Imo2009

International
Mathematical
Olympiad
Bremen Germany 2009
th
Bremen
th
Problem Shortlist
with Solutions
The Problem Selection Committee

The Note of Confidentiality IMPORTANT
We insistently ask everybody to consider the following IMO Regulations rule:
These Shortlist Problems
have to be kept strictly confidential
until IMO 2010.
The Problem Selection Committee
Konrad Engel, Karl Fegert, Andreas Felgenhauer, Hans-Dietrich Gronau,
Roger Labahn, Bernd Mulansky, Jürgen Prestin, Christian Reiher,
Peter Scholze, Eckard Specht, Robert Strich, Martin Welk
gratefully received
132 problem proposals submitted by 39 countries:
Algeria, Australia, Austria, Belarus, Belgium, Bulgaria, Colombia,
Croatia, Czech Republic, El Salvador, Estonia, Finland, France,
Greece, Hong Kong, Hungary, India, Ireland, Islamic Republic of
Iran, Japan, Democratic People’s Republic of Korea, Lithuania,
Luxembourg, The former Yugoslav Republic of Macedonia,
Mongolia, Netherlands, New Zealand, Pakistan, Peru, Poland,
Romania, Russian Federation, Slovenia, South Africa, Taiwan,
Turkey, Ukraine, United Kingdom, United States of America .
Layout: Roger Labahn with LATEX & TEX
Drawings: Eckard Specht with nicefig 2.0

: The Problem Selection Committee
Contents
Problem Shortlist 4
Algebra 12
Combinatorics 26
Geometry 47
Number Theory 69

Algebra Problem Shortlist 50th IMO 2009
Algebra
A1 CZE (Czech Republic)
Find the largest possible integer k, such that the following statement is true:
Let 2009 arbitrary non-degenerated triangles be given. In every triangle the three sides are
colored, such that one is blue, one is red and one is white. Now, for every color separately, let
us sort the lengths of the sides. We obtain
b1 ≤ b2 ≤ . . . ≤ b2009 the lengths of the blue sides,
r1 ≤ r2 ≤ . . . ≤ r2009 the lengths of the red sides,
and w1 ≤ w2 ≤ . . . ≤ w2009 the lengths of the white sides.
Then there exist k indices j such that we can form a non-degenerated triangle with side lengths
bj, rj, wj.
A2 EST (Estonia)
Let a, b, c be positive real numbers such that
1
a
+
1
b
+
1
c
= a + b + c. Prove that
1
(2a + b + c)2
+
1
(2b + c + a)2
+
1
(2c + a + b)2
≤
3
16
.
A3 FRA (France)
Determine all functions f from the set of positive integers into the set of positive integers such
that for all x and y there exists a non degenerated triangle with sides of lengths
x, f(y) and f(y + f(x) − 1).
A4 BLR (Belarus)
Let a, b, c be positive real numbers such that ab + bc + ca ≤ 3abc. Prove that
a2 + b2
a + b
+
b2 + c2
b + c
+
c2 + a2
c + a
+ 3 ≤
√
2
√
a + b +
√
b + c +
√
c + a .
A5 BLR (Belarus)
Let f be any function that maps the set of real numbers into the set of real numbers. Prove
that there exist real numbers x and y such that
f (x − f(y)) > yf(x) + x.
4

50th IMO 2009 Problem Shortlist Algebra
A6 USA (United States of America)
Suppose that s1, s2, s3, . . . is a strictly increasing sequence of positive integers such that the
subsequences
ss1 , ss2 , ss3 , . . . and ss1+1, ss2+1, ss3+1, . . .
are both arithmetic progressions. Prove that s1, s2, s3, . . . is itself an arithmetic progression.
A7 JPN (Japan)
Find all functions f from the set of real numbers into the set of real numbers which satisfy for
all real x, y the identity
f(xf(x + y)) = f(yf(x)) + x2
.
5

