Problems Short List: With Solutions

Problems short list
with solutions
Shortlisted Problems with Solutions
55th International Mathematical Olympiad

Cape Town, South Africa, 2014
Note of Confidentiality
The shortlisted problems should be kept

strictly confidential until IMO 2015.
Contributing Countries
The Organising Committee and the Problem Selection Committee of IMO 2014 thank the
following 43 countries for contributing 141 problem proposals.
Australia, Austria, Belgium, Benin, Bulgaria, Colombia, Croatia,

Cyprus, Czech Republic, Denmark, Ecuador, Estonia, Finland,
France, Georgia, Germany, Greece, Hong Kong, Hungary, Iceland,
India, Indonesia, Iran, Ireland, Japan, Lithuania, Luxembourg,
Malaysia, Mongolia, Netherlands, Nigeria, Pakistan, Russia,
Saudi Arabia, Serbia, Slovakia, Slovenia, South Korea, Thailand,
Turkey, Ukraine, United Kingdom, U.S.A.
Problem Selection Committee
Johan Meyer
Ilya I. Bogdanov
Géza Kós
Waldemar Pompe
Christian Reiher
Stephan Wagner
4 IMO 2014 South Africa
Problems
Algebra
A1. Let z0 ă z1 ă z2 ă ¨ ¨ ¨ be an infinite sequence of positive integers. Prove that there
exists a unique integer n ě 1 such that
z0 ` z1 ` ¨ ¨ ¨ ` zn
zn ă ď zn`1 .
n
(Austria)
A2. Define the function f : p0, 1q Ñ p0, 1q by
#
1
x` 2
if x ă 21 ,
f pxq “
x2 if x ě 21 .
Let a and b be two real numbers such that 0 ă a ă b ă 1. We define the sequences an and bn
by a0 “ a, b0 “ b, and an “ f pan´1 q, bn “ f pbn´1 q for n ą 0. Show that there exists a positive
integer n such that
pan ´ an´1 qpbn ´ bn´1 q ă 0.
(Denmark)
A3. For a sequence x1 , x2 , . . . , xn of real numbers, we define its price as
max |x1 ` ¨ ¨ ¨ ` xi |.
1ďiďn
Given n real numbers, Dave and George want to arrange them into a sequence with a
low price. Diligent Dave checks all possible ways and finds the minimum possible price D.
Greedy George, on the other hand, chooses x1 such that |x1 | is as small as possible; among
the remaining numbers, he chooses x2 such that |x1 ` x2 | is as small as possible, and so on.
Thus, in the ith step he chooses xi among the remaining numbers so as to minimise the value
of |x1 ` x2 ` ¨ ¨ ¨ ` xi |. In each step, if several numbers provide the same value, George chooses
one at random. Finally he gets a sequence with price G.
Find the least possible constant c such that for every positive integer n, for every collection
of n real numbers, and for every possible sequence that George might obtain, the resulting
values satisfy the inequality G ď cD.
(Georgia)
A4. Determine all functions f : Z Ñ Z satisfying
` ˘
f f pmq ` n ` f pmq “ f pnq ` f p3mq ` 2014
for all integers m and n.
(Netherlands)
Shortlisted problems 5
A5. Consider all polynomials P pxq with real coefficients that have the following property:
for any two real numbers x and y one has
|y 2 ´ P pxq| ď 2 |x| if and only if |x2 ´ P pyq| ď 2 |y| .
Determine all possible values of P p0q.

(Belgium)
A6. Find all functions f : Z Ñ Z such that
n2 ` 4f pnq “ f pf pnqq2
for all n P Z.
(United Kingdom)
Combinatorics
C1. Let n points be given inside a rectangle R such that no two of them lie on a line parallel
to one of the sides of R. The rectangle R is to be dissected into smaller rectangles with sides
parallel to the sides of R in such a way that none of these rectangles contains any of the given
points in its interior. Prove that we have to dissect R into at least n ` 1 smaller rectangles.
(Serbia)
C2. We have 2 sheets of paper, with the number 1 written on each of them. We perform
m
the following operation. In every step we choose two distinct sheets; if the numbers on the two
sheets are a and b, then we erase these numbers and write the number a ` b on both sheets.
Prove that after m2m´1 steps, the sum of the numbers on all the sheets is at least 4m .
(Iran)
C3. Let n ě 2 be an integer. Consider an n ˆ n chessboard divided into n2 unit squares.
We call a configuration of n rooks on this board happy if every row and every column contains
exactly one rook. Find the greatest positive integer k such that for every happy configuration
of rooks, we can find a k ˆ k square without a rook on any of its k 2 unit squares.
(Croatia)
C4. Construct a tetromino by attaching two 2 ˆ 1 dominoes along their longer sides such
that the midpoint of the longer side of one domino is a corner of the other domino. This
construction yields two kinds of tetrominoes with opposite orientations. Let us call them S-
and Z-tetrominoes, respectively.
S-tetrominoes Z-tetrominoes
Assume that a lattice polygon P can be tiled with S-tetrominoes. Prove than no matter
how we tile P using only S- and Z-tetrominoes, we always use an even number of Z-tetrominoes.
(Hungary)
C5. Consider n ě 3 lines in the plane such that no two lines are parallel and no three have a
common point. These lines divide the plane into polygonal
Pa regions;
T let F be the set of regions
having finite area. Prove that it is possible to colour n{2 of the lines blue in such a way
that no region in F has a completely blue boundary. (For a real number x, rxs denotes the
least integer which is not smaller than x.)
(Austria)
C6. We are given an infinite deck of cards, each with a real number on it. For every real
number x, there is exactly one card in the deck that has x written on it. Now two players draw
disjoint sets A and B of 100 cards each from this deck. We would like to define a rule that
declares one of them a winner. This rule should satisfy the following conditions:
1. The winner only depends on the relative order of the 200 cards: if the cards are laid down
in increasing order face down and we are told which card belongs to which player, but
not what numbers are written on them, we can still decide the winner.
2. If we write the elements of both sets in increasing order as A “ ta1 , a2 , . . . , a100 u and
B “ tb1 , b2 , . . . , b100 u, and ai ą bi for all i, then A beats B.
3. If three players draw three disjoint sets A, B, C from the deck, A beats B and B beats C,
then A also beats C.
How many ways are there to define such a rule? Here, we consider two rules as different if there
exist two sets A and B such that A beats B according to one rule, but B beats A according to
the other.
(Russia)
C7. Let M be a set of n ě 4 points in the plane, no three of which are collinear. Initially these
points are connected with n segments so that each point in M is the endpoint of exactly two
segments. Then, at each step, one may choose two segments AB and CD sharing a common
interior point and replace them by the segments AC and BD if none of them is present at this
moment. Prove that it is impossible to perform n3 {4 or more such moves.
(Russia)
C8. A card deck consists of 1024 cards. On each card, a set of distinct decimal digits is
written in such a way that no two of these sets coincide (thus, one of the cards is empty). Two
players alternately take cards from the deck, one card per turn. After the deck is empty, each
player checks if he can throw out one of his cards so that each of the ten digits occurs on an
even number of his remaining cards. If one player can do this but the other one cannot, the
one who can is the winner; otherwise a draw is declared.
Determine all possible first moves of the first player after which he has a winning strategy.
(Russia)
C9. There are n circles drawn on a piece of paper in such a way that any two circles
intersect in two points, and no three circles pass through the same point. Turbo the snail slides
along the circles in the following fashion. Initially he moves on one of the circles in clockwise
direction. Turbo always keeps sliding along the current circle until he reaches an intersection
with another circle. Then he continues his journey on this new circle and also changes the
direction of moving, i.e. from clockwise to anticlockwise or vice versa.
Suppose that Turbo’s path entirely covers all circles. Prove that n must be odd.
(India)
Geometry
G1. The points P and Q are chosen on the side BC of an acute-angled triangle ABC so
that =P AB “ =ACB and =QAC “ =CBA. The points M and N are taken on the rays AP
and AQ, respectively, so that AP “ P M and AQ “ QN. Prove that the lines BM and CN
intersect on the circumcircle of the triangle ABC.
(Georgia)
G2. Let ABC be a triangle. The points K, L, and M lie on the segments BC, CA, and AB,
respectively, such that the lines AK, BL, and CM intersect in a common point. Prove that it
is possible to choose two of the triangles ALM, BMK, and CKL whose inradii sum up to at
least the inradius of the triangle ABC.
(Estonia)
G3. Let Ω and O be the circumcircle and the circumcentre of an acute-angled triangle ABC
with AB ą BC. The angle bisector of =ABC intersects Ω at M ‰ B. Let Γ be the circle
with diameter BM. The angle bisectors of =AOB and =BOC intersect Γ at points P and Q,
respectively. The point R is chosen on the line P Q so that BR “ MR. Prove that BR k AC.
(Here we always assume that an angle bisector is a ray.)
(Russia)
G4. Consider a fixed circle Γ with three fixed points A, B, and C on it. Also, let us fix
a real number λ P p0, 1q. For a variable point P R tA, B, Cu on Γ, let M be the point on
the segment CP such that CM “ λ ¨ CP . Let Q be the second point of intersection of the
circumcircles of the triangles AMP and BMC. Prove that as P varies, the point Q lies on a
fixed circle.
(United Kingdom)
G5. Let ABCD be a convex quadrilateral with =B “ =D “ 90 . Point H is the foot of
˝
the perpendicular from A to BD. The points S and T are chosen on the sides AB and AD,
respectively, in such a way that H lies inside triangle SCT and
=SHC ´ =BSC “ 90˝ , =T HC ´ =DT C “ 90˝ .
Prove that the circumcircle of triangle SHT is tangent to the line BD.
(Iran)
G6. Let ABC be a fixed acute-angled triangle. Consider some points E and F lying on
the sides AC and AB, respectively, and let M be the midpoint of EF . Let the perpendicular
bisector of EF intersect the line BC at K, and let the perpendicular bisector of MK intersect
the lines AC and AB at S and T , respectively. We call the pair pE, F q interesting, if the
quadrilateral KSAT is cyclic.
Suppose that the pairs pE1 , F1 q and pE2 , F2 q are interesting. Prove that
E1 E2 F1 F2
“ .
AB AC
(Iran)
G7. Let ABC be a triangle with circumcircle Ω and incentre I. Let the line passing through I
and perpendicular to CI intersect the segment BC and the arc BC (not containing A) of Ω at
points U and V , respectively. Let the line passing through U and parallel to AI intersect AV
at X, and let the line passing through V and parallel to AI intersect AB at Y . Let W and Z be
the midpoints of AX and BC, respectively. Prove that if the points I, X, and Y are collinear,
then the points I, W , and Z are also collinear.
(U.S.A.)
Number Theory
N1. Let n ě 2 be an integer, and let An be the set
An “ t2n ´ 2k | k P Z, 0 ď k ă nu.
Determine the largest positive integer that cannot be written as the sum of one or more (not
necessarily distinct) elements of An .
(Serbia)
N2. Determine all pairs px, yq of positive integers such that
a3
7x2 ´ 13xy ` 7y 2 “ |x ´ y| ` 1 .
(U.S.A.)
N3. A coin is called a Cape Town coin if its value is 1{n for some positive integer n. Given
a collection of Cape Town coins of total value at most 99 ` 21 , prove that it is possible to split
this collection into at most 100 groups each of total value at most 1.
(Luxembourg)
N4. Let n ą 1 be a given integer. Prove that infinitely many terms of the sequence pak qkě1 ,
defined by Z k^
n
ak “ ,
k
are odd. (For a real number x, txu denotes the largest integer not exceeding x.)
(Hong Kong)
N5. Find all triples pp, x, yq consisting of a prime number p and two positive integers x and y
such that xp´1 ` y and x ` y p´1 are both powers of p.
(Belgium)
N6. Let a1 ă a2 ă ¨ ¨ ¨ ă an be pairwise coprime positive integers with a1 being prime
and a1 ě n ` 2. On the segment I “ r0, a1 a2 ¨ ¨ ¨ an s of the real line, mark all integers that are
divisible by at least one of the numbers a1 , . . . , an . These points split I into a number of smaller
segments. Prove that the sum of the squares of the lengths of these segments is divisible by a1 .
(Serbia)
N7. Let c ě 1 be an integer. Define a sequence of positive integers by a1 “ c and
an`1 “ a3n ´ 4c ¨ a2n ` 5c2 ¨ an ` c
for all n ě 1. Prove that for each integer n ě 2 there exists a prime number p dividing an but
none of the numbers a1 , . . . , an´1 .
(Austria)
N8. For every real number x, let }x} denote the distance between x and the nearest integer.
Prove that for every pair pa, bq of positive integers there exist an odd prime p and a positive
integer k satisfying › › › › › ›
› a › › b › ›a ` b›
› ›`› ›`›
› pk › › pk › › pk › “ 1.
›
(Hungary)
Solutions
Algebra
A1. Let z0 ă z1 ă z2 ă ¨ ¨ ¨ be an infinite sequence of positive integers. Prove that there
exists a unique integer n ě 1 such that
z0 ` z1 ` ¨ ¨ ¨ ` zn
zn ă ď zn`1 . p1q
n
(Austria)
Solution. For n “ 1, 2, . . . define
dn “ pz0 ` z1 ` ¨ ¨ ¨ ` zn q ´ nzn .
The sign of dn indicates whether the first inequality in (1) holds; i.e., it is satisfied if and only
if dn ą 0.
Notice that
nzn`1 ´ pz0 ` z1 ` ¨ ¨ ¨ ` zn q “ pn ` 1qzn`1 ´ pz0 ` z1 ` ¨ ¨ ¨ ` zn ` zn`1 q “ ´dn`1 ,
so the second inequality in (1) is equivalent to dn`1 ď 0. Therefore, we have to prove that there
is a unique index n ě 1 that satisfies dn ą 0 ě dn`1 .
By its definition the sequence d1 , d2 , . . . consists of integers and we have
d1 “ pz0 ` z1 q ´ 1 ¨ z1 “ z0 ą 0.
From
` ˘ ` ˘
dn`1 ´ dn “ pz0 ` ¨ ¨ ¨ ` zn ` zn`1 q ´ pn ` 1qzn`1 ´ pz0 ` ¨ ¨ ¨ ` zn q ´ nzn “ npzn ´ zn`1 q ă 0
we can see that dn`1 ă dn and thus the sequence strictly decreases.
Hence, we have a decreasing sequence d1 ą d2 ą . . . of integers such that its first element d1
is positive. The sequence must drop below 0 at some point, and thus there is a unique index n,
that is the index of the last positive term, satisfying dn ą 0 ě dn`1 .
Comment. Omitting the assumption that z1 , z2 , . . . are integers allows the numbers dn to be all
positive. In such cases the desired n does not exist. This happens for example if zn “ 2 ´ 21n for all
integers n ě 0.
Shortlisted problems – solutions 11
A2. Define the function f : p0, 1q Ñ p0, 1q by

#
1
x` 2
if x ă 21 ,
f pxq “
x2 if x ě 21 .
Let a and b be two real numbers such that 0 ă a ă b ă 1. We define the sequences an and bn
by a0 “ a, b0 “ b, and an “ f pan´1 q, bn “ f pbn´1 q for n ą 0. Show that there exists a positive
integer n such that
pan ´ an´1 qpbn ´ bn´1 q ă 0.
(Denmark)
Solution. Note that
1
f pxq ´ x “ 2
ą0
1
if x ă 2
and
f pxq ´ x “ x2 ´ x ă 0
if x ě 21 . So if we consider p0, 1q as being divided into the two subintervals I1 “ p0, 21 q and
I2 “ r 21 , 1q, the inequality
` ˘` ˘
pan ´ an´1 qpbn ´ bn´1 q “ f pan´1 q ´ an´1 f pbn´1 q ´ bn´1 ă 0
holds if and only if an´1 and bn´1 lie in distinct subintervals.

Let us now assume, to the contrary, that ak and bk always lie in the same subinterval.
Consider the distance dk “ |ak ´ bk |. If both ak and bk lie in I1 , then
ˇ ˇ
dk`1 “ |ak`1 ´ bk`1 | “ ˇak ` 21 ´ bk ´ 21 ˇ “ dk .
If, on the other hand, ak and bk both lie in I2 , then minpak , bk q ě 21 and maxpak , bk q “
minpak , bk q ` dk ě 21 ` dk , which implies
ˇ ˇ ˇ ˇ
dk`1 “ |ak`1 ´ bk`1 | “ ˇa2k ´ b2k ˇ “ ˇpak ´ bk qpak ` bk qˇ ě |ak ´ bk | 21 ` 21 ` dk “ dk p1 ` dk q ě dk .
` ˘
This means that the difference dk is non-decreasing, and in particular dk ě d0 ą 0 for all k.
We can even say more. If ak and bk lie in I2 , then
dk`2 ě dk`1 ě dk p1 ` dk q ě dk p1 ` d0 q.
If ak and bk both lie in I1 , then ak`1 and bk`1 both lie in I2 , and so we have
dk`2 ě dk`1p1 ` dk`1 q ě dk`1p1 ` d0 q “ dk p1 ` d0 q.
In either case, dk`2 ě dk p1 ` d0 q, and inductively we get
d2m ě d0 p1 ` d0 qm .
For sufficiently large m, the right-hand side is greater than 1, but since a2m , b2m both lie in
p0, 1q, we must have d2m ă 1, a contradiction.
Thus there must be a positive integer n such that an´1 and bn´1 do not lie in the same
subinterval, which proves the desired statement.
A3. For a sequence x1 , x2 , . . . , xn of real numbers, we define its price as

max |x1 ` ¨ ¨ ¨ ` xi |.
1ďiďn
Given n real numbers, Dave and George want to arrange them into a sequence with a
low price. Diligent Dave checks all possible ways and finds the minimum possible price D.
Greedy George, on the other hand, chooses x1 such that |x1 | is as small as possible; among
the remaining numbers, he chooses x2 such that |x1 ` x2 | is as small as possible, and so on.
Thus, in the ith step he chooses xi among the remaining numbers so as to minimise the value
of |x1 ` x2 ` ¨ ¨ ¨ ` xi |. In each step, if several numbers provide the same value, George chooses
one at random. Finally he gets a sequence with price G.
Find the least possible constant c such that for every positive integer n, for every collection
of n real numbers, and for every possible sequence that George might obtain, the resulting
values satisfy the inequality G ď cD.
(Georgia)
Answer. c “ 2.
Solution. If the initial numbers are 1, ´1, 2, and ´2, then Dave may arrange them as
1, ´2, 2, ´1, while George may get the sequence 1, ´1, 2, ´2, resulting in D “ 1 and G “ 2. So
we obtain c ě 2.
Therefore, it remains to prove that G ď 2D. Let x1 , x2 , . . . , xn be the numbers Dave and
George have at their disposal. Assume that Dave and George arrange them into sequences
d1 , d2, . . . , dn and g1 , g2 , . . . , gn , respectively. Put
M “ max |xi |, S “ |x1 ` ¨ ¨ ¨ ` xn |, and N “ maxtM, Su.

1ďiďn
We claim that
D ě S, (1)
M
Dě , and (2)
2
G ď N “ maxtM, Su. (3)
These inequalities yield the desired estimate, as G ď maxtM, Su ď maxtM, 2Su ď 2D.
The inequality (1) is a direct consequence of the definition of the price.
To prove (2), consider an index i with |di | “ M. Then we have
ˇ ˇ
M “ |di | “ ˇpd1 ` ¨ ¨ ¨ ` di q ´ pd1 ` ¨ ¨ ¨ ` di´1 qˇ ď |d1 ` ¨ ¨ ¨ ` di | ` |d1 ` ¨ ¨ ¨ ` di´1 | ď 2D,
as required.
It remains to establish (3). Put hi “ g1 ` g2 ` ¨ ¨ ¨ ` gi . We will prove by induction on
i that |hi | ď N. The base case i “ 1 holds, since |h1 | “ |g1 | ď M ď N. Notice also that
|hn | “ S ď N.
For the induction step, assume that |hi´1 | ď N. We distinguish two cases.
Case 1. Assume that no two of the numbers gi , gi`1 , . . . , gn have opposite signs.
Without loss of generality, we may assume that they are all nonnegative. Then one has
hi´1 ď hi ď ¨ ¨ ¨ ď hn , thus (
|hi | ď max |hi´1 |, |hn | ď N.
Case 2. Among the numbers gi , gi`1 , . . . , gn there are positive and negative ones.
Then there exists some index j ě i such that hi´1 gj ď 0. By the definition of George’s
sequence we have
(
|hi | “ |hi´1 ` gi | ď |hi´1 ` gj | ď max |hi´1 |, |gj | ď N.
Thus, the induction step is established.
M M
Comment 1. One can establish the weaker inequalities D ě 2 and G ď D ` 2 from which the
result also follows.
Comment 2. One may ask a more specific question to find the maximal suitable c if the number n
is fixed. For n “ 1 or 2, the answer is c “ 1. For n “ 3, the answer is c “ 32 , and it is reached e.g.,
for the collection 1, 2, ´4. Finally, for n ě 4 the answer is c “ 2. In this case the arguments from the
solution above apply, and the answer is reached e.g., for the same collection 1, ´1, 2, ´2, augmented
by several zeroes.
A4. Determine all functions f : Z Ñ Z satisfying
` ˘
f f pmq ` n ` f pmq “ f pnq ` f p3mq ` 2014 (1)
for all integers m and n.

(Netherlands)
Answer. There is only one such function, namely n ÞÝÑ 2n ` 1007.

Solution. Let f be a function satisfying (1). Set C “ 1007 and define the function g : Z Ñ Z
by gpmq “ f p3mq ´ f pmq ` 2C for all m P Z; in particular, gp0q “ 2C. Now (1) rewrites as
` ˘
f f pmq ` n “ gpmq ` f pnq
for all m, n P Z. By induction in both directions it follows that

` ˘
f tf pmq ` n “ tgpmq ` f pnq (2)
` ˘ ` ˘
holds for all m, n, t P Z. Applying this, for any r P Z, to the triples r, 0, f p0q and 0, 0, f prq
in place of pm, n, tq we obtain
` ˘
f p0qgprq “ f f prqf p0q ´ f p0q “ f prqgp0q .
Now if f p0q vanished, then gp0q “ 2C ą 0 would entail that f vanishes identically, contrary
to (1). Thus f p0q ‰ 0 and the previous equation yields gprq “ α f prq, where α “ fgp0q
p0q
is some
nonzero constant.
So the definition of g reveals f p3mq “ p1 ` αqf pmq ´ 2C, i.e.,
` ˘
f p3mq ´ β “ p1 ` αq f pmq ´ β (3)
2C
for all m P Z, where β “ α
. By induction on k this implies
f p3k mq ´ β “ p1 ` αqk f pmq ´ β

` ˘
(4)
for all integers k ě 0 and m.

Since 3 ∤ 2014, there exists by (1) some value d “ f paq attained by f that is not divisible
by 3. Now by (2) we have f pn ` tdq “ f pnq ` tgpaq “ f pnq ` α ¨ tf paq, i.e.,
f pn ` tdq “ f pnq ` α ¨ td (5)
for all n, t P Z.
Let us fix any positive integer k with d | p3k ´ 1q, which is possible, since gcdp3, dq “ 1.
E.g., by the Euler–Fermat theorem, we may take k “ ϕp|d|q. Now for each m P Z we get
f p3k mq “ f pmq ` αp3k ´ 1qm

` ˘` ˘
from (5), which in view of (4) yields p1 ` αqk ´ 1 f pmq ´ β “ αp3k ´ 1qm. Since α ‰ 0,
the right hand side does not vanish for m ‰ 0, wherefore the first factor on the left hand side
cannot vanish either. It follows that
αp3k ´ 1q
f pmq “ ¨m`β.
p1 ` αqk ´ 1
So f is a linear function, say f pmq “ Am ` β for all m P Z with some constant A P Q. Plugging
this into (1) one obtains pA2 ´ 2Aqm ` pAβ ´ 2Cq “ 0 for all m, which is equivalent to the
conjunction of
A2 “ 2A and Aβ “ 2C . (6)
The first equation is equivalent to A P t0, 2u, and as C ‰ 0 the second one gives
A“2 and β “C. (7)
This shows that f is indeed the function mentioned in the answer and as the numbers found
in (7) do indeed satisfy the equations (6) this function is indeed as desired.
Comment 1. One may see that α “ 2. A more pedestrian version of the above solution starts with
a direct proof of this fact, that can be obtained by substituting some special values into (1), e.g., as
follows.
Set D “ f p0q. Plugging m “ 0 into (1) and simplifying, we get
f pn ` Dq “ f pnq ` 2C (8)
for all n P Z. In particular, for n “ 0, D, 2D we obtain f pDq “ 2C ` D, f p2Dq “ f pDq` 2C “ 4C ` D,

and f p3Dq “ f p2Dq ` 2C “ 6C ` D. So substituting m “ D and n “ r ´ D into (1) and applying (8)
with n “ r ´ D afterwards we learn
` ˘
f pr ` 2Cq ` 2C ` D “ f prq ´ 2C ` p6C ` Dq ` 2C ,
i.e., f pr ` 2Cq “ f prq ` 4C. By induction in both directions it follows that
f pn ` 2Ctq “ f pnq ` 4Ct (9)
holds for all n, t P Z.

