Barycentric Coordinates in Olympiad Geometry

Max Schindler

Evan Chen

May 15, 2012

I suppose it is tempting, if the only tool you have is

a hammer, to treat everything as if it were a nail.
Abstract
In this paper we present a powerful computational approach to large class of olympiad
geometry problems barycentric coordinates. We then extend this method using some of the
techniques from vector computations to greatly extend the scope of this technique.
The general method here was popularized by Max Schindler. Theorem 4, nicknamed EFFT,
is either extremely underrated or new, probably the former.
Special thanks to Amir Hossein and the other olympiad moderators for helping to get this
article featured: I certainly did not have such ambitious goals in mind when I first wrote this!

Mewto55555, Missouri. I can be contacted at igoroogenflagenstein@gmail.com.

v Enhance, SFBA. I can be reached at echen42@stanford.edu.

Contents
Title Page

1 Preliminaries

1.1

Advantages of barycentric coordinates . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2

Notations and Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.3

How to Use this Article . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2 The Basics

2.1

The Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2

Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2.1

The Equation of a Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2.2

Ceva and Menelaus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Special points in barycentric coordinates . . . . . . . . . . . . . . . . . . . . . . . . .

2.3

3 Standard Strategies

3.1

EFFT: Perpendicular Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.2

Distance Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.3

Circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.1

Equation of the Circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

4 Trickier Tactics

12

4.1

Areas and Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4.2

Non-normalized Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4.3

O, H, and Strong EFFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4.4

Conways Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4.5

A Few Final Lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

5 Example Problems
5.1

5.2

16

USAMO 2001/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.1.1

Problem and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

5.1.2

Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

USAMO 2008/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2.1

Problem and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2

5.2.2
5.3

5.4

5.5

5.6

5.7

Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

ISL 2001 G1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.3.1

Problem and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

5.3.2

Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2012 WOOT PO4/7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

5.4.1

Problem and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

5.4.2

Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

ISL 2005 G5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.5.1

Problem and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5.5.2

Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

USA TSTST 2011/4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

5.6.1

Problem and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

5.6.2

Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

USA TSTST 2011/2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5.7.1

Problem and Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5.7.2

Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

6 Problems

30

6.1

Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

6.2

Hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

A Vectors Tutorial

32

A.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
A.2 Triangle ABC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
A.3 Special Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
B Formula Sheet

34

B.1 Standard Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

B.2 More Obscure Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
B.2.1 Other Special Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
B.2.2 Special Lines and Circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Bibliography

38

Chapter 1

Preliminaries
So many problems are killed by it. Its just that no one knows it.
Max Schindler
Over the course of olympiad geometry, several computational approaches have surfaced as a
method of producing complete solutions to geometry problems given sufficient computational fortitude. Each has their advantages and drawbacks.
Barycentric coordinates, also called areal coordinates, provide a new bash approach for geometry problems. Barycentric coordinates offer a length-based, coordinate approach to geometry
problems.

1.1

Advantages of barycentric coordinates

The advantages of the system include

Sides of the triangle playing the role of the axes.
Simple expressions for lines in general, making it computationally feasible to intersect lines.
Simple forms for some common points (centroid, incenter, symmedian point...)
Very strong handling of ratios of lengths.
A useful method for dropping arbitrary perpendiculars.
An area formula.
Circle formula.
Distance formula.
This arsenal of tools is far more extensive than that of many other computational techniques.
Cartesian coordinates are woefully inadequate for most olympiad geometry problems because the
forms for special points are typically hideous, and the equation of a circle is difficult to work with.
Complex numbers and vectors are more popular, but the concept of an equation of a line is
complicated in the former and virtually nonexistent in the latter.

1.2

Notations and Conventions

Throughout this paper, 4ABC is a triangle with vertices in counterclockwise order. The lengths
will be abbreviated a = BC, b = CA, c = AB. These correspond with points in the vector plane
~ B,
~ C.
~
A,
For arbitrary points P , Q, R, [P QR] will denote the signed area of 4P QR.1

1.3

How to Use this Article

You could read the entire thing, but the page count makes this prospect rather un-inviting. About
half of it is theory, and half of it is example problems; its probably possible to read either half only
and then go straight to the problems.
Im assuming that the majority of readers would like to read just the examples, and attack the
problems with a formula sheet beside them. Ive included Appendix B to facilitate these needs;
I also have a version which includes just the formula sheet and the problems, which should be
floating around.
Happy bashing!

1
For ABC counterclockwise, this is positive when P , Q and R are in counterclockwise order, and negative otherwise. When ABC is labeled clockwise the convention is reversed; that is, [P QR] is positive if and only if it is oriented
in the same way as ABC. In this article, ABC will always be labeled counterclockwise.

Chapter 2

The Basics
We will bary you!
Schindler-Khruschev

2.1

The Coordinates

Definition. Each point in the plane is assigned an ordered triple of real numbers P = (x, y, z)
such that
~ + yB
~ + zC
~ and x + y + z = 1
P~ = xA
It is not hard to verify that the coordinates of any point are well-defined. These are sometimes
called areal coordinates because if P = (x, y, z), then the signed area [CP B] is equal to x[ABC],
and so on [1, 2]. In other words, these coordinates can viewed as
P =

1
([P BC], [P CA], [P AB])
[ABC]

Of course, notice that A = (1, 0, 0), B = (0, 1, 0) and C = (0, 0, 1)! This is why barycentric
coordinates are substantially more suited for standard triangle geometry problems.
Exercise. Find the coordinates for the midpoint of AB.
Exercise. When is a point (x, y, z) in the interior of a triangle?

2.2
2.2.1

Lines
The Equation of a Line

The equation of a line [1, 2] in barycentic coordinates is astoundingly simple.

Theorem 1 (Line). The equation of a line is ux + vy + wz = 0 where u, v, w are reals. (These u,
v and w are unique up to scaling.)
This is a corollary of the area formula, Theorem 10.
In particular, if a line ` passes through a vertex, say A, then u(1) + v(0) + w(0) = 0 u = 0.
So we rearrange to obtain
Corollary 2 (Line through a vertex). The equation of a line passing through A is simply of the
form y = kz for some constant k.
In particular, the equation for the line AB is simply z = 0, by substituting (1, 0, 0) and (0, 1, 0)
into ux + vy + wz = 0.
6

2.2.2

Ceva and Menelaus

In fact, the above techniques are already sufficient to prove both Cevas and Menelauss Theorem,
as in [1].
Corollary 3 (Cevas Theorem). Let AD, BE and CF be cevians of a triangle ABC. Then the
cevians concur if and only if
BD CE AF
=1
DC EA F B
Proof. Since D lies on BC, the point D has the form D = (0, d, 1 d). So the equation of line AD
is simply
1d
y=
z
d
Similarly, if we let E = (1 e, 0, e) and F = (f, 1 f, 0) then the lines BE and CF have equations
1f
z = 1e
e x and x = f y, respectively.
Notice that this system of three equations is homogeneous, so we may ignore the condition that
x + y + z = 1 temporarily. Then it is easy to see that this equation has solutions if and only if
(1 d)(1 e)(1 f )
=1
def
which is equivalent to Cevas Theorem.
Exercise. Prove Menelauss Theorem.

