SOLUTII SI COMENTARII OIM 2012

MAR DEL PLATA ARGENTINA

Abstract. Comments and solutions for IMO 2012.

Data: 15 iulie 2012.
Autor: Dan Schwarz, Bucuresti.

1. Introducere
Prezentarea solutiilor problemelor propuse la OIM 2012 Mar del Plata
(Argentina), augmentate cu comentarii. Multe dintre solutiile care urmeaza
sunt culese de pe mathlinks.ro, cu clarificarile si adaugirile de rigoare (n
masura posibilitatilor nu garantez ntotdeauna acuratetea afirmatiilor din
unele variante).
Din acest motiv, prezentarea de fata este pastrata n limba Engleza.

2. Day I
Problem 1. Given triangle ABC, the point J is the centre of the excircle
opposite the vertex A. This excircle is tangent to the side BC at M , and
to the lines AB and AC at K and L, respectively. The lines LM and BJ
meet at F , and the lines KM and CJ meet at G. Let S be the point of
intersection of the lines AF and BC, and let T be the point of intersection
of the lines AG and BC. Prove that M is the midpoint of ST.
(The excircle of ABC opposite the vertex A is the circle that is tangent
to the side BC, to the ray AB beyond B, and to the ray AC beyond C.)
Evangelos Psychas - Greece

1
Solution. It is easy to see that LF J = A and so the quadrilateral
2
AF JL is cyclic. But JLA = 90 and so AF J = 90 . Thus AB = BS,
hence M S = AK. Similarly M T = AL, but AK = AL (as tangents from a
same point), and so we are done.

For a slightly different start, let P, Q be the midpoints of M K, M L.

Clearly F P GP and GQ F Q so F P QG is cyclic. Thus F GP =
F QP = M QP = M LK (P Q k KL of course). Thus F GLK is cyclic.
1
Angle-chasing shows that this angle is actually B, which is equal to
2
F JK and so J also lies on this circle. But A is clearly the diameter of
(JKL), so the points A, F, K, J, L, G are concyclic. 
1
2 COMENTARII

Alternative Solutions. Notice that M K and M L are parallel to the internal

bisectors of B and C respectively. Then M K BJ, M L CJ, so M is the
orthocentre of JF G. Since JM BC from the tangency, and JM F G
from the orthocentre, it follows BC k F G. Now we are in a position to
prove that M F k AG and M G k AF with any of the methods above, or by
considering the excentres Ib , Ic , the fact that Ib Ic is antiparallel to BC, and
a short angle chase. With this, F M G is the medial triangle of AST and we
are done.
1
We angle chase on 4F BM and easily get BF L = A, so AF CJ
2
1
is cyclic because LAJ = A. Then AF J = 180 ALJ = 90 , so

2
AS F J KM , and AS k M G. Similarly F M k AT . Since F BJ AF S,
by symmetry F is the midpoint of AS, and similarly G is the midpoint of
AT , so F GM is the medial triangle of 4AST and M is the midpoint of ST .
You can actually prove that AM LT and AM KS are isosceles trapezia,
so SM = AK, M T = AL, AK = AL so done. Use simple trigonometry to
prove it. 
Alternative Solutions. (Computational Considerations)
[Barycentric Coordinates] Use as reference 4ABC. It is obvious
that K = ((s c) : s : 0), M = (0 : s b : s c). Also, J = (a : b : c).
In no time one gets
 
as + (s c)b
G = a : b : .
sb
It follows immediately that
 
as + (s c)
T = 0:b: = (0 : b(s b) : b(s c) as).
sb
 
b b
Normalizing, we see that T = 0, , 1 + , from which we quickly obtain
a a
M T = s. Similarly, M S = s, and we are done.

[Complex Numbers] Let the excircle be the unit circle and m = 1. Then
2k 2l
s = k+l and t = k+l . 