Combinatorics Problem Shortlist 50th IMO 2009
Combinatorics
C1 NZL (New Zealand)
Consider 2009 cards, each having one gold side and one black side, lying in parallel on a long
table. Initially all cards show their gold sides. Two players, standing by the same long side of
the table, play a game with alternating moves. Each move consists of choosing a block of 50
consecutive cards, the leftmost of which is showing gold, and turning them all over, so those
which showed gold now show black and vice versa. The last player who can make a legal move
wins.
(a) Does the game necessarily end?
(b) Does there exist a winning strategy for the starting player?
C2 ROU (Romania)
For any integer n ≥ 2, let N(n) be the maximal number of triples (ai, bi, ci), i = 1, . . . , N(n),
consisting of nonnegative integers ai, bi and ci such that the following two conditions are satis-
ﬁed:
(1) ai + bi + ci = n for all i = 1, . . . , N(n),
(2) If i = j, then ai = aj, bi = bj and ci = cj.
Determine N(n) for all n ≥ 2.
Comment. The original problem was formulated for m-tuples instead for triples. The numbers
N(m, n) are then deﬁned similarly to N(n) in the case m = 3. The numbers N(3, n) and
N(n, n) should be determined. The case m = 3 is the same as in the present problem. The
upper bound for N(n, n) can be proved by a simple generalization. The construction of a set
of triples attaining the bound can be easily done by induction from n to n + 2.
C3 RUS (Russian Federation)
Let n be a positive integer. Given a sequence ε1, . . . , εn−1 with εi = 0 or εi = 1 for each
i = 1, . . . , n − 1, the sequences a0, . . . , an and b0, . . . , bn are constructed by the following rules:
a0 = b0 = 1, a1 = b1 = 7,
ai+1 =
2ai−1 + 3ai, if εi = 0,
3ai−1 + ai, if εi = 1,
for each i = 1, . . . , n − 1,
bi+1 =
2bi−1 + 3bi, if εn−i = 0,
3bi−1 + bi, if εn−i = 1,
for each i = 1, . . . , n − 1.
Prove that an = bn.
C4 NLD (Netherlands)
For an integer m ≥ 1, we consider partitions of a 2m
× 2m
chessboard into rectangles consisting
of cells of the chessboard, in which each of the 2m
cells along one diagonal forms a separate
rectangle of side length 1. Determine the smallest possible sum of rectangle perimeters in such
a partition.
6

50th IMO 2009 Problem Shortlist Combinatorics
Five identical empty buckets of 2-liter capacity stand at the vertices of a regular pentagon.
Cinderella and her wicked Stepmother go through a sequence of rounds: At the beginning of
every round, the Stepmother takes one liter of water from the nearby river and distributes it
arbitrarily over the five buckets. Then Cinderella chooses a pair of neighboring buckets, empties
them into the river, and puts them back. Then the next round begins. The Stepmother’s goal
is to make one of these buckets overflow. Cinderella’s goal is to prevent this. Can the wicked
Stepmother enforce a bucket overflow?
C6 BGR (Bulgaria)
On a 999×999 board a limp rook can move in the following way: From any square it can move
to any of its adjacent squares, i.e. a square having a common side with it, and every move
must be a turn, i.e. the directions of any two consecutive moves must be perpendicular. A non-
intersecting route of the limp rook consists of a sequence of pairwise different squares that the
limp rook can visit in that order by an admissible sequence of moves. Such a non-intersecting
route is called cyclic, if the limp rook can, after reaching the last square of the route, move
directly to the first square of the route and start over.
How many squares does the longest possible cyclic, non-intersecting route of a limp rook
visit?
Variant 1. A grasshopper jumps along the real axis. He starts at point 0 and makes 2009
jumps to the right with lengths 1, 2, . . . , 2009 in an arbitrary order. Let M be a set of 2008
positive integers less than 1005 · 2009. Prove that the grasshopper can arrange his jumps in
such a way that he never lands on a point from M.
Variant 2. Let n be a nonnegative integer. A grasshopper jumps along the real axis. He starts
at point 0 and makes n + 1 jumps to the right with pairwise different positive integral lengths
a1, a2, . . . , an+1 in an arbitrary order. Let M be a set of n positive integers in the interval (0, s),
where s = a1 + a2 + · · · + an+1. Prove that the grasshopper can arrange his jumps in such a
way that he never lands on a point from M.
C8 AUT (Austria)
For any integer n ≥ 2, we compute the integer h(n) by applying the following procedure to its
decimal representation. Let r be the rightmost digit of n.
(1) If r = 0, then the decimal representation of h(n) results from the decimal representation
of n by removing this rightmost digit 0.
(2) If 1 ≤ r ≤ 9 we split the decimal representation of n into a maximal right part R that
solely consists of digits not less than r and into a left part L that either is empty or ends
with a digit strictly smaller than r. Then the decimal representation of h(n) consists of the
decimal representation of L, followed by two copies of the decimal representation of R − 1.
For instance, for the number n = 17,151,345,543, we will have L = 17,151, R = 345,543
and h(n) = 17,151,345,542,345,542.
Prove that, starting with an arbitrary integer n ≥ 2, iterated application of h produces the
integer 1 after finitely many steps.
7

