Mock IMO, Problem 5
Evan Chen
June 16, 2013
Let ABC be a triangle with circumcenter O and incenter I. Points D, E,
F lie on BC, CA, AB so that BD + BE = AC and CD + CF = AB. The
circumcircles of 4BDE and 4BDF meet again at P . Prove that P O = OI.
Contents
1 Setup 1
1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Synthetic Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Length Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Computing the coordinates of P 2
2.1 Circles and Radical Axis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Cramers Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3 Computing OP 3
4 Comparing Coefficients 3
4.1 Constant Term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
4.2 Linear Term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4.3 Quadratic Term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4.4 Cubic Term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4.5 Leading Term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
5 Conclusion 5
1 Setup
1.1 Definitions
Let a = BC, b = CA, c = AB. Let R and r denote the circumradius and inradius,
respectively.
Let p = a + b + c, q = ab + bc + ca, r = abc. The pqr notation is useful for expanding
symmetric sums, since any symmetric sum is uniquely given by a pqr polynomial.
1.2 Synthetic Observations
It is easy to see that AE + AF = a. Furthermore, P (AEF ) by Miquels Theorem.
1
Evan Chen 2 Computing the coordinates of P
1.3 Length Setup
We need some way to represent the lengths. Let us set BD = k c,1 where k is some
(negative) real. Now we can repeatedly apply BD + DC = a, DC + CF = c to obtain
the following cyclic equations:
BD = k c
DC = a + c + k
CF = k a
FA = b + a + k
AE = k b
EB = c + b + k
2 Computing the coordinates of P
2.1 Circles and Radical Axis
Firstly it is straightforward to derive the equation of the three circles; it is merely
c BF y + b CE z (AEF )
a2 yz + b2 zx + c2 xy
= a CD z + c AF x (BF D)
x+y+z
b AE x + a BD y (CDE)
Being the radical center of the three circles, we see P can be defined as the point such
that the three quantities are all equal. Letting P = (au : bv : cw) to factor out the side
lengths, it follows that we need to solve
aK = BF v + CE w
bK = CD w + AF u
cK = AE u + BD v
2.2 Cramers Rule
In light of Cramers Rule, if we let K be the determinant of the coefficient matrix, then
the solution is given by
a BF CE
u = det b 0 CD
c BD 0
and its cyclic counterparts. But using row operations, this is
a+b+c b c
u = det b 0 CD
c BD 0
= bc(BD + CD) (a + b + c) BD CD
= abc + (a + b + c)(k + c)(k + (a + c))
= pk 2 + p(a + 2c)k + pc(c + a) + r
1
This can be motivated by letting t = AE + BD + CF , expressing t in terms of BD, and then BD in
terms of t. This representation then becomes natural. The choice of a k cleans up signs
2
Evan Chen 4 Comparing Coefficients
Hence, with P = (au : bv : cw), we obtain
P = (pak 2 + pa(a + 2c)k + pca(c + a) + ar : : )
Now is a convenient time to find the coordinates of P .
def
X
pak 2 + pa(a + 2c)k + pac(a + c) + ar
P =
cyc
X
= p p k 2 + p (a2 + b2 + c2 + 2ab + 2bc + 2ca)k + p a2 c + pr
sum
2 2
= p(pk + p k + (pq 3r) + r)
= p(pk 2 + p2 k + (pq 2r))
3 Computing OP
Let us return to our normalized coordinate definition. ~ ~ ~
In view
of (x,
y, z) = xA+y B+z C,
~ as the origin (so that A
if we set the circumcenter O
~ = B
~ = C
~ = R), we can derive
via the dot product that the distance to O is given by
(x + y + z)2 R2 a2 yz b2 zx c2 xy = R2 a2 yz b2 zx c2 xy.
In that sense, we see that
1 X 2
OP 2 = R2 2 a pbk 2 + pb(b + 2a) + pab(a + b) + br pck 2 + pc(c + 2b) + pbc(b + c) + cr
P cyc
abc X
= R2 2 a pk 2 + p(b + 2a) + pa(a + b) + r pk 2 + p(c + 2b) + pb(b + c) + r
P cyc
This leads us to the following claim, which we prove by comparing coefficients.
2
Claim (Main Computation). The cyclic sum equals p pk 2 + p2 k + (pq 2r) = p1 2P .
4 Comparing Coefficients
4.1 Constant Term
The right-hand side clearly has constant term p(pq 2r)2 , and
X
[k 0 ] LHS = a (r + pa(a + b)) (r + pb(b + c))
cyc
!
X X
= pr2 + pr a2 (a + b) + ab(b + c) + p2 a2 b(a + b)(a + c)
cyc cyc
!
X X
= pr2 + pr a3 + a2 b + ab2 + abc + p2 a3 b2 + a3 bc + a2 b3 + a2 b2 c
cyc cyc
= pr + pr p3 2(pq 3r) 3r + p2 (a2 b2 + b2 c2 + c2 a2 )(a + b + c) + abc(a2 + b2 + c2 )
2
= pr2 + pr(p3 2pq + 3r) + p2 (q 2 2pr)p + r(p2 2q)
= p r2 + p3 r 2prq + 3r2 + p2 q 2 2p3 r + p3 r 2pqr
= p[4r2 4pqr + p2 q 2 ]
= p(pq 2r)2
3
Evan Chen 4 Comparing Coefficients
4.2 Linear Term
Compute
X
[k 1 ] LHS = a [p(2a + b) (r + pb(b + c)) + p(2b + c)(r + pa(a + b))]
cyc
X
a pr(2a + 3b + c) + p2 (b(2a + b)(b + c) + a(2b + c)(a + b))
=
cyc
" #
X
2 2 2
= pr 2p + p ab(2a + b)(b + c) + a (2b + c)(a + b)
cyc
" #
X X X
= 2p3 r + p2 4 a2 b2 + 4 a2 bc + 2 ab3 + a3 b
cyc cyc cyc
3 2 2 2
= 2p r + p 4(q 2pr) + 4pr + 2 q(p 2q) pr
= 2p3 r + p2 4q 2 + 2p2 q 2 4q 2 2pr
= 2p4 q 2p3 r
Meanwhile, the right-hand side is merely p 2 p2 (pq 2r), so these match.
4.3 Quadratic Term
The left-hand side is
X
[k 2 ] LHS = a p (2r + p (a(a + b) + b(b + c))) + p2 (2a + b)(2b + c)
| {z } | {z }
cyc
[k2 ][k0 ]+[k0 ][k2 ] [k1 ][k1 ]
X
2pra + p2 (a3 + a2 b + ab2 + abc) + p2 (4a2 b + 2a2 c + 2ab2 + abc)
=
cyc
! !
X X
= 2pr p + p2 a3 + a2 b + ab2 + 3abc + p2 a2 b + ab2 + 3abc
4
cyc cyc
2 2 2 2
= 2p r + p p(p 2q) + 3r + p (4(pq 3r) + 3r)
= p(2pr + p 2p2 q + 3pr + 4p2 q 9pr)
4
= p(p4 + 2p2 q 4pr)
Meanwhile, the right-hand side as a quadratic term of p p(pq 2r) + (p2 )2 + (pq 2r)p =
p(p4 + 2p2 q 4pr), so the two coincide.
4.4 Cubic Term
As the degree gets ever smaller,
X
[k 3 ] LHS = a p2 (2a + b) + p2 (2b + c)
cyc
X
= p2 2a3 + 3ab + ac
cyc
4
=p
which is indeed the coefficient of the left-hand side.
4
Evan Chen 5 Conclusion
4.5 Leading Term
This is very easy, both are p3 .
5 Conclusion
Applying the claim directly, we find that
2 2abc 1 2
OP = R 2 P
P p
1
= R2 (4Rrs)
p
= R2 2Rr
= OI 2
as desired.