Claim. If a and b denote two integers with the property that f pn ` aq “ f pnq ` b holds for all n P Z,
then b “ 2a.
Proof. Applying induction in both directions to the assumption we get f pn ` taq “ f pnq ` tb for all
n, t P Z. Plugging pn, tq “ p0, 2Cq into this equation and pn, tq “ p0, aq into (9) we get f p2aCq´ f p0q “
2bC “ 4aC, and, as C ‰ 0, the claim follows. l
Now by (1), for any m P Z, the numbers a “ f pmq and b “ f p3mq ´ f pmq ` 2C have the property
mentioned in the claim, whence we have
` ˘
f p3mq ´ C “ 3 f pmq ´ C .
In view of (3) this tells us indeed that α “ 2.

Now the solution may be completed as above, but due to our knowledge of α “ 2 we get the
desired formula f pmq “ 2m ` C directly without having the need to go through all linear functions.
Now it just remains to check that this function does indeed satisfy (1).
Comment 2. It is natural to wonder what happens if one replaces the number 2014 appearing in
the statement of the problem by some arbitrary integer B.
If B is odd, there is no such function, as can be seen by using the same ideas as in the above
solution.
If B ‰ 0 is even, however, then the only such function is given by n ÞÝÑ 2n`B{2. In case 3 ∤ B this
was essentially proved above, but for the general case one more idea seems to be necessary. Writing
B “ 3ν ¨ k with some integers ν and k such that 3 ∤ k one can obtain f pnq “ 2n ` B{2 for all n that
are divisible by 3ν in the same manner as usual; then one may use the formula f p3nq “ 3f pnq ´ B to
establish the remaining cases.
Finally, in case B “ 0 there are more solutions than just the function n ÞÝÑ 2n. It can be shown
that all these other functions are periodic; to mention just one kind of example, for any even integers
r and s the function #
r if n is even,
f pnq “
s if n is odd,
also has the property under discussion.
A5. Consider all polynomials P pxq with real coefficients that have the following property:
for any two real numbers x and y one has
|y 2 ´ P pxq| ď 2 |x| if and only if |x2 ´ P pyq| ď 2 |y| . (1)
Determine all possible values of P p0q.

(Belgium)
Answer. The set of possible values of P p0q is p´8, 0q Y t1u.

Solution.
Part I. We begin by verifying that these numbers are indeed possible values of P p0q. To see
that each negative real´ number¯Ć can be P p0q, it suffices to check that for every C ą 0 the
2
polynomial P pxq “ ´ 2xC ` C has the property described in the statement of the problem.
Due to symmetry it is enough for this purpose to prove |y 2 ´ P pxq| ą 2 |x| for any two real
numbers x and y. In fact we have
2 x2 p|x| ´ Cq2
2 x2
|y ´ P pxq| “ y ` ` ` 2 |x| ě ` 2 |x| ě 2 |x| ,
C C C
where in the first estimate equality can only hold if |x| “ C, whilst in the second one it can
only hold if x “ 0. As these two conditions cannot be met at the same time, we have indeed
|y 2 ´ P pxq| ą 2 |x|.
To show that P p0q “ 1 is possible as well, we verify that the polynomial P pxq “ x2 ` 1
satisfies (1). Notice that for all real numbers x and y we have
|y 2 ´ P pxq| ď 2 |x| ðñ py 2 ´ x2 ´ 1q2 ď 4x2
ðñ 0 ď py 2 ´ px ´ 1q2 px ` 1q2 ´ y 2
` ˘` ˘
ðñ 0 ď py ´ x ` 1qpy ` x ´ 1qpx ` 1 ´ yqpx ` 1 ` yq

ðñ 0 ď px ` yq2 ´ 1 1 ´ px ´ yq2 .
` ˘` ˘
Since this inequality is symmetric in x and y, we are done.

Part II. Now we show that no values other than those mentioned in the answer are possible
for P p0q. To reach this we let P denote any polynomial satisfying (1) and P p0q ě 0; as we shall
see, this implies P pxq “ x2 ` 1 for all real x, which is actually more than what we want.
First step: We prove that P is even.
By (1) we have
|y 2 ´ P pxq| ď 2 |x| ðñ |x2 ´ P pyq| ď 2 |y| ðñ |y 2 ´ P p´xq| ď 2 |x|
for all real numbers x and y. Considering just the equivalence of the first and third statement
and taking into account that y 2 may vary through Rě0 we infer that
“ ‰ “ ‰
P pxq ´ 2 |x|, P pxq ` 2 |x| X Rě0 “ P p´xq ´ 2 |x|, P p´xq ` 2 |x| X Rě0
holds for all x P R. We claim that there are infinitely many real numbers x such that
P pxq ` 2 |x| ě 0. This holds in fact for any real polynomial with P p0q ě 0; in order to see
this, we may assume that the coefficient of P appearing in front of x is nonnegative. In this
case the desired inequality holds for all sufficiently small positive real numbers.
For such numbers x satisfying P pxq ` 2 |x| ě 0 we have P pxq ` 2 |x| “ P p´xq ` 2 |x| by
the previous displayed formula, and hence also P pxq “ P p´xq. Consequently the polynomial
P pxq ´ P p´xq has infinitely many zeros, wherefore it has to vanish identically. Thus P is
indeed even.
Second step: We prove that P ptq ą 0 for all t P R.

Let us assume for a moment that there exists a real number t ‰ 0 with P ptq “ 0. Then
there is some open interval I around t such that |P pyq| ď 2 |y| holds for all y P I. Plugging
x “ 0 into (1) we learn that y 2 “ P p0q holds for all y P I, which is clearly absurd. We have
thus shown P ptq ‰ 0 for all t ‰ 0.
In combination with P p0q ě 0 this informs us that our claim could only fail if P p0q “ 0. In
this case there is by our first step a polynomial Qpxq such that P pxq “ x2 Qpxq. Applying (1)
to x “ 0 and an arbitrary y ‰ 0 we get |y Qpyq| ą 2, which is surely false when y is sufficiently
small.
Third step: We prove that P is a quadratic polynomial.
a
Notice that P cannot be constant, for otherwise if x “ P p0q and y is sufficiently large, the
first part of (1) is false whilst the second part is true. So the degree n of P has to be at least 1.
By our first step n has to be even as well, a whence in particular nˇě 2. à ˘ˇ a
Now assume that n ě 4. Plugging y “ P pxq into (1) we get ˇx2 ´ P P pxq ˇ ď 2 P pxq
and hence à a
P pxq ď x2 ` 2 P pxq
˘
P
for all real x. Choose positive real numbers x0 , a, and b such that if x P px0 , 8q, then axn ă
P pxq ă bxn ; this is indeed possible, for if d ą 0 denotes the leading coefficient of P , then
lim P pxq
n “ d, whence for instance the numbers a “ d2 and b “ 2d work provided that x0 is
xÑ8 x
chosen large enough.
Now for all sufficiently large real numbers x we have
2 à a
an{2`1 xn {2 ă aP pxqn{2 ă P P pxq ď x2 ` 2 P pxq ă xn{2 ` 2b1{2 xn{2 ,
˘
i.e.
1 ` 2b1{2
2 ńq{2
xpn ă ,
an{2`1
which is surely absurd. Thus P is indeed a quadratic polynomial.
Fourth step: We prove that P pxq “ x2 ` 1.
In the light of our first three steps there are?two real numbers a ą 0 and b such that P pxq “
ax2 ` b. Now if x is large ? enough and y “ a x, the left part of (1) holds and the right part
2 2
reads |p1 ´ a qx ´ b| ď 2 a x. In view of the fact that a ą 0 this is only possible if a “ 1.
Finally, substituting y “ x ` 1 with x ą 0 into (1) we get
|2x ` 1 ´ b| ď 2x ðñ |2x ` 1 ` b| ď 2x ` 2 ,
i.e.,
b P r1, 4x ` 1s ðñ b P r´4x ´ 3, 1s
for all x ą 0. Choosing x large enough, we can achieve that at least one of these two statements
holds; then both hold, which is only possible if b “ 1, as desired.
Comment 1. There are some issues with this problem in that its most natural solutions seem to
use some basic facts from analysis, such as the continuity of polynomials or the intermediate value
theorem. Yet these facts are intuitively obvious and implicitly clear to the students competing at this
level of difficulty, so that the Problem Selection Committee still thinks that the problem is suitable
for the IMO.
Comment 2. It seems that most solutions will in the main case, where P p0q is nonnegative, contain
an argument that is somewhat asymptotic in nature showing that P is quadratic, and some part
narrowing that case down to P pxq “ x2 ` 1.
Comment 3. It is also possible to skip the first step and start with the second step directly, but
then one has to work a bit harder to rule out the case P p0q “ 0. Let us sketch one possibility of doing
this: Take the auxiliary polynomial Qpxq such that P pxq “ xQpxq. Applying (1) to x “ 0 and an
arbitrary y ‰ 0 we get |Qpyq| ą 2. Hence we either have Qpzq ě 2 for all real z or Qpzq ď ´2 for all
real z. In particular there is some η P t´1, `1u such that P pηq ě 2 and P p´ηq ď ´2. Substituting
x “ ˘η into (1) we learn
|y 2 ´ P pηq| ď 2 ðñ |1 ´ P pyq| ď 2 |y| ðñ |y 2 ´ P p´ηq| ď 2 .

a
But for y “ P pηq the first statement is true, whilst the third one is false.
Also, if one has not obtained the evenness of P before embarking on the fourth step, one needs to
work a bit harder there, but not in a way that is likely to cause major difficulties.
Comment 4. Truly curious people may wonder about the set of all polynomials having property (1).
As explained in the solution above, P pxq “ x2 ` 1 is the only one with P p0q “ 1. On the other hand,
it is not hard to notice that for negative P p0q there are more possibilities than those mentioned above.
E.g., as remarked by the proposer, if a and b denote two positive real `numbers with ab˘ ą 1 and Q
denotes a polynomial attaining nonnegative values only, then P pxq “ ´ ax2 ` b ` Qpxq works.
More generally, it may be proved that if P pxq satisfies (1) and P p0q ă 0, then ´P pxq ą 2 |x| holds
for all x P R so that one just considers the equivalence of two false statements. One may generate all
such polynomials P by going through all combinations of a solution of the polynomial equation
x “ ApxqBpxq ` CpxqDpxq
and a real E ą 0, and setting
P pxq “ ´ Apxq2 ` Bpxq2 ` Cpxq2 ` Dpxq2 ` E

` ˘
for each of them.

A6. Find all functions f : Z Ñ Z such that

n2 ` 4f pnq “ f pf pnqq2 (1)
for all n P Z.
(United Kingdom)
Answer. The possibilities are:
• f pnq “ n ` 1 for all n;

#
n ` 1, n ą á,
• or, for some a ě 1, f pnq “
ń ` 1, n ď á;
$
&n ` 1,
’ n ą 0,
• or f pnq “ 0, n “ 0,
’
ń ` 1, n ă 0.
%
Solution 1.
Part I. Let us first check that each of the functions above really satisfies the given functional
equation. If f pnq “ n ` 1 for all n, then we have
n2 ` 4f pnq “ n2 ` 4n ` 4 “ pn ` 2q2 “ f pn ` 1q2 “ f pf pnqq2 .
If f pnq “ n ` 1 for n ą á and f pnq “ ń ` 1 otherwise, then we have the same identity for
n ą á and
n2 ` 4f pnq “ n2 ´ 4n ` 4 “ p2 ´ nq2 “ f p1 ´ nq2 “ f pf pnqq2
otherwise. The same applies to the third solution (with a “ 0), where in addition one has
02 ` 4f p0q “ 0 “ f pf p0qq2.
Part II. It remains to prove that these are really the only functions that satisfy our func-
tional equation. We do so in three steps:
Step 1: We prove that f pnq “ n ` 1 for n ą 0.
Consider the sequence pak q given by ak “ f k p1q for k ě 0. Setting n “ ak in (1), we get
a2k ` 4ak`1 “ a2k`2 .
Of course, a0 “ 1 by definition. Since a22 “ 1 ` 4a1 is odd, a2 has to be odd as well, so we set
a2 “ 2r ` 1 for some r P Z. Then a1 “ r 2 ` r and consequently
a23 “ a21 ` 4a2 “ pr 2 ` rq2 ` 8r ` 4.
Since 8r ` 4 ‰ 0, a23 ‰ pr 2 ` rq2 , so the difference between a23 and pr 2 ` rq2 is at least the
distance from pr 2 ` rq2 to the nearest even square (since 8r ` 4 and r 2 ` r are both even). This
implies that
ˇ ˇ
|8r ` 4| “ ˇa23 ´ pr 2 ` rq2 ˇ ě pr 2 ` rq2 ´ pr 2 ` r ´ 2q2 “ 4pr 2 ` r ´ 1q,
(for r “ 0 and r “ ´1, the estimate is trivial, but this does not matter). Therefore, we ave
4r 2 ď |8r ` 4| ´ 4r ` 4.
If |r| ě 4, then
4r 2 ě 16|r| ě 12|r| ` 16 ą 8|r| ` 4 ` 4|r| ` 4 ě |8r ` 4| ´ 4r ` 4,
a contradiction. Thus |r| ă 4. Checking all possible remaining values of r, we find that
pr 2 ` rq2 ` 8r ` 4 is only a square in three cases: r “ ´3, r “ 0 and r “ 1. Let us now
distinguish these three cases:
• r “ ´3, thus a1 “ 6 and a2 “ ´5. For each k ě 1, we have

b
ak`2 “ ˘ a2k ` 4ak`1,
and the sign needs to be chosen in such a way that a2k`1 ` 4ak`2 is again a square. This
yields a3 “ ´4, a4 “ ´3, a5 “ ´2, a6 “ ´1, a7 “ 0, a8 “ 1, a9 “ 2. At this point
we have reached a contradiction, since f p1q “ f pa0 q “ a1 “ 6 and at the same time
f p1q “ f pa8 q “ a9 “ 2.
• r “ 0, thus a1 “ 0 and a2 “ 1. Then a23 “ a21 ` 4a2 “ 4, so a3 “ ˘2. This, however,

is a contradiction again, since it gives us f p1q “ f pa0 q “ a1 “ 0 and at the same time
f p1q “ f pa2 q “ a3 “ ˘2.
• r “ 1, thus a1 “ 2 and a2 “ 3. We prove by induction that ak “ k ` 1 for all k ě 0

in this case, which we already know for k ď 2 now. For the induction step, assume that
ak´1 “ k and ak “ k ` 1. Then
a2k`1 “ a2k´1 ` 4ak “ k 2 ` 4k ` 4 “ pk ` 2q2 ,
so ak`1 “ ˘pk ` 2q. If ak`1 “ ´pk ` 2q, then
a2k`2 “ a2k ` 4ak`1 “ pk ` 1q2 ´ 4k ´ 8 “ k 2 ´ 2k ´ 7 “ pk ´ 1q2 ´ 8.
The latter can only be a square if k “ 4 (since 1 and 9 are the only two squares whose
difference is 8). Then, however, a4 “ 5, a5 “ ´6 and a6 “ ˘1, so
a27 “ a25 ` 4a6 “ 36 ˘ 4,
but neither 32 nor 40 is a perfect square. Thus ak`1 “ k ` 2, which completes our
induction. This also means that f pnq “ f pan´1 q “ an “ n ` 1 for all n ě 1.
Step 2: We prove that either f p0q “ 1, or f p0q “ 0 and f pnq ‰ 0 for n ‰ 0.

Set n “ 0 in (1) to get
4f p0q “ f pf p0qq2.
This means that f p0q ě 0. If f p0q “ 0, then f pnq ‰ 0 for all n ‰ 0, since we would otherwise
have
n2 “ n2 ` 4f pnq “ f pf pnqq2 “ f p0q2 “ 0.
If f p0q ą 0, then we know that f pf p0qq “ f p0q ` 1 from the first step, so
` ˘2
4f p0q “ f p0q ` 1 ,
which yields f p0q “ 1.

Step 3: We discuss the values of f pnq for n ă 0.

Lemma. For every n ě 1, we have f pńq “ ń ` 1 or f pńq “ n ` 1. Moreover, if f pńq “
ń ` 1 for some n ě 1, then also f pń ` 1q “ ń ` 2.
Proof. We prove this statement by strong induction on n. For n “ 1, we get
1 ` 4f p´1q “ f pf p´1qq2.
Thus f p´1q needs to be nonnegative. If f p´1q “ 0, then f pf p´1qq “ f p0q “ ˘1, so f p0q “ 1
(by our second step). Otherwise, we know that f pf p´1qq “ f p´1q ` 1, so
` ˘2
1 ` 4f p´1q “ f p´1q ` 1 ,
which yields f p´1q “ 2 and thus establishes the base case. For the induction step, we consider
two cases:
• If f pńq ď ń, then
f pf pńqq2 “ pńq2 ` 4f pńq ď n2 ´ 4n ă pn ´ 2q2 ,
so |f pf pńqq| ď n ´ 3 (for n “ 2, this case cannot even occur). If f pf pńqq ě 0, then we
already know from the first two steps that f pf pf pńqqq “ f pf pńqq ` 1, unless perhaps
if f p0q “ 0 and f pf pńqq “ 0. However, the latter would imply f pńq “ 0 (as shown in
Step 2) and thus n “ 0, which is impossible. If f pf pńqq ă 0, we can apply the induction
hypothesis to f pf pńqq. In either case, f pf pf pńqqq “ ˘f pf pńqq ` 1. Therefore,
˘2
f pńq2 ` 4f pf pńqq “ f pf pf pńqqq2 “ ˘f pf pńqq ` 1 ,
`
which gives us
˘2
n2 ď f pńq2 “ ˘f pf pńqq ` 1 ´ 4f pf pńqq ď f pf pńqq2 ` 6|f pf pńqq| ` 1
`
ď pn ´ 3q2 ` 6pn ´ 3q ` 1 “ n2 ´ 8,
a contradiction.
• Thus, we are left with the case that f pńq ą ń. Now we argue as in the previous
case: if f pńq ě 0, then f pf pńqq “ f pńq ` 1 by the first two steps, since f p0q “ 0
and f pńq “ 0 would imply n “ 0 (as seen in Step 2) and is thus impossible. If
f pńq ă 0, we can apply the induction hypothesis, so in any case we can infer that
f pf pńqq “ ˘f pńq ` 1. We obtain
˘2
pńq2 ` 4f pńq “ ˘f pńq ` 1 ,
`
so either ˘2
n2 “ f pńq2 ´ 2f pńq ` 1 “ f pńq ´ 1 ,
`
which gives us f pńq “ ˘n ` 1, or

˘2
n2 “ f pńq2 ´ 6f pńq ` 1 “ f pńq ´ 3 ´ 8.
`
Since 1 and 9 are the only perfect squares whose difference is 8, we must have n “ 1,
which we have already considered.
Finally, suppose that f pńq “ ń ` 1 for some n ě 2. Then
f pń ` 1q2 “ f pf pńqq2 “ pńq2 ` 4f pńq “ pn ´ 2q2 ,
so f pń` 1q “ ˘pn´ 2q. However, we already know that f pń` 1q “ ń` 2 or f pń` 1q “ n,
so f pń ` 1q “ ń ` 2. l
Combining everything we know, we find the solutions as stated in the answer:

• One solution is given by f pnq “ n ` 1 for all n.
• If f pnq is not always equal to n ` 1, then there is a largest integer m (which cannot be
positive) for which this is not the case. In view of the lemma that we proved, we must
then have f pnq “ ń ` 1 for any integer n ă m. If m “ á ă 0, we obtain f pnq “ ń ` 1
for n ď á (and f pnq “ n ` 1 otherwise). If m “ 0, we have the additional possibility
that f p0q “ 0, f pnq “ ń ` 1 for negative n and f pnq “ n ` 1 for positive n.
Solution 2. Let us provide an alternative proof for Part II, which also proceeds in several
steps.
Step 1. Let a be an arbitrary integer and b “ f paq. We first concentrate on the case where
|a| is sufficiently large.
1. If b “ 0, then (1) applied to a yields a2 “ f pf paqq2, thus
f paq “ 0 ñ a “ ˘f p0q. (2)
From now on, we set D “ |f p0q|. Throughout Step 1, we will assume that a R t´D, 0, Du,
thus b ‰ 0.
2. From (1), noticing that f pf paqq and a have the same parity, we get
ˇ ˇ ˘2
0 ‰ 4|b| “ ˇf pf paqq2 ´ a2 ˇ ě a2 ´ |a| ´ 2 “ 4|a| ´ 4.
`
Hence we have
|b| “ |f paq| ě |a| ´ 1 for a R t´D, 0, Du. (3)
For the rest of Step 1, we also assume that |a| ě E “ maxtD ` 2, 10u. Then by (3) we
have |b| ě D ` 1 and thus |f pbq| ě D.
3. Set c “ f pbq; by (3), we have |c| ě |b| ´ 1. Thus (1) yields