Geometry Problem Shortlist 50th IMO 2009
Geometry
G1 BEL (Belgium)
Let ABC be a triangle with AB = AC. The angle bisectors of A and B meet the sides BC
and AC in D and E, respectively. Let K be the incenter of triangle ADC. Suppose that
∠BEK = 45◦
. Find all possible values of ∠BAC.
G2 RUS (Russian Federation)
Let ABC be a triangle with circumcenter O. The points P and Q are interior points of the
sides CA and AB, respectively. The circle k passes through the midpoints of the segments BP,
CQ, and PQ. Prove that if the line PQ is tangent to circle k then OP = OQ.
G3 IRN (Islamic Republic of Iran)
Let ABC be a triangle. The incircle of ABC touches the sides AB and AC at the points Z
and Y , respectively. Let G be the point where the lines BY and CZ meet, and let R and S be
points such that the two quadrilaterals BCY R and BCSZ are parallelograms.
Prove that GR = GS.
G4 UNK (United Kingdom)
Given a cyclic quadrilateral ABCD, let the diagonals AC and BD meet at E and the lines AD
and BC meet at F. The midpoints of AB and CD are G and H, respectively. Show that EF
is tangent at E to the circle through the points E, G, and H.
G5 POL (Poland)
Let P be a polygon that is convex and symmetric to some point O. Prove that for some
parallelogram R satisfying P ⊂ R we have
|R|
|P|
≤
√
2
where |R| and |P| denote the area of the sets R and P, respectively.
G6 UKR (Ukraine)
Let the sides AD and BC of the quadrilateral ABCD (such that AB is not parallel to CD)
intersect at point P. Points O1 and O2 are the circumcenters and points H1 and H2 are the
orthocenters of triangles ABP and DCP, respectively. Denote the midpoints of segments
O1H1 and O2H2 by E1 and E2, respectively. Prove that the perpendicular from E1 on CD, the
perpendicular from E2 on AB and the line H1H2 are concurrent.
Let ABC be a triangle with incenter I and let X, Y and Z be the incenters of the triangles
BIC, CIA and AIB, respectively. Let the triangle XY Z be equilateral. Prove that ABC is
equilateral too.
8

50th IMO 2009 Problem Shortlist Geometry
G8 BGR (Bulgaria)
Let ABCD be a circumscribed quadrilateral. Let g be a line through A which meets the
segment BC in M and the line CD in N. Denote by I1, I2, and I3 the incenters of ABM,
MNC, and NDA, respectively. Show that the orthocenter of I1I2I3 lies on g.
9