2.3

Special points in barycentric coordinates

Here we give explicit forms for several special points in barycentric coordinates (compiled from
1
(u, v, w);
[1, 3, 5]). In this section, it will be understood that (u : v : w) refers to the point u+v+w
1
that is, we are not normalizing the coordinates such that they sum to 1.
Again, the coordinates here are not homogenized!
Point
Centroid
Incenter
Symmedian point
Excenter
Orthocenter
Circumcenter

Coordinates
G = (1 : 1 : 1)
I = (a : b : c)
K = (a2 : b2 : c2 )
Ia = (a : b : c), etc.
H = (tan A : tan B : tan C)
O = (sin 2A : sin 2B : sin 2C)

Sketch of Proof
Trivial
Angle bisector theorem
Similar to above
Similar to above
Use area definition
Use area definition

One will notice that O and H are not particularly nice in barycentric coordinates (as compared
to in, say, complex numbers), but I and K are particularly pretty.
If absolutely necessary, it is sometimes useful to convert the trigonometric forms of H and O
into expressions entirely in terms of the side lengths (cf. [3, 5]) by
O = (a2 (b2 + c2 a2 ) : b2 (c2 + a2 b2 ) : c2 (a2 + b2 c2 ))
1
These standard coordinates with u + v + w = 1 are sometimes referred to homogenized barycentric coordinates
to emphasize the distinction.

and
H = ((a2 + b2 c2 )(c2 + a2 b2 ) : (b2 + c2 a2 )(a2 + b2 c2 ) : (c2 + a2 b2 )(b2 + c2 a2 ))
As we will see in section 4.4, Conways notation gives a more viable way of manipulating the
expressions.
Exercise. Prove the formulas given in the table above are correct (or email me telling me about
a typo.)
Some of these proofs are given in [2].
When calculating ratios and areas, etc., it is important that coordinates are normalized. There
are circumstances where it is both legitimate and computationally easier to use (kx : ky : kz)
in place of (x, y, z). See section 4.2 for details. It is recommended that one keep coordinates
normalized when they are still learning this technique in order to avoid pitfalls.

Chapter 3

Standard Strategies
Uh, perpendiculars should be okay right? Dot product?
Evan Chen, on EFFT
This chapter will cover some typical techniques which arise in olympiad problems.
First, we begin the following definition [1]:
Definition. The displacement vector of two (normalized) points P = (p1 , p2 , p3 ) and Q = (q1 , q2 , q3 )

is denoted by P Q and is equal to (p1 q1 , p2 q2 , p3 q3 ).

Note that the sum of the coordinates of a displacement vector is 0.
In the proofs of this chapter, we will invoke some properties which will be familiar to those who
~ = ~0
have done vector bashing; namely, when O
~A
~ = R2
A
~B
~ = R2 c2 /2
A
where R is the circumradius.
If you dont care about the proofs, then this doesnt matter; otherwise, see appendix A for a
treatise on some elementary vector bashing.
This chapter will frequently involve translating O to the null vector ~0, so that we may invoke
useful properties about dot products mentioned above. This is valid since the point (x, y, z) satisfies
x + y + z = 1, so translation is actually an invariant. As a consequence, however, take heed of the
following.
Note. It is important that x + y + z = 1 when doing calculations!

3.1

EFFT: Perpendicular Lines

In fact there is a fast characterization for when four points M, N, P, Q are perpendicular.

Theorem 4 (Evans Favorite Forgotten Trick). Consider displacement vectors M N = (x1 , y1 , z1 )

and P Q = (x2 , y2 , z2 ). Then M N P Q if and only if

0 = a2 (z1 y2 + y1 z2 ) + b2 (x1 z2 + z1 x2 ) + c2 (y1 x2 + x1 y2 )
~ to ~0. It is necessary and sufficient that (x1 A+y
~ 1 B+z
~ 1 C)(x
~
~
~
~
Proof. Translate O
2 A+y2 B+z2 C) = 0.
Expanding, this is just
 X

X
~A
~ +
~B
~ =0
x1 x2 A
(x1 y2 + x2 y1 ) A
cyc

cyc

~ = ~0, we make the standard substitutions and rearrange:

Utilizing the fact that O


c2
2
=0

(x1 x2 R ) +
(x1 y2 + x2 y1 ) R
2
cyc
cyc
!
X
X


1X
(x1 y2 + x2 y1 ) =
R2
(x1 x2 ) +
(x1 y2 + x2 y1 )(c2 )
2 cyc
cyc
cyc


1X
(x1 y2 + x2 y1 )(c2 )
R2 (x1 + y1 + z1 )(x2 + y2 + z2 ) =
2 cyc


1X
R2 0 0 =
(x1 y2 + x2 y1 )(c2 )
2 cyc
X


0=
(x1 y2 + x2 y1 )(c2 )
X

cyc

since x1 + y1 + z1 = x2 + y2 + z2 = 0 (remember that (x1 , y1 , z1 ) is a displacement vector!).

We will see later, in Section 4.2, that this holds for scaled displacement vectors.
This will be abbreviated EFFT1 throughout this document. As we shall see, EFFT increases
the scope of solvable problems by providing a usable method for constructing perpendiculars, and
in particular, perpendicular bisectors.
EFFT is particularly useful when one of the displacement vectors is the side of the triangle.
For instance, we may deduce

Corollary 5. Consider a displacement vector P Q = (x1 , y1 , z1 ). Then P Q BC if and only if

0 = a2 (z1 y1 ) + x1 (c2 b2 )
Corollary 6. The perpendicular bisector of BC has equation
0 = a2 (z y) + x(c2 b2 )
Exercise. Prove the above corollaries, and deduce the Pythagorean Theorem.
EFFT is used in many problems, e.g. problems 5.2 and 5.4. Its generalization Strong EFFT
(cf. Theorem 4.3) is also useful and is invoked in problem 5.5.

3.2

Distance Formula

The distance formula [1] can be proved by precisely the same means as EFFT:

Theorem 7 (Distance Formula). Consider a displacement vector P Q = (x, y, z). Then

|P Q|2 = a2 yz b2 zx c2 xy
Exercise. Prove the distance formula by using
~ + yB
~ + z C)
~ (xA
~ + yB
~ + z C)
~
P Q2 = (xA
and the fact that x + y + z = 0.
1

Hi Simon.

10

3.3
3.3.1

Circles
Equation of the Circle

Not surprisingly, the equation of a circle [1] is much more annoying than anything that has showed
up so far. Nonetheless, it is certainly usable, particularly in certain situations described below.
Theorem 8. The general equation of a circle is
a2 yz b2 zx c2 xy + (ux + vy + wz)(x + y + z) = 0
for reals u, v, w.
Proof. Assume the circle has center (j, k, l) and radius r. Then this is just
a2 (y k)(z l) b2 (z l)(x j) c2 (x j)(y k) = r2
Expand everything, and collect terms to get
a2 yz b2 zx c2 xy + C1 x + C2 y + C3 z = C
for some hideous constants Ci and C. Since x + y + z = 1, we can rewrite this as
a2 yz b2 zx c2 xy + ux + vy + wz = 0 a2 xy b2 yz c2 zx + (ux + vy + wz)(x + y + z) = 0
where u = C1 C, et cetera.
Fortunately, most circles that we are concerned with have much cleaner forms. For example, as
in [1],
Corollary 9. The circumcircle has equation
a2 yz + b2 zx + c2 xy = 0
Exercise. Prove the above corollary.
Furthermore, circles passing through vertices of the triangle make for more elegant circle formulas, since many terms in the circle equation vanish.
Note. Notice that the equation of the circle is homogeneous, so it is possible to scale (x, y, z)
arbitrarily and still have a true equation. See section 4.2.

11

Chapter 4

Trickier Tactics
All you have to do is construct a parallelogram!
James Tao
Here are some strategies which are occasionally useful.