˘2
a2 ` 4b “ c2 ě |b| ´ 1 ,
`
which implies ˘2 ˘2 ˘2
a2 ě |b| ´ 1 ´ 4|b| “ |b| ´ 3 ´ 8 ą |b| ´ 4
` ` `
because |b| ě |a| ´ 1 ě 9. Thus (3) can be refined to
|a| ` 3 ě |f paq| ě |a| ´ 1 for |a| ě E.
Now, from c2 “ a2 ` 4b with |b| P r|a| ´ 1, |a| ` 3s we get c2 “ pa ˘ 2q2 ` d, where

d P t´16, ´12, ´8, ´4, 0, 4, 8u. Since |a ˘ 2| ě 8, this can happen only if c2 “ pa ˘ 2q2 ,
which in turn yields b “ ˘a ` 1. To summarise,
f paq “ 1 ˘ a for |a| ě E. (4)
We have shown that, with at most finitely many exceptions, f paq “ 1 ˘ a. Thus it will be
convenient for our second step to introduce the sets
( ( ` ˘
Z` “ a P Z : f paq “ a ` 1 , Z´ “ a P Z : f paq “ 1 ´ a , and Z0 “ Zz Z` Y Z´ .
Step 2. Now we investigate the structure of the sets Z` , Z´ , and Z0 .
4. Note that f pE `1q “ 1˘pE `1q. If f pE `1q “ E `2, then E `1 P Z` . Otherwise we have
f p1Èq “ É; then the original equation (1) with n “ E `1 gives us pE ´1q2 “ f pÉq2 ,
so f pÉq “ ˘pE ´ 1q. By (4) this may happen only if f pÉq “ 1 ´ E, so in this case
É P Z` . In any case we find that Z` ‰ ∅.
5. Now take any a P Z` . We claim that every integer x ě a also lies in Z` . We proceed by
induction on x, the base case x “ a being covered by our assumption. For the induction
step, assume that f px ´ 1q “ x and plug n “ x ´ 1 into (1). We get f pxq2 “ px ` 1q2 , so
either f pxq “ x ` 1 or f pxq “ ´px ` 1q.
Assume that f pxq “ ´px ` 1q and x ‰ ´1, since otherwise we already have f pxq “ x ` 1.
Plugging n “ x into (1), we obtain f p´x ´ 1q2 “ px ´ 2q2 ´ 8, which may happen only if
x´2 “ ˘3 and f p´x´1q “ ˘1. Plugging n “ ´x´1 into (1), we get f p˘1q2 “ px`1q2 ˘4,
which in turn may happen only if x ` 1 P t´2, 0, 2u.
Thus x P t´1, 5u and at the same time x P t´3, ´1, 1u, which gives us x “ ´1. Since this
has already been excluded, we must have f pxq “ x ` 1, which completes our induction.
6. Now we know that either Z` “ Z (if Z` is not bounded below), or Z` “ ta P Z : a ě a0 u,

where a0 is the smallest element of Z` . In the former case, f pnq “ n ` 1 for all n P Z,
which is our first solution. So we assume in the following that Z` is bounded below and
has a smallest element a0 .
If Z0 “ ∅, then we have f pxq “ x ` 1 for x ě a0 and f pxq “ 1 ´ x for x ă a0 . In
particular, f p0q “ 1 in any case, so 0 P Z` and thus a0 ď 0. Thus we end up with the
second solution listed in the answer. It remains to consider the case where Z0 ‰ ∅.
7. Assume that there exists some a P Z0 with b “ f paq R Z0 , so that f pbq “ 1 ˘ b. Then we
have a2 ` 4b “ p1 ˘ bq2 , so either a2 “ pb ´ 1q2 or a2 “ pb ´ 3q2 ´ 8. In the former case
we have b “ 1 ˘ a, which is impossible by our choice of a. So we get a2 “ pb ´ 3q2 ´ 8,
which implies f pbq “ 1 ´ b and |a| “ 1, |b ´ 3| “ 3.
If b “ 0, then we have f pbq “ 1, so b P Z` and therefore a0 ď 0; hence a “ ´1. But then
f paq “ 0 “ a ` 1, so a P Z` , which is impossible.
If b “ 6, then we have f p6q “ ´5, so f p´5q2 “ 16 and f p´5q P t´4, 4u. Then f pf p´5qq2 “
25 ` 4f p´5q P t9, 41u, so f p´5q “ ´4 and ´5 P Z` . This implies a0 ď ´5, which
contradicts our assumption that ˘1 “ a R Z` .
8. Thus we have shown that f pZ0 q Ď Z0 , and Z0 is finite. Take any element c P Z0 , and
consider the sequence defined by ci “ f i pcq. All elements of the sequence pci q lie in Z0 ,
hence it is bounded. Choose an index k for which |ck | is maximal, so that in particular
|ck`1 | ď |ck | and |ck`2| ď |ck |. Our functional equation (1) yields
p|ck | ´ 2q2 ´ 4 “ |ck |2 ´ 4|ck | ď c2k ` 4ck`1 “ c2k`2 .
Since ck and ck`2 have the same parity and |ck`2 | ď |ck |, this leaves us with three possi-
bilities: |ck`2 | “ |ck |, |ck`2 | “ |ck | ´ 2, and |ck | ´ 2 “ ˘2, ck`2 “ 0.
If |ck`2 | “ |ck | ´ 2, then f pck q “ ck`1 “ 1 ´ |ck |, which means that ck P Z´ or ck P Z` ,
and we reach a contradiction.
If |ck`2 | “ |ck |, then ck`1 “ 0, thus c2k`3 “ 4ck`2. So either ck`3 ‰ 0 or (by maximality
of |ck`2 | “ |ck |) ci “ 0 for all i. In the former case, we can repeat the entire argument
with ck`2 in the place of ck . Now |ck`4 | “ |ck`2 | is not possible any more since ck`3 ‰ 0,
leaving us with the only possibility |ck | ´ 2 “ |ck`2 | ´ 2 “ ˘2.
Thus we know now that either all ci are equal to 0, or |ck | “ 4. If ck “ ˘4, then either
ck`1 “ 0 and |ck`2 | “ |ck | “ 4, or ck`2 “ 0 and ck`1 “ ´4. From this point onwards, all
elements of the sequence are either 0 or ˘4.
Let cr be the last element of the sequence that is not equal to 0 or ˘4 (if such an element
exists). Then cr`1 , cr`2 P t´4, 0, 4u, so
c2r “ c2r`2 ´ 4cr`1 P t´16, 0, 16, 32u,
which gives us a contradiction. Thus all elements of the sequence are equal to 0 or ˘4,
and since the choice of c0 “ c was arbitrary, Z0 Ď t´4, 0, 4u.
9. Finally, we show that 4 R Z0 and ´4 R Z0 . Suppose that 4 P Z0 . Then in particular a0

(the smallest element of Z` ) cannot be less than 4, since this would imply 4 P Z` . So
´3 P Z´ , which means that f p´3q “ 4. Then 25 “ p´3q2 ` 4f p´3q “ f pf p´3qq2 “ f p4q2 ,
so f p4q “ ˘5 R Z0 , and we reach a contradiction.
Suppose that ´4 P Z0 . The only possible values for f p´4q that are left are 0 and ´4. Note
that 4f p0q “ f pf p0qq2 , so f p0q ě 0. If f p´4q “ 0, then we get 16 “ p´4q2 ` 0 “ f p0q2 ,
thus f p0q “ 4. But then f pf p´4qq R Z0 , which is impossible. Thus f p´4q “ ´4, which
gives us 0 “ p´4q2 ` 4f p´4q “ f pf p´4qq2 “ 16, and this is clearly absurd.
Now we are left with Z0 “ t0u and f p0q “ 0 as the only possibility. If 1 P Z´ , then
f p1q “ 0, so 1 “ 12 ` 4f p1q “ f pf p1qq2 “ f p0q2 “ 0, which is another contradiction. Thus
1 P Z` , meaning that a0 ď 1. On the other hand, a0 ď 0 would imply 0 P Z` , so we can
only have a0 “ 1. Thus Z` comprises all positive integers, and Z´ comprises all negative
integers. This gives us the third solution.
Comment. All solutions known to the Problem Selection Committee are quite lengthy and technical,
as the two solutions presented here show. It is possible to make the problem easier by imposing
additional assumptions, such as f p0q ‰ 0 or f pnq ě 1 for all n ě 0, to remove some of the technicalities.
Combinatorics
C1. Let n points be given inside a rectangle R such that no two of them lie on a line parallel
to one of the sides of R. The rectangle R is to be dissected into smaller rectangles with sides
parallel to the sides of R in such a way that none of these rectangles contains any of the given
points in its interior. Prove that we have to dissect R into at least n ` 1 smaller rectangles.
(Serbia)
Solution 1. Let k be the number of rectangles in the dissection. The set of all points that
are corners of one of the rectangles can be divided into three disjoint subsets:
• A, which consists of the four corners of the original rectangle R, each of which is the
corner of exactly one of the smaller rectangles,
• B, which contains points where exactly two of the rectangles have a common corner
(T-junctions, see the figure below),
• C, which contains points where four of the rectangles have a common corner (crossings,
see the figure below).
Figure 1: A T-junction and a crossing
We denote the number of points in B by b and the number of points in C by c. Since each
of the k rectangles has exactly four corners, we get
4k “ 4 ` 2b ` 4c.
It follows that 2b ď 4k ´ 4, so b ď 2k ´ 2.
Each of the n given points has to lie on a side of one of the smaller rectangles (but not
of the original rectangle R). If we extend this side as far as possible along borders between
rectangles, we obtain a line segment whose ends are T-junctions. Note that every point in B
can only be an endpoint of at most one such segment containing one of the given points, since
it is stated that no two of them lie on a common line parallel to the sides of R. This means
that
b ě 2n.
Combining our two inequalities for b, we get
2k ´ 2 ě b ě 2n,
thus k ě n ` 1, which is what we wanted to prove.

Solution 2. Let k denote the number of rectangles. In the following, we refer to the directions
of the sides of R as ‘horizontal’ and ‘vertical’ respectively. Our goal is to prove the inequality
k ě n ` 1 for fixed n. Equivalently, we can prove the inequality n ď k ´ 1 for each k, which
will be done by induction on k. For k “ 1, the statement is trivial.
Now assume that k ą 1. If none of the line segments that form the borders between the
rectangles is horizontal, then we have k ´ 1 vertical segments dividing R into k rectangles. On
each of them, there can only be one of the n points, so n ď k ´ 1, which is exactly what we
want to prove.
Otherwise, consider the lowest horizontal line h that contains one or more of these line
segments. Let R1 be the rectangle that results when everything that lies below h is removed
from R (see the example in the figure below).
The rectangles that lie entirely below h form blocks of rectangles separated by vertical line
segments. Suppose there are r blocks and ki rectangles in the ith block. The left and right
border of each block has to extend further upwards beyond h. Thus we can move any points
that lie on these borders upwards, so that they now lie inside R1 . This can be done without
violating the conditions, one only needs to make sure that they do not get to lie on a common
horizontal line with one of the other given points.
All other borders between rectangles in the ith block have to lie entirely below h. There are
ki ´ 1 such line segments, each of which can contain at most one of the given points. Finally,
there can be one point that lies on h. All other points have to lie in R1 (after moving some of
them as explained in the previous paragraph).
R′
Figure 2: Illustration of the inductive argument

řr
We see that R1 is divided into k ´ i“1 ki rectangles. Applying the induction hypothesis
to R1 , we find that there are at most
´ r
ÿ ¯ r
ÿ
k´ ki ´ 1 ` pki ´ 1q ` 1 “ k ´ r
i“1 i“1
points. Since r ě 1, this means that n ď k ´ 1, which completes our induction.

C2. We have 2m sheets of paper, with the number 1 written on each of them. We perform
the following operation. In every step we choose two distinct sheets; if the numbers on the two
sheets are a and b, then we erase these numbers and write the number a ` b on both sheets.
Prove that after m2m´1 steps, the sum of the numbers on all the sheets is at least 4m .
(Iran)
Solution. Let Pk be the product of the numbers on the sheets after k steps.
Suppose that in the pk ` 1qth step the numbers a and b are replaced by a` b. In the product,
the number ab is replaced by pa ` bq2 , and the other factors do not change. Since pa ` bq2 ě 4ab,
we see that Pk`1 ě 4Pk . Starting with P0 “ 1, a straightforward induction yields
Pk ě 4 k
for all integers k ě 0; in particular
m´1 m
Pm¨2m´1 ě 4m¨2 “ p2m q2 ,
so by the AM–GM inequality, the sum of the numbers written on the sheets after m2m´1 steps
is at least
m
a
2m ¨ 2 Pm¨2m´1 ě 2m ¨ 2m “ 4m .
Comment 1. It is possible to achieve the sum 4m in m2m´1 steps. For example, starting from 2m
equal numbers on the sheets, in 2m´1 consecutive steps we can double all numbers. After m such
doubling rounds we have the number 2m on every sheet.
Comment 2. There are several versions of the solution above. E.g., one may try to assign to each
positive integer n a weight wn in such a way that the sum of the weights of the numbers written on
the sheets increases, say, by at least 2 in each step. For this purpose, one needs the inequality
2wa`b ě wa ` wb ` 2 (1)
to be satisfied for all positive integers a and b.
Starting from w1 “ 1 and trying to choose the weights as small as possible, one may find that
these weights can be defined as follows: For every positive integer n, one chooses k to be the maximal
integer such that n ě 2k , and puts
n ´ n¯
wn “ k ` k “ min d ` d . (2)
2 dPZě0 2
Now, in order to prove that these weights satisfy (1), one may take arbitrary positive integers a and b,
and choose an integer d ě 0 such that wa`b “ d ` a`b 2d
. Then one has
ˆ ˙
a`b ´ a ¯ b
2wa`b “ 2d ` 2 ¨ d “ pd ´ 1q ` d´1 ` pd ´ 1q ` d´1 ` 2 ě wa ` wb ` 2.
2 2 2
Since the initial sum of the weights was 2m , after m2m´1 steps the sum is at least pm ` 1q2m . To
finish the solution, one may notice that by (2) for every positive integer a one has
a
wa ď m ` m , i.e., a ě 2m p´m ` wa q. (3)
2
So the sum of the numbers a1 , a2 , . . . , a2m on the sheets can be estimated as
2m 2m 2m
ÿ ÿ ÿ
m m m m
ai ě 2 p´m ` wai q “ ´m2 ¨ 2 ` 2 wai ě ´m4m ` pm ` 1q4m “ 4m ,
i“1 i“1 i“1
as required.
For establishing the inequalities (1) and (3), one may also use the convexity argument, instead of
the second definition of wn in (2).
One may check that log2 n ď wn ď log2 n ` 1; thus, in some rough sense, this approach is obtained
by “taking the logarithm” of the solution above.
Comment 3. An intuitive strategy to minimise the sum of numbers is that in every step we choose
the two smallest numbers. We may call this the greedy strategy. In the following paragraphs we prove
that the greedy strategy indeed provides the least possible sum of numbers.
Claim. Starting from any sequence x1 , . . . , xN of positive real numbers on N sheets, for any number
k of steps, the greedy strategy achieves the lowest possible sum of numbers.
Proof. We apply induction on k; for k “ 1 the statement is obvious. Let k ě 2, and assume that the
claim is true for smaller values.
` ˘
Every sequence of k steps can be encoded as S “ pi1 , j1 q, . . . , pik , jk q , where, for r “ 1, 2, . . . , k,
the numbers ir and jr are the indices of the two sheets that are chosen in the r th step. The resulting
final sum will be some linear combination of x1 , . . . , xN , say, c1 x1 ` ¨ ¨ ¨ ` cN xN with positive integers
c1 , . . . , cN that depend on S only. Call the numbers pc1 , . . . , cN q the characteristic vector of S.
` ˘
Choose a sequence S0 “ pi1 , j1 q, . . . , pik , jk q of steps that produces the minimal sum, starting
from x1 , . . . , xN , and let pc1 , . . . , cN q be the characteristic vector of S. We may assume that the sheets
are indexed in such an order that c1 ě c2 ě ¨ ¨ ¨ ě cN . If the sheets (and the numbers) are permuted by
a permutation π of the indices p1, 2, . . . , N q and then the same steps are performed, we can obtain the
řN
sum ct xπptq . By the rearrangement inequality, the smallest possible sum can be achieved when the
t“1
numbers px1 , . . . , xN q are in non-decreasing order. So we can assume that also x1 ď x2 ď ¨ ¨ ¨ ď xN .
Let ℓ be the largest index with c1 “ ¨ ¨ ¨ “ cℓ , and let the r th step be the first step for which cir “ c1
or cjr “ c1 . The role of ir and jr is symmetrical, so we can assume cir “ c1 and thus ir ď ℓ. We show
that cjr “ c1 and jr ď ℓ hold, too.
Before the r th step, on the ir th sheet we had the number xir . On the jr th sheet there was a linear
combination that contains the number xjr with a positive integer coefficient, and possibly some other
terms. In the r th step, the number xir joins that linear combination. From this point, each sheet
contains a linear combination of x1 , . . . , xN , with the coefficient of xjr being not smaller than the
coefficient of xir . This is preserved to the end of the procedure, so we have cjr ě cir . But cir “ c1 is
maximal among the coefficients, so we have cjr “ cir “ c1 and thus jr ď ℓ.
Either from cjr “ cir “ c1 or from the arguments in the previous paragraph we can see that none
of the ir th and the jr th sheets were used before step r. Therefore, the final linear combination of the
numbers does not change if the step pir , jr q is performed first: the sequence of steps
` ˘
S1 “ pir , jr q, pi1 , j1 q, . . . , pir´1 , jr´1 q, pir`1 , jr`1 q, . . . , piN , jN q
also produces the same minimal sum at the end. Therefore, we can replace S0 by S1 and we may
assume that r “ 1 and ci1 “ cj1 “ c1 .
As i1 ‰ j1 , we can see that ℓ ě 2 and c1 “ c2 “ ci1 “ cj1 . Let π be such a permutation of the
indices p1, 2, . . . , N q that exchanges 1, 2 with ir , jr and does not change the remaining indices. Let
` ˘
S2 “ pπpi1 q, πpj1 qq, . . . , pπpiN q, πpjN qq .
Since cπpiq “ ci for all indices i, this sequence of steps produces the same, minimal sum. Moreover, in
the first step we chose xπpi1 q “ x1 and xπpj1 q “ x2 , the two smallest numbers.
Hence, it is possible to achieve the optimal sum if we follow the greedy strategy in the first step.
By the induction hypothesis, following the greedy strategy in the remaining steps we achieve the
optimal sum.
C3. Let n ě 2 be an integer. Consider an n ˆ n chessboard divided into n2 unit squares.

We call a configuration of n rooks on this board happy if every row and every column contains
exactly one rook. Find the greatest positive integer k such that for every happy configuration
of rooks, we can find a k ˆ k square without a rook on any of its k 2 unit squares.
(Croatia)
X? \
Answer. n´1 .
Solution. Let ℓ be a positive integer. We will show that (i) if n ą ℓ2 then each happy
configuration contains an empty ℓ ˆ ℓ square, but (ii) if n ď ℓ2 then there exists a happy
configuration not containing such a square. These two statements together yield the answer.
(i). Assume that n ą ℓ2 . Consider any happy configuration. There exists a row R containing
a rook in its leftmost square. Take ℓ consecutive rows with R being one of them. Their union
U contains exactly ℓ rooks. Now remove the n ´ ℓ2 ě 1 leftmost columns from U (thus at least
one rook is also removed). The remaining part is an ℓ2 ˆ ℓ rectangle, so it can be split into ℓ
squares of size ℓ ˆ ℓ, and this part contains at most ℓ ´ 1 rooks. Thus one of these squares is
empty.
(ii). Now we assume that n ď ℓ2 . Firstly, we will construct a happy configuration with no
empty ℓ ˆ ℓ square for the case n “ ℓ2 . After that we will modify it to work for smaller values
of n.
Let us enumerate the rows from bottom to top as well as the columns from left to right
by the numbers 0, 1, . . . , ℓ2 ´ 1. Every square will be denoted, as usual, by the pair pr, cq of
its row and column numbers. Now we put the rooks on all squares of the form piℓ ` j, jℓ ` iq
with i, j “ 0, 1, . . . , ℓ ´ 1 (the picture below represents this arrangement for ℓ “ 3). Since each
number from 0 to ℓ2 ´ 1 has a unique representation of the form iℓ ` j (0 ď i, j ď ℓ ´ 1), each
row and each column contains exactly one rook.
8
r r
r
7
r
6
r r r
5
4
3
r
2
r
1
0
0 1 2 3 4 5 6 7 8
Next, we show that each ℓ ˆ ℓ square A on the board contains a rook. Consider such a
square A, and consider ℓ consecutive rows the union of which contains A. Let the lowest of
these rows have number pℓ ` q with 0 ď p, q ď ℓ ´ 1 (notice that pℓ ` q ď ℓ2 ´ ℓ). Then the
rooks in this union are placed in the columns with numbers qℓ ` p, pq ` 1qℓ ` p, . . . , pℓ ´ 1qℓ ` p,
p ` 1, ℓ ` pp ` 1q, . . . , pq ´ 1qℓ ` p ` 1, or, putting these numbers in increasing order,
p ` 1, ℓ ` pp ` 1q, . . . , pq ´ 1qℓ ` pp ` 1q, qℓ ` p, pq ` 1qℓ ` p, . . . , pℓ ´ 1qℓ ` p.
One readily checks that the first number in this list is at most ℓ ´ 1 (if p “ ℓ ´ 1, then q “ 0,
and the first listed number is qℓ ` p “ ℓ ´ 1), the last one is at least pℓ ´ 1qℓ, and the difference
between any two consecutive numbers is at most ℓ. Thus, one of the ℓ consecutive columns
intersecting A contains a number listed above, and the rook in this column is inside A, as
required. The construction for n “ ℓ2 is established.
It remains to construct a happy configuration of rooks not containing an empty ℓ ˆ ℓ square

for n ă ℓ2 . In order to achieve this, take the construction for an ℓ2 ˆ ℓ2 square described above
and remove the ℓ2 ´ n bottom rows together with the ℓ2 ´ n rightmost columns. We will have
a rook arrangement with no empty ℓ ˆ ℓ square, but several rows and columns may happen to
be empty. Clearly, the number of empty rows is equal to the number of empty columns, so one
can find a bijection between them, and put a rook on any crossing of an empty row and an
empty column corresponding to each other.
Comment. Part (i) allows several different proofs. E.g., in the last paragraph of the solution, it
suffices to deal only with the case n “ ℓ2 ` 1. Notice now that among the four corner squares, at
least one is empty. So the rooks in its row and in its column are distinct. Now, deleting this row and
column we obtain an ℓ2 ˆ ℓ2 square with ℓ2 ´ 1 rooks in it. This square can be partitioned into ℓ2
squares of size ℓ ˆ ℓ, so one of them is empty.
C4. Construct a tetromino by attaching two 2 ˆ 1 dominoes along their longer sides such
that the midpoint of the longer side of one domino is a corner of the other domino. This
construction yields two kinds of tetrominoes with opposite orientations. Let us call them S-
and Z-tetrominoes, respectively.
S-tetrominoes Z-tetrominoes
Assume that a lattice polygon P can be tiled with S-tetrominoes. Prove than no matter
how we tile P using only S- and Z-tetrominoes, we always use an even number of Z-tetrominoes.
(Hungary)
Solution 1. We may assume that polygon P is the union of some squares of an infinite
chessboard. Colour the squares of the chessboard with two colours as the figure below illustrates.
Observe that no matter how we tile P , any S-tetromino covers an even number of black
squares, whereas any Z-tetromino covers an odd number of them. As P can be tiled exclusively
by S-tetrominoes, it contains an even number of black squares. But if some S-tetrominoes and
some Z-tetrominoes cover an even number of black squares, then the number of Z-tetrominoes
must be even.
Comment. An alternative approach makes use of the following two colourings, which are perhaps
somewhat more natural:
Let s1 and s2 be the number of S-tetrominoes of the first and second type (as shown in the figure above)
respectively that are used in a tiling of P . Likewise, let z1 and z2 be the number of Z-tetrominoes of
the first and second type respectively. The first colouring shows that s1 ` z2 is invariant modulo 2, the
second colouring shows that s1 ` z1 is invariant modulo 2. Adding these two conditions, we find that
z1 ` z2 is invariant modulo 2, which is what we have to prove. Indeed, the sum of the two colourings
(regarding white as 0 and black as 1 and adding modulo 2) is the colouring shown in the solution.
Solution 2. Let us assign coordinates to the squares of the infinite chessboard in such a way
that the squares of P have nonnegative coordinates only, and that the first coordinate increases
as one moves to the right, while the second coordinate increases as one moves upwards. Write
the integer 3i ¨ p´3qj into the square with coordinates pi, jq, as in the following figure:
..
.
..
81 .
..
27 81 .
9 27 81
3 9 27 81
1 3 9 27 81
The sum of the numbers written into four squares that can be covered by an S-tetromino
is either of the form
3i ¨ p´3qj ¨ 1 ` 3 ` 3 ¨ p´3q ` 32 ¨ p´3q “ ´32 ¨ 3i ¨ p´3qj

` ˘
(for the first type of S-tetrominoes), or of the form
3i ¨ p´3qj ¨ 3 ` 3 ¨ p´3q ` p´3q ` p´3q2 “ 0,

` ˘
and thus divisible by 32. For this reason, the sum of the numbers written into the squares
of P , and thus also the sum of the numbers covered by Z-tetrominoes in the second covering,
is likewise divisible by 32. Now the sum of the entries of a Z-tetromino is either of the form
3i ¨ p´3qj ¨ 3 ` 32 ` p´3q ` 3 ¨ p´3q “ 0

` ˘
(for the first type of Z-tetrominoes), or of the form
3i ¨ p´3qj ¨ 1 ` p´3q ` 3 ¨ p´3q ` 3 ¨ p´3q2 “ 16 ¨ 3i ¨ p´3qj ,

` ˘
i.e., 16 times an odd number. Thus in order to obtain a total that is divisible by 32, an even
number of the latter kind of Z-tetrominoes needs to be used. Rotating everything by 90˝ , we
find that the number of Z-tetrominoes of the first kind is even as well. So we have even proven
slightly more than necessary.
Comment 1. In the second solution, 3 and ´3 can be replaced by other combinations as well.
For example, for any positive integer a ” 3 pmod 4q, we can write ai ¨ páqj into the square with
coordinates pi, jq and apply the same argument.
Comment 2. As the second solution shows, we even have the stronger result that the parity of the
number of each of the four types of tetrominoes in a tiling of P by S- and Z-tetrominoes is an invariant
of P . This also remains true if there is no tiling of P that uses only S-tetrominoes.
C5. Consider n ě 3 lines in the plane such that no two lines are parallel and no three have a
common point. These lines divide the plane into polygonal
Pa regions;
T let F be the set of regions
having finite area. Prove that it is possible to colour n{2 of the lines blue in such a way
that no region in F has a completely blue boundary. (For a real number x, rxs denotes the
least integer which is not smaller than x.)
(Austria)
Solution. Let L be the given set of lines. Choose a maximal (by inclusion) subset B Ď L such
that when we colour the lines Paof B Tblue, no region in F has a completely blue boundary. Let
|B| “ k. We claim that k ě n{2 .
Let us colour all the
`k˘lines of LzB red. Call a point blue if it is the intersection of two blue
lines. Then there are 2 blue points.
Now consider any red line ℓ. By the maximality of B, there exists at least one region A P F
whose only red side lies on ℓ. Since A has at least three sides, it must have at least one blue
vertex. Let us take one such vertex and associate it to ℓ.
Since each blue point belongs to four regions (some of which may be unbounded), ` ˘ it is
associated to at most four red lines. Thus the total number of red lines is at most 4 k2 . On
the other hand, this number is n ´ k, so
n ´ k ď 2kpk ´ 1q, thus n ď 2k 2 ´ k ď 2k 2 ,

