Let $\triangle ABC$ be a triangle with incenter $I$ and orthocenter $H$. The incircle of $\triangle ABC$ touches $BC$, $CA$, $AB$ at $D, E, F$ respectively. Let $D'$ be the reflection of $D$ through $I$ and let $S$ be the midpoint of $AI$. How can we prove that $SD'$, $EF$and $HI$ are concurrent?

enter image description here

  • $\begingroup$ I can bash this in a reasonable length though I would like to see an elegant solution. $\endgroup$ – dezdichado Oct 18 '18 at 17:20

I'll prove the statement using complex coordinates so I'll denote all points with small letters, sorry. I won't elaborate every single detail but you should still be able to follow the proof.

enter image description here

Put the triangle into a complex plane and, WLOG, assume that the radius of the inscribed circle is equal to 1. Choose point $i$ as the origin, with real axis along the line $id$ (see picture).

Draw point $v$ symmetric to $d'$ with respect to $s$. Because $as=si$, $sd'=sv$ and $\angle asv=\angle isd'$ triangles $\triangle asv$ and $\triangle isd'$ are congruent (SAS). Consequentially $av=id'=1$

As a consequence $\angle sav=\angle sid'$ so lines $av$ and $id'$ must be parallel. Line $d'i$ is perpendicular to $bc$ so line $av$ must be perpendicular to $bc$ too, thus passing through orthocenter $h$.

Taking all that into account, it means that triangles $\triangle ud'i$ and $\triangle uvh$ are similar.

To finish the proof we need two lemmas and I won't prove them in detail:

Lemma 1:


You can find the proof here on page 5. Even without a proof you should be able to figure it out by yourself.

Lemma 2:


The proof can be found here, on page 4.

So we have the following coordinates:

$$i=0\tag{1}$$ $$d'=-1\tag{2}$$ $$v=a+1=\frac{2ef}{e+f}+1\tag{3}$$ $$h=\frac{2(d^2e^2+e^2f^2+f^2d^2+def(d+e+f))}{(d+e)(e+f)(f+d)}\tag{4}$$

Because triangles $\triangle ud'i$ and $\triangle uvh$ are similar:


Replace (1), (2), (3), (4) into (5) and solve for $u$:

$$u=\frac{e^2 f^2+e^2 f+e^2+e f^2+e f+f^2}{e^2 f+e f^2+2 e f+e+f}\tag{6}$$

The points $u$, $e$ and $f$ are collinear if and only if:

$$\lambda=\frac{u-e}{f-u}\in R\tag{7}$$

...or, if you replace (6) into (7):

$$\lambda=\frac{(e+1) f \left(e^2-f\right)}{e (f+1) \left(e-f^2\right)}\in R\tag{8}$$

The trick is to prove that (8) is a real number for any $e,f$ from the unit circle. This is equivalent to proving:

$$\lambda - \bar\lambda=0$$


$$\frac{(e+1) f \left(e^2-f\right)}{e (f+1) \left(e-f^2\right)} - \frac{(\bar e+1) \bar f \left(\bar e^2-\bar f\right)}{\bar e (\bar f+1) \left(\bar e-\bar f^2\right)}=0$$

...which leads to:

$$e^3 \bar{e}^2 f \bar{f}+e^3 \bar{e}^2 f+e^3 (-\bar{e}) f \bar{f}^3-e^3 \bar{e} f \bar{f}^2-e^2 \bar{e}^3 f \bar{f}\\ -e^2 \bar{e}^3 \bar{f}+e^2 \bar{e}^2 f-e^2 \bar{e}^2 \bar{f}-e^2 \bar{e} f \bar{f}^3+e^2 \bar{e} \bar{f}^2\\ +e^2 f \bar{f}^2+e^2 \bar{f}^2+e \bar{e}^3 f^3 \bar{f}+e \bar{e}^3 f^2 \bar{f}+e \bar{e}^2 f^3 \bar{f}-e \bar{e}^2 f^2\\ -e \bar{e} f^3 \bar{f}^2+e \bar{e} f^2 \bar{f}^3-e f^3 \bar{f}^2-e f^2 \bar{f}^2-\bar{e}^2 f^2 \bar{f}-\bar{e}^2 f^2\\ +\bar{e} f^2 \bar{f}^3+\bar{e} f^2 \bar{f}^2=0\tag{9}$$

It looks pretty hopeless, but for points $e$ and $f$ on the unit circle:

$$e\bar e = f \bar f =1\tag{10}$$

Replace (10) into (9) and you will prove that (9) is indeed zero. In other words $\lambda - \bar\lambda=0$ for any $e,f$ so $\lambda$ must be a real number. Therefore points $e,f,u$ are collinear and lines $sd'$, $hi$ amd $ef$ are congruent.

  • $\begingroup$ Thanks for your answer. However can you solve it without using complex numbers or coordinates ? Using elementary geometry ? $\endgroup$ – apple Oct 20 '18 at 7:32
  • $\begingroup$ @apple, there is certainly an elegant solution; however, that solution might be very elusive. I used to be obsessed with non-bashing solutions when I was IMO student. But sometimes it's just better to use complex or coordinate to brute-force the problem. $\endgroup$ – dezdichado Oct 21 '18 at 7:38
  • $\begingroup$ @dezdichado I agree with you. There are certain classes of problems that can be solved only by using brute force of complex numbers. However, I still suspect that this problem might have a simpler solution. $\endgroup$ – Oldboy Oct 21 '18 at 9:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.