Alternative Solution. (Darij Grinberg) Guess Im hardly breaking any news

here, but the problem is pretty close to known facts.
First, replace excircle opposite the vertex A by incircle throughout
the problem. This doesnt change the validity of the problem (a phenomenon
called extraversion, and somewhat subtle in the cases when several excir-
cles are concerned; but in our case its very obvious: just rewrite the problem
in terms of triangle KLM instead of triangle ABC, and use the fact that
an algebraic-identity type problem that holds for any obtuse-angled triangle
 COMENTARII 3

(with obtuse angle at a specified vertex) must also hold for any acute-angled
triangle).
An Unlikely Concurrence (Chapter 3, 4, page 31 in Ross Honsberger,
Episodes in Nineteenth and Twentieth Century Euclidean Geometry) now
states that the lines BJ, M L and the perpendicular from A to BJ are
concurrent. (I have switched A and B.) In other words, AF BJ. As
a consequence, T is the reflection of A in BJ (because AF BJ and
ABF = F BT ). Combined with the fact that K is the reflection of M
in BJ, we see that AK = SM (since reflections leave lengths invariant).
Similarly, AL = T M . Thus, SM = AK = AL = T M (where AK = AL
is for obvious reasons). This is not an equality of directed lengths, but it
is easy to see (by reflection again) that S and T lie on different sides of M
along the line BC, so we get SM = M T as an equality of directed lengths,
and thus M is the midpoint of ST . 
Remarks. (Alexander Magazinov) I have a strong negative opinion on this
problem. I dislike the problems where simultaneously a) simple objects are
constructed quite complicatedly, b) the question is not to determine this
simple object, but to prove some property of it. In this case b) is really
annoying.

Problem 2. Let n 3 be an integer, and let a2 , a3 , . . . , an be positive real

numbers such that a2 a3 an = 1. Prove that
(1 + a2 )2 (1 + a3 )3 (1 + an )n > nn .
Angelo di Pasquale - Australia

1 1 ak
r
Solution. Notice that 1 + ak = + + + ak k k .
|k 1 {z k 1} (k 1)k1
k1 times
kk
Therefore (1+ak )k ak . Taking the product from k = 2 to k = n
(k 1)k1
we see that it telescopes to
n
Y
(1 + ak )k nn a2 a3 an = nn .
k=2
1
Equality holds if and only if ak = for all 2 k n, which is not
k1
n
Y 1
possible since then ak = 6= 1. 
(n 1)!
k=2
4 COMENTARII

Alternative Solutions. Actually, if we consider the functions fk : R+ R+

(1 + x)k (1 + x)k1 ((k 1)x 1)
given by fk (x) = , then fk0 (x) = , so fk
x x2
1
reaches its minimum at xk = . This proves everything, since fk (xk ) =
k1
kk
yields (by telescoping) that LHS is at least nn a2 a3 an = nn ,
(k 1)k1
1
with equality if and only if ak = for each k, when it is impossible to
k1
have a2 a3 an = 1.

(Dan Schwarz) This result allows a proof by induction. Denote E(n) =

n
Y n
Y
k n
(1 + ak ) and assume as induction hypothesis that E(n) n ak .
k=2 k=2
The starting ! case is trivial for n = 2, and!E(n + 1) = E(n)(1 + an+1 )n+1
n+1 n+1 n+1
n
Y (1 + an+1 )n+1 n
Y (n + 1)n+1 n+1
Y
n ak n ak = (n + 1) ak ,
an+1 nn
k=2 k=2 k=2
while equality cannot occur, since all equality cases are not simultaneously
compatible. 

Alternative Solution. For k 2

kk
(ak + 1)k ak
(k 1)k1
is equivalent to
 k  
k1 k1
(ak + 1) k (ak + 1) 1 + 1,
k k

which is true from the well known inequality xk k(x 1) + 1 for x 0,

which is just the Bernoully inequality (1 + (x 1))k 1 + k(x 1). Since
in our case the exponent k is a positive integer, it can be proved by simple
induction. For k = 1 it is true, being an identity; and (1 + (x 1))k+1 =
(1 + (x 1))k (1 + (x 1)) (1 + k(x 1))(1 + (x 1)) = k(x 1)2 + 1 +
(k + 1)(x 1) 1 + (k + 1)(x 1).
As it can be seen, the problem is related to the tangent line. 