Pa T
and finally k ě n{2 , which gives the desired result.
Comment 1. The constant factor in the estimate can be improved in different ways; we sketch
two of them below. On the other hand, the Problem Selection Committee is not aware of any results
?
showing that it is sometimes impossible to colour k lines satisfying the desired condition for k " n.
In this situation we find it more suitable to keep the original formulation of the problem.
Pa T
1. Firstly, we show that in the proof above one has in fact k “ |B| ě 2n{3 .
Let us make weighted associations as follows. Let a region A whose only red side lies on ℓ have
k vertices, so that k ´ 2 of them are blue. We associate each of these blue vertices to ℓ, and put the
1
weight k´2 on each such association. So the sum of the weights of all the associations is exactly n ´ k.
Now, one may check that among the four regions adjacent to a blue vertex v, at most two are trian-
gles. This means that the sum of the weights of all associations involving v is at most 1 ` 1 ` 21 ` 12 “ 3.
This leads to the estimate ˆ ˙
k
n´k ď3 ,
2
or
2n ď 3k2 ´ k ă 3k2 ,
Pa T
which yields k ě 2n{3 .
?
2. Next, we even show that k “ |B| ě r n s. For this, we specify the process of associating points
to red lines in one more different way.
Call a point red if it lies on a red line as well as on a blue line. Consider any red line ℓ, and take an
arbitrary region A P F whose only red side lies on ℓ. Let r 1 , r, b1 , . . . , bk be its vertices in clockwise
order with r 1 , r P ℓ; then the points r 1 , r are red, while all the points b1 , . . . , bk are blue. Let us
associate to ℓ the red point r and the blue point b1 . One may notice that to each pair of a red point r
and a blue point b, at most one red line can be associated, since there is at most one region A having
r and b as two clockwise consecutive vertices.
We claim now that at most two red lines are associated to each blue point b; this leads to the
desired bound ˆ ˙
k
n´k ď2 ðñ n ď k2 .
2
Assume, to the contrary, that three red lines ℓ1 , ℓ2 , and ℓ3 are associated to the same blue point b.
Let r1 , r2 , and r3 respectively be the red points associated to these lines; all these points are distinct.
The point b defines four blue rays, and each point ri is the red point closest to b on one of these rays.
So we may assume that the points r2 and r3 lie on one blue line passing through b, while r1 lies on
the other one.
ℓ1
r1
r3
A b
r2
Now consider the region A used to associate r1 and b with ℓ1 . Three of its clockwise consecutive
vertices are r1 , b, and either r2 or r3 (say, r2 ). Since A has only one red side, it can only be the
triangle r1 br2 ; but then both ℓ1 and ℓ2 pass through r2 , as well as some blue line. This is impossible
by the problem assumptions.
Comment 2. The condition that the lines be non-parallel is essentially not used in the solution, nor
in the previous comment; thus it may be omitted.
C6. We are given an infinite deck of cards, each with a real number on it. For every real
number x, there is exactly one card in the deck that has x written on it. Now two players draw
disjoint sets A and B of 100 cards each from this deck. We would like to define a rule that
declares one of them a winner. This rule should satisfy the following conditions:
1. The winner only depends on the relative order of the 200 cards: if the cards are laid down
in increasing order face down and we are told which card belongs to which player, but
not what numbers are written on them, we can still decide the winner.
2. If we write the elements of both sets in increasing order as A “ ta1 , a2 , . . . , a100 u and
B “ tb1 , b2 , . . . , b100 u, and ai ą bi for all i, then A beats B.
3. If three players draw three disjoint sets A, B, C from the deck, A beats B and B beats C,
then A also beats C.
How many ways are there to define such a rule? Here, we consider two rules as different if there
exist two sets A and B such that A beats B according to one rule, but B beats A according to
the other.
(Russia)
Answer. 100.
Solution 1. We prove a more general statement for sets of cardinality n (the problem being
the special case n “ 100, then the answer is n). In the following, we write A ą B or B ă A for
“A beats B”.
Part I. Let us first define n different rules that satisfy the conditions. To this end, fix an
index k P t1, 2, . . . , nu. We write both A and B in increasing order as A “ ta1 , a2 , . . . , an u and
B “ tb1 , b2 , . . . , bn u and say that A beats B if and only if ak ą bk . This rule clearly satisfies all
three conditions, and the rules corresponding to different k are all different. Thus there are at
least n different rules.
Part II. Now we have to prove that there is no other way to define such a rule. Suppose
that our rule satisfies the conditions, and let k P t1, 2, . . . , nu be minimal with the property
that
Ak “ t1, 2, . . . , k, n ` k ` 1, n ` k ` 2, . . . , 2nu ă Bk “ tk ` 1, k ` 2, . . . , n ` ku.
Clearly, such a k exists, since this holds for k “ n by assumption. Now consider two disjoint sets
X “ tx1 , x2 , . . . , xn u and Y “ ty1 , y2 , . . . , yn u, both in increasing order (i.e., x1 ă x2 ă ¨ ¨ ¨ ă xn
and y1 ă y2 ă ¨ ¨ ¨ ă yn ). We claim that X ă Y if (and only if – this follows automatically)
xk ă yk .
To prove this statement, pick arbitrary real numbers ui , vi , wi R X Y Y such that
u1 ă u2 ă ¨ ¨ ¨ ă uk´1 ă minpx1 , y1q, maxpxn , yn q ă vk`1 ă vk`2 ă ¨ ¨ ¨ ă vn ,
and
xk ă v1 ă v2 ă ¨ ¨ ¨ ă vk ă w1 ă w2 ă ¨ ¨ ¨ ă wn ă uk ă uk`1 ă ¨ ¨ ¨ ă un ă yk ,
and set
U “ tu1 , u2, . . . , un u, V “ tv1 , v2 , . . . , vn u, W “ tw1 , w2, . . . , wn u.
Then
• ui ă yi and xi ă vi for all i, so U ă Y and X ă V by the second condition.
• The elements of U Y W are ordered in the same way as those of Ak´1 Y Bk´1 , and since
Ak´1 ą Bk´1 by our choice of k, we also have U ą W (if k “ 1, this is trivial).
• The elements of V Y W are ordered in the same way as those of Ak Y Bk , and since
Ak ă Bk by our choice of k, we also have V ă W .
It follows that
X ă V ă W ă U ă Y,
so X ă Y by the third condition, which is what we wanted to prove.
Solution 2. Another possible approach to Part II of this problem is induction on n. For

n “ 1, there is trivially only one rule in view of the second condition.
In the following, we assume that our claim (namely, that there are no possible rules other
than those given in Part I) holds for n´1 in place of n. We start with the following observation:
Claim. At least one of the two relations
` ˘ ` ˘
t2u Y t2i ´ 1 | 2 ď i ď nu ă t1u Y t2i | 2 ď i ď nu
and ` ˘ ` ˘
t2i ´ 1 | 1 ď i ď n ´ 1u Y t2nu ă t2i | 1 ď i ď n ´ 1u Y t2n ´ 1u
holds.
Proof. Suppose that the first relation does not hold. Since our rule may only depend on the
relative order, we must also have
` ˘ ` ˘
t2u Y t3i ´ 2 | 2 ď i ď n ´ 1u Y t3n ´ 2u ą t1u Y t3i ´ 1 | 2 ď i ď n ´ 1u Y t3nu .
Likewise, if the second relation does not hold, then we must also have
` ˘ ` ˘
t1u Y t3i ´ 1 | 2 ď i ď n ´ 1u Y t3nu ą t3u Y t3i | 2 ď i ď n ´ 1u Y t3n ´ 1u .
Now condition 3 implies that

` ˘ ` ˘
t2u Y t3i ´ 2 | 2 ď i ď n ´ 1u Y t3n ´ 2u ą t3u Y t3i | 2 ď i ď n ´ 1u Y t3n ´ 1u ,
which contradicts the second condition. l

Now we distinguish two cases, depending on which of the two relations actually holds:
` ˘ ` ˘
First case: t2u Y t2i ´ 1 | 2 ď i ď nu ă t1u Y t2i | 2 ď i ď nu .
Let A “ ta1 , a2 , . . . , an u and B “ tb1 , b2 , . . . , bn u be two disjoint sets, both in increasing
order. We claim that the winner can be decided only from the values of a2 , . . . , an and b2 , . . . , bn ,
while a1 and b1 are actually irrelevant. Suppose that this was not the case, and assume without
loss of generality that a2 ă b2 . Then the relative order of a1 , a2 , . . . , an , b2 , . . . , bn is fixed, and
the position of b1 has to decide the winner. Suppose that for some value b1 “ x, B wins, while
for some other value b1 “ y, A wins.
Write Bx “ tx, b2 , . . . , bn u and By “ ty, b2 , . . . , bn u, and let ε ą 0 be smaller than half the
distance between any two of the numbers in Bx Y By Y A. For any set M, let M ˘ ε be the set
obtained by adding/subtracting ε to all elements of M. By our choice of ε, the relative order
of the elements of pBy ` εq Y A is still the same as for By Y A, while the relative order of the
elements of pBx ´ εq Y A is still the same as for Bx Y A. Thus A ă Bx ´ ε, but A ą By ` ε.
Moreover, if y ą x, then Bx ´ ε ă By ` ε by condition 2, while otherwise the relative order of
the elements in pBx ´ εq Y pBy ` εq is the same as for the two sets t2u Y t2i ´ 1 | 2 ď i ď nu
and t1u Y t2i | 2 ď i ď nu, so that Bx ´ ε ă By ` ε. In either case, we obtain
A ă Bx ´ ε ă By ` ε ă A,
which contradicts condition 3.

So we know now that the winner does not depend on a1 , b1 . Therefore, we can define a new
rule ă˚ on sets of cardinality n ´ 1 by saying that A ă˚ B if and only if A Y tau ă B Y tbu for
some a, b (or equivalently, all a, b) such that a ă min A, b ă min B and A Y tau and B Y tbu
are disjoint. The rule ă˚ satisfies all conditions again, so by the induction hypothesis, there
exists an index i such that A ă˚ B if and only if the ith smallest element of A is less than the
ith smallest element of B. This implies that C ă D if and only if the pi ` 1qth smallest element
of C is less than the pi ` 1qth smallest element of D, which completes our induction.
` ˘ ` ˘
Second case: t2i ´ 1 | 1 ď i ď n ´ 1u Y t2nu ă t2i | 1 ď i ď n ´ 1u Y t2n ´ 1u .
Set Á “ tá | a P Au for any A Ď R. For any two disjoint sets A, B Ď R of cardinality n,
we write A ă˝ B to mean p´Bq ă pÁq. It is easy to see that ă˝ defines a rule to determine
a winner that satisfies the three conditions of our problem as well as the relation of the first
case. So it follows in the same way as in the first case that for some i, A ă˝ B if and only if
the ith smallest element of A is less than the ith smallest element of B, which is equivalent to
the condition that the ith largest element of Á is greater than the ith largest element of ´B.
This proves that the original rule ă also has the desired form.
Comment. The problem asks for all possible partial orders on the set of n-element subsets of R such
that any two disjoint sets are comparable, the order relation only depends on the relative order of the
elements, and ta1 , a2 , . . . , an u ă tb1 , b2 , . . . , bn u whenever ai ă bi for all i.
As the proposer points out, one may also ask for all total orders on all n-element subsets of R
(dropping the condition of disjointness and requiring that ta1 , a2 , . . . , an u ĺ tb1 , b2 , . . . , bn u whenever
ai ď bi for all i). It turns out that the number of possibilities in this case is n!, and all possible total
orders are obtained in the following way. Fix a permutation σ P Sn . Let A “ ta1 , a2 , . . . , an u and
B “ tb1 , b2 , . . . , bn u be two subsets of R with a1 ă a2 ă ¨ ¨ ¨ ă an and b1 ă b2 ă ¨ ¨ ¨ ă bn . Then we say
that A ąσ B if and only if paσp1q , . . . , aσpnq q is lexicographically greater than pbσp1q , . . . , bσpnq q.
It seems, however, that this formulation adds rather more technicalities to the problem than
additional ideas.
This page is intentionally left blank

C7. Let M be a set of n ě 4 points in the plane, no three of which are collinear. Initially these
points are connected with n segments so that each point in M is the endpoint of exactly two
segments. Then, at each step, one may choose two segments AB and CD sharing a common
interior point and replace them by the segments AC and BD if none of them is present at this
moment. Prove that it is impossible to perform n3 {4 or more such moves.
(Russia)
Solution. A line is said to be red if it contains two points of M. As no three points of M are
collinear,
`n˘ n2 each red line determines a unique pair of points of M. Moreover, there are precisely
2
ă 2 red lines. By the value of a segment we mean the number of red lines intersecting it
in its interior, and the value of a set of segments is defined to be the sum of the values of its
elements. We will prove that piq the value of the initial set of segments is smaller than n3 {2
and that piiq each step decreases the value of the set of segments present by at least 2. Since
such a value can never be negative, these two assertions imply the statement of the problem.
To show piq we just need to observe that each segment has a value that is smaller than n2 {2.
Thus the combined value of the n initial segments is indeed below n ¨ n2 {2 “ n3 {2.
It remains to establish piiq. Suppose that at some moment we have two segments AB
and CD sharing an interior point S, and that at the next moment we have the two segments
AC and BD instead. Let XAB denote the set of red lines intersecting the segment AB in
its interior and let the sets XAC , XBD , and XCD be defined similarly. We are to prove that
|XAC | ` |XBD | ` 2 ď |XAB | ` |XCD |.
As a first step in this direction, we claim that
|XAC Y XBD | ` 2 ď |XAB Y XCD | . (1)
Indeed, if g is a red line intersecting, e.g. the segment AC in its interior, then it has to
intersect the triangle ACS once again, either in the interior of its side AS, or in the interior of
its side CS, or at S, meaning that it belongs to XAB or to XCD (see Figure 1). Moreover, the
red lines AB and CD contribute to XAB Y XCD but not to XAC Y XBD . Thereby (1) is proved.
h h
D D D
B B B
g S S S
C A C A C A
Figure 1 Figure 2 Figure 3
Similarly but more easily one obtains
|XAC X XBD | ď |XAB X XCD | . (2)
Indeed, a red line h appearing in XAC X XBD belongs, for similar reasons as above, also to
XAB X XCD . To make the argument precise, one may just distinguish the cases S P h (see
Figure 2) and S R h (see Figure 3). Thereby (2) is proved.
Adding (1) and (2) we obtain the desired conclusion, thus completing the solution of this
problem.
Comment 1. There is a problem belonging to the folklore, in the solution of which one may use the
same kind of operation:
Given n red and n green points in the plane, prove that one may draw n nonintersecting segments
each of which connects a red point with a green point.
A standard approach to this problem consists in taking n arbitrary segments connecting the red
points with the green points, and to perform the same operation as in the above proposal whenever
an intersection occurs. Now each time one performs such a step, the total length of the segments that
are present decreases due to the triangle inequality. So, as there are only finitely many possibilities
for the set of segments present, the process must end at some stage.
In the above proposal, however, considering the sum of the Euclidean lengths of the segment that
are present does not seem to help much, for even though it shows that the process must necessarily
terminate after some finite number of steps, it does not seem to easily yield any upper bound on the
number of these steps that grows polynomially with n.
One may regard the concept of the value of a segment introduced in the above solution as an
appropriately discretised version of Euclidean length suitable for obtaining such a bound.
The Problem Selection Committee still believes the problem to be sufficiently original for the
competition.
Comment 2. There are some other essentially equivalent ways of presenting the same solution. E.g.,
put M “ tA1 , A2 , . . . , An u, denote the set of segments present at any moment by te1 , e2 , . . . , en u, and
called a triple pi, j, kq of indices with i ‰ j intersecting, if the line Ai Aj intersects the segment ek . It
may then be shown that the number S of intersecting triples satisfies 0 ď S ă n3 at the beginning
and decreases by at least 4 in each step.
Comment 3. It is not difficult to construct an example where cn2 moves are possible (for some
absolute constant c ą 0). It would be interesting to say more about the gap between cn2 and cn3 .
C8. A card deck consists of 1024 cards. On each card, a set of distinct decimal digits is
written in such a way that no two of these sets coincide (thus, one of the cards is empty). Two
players alternately take cards from the deck, one card per turn. After the deck is empty, each
player checks if he can throw out one of his cards so that each of the ten digits occurs on an
even number of his remaining cards. If one player can do this but the other one cannot, the
one who can is the winner; otherwise a draw is declared.
Determine all possible first moves of the first player after which he has a winning strategy.
(Russia)
Answer. All the moves except for taking the empty card.
Solution. Let us identify each card with the set of digits written on it. For any collection of
cards C1 , C2 , . . . , Ck denote by their sum the set C1 △ C2 △ ¨ ¨ ¨ △ Ck consisting of all elements
belonging to an odd number of the Ci ’s. Denote the first and the second player by F and S,
respectively.
Since each digit is written on exactly 512 cards, the sum of all the cards is ∅. Therefore,
at the end of the game the sum of all the cards of F will be the same as that of S; denote this
sum by C. Then the player who took C can throw it out and get the desired situation, while
the other one cannot. Thus, the player getting card C wins, and no draw is possible.
Now, given a nonempty card B, one can easily see that all the cards can be split into 512
pairs of the form pX, X △Bq because pX △Bq△B “ X. The following lemma shows a property
of such a partition that is important for the solution.
Lemma. Let B ‰ ∅ be some card. Let us choose 512 cards so that exactly one card is chosen
from every pair pX, X △ Bq. Then the sum of all chosen cards is either ∅ or B.
Proof. Let b be some element of B. Enumerate the pairs; let Xi be the card not containing b
in the ith pair, and let Yi be the other card in this pair. Then the sets Xi are exactly all the
sets not containing b, therefore each digit a ‰ b is written on exactly 256 of these cards, so
X1 △ X2 △ ¨ ¨ ¨ △ X512 “ ∅. Now, if we replace some summands in this sum by the other
elements from their pairs, we will simply add B several times to this sum, thus the sum will
either remain unchanged or change by B, as required. l
Now we consider two cases.
Case 1. Assume that F takes the card ∅ on his first move. In this case, we present a
winning strategy for S.
Let S take an arbitrary card A. Assume that F takes card B after that; then S takes A △ B.
Split all 1024 cards into 512 pairs of the form pX, X △Bq; we call two cards in one pair partners.
Then the four cards taken so far form two pairs p∅, Bq and pA, A △ Bq belonging to F and S,
respectively. On each of the subsequent moves, when F takes some card, S should take the
partner of this card in response.
Consider the situation at the end of the game. Let us for a moment replace card A belonging
to S by ∅. Then he would have one card from each pair; by our lemma, the sum of all these
cards would be either ∅ or B. Now, replacing ∅ back by A we get that the actual sum of the
cards of S is either A or A △ B, and he has both these cards. Thus S wins.
Case 2. Now assume that F takes some card A ‰ ∅ on his first move. Let us present a
winning strategy for F in this case.
Assume that S takes some card B ‰ ∅ on his first move; then F takes A △ B. Again, let
us split all the cards into pairs of the form pX, X △ Bq; then the cards which have not been
taken yet form several complete pairs and one extra element (card ∅ has not been taken while
its partner B has). Now, on each of the subsequent moves, if S takes some element from a
complete pair, then F takes its partner. If S takes the extra element, then F takes an arbitrary
card Y , and the partner of Y becomes the new extra element.
Thus, on his last move S is forced to take the extra element. After that player F has cards
A and A △ B, player S has cards B and ∅, and F has exactly one element from every other
pair. Thus the situation is the same as in the previous case with roles reversed, and F wins.
Finally, if S takes ∅ on his first move then F denotes any card which has not been taken
yet by B and takes A △ B. After that, the same strategy as above is applicable.
Comment 1. If one wants to avoid the unusual question about the first move, one may change the
formulation as follows. (The difficulty of the problem would decrease somewhat.)
A card deck consists of 1023 cards; on each card, a nonempty set of distinct decimal digits is
written in such a way that no two of these sets coincide. Two players alternately take cards from
the deck, one card per turn. When the deck is empty, each player checks if he can throw out one of
his cards so that for each of the ten digits, he still holds an even number of cards with this digit. If
one player can do this but the other one cannot, the one who can is the winner; otherwise a draw is
declared.
Determine which of the players (if any) has a winning strategy.
The winner in this version is the first player. The analysis of the game from the first two paragraphs
of the previous solution applies to this version as well, except for the case C “ ∅ in which the result
is a draw. Then the strategy for S in Case 1 works for F in this version: the sum of all his cards at
the end is either A or A △ B, thus nonempty in both cases.
Comment 2. Notice that all the cards form a vector space over F2 , with △ the operation of addition.
Due to the automorphisms of this space, all possibilities for F’s first move except ∅ are equivalent.
The same holds for the response by S if F takes the card ∅ on his first move.
Comment 3. It is not that hard to show that in the initial game, F has a winning move, by the
idea of “strategy stealing”.
Namely, assume that S has a winning strategy. Let us take two card decks and start two games, in
which S will act by his strategy. In the first game, F takes an arbitrary card A1 ; assume that S takes
some B1 in response. Then F takes the card B1 at the second game; let the response by S be A2 .
Then F takes A2 in the first game and gets a response B2 , and so on.
This process stops at some moment when in the second game S takes Ai “ A1 . At this moment
the players hold the same sets of cards in both games, but with roles reversed. Now, if some cards
remain in the decks, F takes an arbitrary card from the first deck starting a similar cycle.
At the end of the game, player F’s cards in the first game are exactly player S’s cards in the second
game, and vice versa. Thus in one of the games F will win, which is impossible by our assumption.
One may notice that the strategy in Case 2 is constructed exactly in this way from the strategy
in Case 1. This is possible since every response by S wins if F takes the card ∅ on his first move.
C9. There are n circles drawn on a piece of paper in such a way that any two circles
intersect in two points, and no three circles pass through the same point. Turbo the snail slides
along the circles in the following fashion. Initially he moves on one of the circles in clockwise
direction. Turbo always keeps sliding along the current circle until he reaches an intersection
with another circle. Then he continues his journey on this new circle and also changes the
direction of moving, i.e. from clockwise to anticlockwise or vice versa.
Suppose that Turbo’s path entirely covers all circles. Prove that n must be odd.
(India)
Solution 1. Replace every cross (i.e. intersection of two circles) by two small circle arcs that
indicate the direction in which the snail should leave the cross (see Figure 1.1). Notice that
the placement of the small arcs does not depend on the direction of moving on the curves; no
matter which direction the snail is moving on the circle arcs, he will follow the same curves
(see Figure 1.2). In this way we have a set of curves, that are the possible paths of the snail.
Call these curves snail orbits or just orbits. Every snail orbit is a simple closed curve that has
no intersection with any other orbit.
anticlockwise anticlockwise
clockwise clockwise
anticlockwise anticlockwise
clockwise
clockwise
Figure 1.1 Figure 1.2

We prove the following, more general statement.
p˚q In any configuration of n circles such that no two of them are tangent, the
number of snail orbits has the same parity as the number n. (Note that it is not
assumed that all circle pairs intersect.)
This immediately solves the problem.
Let us introduce the following operation that will be called flipping a cross. At a cross,
remove the two small arcs of the orbits, and replace them by the other two arcs. Hence, when
the snail arrives at a flipped cross, he will continue on the other circle as before, but he will
preserve the orientation in which he goes along the circle arcs (see Figure 2).
a a
c d c d
b b
Figure 2
Consider what happens to the number of orbits when a cross is flipped. Denote by a, b, c,
and d the four arcs that meet at the cross such that a and b belong to the same circle. Before
the flipping a and b were connected to c and d, respectively, and after the flipping a and b are
connected to d and c, respectively.
The orbits passing through the cross are closed curves, so each of the arcs a, b, c, and d is
connected to another one by orbits outside the cross. We distinguish three cases.
Case 1: a is connected to b and c is connected to d by the orbits outside the cross (see
Figure 3.1).
We show that this case is impossible. Remove the two small arcs at the cross, connect a
to b, and connect c to d at the cross. Let γ be the new closed curve containing a and b, and
let δ be the new curve that connects c and d. These two curves intersect at the cross. So one
of c and d is inside γ and the other one is outside γ. Then the two closed curves have to meet
at least one more time, but this is a contradiction, since no orbit can intersect itself.
a a a
γ a a
c d d d
c c c d c d
?!
b δ b b b b
Figure 3.1 Figure 3.2 Figure 3.3