4.1

Areas and Lines

The area formula [1, 2, 5] is as follows:

Theorem 10 (Area Formula). The area of a triangle with vertices P = (x1 , y1 , z1 ), Q = (x2 , y2 , z2 )
and R = (x3 , y3 , z3 ) is


 x1 y1 z1 


[P QR] = [ABC]  x2 y2 z2 
 x3 y3 z3 
A proof of the area formula can be found in [2].
From this, we immediately find two corollaries, which also appear in [1] and [2].
Corollary 11 (First Collinearity Criteria).
R = (x3 : y3 : z3 ) are collinear if and only if

 x1

 x2

 x3

The points P = (x1 : y1 : z1 ), Q = (x2 : y2 : z2 ) and


y1 z1 
y2 z2  = 0
y3 z3 

Proof. The points are collinear iff the area of the triangle they determine is zero, and this doesnt
change under scaling (x, y, z) 7 (kx : ky : kz).
Corollary 12 (Line Through 2 Points). The
and Q = (x2 : y2 : z2 ) is

 x1

 x2

 x

equation of a line through the points P = (x1 : y1 : z1 )


y1 z1 
y2 z2  = 0
y z 

The previous corollary also implies the line formula.

On the other hand, when normalized, Corollary 11 can be simplified by

12

Corollary 13 (Second Collinearity Criteria). The

R = (x3 , y3 , z3 ), are collinear if and only if

 x1 y1 1

 x2 y2 1

 x3 y3 1

points P = (x1 , y1 , z1 ), Q = (x2 , y2 , z2 ) and




=0



Cyclic variations hold.

Proof. It is a property of the







determinant that
 
x1 y1 z1   x1 y1 x1 + y1 + z1
x2 y2 z2  =  x2 y2 x2 + y2 + z2
x3 y3 z3   x3 y3 x3 + y3 + z3

This has a substantial computational advantage.

4.2

Non-normalized Coordinates

Suppose x + y + z = 1. Non-normalized coordinates involve the notion of extending the notation

(x, y, z) as follows: for all nonzero k, the triple (kx : ky : kz) refers to the point (x, y, z). We will
also make the same scaling for displacement vectors: the displacement vector (x, y, z) can also be
written (kx : ky : kz).
It turns out that, if one is merely intersecting lines and circles, we can actually use nonnormalized coordinates during the computation, simplifying calculations greatly (cf. [1]). This
holds because the equations for lines and circles are homogeneous; that is, scaling x, y, z in them
does not affect the validity of the equation. However, when using the area formula, distance formula,
etc. it is extremely important that points be normalized before the calculations are applied.
Exercise. Check that the distance formula fails for non-normalized coordinates, but holds for
EFFT.
Some sources, including the first version of this article, simply use (kx, ky, kz) instead of (kx :
ky : kz). We use the latter to avoid confusion, but a reader should keep in mind that not all authors
choose to do this.

4.3

O, H, and Strong EFFT

As we can see easily in Section 2.3, in fact the least elegant points are H and O. This seems contrary
~ as zero, and H
~ =A
~ +B
~ + C.
~
to expectation: in complex numbers and vectors, we frequently set O
One might be inclined to think that, given the vector-based definition of barycentric coordinates,
we might be able to utilize this.
Unfortunately, the area formula, and hence the line equation, relies on the fact that (x, y, z)
reflect the actual ratio of areas. However, we can get some partial tools to simplify computations
related to O and H.
13

Theorem 14 (Strong EFFT). Suppose M , N , P and Q are points with

M N = x1 AO + y1 BO + z1 CO

P Q = x2 AO + y2 BO + z2 CO
If either x1 + y1 + z1 = 0 or x2 + y2 + z2 = 0, then M N P Q if and only if
0 = a2 (z1 y2 + y1 z2 ) + b2 (x1 z2 + z1 x2 ) + c2 (y1 x2 + x1 y2 )
Proof. Reproduce the proof of Theorem 4, with a few modifications. Here, the displacement
vector P~ Q has already been shifted, so that the properties of the dot product A A = R2 and
A B = R2 c2 /2 hold.
In the final step, instead of 0 0 = 0 we have 0 (x2 + y2 + z2 ) = 0, which is just as well.
Note. If M N = (x1 , y1 , z1 ) is a displacement vector, then we automatically have both x1 +y1 +z1 =
~ N = x1 A
~ + y1 B
~ + z1 C,
~ since this is independent of the location of the null vector (here O).
~
0 and M
Therefore, Strong EFFT is usually useful for showing some more unusual vector is perpendicular
to a standard displacement vector.
Once again, scaling is valid: the condition is homogenous in each of x1 , y1 , z1 and x2 , y2 , z2 .
~ =A
~+B
~ + C.
~ See problem
This is particularly useful if one of the points involved is H, for H
5.5 for an example.
The other point for which Strong EFFT is useful is O; for example, the following corollary is
immediate
Corollary 15. The equation for the tangent to the circumcircle at A is b2 z + c2 y = 0.
Exercise. Prove the above corollary.

4.4

Conways Formula

Conways Formula gives a nice way to work with (some) angles in a triangle. Most of the results
here are from [4].
We start by defining some notation:
Definition (Conways Notation). Let S be twice the area of the triangle. Define S = S cot , and
define the shorthand notation S = S S .
2

+c
We have SA = a +b
= bc cos A and its cyclic variations. (We also have S =
2
where is the Brocard angle. This follows from cot = cot A + cot B + cot C.)

Fact. We have the identities

SB + SC = a2
and
SAB + SBC + SCA = S 2
14

a2 +b2 +c2
,
2

Exercise. Prove the above facts.

This gives us slightly more convenient ways to express some things. For example


Fact. O = (a2 SA : b2 SB : c2 SC ) and H = (SBC : SCA : SAB ) = S1A : S1B : S1C .
(Note that these do not sum to 1; we are using the conventions spelled out in section 4.2).
Now for the main result of this chapter:
Theorem 16 (Conways Formula). Given point P, with counter-clockwise directed angles ]P BC =
and ]BCP = , we have P = (a2 : SC + S : SB + S ).

Proof. By the Law of Sines on triangle P BC, BP = sina sin

(+) , and CP =
triangle P BC is
1
a2 sin sin
BC BP sin =
2
2 sin +

Similarly, the area of P CA and P BA can be found to be

tively.

ab sin sin +C
2 sin (+)

a sin
sin (+) ,

and

so the area of

ac sin sin +B
2 sin (+)

respec-

Eliminating the denominator gives that P s barycentric coordinates are at:

(a2 sin sin : ab sin sin + C : ac sin sin ( + B))


ac sin ( + B)
2
= a : ab sin + Csin :
sin
= (a2 : ab cos C + ab sin C cot : ca cos B + ca sin B cot )
= (a2 : SC + S : SB + S )
as desired.
An example of an application of Conways Formula can be found in problem 5.3.

4.5

A Few Final Lemmas

Lemma 17 (Parallelogram Lemma). The points ABCD form a parallelogram iff A + C = B + D

(here the points are normalized), where addition is done component-wise.
Lemma 18 (Concurrence Lemma). The three lines ui x+vi y+wi z = 0, for i = 1, 2, 3 are concurrent
if and only if


 u1 v1 w1 


 u2 v2 w2  = 0


 u3 v3 w3 

15

Chapter 5

Example Problems
Graders received some elegant solutions, some not-so-elegant solutions, and some so not
elegant solutions.
Delong Meng
Let us now demonstrate the power of the strategy on a few... victims.

5.1

USAMO 2001/2

We begin with a USAMO problem which literally (figuratively?) screams for an analytic solution.