Number Theory Problem Shortlist 50th IMO 2009
Number Theory
N1 AUS (Australia)
A social club has n members. They have the membership numbers 1, 2, . . . , n, respectively.
From time to time members send presents to other members, including items they have already
received as presents from other members. In order to avoid the embarrassing situation that a
member might receive a present that he or she has sent to other members, the club adds the
following rule to its statutes at one of its annual general meetings:
“A member with membership number a is permitted to send a present to a member with
membership number b if and only if a(b − 1) is a multiple of n.”
Prove that, if each member follows this rule, none will receive a present from another member
that he or she has already sent to other members.
Alternative formulation: Let G be a directed graph with n vertices v1, v2, . . . , vn, such that
there is an edge going from va to vb if and only if a and b are distinct and a(b − 1) is a multiple
of n. Prove that this graph does not contain a directed cycle.
N2 PER (Peru)
A positive integer N is called balanced, if N = 1 or if N can be written as a product of an
even number of not necessarily distinct primes. Given positive integers a and b, consider the
polynomial P defined by P(x) = (x + a)(x + b).
(a) Prove that there exist distinct positive integers a and b such that all the numbers P(1), P(2),
. . . , P(50) are balanced.
(b) Prove that if P(n) is balanced for all positive integers n, then a = b.
N3 EST (Estonia)
Let f be a non-constant function from the set of positive integers into the set of positive integers,
such that a − b divides f(a) − f(b) for all distinct positive integers a, b. Prove that there exist
infinitely many primes p such that p divides f(c) for some positive integer c.
N4 PRK (Democratic People’s Republic of Korea)
Find all positive integers n such that there exists a sequence of positive integers a1, a2, . . . , an
satisfying
ak+1 =
a2
k + 1
ak−1 + 1
− 1
for every k with 2 ≤ k ≤ n − 1.
N5 HUN (Hungary)
Let P(x) be a non-constant polynomial with integer coefficients. Prove that there is no function
T from the set of integers into the set of integers such that the number of integers x with
Tn
(x) = x is equal to P(n) for every n ≥ 1, where Tn
denotes the n-fold application of T.
10

50th IMO 2009 Problem Shortlist Number Theory
N6 TUR (Turkey)
Let k be a positive integer. Show that if there exists a sequence a0, a1, . . . of integers satisfying
the condition
an =
an−1 + nk
n
for all n ≥ 1,
then k − 2 is divisible by 3.
N7 MNG (Mongolia)
Let a and b be distinct integers greater than 1. Prove that there exists a positive integer n such
that (an
− 1)(bn
− 1) is not a perfect square.
11

A1 Algebra 50th IMO 2009
Algebra
A1 CZE (Czech Republic)
Find the largest possible integer k, such that the following statement is true:
Let 2009 arbitrary non-degenerated triangles be given. In every triangle the three sides are
colored, such that one is blue, one is red and one is white. Now, for every color separately, let
us sort the lengths of the sides. We obtain
b1 ≤ b2 ≤ . . . ≤ b2009 the lengths of the blue sides,
r1 ≤ r2 ≤ . . . ≤ r2009 the lengths of the red sides,
and w1 ≤ w2 ≤ . . . ≤ w2009 the lengths of the white sides.
Then there exist k indices j such that we can form a non-degenerated triangle with side lengths
bj, rj, wj.
Solution. We will prove that the largest possible number k of indices satisfying the given
condition is one.
Firstly we prove that b2009, r2009, w2009 are always lengths of the sides of a triangle. Without
loss of generality we may assume that w2009 ≥ r2009 ≥ b2009. We show that the inequality
b2009 + r2009 > w2009 holds. Evidently, there exists a triangle with side lengths w, b, r for the
white, blue and red side, respectively, such that w2009 = w. By the conditions of the problem
we have b + r > w, b2009 ≥ b and r2009 ≥ r. From these inequalities it follows
b2009 + r2009 ≥ b + r > w = w2009.
Secondly we will describe a sequence of triangles for which wj, bj, rj with j < 2009 are not the
lengths of the sides of a triangle. Let us deﬁne the sequence ∆j, j = 1, 2, . . . , 2009, of triangles,
where ∆j has
a blue side of length 2j,
a red side of length j for all j ≤ 2008 and 4018 for j = 2009,
and a white side of length j + 1 for all j ≤ 2007, 4018 for j = 2008 and 1 for j = 2009.
Since
(j + 1) + j > 2j ≥ j + 1> j, if j ≤ 2007,
2j + j > 4018 > 2j > j, if j = 2008,
4018 + 1 > 2j = 4018 > 1, if j = 2009,
such a sequence of triangles exists. Moreover, wj = j, rj = j and bj = 2j for 1 ≤ j ≤ 2008.
Then
wj + rj = j + j = 2j = bj,
i.e., bj, rj and wj are not the lengths of the sides of a triangle for 1 ≤ j ≤ 2008.
12