Case 2: a is connected to c and b is connected to d (see Figure 3.2).
Before the flipping a and c belong to one orbit and b and d belong to another orbit. Flipping
the cross merges the two orbits into a single orbit. Hence, the number of orbits decreases by 1.
Case 3: a is connected to d and b is connected to c (see Figure 3.3).
Before the flipping the arcs a, b, c, and d belong to a single orbit. Flipping the cross splits
that orbit in two. The number of orbits increases by 1.
As can be seen, every flipping decreases or increases the number of orbits by one, thus
changes its parity.
Now flip every cross, one by one. Since every pair of circles has 0 or 2 intersections, the
number of crosses is even. Therefore, when all crosses have been flipped, the original parity of
the number of orbits is restored. So it is sufficient to prove p˚q for the new configuration, where
all crosses are flipped. Of course also in this new configuration the (modified) orbits are simple
closed curves not intersecting each other.
Orient the orbits in such a way that the snail always moves anticlockwise along the circle
arcs. Figure 4 shows the same circles as in Figure 1 after flipping all crosses and adding
orientation. (Note that this orientation may be different from the orientation of the orbit as a
planar curve; the orientation of every orbit may be negative as well as positive, like the middle
orbit in Figure 4.) If the snail moves around an orbit, the total angle change in his moving
direction, the total curvature, is either `2π or ´2π, depending on the orientation of the orbit.
Let P and N be the number of orbits with positive and negative orientation, respectively. Then
the total curvature of all orbits is pP ´ Nq ¨ 2π.
change (±)
Figure 4 Figure 5
Double-count the total curvature of all orbits. Along every circle the total curvature is 2π.
At every cross, the two turnings make two changes with some angles having the same absolute
value but opposite signs, as depicted in Figure 5. So the changes in the direction at the crosses
cancel out. Hence, the total curvature is n ¨ 2π.
Now we have pP ´ Nq ¨ 2π “ n ¨ 2π, so P ´ N “ n. The number of (modified) orbits is
P ` N, that has a same parity as P ´ N “ n.
Solution 2. We present a different proof of p˚q.
We perform a sequence of small modification steps on the configuration of the circles in

such a way that at the end they have no intersection at all (see Figure 6.1). We use two kinds
of local changes to the structure of the orbits (see Figure 6.2):
• Type-1 step: An arc of a circle is moved over an arc of another circle; such a step creates
or removes two intersections.
• Type-2 step: An arc of a circle is moved through the intersection of two other circles.
Type-1 Type-2
Figure 6.1 Figure 6.2
We assume that in every step only one circle is moved, and that this circle is moved over at
most one arc or intersection point of other circles.
We will show that the parity of the number of orbits does not change in any step. As every
circle becomes a separate orbit at the end of the procedure, this fact proves p˚q.
Consider what happens to the number of orbits when a Type-1 step is performed. The two
intersection points are created or removed in a small neighbourhood. Denote some points of the
two circles where they enter or leave this neighbourhood by a, b, c, and d in this order around
the neighbourhood; let a and b belong to one circle and let c and d belong to the other circle.
The two circle arcs may have the same or opposite orientations. Moreover, the four end-points
of the two arcs are connected by the other parts of the orbits. This can happen in two ways
without intersection: either a is connected to d and b is connected to c, or a is connected to b
and c is connected to d. Altogether we have four cases, as shown in Figure 7.
d d d d a d a d
a d a d
a a a a
b b b b
c b c b c
c c c b c b c
Figure 7
We can see that the number of orbits is changed by ´2 or `2 in the leftmost case when the
arcs have the same orientation, a is connected to d, and b is connected to c. In the other three
cases the number of orbits is not changed. Hence, Type-1 steps do not change the parity of the
number of orbits.
Now consider a Type-2 step. The three circles enclose a small, triangular region; by the
step, this triangle is replaced by another triangle. Again, the modification of the orbits is done
in some small neighbourhood; the structure does not change outside. Each side of the triangle
shaped region can be convex or concave; the number of concave sides can be 0, 1, 2 or 3, so
there are 4 possible arrangements of the orbits inside the neighbourhood, as shown in Figure 8.
a f a f a f a f
b e b e b e b e
c d c d c d c d
all convex 1 concave 2 concave 3 concave

Figure 8
Denote the points where the three circles enter or leave the neighbourhood by a, b, c, d,
e, and f in this order around the neighbourhood. As can be seen in Figure 8, there are only
two essentially different cases; either a, c, e are connected to b, d, f , respectively, or a, c, e are
connected to f, b, d, respectively. The step either preserves the set of connections or switches
to the other arrangement. Obviously, in the earlier case the number of orbits is not changed;
therefore we have to consider only the latter case.
The points a, b, c, d, e, and f are connected by the orbits outside, without intersection. If
a was connected to c, say, then this orbit would isolate b, so this is impossible. Hence, each of
a, b, c, d, e and f must be connected either to one of its neighbours or to the opposite point.
If say a is connected to d, then this orbit separates b and c from e and f , therefore b must be
connected to c and e must be connected to f . Altogether there are only two cases and their
reverses: either each point is connected to one of its neighbours or two opposite points are
connected and the the remaining neigh boring pairs are connected to each other. See Figure 9.
a a a a
f f f f
b b b b
e e e e
c d c d c d c d
Figure 9
We can see that if only neighbouring points are connected, then the number of orbits is
changed by `2 or ´2. If two opposite points are connected (a and d in the figure), then the
orbits are re-arranged, but their number is unchanged. Hence, Type-2 steps also preserve the
parity. This completes the proof of p˚q.
Solution 3. Like in the previous solutions, we do not need all circle pairs to intersect but we
assume that the circles form a connected set. Denote by C and P the sets of circles and their
intersection points, respectively.
The circles divide the plane into several simply connected, bounded regions and one un-
bounded region. Denote the set of these regions by R. We say that an intersection point or
a region is odd or even if it is contained inside an odd or even number of circles, respectively.
Let Podd and Rodd be the sets of odd intersection points and odd regions, respectively.
Claim.
|Rodd | ´ |Podd | ” n pmod 2q. p1q
Proof. For each circle c P C, denote by Rc , Pc , and Xc the number of regions inside c, the
number of intersection points inside c, and the number of circles intersecting c, respectively.
The circles divide each other into several arcs; denote by Ac the number of such arcs inside c.
By double counting the regions and intersection points inside the circles we get
ÿ ÿ
|Rodd | ” Rc pmod 2q and |Podd | ” Pc pmod 2q.
cPC cPC
For each circle c, apply Euler’s polyhedron theorem to the (simply connected) regions in c.
There are 2Xc intersection points on c; they divide the circle into 2Xc arcs. The polyhedron
theorem yields pRc ` 1q ` pPc ` 2Xc q “ pAc ` 2Xc q ` 2, considering the exterior of c as a single
region. Therefore,
Rc ` Pc “ Ac ` 1. p2q
Moreover, we have four arcs starting from every interior points inside c and a single arc
starting into the interior from each intersection point on the circle. By double-counting the
end-points of the interior arcs we get 2Ac “ 4Pc ` 2Xc , so
Ac “ 2Pc ` Xc . p3q
The relations (2) and (3) together yield
Rc ´ Pc “ Xc ` 1. p4q
By summing up (4) for all circles we obtain

ÿ ÿ ÿ
Rc ´ Pc “ Xc ` |C|,
cPC cPC cPC
which yields ÿ
|Rodd | ´ |Podd | ” Xc ` n pmod 2q. p5q
cPC
ř
Notice that in Xc each intersecting circle pair is counted twice, i.e., for both circles in the
cPC
pair, so ÿ
Xc ” 0 pmod 2q,
cPC
which finishes the proof of the Claim. l
Now insert the same small arcs at the intersections as in the first solution, and suppose that
there is a single snail orbit b.
First we show that the odd regions are inside the curve b, while the even regions are outside.
Take a region r P R and a point x in its interior, and draw a ray y, starting from x, that does
not pass through any intersection point of the circles and is neither tangent to any of the circles.
As is well-known, x is inside the curve b if and only if y intersects b an odd number of times
(see Figure 10). Notice that if an arbitrary circle c contains x in its interior, then c intersects y
at a single point; otherwise, if x is outside c, then c has 2 or 0 intersections with y. Therefore,
y intersects b an odd number of times if and only if x is contained in an odd number of circles,
so if and only if r is odd.
y
x
r
Figure 10
Now consider an intersection point p of two circles c1 and c2 and a small neighbourhood
around p. Suppose that p is contained inside k circles.
We have four regions that meet at p. Let r1 be the region that lies outside both c1 and c2 ,
let r2 be the region that lies inside both c1 and c2 , and let r3 and r4 be the two remaining
regions, each lying inside exactly one of c1 and c2 . The region r1 is contained inside the same
k circles as p; the region r2 is contained also by c1 and c2 , so by k ` 2 circles in total; each of
the regions r3 and r4 is contained inside k ` 1 circles. After the small arcs have been inserted
at p, the regions r1 and r2 get connected, and the regions r3 and r4 remain separated at p (see
Figure 11). If p is an odd point, then r1 and r2 are odd, so two odd regions are connected at p.
Otherwise, if p is even, then we have two even regions connected at p.
r1
c1 c2
p
r3 r4
r2
Figure 11 Figure 12
Consider the system of odd regions and their connections at the odd points as a graph.
In this graph the odd regions are the vertices, and each odd point establishes an edge that
connects two vertices (see Figure 12). As b is a single closed curve, this graph is connected and
contains no cycle, so the graph is a tree. Then the number of vertices must be by one greater
than the number of edges, so
|Rodd | ´ |Podd | “ 1. p9q
The relations (1) and (9) together prove that n must be odd.
Comment. For every odd n there exists at least one configuration of n circles with a single snail orbit.
˝
Figure 13 shows a possible configuration with 5 circles. In general, if a circle is rotated by k ¨ 360 n
(k “ 1, 2, . . . , n ´ 1q around an interior point other than the centre, the circle and its rotated copies
together provide a single snail orbit.
Figure 13
Geometry
G1. The points P and Q are chosen on the side BC of an acute-angled triangle ABC so
that =P AB “ =ACB and =QAC “ =CBA. The points M and N are taken on the rays AP
and AQ, respectively, so that AP “ P M and AQ “ QN. Prove that the lines BM and CN
intersect on the circumcircle of the triangle ABC.
(Georgia)
Solution 1. Denote by S the intersection point of the lines BM and CN. Let moreover
β “ =QAC “ =CBA and γ “ =P AB “ =ACB. From these equalities it follows that the
triangles ABP and CAQ are similar (see Figure 1). Therefore we obtain
BP BP AQ NQ
“ “ “ .
PM PA QC QC
Moreover,
=BP M “ β ` γ “ =CQN .
Hence the triangles BP M and NQC are similar. This gives =BMP “ =NCQ, so the trian-
gles BP M and BSC are also similar. Thus we get
=CSB “ =BP M “ β ` γ “ 180˝ ´ =BAC ,
which completes the solution.
A
A
γ β
β β +γ γ B C
B C Q P
Q P
L
S
S
K
N M N X M
Figure 1 Figure 2
Solution 2. As in the previous solution, denote by S the intersection point of the lines BM
and NC. Let moreover the circumcircle of the triangle ABC intersect the lines AP and AQ
again at K and L, respectively (see Figure 2).
Note that =LBC “ =LAC “ =CBA and similarly =KCB “ =KAB “ =BCA. It implies
that the lines BL and CK meet at a point X, being symmetric to the point A with respect
to the line BC. Since AP “ P M and AQ “ QN, it follows that X lies on the line MN.
Therefore, using Pascal’s theorem for the hexagon ALBSCK, we infer that S lies on the
circumcircle of the triangle ABC, which finishes the proof.
Comment. Both solutions can be modified to obtain a more general result, with the equalities
AP “ P M and AQ “ QN
replaced by
AP QN
“ .
PM AQ
G2. Let ABC be a triangle. The points K, L, and M lie on the segments BC, CA, and AB,
respectively, such that the lines AK, BL, and CM intersect in a common point. Prove that it
is possible to choose two of the triangles ALM, BMK, and CKL whose inradii sum up to at
least the inradius of the triangle ABC.
(Estonia)
Solution. Denote
BK CL AM
a“ , b“ , c“ .
KC LA MB
By Ceva’s theorem, abc “ 1, so we may, without loss of generality, assume that a ě 1. Then at
least one of the numbers b or c is not greater than 1. Therefore at least one of the pairs pa, bq,
pb, cq has its first component not less than 1 and the second one not greater than 1. Without
loss of generality, assume that 1 ď a and b ď 1.
Therefore, we obtain bc ď 1 and 1 ď ca, or equivalently
AM LA MB BK
ď and ď .
MB CL AM KC
The first inequality implies that the line passing through M and parallel to BC intersects the
segment AL at a point X (see Figure 1). Therefore the inradius of the triangle ALM is not
less than the inradius r1 of triangle AMX.
Similarly, the line passing through M and parallel to AC intersects the segment BK at
a point Y , so the inradius of the triangle BMK is not less than the inradius r2 of the trian-
gle BMY . Thus, to complete our solution, it is enough to show that r1 ` r2 ě r, where r is
the inradius of the triangle ABC. We prove that in fact r1 ` r2 “ r.
C
L
K
X
Y
r
r1 r2
A M B
Figure 1
Since MX k BC, the dilation with centre A that takes M to B takes the incircle of the
triangle AMX to the incircle of the triangle ABC. Therefore
r1 AM r2 MB
“ , and similarly “ .
r AB r AB
Adding these equalities gives r1 ` r2 “ r, as required.
Comment. Alternatively, one can use Desargues’ theorem instead of Ceva’s theorem, as follows:
The lines AB, BC, CA dissect the plane into seven regions. One of them is bounded, and amongst
the other six, three are two-sided and three are three-sided. Now define the points P “ BC X LM ,
Q “ CA X M K, and R “ AB X KL (in the projective plane). By Desargues’ theorem, the points P ,
Q, R lie on a common line ℓ. This line intersects only unbounded regions. If we now assume (without
loss of generality) that P , Q and R lie on ℓ in that order, then one of the segments P Q or QR lies
inside a two-sided region. If, for example, this segment is P Q, then the triangles ALM and BM K
will satisfy the statement of the problem for the same reason.
G3. Let Ω and O be the circumcircle and the circumcentre of an acute-angled triangle ABC
with AB ą BC. The angle bisector of =ABC intersects Ω at M ‰ B. Let Γ be the circle
with diameter BM. The angle bisectors of =AOB and =BOC intersect Γ at points P and Q,
respectively. The point R is chosen on the line P Q so that BR “ MR. Prove that BR k AC.
(Here we always assume that an angle bisector is a ray.)
(Russia)
Solution. Let K be the midpoint of BM, i.e., the centre of Γ. Notice that AB ‰ BC implies
K ‰ O. Clearly, the lines OM and OK are the perpendicular bisectors of AC and BM,
respectively. Therefore, R is the intersection point of P Q and OK.
Let N be the second point of intersection of Γ with the line OM. Since BM is a diameter
of Γ, the lines BN and AC are both perpendicular to OM. Hence BN k AC, and it suffices to
prove that BN passes through R. Our plan for doing this is to interpret the lines BN, OK,
and P Q as the radical axes of three appropriate circles.
Let ω be the circle with diameter BO. Since =BNO “ =BKO “ 90˝ , the points N and K
lie on ω.
Next we show that the points O, K, P , and Q are concyclic. To this end, let D and E
be the midpoints of BC and AB, respectively. Clearly, D and E lie on the rays OQ and OP ,
respectively. By our assumptions about the triangle ABC, the points B, E, O, K, and D
lie in this order on ω. It follows that =EOR “ =EBK “ =KBD “ =KOD, so the line
KO externally bisects the angle P OQ. Since the point K is the centre of Γ, it also lies on
the perpendicular bisector of P Q. So K coincides with the midpoint of the arc P OQ of the
circumcircle γ of triangle P OQ.
Thus the lines OK, BN, and P Q are pairwise radical axes of the circles ω, γ, and Γ. Hence
they are concurrent at R, as required.
R N B γ
Q
E
E
D

O
ω
ω
K
A C
M
G4. Consider a fixed circle Γ with three fixed points A, B, and C on it. Also, let us fix
a real number λ P p0, 1q. For a variable point P R tA, B, Cu on Γ, let M be the point on
the segment CP such that CM “ λ ¨ CP . Let Q be the second point of intersection of the
circumcircles of the triangles AMP and BMC. Prove that as P varies, the point Q lies on a
fixed circle.
(United Kingdom)
Solution 1. Throughout the solution, we denote by >pa, bq the directed angle between the
lines a and b.
Let D be the point on the segment AB such that BD “ λ ¨ BA. We will show that either
Q “ D, or >pDQ, QBq “ >pAB, BCq; this would mean that the point Q varies over the
constant circle through D tangent to BC at B, as required.
Denote the circumcircles of the triangles AMP and BMC by ωA and ωB , respectively. The
lines AP , BC, and MQ are pairwise radical axes of the circles Γ, ωA , and ωB , thus either they
are parallel, or they share a common point X.
Assume that these lines are parallel (see Figure 1). Then the segments AP , QM, and BC
have a common perpendicular bisector; the reflection in this bisector maps the segment CP
to BA, and maps M to Q. Therefore, in this case Q lies on AB, and BQ{AB “ CM{CP “
BD{AB; so we have Q “ D.
P

ωA
ωA
A
M
Y
A P D
Q
Q ωB
D=
D =Q
Q M
M
X
B
B C C
ωB
Figure 1 Figure 2
Now assume that the lines AP , QM, and BC are concurrent at some point X (see Figure 2).
Notice that the points A, B, Q, and X lie on a common circle Ω by Miquel’s theorem
applied to the triangle XP C. Let us denote by Y the symmetric image of X about the
perpendicular bisector of AB. Clearly, Y lies on Ω, and the triangles Y AB and △XBA are
congruent. Moreover, the triangle XP C is similar to the triangle XBA, so it is also similar to
the triangle Y AB.
Next, the points D and M correspond to each other in similar triangles Y AB and XP C,
since BD{BA “ CM{CP “ λ. Moreover, the triangles Y AB and XP C are equi-oriented, so
>pMX, XP q “ >pDY, Y Aq. On the other hand, since the points A, Q, X, and Y lie on Ω, we
have >pQY, Y Aq “ >pMX, XP q. Therefore, >pQY, Y Aq “ >pDY, Y Aq, so the points Y , D,
and Q are collinear.
Finally, we have >pDQ, QBq “ >pY Q, QBq “ >pY A, ABq “ >pAB, BXq “ >pAB, BCq,
as desired.
Comment. In the original proposal, λ was supposed to be an arbitrary real number distinct from 0
ÝÝÑ ÝÝ
Ñ
and 1, and the point M was defined by CM “ λ ¨ CP . The Problem Selection Committee decided to
add the restriction λ P p0, 1q in order to avoid a large case distinction.
Solution 2. As in the previous solution, we introduce the radical centre X “ AP X BC X MQ
of the circles ωA , ωB , and Γ. Next, we also notice that the points A, Q, B, and X lie on a
common circle Ω.
If the point P lies on the arc BAC of Γ, then the point X is outside Γ, thus the point Q
belongs to the ray XM, and therefore the points P , A, and Q lie on the same side of BC.
Otherwise, if P lies on the arc BC not containing A, then X lies inside Γ, so M and Q lie on
different sides of BC; thus again Q and A lie on the same side of BC. So, in each case the
points Q and A lie on the same side of BC.
P

ωA
A
M
Q
Q ωB
B
C
Figure 3
Now we prove that the ratio

QB QB QX
“ ¨
sin =QBC QX sin =QBX
is constant. Since the points A, Q, B, and X are concyclic, we have
QX AX
“ .
sin =QBX sin =ABC
Next, since the points B, Q, M, and C are concyclic, the triangles XBQ and XMC are similar,
so
QB CM CP
“ “λ¨ .
QX CX CX
Analogously, the triangles XCP and XAB are also similar, so
CP AB
“ .
CX AX
Therefore, we obtain
QB AB AX AB
“λ¨ ¨ “λ¨ ,
sin =QBC AX sin =ABC sin =ABC
so this ratio is indeed constant. Thus the circle passing through Q and tangent to BC at B is
also constant, and Q varies over this fixed circle.
Comment. It is not hard to guess that the desired circle should be tangent to BC at B. Indeed, the
second paragraph of this solution shows that this circle lies on one side of BC; on the other hand, in
the limit case P “ B, the point Q also coincides with B.
Solution 3. Let us perform an inversion centred at C. Denote by X 1 the image of a point X
under this inversion.
The circle Γ maps to the line Γ1 passing through the constant points A1 and B 1 , and con-
taining the variable point P 1 . By the problem condition, the point M varies over the circle γ
which is the homothetic image of Γ with centre C and coefficient λ. Thus M 1 varies over the
constant line γ 1 k A1 B 1 which is the homothetic image of A1 B 1 with centre C and coefficient 1{λ,
and M “ γ 1 X CP 1 . Next, the circumcircles ωA and ωB of the triangles AMP and BMC map
to the circumcircle ωA1 of the triangle A1 M 1 P 1 and to the line B 1 M 1 , respectively; the point Q
thus maps to the second point of intersection of B 1 M 1 with ωA1 (see Figure 4).
′
ωB
ℓ
P′ A′′′′
A B′
′
′
ωA Q′′′′
Q
V

M′ J γ′
Figure 4
Let J be the (constant) common point of the lines γ 1 and CA1 , and let ℓ be the (constant)
line through J parallel to CB 1 . Let V be the common point of the lines ℓ and B 1 M 1 . Applying
Pappus’ theorem to the triples pC, J, A1 q and pV, B 1 , M 1 q we get that the points CB 1 X JV ,
JM 1 X A1 B 1 , and CM 1 X A1 V are collinear. The first two of these points are ideal, hence so is
the third, which means that CM 1 k A1 V .
Now we have >pQ1 A1 , A1 P 1 q “ >pQ1 M 1 , M 1 P 1q “ =pV M 1 , A1 V q, which means that the
triangles B 1 Q1 A1 and B 1 A1 V are similar, and pB 1 A1 q2 “ B 1 Q1 ¨ B 1 V . Thus Q1 is the image of V
under the second (fixed) inversion with centre B 1 and radius B 1 A1 . Since V varies over the
constant line ℓ, Q1 varies over some constant circle Θ. Thus, applying the first inversion back
we get that Q also varies over some fixed circle.
One should notice that this last circle is not a line; otherwise Θ would contain C, and thus
ℓ would contain the image of C under the second inversion. This is impossible, since CB 1 k ℓ.
G5. Let ABCD be a convex quadrilateral with =B “ =D “ 90˝ . Point H is the foot of
the perpendicular from A to BD. The points S and T are chosen on the sides AB and AD,
respectively, in such a way that H lies inside triangle SCT and
=SHC ´ =BSC “ 90˝ , =T HC ´ =DT C “ 90˝ .
Prove that the circumcircle of triangle SHT is tangent to the line BD.
(Iran)
Solution. Let the line passing through C and perpendicular to the line SC intersect the line AB
at Q (see Figure 1). Then
=SQC “ 90˝ ´ =BSC “ 180˝ ´ =SHC ,
which implies that the points C, H, S, and Q lie on a common circle. Moreover, since SQ is a
diameter of this circle, we infer that the circumcentre K of triangle SHC lies on the line AB.
Similarly, we prove that the circumcentre L of triangle CHT lies on the line AD.
A
D
K
H
H
B
C
L
Q
Figure 1
In order to prove that the circumcircle of triangle SHT is tangent to BD, it suffices to show
that the perpendicular bisectors of HS and HT intersect on the line AH. However, these two
perpendicular bisectors coincide with the angle bisectors of angles AKH and ALH. Therefore,
in order to complete the solution, it is enough (by the bisector theorem) to show that
AK AL
“ . p1q
KH LH
We present two proofs of this equality.
First proof. Let the lines KL and HC intersect at M (see Figure 2). Since KH “ KC
and LH “ LC, the points H and C are symmetric to each other with respect to the line KL.
Therefore M is the midpoint of HC. Denote by O the circumcentre of quadrilateral ABCD.
Then O is the midpoint of AC. Therefore we have OM k AH and hence OM K BD. This
together with the equality OB “ OD implies that OM is the perpendicular bisector of BD
and therefore BM “ DM.
Since CM K KL, the points B, C, M, and K lie on a common circle with diameter KC.
Similarly, the points L, C, M, and D lie on a circle with diameter LC. Thus, using the sine
law, we obtain
AK sin =ALK DM CK CK KH
“ “ ¨ “ “ ,
AL sin =AKL CL BM CL LH
which finishes the proof of p1q.

