Directed Angles

Kada Williams

March 28, 2019

Contents
1 Introduction: analytic geometry in Olympiads 1

2 Recap of what R2 and C are 2

3 Directed lines 2

4 Directed angles 3

5 The cosine and sine rules 4

6 The Ratio Lemma, Menelaus, Ceva 4

7 Correct trig solution to the USAMO problem 5

8 Bonus: proof of circle theorems using C 6

1 Introduction: analytic geometry in Olympiads

This note is inspired by the imprecision of all the trigonometric solutions to US-
AMO problem 2012/5 on AoPS. The problem statement, by the way, is this:

Let P be a point in the plane of 4ABC, and γ a line passing through P .

Let A0 , B 0 , C 0 be the points where the reflections of lines P A, P B, P C with
respect to γ intersect lines BC, AC, AB respectively. Prove that A0 , B 0 , C 0 are
collinear.

I’m also reflecting to some extent on a sense I have that in Olympiads, synthetic
geometry mixes perhaps a bit too freely with coordinate methods, trigonometric
methods, and projective methods.

1
 You’ve might have heard about the distinction between synthetic and analytic
geometry: in synthetic geometry, you’ve ≈ 20 axioms and are supposed to create
ever-beautiful proofs using small deductions, but in analytic geometry, the plane is
defined as R2 , and proofs proceed algebraically. I’m unsure of the extent to which
the two setups contain each other, but somehow that’s something we rarely or never
need to worry about. In any case, I’m going to treat angles purely analytically. This
will come at the cost of having to re-prove a number of synthetic facts we take for
granted, but seeing them in an algebraic light gives them a lot more depth, I believe.

2 Recap of what R2 and C are

I’m going to run through this quickly. There are a million details to check – that’s
higher math for you. Please don’t waste your time on fleshing them all out.
Points of the plane are pairs (x, y) ∈ R2 . We may add points and multiply them
by λ ∈ R (component-wise). If we want to multiply points, we can do it like this:

(a, b) · (c, d) = (ac − bd, ad + bc).

To stress that we’ve got this special multiplication too, we now call our plane C. We
can show that multiplication is commutative, associative, and distributive over +.
Moreover, we’ve a copy of R in C if we label (x, 0) ∈ C as x ∈ R. This way, with
i := (0, 1), we have x + iy = (x, y) and i2 = −1.
The numbers i and −i behave the same, so it’s worth comparing z = (a, b) with
the so-called conjugate z = (a, −b). We then find

zz = (a + bi)(a − bi) = a2 + b2 =: |z|2 .

That’s useful, first because we can now show that c + di = (a, b) implies c = a, d = b
(subtract and use ≥ 0), second because we can now divide: z −1 = |z|z 2 .
I want to talk about angles, so what I need is directions, so what I need is one
point on each ray, e.g. the unit circle S 1 = {z ∈ C : |z| = 1}. We need trig functions,
so my claim is that there exists cos and sin with the usual properties (period 2π,
addition formulas) such that every point of S 1 can be written in the form

(cos θ, sin θ) = cos θ + i sin θ = ” eiθ ”.

(In case you’re

P∞ interested, arguably the best proof of my claim here is to define
n
exp(z) = n=0 zn! for all z ∈ C (uses Analysis), then define cos, sin using power se-

n define π has half the period of cos, finally noting that
ries, then prove periodicity and
exp(1) = e := limn→∞ 1 + n1 so we can abuse notation and write exp(z) = ez .)