50th IMO 2009 Algebra A2
A2 EST (Estonia)
Let a, b, c be positive real numbers such that
1
a
+
1
b
+
1
c
= a + b + c. Prove that
1
(2a + b + c)2
+
1
(2b + c + a)2
+
1
(2c + a + b)2
≤
3
16
.
Solution 1. For positive real numbers x, y, z, from the arithmetic-geometric-mean inequality,
2x + y + z = (x + y) + (x + z) ≥ 2 (x + y)(x + z),
we obtain
1
(2x + y + z)2
≤
1
4(x + y)(x + z)
.
Applying this to the left-hand side terms of the inequality to prove, we get
1
(2a + b + c)2
+
1
(2b + c + a)2
+
1
(2c + a + b)2
≤
1
4(a + b)(a + c)
+
1
4(b + c)(b + a)
+
1
4(c + a)(c + b)
=
(b + c) + (c + a) + (a + b)
4(a + b)(b + c)(c + a)
=
a + b + c
2(a + b)(b + c)(c + a)
. (1)
A second application of the inequality of the arithmetic-geometric mean yields
a2
b + a2
c + b2
a + b2
c + c2
a + c2
b ≥ 6abc,
or, equivalently,
9(a + b)(b + c)(c + a) ≥ 8(a + b + c)(ab + bc + ca). (2)
The supposition 1
a
+ 1
b
+ 1
c
= a + b + c can be written as
ab + bc + ca = abc(a + b + c). (3)
Applying the arithmetic-geometric-mean inequality x2
y2
+ x2
z2
≥ 2x2
yz thrice, we get
a2
b2
+ b2
c2
+ c2
a2
≥ a2
bc + ab2
c + abc2
,
which is equivalent to
(ab + bc + ca)2
≥ 3abc(a + b + c). (4)
13

Combining (1), (2), (3), and (4), we will ﬁnish the proof:
a + b + c
2(a + b)(b + c)(c + a)
=
(a + b + c)(ab + bc + ca)
2(a + b)(b + c)(c + a)
·
ab + bc + ca
abc(a + b + c)
·
abc(a + b + c)
(ab + bc + ca)2
≤
9
2 · 8
· 1 ·
1
3
=
3
16
.
Solution 2. Equivalently, we prove the homogenized inequality
(a + b + c)2
(2a + b + c)2
+
(a + b + c)2
(a + 2b + c)2
+
(a + b + c)2
(a + b + 2c)2
≤
3
16
(a + b + c)
1
a
+
1
b
+
1
c
for all positive real numbers a, b, c. Without loss of generality we choose a + b + c = 1. Thus,
the problem is equivalent to prove for all a, b, c > 0, fulﬁlling this condition, the inequality
1
(1 + a)2
+
1
(1 + b)2
+
1
(1 + c)2
≤
3
16
1
a
+
1
b
+
1
c
. (5)
Applying Jensen’s inequality to the function f(x) =
x
(1 + x)2
, which is concave for 0 ≤ x ≤ 2
and increasing for 0 ≤ x ≤ 1, we obtain
α
a
(1 + a)2
+ β
b
(1 + b)2
+ γ
c
(1 + c)2
≤ (α + β + γ)
A
(1 + A)2
, where A =
αa + βb + γc
α + β + γ
.
Choosing α =
1
a
, β =
1
b
, and γ =
1
c
, we can apply the harmonic-arithmetic-mean inequality
A =
3
1
a
+ 1
b
+ 1
c
≤
a + b + c
3
=
1
3
< 1.
Finally we prove (5):
1
(1 + a)2
+
1
(1 + b)2
+
1
(1 + c)2
≤
1
a
+
1
b
+
1
c
A
(1 + A)2
≤
1
a
+
1
b
+
1
c
1
3
1 + 1
3
2 =
3
16
1
a
+
1
b
+
1
c
.
14