A
ω
A
D
K
H
H D
B K
M B H
N
C
L C
L
Figure 2 Figure 3
Second proof. If the points A, H, and C are collinear, then AK “ AL and KH “ LH, so
the equality p1q follows. Assume therefore that the points A, H, and C do not lie in a line and
consider the circle ω passing through them (see Figure 3). Since the quadrilateral ABCD is
cyclic,
=BAC “ =BDC “ 90˝ ´ =ADH “ =HAD .
Let N ‰ A be the intersection point of the circle ω and the angle bisector of =CAH. Then
AN is also the angle bisector of =BAD. Since H and C are symmetric to each other with
respect to the line KL and HN “ NC, it follows that both N and the centre of ω lie on the
line KL. This means that the circle ω is an Apollonius circle of the points K and L. This
immediately yields p1q.
Comment. Either proof can be used to obtain the following generalised result:
Let ABCD be a convex quadrilateral and let H be a point in its interior with =BAC “ =DAH. The
points S and T are chosen on the sides AB and AD, respectively, in such a way that H lies inside
triangle SCT and
=SHC ´ =BSC “ 90˝ , =T HC ´ =DT C “ 90˝ .
Then the circumcentre of triangle SHT lies on the line AH (and moreover the circumcentre of trian-
gle SCT lies on AC).
G6. Let ABC be a fixed acute-angled triangle. Consider some points E and F lying on
the sides AC and AB, respectively, and let M be the midpoint of EF . Let the perpendicular
bisector of EF intersect the line BC at K, and let the perpendicular bisector of MK intersect
the lines AC and AB at S and T , respectively. We call the pair pE, F q interesting, if the
quadrilateral KSAT is cyclic.
Suppose that the pairs pE1 , F1 q and pE2 , F2 q are interesting. Prove that
E1 E2 F1 F2
“ .
AB AC
(Iran)
Solution 1. For any interesting pair pE, F q, we will say that the corresponding triangle EF K
is also interesting.
Let EF K be an interesting triangle. Firstly, we prove that =KEF “ =KF E “ =A, which
also means that the circumcircle ω1 of the triangle AEF is tangent to the lines KE and KF .
Denote by ω the circle passing through the points K, S, A, and T . Let the line AM intersect
the line ST and the circle ω (for the second time) at N and L, respectively (see Figure 1).
Since EF k T S and M is the midpoint of EF , N is the midpoint of ST . Moreover, since K
and M are symmetric to each other with respect to the line ST , we have =KNS “ =MNS “
=LNT . Thus the points K and L are symmetric to each other with respect to the perpendicular
bisector of ST . Therefore KL k ST .
Let G be the point symmetric to K with respect to N. Then G lies on the line EF , and we
may assume that it lies on the ray MF . One has
=KGE “ =KNS “ =SNM “ =KLA “ 180˝ ´ =KSA
(if K “ L, then the angle KLA is understood to be the angle between AL and the tangent
to ω at L). This means that the points K, G, E, and S are concyclic. Now, since KSGT is a
parallelogram, we obtain =KEF “ =KSG “ 180˝ ´ =T KS “ =A. Since KE “ KF , we also
have =KF E “ =KEF “ =A.
After having proved this fact, one may finish the solution by different methods.
A
ω ω1
E A
E
M
F S
E1
G F2
B
B N
N E2
K C F1
Z1 Z2
T
L B K2 K1 C
Figure 1 Figure 2
First method. We have just proved that all interesting triangles are similar to each other.
This allows us to use the following lemma.
Lemma. Let ABC be an arbitrary triangle. Choose two points E1 and E2 on the side AC, two
points F1 and F2 on the side AB, and two points K1 and K2 on the side BC, in a way that the
triangles E1 F1 K1 and E2 F2 K2 are similar. Then the six circumcircles of the triangles AEi Fi ,
BFi Ki , and CEi Ki (i “ 1, 2) meet at a common point Z. Moreover, Z is the centre of the
spiral similarity that takes the triangle E1 F1 K1 to the triangle E2 F2 K2 .
Proof. Firstly, notice that for each i “ 1, 2, the circumcircles of the triangles AEi Fi , BFi Ki ,
and CKi Ei have a common point Zi by Miquel’s theorem. Moreover, we have
>pZi Fi , Zi Ei q “ >pAB, CAq , >pZi Ki , Zi Fi q “ >pBC, ABq , >pZi Ei , Zi Ki q “ >pCA, BCq .
This yields that the points Z1 and Z2 correspond to each other in similar triangles E1 F1 K1
and E2 F2 K2 . Thus, if they coincide, then this common point is indeed the desired centre of a
spiral similarity.
Finally, in order to show that Z1 “ Z2 , one may notice that >pAB, AZ1 q “ >pE1 F1 , E1 Z1 q “
>pE2 F2 , E2 Z2 q “ >pAB, AZ2 q (see Figure 2). Similarly, one has >pBC, BZ1 q “ >pBC, BZ2 q
and >pCA, CZ1 q “ >pCA, CZ2 q. This yields Z1 “ Z2 . l
Now, let P and Q be the feet of the perpendiculars from B and C onto AC and AB,
respectively, and let R be the midpoint of BC (see Figure 3). Then R is the circumcentre
of the cyclic quadrilateral BCP Q. Thus we obtain =AP Q “ =B and =RP C “ =C, which
yields =QP R “ =A. Similarly, we show that =P QR “ =A. Thus, all interesting triangles are
similar to the triangle P QR.
A A
F1
P
P
F2 E2
Z
E1 Q
Q Z
B R C
B K1 K2 R C
Figure 3 Figure 4
Denote now by Z the common point of the circumcircles of AP Q, BQR, and CP R. Let
E1 F1 K1 and E2 F2 K2 be two interesting triangles. By the lemma, Z is the centre of any
spiral similarity taking one of the triangles E1 F1 K1 , E2 F2 K2 , and P QR to some other of them.
Therefore the triangles ZE1 E2 and ZF1 F2 are similar, as well as the triangles ZE1 F1 and ZP Q.
Hence
E1 E2 ZE1 ZP
“ “ .
F1 F2 ZF1 ZQ
Moreover, the equalities =AZQ “ =AP Q “ =ABC “ 180˝ ´ =QZR show that the point Z
lies on the line AR (see Figure 4). Therefore the triangles AZP and ACR are similar, as well
as the triangles AZQ and ABR. This yields
ZP ZP RB AZ AB AB
“ ¨ “ ¨ “ ,
ZQ RC ZQ AC AZ AC

Second method. Now we will start from the fact that ω1 is tangent to the lines KE and KF
(see Figure 5). We prove that if pE, F q is an interesting pair, then
AE AF
` “ 2 cos =A. (1)
AB AC
Let Y be the intersection point of the segments BE and CF . The points B, K, and C are
collinear, hence applying Pascal’s theorem to the degenerated hexagon AF F Y EE, we infer
that Y lies on the circle ω1 .
Denote by Z the second intersection point of the circumcircle of the triangle BF Y with
the line BC (see Figure 6). By Miquel’s theorem, the points C, Z, Y , and E are concyclic.
Therefore we obtain
BF ¨ AB ` CE ¨ AC “ BY ¨ BE ` CY ¨ CF “ BZ ¨ BC ` CZ ¨ BC “ BC 2 .
On the other hand, BC 2 “ AB 2 ` AC 2 ´ 2AB ¨ AC cos =A, by the cosine law. Hence
pAB ´ AF q ¨ AB ` pAC ´ AEq ¨ AC “ AB 2 ` AC 2 ´ 2AB ¨ AC cos =A ,
which simplifies to the desired equality (1).

Let now pE1 , F1 q and pE2 , F2 q be two interesting pairs of points. Then we get
AE1 AF1 AE2 AF2

` “ ` ,
AB AC AB AC
which gives the desired result.
A
E
ω1 A
M
S E
ω1
F Y
F
Y
B
K C
B Z C
T
Figure 5 Figure 6
Third method. Again, we make use of the fact that all interesting triangles are similar (and
equi-oriented). Let us put the picture onto a complex plane such that A is at the origin, and
identify each point with the corresponding complex number.
Let EF K be any interesting triangle. The equalities =KEF “ =KF E “ =A yield that the
ratio ν “ KÉ
F É
is the same for all interesting triangles. This in turn means that the numbers E,
F , and K satisfy the linear equation
K “ µE ` νF, where µ “ 1 ´ ν. (2)

Now let us choose the points X and Y on the rays AB and AC, respectively, so that
=CXA “ =AY B “ =A “ =KEF (see Figure 7). Then each of the triangles AXC and Y AB
is similar to any interesting triangle, which also means that
C “ µA ` νX “ νX and B “ µY ` νA “ µY. (3)
Moreover, one has X{Y “ C{B.

Since the points E, F , and K lie on AC, AB, and BC, respectively, one gets
E “ ρY, F “ σX, and K “ λB ` p1 ´ λqC
for some real ρ, σ, and λ. In view of (3), the equation (2) now reads λB ` p1 ´ λqC “ K “
µE ` νF “ ρB ` σC, or
pλ ´ ρqB “ pσ ` λ ´ 1qC.
Since the nonzero complex numbers B and C have different arguments, the coefficients in the
brackets vanish, so ρ “ λ and σ “ 1 ´ λ. Therefore,
E F
` “ ρ ` σ “ 1. (4)
Y X
Now, if pE1 , F1 q and pE2 , F2 q are two distinct interesting pairs, one may apply (4) to both
pairs. Subtracting, we get
E1 ´ E2 F2 ´ F1 E1 ´ E2 Y B
“ , so “ “ .
Y X F2 ´ F1 X C
Taking absolute values provides the required result.
E
K
A
F B X
Figure 7
Comment 1. One may notice that the triangle P QR is also interesting.
Comment 2. In order to prove that =KEF “ =KF E “ =A, one may also use the following
well-known fact:
Let AEF be a triangle with AE ‰ AF , and let K be the common point of the symmedian taken from A
and the perpendicular bisector of EF . Then the lines KE and KF are tangent to the circumcircle ω1
of the triangle AEF .
In this case, however, one needs to deal with the case AE “ AF separately.
Solution 2. Let pE, F q be an interesting pair. This time we prove that
AM
“ cos =A . (5)
AK
As in Solution 1, we introduce the circle ω passing through the points K, S, A, and T , together
with the points N and L at which the line AM intersect the line ST and the circle ω for the
second time, respectively. Let moreover O be the centre of ω (see Figures 8 and 9). As in
Solution 1, we note that N is the midpoint of ST and show that KL k ST , which implies
=F AM “ =EAK.
X A
ω
E
A
M S ω
F O
O
E
N M
B F S
K C N
T T
L B K =L C
Figure 8 Figure 9
Suppose now that K ‰ L (see Figure 8). Then KL k ST , and consequently the lines KM
and KL are perpendicular. It implies that the lines LO and KM meet at a point X lying on the
circle ω. Since the lines ON and XM are both perpendicular to the line ST , they are parallel
to each other, and hence =LON “ =LXK “ =MAK. On the other hand, =OLN “ =MKA,
so we infer that triangles NOL and MAK are similar. This yields
AM ON ON
“ “ “ cos =T ON “ cos =A .
AK OL OT
If, on the other hand, K “ L, then the points A, M, N, and K lie on a common line, and
this line is the perpendicular bisector of ST (see Figure 9). This implies that AK is a diameter
of ω, which yields AM “ 2OK ´ 2NK “ 2ON. So also in this case we obtain
AM 2ON
“ “ cos =T ON “ cos =A .
AK 2OT
Thus (5) is proved.
Let P and Q be the feet of the perpendiculars from B and C onto AC and AB, respectively
(see Figure 10). We claim that the point M lies on the line P Q. Consider now the composition
of the dilatation with factor cos =A and centre A, and the reflection with respect to the angle
bisector of =BAC. This transformation is a similarity that takes B, C, and K to P , Q, and M,
respectively. Since K lies on the line BC, the point M lies on the line P Q.
A
ω
E
E
P
P
M
Q S
F
B
K C
Figure 10
Suppose that E ‰ P . Then also F ‰ Q, and by Menelaus’ theorem, we obtain
AQ F M EP
¨ ¨ “ 1.
F Q EM AP
Using the similarity of the triangles AP Q and ABC, we infer that
EP AP AB EP FQ
“ “ , and hence “ .
FQ AQ AC AB AC
The last equality holds obviously also in case E “ P , because then F “ Q. Moreover, since
the line P Q intersects the segment EF , we infer that the point E lies on the segment AP if
and only if the point F lies outside of the segment AQ.
Let now pE1 , F1 q and pE2 , F2 q be two interesting pairs. Then we obtain
E1 P F1 Q E2 P F2 Q
“ and “ .
AB AC AB AC
If P lies between the points E1 and E2 , we add the equalities above, otherwise we subtract
them. In any case we obtain
E1 E2 F1 F2
“ ,
AB AC
G7. Let ABC be a triangle with circumcircle Ω and incentre I. Let the line passing through I
and perpendicular to CI intersect the segment BC and the arc BC (not containing A) of Ω at
points U and V , respectively. Let the line passing through U and parallel to AI intersect AV
at X, and let the line passing through V and parallel to AI intersect AB at Y . Let W and Z be
the midpoints of AX and BC, respectively. Prove that if the points I, X, and Y are collinear,
then the points I, W , and Z are also collinear.
(U.S.A.)
Solution 1. We start with some general observations. Set α “ =A{2, β “ =B{2, γ “ =C{2.
Then obviously α ` β ` γ “ 90˝ . Since =UIC “ 90˝ , we obtain =IUC “ α ` β. Therefore
=BIV “ =IUC ´ =IBC “ α “ =BAI “ =BY V , which implies that the points B, Y , I,
and V lie on a common circle (see Figure 1).
Assume now that the points I, X and Y are collinear. We prove that =Y IA “ 90˝ .
Let the line XU intersect AB at N. Since the lines AI, UX, and V Y are parallel, we get
NX YN VU XU
“ “ “ ,
AI YA VI AI
implying NX “ XU. Moreover, =BIU “ α “ =BNU. This implies that the quadrilat-
eral BUIN is cyclic, and since BI is the angle bisector of =UBN, we infer that NI “ UI.
Thus in the isosceles triangle NIU, the point X is the midpoint of the base NU. This gives
=IXN “ 90˝ , i.e., =Y IA “ 90˝ .
A

x α
N
T
α
I
X
Y α
β γ
B U x C
Figure 1
Let S be the midpoint of the segment V C. Let moreover T be the intersection point of the
lines AX and SI, and set x “ =BAV “ =BCV . Since =CIA “ 90˝ ` β and SI “ SC, we
obtain
=T IA “ 180˝ ´ =AIS “ 90˝ ´ β ´ =CIS “ 90˝ ´ β ´ γ ´ x “ α ´ x “ =T AI ,
which implies that T I “ T A. Therefore, since =XIA “ 90˝ , the point T is the midpoint
of AX, i.e., T “ W .
To complete our solution, it remains to show that the intersection point of the lines IS
and BC coincide with the midpoint of the segment BC. But since S is the midpoint of the
segment V C, it suffices to show that the lines BV and IS are parallel.
Since the quadrilateral BY IV is cyclic, =V BI “ =V Y I “ =Y IA “ 90˝ . This implies that

BV is the external angle bisector of the angle ABC, which yields =V AC “ =V CA. Therefore
2α ´ x “ 2γ ` x, which gives α “ γ ` x. Hence =SCI “ α, so =V SI “ 2α.
On the other hand, =BV C “ 180˝ ´ =BAC “ 180˝ ´ 2α, which implies that the lines BV
and IS are parallel. This completes the solution.
Solution 2. As in Solution 1, we first prove that the points B, Y , I, V lie on a common circle
and =Y IA “ 90˝ . The remaining part of the solution is based on the following lemma, which
holds true for any triangle ABC, not necessarily with the property that I, X, Y are collinear.
Lemma. Let ABC be the triangle inscribed in a circle Γ and let I be its incentre. Assume
that the line passing through I and perpendicular to the line AI intersects the side AB at the
point Y . Let the circumcircle of the triangle BY I intersect the circle Γ for the second time
at V , and let the excircle of the triangle ABC opposite to the vertex A be tangent to the
side BC at E. Then
=BAV “ =CAE .
?
Proof. Let ρ be the composition of the inversion with centre A and radius AB ¨ AC, and the
symmetry with respect to AI. Clearly, ρ interchanges B and C.
Let J be the excentre of the triangle ABC opposite to A (see Figure 2). Then we have
=JAC “ =BAI and =JCA “ 90˝ ` γ “ =BIA, so the triangles ACJ and AIB are similar,
and therefore AB ¨ AC “ AI ¨ AJ. This means that ρ interchanges I and J. Moreover, since
Y lies on AB and =AIY “ 90˝ , the point Y 1 “ ρpY q lies on AC, and =JY 1 A “ 90˝ . Thus ρ
maps the circumcircle γ of the triangle BY I to a circle γ 1 with diameter JC.
Finally, since V lies on both Γ and γ, the point V 1 “ ρpV q lies on the line ρpΓq “ AB as
well as on γ 1 , which in turn means that V 1 “ E. This implies the desired result. l
F
A W
I K
X
X ω1
Y
I
Y γ
E D
C B C
B U Z E
Y′
V
ω2
γ′
V
J
Figure 2 Figure 3
Now we turn to the solution of the problem.

Assume that the incircle ω1 of the triangle ABC is tangent to BC at D, and let the
excircle ω2 of the triangle ABC opposite to the vertex A touch the side BC at E (see Figure 3).
The homothety with centre A that takes ω2 to ω1 takes the point E to some point F , and the
tangent to ω1 at F is parallel to BC. Therefore DF is a diameter of ω1 . Moreover, Z is the

midpoint of DE. This implies that the lines IZ and F E are parallel.
Let K “ Y I X AE. Since =Y IA “ 90˝ , the lemma yields that I is the midpoint of XK.
This implies that the segments IW and AK are parallel. Therefore, the points W , I and Z are
collinear.
Comment 1. The properties =Y IA “ 90˝ and V A “ V C can be established in various ways. The
main difficulty of the problem seems to find out how to use these properties in connection to the points
W and Z.
In Solution 2 this principal part is more or less covered by the lemma, for which we have presented
a direct proof. On the other hand, this lemma appears to be a combination of two well-known facts;
let us formulate them in terms of the lemma statement.
Let the line IY intersect AC at P (see Figure 4). The first fact states that the circumcircle ω of
the triangle V Y P is tangent to the segments AB and AC, as well as to the circle Γ. The second fact
states that for such a circle, the angles BAV and CAE are equal.
The awareness of this lemma may help a lot in solving this problem; so the Jury might also consider
a variation of the proposed problem, for which the lemma does not seem to be useful; see Comment 3.
N W Q
A
I ω1
X′′′′
X
Y X
P
P D
B C
U Z E
Y I
ω2
B E C
ω V V′
ω2
V
Figure 4 Figure 5
Comment 2. The proposed problem stated the equivalence: the point I lies on the line XY if and
only if I lies on the line W Z. Here we sketch the proof of the “if” part (see Figure 5).
As in Solution 2, let BC touch the circles ω1 and ω2 at D and E, respectively. Since IZ k AE and W
lies on IZ, the line DX is also parallel to AE. Therefore, the triangles XU P and AIQ are similar.
Moreover, the line DX is symmetric to AE with respect to I, so IP “ IQ, where P “ U V X XD and
Q “ U V X AE. Thus we obtain
UV UX UP UP
“ “ “ .
VI IA IQ IP
So the pairs IU and P V are harmonic conjugates, and since =U DI “ 90˝ , we get =V DB “ =BDX “
=BEA. Therefore the point V 1 symmetric to V with respect to the perpendicular bisector of BC lies
on the line AE. So we obtain =BAV “ =CAE.
The rest can be obtained by simply reversing the arguments in Solution 2. The points B, V , I, and Y
are concyclic. The lemma implies that =Y IA “ 90˝ . Moreover, the points B, U , I, and N , where
N “ U X X AB, lie on a common circle, so IN “ IU . Since IY K U N , the point X 1 “ IY X U N is
the midpoint of U N . But in the trapezoid AY V I, the line XU is parallel to the sides AI and Y V , so
N X “ U X 1 . This yields X “ X 1 .
The reasoning presented in Solution 1 can also be reversed, but it requires a lot of technicalities.
Therefore the Problem Selection Committee proposes to consider only the “only if” part of the original
proposal, which is still challenging enough.
Comment 3. The Jury might also consider the following variation of the proposed problem.
Let ABC be a triangle with circumcircle Ω and incentre I. Let the line through I perpendicular to CI
intersect the segment BC and the arc BC (not containing A) of Ω at U and V , respectively. Let the
line through U parallel to AI intersect AV at X. Prove that if the lines XI and AI are perpendicular,
then the midpoint of the segment AC lies on the line XI (see Figure 6).
α
N
α
M
I
X
A Y

α
β γ
B C
U
I
X
B C
U
V
Figure 6 Figure 7
Since the solution contains the arguments used above, we only sketch it.
Let N “ XU X AB (see Figure 7). Then =BN U “ =BAI “ =BIU , so the points B, U , I, and N lie
on a common circle. Therefore IU “ IN , and since IX K N U , it follows that N X “ XU .
Now set Y “ XI X AB. The equality N X “ XU implies that
VX XU NX YX
“ “ “ ,
VA AI AI YI
and therefore Y V k AI. Hence =BY V “ =BAI “ =BIV , so the points B, V , I, Y are concyclic.
Next we have IY K Y V , so =IBV “ 90˝ . This implies that BV is the external angle bisector of the
angle ABC, which gives =V AC “ =V CA.
So in order to show that M “ XI X AC is the midpoint of AC, it suffices to prove that =V M C “ 90˝ .
But this follows immediately from the observation that the points V , C, M , and I are concyclic, as
=M IV “ =Y BV “ 180˝ ´ =ACV .
The converse statement is also true, but its proof requires some technicalities as well.
Number Theory
N1. Let n ě 2 be an integer, and let An be the set
An “ t2n ´ 2k | k P Z, 0 ď k ă nu.
Determine the largest positive integer that cannot be written as the sum of one or more (not
necessarily distinct) elements of An .
(Serbia)
Answer. pn ´ 2q2n ` 1.
Solution 1.
Part I. First we show that every integer greater than pn ´ 2q2n ` 1 can be represented as
such a sum. This is achieved by induction on n.
For n “ 2, the set An consists of the two elements 2 and 3. Every positive integer m except
for 1 can be represented as the sum of elements of An in this case: as m “ 2 ` 2 ` ¨ ¨ ¨ ` 2 if m
is even, and as m “ 3 ` 2 ` 2 ` ¨ ¨ ¨ ` 2 if m is odd.
Now consider some n ą 2, and take an integer m ą pn´ 2q2n ` 1. If m is even, then consider
m pn ´ 2q2n ` 2
ě “ pn ´ 2q2n´1 ` 1 ą pn ´ 3q2n´1 ` 1.
2 2
By the induction hypothesis, there is a representation of the form
m
“ p2n´1 ´ 2k1 q ` p2n´1 ´ 2k2 q ` ¨ ¨ ¨ ` p2n´1 ´ 2kr q
2
for some ki with 0 ď ki ă n ´ 1. It follows that
m “ p2n ´ 2k1 `1 q ` p2n ´ 2k2 `1 q ` ¨ ¨ ¨ ` p2n ´ 2kr `1 q,
giving us the desired representation as a sum of elements of An . If m is odd, we consider

m ´ p2n ´ 1q pn ´ 2q2n ` 1 ´ p2n ´ 1q
ą “ pn ´ 3q2n´1 ` 1.
2 2
By the induction hypothesis, there is a representation of the form
m ´ p2n ´ 1q
“ p2n´1 ´ 2k1 q ` p2n´1 ´ 2k2 q ` ¨ ¨ ¨ ` p2n´1 ´ 2kr q
2
for some ki with 0 ď ki ă n ´ 1. It follows that
m “ p2n ´ 2k1 `1 q ` p2n ´ 2k2 `1 q ` ¨ ¨ ¨ ` p2n ´ 2kr `1 q ` p2n ´ 1q,
giving us the desired representation of m once again.