5.1.1

Problem and Solution

Problem (USAMO 2001/2). Let ABC be a triangle and let be its incircle. Denote by D1 and
E1 the points where is tangent to sides BC and AC, respectively. Denote by D2 and E2 the
points on sides BC and AC, respectively, such that CD2 = BD1 and CE2 = AE1 , and denote by
P the point of intersection of segments AD2 and BE2 . Circle intersects segment AD2 at two
points, the closer of which to the vertex A is denoted by Q. Prove that AQ = D2 P .
Solution. (Evan Chen) So we use barycentric coordinates.
Its obvious that un-normalized,
D1 = (0 : s c : s b)
D2 = (0 : s b : s c), so we get a


sb sc
sc
normalized D2 = 0, a , a: . Similarly, E2 = sa
b , 0, b .


sb sc
by intersecting the lines AD2 : (s c)y = (s b)z
Now we obtain the points P = sa
s , s , s
and BE2 : (s c)x = (s a)z.
Let Q0 be such that AQ0 = P D2 . Its obvious that Q0y + Py = Ay + D2y , so we find that Q0y =
sb
a

sb
s =
Hence,

(sa)(sb)
.
sa

Also, since it lies on the line AD2 , we get that Q0z =

Q0x = 1

((s b) + (s c))(s a)
sa a(s a)
a
=
=
sa
sa
s

Hence,
0

Q =

a (s a)(s b) (s a)(s c)
,
,
s
sa
sa

16

sc
sb

Q0y =

(sa)(sc)
.
sa

Q
E1

E2
I

D2

D1

Figure 5.1: USAMO 2001/2

Let I =

a b c
2s , 2s , 2s

. We claim that, in fact, I is the midpoint of Q0 D1 . Indeed,

1
a
a
0+
=
2
s 2s

(s a)(s b) + s(s c)
1 (s a)(s b) s c
+
=
2
sa
a
2sa
ab
=
2sa
b
=
2s


c
1 (s a)(s c) s b
+
=
2
sa
a
2s

Implying that Q lies on the circle; in particular, diametrically opposite from D1 , so it is the
closer of the two points. Hence, Q = Q0 , so were done.

5.1.2

Commentary

The unusual points D2 , and E2 make this problem ripe for barycentric coordinates, since they can
be written as ratios on the appropriate sides. The point Q0 is also easy to construct because of
lengths.
Now it turns out that its well known that AD2 passes through the point directly above D1 , so
it is merely a matter of length chasing to construct Q0 and subsequently use midpoints.
17

5.2

USAMO 2008/2

We follow this with another problem, made possible by EFFT (theorem 4). As we shall see, EFFT
is particularly useful for constructing perpendiculars to the sides of the triangle.

5.2.1

Problem and Solution

Problem (USAMO 2008/2). Let ABC be an acute, scalene triangle, and let M , N , and P be
the midpoints of BC, CA, and AB, respectively. Let the perpendicular bisectors of AB and AC
intersect ray AM in points D and E respectively, and let lines BD and CE intersect in point F ,
inside of triangle ABC. Prove that points A, N , F , and P all lie on one circle.

N
E
F
D

Figure 5.2: USAMO 2008/2

Solution. (Evan Chen) Set A = (1, 0, 0), B = (0, 1, 0) and C = (0, 0, 1). Evidently P = ( 21 , 12 , 0),
etc. Check that the equation of the line AM is y = z.
We will now compute the coordinates of the point D. Write D = (1 2t, t, t) as it lies on the
line AM . Applying EFFT to DP AB,
1
1
((t ) ( 2t))(c2 /2) + t(a2 b2 /2) = 0
2
2
(3t 1)(c2 ) + t(a2 b2 ) = 0
(a2 b2 3c2 )t = c2
t=

c2
3c2 + b2 a2

Call this value j. Then D = (1 2j, j, j). Similarly, E = (1 2k, k, k) where k =

18

b2
.
3b2 +c2 a2

Now the line BD has equation

F = (p, q, r), then p + q + r = 1,
In fact,
1+

b2 +c2
c2 +b2 a2

r
p

z
x

j
12j ,

j
12j ,

and

whilst the line CE has equation

q
p

k
12k .

So if

k
12k .

2
2
c2
= c2 +bc2 a2 , and pq = b2 +cb2 a2 . Evidently p1
(3c2 +b2 a2 )2c2
2
2
= 2S+a
= 2 + aS . Let S = c2 + b2 a2 for convenience.
S
to F 0 = 2F A = (2p 1, 2q, 2r). It suffices to show this lies on

r
p

y
x

= 1+

r
p

q
p

Dilate F
the circumcircle
of triangle ABC (since this will send P and N to B and C, respectively), which has equation
a2 yz + b2 zx + c2 xy = 0. Hence it suffices to show that
0 = a2 (2q)(2r) + b2 (2r)(2p 1) + c2 (2p 1)(2q)
qr
1
r
q
0 = 4a2
+ (2 + )(2b2 + 2c2 )
pp
p
p
p
2
2
2 c2
a2
c
b
b
(2 (2 + ))(2b2 + 2c2 ) = 4a2 2
S
S
S
S
2
2 2
2 2
2 2 2
a (2b c + 2c b ) = 4a b c
4a2 b2 c2 = 4a2 b2 c2
which is true.

5.2.2

Commentary

EFFT provides a method for constructing the perpendicular bisectors (in actuality this is just
corollary 6), from which it is not hard at all to intersect lines. The final conclusion involves a circle,
so we dilate the points such that this circle is simply the circumcircle, which has a very simple
form.

5.3

ISL 2001 G1

We now present an application of Conways Formula to an ISL G1.

5.3.1

Problem and Solution

Problem (ISL 2001/G1). Let A1 be the center of the square inscribed in acute triangle ABC
with two vertices of the square on side BC. Thus one of the two remaining vertices of the square
is on side AB and the other is on AC. Points B1 , C1 are defined in a similar way for inscribed
squares with two vertices on sides AC and AB, respectively. Prove that lines AA1 , BB1 , CC1 are
concurrent.
Solution. (Max Schindler) There is an obvious homothety centered on A that maps the inscribed
square with center A1 to one with center A2 , which is constructed externally off side BC. B2 and
C2 can be defined similarly. It then suffices to show that AA2 , BB2 , CC2 concur.
By Conways formula, A2 = (a2 : SC + S cot 45 : SB + S cot 45) = (a2 : SC + S : SB + S).
It can be very easily verified that AA2 , and similarly BB2 and CC2 , all go through the point
((SB + S)(SC + S) : (SA + S)(SC + S) : (SB + S)(SA + S)) and are thus concurrent (this point is
called the Outer Vecten Point).
19

L
A1

A2

M0

N0

Figure 5.3: ISL 2001 G1

5.3.2

Commentary

Homothety (similar to that in problem 5.2) allowed us to create a convenient expression for S by
Conways Formula, at which point Conways Formula cleared the problem.

5.4

2012 WOOT PO4/7

We now exhibit yet another problem in which perpendicular bisectors can be constructed via EFFT,
which are then used to determine a circumcenter. The method was used by v Enhance during the
actual test and received 7/.7. See [6].

5.4.1

Problem and Solution

Problem (2012 WOOT PO4/7). Let be the circumcircle of acute triangle ABC. The tangents
to at B and C intersect at P , and AP and BC intersect at D. Points E, F are on AC and AB,
respectively, such that DE k BA and DF k CA.
(a) Prove that points F , B, C, and E are concyclic.
(b) Let A1 denote the circumcenter of cyclic quadrilateral F BCE. Points B1 and C1 are defined
similarly. Prove that AA1 , BB1 , and CC1 are concurrent.
Solution. (Evan Chen) Let A = (1, 0,0), etc. Since the symmedian point
has K = (a2 : b2 :

2
2
2
2
c2 ), we find that D = 0, b2b+c2 , b2c+c2 . Then, its obvious that E = b2b+c2 , 0, b2c+c2 and F =

20

F
E

P
Figure 5.4: WOOT Practice Olympiad 4, Problem 7



2
, b2b+c2 , 0 . In that case, notice that since the general form of a circle is a2 yz + b2 zx + c2 xy +
(ux + vy + wz)(x + y + z) = 0, the circle passing through
c2
b2 +c2

: a2 yz + b2 zx + c2 xy

b2 c2
x(x + y + z)
b2 + c2

passes through BFEC, so its cyclic.

Suppose A1 = (x0 , y0 , z0 ). Let M1 =
applying EFFT and rearranging yields
x 0 y0 +

B+F
2

1
1 c2
2 b2 +c2 , 2

1 b2
2 b2 +c2 , 0

. Then A1 M1 AB,

c2
a2 b2
z0 = 2
2
c
b + c2

Doing a similar thing, we find that A1 is given by

x0 + y0 + z0 = 1
a2 b2
c2
z
=
0
c2
b2 + c2
2
2
a c
b2
x0 +
y

z
=
0
0
b2
b2 + c2
x0 y0 +

Then by Cramers rule, we can bash and get

conclusion by Ceva.

5.4.2

y0
z0

a2 +b2 c2
.
a2 b2 +c2

Then its easy to get the desired

Commentary

Once we realize that AD is a symmedian, the symmedian point immediately gives us a nice form
for D, and hence E and F by similar triangles. The condition that F BCE is cyclic is made easier
21

since two of the points are vertices of the triangle, making both v and w vanish; hence the end
computation is no longer hard.
For the second part, EFFT allows us to use perpendicular bisectors to construct the circumcenter; we can then compute the ratio that the cevian AA1 divides BC into. Cevas Theorem
guarantees that these will cancel cyclically, so we just compute the proper determinants, knowing
that they will cancel cyclically at the end.
The part about Ceva is worth remembering: any time you are trying to prove three cevians
are concurrent, barycentric coordinates may provide an easy way to compute the ratios in Cevas
Theorem.

5.5

ISL 2005 G5

We now present a solution to an ISL problem involving Strong EFFT (theorem 4.3). This problem
was proposed by Romania, and is relatively difficult to approach synthetically.

5.5.1

Problem and Solution

Problem (ISL 2005/G5). Let 4ABC be an acute-angled triangle with AB 6= AC. Let H be the
orthocenter of triangle ABC, and let M be the midpoint of the side BC. Let D be a point on the
side AB and E a point on the side AC such that AE = AD and the points D, H, E are on the
same line. Prove that the line HM is perpendicular to the common chord of the circumscribed
circles of triangle 4ABC and triangle 4ADE.
Solution. (Max Schindler) We use barycentric co-ordinates.
Let the length AD = AE = `.
Then, D = (c` : ` : 0), E = (b` : 0 : `), and, using Conways Notation, H =

1
SA

1
SB

1
SC

Since they are collinear,

c`
det b `
1
SA

`(c `)

`
SB

Thus, ` =

`
0
1
SB

0
` =0

1
SC


b`
`

=0
SC
SA


cSC bSB
1
1
1

=`
+
+
SB SC
SA SB
SC

SA (cSC bSB )
SB SC +SA SC +SA SB .

Substituting in SA =

b2 +c2 a2
2

`=

and similar, remembering that SB SC + SA SC + SA SB = S 2 gives:

(b2 + c2 a2 )[c(a2 + b2 c2 ) + b(a2 b2 + c2 )]

(a + b + c)(a + b c)(b + c a)(c + b a)
22

F 90
E
H

Figure 5.5: ISL 2005 G5

It can easily be seen that c = b is a root of the bracketed part of the numerator; factoring it
out leaves a2 + 2bc b2 c2 = a2 (b c)2 = (a b + c)(a + b c).
Thus,
`=

(b2 + c2 a2 )(b + c)(a b + c)(a + b c)

(b2 + c2 a2 )(b + c)
`=
(a + b + c)(a + b c)(b + c a)(c + b a)
(a + b + c)(b + c a)

Meanwhile, the circumcircle of ADE has equation given by a2 yz b2 zx c2 xy + (x + y +

z)(ux + vy + wz) = 0, as usual, so upon substituting the points A, D and E we find that

u=0
2

c (c `)(`) + c ((c `)u + `v) = 0

b2 (b `)(`) + b ((b `)u + `w) = 0
Conveniently enough, the first equation immediately gives u = 0, and we can easily solve for v
and w to get v = c(c `) and w = b(b `).
Now we need the common chord of the two circles

23

a2 yz b2 zx c2 xy = 0
a2 yz b2 zx c2 xy + (x + y + z)(c(c `)y + b(b `)z) = 0

But this is easy: subtract the two equations. We find that the common chord has equation
c(c `)y + b(b `)z = 0
Now, we have everything we need to finish this problem off.
Consider two points on this chord, A = (1, 0, 0) and P = (0 : b(b `) : c(c `)).

Then, the (scaled) displacement vector P A is (b(b `) c(c `) : b(b `) : c(c `)).
~ ~ ~ B+
~ ~
Let
the circumcenter be the null vector. Then, the displacement vector M H is A+
B+C 2 C =
1, 21 , 21 which can be scaled as (2 : 1 : 1)
Since P A has coefficient-sum 0, we can apply Strong EFFT:
M H P A a2 [c(c `) b(b `)] + b2 [c(c `) + b(b `)] + c2 [b(b `) c(c `)] = 0
which becomes
b(a2 b2 + c2 )(b `) = c(a2 + b2 c2 )(c `)
But we have
b`=

b(a + b + c)(b + c a) (b2 + c2 a2 )(b + c)

c(a2 + b2 c2 )
=
(a + b + c)(b + c a)
(a + b + c)(b + c a)

Similarly,
c`=

b(a2 b2 + c2 )
(a + b + c)(b + c a)

Thus,
c(a2 + b2 c2 )
(a + b + c)(b + c a)
2
2
2
= c(a + b c )(c `)

b(a2 b2 + c2 )(b `) = b(a2 b2 + c2 )

MH PA

5.5.2

Commentary

While we eventually did have to use the somewhat messy form of H, Strong EFFT permits us to
avoid doing so when finding the condition for perpendicular the perpendicular chords. The other
useful insight was computing the radical axis by simply subtracting the two equations: keep this
in mind!
By the way, it is true in general (as the diagram implies) that M , H and F are collinear. To
my best knowledge, the easiest way to prove this is to first prove the original problem.
24

5.6

USA TSTST 2011/4

The solution to this problem involves a synthetic observation, followed by an application of Vieta.
Normally, solving for P involves an incredibly painful quadratic. However, by first computing the
other intersection, we can actually use Vieta to deduce the coordinate of the point we care about.

5.6.1

Problem and Solution

Problem (TSTST Problem 4). Acute triangle ABC is inscribed in circle . Let H and O denote
its orthocenter and circumcenter, respectively. Let M and N be the midpoints of sides AB and
AC, respectively. Rays M H and N H meet at P and Q, respectively. Lines M N and P Q meet
at R. Prove that OA RA.
A

N
H

Figure 5.6: 2011 TSTST, Problem 4

Solution. (Evan Chen) We use barycentric coordinates. Set A = (1, 0, 0), B = (0, 1, 0) and C =
(0, 0, 1). Let TA = S1A in Conways Notation, so that H = (TA : TB : TC ).
~ = 0, and
First we will compute the point R0 = (x : y : z) on M N with R0 A OA. Setting O
applying Strong EFFT, we find that the point satisfies equations c2 y + b2 z = 0 (per EFFT) and
x = y + z (per midline). Thus, we may write R0 = (b2 c2 : b2 : c2 ) .
Next, we will compute the coordinates of P . Check that we have the line
HM : x y +