A3 FRA (France)
Determine all functions f from the set of positive integers into the set of positive integers such
that for all x and y there exists a non degenerated triangle with sides of lengths
x, f(y) and f(y + f(x) − 1).
Solution. The identity function f(x) = x is the only solution of the problem.
If f(x) = x for all positive integers x, the given three lengths are x, y = f(y) and z =
f (y + f(x) − 1) = x + y − 1. Because of x ≥ 1, y ≥ 1 we have z ≥ max{x, y} > |x − y| and
z < x + y. From this it follows that a triangle with these side lengths exists and does not
degenerate. We prove in several steps that there is no other solution.
Step 1. We show f(1) = 1.
If we had f(1) = 1+m > 1 we would conclude f(y) = f(y+m) for all y considering the triangle
with the side lengths 1, f(y) and f(y + m). Thus, f would be m-periodic and, consequently,
bounded. Let B be a bound, f(x) ≤ B. If we choose x > 2B we obtain the contradiction
x > 2B ≥ f(y) + f(y + f(x) − 1).
Step 2. For all positive integers z, we have f(f(z)) = z.
Setting x = z and y = 1 this follows immediately from Step 1.
Step 3. For all integers z ≥ 1, we have f(z) ≤ z.
Let us show, that the contrary leads to a contradiction. Assume w + 1 = f(z) > z for some
z. From Step 1 we know that w ≥ z ≥ 2. Let M = max{f(1), f(2), . . . , f(w)} be the largest
value of f for the ﬁrst w integers. First we show, that no positive integer t exists with
f(t) >
z − 1
w
· t + M, (1)
otherwise we decompose the smallest value t as t = wr +s where r is an integer and 1 ≤ s ≤ w.
Because of the deﬁnition of M, we have t > w. Setting x = z and y = t − w we get from the
triangle inequality
z + f(t − w) > f((t − w) + f(z) − 1) = f(t − w + w) = f(t).
Hence,
f(t − w) ≥ f(t) − (z − 1) >
z − 1
w
(t − w) + M,
a contradiction to the minimality of t.
Therefore the inequality (1) fails for all t ≥ 1, we have proven
f(t) ≤
z − 1
w
· t + M, (2)
instead.
15

Now, using (2), we finish the proof of Step 3. Because of z ≤ w we have
z − 1
w
< 1 and we can
choose an integer t sufficiently large to fulfill the condition
z − 1
w
2
t +
z − 1
w
+ 1 M < t.
Applying (2) twice we get
f (f(t)) ≤
z − 1
w
f(t) + M ≤
z − 1
w
z − 1
w
t + M + M < t
in contradiction to Step 2, which proves Step 3.
Final step. Thus, following Step 2 and Step 3, we obtain
z = f(f(z)) ≤ f(z) ≤ z
and f(z) = z for all positive integers z is proven.
16

A4 BLR (Belarus)
Let a, b, c be positive real numbers such that ab + bc + ca ≤ 3abc. Prove that
a2 + b2
a + b
+
b2 + c2
b + c
+
c2 + a2
c + a
+ 3 ≤
√
2
√
a + b +
√
b + c +
√
c + a .
Solution. Starting with the terms of the right-hand side, the quadratic-arithmetic-mean in-
equality yields
√
2
√
a + b = 2
ab
a + b
1
2
2 +
a2 + b2
ab
≥ 2
ab
a + b
·
1
2
√
2 +
a2 + b2
ab
=
2ab
a + b
+
a2 + b2
a + b
and, analogously,
√
2
√
b + c ≥
2bc
b + c
+
b2 + c2
b + c
,
√
2
√
c + a ≥
2ca
c + a
+
c2 + a2
c + a
.
Applying the inequality between the arithmetic mean and the squared harmonic mean will
ﬁnish the proof:
2ab
a + b
+
2bc
b + c
+
2ca
c + a
≥ 3 ·
3
a+b
2ab
2
+ b+c
2bc
2
+ c+a
2ca
2
= 3 ·
3abc
ab + bc + ca
≥ 3.
17