3 Directed lines
A line is defined as a set

L = {(x, y) ∈ R2 : Ax + By + C = 0},

2
where A, B are not both zero. Introducing the notion of vector scalar multiplication
(x1 , y1 ) · (x2 , y2 ) = x1 x2 + y1 y2 (all of a sudden we refer to points in R2 as vectors),
we’re allowed to say u ⊥ v iff u · v = 0. We can write a line in terms of its normal
vector:
L = {x : n · x = c},
where |n| > 0.1 We lose nothing if we divide by a positive number so that n ∈ S 1 .
However, with our use of n, the line became directed. If we take the scalar prod-
uct of x with n, we’re saying how far we’re going in a direction perpendicular to L,
except that now we’ve a signed distance n·x−c which is positive on the left half-plane
and negative on the right half-plane.
We’ve signed lengths on directed lines (as seen, for instance, in Menelaus’ the-
orem). If A, B ∈ L and we rotate n by −90◦ to get the direction vector u (so
u = −in = (n2 , −n1 )), we define the signed length AB ∈ R by the equation
−−→
AB := B − A = ABu.

4 Directed angles
Given directed lines a and b with direction vectors a and b, their directed angle is
the angle between the direction vectors, namely β − α where a = eiα and b = eiβ .
Issue: the angle is not unique, the set of its possible values is a residue class of R
modulo 2π. Fine, then we let the directed angle be a residue class of reals rather than
a real number. Nothing wrong with that. Alternatively, we may simply identify the
angle between a and b with the complex number b/a ∈ S 1 .
If a = (a1 , a2 ) and b = (b1 , b2 ), and their directed angle is θ, then we have

cos θ + i sin θ = (a1 b1 + a2 b2 ) + i(a1 b2 − a2 b1 ).

We identify cos θ = a · b and sin θ as the third coordinate of a × b.2

In Olympiad settings, due to circle theorems, it is more comfortable to use di-
rected angles between undirected lines. Given a line, we’ve two opposite normals
(±n) we can choose from, and so the set of possible angle values between the two
normals is a residue class of R modulo π.3
For directed lines a, b, c, the equality b/a · c/b · a/c implies

∠(a, b) + ∠(b, c) + ∠(c, a) = 0 (mod 2π).

The same modulo π, then, holds for undirected lines.

1 Yes,
√ √
it’s also customary to define the length of a vector as x · x, but it’s the same as the zz defini-
tion.
2 We embed (x, y) ∈ R2 into R3 as (x, y, 0) and then define the vector product

(x1 , x2 , x3 ) × (y1 , y2 , y3 ) = (x2 y3 − x3 y2 , x3 y1 − x1 y3 , x1 y2 − x2 y1 ).

The vector product obeys x × x = 0 as well as bilinearity.
3 One memorable theorem to connect with this is that if our two lines have gradients m and m , then
1 2
m2 −m1
the tangent of the angle between them is 1+m m
. And tangents have mod π periodicity.
1 2

3
5 The cosine and sine rules
The cosine and sine rules are, fundamentally, algebraic theorems concerning three
directed lines a, b, c with direction vectors a, b, c. Suppose a ∩ b = {C}, b ∩ c = {A},
c ∩ a = {B}.
The cosine rule.
−→ −−→ −→ −−→
BC 2 = (AC − AB) · (AC − AB)
= AC 2 (b · b) − 2 · AB · AC(b · c) + AB 2 (c · c) = AB 2 + AC 2 − 2 · AB · AC · cos ∠(b, c).

The sine rule. We use the vector area of triangle ABC in R3 , given by

2T = (B − A) × (C − A) = A × B + B × C + C × A.

We may use the symmetry of the resulting expression to write this in three ways.
Notice that (B − A) × (C − A) = AB · AC · (c × b), so if we take the third coordinates,
we’ll find

CA · AB · sin ∠(b, c) = AB · BC · sin ∠(c, a) = BC · CA · sin ∠(a, b),

which, when divided by AB · BC · CA, gives the directed version of the sine rule.