TB TA
z=0
TC

25

Furthermore, the circumcircle has equation z[a2 y + b2 x] + c2 xy = 0. WLOG, homogenize so that

z = TC ; then the first equation becomes y = x + TA TB . Substituting in the second equation,
this becomes simply
c2 x(x + (TA TB )) + a2 (TC )(x + TA TB ) + b2 (TC )(x) = 0
Expanding, and collecting coefficients of x, this reduces to
c2 x2 + [c2 (TA TB ) a2 TC b2 TC ]x + a2 (TC )(TA TB ) = 0
Now heres the trick: its well known that the reflection of H across M lies on the circle. So the
~0 =A
~ +B
~ H
~ (since H 0 AHB is a parallelogram). We can thus
other solution to this equation is H
quickly deduce that, in barycentric coordinates, H 0 = (TB + TC : TA + TC : TC ). So by Vietas
Formulas, the x-coordinate we seek is just
x=

c2 (TA TB ) a2 TC b2 TC
a2 + b2 c2

(T
+
T
)
=
TC TA
B
C
c2
c2

Hence,
y=

a2 + b2 c2
TC TB
c2

So finally, we obtain


a2 + b2 c2
a2 + b2 c2
T

T
:
TC TB : TC
C
A
c2
c2

a2 + c2 b2
a2 + c2 b2
T

T
:
T
:
TB TC
B
A
B
b2
b2

P =
Similarly,
Q=

It remains to show that P , Q and R0 are collinear. This occurs if and only if
2 2 2

a +b c
a2 +b2 c2
T

T
T

T
T
C
A
C
B
C
2
c2
2 c2 2

a2 +c2 b2
0 = det a +c2 b TB TA
T
T
B
B TC
c
b2
b2 c2
b2
c2
Subtracting the second and third column from the first, this is equivalent to

2
2
2
TC
TC + TB TA a +bc2 c TC TB

a2 +c2 b2
0 = det TC + TB TA
TB
TB TC
b2
0
b2
c2
Expanding the determinant, and factoring out the term TC + TB TA , we see that it suffices
to show that

 2

 2

a + b2 c2
a + c2 b2
2
2
2
2
0 = c TB b TC (c )
TC TB + (b )
TB TC
c2
b2
Expanding, this is just
(a2 b2 + c2 )TB + (a2 + b2 c2 )TC
But TB = S1B =
points are collinear.

2
a2 b2 +c2

and TC =

2
.
a2 +b2 c2

26

Since 2 + 2 = 0, we conclude that the three

5.6.2

Commentary

The application of Vieta played a critical role in this problem. BTW, during the actual TSTST,
only one person successfully bashed this problem (via complex numbers). Had I known about this
technique at the time, I think I could have gotten a positive score on Day 2.

5.7

USA TSTST 2011/2

Finally, to complete the section, we present an example of a bash which is possibly not feasible
by hand. Incidentally, this is the only problem in the section that was not done by hand. Max
Schindler claims this is doable if you have the computational fortitude.
The problem has been inverted around the point A, and the constant c has been replaced by
1/c.

5.7.1

Problem and Solution

Problem (TSTST Problem 2, Inverted). Let AP Q be an obtuse triangle with A > 90 . The
tangents at P and Q to the circumcircle of AP Q meet at the point B. Points X and Y are selected
on segments P B and QS, respectively, such that P X = c QA, and QY = c P A. Then, P A and
QA are extended through A to R and S, so that AR = AS = c P AQA
P Q . O is the point such that
AROS is a parallelogram. Given that AXOY is cyclic, prove that P XA = QY A.

Center of APQ

A
S
R

Y
B

Figure 5.7: 2011 TSTST, Problem 2

27

Solution. The reference triangle is AP Q, with A = (1, 0, 0), P = (0, 1, 0) and Q = (0, 0, 1). To
avoid confusion, let a = P Q, q = AP and p = AQ.
It is not hard at all to compute R and S. Since AR = AS =
and Q = (0, 0, 1), a quick application of ratios of lengths yields


cq
cq
R=
+ 1,
,0
a
a
and


S=

cp
cp
+ 1, 0,
a
a

cpq
a ,

and A = (1, 0, 0), P = (0, 1, 0)

Next we want to compute X = (x, y, z). Since X lies on the tangent P B, it satisfies 0 = b2 y+c2 z
(see the earlier problem, 5.6). Letting x + y + z = 1, and invoking the distance formula, we find
that (cq)2 = a2 (y 1)z + b2 zx + c2 x(y
Solving this system by writing everything in z, and
 1).
2
ac a2 c+ap+cq 2
taking the negative root, gives X = p ,
, cq
ap
ap .


cp2 a2 (c)+aq+cp2
cq
cp
cq

Similarly, Y = ac
,

,
. We then obtain that O = 1 + cp
q
aq
aq
a + a , a , a
Wed like to find criteria for when P XA = QY A. Noting the equal angles BP Q = BQP ,
we extend XA and Y A through A to points X 0 and Y 0 respectively. Then X 0 = (0 : pa + q 2 c a2 c :
q 2 c) and Y 0 = (0 : p2 c : qa + p2 c a2 c) (arent cevians wonderful?). In that case, we obtain the
lengths
q 2 c
q 2 c
P X0 =
a
=
pa a2 c
p ac
and similarly,
QY 0 =

p2 c
q ac

Now P XA = QY A if and only if the triangles XP X 0 and Y QY 0 are similar! Recall that
P X = cq and QY = cp; by SAS, this occurs if and only if
q 2 c
pac

cq

p2 c
qac

cp

q(q ac) = p(p ac) c =

So we just need to show that if AROS is cyclic, then c =

p+q
a

p+q
a .

Unfortunately, this is where the calculations get heavy.1

Putting A, X, Y into the circle formula and solving for u, v, w yields
u=0
v = (c2 pq 2 (a2 c + aq + cp(p + q)))/(a3 c2 + a2 c(p + q) c(p3 + q 3 ) + a(pq + c2 (p2 + q 2 )))
w = (c2 p2 q(a2 c + ap + cq(p + q)))/(a3 c2 + a2 c(p + q) c(p3 + q 3 ) + a(pq + c2 (p2 + q 2 )))
1
I wonder if theres any synthetic interpretation that would lead directly to this expression for c. That would be
nice too. Tell me if you find one!

28

Substituting the point O into the equation and simplifying, we get that
(cpq(a5 c3 2a4 c2 (p+q)c2 (p+q)3 (p2 pq+q 2 )+a3 c((12c2 )p2 +3pq+(12c2 )q 2 )+ac(p+q)2 ((1+c2 )p2 c2 pq+(1+c2 )q 2 )+a2 (p+q)(pq+3c2 (p2 +q 2 ))))
=0
(a2 (a3 c2 a2 c(p+q)+c(p3 +q 3 )a(pq+c2 (p2 +q 2 ))))

Clearing the denominator, this is just

0 = cpq(a5 c3 2a4 c2 (p + q)
c2 (p + q)3 (p2 pq + q 2 )
+ a3 c((1 2c2 )p2 + 3pq + (1 2c2 )q 2 )
+ ac(p + q)2 ((1 + c2 )p2 c2 pq + (1 + c2 )q 2 )
+ a2 (p + q)(pq + 3c2 (p2 + q 2 )))
Noting that c = 0 and c =
and collect terms in c to get

p+q
a

are solutions (the latter by inspection) we can divide this out

c2 a4 p4 p3 q pq 3 q 4 + 2a2 p2 + q 2

+ c a3 (p + q) a p3 + p2 q + pq 2 + q 3

a2 pq = 0

We need to show that there are no positive reals c satisfying this. A quick glance at the constant
term tells us that we should probably use the discriminant; our only hope is that there are no real
roots at all. Computing, we get
=a6 p2 2a4 p4 + a2 p6 2a6 pq + 4a4 p3 q 2a2 p5 q
+ a6 q 2 4a4 p2 q 2 a2 p4 q 2 + 4a4 pq 3 + 4a2 p3 q 3
2a4 q 4 a2 p2 q 4 2a2 pq 5 + a2 q 6
which factors as
= a2 (a + p + q)(p q)2 (a + p q)(a p + q)(a + p + q)
Since a, p and q are the sides of a triangle, this is always negative.