A5 BLR (Belarus)
Let f be any function that maps the set of real numbers into the set of real numbers. Prove
that there exist real numbers x and y such that
f (x − f(y)) > yf(x) + x.
Solution 1. Assume that
f(x − f(y)) ≤ yf(x) + x for all real x, y. (1)
Let a = f(0). Setting y = 0 in (1) gives f(x − a) ≤ x for all real x and, equivalently,
f(y) ≤ y + a for all real y. (2)
Setting x = f(y) in (1) yields in view of (2)
a = f(0) ≤ yf(f(y)) + f(y) ≤ yf(f(y)) + y + a.
This implies 0 ≤ y(f(f(y)) + 1) and thus
f(f(y)) ≥ −1 for all y > 0. (3)
From (2) and (3) we obtain −1 ≤ f(f(y)) ≤ f(y) + a for all y > 0, so
f(y) ≥ −a − 1 for all y > 0. (4)
Now we show that
f(x) ≤ 0 for all real x. (5)
Assume the contrary, i.e. there is some x such that f(x) > 0. Take any y such that
y < x − a and y <
−a − x − 1
f(x)
.
Then in view of (2)
x − f(y) ≥ x − (y + a) > 0
and with (1) and (4) we obtain
yf(x) + x ≥ f(x − f(y)) ≥ −a − 1,
whence
y ≥
−a − x − 1
f(x)
contrary to our choice of y. Thereby, we have established (5).
Setting x = 0 in (5) leads to a = f(0) ≤ 0 and (2) then yields
f(x) ≤ x for all real x. (6)
Now choose y such that y > 0 and y > −f(−1) − 1 and set x = f(y) − 1. From (1), (5) and
18

(6) we obtain
f(−1) = f(x − f(y)) ≤ yf(x) + x = yf(f(y) − 1) + f(y) − 1 ≤ y(f(y) − 1) − 1 ≤ −y − 1,
i.e. y ≤ −f(−1) − 1, a contradiction to the choice of y.
Solution 2. Assume that
f(x − f(y)) ≤ yf(x) + x for all real x, y. (7)
Let a = f(0). Setting y = 0 in (7) gives f(x − a) ≤ x for all real x and, equivalently,
f(y) ≤ y + a for all real y. (8)
Now we show that
f(z) ≥ 0 for all z ≥ 1. (9)
Let z ≥ 1 be fixed, set b = f(z) and assume that b < 0. Setting x = w + b and y = z in (7)
gives
f(w) − zf(w + b) ≤ w + b for all real w. (10)
Applying (10) to w, w + b, . . . , w + (n − 1)b, where n = 1, 2, . . . , leads to
f(w) − zn
f(w + nb) = (f(w) − zf(w + b)) + z (f(w + b) − zf(w + 2b))
+ · · · + zn−1
(f(w + (n − 1)b) − zf(w + nb))
≤(w + b) + z(w + 2b) + · · · + zn−1
(w + nb).
From (8) we obtain
f(w + nb) ≤ w + nb + a
and, thus, we have for all positive integers n
f(w) ≤ (1 + z + · · · + zn−1
+ zn
)w + (1 + 2z + · · · + nzn−1
+ nzn
)b + zn
a. (11)
With w = 0 we get
a ≤ (1 + 2z + · · · + nzn−1
+ nzn
)b + azn
. (12)
In view of the assumption b < 0 we find some n such that
a > (nb + a)zn
(13)
because the right hand side tends to −∞ as n → ∞. Now (12) and (13) give the desired
contradiction and (9) is established. In addition, we have for z = 1 the strict inequality
f(1) > 0. (14)
Indeed, assume that f(1) = 0. Then setting w = −1 and z = 1 in (11) leads to
f(−1) ≤ −(n + 1) + a
which is false if n is sufficiently large.
To complete the proof we set t = min{−a, −2/f(1)}. Setting x = 1 and y = t in (7) gives
f(1 − f(t)) ≤ tf(1) + 1 ≤ −2 + 1 = −1. (15)
19