6 The Ratio Lemma, Menelaus, Ceva

Let directed lines a, b, c pass through O, and let them intersect directed line ` at
A, B, C. Then we have (note: sin(`1 , `2 ) 6= 0 if `1 , `2 aren’t parallel)

AC sin ∠(a, c) · sinOA

∠(c,`) sin ∠(a, c) OA
= OB
= · .
CB sin ∠(c, b) · sin ∠(c,`) sin ∠(c, b) OB

This is the Ratio Lemma. Note the sine ratio involved: (abc) := sin ∠(a,c)
sin ∠(c,b) is negated if
the direction of a or b is reversed, but is independent of the direction of c.
Ceva’s theorem involves a triangle ABC and a point O not on its sides. Directing
lines OA, OB, OC (imagine planting a direction vector on each line) which intersect
the opposite sides at D, E, F , we find

BD sin ∠(OB, OD) OB

= · ,
DC sin ∠(OD, OC) OC
where OD and OA are the same line. Multiplying these gives
BD CE AF
· · = 1.
DC EA F B
We can talk about signed ratios on undirected lines as well, because reversing the
direction leaves them unchanged. More nuanced is the trigonometric form of Ceva’s
theorem (to prove it, we apply the ratio lemma on AB, AC, AO etc.):
sin BAD sin CBE sin ACF
· · = 1.
sin DAC sin EBA sin F CB

4
Here, sin BAD really means sin ∠(BA, AD), which does depend on the orientation.
So Ceva sine ratios are strange: if planting a direction vector on each line doesn’t
work, just show that the absolute value of the left-hand side is 1 and show that the
signs of the two sides in length-Ceva are equal.
We can similarly prove Menelaus’ theorem and its converse: given a triangle
ABC and points D, E, F on sides BC, CA, AB (A, B, C, D, E, F all distinct), we have
BD CE AF
· · = −1
DC EA F B
if and only if D, E, F are collinear. To prove this, we center things at D, say, directing
every line involved. Then
CE sin CDE DC AF sin ADF DA
= · , = · .
EA sin EDA DA FB sin F DB DB
Therefore,
BD CE AF sin CDE sin ADF
· · =− · ,
DC EA F B sin EDA sin F DB
where CD and DB are the same line, so this equals −1 if and only if
sin BDE sin BDF
= ,
sin EDA sin F DA
which holds if and only if DE and DF are the same line (exercise: check this deduc-
tion, using the ratio lemma).

7 Correct trig solution to the USAMO problem

Let P be a point in the plane of 4ABC, and γ a line passing through P .

Let A0 , B 0 , C 0 be the points where the reflections of lines P A, P B, P C with
respect to γ intersect lines BC, AC, AB respectively. Prove that A0 , B 0 , C 0 are
collinear.

Configuration issues: P 6= A, B, C is implicitly assumed; if P ∈ AB then A0 = B,

B = A, C 0 = P ; from now on assume P isn’t on any of the sides. For the proof, we’ll
0

need lines P A, P B, P C, P A0 , P B 0 , P C 0 to be distinct; if P A, P B 0 are the same then

P A, P B are symmetric wrt γ, so A0 = B, B 0 = A, and C 0 lies on AB so done.
Solution. Put a direction vector on all the lines involved. We use the converse of
Menelaus’ theorem: we prove
BA0 CB 0 AC 0
· · = −1.
A0 C B 0 A C 0 B
By the ratio lemma applied to P B, P A0 , P C,
BA0 sin BP A0 P B
= · .
A0 C sin A0 P C P C
The reflection construction entails ∠(P A, γ) = ∠(γ, P A0 ) etc. So
∠(BP, P A0 ) = ∠(BP, γ) + ∠(γ, P A0 ) = ∠(BP, γ) + ∠(P A, γ).
Also, ∠A0 P C = −∠CP A0 . Using all these, we can see that the cyclic product for
Menelaus is indeed −1. 

5
8 Bonus: proof of circle theorems using C
The story begins with complex numbers and Möbius maps, which have the form
z 7→ az+b
cz+d where a, b, c, d ∈ C with ad − bc 6= 0. Compositions of Möbius maps are
Möbius maps (in fact, there’s an isomorphism with matrix multiplication), and all
Möbius maps can be composed using z 7→ z + a (a ∈ C) and z 7→ z1 .
z+z z−z
Since x = 2 and y = 2i , the equation of a line in C has the form

Bz + Cz + D = 0

where B, C, D ∈ C and B, C are conjugates, D ∈ R. (The additional condition just

means that conjugating the equation gives the same equation. Observe that if we
wanted to intersect two lines, we’d be solving a system for z, z instead of for x, y.)
The equation of a circle in C is

|z − w| = r ⇐⇒ (z − w)(z − w) = r2 .

Therefore, the equation of a line/circle is

Azz + Bz + Cz + D = 0,
1
√
where the radius is A BC − AD (indicating how BC > AD is required, and how
the radius is ”∞” for a line). Now it’s very easy to check that z 7→ z + a and z 7→ z1
map line/circles to line/circles, so all Möbius maps map line/circles to line/circles.
We’ve so far glossed over details such as where z 7→ z1 sends 0. The elegant
way to settle this is to talk about C∞ := C ∪ {∞}.4 What this really means is that
we’re working in C2 quotiented by the equivalence relation (a, b) ∼ (c, d) if (c, d) =
(λa, λb) for some nonzero λ ∈ C. In these homogeneous coordinates, a Möbius map is
(z1 , z2 ) 7→ (az1 + bz2 , cz1 + dz2 ) (so literally matrix multiplication), and a line/circle
is Az1 z1 + Bz1 z2 + Cz1 z2 + Dz2 z2 = 0. Then ∞ can be seen as the equivalence class
of (1, 0). Sorry if this doesn’t feel completely logical: there are indeed a lot of details
to work out.
The reason I bring up ∞ is because it’s easy to show that only the identity map
preserves 0, 1, ∞, and so given u, v, w
u−w u−z
z 7→ /
w−v z−v

is the unique Möbius map sending u, v, w to ∞, 0, 1. Define the cross-ratio5 of a, b, c, d ∈

C to equal the image of d in the Möbius map that sends a, b, c to ∞, 0, 1, namely

a−c a−d
(abcd) := / .
c−b d−c
Our definition immediately tells us that cross-ratio is preserved under Möbius maps!
Therefore, a, b, c, d are on the same line/circle if and only if 0, 1, ∞, (abcd) are on
the same line/circle, i.e. if and only if (abcd) ∈ R (or is ∞).
4 Also called the Riemann sphere, as by stereographic projection it bijects with the sphere in R3 . Actu-

ally, every rotation of this sphere corresponds to a Möbius map.

5 no, not the projective cross-ratio: that’s a bit different

6
 So... what about cyclic quadrilaterals? Well, now we can say that a, b, c, d are
collinear/concyclic if and only if a−c a−d
c−b and d−b are real multiples of each other. There
is a unique θ and φ in [0, π) with which they can be written as a real number times eiθ
and eiφ , respectively. Furthermore, the residue class of θ mod π is the directed angle
∠(cb, ca): if u is a possible direction vector for cb, then eiθ u is a possible direction
vector for ca. It follows that a, b, c, d are collinear/concyclic if and only if with mod
π directed angles, ∠(cb, ca) = ∠(db, da).
Or, if we’d like a much more elementary and simple proof, we can show that if
A, B, P lie on a circle with center O, then doubling the mod π angle ∠(P A, P B) gives
−→ −−→ p−b
the mod 2π angle ∠(OA, OB). In other words, if p−a = ceiθ for (nonzero) c ∈ R, then
b−o 2iθ
a−o = e . Translation and multiplying by a complex number, we may assume that
a, b, p ∈ S 1 with O = 0.
p−b p−b
p
Let’s write p−a = ·
. Imagining a diagram with two isosceles triangles,
p
p−a
2
we want to claim that p−b
p
b
is a positive real times −p . So we calculate
2
2pb − p2 − b2
  
p−b −p p b
= =2− + ,
p b pb b p

and wonder at how in the bracket there’s a unit complex number (6= 1) plus its
conjugate, which certainly is a real less than 2. So our claim is established and our
theorem proven. It follows that ∠(P A, P B) is independent of P .