Part II. It remains to show that there is no representation for pn ´ 2q2n ` 1. Let N be
the smallest positive integer that satisfies N ” 1 pmod 2n q, and which can be represented as a
sum of elements of An . Consider a representation of N, i.e.,
N “ p2n ´ 2k1 q ` p2n ´ 2k2 q ` ¨ ¨ ¨ ` p2n ´ 2kr q, (1)
where 0 ď k1 , k2 , . . . , kr ă n. Suppose first that two of the terms in the sum are the same, i.e.,
ki “ kj for some i ‰ j. If ki “ kj “ n ´ 1, then we can simply remove these two terms to get a
representation for
N ´ 2p2n ´ 2n´1 q “ N ´ 2n
as a sum of elements of An , which contradicts our choice of N. If ki “ kj “ k ă n ´ 1, replace

the two terms by 2n ´ 2k`1, which is also an element of An , to get a representation for
N ´ 2p2n ´ 2k q ` 2n ´ 2k`1 “ N ´ 2n .
This is a contradiction once again. Therefore, all ki have to be distinct, which means that
2k1 ` 2k2 ` ¨ ¨ ¨ ` 2kr ď 20 ` 21 ` 22 ` ¨ ¨ ¨ ` 2n´1 “ 2n ´ 1.
On the other hand, taking (1) modulo 2n , we find
2k1 ` 2k2 ` ¨ ¨ ¨ ` 2kr ” Ń ” ´1 pmod 2n q.
Thus we must have 2k1 ` 2k2 ` ¨ ¨ ¨ ` 2kr “ 2n ´ 1, which is only possible if each element of
t0, 1, . . . , n ´ 1u occurs as one of the ki . This gives us
N “ n2n ´ p20 ` 21 ` ¨ ¨ ¨ ` 2n´1 q “ pn ´ 1q2n ` 1.
In particular, this means that pn ´ 2q2n ` 1 cannot be represented as a sum of elements of An .
Solution 2. The fact that m “ pn ´ 2q2n ` 1 cannot be represented as a sum of elements

of An can also be shown in other ways. We prove the following statement by induction on n:
Claim. If a, b are integers with a ě 0, b ě 1, and a ` b ă n, then a2n ` b cannot be written as
a sum of elements of An .
Proof. The claim is clearly true for n “ 2 (since a “ 0, b “ 1 is the only possibility). For
n ą 2, assume that there exist integers a, b with a ě 0, b ě 1 and a ` b ă n as well as elements
m1 , m2 , . . . , mr of An such that
a2n ` b “ m1 ` m2 ` ¨ ¨ ¨ ` mr .
We can suppose, without loss of generality, that m1 ě m2 ě ¨ ¨ ¨ ě mr . Let ℓ be the largest

index for which mℓ “ 2n ´ 1 (ℓ “ 0 if m1 ‰ 2n ´ 1). Clearly, ℓ and b must have the same parity.
Now
pa ´ ℓq2n ` pb ` ℓq “ mℓ`1 ` mℓ`2 ` ¨ ¨ ¨ ` mr
and thus
b`ℓ mℓ`1 mℓ`2 mr
pa ´ ℓq2n´1 ` “ ` ` ¨¨¨` .
2 2 2 2
Note that mℓ`1 {2, mℓ`2 {2, . . . , mr {2 are elements of An´1 . Moreover, a ´ ℓ and pb ` ℓq{2 are
integers, and pb ` ℓq{2 ě 1. If a ´ ℓ was negative, then we would have
a2n ` b ě ℓp2n ´ 1q ě pa ` 1qp2n ´ 1q “ a2n ` 2n ´ a ´ 1,
thus n ě a ` b ` 1 ě 2n , which is impossible. So a ´ ℓ ě 0. By the induction hypothesis, we

must have a ´ ℓ ` b`ℓ
2
ě n ´ 1, which gives us a contradiction, since
b`ℓ
a´ℓ` ď a ´ ℓ ` b ` ℓ ´ 1 “ a ` b ´ 1 ă n ´ 1. l
2
Considering the special case a “ n ´ 2, b “ 1 now completes the proof.

Solution 3. Denote by Bn the set of all positive integers that can be written as a sum of
elements of An . In this solution, we explicitly describe all the numbers in Bn by an argument
similar to the first solution.
For a positive integer n, we denote by σ2 pnq the sum of its digits in the binary representation.
Notice that every positive integer m has a unique representation of the form m “ s2n ´ t with
some positive integer s and 0 ď t ď 2n ´ 1.
Lemma. For any two integers s ě 1 and 0 ď t ď 2n ´ 1, the number m “ s2n ´ t belongs to Bn
if and only if s ě σ2 ptq.
Proof. For t “ 0, the statement of the Lemma is obvious, since m “ 2s ¨ p2n ´ 2n´1 q.
Now suppose that t ě 1, and let
t “ 2k1 ` ¨ ¨ ¨ ` 2kσ p0 ď k1 ă ¨ ¨ ¨ ă kσ ď n ´ 1, σ “ σ2 ptqq
be its binary expansion. If s ě σ, then m P Bn since

σ
ÿ
n n n n´1
m “ ps ´ σq2 ` pσ2 ´ tq “ 2ps ´ σq ¨ p2 ´ 2 q` p2n ´ 2ki q.
i“1
Assume now that there exist integers s and t with 1 ď s ă σ2 ptq and 0 ď t ď 2n ´ 1 such
that the number m “ s2n ´ t belongs to Bn . Among all such instances, choose the one for
which m is smallest, and let
d
ÿ
m“ p2n ´ 2ℓi q p0 ď ℓi ď n ´ 1q
i“1
be the corresponding representation. If all the ℓi ’s are distinct, then di“1 2ℓi ď j“0
řn´1 j
2 “ 2n ´1,
ř
řd ℓi
so one has s “ d and t “ i“1 2 , whence s “ d “ σ2 ptq; this is impossible. Therefore, two of
the ℓi ’s must be equal, say ℓd´1 “ ℓd . Then m ě 2p2n ´ 2ℓd q ě 2n , so s ě 2.
Now we claim that the number m1 “ m ´ 2n “ ps ´ 1q2n ´ t also belongs to Bn , which
contradicts the minimality assumption. Indeed, one has
p2n ´ 2ℓd´1 q ` p2n ´ 2ℓd q “ 2p2n ´ 2ℓd q “ 2n ` p2n ´ 2ℓd `1 q,
so
d´2
ÿ
1
m “ p2n ´ 2ℓi q ` p2n ´ 2ℓd `1 q
i“1
is the desired representation of m1 (if ℓd “ n ´ 1, then the last summand is simply omitted).
This contradiction finishes the proof. l
By our lemma, the largest number M which does not belong to Bn must have the form
mt “ pσ2 ptq ´ 1q2n ´ t
for some t with 1 ď t ď 2n ´ 1, so M is just the largest of these numbers. For t0 “ 2n ´ 1 we

have mt0 “ pn ´ 1q2n ´ p2n ´ 1q “ pn ´ 2q2n ` 1; for every other value of t one has σ2 ptq ď n ´ 1,
thus mt ď pσptq ´ 1q2n ď pn ´ 2q2n ă mt0 . This means that M “ mt0 “ pn ´ 2q2n ` 1.
N2. Determine all pairs px, yq of positive integers such that

a
3
7x2 ´ 13xy ` 7y 2 “ |x ´ y| ` 1 . (1)
(U.S.A.)
Answer. Either px, yq “ p1, 1q or tx, yu “ tm3 ` m2 ´ 2m ´ 1, m3 ` 2m2 ´ m ´ 1u for some

positive integer m ě 2.
Solution. Let px, yq be any pair of positive integers solving (1). We shall prove that it appears
in the list displayed above. The converse assertion that all these pairs do actually satisfy (1)
either may be checked directly by means of a somewhat laborious calculation, or it can be seen
by going in reverse order through the displayed equations that follow.
In case x “ y the given equation reduces to x2{3 “ 1, which is equivalent to x “ 1, whereby
he have found the first solution.
To find the solutions with x ‰ y we may assume x ą y due to symmetry. Then the integer
n “ x ´ y is positive and (1) may be rewritten as
a
3
7py ` nq2 ´ 13py ` nqy ` 7y 2 “ n ` 1 .
Raising this to the third power and simplifying the result one obtains
y 2 ` yn “ n3 ´ 4n2 ` 3n ` 1 .
To complete the square on the left hand side, we multiply by 4 and add n2 , thus getting
p2y ` nq2 “ 4n3 ´ 15n2 ` 12n ` 4 “ pn ´ 2q2 p4n ` 1q .
This shows that the cases n “ 1 and n “ 2 are impossible, whence n ą 2, and 4n ` 1 is the
square of the rational number 2y`nn´2
. Consequently, it has to be a perfect square, and, since it
is odd as well, there has to exist some nonnegative integer m such that 4n ` 1 “ p2m ` 1q2 , i.e.
n “ m2 ` m .
Notice that n ą 2 entails m ě 2. Substituting the value of n just found into the previous
displayed equation we arrive at
p2y ` m2 ` mq2 “ pm2 ` m ´ 2q2 p2m ` 1q2 “ p2m3 ` 3m2 ´ 3m ´ 2q2 .
Extracting square roots and taking 2m3 ` 3m2 ´ 3m ´ 2 “ pm ´ 1qp2m2 ` 5m ` 2q ą 0 into
account we derive 2y ` m2 ` m “ 2m3 ` 3m2 ´ 3m ´ 2, which in turn yields
y “ m3 ` m2 ´ 2m ´ 1 .
Notice that m ě 2 implies that y “ pm3 ´ 1q ` pm ´ 2qm is indeed positive, as it should be. In
view of x “ y ` n “ y ` m2 ` m it also follows that
x “ m3 ` 2m2 ´ m ´ 1 ,
and that this integer is positive as well.
Comment. Alternatively one could ask to find all pairs px, yq of – not necessarily positive – integers
solving (1). The answer to that question is a bit nicer than the answer above: the set of solutions are
now described by
tx, yu “ tm3 ` m2 ´ 2m ´ 1, m3 ` 2m2 ´ m ´ 1u ,
where m varies through Z. This may be shown using essentially the same arguments as above. We
finally observe that the pair px, yq “ p1, 1q, that appears to be sporadic above, corresponds to m “ ´1.
N3. A coin is called a Cape Town coin if its value is 1{n for some positive integer n. Given
a collection of Cape Town coins of total value at most 99 ` 21 , prove that it is possible to split
this collection into at most 100 groups each of total value at most 1.
(Luxembourg)
Solution. We will show that for every positive integer N any collection of Cape Town coins
of total value at most N ´ 21 can be split into N groups each of total value at most 1. The
problem statement is a particular case for N “ 100.
We start with some preparations. If several given coins together have a total value also of
the form k1 for a positive integer k, then we may merge them into one new coin. Clearly, if the
resulting collection can be split in the required way then the initial collection can also be split.
After each such merging, the total number of coins decreases, thus at some moment we
come to a situation when no more merging is possible. At this moment, for every even k there
is at most one coin of value k1 (otherwise two such coins may be merged), and for every odd
k ą 1 there are at most k ´ 1 coins of value k1 (otherwise k such coins may also be merged).
Now, clearly, each coin of value 1 should form a single group; if there are d such coins then
we may remove them from the collection and replace N by N ´ d. So from now on we may
assume that there are no coins of value 1.
Finally, we may split all the coins in the following way. For each k “ 1, 2, . . . , N we put all
1 1
the coins of values 2k´1 and 2k into a group Gk ; the total value of Gk does not exceed
1 1
p2k ´ 2q ¨ ` ă 1.
2k ´ 1 2k
1
It remains to distribute the “small” coins of values which are less than 2N ; we will add them one
by one. In each step, take any remaining small coin. The total value of coins` in the˘ groups at
this moment is at most N ´ 21 , so there exists a group of total value at most N1 N ´ 12 “ 1 ´ 2N1
;
thus it is possible to put our small coin into this group. Acting so, we will finally distribute all
the coins.
Comment 1. The algorithm may be modified, at least the step where one distributes the coins of
1 1
values ě 2N . One different way is to put into Gk all the coins of values p2k´1q2s for all integer s ě 0.
One may easily see that their total value also does not exceed 1.
Comment 2. The original proposal also contained another part, suggesting to show that a required
splitting may be impossible if the total value of coins is at most 100. There are many examples of
such a collection, e.g. one may take 98 coins of value 1, one coin of value 21 , two coins of value 13 , and
four coins of value 51 .
The Problem Selection Committee thinks that this part is less suitable for the competition.
N4. Let n ą 1 be a given integer. Prove that infinitely many terms of the sequence pak qkě1 ,
defined by
nk
Z ^
ak “ ,
k
are odd. (For a real number x, txu denotes the largest integer not exceeding x.)
(Hong Kong)
nm ´m
Solution 1. If n is odd, let k “ nm for m “ 1, 2, . . .. Then ak “ n , which is odd for
each m.
Henceforth, assume that n is even, say n “ 2t for some integer t ě 1. Then, for any m ě 2,
m m m
the integer n2 ´ 2m “ 2m p22 ´m ¨ t2 ´ 1q has an odd prime divisor p, since 2m ´ m ą 1. Then,
for k “ p ¨ 2m , we have
m
nk “ pn2 qp ” p2m qp “ p2p qm ” 2m ,
where the congruences are taken modulo p (recall that 2p ” 2 pmod pq, by Fermat’s little
nk
theorem). Also, from nk ´ 2m ă nk ă nk ` 2m pp ´ 1q, we see that the fraction lies strictly
k
nk ´ 2m nk ` 2m pp ´ 1q
between the consecutive integers and , which gives
p ¨ 2m p ¨ 2m
Z k^
n nk ´ 2m
“ .
k p ¨ 2m
n k
nk ´ 2m 2m
´1 nk
We finally observe that “ is an odd integer, since the integer ´ 1 is odd
p ¨ 2m p 2m
(recall that k ą m). Note that for different values of m, we get different values of k, due to the
different powers of 2 in the prime factorisation of k.
Solution 2. Treat the (trivial) case when n is odd as in Solution 1.

Now assume that n is even and n ą 2. Let p be a prime divisor of n ´ 1.
i
Proceed by induction on i to prove that pi`1 is a divisor of np ´ 1 for every i ě 0. The case
i “ 0 is true by the way in which p is chosen. Suppose the result is true for some i ě 0. The
factorisation
i`1 i i i i
np ´ 1 “ pnp ´ 1qrnp pp´1q ` np pp´2q ` ¨ ¨ ¨ ` np ` 1s,
together with the fact that each of the p terms between the square brackets is congruent to 1
modulo p, Zimplies that the result is also true for i ` 1.
i^ i
np np ´ 1
Hence “ , an odd integer for each i ě 1.
pi pi
i
Finally, we consider the case n “ 2. We observe that 3 ¨ 4i is a divisor of 23¨4 ´ 4i for every
i i
i ě 1: Trivially, 4i is a divisor of 23¨4 ´ 4i , since 3 ¨ 4i ą 2i. ZFurthermore, since 23¨4 and 4i are
i i i
23¨4 23¨4 ´ 4i 23¨4 ´2i ´ 1
^
ˇ 3¨4i i
both congruent to 1 modulo 3, we have 3 2 ´ 4 . Hence,
ˇ “ “ ,
3 ¨ 4i 3 ¨ 4i 3
which is odd for every i ě 1.
Comment. The case n even and n ą 2 can also be solved by recursively defining the sequence pki qiě1
by k1 “ 1 and ki`1 “ nki ´ 1 for i ě 1. Then pki q is strictly increasing and it follows (by induction
on i) that ki | nki ´ 1 for all i ě 1, so the ki are as desired.
The case n “ 2 can also be solved as follows: Let i ě 2. By Bertrand’s postulate, there exists a
i i
prime number p such that 22 ´1 ă p ¨ 2i ă 22 . This gives
i
p ¨ 2i ă 22 ă 2p ¨ 2i . (1)
i i
Also, we have that p ¨ 2i is a divisor of 2p¨2 ´ 22 , hence, using (1), we get that
i ^ i i i i
2p¨2 2p¨2 ´ 22 ` p ¨ 2i 2p¨2 í ´ 22 í ` p
Z
“ “ ,
p ¨ 2i p ¨ 2i p
which is an odd integer.

Solution 3. Treat the (trivial) case when n is odd as in Solution 1.
Let n be even, and let p be a prime divisor of n ` 1. Define the sequence pai qiě1 by
ai “ min a P Zą0 : 2i divides ap ` 1 .

(
Recall that there exists a with 1 ď a ă 2i such that ap ” ´1 pmod 2i q, so each ai satisfies
1 ď ai ă 2i . This implies that ai p ` 1 ă p ¨ 2i . Also, ai Ñ 8 as i Ñ 8, whence there are
infinitely many i such that ai ă ai`1 . From now on, we restrict ourselves only to these i.
i
Notice that p is a divisor of np ` 1, which, in turn, divides np¨2 ´Z1. It follows that p ¨ 2i is a
i i
np¨2 np¨2 ´ pai p ` 1q
^
p¨2i
divisor of n ´ pai p ` 1q, and we consequently see that the integer “
p ¨ 2i p ¨ 2i
i
is odd, since 2i`1 divides np¨2 , but not ai p ` 1.
N5. Find all triples pp, x, yq consisting of a prime number p and two positive integers x and y
such that xp´1 ` y and x ` y p´1 are both powers of p.
(Belgium)
( (
Answer. pp, x, yq P p3, 2, 5q, p3, 5, 2q Y p2, n, 2k ´ nq | 0 ă n ă 2k .
Solution 1. For p “ 2, clearly all pairs of two positive integers x and y whose sum is a power
of 2 satisfy the condition. Thus we assume in the following that p ą 2, and we let a and b be
positive integers such that xp´1 ` y “ pa and x ` y p´1 “ pb . Assume further, without loss of
generality, that x ď y, so that pa “ xp´1 ` y ď x ` y p´1 “ pb , which means that a ď b (and
thus pa | pb ).
Now we have
pb “ y p´1 ` x “ ppa ´ xp´1 qp´1 ` x.
We take this equation modulo pa and take into account that p ´ 1 is even, which gives us
2
0 ” xpp´1q ` x pmod pa q.
2
If p | x, then pa | x, since xpp´1q ´1 ` 1 is not divisible by p in this case. However, this is
impossible, since x ď xp´1 ă pa . Thus we know that p ∤ x, which means that
2 ´1
pa | xpp´1q ` 1 “ xppp´2q ` 1.
2
By Fermat’s little theorem, xpp´1q ” 1 pmod pq, thus p divides x`1. Let pr be the highest
power of p that divides x ` 1. By the binomial theorem, we have
ppp´2q
ÿ ˆppp ´ 2q˙
ppp´2q
x “ p´1qppp´2q´k px ` 1qk .
k“0
k
Except for the terms corresponding to k “ 0, k “ 1 and k “ 2, all terms in the sum are clearly
divisible by p3r and thus by pr`2 . The remaining terms are
ppp ´ 2qpp2 ´ 2p ´ 1q
´ px ` 1q2 ,
2
which is divisible by p2r`1 and thus also by pr`2 ,
ppp ´ 2qpx ` 1q,
which is divisible by pr`1 , but not pr`2 by our choice of r, and the final term ´1 corresponding
to k “ 0. It follows that the highest power of p that divides xppp´2q ` 1 is pr`1 .
On the other hand, we already know that pa divides xppp´2q ` 1, which means that a ď r ` 1.
Moreover,
pr ď x ` 1 ď xp´1 ` y “ pa .
Hence we either have a “ r or a “ r ` 1.
If a “ r, then x “ y “ 1 needs to hold in the inequality above, which is impossible for
p ą 2. Thus a “ r ` 1. Now since pr ď x ` 1, we get
x2 ` x xp´1 ` y pa pa
x“ ď “ ď r “ p,
x`1 x`1 x`1 p
so we must have x “ p ´ 1 for p to divide x ` 1.
It follows that r “ 1 and a “ 2. If p ě 5, we obtain
pa “ xp´1 ` y ą pp ´ 1q4 “ pp2 ´ 2p ` 1q2 ą p3pq2 ą p2 “ pa ,
a contradiction. So the only case that remains is p “ 3, and indeed x “ 2 and y “ pa ´ xp´1 “ 5
satisfy the conditions.
Comment 1. In this solution, we are implicitly using a special case of the following lemma known
as “lifting the exponent”:
Lemma. Let n be a positive integer, let p be an odd prime, and let vp pmq denote the exponent of the
highest power of p that divides m.
If x and y are integers not divisible by p such that p | x ´ y, then we have
vp pxn ´ y n q “ vp px ´ yq ` vp pnq.
Likewise, if x and y are integers not divisible by p such that p | x ` y, then we have
vp pxn ` y n q “ vp px ` yq ` vp pnq.
Comment 2. There exist various ways of solving the problem involving the “lifting the exponent”
lemma. Let us sketch another one.
The cases x “ y and p | x are ruled out easily, so we assume that p ą 2, x ă y, and p ∤ x. In this
case we also have pa ă pb and p | x ` 1.
Now one has
y p ´ xp ” ypy p´1 ` xq ´ xpxp´1 ` yq ” 0 pmod pa q,
so by the lemma mentioned above one has pa´1 | y ´ x and hence y “ x ` tpa´1 for some positive
integer t. Thus one gets
xpxp´2 ` 1q “ xp´1 ` x “ pxp´1 ` yq ´ py ´ xq “ pa´1 pp ´ tq.
The factors on the left-hand side are coprime. So if p | x, then xp´2 ` 1 | p ´ t, which is impossible
since x ă xp´2 ` 1. Therefore, p ∤ x, and thus x | p ´ t. Since p | x ` 1, the only remaining case is
x “ p ´ 1, t “ 1, and y “ pa´1 ` p ´ 1. Now the solution can be completed in the same way as before.
Solution 2. Again, we can focus on the case that p ą 2. If p | x, then also p | y. In this case,
let pk and pℓ be the highest powers of p that divide x and y respectively, and assume without
loss of generality that k ď ℓ. Then pk divides x ` y p´1 while pk`1 does not, but pk ă x ` y p´1,
which yields a contradiction. So x and y are not divisible by p. Fermat’s little theorem yields
0 ” xp´1 ` y ” 1 ` y pmod pq, so y ” ´1 pmod pq and for the same reason x ” ´1 pmod pq.
In particular, x, y ě p ´ 1 and thus xp´1 ` y ě 2pp ´ 1q ą p, so xp´1 ` y and y p´1 ` x are
both at least equal to p2 . Now we have
xp´1 ” ý pmod p2 q and y p´1 ” ´x pmod p2 q.
These two congruences, together with the Euler–Fermat theorem, give us
1 ” xppp´1q ” pýqp ” ý p ” xy pmod p2 q.
Since x ” y ” ´1 pmod pq, x ´ y is divisible by p, so px ´ yq2 is divisible by p2 . This means

that
px ` yq2 “ px ´ yq2 ` 4xy ” 4 pmod p2 q,
so p2 divides px ` y ´ 2qpx ` y ` 2q. We already know that x ` y ” ´2 pmod pq, so x ` y ´ 2 ”
´4 ı 0 pmod pq. This means that p2 divides x ` y ` 2.
Using the same notation as in the first solution, we subtract the two original equations to
obtain
pb ´ pa “ y p´1 ´ xp´1 ` x ´ y “ py ´ xqpy p´2 ` y p´3x ` ¨ ¨ ¨ ` xp´2 ´ 1q. (1)
The second factor is symmetric in x and y, so it can be written as a polynomial of the elementary
symmetric polynomials x ` y and xy with integer coefficients. In particular, its value modulo
p2 is characterised by the two congruences xy ” 1 pmod p2 q and x ` y ” ´2 pmod p2 q. Since

both congruences are satisfied when x “ y “ ´1, we must have
y p´2 ` y p´3x ` ¨ ¨ ¨ ` xp´2 ´ 1 ” p´1qp´2 ` p´1qp´3 p´1q ` ¨ ¨ ¨ ` p´1qp´2 ´ 1 pmod p2 q,
which simplifies to y p´2 ` y p´3x ` ¨ ¨ ¨ ` xp´2 ´ 1 ” ´p pmod p2 q. Thus the second factor in (1)
is divisible by p, but not p2 .
This means that pa´1 has to divide the other factor y ´ x. It follows that
0 ” xp´1 ` y ” xp´1 ` x ” xpx ` 1qpxp´3 ´ xp´4 ` ¨ ¨ ¨ ` 1q pmod pa´1 q.
Since x ” ´1 pmod pq, the last factor is xp´3 ´ xp´4 ` ¨ ¨ ¨ ` 1 ” p ´ 2 pmod pq and in particular
not divisible by p. We infer that pa´1 | x ` 1 and continue as in the first solution.
Comment. Instead of reasoning by means of elementary symmetric polynomials, it is possible to

provide a more direct argument as well. For odd r, px ` 1q2 divides pxr ` 1q2 , and since p divides
x ` 1, we deduce that p2 divides pxr ` 1q2 . Together with the fact that xy ” 1 pmod p2 q, we obtain
0 ” y r pxr ` 1q2 ” x2r y r ` 2xr y r ` y r ” xr ` 2 ` y r pmod p2 q.
We apply this congruence with r “ p ´ 2 ´ 2k (where 0 ď k ă pp ´ 2q{2) to find that
xk y p´2´k ` xp´2´k y k ” pxyqk pxp´2´2k ` y p´2´2k q ” 1k ¨ p´2q ” ´2 pmod p2 q.
Summing over all k yields

p´1
y p´2 ` y p´3 x ` ¨ ¨ ¨ ` xp´2 ´ 1 ” 2 ¨ p´2q ´ 1 ” ´p pmod p2 q
once again.
N6. Let a1 ă a2 ă ¨ ¨ ¨ ă an be pairwise coprime positive integers with a1 being prime

and a1 ě n ` 2. On the segment I “ r0, a1 a2 ¨ ¨ ¨ an s of the real line, mark all integers that are
divisible by at least one of the numbers a1 , . . . , an . These points split I into a number of smaller
segments. Prove that the sum of the squares of the lengths of these segments is divisible by a1 .
(Serbia)
Solution 1. Let A “ a1 ¨ ¨ ¨ an . Throughout the solution, all intervals will be nonempty and
have integer end-points. For any interval X, the length of X will be denoted by |X|.
Define the following two families of intervals:

(
S “ rx, ys : x ă y are consecutive marked points
(
T “ rx, ys : x ă y are integers, 0 ď x ď A ´ 1, and no point is marked in px, yq
|X|2 modulo a1 .
ř
We are interested in computing
XPS
Note that the number A is marked, so in the definition of T the condition y ď A is enforced
without explicitly prescribing it.
Assign weights to the intervals` in ˘T , depending only on their lengths. The weight of an
arbitrary interval Y P T will be w |Y | , where
#
1 if k “ 1,
wpkq “
2 if k ě 2 .
Consider an arbitrary interval X P S and its sub-intervals Y P T . Clearly, X has one

sub-interval of length |X|, two sub-intervals of length |X| ´ 1 and so on; in general X has
|X| ´ d ` 1 sub-intervals of length d for every d “ 1, 2, . . . , |X|. The sum of the weights of the
sub-intervals of X is
ÿ |X|
` ˘ ÿ
p|X| ´ d ` 1q ¨ wpdq “ |X| ¨ 1 ` p|X| ´ 1q ` p|X| ´ 2q ` ¨ ¨ ¨ ` 1 ¨ 2 “ |X|2 .
` ˘
w |Y | “
Y PT , Y ĎX d“1
Since the intervals in S are non-overlapping, every interval Y P T is a sub-interval of a single

interval X P S. Therefore,
˜ ¸
ÿ ÿ ÿ ÿ ` ˘
|X|2 “
` ˘
w |Y | “ w |Y | . (1)
XPS XPS Y PT , Y ĎX Y PT
For every d “ 1, 2, . . . , a1 , we count how many intervals in T are of length d. Notice that
the multiples of a1 are all marked, so the lengths of the intervals in S and T cannot exceed a1 .
Let x be an arbitrary integer with 0 ď x ď A ´ 1 and consider the interval rx, x ` ds. Let r1 ,
. . . , rn be the remainders of x modulo a1 , . . . , an , respectively. Since a1 , . . . , an are pairwise
coprime, the number x is uniquely identified by the sequence pr1 , . . . , rn q, due to the Chinese
remainder theorem.
For every i “ 1, . . . , n, the property that the interval px, x`dq does not contain any multiple
of ai is equivalent with ri ` d ď ai , i.e. ri P t0, 1, . . . , ai ´ du, so there are ai ´ d ` 1 choices for
the number ri for each i. Therefore, the number of the remainder sequences pr1 , . . . , rn q that
satisfy rx, x ` ds P T is precisely pa1 ` 1 ´ dq ¨ ¨ ¨ pan ` 1 ´ dq. Denote this product by f pdq.
Now we can group the last sum in (1) by length of the intervals. As we have seen, for every
d “ 1, . . . , a1 there are f pdq intervals Y P T with |Y | “ d. Therefore, (1) can be continued as
ÿ ÿ a1 a1
2
` ˘ ÿ ÿ
|X| “ w |Y | “ f pdq ¨ wpdq “ 2 f pdq ´ f p1q. (2)
XPS Y PT d“1 d“1
Having the formula (2), the solution can be finished using the following well-known fact:
Lemma. If p is a prime, F pxq is a polynomial with integer coefficients, and deg F ď p ´ 2, then
řp
F pxq is divisible by p.
x“1
Proof. Obviously, it is sufficient to prove the lemma for monomials of the form xk with k ď p´2.
Apply induction on k. If k “ 0 then F “ 1, and the statement is trivial.
Let 1 ď k ď p ´ 2, and assume that the lemma is proved for all lower degrees. Then
p p
˜ ˙ ¸
k ˆ
ÿ ÿ ÿ k ` 1
0 ” pk`1 “ xk`1 ´ px ´ 1qk`1 “ p´1qk´ℓ xℓ
` ˘
x“1 x“1 ℓ“0
ℓ
p k´1 ˆ ˙ÿ p p
k´ℓ k ` 1
ÿ ÿ ÿ
k ℓ
“ pk ` 1q x ` p´1q x ” pk ` 1q xk pmod pq.
x“1 ℓ“0
ℓ x“1 x“1
p
xk ” 0 pmod pq.
ř
Since 0 ă k ` 1 ă p, this proves l
x“1
a1
ř
In (2), by applying the lemma to the polynomial f and the prime a1 , we obtain that f pdq
d“1
ř a1 . 2The term f p1q “ a1 ¨ ¨ ¨ an is also divisible by a1 ; these two facts together

is divisible by
prove that |X| is divisible by a1 .
XPS
Comment 1. With suitable sets of weights, the same method can be used to sum up other expressions
ř of3the segments. For example, wp1q “ 1 and wpkq “ 6pk ´ 1q for k ě 2 can be used to
on the lengths
compute |X| and to prove that this sum is divisible by a1 if a1 is a prime with a1 ě n ` 3. See
XPS
also Comment 2 after the second solution.
Solution 2. The conventions from the first paragraph of the first solution are still in force.
We shall prove the following more general statement:
p‘q Let p denote a prime number, let p “ a1 ă a2 ă ¨ ¨ ¨ ă an be n pairwise

coprime positive integers, and let d be an integer with 1 ď d ď p ´ n. Mark all
integers that are divisible by at least one of the numbers a1 , . . . , an on the interval
I “ r0, a1a2 ¨ ¨ ¨ an s of the real line. These points split I into a number of smaller
k ` ˘
ř bi
segments, say of lengths b1 , . . . , bk . Then the sum d
is divisible by p.
i“1
Applying p‘q to d “ 1 and d “ 2 and using the equation x2 “ 2 x2 ` x1 , one easily gets
`˘ `˘
the statement of the problem.

To prove p‘q itself, we argue by induction
`p ˘ on n. The base case n “ 1 follows from the
known fact that the binomial coefficient d is divisible by p whenever 1 ď d ď p ´ 1.
Let us now assume that n ě 2, and that the statement is known whenever n ´ 1 rather
than n coprime integers are given together with some integer d P r1, p ´ n ` 1s. Suppose that
śn´1
the numbers p “ a1 ă a2 ă ¨ ¨ ¨ ă an and d are as above. Write A1 “ i“1 ai and A “ A1 an .
Mark the points on the real axis divisible by one of the numbers a1 , . . . , an´1 green and those
divisible by an red. The green points divide r0, A1 s into certain sub-intervals, say J1 , J2 , . . . ,
and Jℓ .
To translate intervals we use the notation ra, bs ` m “ ra ` m, b ` ms whenever a, b, m P Z.
For each i P t1, 2, . . . , ℓu let Fi be the family of intervals into which the red points partition
the intervals Ji , Ji ` A1 , . . . , and Ji ` pan ´ 1qA1 . We are to prove that
ℓ ÿ ˆ ˙
ÿ |X|
i“1 XPF
d
i
is divisible by p.
Let us fix any index i with 1 ď i ď ℓ for a while. Since the numbers A1 and an are coprime
by hypothesis, the numbers 0, A1 , . . . , pan ´ 1qA1 form a complete system of residues modulo an .
Moreover, we have |Ji | ď p ă an , as in particular all multiples of p are green. So each of the
intervals Ji , Ji ` A1 , . . . , and Ji ` pan ´ 1qA1 contains at most one red point. More precisely,
for each j P t1, . . . , |Ji | ´ 1u there is exactly one amongst those intervals containing a red point
splitting it into an interval of length j followed by an interval of length |Ji | ´ j, while the
remaining an ´ |Ji | ` 1 such intervals have no red points in their interiors. For these reasons
ÿ ˆ|X|˙ ˆˆ ˙
1
ˆ
|Ji | ´ 1
˙˙ ˆ ˙
|Ji |
“2 ` ¨¨¨` ` pan ´ |Ji | ` 1q
XPFi
d d d d
ˆ ˙ ˆ ˙ ˆ ˙
|Ji | |Ji | |Ji |
“2 ` pan ´ d ` 1q ´ pd ` 1q
d`1 d d`1
ˆ ˙ ˆ ˙
|Ji | |Ji |
“ p1 ´ dq ` pan ´ d ` 1q .
d`1 d
So it remains to prove that

ℓ ˆ ˙ ℓ ˆ ˙
ÿ |Ji | ÿ |Ji |
p1 ´ dq ` pan ´ d ` 1q
i“1
d`1 i“1
d
is divisible by p. By the induction hypothesis, however, it is even true that both summands
are divisible by p, for 1 ď d ă d ` 1 ď p ´ pn ´ 1q. This completes the proof of p‘q and hence
the solution of the problem.
Comment 2. `The ˘statement p‘q can also be proved by the method of the first solution, using the
weights wpxq “ x´2
d´2 .
This page is intentionally left blank

N7. Let c ě 1 be an integer. Define a sequence of positive integers by a1 “ c and

an`1 “ a3n ´ 4c ¨ a2n ` 5c2 ¨ an ` c
for all n ě 1. Prove that for each integer n ě 2 there exists a prime number p dividing an but
none of the numbers a1 , . . . , an´1 .
(Austria)
Solution. Let us define x0 “ 0 and xn “ an {c for all integers n ě 1. It is easy to see that the
sequence pxn q thus obtained obeys the recursive law
xn`1 “ c2 px3n ´ 4x2n ` 5xn q ` 1 (1)
for all integers n ě 0. In particular, all of its terms are positive integers; notice that x1 “ 1
and x2 “ 2c2 ` 1. Since
xn`1 “ c2 xn pxn ´ 2q2 ` c2 xn ` 1 ą xn (2)
holds for all integers n ě 0, it is also strictly increasing. Since xn`1 is by (1) coprime to c for
any n ě 0, it suffices to prove that for each n ě 2 there exists a prime number p dividing xn
but none of the numbers x1 , . . . , xn´1 . Let us begin by establishing three preliminary claims.
Claim 1. If i ” j pmod mq holds for some integers i, j ě 0 and m ě 1, then xi ” xj pmod xm q
holds as well.
Proof. Evidently, it suffices to show xi`m ” xi pmod xm q for all integers i ě 0 and m ě 1. For
this purpose we may argue for fixed m by induction on i using x0 “ 0 in the base case i “ 0.
Now, if we have xi`m ” xi pmod xm q for some integer i, then the recursive equation (1) yields
xi`m`1 ” c2 px3i`m ´ 4x2i`m ` 5xi`m q ` 1 ” c2 px3i ´ 4x2i ` 5xi q ` 1 ” xi`1 pmod xm q ,
which completes the induction. l
Claim 2. If the integers i, j ě 2 and m ě 1 satisfy i ” j pmod mq, then xi ” xj pmod x2m q
holds as well.
Proof. Again it suffices to prove xi`m ” xi pmod x2m q for all integers i ě 2 and m ě 1. As
above, we proceed for fixed m by induction on i. The induction step is again easy using (1),
but this time the base case i “ 2 requires some calculation. Set L “ 5c2 . By (1) we have
xm`1 ” Lxm ` 1 pmod x2m q, and hence
x3m`1 ´ 4x2m`1 ` 5xm`1 ” pLxm ` 1q3 ´ 4pLxm ` 1q2 ` 5pLxm ` 1q

” p3Lxm ` 1q ´ 4p2Lxm ` 1q ` 5pLxm ` 1q ” 2 pmod x2m q ,
which in turn gives indeed xm`2 ” 2c2 ` 1 ” x2 pmod x2m q. l
Claim 3. For each integer n ě 2, we have xn ą x1 ¨ x2 ¨ ¨ ¨ xn´2 .

Proof. The cases n “ 2 and n “ 3 are clear. Arguing inductively, we assume now that
the claim holds for some n ě 3. Recall that x2 ě 3, so by monotonicity and (2) we get
xn ě x3 ě x2 px2 ´ 2q2 ` x2 ` 1 ě 7. It follows that
xn`1 ą x3n ´ 4x2n ` 5xn ą 7x2n ´ 4x2n ą x2n ą xn xn´1 ,
which by the induction hypothesis yields xn`1 ą x1 ¨ x2 ¨ ¨ ¨ xn´1 , as desired. l

Now we direct our attention to the problem itself: let any integer n ě 2 be given. By Claim 3
there exists a prime number p appearing with a higher exponent in the prime factorisation of xn
than in the prime factorisation of x1 ¨ ¨ ¨ xn´2 . In particular, p | xn , and it suffices to prove that
p divides none of x1 , . . . , xn´1 .
Otherwise let k P t1, . . . , n ´ 1u be minimal such that p divides xk . Since xn´1 and xn are
coprime by (1) and x1 “ 1, we actually have 2 ď k ď n ´ 2. Write n “ qk ` r with some
integers q ě 0 and 0 ď r ă k. By Claim 1 we have xn ” xr pmod xk q, whence p | xr . Due to
the minimality of k this entails r “ 0, i.e. k | n.
Thus from Claim 2 we infer
xn ” xk pmod x2k q .
Now let α ě 1 be maximal with the property pα | xk . Then x2k is divisible by pα`1 and by our
choice of p so is xn . So by the previous congruence xk is a multiple of pα`1 as well, contrary to
our choice of α. This is the final contradiction concluding the solution.
N8. For every real number x, let }x} denote the distance between x and the nearest integer.
Prove that for every pair pa, bq of positive integers there exist an odd prime p and a positive
integer k satisfying › › › › › ›
› a › › b › ›a ` b›
› ›`› ›`›
› pk › › pk › › pk › “ 1. (1)
›
(Hungary)
1
ˇX ˇ
is an integer nearest to x, so }x} “ ˇ x ` 12 ´ xˇ. Thus we
X \ \
Solution. Notice first that x ` 2
have Z ^
1
x` “ x ˘ }x}. (2)
2
For every rational number r and every prime number p, denote by vp prq the exponent of p
in the prime factorisation of r. Recall the notation p2n ´ 1q!! for the product of all odd positive
integers not exceeding 2n ´ 1, i.e., p2n ´ 1q!! “ 1 ¨ 3 ¨ ¨ ¨ p2n ´ 1q.
Lemma. For every positive integer n and every odd prime p, we have
8 Z ^
` ˘ ÿ n 1
vp p2n ´ 1q!! “ ` .
k“1
pk 2
Proof. For every positive integer k, let us count the multiples of pk among the factors 1, 3, . . . ,
2n ´ 1. If ℓ is an arbitrary integer, the number p2ℓ ´ 1qpk is listed above if and only if
Z ^
k 1 n 1 n 1
0 ă p2ℓ ´ 1qp ď 2n ðñ ăℓď k ` ðñ 1 ď ℓ ď k ` .
2 p 2 p 2
Hence, the number of multiples of pk among the factors is precisely mk “ pnk ` 21 . Thus we
X \
obtain
n n vp p2i´1q 8 ÿmk 8 Z ^
` ˘ ÿ ÿ ÿ ÿ ÿ n 1
vp p2n ´ 1q!! “ vp p2i ´ 1q “ 1“ 1“ ` . l
i“1 i“1 k“1 k“1 ℓ“1 k“1
pk 2
In order to prove the problem statement, consider the rational number

p2a ` 2b ´ 1q!! p2a ` 1qp2a ` 3q ¨ ¨ ¨ p2a ` 2b ´ 1q
N“ “ .
p2a ´ 1q!! ¨ p2b ´ 1q!! 1 ¨ 3 ¨ ¨ ¨ p2b ´ 1q
Obviously, N ą 1, so there exists a prime p with vp pNq ą 0. Since N is a fraction of two odd
numbers, p is odd.
By our lemma,
8 ˆZ ^ Z ^ Z ^˙
ÿ a`b 1 a 1 b 1
0 ă vp pNq “ ` ´ k` ´ k` .
k“1
pk 2 p 2 p 2
Therefore, there exists some positive integer k such that the integer number
Z ^ Z ^ Z ^
a`b 1 a 1 b 1
dk “ ` ´ k` ´ k`
pk 2 p 2 p 2
is positive, so dk ě 1. By (2) we have

› › › › › › › › › › › ›
a`b a b ›a ` b› › a › › b › ›a ` b› › a › › b ›
1 ď dk “ k ´ k ´ k ˘ › k › ˘ › k › ˘ › k › “ ˘ › k ›› ˘ ›› k ›› ˘ ›› k ›› .
› › › › › › › (3)
p p p p p p p p p
Since }x} ă 21 for every rational x with odd denominator, the relation (3) can only be satisfied
if all three signs on the right-hand side are positive and dk “ 1. Thus we get
› › › › › ›
› a › › b › ›a ` b›
› ›`› ›`›
› pk › › pk › › pk › “ dk “ 1,
›
as required.
Comment 1. There are various choices for the number N in the solution. Here we sketch such a
version.
Let x and y be two rational numbers with odd denominators. It is easy to see that the condi-
tion }x} ` }y} ` }x ` y} “ 1 is satisfied if and only if
either txu ă 21 , tyu ă 12 , tx ` yu ą 12 , or txu ą 21 , tyu ą 12 , tx ` yu ă 12 ,
where txu denotes the fractional part of x.
In the context of our problem, the first condition seems easier to deal with. Also, one may notice
that
txu ă 12 ðñ κpxq “ 0 and txu ě 12 ðñ κpxq “ 1, (4)
where
κpxq “ t2xu ´ 2txu.
Now it is natural to consider the number
ˆ ˙
2a ` 2b
a`b
M “ ˆ ˙ˆ ˙ ,
2a 2b
a b
since ˜ ˆ ˆ ˙¸
8 ˙ ˆ ˙
ÿ 2pa ` bq 2a 2b
vp pM q “ κ k
´κ k
´κ .
k“1
p p pk
One may see that M ą 1, and that v2 pM q ď 0. Thus, there exist an odd prime p and a positive
integer k with ˆ ˙ ˆ ˙ ˆ ˙
2pa ` bq 2a 2b
κ k
´κ k
´κ ą 0.
p p pk
In view of (4), the last inequality yields
" * " * " *
a 1 b 1 a`b 1
k
ă , k
ă , and ą , (5)
p 2 p 2 pk 2
which is what we wanted to obtain.
Comment 2. Once one tries to prove the existence of suitable p and k satisfying (5), it seems somehow
natural to suppose that a ď b and to add the restriction pk ą a. In this case the inequalities (5) can
be rewritten as
2a ă pk , 2mpk ă 2b ă p2m ` 1qpk , and p2m ` 1qpk ă 2pa ` bq ă p2m ` 2qpk
for some positive integer m. This means exactly that one of the numbers 2a ` 1, 2a ` 3, . . . , 2a ` 2b ´ 1
is divisible by some number of the form pk which is greater than 2a.
Using more advanced techniques, one can show that such a number pk exists even with k “ 1.
This was shown in 2004 by Laishram and Shorey; the methods used for this proof are elementary
but still quite involved. In fact, their result generalises a theorem by Sylvester which states that
for every pair of integers pn, kq with n ě k ě 1, the product pn ` 1qpn ` 2q ¨ ¨ ¨ pn ` kq is divisible by
some prime p ą k. We would like to mention here that Sylvester’s theorem itself does not seem to
suffice for solving the problem.

Problems Short List: With Solutions

Uploaded by

Copyright:

Available Formats

Problems Short List: With Solutions

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Problems Short List: With Solutions

Uploaded by

Copyright:

Available Formats

Problems short list

55th International Mathematical Olympiad

The shortlisted problems should be kept

Australia, Austria, Belgium, Benin, Bulgaria, Colombia, Croatia,

Problem Selection Committee

|y 2 ´ P pxq| ď 2 |x| if and only if |x2 ´ P pyq| ď 2 |y| .

Determine all possible values of P p0q.

=SHC ´ =BSC “ 90˝ , =T HC ´ =DT C “ 90˝ .

nzn`1 ´ pz0 ` z1 ` ¨ ¨ ¨ ` zn q “ pn ` 1qzn`1 ´ pz0 ` z1 ` ¨ ¨ ¨ ` zn ` zn`1 q “ ´dn`1 ,

By its definition the sequence d1 , d2 , . . . consists of integers and we have

A2. Define the function f : p0, 1q Ñ p0, 1q by

holds if and only if an´1 and bn´1 lie in distinct subintervals.

dk`2 ě dk`1p1 ` dk`1 q ě dk`1p1 ` d0 q “ dk p1 ` d0 q.

In either case, dk`2 ě dk p1 ` d0 q, and inductively we get

A3. For a sequence x1 , x2 , . . . , xn of real numbers, we define its price as

M “ max |xi |, S “ |x1 ` ¨ ¨ ¨ ` xn |, and N “ maxtM, Su.

Thus, the induction step is established.

A4. Determine all functions f : Z Ñ Z satisfying

for all integers m and n.

Answer. There is only one such function, namely n ÞÝÑ 2n ` 1007.

for all m, n P Z. By induction in both directions it follows that

f p3k mq ´ β “ p1 ` αqk f pmq ´ β

for all integers k ě 0 and m.

f pn ` tdq “ f pnq ` α ¨ td (5)

f p3k mq “ f pmq ` αp3k ´ 1qm

A“2 and β “C. (7)

for all n P Z. In particular, for n “ 0, D, 2D we obtain f pDq “ 2C ` D, f p2Dq “ f pDq` 2C “ 4C ` D,

i.e., f pr ` 2Cq “ f prq ` 4C. By induction in both directions it follows that

f pn ` 2Ctq “ f pnq ` 4Ct (9)

holds for all n, t P Z.

In view of (3) this tells us indeed that α “ 2.

Determine all possible values of P p0q.

Answer. The set of possible values of P p0q is p´8, 0q Y t1u.

ðñ 0 ď py ´ x ` 1qpy ` x ´ 1qpx ` 1 ´ yqpx ` 1 ` yq

Since this inequality is symmetric in x and y, we are done.

Second step: We prove that P ptq ą 0 for all t P R.

|y 2 ´ P pηq| ď 2 ðñ |1 ´ P pyq| ď 2 |y| ðñ |y 2 ´ P p´ηq| ď 2 .

and a real E ą 0, and setting

P pxq “ ´ Apxq2 ` Bpxq2 ` Cpxq2 ` Dpxq2 ` E

for each of them.

A6. Find all functions f : Z Ñ Z such that

Answer. The possibilities are:

• f pnq “ n ` 1 for all n;

n2 ` 4f pnq “ n2 ` 4n ` 4 “ pn ` 2q2 “ f pn ` 1q2 “ f pf pnqq2 .

a2k ` 4ak`1 “ a2k`2 .

a23 “ a21 ` 4a2 “ pr 2 ` rq2 ` 8r ` 4.

4r 2 ě 16|r| ě 12|r| ` 16 ą 8|r| ` 4 ` 4|r| ` 4 ě |8r ` 4| ´ 4r ` 4,

• r “ ´3, thus a1 “ 6 and a2 “ ´5. For each k ě 1, we have

• r “ 0, thus a1 “ 0 and a2 “ 1. Then a23 “ a21 ` 4a2 “ 4, so a3 “ ˘2. This, however,

• r “ 1, thus a1 “ 2 and a2 “ 3. We prove by induction that ak “ k ` 1 for all k ě 0

a2k`1 “ a2k´1 ` 4ak “ k 2 ` 4k ` 4 “ pk ` 2q2 ,

so ak`1 “ ˘pk ` 2q. If ak`1 “ ´pk ` 2q, then

a2k`2 “ a2k ` 4ak`1 “ pk ` 1q2 ´ 4k ´ 8 “ k 2 ´ 2k ´ 7 “ pk ´ 1q2 ´ 8.

a27 “ a25 ` 4a6 “ 36 ˘ 4,

Step 2: We prove that either f p0q “ 1, or f p0q “ 0 and f pnq ‰ 0 for n ‰ 0.

which yields f p0q “ 1.

Step 3: We discuss the values of f pnq for n ă 0.

which gives us f p´nq “ ˘n ` 1, or

Combining everything we know, we find the solutions as stated in the answer: