Professional Documents
Culture Documents
Matija Delić
July 2020
• Inversion
• Harmonic bundles
• Common projective geometry lemmas1
• Pascal’s theorem
• Brocard’s theorem
• Pole, polar and La Hire’s theorem
The list is (as you will see) far from complete, but is a place to start. To
demonstrate actual techniques, we will follow a series of problems to explain.
1
Problem 1: In 4ABC with centroid G, M and N are the midpoints of
AB and AC, and the tangents from M and N to the circumcircle of 4AM N
meet BC at R and S, respectively. Point X lies on side BC satisfying ∠CAG =
∠BAX. Show that GX is the radical axis of the circumcircles of 4BM S and
4CN R.
First thing first, you should notice immediately that AX is the A-symmedian
line to 4ABC, and know that it implies XB AB 2
XC = ( AC ) . Remember Technique 0?
Now, let’s translate the problem requirement. GX being the desired radical
axis is equivalent to both G and X lying on the radical axis, and this allows us
to think about the separately. Let’s go for X first, and remember, equivalences
are your friend!
2
XR AB 2
XR · XC = XS · XB ⇐⇒ =
XS AC 2
⇐⇒ XR · AC 2 = XS · AB 2
⇐⇒ (XB − XR) · AC 2 = (XC − CS) · AB 2
⇐⇒ XB · AC 2 + CS · AB 2 = XC · AB 2 + BR · AC 2
AC 2 AB 2
XB · AC 2 + AB 2 = XC · AB 2 + AC 2 ⇐⇒ XB · AC 2 = XC · AB 2
2BC 2BC
XB AB 2
⇐⇒ =
XC AC 2
Which is exactly the first thing we were able to conclude was true, so we got
X out of the way, moving on to G. Here Technique 1 comes to the rescue once
more: forget about G as a centroid and think about G as BN ∩ CM . Another
tip, (Technique 2 spoiler) draw a good diagram! If you do so, you might as well
notice that G looks to be the radical center of AM N , CN R and BM S.
This is obviously sufficient for us, so let’s prove it.
3
Problem 2: Let ABC be an acute, scalene triangle with circumcenter O,
and let D be a point lying on side BC. The perpendicular bisectors of DB and
DC meet lines AB and AC at points P and Q, respectively. Point X is the
reflection of D in P Q. Show that AX||BC if and only if D, O, X are collinear.
First step, play around a little. The circle through AP Q and O is looking
like something pulled out of Miquel’s theorem, so let’s add circles through CQO
and BP O. Because we know O is going to be a part of AP Q, we should get
4
that the two new circles intersect at BC.
Here we might notice that ∠OCA + ∠OBA = ∠BAC, so the intersection of
these two circles will actually lie on P QD! This is where our friendly point
comes into play, define G = P QD ∩ BC.
The point (no pun intended) of this is to add the missing pieces of the ”mech-
anism” that forces O to be on the desired circle, i.e. looking for the cause of O
being there in the first place, which gets me to my next point:
Why O? This is the heart of Technique 1 actually, you need to figure out
what property of O places it on AP Q. Let’s not get too philosophical, it will
make itself clear later. First, it is useful to tinker with G a little, notice that
∠GP Q = ∠GDQ = ∠ACB and ∠GQP = π − ∠GDP − ∠ABC =⇒ 4DQP ∼
4ABC.2
We have note proven it yet, but we secretly do know that G is going lie
on OQC, which is equivalent to ∠OGD = ∠OCA. and also ∠QOG =
π − ∠ACB = π − ∠QP G, and all of this basically means O is supposed to
be the orthocenter of 4DQP .
So, let’s add friendly point number two: let H be the orthocenter of 4DQP .
We know ∠QHG = π − ∠QP G = π − ∠QCG =⇒ HECG is cyclic, and
∠HGQ = π − ∠ABC =⇒ ∠ACH = ∠HGQ = ∠ACO. Doing this similarly
for the other side forces H = O!
Quick Note: ”Add and intersection of P QD with BC and then the orthocen-
ter of a new triangle and voila” sure is a pretty solution, but in that form completely
unmotivated, so you can’t learn much from it. Hopefully these techniques help you
to see the ”behind the scenes” part of each solution. Also, I know this belongs in the
Acknowledgements section, but it was more appropriate here. My paper, my rules.
Actually, you can set R to be any point on the arc BC of Ω not containing
A, set Q = AR ∩ EF and X = P Q ∩ γ, and the problem will remain true.
2 You may notice that I am completely ignoring configuration issues, to follow just assume
5
Technique 4: Simplify, simplify, simplify
Pretty self explanatory- we will get to it after moving a few things out of the
way.
You have probably noticed that the concurrence we need to prove is just
silly. Well, Technique 1 can never get old, so let’s reduce it to something more
reasonable. If you denote KR ∩ γ = G, and ∠GKA = ∠RBA = ∠AQF , be-
cause F QRB is cyclic. Note that G ∈ XD ⇐⇒ ∠DXA = ∠AQF . Now that
is something we can work with.
6
Now, please notice how much simpler the new picture is, and in majority of
the cases, this also means the problem itself gets much simpler.
Easy to see ∠AXQ = ∠AXP = ∠AHP = ∠AH ∗ Q, and from this the con-
clusion easily follows.
This is the kind of problem for which I want to emphasize a certain strategy,
and that is, guess the point you are looking for. Seriously, this works more
often then you would expect.5
4 If you find this jumping from one diagram to another then back unnecessary, that is
completely valid, I only consider this path to be the most natural one
5 The reason I say this is because I have guesses XY ∩ BC is going to be the orthocenter
7
Technique 5: Point Bash
A.k.a. (Technique 4)−1 . There is a circle, and quite a few points on in. Here
you aim to leverage Paskal’s and Brocard’s theorems as much as possible, so
don’t be afraid to consider adding more points.
Now rephrase the wanted result:6 4OP Q seems a bit random, but notice
that a point say S is the orthocenter of 4OP Q iff O is the orthocenter of
4SP Q! Now that seems like something straight out of Brocard’s theorem, so
let’s go that route.
8
The point we are looking for should be the pole of QP , but if you draw a
nice enough diagram, you should notice that Q, P and I seem collinear. That
should be enough for you to conjecture that the point we have denoted as H
above is the desired orthocenter.
Now what remains is that HQ is the polar of P ; notice that for this we could
use Brocard’s theorem if we had HA, BP and CQ concurrent on ABC. For
this, add D = BP ∩ ABC and aim to leverage Paskal’s theorem again: on
CABBDC it tells us that CD ∩ AB lies on the line through P and CC ∩ BB,
so it has to be Q!, i.e. CQ ∩ BP = D ∈ ABC.
Note on Paskal’s theorem: Aim to use it when you have two of the three points
that are the intersections already known, otherwise you might make a mess.
Examples
Talking about a mess, these technique explanation didn’t really work out as
planned, so let’s combine them through a few more problems! I strongly sug-
gest you try them on your own first.
9
Also it is easy to see P, D and F are collinear, so we can translate ∠CAE =
∠BAP to ∠BAP = ∠BF P . Notice how similar this looks to what we actually
need to prove.
From here you can conjecture that you can ditch the majority of the picture,
and that it is enough to have ∠P AB = ∠P F B =⇒ ∠P BA = ∠P CA. In fact,
this is true, so let’s draw another picture to prove it.
The best way of thinking about the mentioned angles is to draw the lines
10
through P parallel to the sides of the parallelogram, and denote the intersec-
tions as on the diagram. You may notice that the problem condition may be
translated to pairs of circles P AB and P BF , and P AC and P F C have
the same radius, and the goal is to prove all these radii are equal.
This is equivalent to sin ∠AP B = sin ∠CP F . We can show in fact that they
are supplementary, which is equivalent to ZC · ZF = ZP · P X(figure it out by
sliding 4W Y P such that W Y matches CF ). But this is finally equivalent to
P X · P Z = P Y · P W , i.e. we want to show XY ZW is cyclic.
The last two paragraphs explain how your thought process would be like
when you try to solve this, but you actually don’t need any of it if you really
prove the cyclicity. It’s just how you arrive to try it.
The point K is very loosely connected with the rest of the diagram; you
can’t do much angle chasing with it, or anything else really. So we might as well
redefine it: let K be the point such that I is the incenter of 4ABK instead.
Another reason we are doing this, having I be the incenter of 4ABK provides
more information than just having K = BD ∩ IC.
11
First, by redefining K we lose the condition that it lies on BD, but it is easy
to prove. It is equivalent to ∠ABE = ∠ABD/2, for which E being the incenter
of 4ABD is sufficient. But we have ∠AED = π − ∠DCB = π/2 + ∠BAC/2
together with ∠DAE = ∠BAE, so this part is done.
2π−∠BEC+∠ABC π+∠ADB+∠ABD
To finish, you have ∠AIB = ∠AEB+∠EBF = 2 = 2 =
π − ∠BAC/2.
12
The difficulty of this problem lies on actually finding a good equivalence to
what we need to prove, quite the opposite of the last example. The problem
can’t be simplified in an obvious manner either, so we resort to either point
bash or friendly points.
2
Firstly, we may get the obvious out of the way: CH 0 = AH = AD AC =⇒
CH 0 ·CA = AD2 , so symmetry implies ABH 0 I 0 is cyclic. Obviously this implies
also F GI 0 H 0 is cyclic. Also notice DF ||BC and DG||AC.
If you draw a good enough diagram, or think for long enough,7 you may
notice DF and H 0 I 0 intersect on Γ, so add DF ∩ Γ = X and DG ∩ Γ = Y to
try and prove the conjecture. Notice that F X · F D = F H · F C = F H 0 · F A, so
AXH 0 D is cyclic =⇒ ∠H 0 XD = ∠BAC. Similarly we find ∠GY I 0 = ∠ABC,
and because we have ∠XDY = ∠ACB, we must have XH 0 and Y I 0 to be the
same line, i.e. XY and H 0 I 0 are the same line.
Also by the angle chase we have done here we get ∠DGF = ∠DXY =⇒
F GXY is cyclic. Now we can leverage DX||BC : ∠QCI 0 = ∠XDC = ∠CY I 0 =⇒
4QI 0 C ∼ 4QCY . This allows us to deduce QI 0 · QY = QC 2 , so think about
what would happen if we inverted about Q with radius QC: circles XCY and
I 0 H 0 C would switch places, but we know QC = QP is true, so P would stay
in place.
7 In the lack of ability to find a nice equivalence for the main claim, this search for seemingly
13
This tells us the missing claim is that P is on H 0 I 0 C, which will then be suf-
ficient. For this we use the Radical axis theorem, it will be sufficient to prove
F G is the radical axis of ABI 0 H 0 and Γ, and we would use the theorem on
these two circles and H 0 I 0 C. But we have F H · F C = F H 0 · F A =⇒ F is on
the radical axis of ABI 0 H 0 and Γ, and by symmetry so is G, so we are finished.
14