Remarks. (Dan Schwarz) For the least eligible value n = 3 it is elementary

to find the true minimum, which is obtained at (a1 , a2 ) = (3/2, 2/3) and
3125
is equal to 28.93 > 27 = 33 . For n = 4 WolframAlpha offers the
108
minimum to be 359.68 > 256 = 44 . It would be interesting to know the
true asymptotic for the minimum, and I have some hope Ilya Bogdanov will
provide one in his official solution and comments.
 COMENTARII 5

Remarks (Sequel). Equivalent to find minimum of (i ln (1 + exi )) subject

P
to x2 + + xn = 0. Using Lagrange multipliers on the function
X
(i ln (1 + exi )) (x2 + ... + xn )
exi
it is easy to see that the minimum occurs when i = for each
1 + ex1
2 i n.

This implies the minimum value occurs when ai = . Its not hard
i
to check
Qthat theres a unique (via monotonicity) which makes this all
satisfy ai = 1, and the boundary cases are all trivial (unless Ive missed
something). Anyways this satisfies
Y
n1 = (i )
and we wish to minimize Q i

i
Q 
(i )i
but I cant think of any way to bound either or the product in the denom-
inator.
Here is another way of seeing that must be very close to 2 as n becomes
large.
e xi
First, if in i = we set i = 2, then we derive < 2.
1 + exi
Secondly, with < 2 the right hand side of
Y
n1 = (i )

is at least (2 ) 1 2 3 (n 2). The left hand side is at most 2n1 .

This implies (2 ) < 2n1 /(n 2)!. Since (n 2)! grows much much faster
than 2n1 , we conclude that 2 must be very close to 0.
ai
(Alexander Magazinov) As before, under the assumption that =
1 + ai i
22 33 nn
we have A = (1 + a2 )2 (1 + an )n =
(2 )2 (3 )3 (n )n
Since (2 ) (n ) = n1 , we have
n 2 n2
 
1
n
A = n (n 1) n1
2(n1) .
n 3
k 2 k2
 
(k 2)(2 )
Further, by Bernoulli, 1 1.
 k  k
So, A nn (n 1)n1 2(n1) ( 1)n2 . Or, one can try a more
1 n1
 
compact, slightly rougher estimate, namely A nn (n 1)n1 .
2
6 COMENTARII

Problem 3. The liars guessing game is a game played between two

players A and B. The rules of the game depend on two positive integers k
and n which are known to both players.
At the start of the game A chooses integers x and N with 1 x N.
Player A keeps x secret, and truthfully tells N to player B. Player B now
tries to obtain information about x by asking player A questions as follows:
each question consists of B specifying an arbitrary set S of positive integers
(possibly one specified in some previous question), and asking A whether x
belongs to S. Player B may ask as many questions as he wishes. After
each question, player A must immediately answer it with YES or NO, but
is allowed to lie as many times as she wants; the only restriction is that,
among any k + 1 consecutive answers, at least one answer must be truthful.
After B has asked as many questions as he wants, he must specify a set
X of at most n positive integers. If x belongs to X, then B wins; otherwise,
he loses. Prove that
1. If n 2k , then B can guarantee a win.
2. For all sufficiently large k, there exists an integer n 1.99k such that
B cannot guarantee a win.
David Arthur - Canada
Solution. For just part 1 (although this probably helps part 2).
First, we notice that it obviously doesnt matter what the actual elements
that are being guessed are. So well generalize the game such that A chooses
a finite set D, tells the entire set to B, and picks x D for B to guess; clearly
this game is still equivalent to the original one.
Lemma. With a fixed k and n, player B can guarantee a win for all N if
and only if B can guarantee a win for N = n + 1.
Proof. The only if part is trivial. For the if part, let us use induction
on N n + 1 (for N n just pick X = D). The starting case N = n + 1 is
now given to be winning.
For larger N > n + 1, arbitrarily partition D into n + 1 nonempty sets
E1 , E2 , . . . , En+1 (they will act as n+1 molecules made of initial elements,
acting as atoms).
B can use his strategy for n + 1 elements[on D0 := {1, 2, . . . , n + 1},
replacing each question S 0 with the set S = Ei . Then his strategy will
iS 0
yield a subset 0 0
[ X of D that has size at most n,[and B will
[ know that
xX = 0 0
Ei . Since X ( D , we have X = Ei ( Ei = D so
iX 0 iX 0 iD0
|X| < |D| = N . From here, B has a winning strategy by the induction
hypothesis. 
 COMENTARII 7

Trivially, a strategy for n = 2k is also a strategy for n 2k , so we only

need to consider n = 2k , and by the lemma we only have to consider the
case N = 2k + 1. Now all B needs to win is to know one element d D that
cannot be x.
Identify each element of D with a binary string of length k, with one
extra element e left over. Let Si be the set of elements of D corresponding
to binary strings with 0 in the i-th position.
First, B asks the questions S1 , S2 , . . . , Sk . Let ai = 0 if A answers x Si ,
and ai = 1 otherwise. Take the binary digit ai = (1 ai ). Now, let wi be
the element in D corresponding to the binary word a1 a2 . . . ai ai+1 ai+2 . . . ak
for i = 0, 1, . . . , k. B next asks the questions {w0 }, {w1 }, . . . , {wk } in order.
If A answers at least once that x 6= wi , let j be the smallest nonnegative
integer for which A answers this. Then if x = wj , A will have lied for the k+1
consecutive questions Sj+1 , Sj+2 , . . . , Sk , {w0 }, {w1 }, . . . , {wj }. Therefore B
knows that x 6= wj and he wins.
If A always answers that x = wi , then if x = e it means A will have lied
for k + 1 consecutive questions {w0 }, {w1 }, . . . , {wk }. Therefore B knows
that x 6= e and he again wins. 

3. Day II
Problem 4. Find all functions f : Z Z such that, for all integers a, b, c
that satisfy a + b + c = 0, the following equality holds
f (a)2 + f (b)2 + f (c)2 = 2f (a)f (b) + 2f (b)f (c) + 2f (c)f (a).
Liam Baker - South Africa
Solution. (Dan Schwarz) I will try to put some order and method into solving
this problem, since all other solutions Ive seen are either flawed, or else sway
allover the place.
We immediately get, like everybody else does, f (0) = 0, and also the
completely equivalent form

2
f (a) + f (b) f (a b) = 4f (a)f (b).
For b = 0 we get f (a) = f (a) (so f is an even function). Our relation,
liberally used in the sequel, writes now

2
f (a + b) f (a) f (b) = 4f (a)f (b).

Clearly, if f (1) = 0, having (f (a + 1) f (a) f (1))2 = 4f (a)f (1) = 0

yields f (a + 1) = f (a) for all a, hence f 0 (which trivially verifies).
Assume therefore f (1) 6= 0. If f (k) = k 2 f (1) for all integers k, this
clearly provides a solution, easily verified. Since f (k) = k 2 f (1) is true for
k = 1, assume then f (k) = k 2 f (1) for all 1 k m, but not for k = m + 1.
We do have (f (m + 1) f (m) f (1))2 = 4f (m)f (1), yielding (f (m + 1)
(m2 + 1)f (1))2 = 4m2 f (1)2 , whence f (m + 1) = (m2 + 1)f (1) 2mf (1) =
8 COMENTARII