5.7.2

Commentary

The calculations are relatively easy up until incorporating the condition that AROS is cyclic.
Despite the fact that the cyclic condition was fairly brutal, it illustrates a few points worth
making: (i) similar triangles are your friends in barycentrics too, because they can yield equal
angles, (ii) inversion can provide a way to eliminate awkward circle conditions, (iii) unusual side
length conditions related to the sides of the triangles is not a problem, and (iv) cyclic quadrilaterals
suck.
Please let us know if you find a cyclic criteria which completes a reasonable solution! Wed like
to have a human to attribute the solution to...

29

Chapter 6

Problems
There is light at the end of the tunnel, but it is moving away at speed c.
Here, we provide some problems for the reader to attempt. Some are easier than others. Infinity
[1] has a longer list than is included here; see http://www.bmoc.maths.org/home/areals.pdf for
the relevant excerpt.
For practice in real life, i.e. determining when barycentric coordinates are feasible for a
problem (here, they are), you can often simply take an arbitrary geometry problem that begins
with the words triangle ABC, and see if its possible to get a barycentric solution. Some will, of
course, take much longer than others, but a good portion will have a reasonable solution.

6.1

Problems

1. Prove Stewarts Theorem.

2. Prove Rouths Theorem.
3. (USA TST 2003/2) Let ABC be a triangle and let P be a point in its interior. Lines P A,
P B, P C intersect sides BC, CA, AB at D, E, F , respectively. Prove that
1
[P AF ] + [P BD] + [P CE] = [ABC]
2
if and only if P lies on at least one of the medians of triangle ABC. (Here [XY Z] denotes
the area of triangle XY Z.)
4. (Mongolia TST 2000/6) In a triangle ABC, the angle bisector at A,B,C meet the opposite
sides at A1 ,B1 ,C1 , respectively. Prove that if the quadrilateral BA1 B1 C1 is cyclic, then
AB
BC
AC
=
+
AB + BC
AC + BC
BA + AC
5. (ISL 2005/G1) Given a triangle ABC satisfying AC + BC = 3 AB. The incircle of triangle
ABC has center I and touches the sides BC and CA at the points D and E, respectively. Let
K and L be the reflections of the points D and E with respect to I. Prove that the points
A, B, K, L lie on one circle.
6. (APMO 2005/5) In a triangle ABC, points M and N are on sides AB and AC, respectively,
such that M B = BC = CN . Let R and r denote the circumradius and the inradius of the
triangle ABC, respectively. Express the ratio M N/BC in terms of R and r.

30

7. (ISL 2008/G4) In an acute triangle ABC segments BE and CF are altitudes. Two circles
passing through the point A and F and tangent to the line BC at the points P and Q so
that B lies between C and Q. Prove that lines P E and QF intersect on the circumcircle of
triangle AEF .
8. (ISL 2001/G6) Let ABC be a triangle and P an exterior point in the plane of the triangle.
Suppose the lines AP , BP , CP meet the sides BC, CA, AB (or extensions thereof) in D, E,
F , respectively. Suppose further that the areas of triangles P BD, P CE, P AF are all equal.
Prove that each of these areas is equal to the area of triangle ABC itself.
9. (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 ` be a line passing through A. Suppose that `
intersects the interior of the segment DC at F and intersects line BC at G. Suppose also
that EF = EG = EC. Prove that ` is the bisector of angle DAB.

6.2

Hints

Hints have been obfuscated by ROT13.

1. ABG UNEQ
2. FGENVTUGSBEJNEQ
3. HFR NERN SBEZ
4. PVEPYR SBEZ FBYIR HIJ
5. PVEPF GURA J VF ZVAHF N O
6. QVFG SNPGBE GURA TRB VQ
7. CBJRE CBVAG FUBJF NRCD PLPYVP GURA NATYR PUNFR
8. PNERSHY FVTARQ NERNF
9. ERS OPQ... SVAQ S T FLF RDA RYVZ P M SNPGBE

31

Appendix A

Vectors Tutorial
Are you making progress on these problems? Or are you making Congress?
Zuming Feng

A.1

Definitions

Our definition of a vector is just an arrow that points from some reference point, denoted ~0, to
another point. In this paper, we are only considering vectors in the plane (e.g. two-dimensional
vectors), but the concept extends easily to multiple dimensions. It is assumed that the reader is
familiar with vector addition, the concept of magnitude, etc.
In order to make the theory of vectors applicable to geometry, we can associate every point
with the vector that points to it.
Of particular use in the world of vectors is the dot product defined by
Definition. Let ~u and ~v be vectors. Then the dot product ~u ~v is defined as
~u ~v = |~u||~v | cos
where is the angle between the two vectors.
The dot product has the following very nice properties:
1. ~u ~u = |u|2 .
2. ~u ~v = 0 if and only if the vectors are perpendicular.
3. The dot product is commutative, distributive, and associative.
~ C)
~
Exercise. Let A, B, C, D be points in the plane. Prove that AC BD if and only if (A
~ D)
~ = 0.
(B

A.2

Triangle ABC

~ = ~0 be the reference point. Then

Let ABC be a triangle in the plane with circumcenter O. Let O
~ = |B|
~ = |C|
~ = R, where R is the circumradius of the triangle.
its obvious that |A|
Now for the key fact:
~B
~ = R2
Fact. A

c2
2.

Cyclic variations hold.

32

Proof. Using the definition,

~B
~ = |A||
~ B|
~ cos AOB
A
= R2 cos 2ACB
= R2 (1 2 sin2 ACB)
= R2 (1 2 sin2 C)
1
= R2 (2R sin C)2
2
c2
= r2
2
the last step following by the Law of Sines.
~ + yB
~ + z C.
~
Barycentric coordinates allow us to write any point in the plane in the form xA
~ B
~
In that case, displacement vectors, points, etc. can all be written as a linear combination of A,
~
and C. The above fact allows us to compute arbitrary dot products via the distributive law. Using
properties 1 and 2 above, we can construct criterion for perpendiculars and lengths; this leads to
EFFT and the distance formula.

A.3

Special Points

~ = ~0, we have the orthocenter1 H

~ =A
~+B
~ + C,
~ the centroid G
~ = 1 H,
~ and the
Again setting O
3
1
~ In other words, the points on the Euler Line are very, very nice.
~ = H.
nine-point center N
2
This allows us, for example, to compute OH 2 as
~ H
~
OH 2 = H
~+B
~ + C)
~ (A
~+B
~ + C)
~
= (A
= 3R2 + 2(R2 a2 /2) + 2(R2 b2 /2) + 2(R2 c2 /2)
= 9R2 a2 b2 c2

~ A)
~ (B
~ C)
~ = 0, implying HA BC.
To check that this is correct, verify that (H

33

Appendix B

Formula Sheet
My formula sheets were pretty thorough, but perhaps they were missing something.
Richard Rusczyk

B.1

Standard Formulas

Throughout this paper, 4ABC is a triangle with vertices in counterclockwise order. The lengths
will be abbreviated a = BC, b = CA, c = AB. These correspond with points in the vector plane
~ B,
~ C.
~
A,
For arbitrary points P , Q, R, [P QR] will denote the signed area of 4P QR.1
Definition. Each point in the plane is assigned an ordered triple of real numbers P = (x, y, z)
such that
~ + yB
~ + zC
~ and x + y + z = 1
P~ = xA
Theorem 1 (Line). The equation of a line is ux + vy + wz = 0 where u, v, w are reals. (These u,
v and w are unique up to scaling.)
Coordinates of Special Points
From this point on, the point (kx : ky : kz) will refer to the point (x, y, z) for k 6= 0. In fact, the
equations for the line and circle are still valid; hence, when one is simply intersecting lines and
circles, it is permissible to use these un-homogenized forms in place of their normal forms. Again,
the coordinates here are not homogenized!
Point
Centroid
Incenter
Symmedian point
Excenter
Orthocenter
Circumcenter

Coordinates
G = (1 : 1 : 1)
I = (a : b : c)
K = (a2 : b2 : c2 )
Ia = (a : b : c), etc.
H = (tan A : tan B : tan C)
O = (sin 2A : sin 2B : sin 2C)

Sketch of Proof
Trivial
Angle bisector theorem
Similar to above
Similar to above
Use area definition
Use area definition

If absolutely necessary, it is sometimes useful to convert the trigonometric forms of H and O

into expressions entirely in terms of the side lengths (cf. [3, 5]) by
O = (a2 (b2 + c2 a2 ) : b2 (c2 + a2 b2 ) : c2 (a2 + b2 c2 ))
1

For ABC counterclockwise, this is positive when P , Q and R are in counterclockwise order, and negative otherwise. When ABC is labeled clockwise the convention is reversed; that is, [P QR] is positive if and only if it is oriented
in the same way as ABC. In this article, ABC will always be labeled counterclockwise.

