elementary geometry (in Bulgarian). 1995.
[2] J. Gallier. Basics of projective geometry. In Geometric Methods and Appli-
cations, pages 103–175. Springer, 2011.
[3] N. Hitchin, J. Derakhshan, and B. Szendroi. Algebraic curves. 2014.
[4] F. C. Kirwan and F. Kirwan. Complex algebraic curves, volume 23. Cambridge
University Press, 1992.

[5] M. Reid. Undergraduate algebraic geometry. Cambridge University Press

Cambridge, 1988.
[6] J. Richter-Gebert. Perspectives on projective geometry: A guided tour through
real and complex geometry. Springer Science & Business Media, 2011.

[7] T. W. Sederberg. Computer aided geometric design. 2012.

43