(m 1)2 f (1) (the other possibility, f (m + 1) = (m2 + 1)f (1) + 2mf (1) =
(m + 1)2 f (1), is ruled out by our assumption this formula does not continue
for k = m + 1).
Now, on one hand (f (2m)f (m)f (m))2 = 4f (m)f (m) yields (f (2m)
2m2 f (1))2 = 4m4 f (1)2 , leading to
f (2m)(f (2m) 4m2 f (1)) = 0,
while on the other hand (f (2m)f (m+1)f (m1))2 = 4f (m+1)f (m1)
yields (f (2m) 2(m 1)2 f (1))2 = 4(m 1)4 f (1)2 , leading to
f (2m)(f (2m) 4(m 1)2 f (1)) = 0.
The only possibility to satisfy both relations is f (2m) = 0. This implies
(f (n + 2m) f (n) f (2m))2 = 4f (n)f (2m) = 0, forcing f (n + 2m) = f (n)
for all n Z, thus f is periodic of period length 2m. Then
f (m + k) = f ((m + k) 2m) = f ((m k)) = f (m k) = (m k)2 f (1)
for all 1 k m.
This offers a solution, both for m = 1 (when f (n) = 0 for n 0 (mod 2),
and f (n) = f (1) for n 1 (mod 2)) and for m = 2 (when f (n) = 0 for
n 0 (mod 4), f (n) = f (1) for n 1, 3 (mod 4), and f (n) = 4f (1) for
n 2 (mod 4)), as is immediately verified by what just has been proved in
the above.
Let us finally prove that for m > 2 we get a contradiction. Indeed,
then (f (m + 1) f (m 1) f (2))2 = 4f (m 1)f (2) yields (4f (1))2 =
16(m 1)2 f (1)2 , whence (m 1)2 = 1, absurd. 
Problem 5. Let ABC be a triangle with BCA = 90 , and let D be the
foot of the altitude from C. Let X be a point in the interior of the segment
CD. Let K be the point on the segment AX such that BK = BC. Similarly,
let L be the point on the segment BX such that AL = AC. Let M be the
point of intersection of AL and BK. Show that M K = M L.
Josef Pepa Tkadlec - Czech Republic
Solution. AL2 = AD AB; then ALD = LBA and a tangent to circum-
circles is obvious.
Let F be the intersection of CD and the perpendicular to AL through L;
then F LDA is cyclic and DF A = DLA = LBA.
Let T be the intersection of BX with AF ; then BF T D is cyclic, and X
is orthocentre, with AX perpendicular to BF .
In a similar way, the perpendicular for BK through K passes through F ,
and with similar triangles we can prove that LF = KF ; then the circum-
circle of LKF is tangent to LM and KM , and so M is the radical center,
thus finish. 
Alternative Solutions. M K and M L are both tangents to a circle. Let BX
meet the circle (A, AC) at J and AX meet the circle (B, BC) at I. Easily
we can find that JKLI is cyclic, and easily we can find that BK is tangent
 COMENTARII 9

to that circle (because we have BC BC = BI BJ, and we have BC = BK,

so BK BK = BL BJ). Similarly we can see that AL is tangent to the
circle too. So M K and M L are both tangents to the circle, so M K = M L.

Obviously the perpendiculars through K, L and to BK, CL intersect at

a point E which lies on CD, which is also the intersection of BDK and
CDL. So, by some angle chasing, it follows that EKD = AXE, this
meaning that EK is tangent to KDX. So, by the power of the point E, it
follows that EK 2 = ED EX = EL2 , because EL is also tangent to LDX.
Therefore EK = EL, and M KE = M LE, leading to M K = M L. 

Alternative Solution. [Analytical] (Pavel Kozlov)

Let us introduce carthesian coordinates
D(0, 0), A(a, 0), B(b, 0), X(0, d).
So the point C has coordinates (0, ab). Suppose the points L and K have
coordinates (xL , yL ) and (xK , yK ) respectively. The point L lies on the line
BX given by equation y = d x db and on the circle with the center A(a, 0)

and radius |AC| = a2 + ab given by equation (x + a)2 + y 2 = a2 + ab.
Then its easy to see that xL is positive root of the quadratic equation
(x + a)2 + (d x db )2 a2 ab = 0. Accurate calculation give us the next

b(abd2 )
relation xL = b b2 +d2
, where = (ab d2 )(a + b) is symmetrical
2 +ab

with respect to a, b. Hence yL = d xL db = d b b2 +d2
b
. Analogously

a(abd2 ) 2
xK = a a2 +d2
, yK = d a +ab
a2 +d2
a
.
So, we are already have some roots and fractions, and its quite impractical
to involve the point M in our calculations. We exclude it with help of the
sines theorem.
From the triangles AKM and BLM we get KM sin KAM
AK = sin KM A and BL =
LM
sin LBK KM AK sin KAL
sin LN B therefore LM = BL sin LBK .
To exlude angles we apply the sines theorem to the triangles AXL and
BXK: XL sin KAL XK sin LBK sin KAL
AL = sin AXB and BK = sin BXA therefore sin LBK = XK AL .
XL BK

Combining the last identities of last two subparagraphs we get KM LM =

AK XL BK AK XL BC
BL XK AL = XK BL AC .
AX xK +a a2 +ab (abd2 )((a2 +ab)
Lets calculate the first fraction: XK = xK = =
(abd2 )(a2 +ab)(abd2 )
q
a2 +ab ( a2 +ab abd2 ) a2 +ab

abd

2 ( a2 +ab abd2 )
= abd2
. In the same way we conclude XL BL =
q q
abd2 b2 +ab
b2 +ab
. Taken evident relation BC AC = a2 +ab
into account we finally get
q q q
KM a2 +ab abd2 b2 +ab
LM = abd2 b2 +ab a2 +ab
= 1.
P.S. This solution lets deduce that the foot of the angle bisector of ACB
lies on the line XM . 