34

and
H = ((a2 + b2 c2 )(c2 + a2 b2 ) : (b2 + c2 a2 )(a2 + b2 c2 ) : (c2 + a2 b2 )(b2 + c2 a2 ))
Definition. The displacement vector of two (normalized) points P = (p1 , p2 , p3 ) and Q = (q1 , q2 , q3 )

is denoted by P Q and is equal to (p1 q1 , p2 q2 , p3 q3 ).

A Note on Scaling Displacement Vectors
In EFFT, one can write a displacement vector (x, y, z) as (kx : ky : kz), and the theorem will still
be true. This is also true for Strong EFFT, but NOT for the distance formula.

Theorem 4 (Evans Favorite Forgotten Trick). Consider displacement vectors M N = (x1 , y1 , z1 )

and P Q = (x2 , y2 , z2 ). Then M N P Q if and only if

0 = a2 (z1 y2 + y1 z2 ) + b2 (x1 z2 + z1 x2 ) + c2 (y1 x2 + x1 y2 )

Corollary 5. Consider a displacement vector P Q = (x1 , y1 , z1 ). Then P Q BC if and only if

0 = a2 (z1 y1 ) + x1 (c2 b2 )
Corollary 6. The perpendicular bisector of BC has equation
0 = a2 (z y) + x(c2 b2 )

Theorem 7 (Distance Formula). Consider a displacement vector P Q = (x, y, z). Then

|P Q|2 = a2 yz b2 zx c2 xy
Theorem 8. The general equation of a circle is
a2 yz b2 zx c2 xy + (ux + vy + wz)(x + y + z) = 0
for reals u, v, w.
Corollary 9. The circumcircle has equation
a2 yz + b2 zx + c2 xy = 0
Theorem 10 (Area Formula). The area of a triangle with vertices P = (x1 , y1 , z1 ), Q = (x2 , y2 , z2 )
and R = (x3 , y3 , z3 ) is


 x1 y1 z1 


[P QR] = [ABC]  x2 y2 z2 
 x 3 y3 z3 
Corollary 11 (First Collinearity Criteria).
R = (x3 : y3 : z3 ) are collinear if and only if

 x1

 x2

 x3

The points P = (x1 : y1 : z1 ), Q = (x2 : y2 : z2 ) and


y1 z1 
y2 z2  = 0
y3 z3 
35

Corollary 12 (Line Through 2 Points). The

and Q = (x2 : y2 : z2 ) is

 x1

 x2

 x

equation of a line through the points P = (x1 : y1 : z1 )


y1 z1 
y2 z2  = 0
y z 

Corollary 13 (Second Collinearity Criteria). The

R = (x3 , y3 , z3 ), are collinear if and only if

 x1 y1 1

 x2 y2 1

 x3 y3 1

points P = (x1 , y1 , z1 ), Q = (x2 , y2 , z2 ) and




=0



Cyclic variations hold.

Theorem 14 (Strong EFFT). Suppose M , N , P and Q are points with

M N = x1 AO + y1 BO + z1 CO

P Q = x2 AO + y2 BO + z2 CO
If either x1 + y1 + z1 = 0 or x2 + y2 + z2 = 0, then M N P Q if and only if
0 = a2 (z1 y2 + y1 z2 ) + b2 (x1 z2 + z1 x2 ) + c2 (y1 x2 + x1 y2 )
Corollary 15. The equation for the tangent to the circumcircle at A is b2 z + c2 y = 0.
Definition (Conways Notation). Let S be twice the area of the triangle. Define S = S cot , and
define the shorthand notation S = S S .
Fact. We have the identities
SB + SC = a2
and
SAB + SBC + SCA = S 2
Fact. O = (a2 SA : b2 SB : c2 SC ) and H = (SBC : SCA : SAB ) =

1
SA

1
SB

1
SC

Theorem 16 (Conways Formula). Given point P, with counter-clockwise directed angles ]P BC =

and ]BCP = , we have P = (a2 : SC + S : SB + S ).
Lemma 17 (Parallelogram Lemma). The points ABCD form a parallelogram iff A + C = B + D
(here the points are normalized), where addition is done component-wise.
Lemma 18 (Concurrence Lemma). The three lines ui x+vi y+wi z = 0, for i = 1, 2, 3 are concurrent
if and only if


 u1 v1 w1 


 u2 v2 w2  = 0


 u3 v3 w3 

36

B.2

More Obscure Formulas

Heres some miscellaneous formulas and the like. These were not included in the main text.
From [7],
Theorem 19 (Leibniz Theorem). Leibniz theorem: Let Q be a point with homogeneous barycentric
coordinates (u : v : w) with respect to 4ABC. For any point P on the plane ABC the following
relation holds:
uP A2 + vP B 2 + wP C 2 = (u + v + w)P Q2 + uQA2 + vQB 2 + wQC 2

B.2.1

Other Special Points

Point
Gregonne Point [3]
Nagel Point [3]
Isogonal Conjugate [1]
Isotomic Conjugate [1]
Feuerbach Point [8]
Nine-point Center

B.2.2

Coordinates
Ge = ((s b)(s c) : (s c)(s a) : (s a)(s b))
Na =(s a : s b : s c)
P =

Pt =

a2

x

1
x

b2
c2
y : z
1 1
y : z

F = (b + c a)(b c)2 : (c + a b)(c a)2 : (a + b c)(a b)2

N = (a cos(B C) : b cos(C A) : c cos(A B))

Special Lines and Circles

Nine-point Circle
Incircle
Euler Line

a2 yz b2 xz c2 xy + 21 (x + y + z)(SA x + SB y + SC z) = 0


a2 yz b2 zx c2 xy + (x + y + z) (s a)2 x + (s b)2 y + (s c)2 z = 0
SA (SB SC )x + SB (SC SA )y + SC (SA SB )z = 0

37

Bibliography
[1] Hojoo Lee, Tom Lovering, and Cosmin Pohoata. Infinity.
Relevant excerpt at http://www.bmoc.maths.org/home/areals.pdf.
[2] Zachery Abel. Barycentric Coordinates.
http://zacharyabel.com/papers/Barycentric_A07.pdf
[3] Eric W. Weisstein Barycentric Coordinates. From MathWorldA Wolfram Web Resource.
http://mathworld.wolfram.com/BarycentricCoordinates.html
[4] Francisco J. Garca Capit
an. Coordenadas Baricentricas.
http://www.aloj.us.es/rbarroso/trianguloscabri/sol/sol202garcap/cb.pdf
[5] Paul Yiu. The use of homogenous barycentric coordinates in plane euclidean geometry.
http://math.fau.edu/Yiu/barycentricpaper.pdf
[6] AoPS. Practice Olympiad 4. WOOT 2011-2012.
[7] AoPS Forum.
http://www.artofproblemsolving.com/Forum/viewtopic.php?p=2162335#p2162335
[8] AoPS Forum.
http://www.artofproblemsolving.com/Forum/viewtopic.php?f=46&t=399047&p=
2225645

38