Remarks. Another addition to this problem. Prove the angle bisector of

ACB and the lines XM and AB meet each other at one point.
10 COMENTARII

Problem 6. Find all positive integers n for which there exist non-negative
integers a1 , a2 , . . . , an such that
1 1 1 1 2 n
a
+ a2 + + an = a1 + a2 + + an = 1.
2 1 2 2 3 3 3
Dusan Djukic - Serbia
n n
X k X n(n + 1)
Solution. We do need a
k = 1 (mod 2), whence-
3 k 2
k=1 k=1
forth n 1 or 2 (mod 4) , and we will show all of them are in fact eligible,
by strong induction. Before the inductive part though, we shall introduce
two types of substitution, as follows
S1 : 2a1k = 2ak1+1 + 2ak1+1
k k 2k
and 3ak = 3ak +1
+ 3ak +1
.
1 1 1
S2 : 2ak = 2ak +2
+ 2ak +3
+ 2ak1+3 + 1
2ak +3
+ 1
2ak +3
+ 1
2ak +3
+ 1
2ak +3
k k 4k5
and =
3ak +
3ak +2
+
3ak +3
+ 34k3
+
ak +3 + 4k1
+
3ak +3
4k+1
3ak +3
4k+3
3ak +3
4k+5
3ak +3
.
Starting from the base cases n = 1, 5, 9, which can be easily found
1 1
n=1: 1 = 1 =1
1 1 1 1 1 1 2 3 4 5
n=5: 4 + 4 + 4 + 8 + 8 = 9 + 9 + 9 + 27 + 27 = 1
1 1 1 1 1 1 1 1 1
n=9: 4 + 8 + 8 + 8 + 8 + 16 + 16 + 16 + 16 =
1 2 3 4 5 6 7 8 9
9 + 27 + 27 + 27 + 27 + 81 + 81 + 81 + 81 = 1
we will show that
a) n = 4m + 1 7 n = 4m + 2. Simply do the first substitution once (by
taking k = 2m + 1).
b) n = 4m + 2 7 n = 4(m + 3) + 1. First we do the second substitution
(by taking k = m + 2); after that we apply the first substitution repeatedly
(by taking k = 2m + 2, 2m + 3, 2m + 4, 2m + 5, 2m + 6).
The proof is complete by induction hypothesis. 

4. Incheiere
Problema 1 este o usoara problema de geometrie sintetica nimic rau
cu asta! Problema 2 este o inegalitate cu o margine extrem de proasta ca
acuratete, si care sucumba imediat la trucul cu inegalitatea ponderata a
mediilor, sau cu metode (mai mult sau mai putin) analitice, legate de functia
(1 + x)k
f (x) = . Era mai potrivita ca problema 4. Problema 4 contine o
x
analiza delicata (dar fundamental simpla), unde dificultatea este n eleganta
si claritatea unei expuneri unde niciun caz nu este pierdut, iar toate cazurile
parazite sunt eliminate. Era mai potrivita ca problema 2. Problema 5 este
o problema de geometrie sintetica mai dificila, care a produs nsa mai multe
accidente decat se putea prevedea (n plus, exista si solutii analitice, ceea ce
o cam strica putin).
 COMENTARII 11

Problema 6 nu si merita locul. Nu o consider o problema de Teoria

Numerelor, ci de elementara Aritmetica, si orice solutie va recurge la gasirea
unor identitati (plicticoase) implicand puteri negative ale lui 2 si 3. Un
exercitiu fara nsemnatate (comparati de exemplu cu implicatiile pline de
miez ale teoremei Zeckendorf, relativa la reprezentari ca sume de numere
Fibonacci).
Problema 3 este problema Olimpiadei! probabil merita un articol separat
pentru ea nsasi (a fost de altfel aleasa de Terence Tao pentru discutie n
proiectul PolyMath). De aceea nu am insistat prea mult, si nici macar nu am
atins punctul 2), unde sunt multe ntrebari legate de cea mai buna margine
care poate fi demonstrata.