On the other hand, by (8) and the choice of t we have f(t) ≤ t + a ≤ 0 and hence 1 − f(t) ≥ 1.
The inequality (9) yields
f(1 − f(t)) ≥ 0,
which contradicts (15).
20

A6 USA (United States of America)
Suppose that s1, s2, s3, . . . is a strictly increasing sequence of positive integers such that the
subsequences
ss1 , ss2 , ss3 , . . . and ss1+1, ss2+1, ss3+1, . . .
are both arithmetic progressions. Prove that s1, s2, s3, . . . is itself an arithmetic progression.
Solution 1. Let D be the common difference of the progression ss1 , ss2 , . . . . Let for n =
1, 2, . . .
dn = sn+1 − sn.
We have to prove that dn is constant. First we show that the numbers dn are bounded. Indeed,
by supposition dn ≥ 1 for all n. Thus, we have for all n
dn = sn+1 − sn ≤ dsn + dsn+1 + · · · + dsn+1−1 = ssn+1 − ssn = D.
The boundedness implies that there exist
m = min{dn : n = 1, 2, . . . } and M = max{dn : n = 1, 2, . . . }.
It suffices to show that m = M. Assume that m < M. Choose n such that dn = m. Considering
a telescoping sum of m = dn = sn+1 − sn items not greater than M leads to
D = ssn+1 − ssn = ssn+m − ssn = dsn + dsn+1 + · · · + dsn+m−1 ≤ mM (1)
and equality holds if and only if all items of the sum are equal to M. Now choose n such that
dn = M. In the same way, considering a telescoping sum of M items not less than m we obtain
D = ssn+1 − ssn = ssn+M − ssn = dsn + dsn+1 + · · · + dsn+M−1 ≥ Mm (2)
and equality holds if and only if all items of the sum are equal to m. The inequalities (1) and
(2) imply that D = Mm and that
dsn = dsn+1 = · · · = dsn+1−1 = M if dn = m,
dsn = dsn+1 = · · · = dsn+1−1 = m if dn = M.
Hence, dn = m implies dsn = M. Note that sn ≥ s1 +(n−1) ≥ n for all n and moreover sn > n
if dn = n, because in the case sn = n we would have m = dn = dsn = M in contradiction to
the assumption m < M. In the same way dn = M implies dsn = m and sn > n. Consequently,
there is a strictly increasing sequence n1, n2, . . . such that
dsn1
= M, dsn2
= m, dsn3
= M, dsn4
= m, . . . .
The sequence ds1 , ds2 , . . . is the sequence of pairwise differences of ss1+1, ss2+1, . . . and ss1 , ss2 , . . . ,
hence also an arithmetic progression. Thus m = M.
21

Solution 2. Let the integers D and E be the common differences of the progressions ss1 , ss2 , . . .
and ss1+1, ss2+1, . . . , respectively. Let briefly A = ss1 − D and B = ss1+1 − E. Then, for all
positive integers n,
ssn = A + nD, ssn+1 = B + nE.
Since the sequence s1, s2, . . . is strictly increasing, we have for all positive integers n
ssn < ssn+1 ≤ ssn+1 ,
which implies
A + nD < B + nE ≤ A + (n + 1)D,
and thereby
0 < B − A + n(E − D) ≤ D,
which implies D − E = 0 and thus
0 ≤ B − A ≤ D. (3)
Let m = min{sn+1 − sn : n = 1, 2, . . . }. Then
B − A = (ss1+1 − E) − (ss1 − D) = ss1+1 − ss1 ≥ m (4)
and
D = A + (s1 + 1)D − (A + s1D) = sss1+1 − sss1
= sB+D − sA+D ≥ m(B − A). (5)
From (3) we consider two cases.
Case 1. B − A = D.
Then, for each positive integer n, ssn+1 = B + nD = A + (n + 1)D = ssn+1 , hence sn+1 = sn + 1
and s1, s2, . . . is an arithmetic progression with common difference 1.
Case 2. B − A < D. Choose some positive integer N such that sN+1 − sN = m. Then
m(A − B + D − 1) = m((A + (N + 1)D) − (B + ND + 1))
≤ sA+(N+1)D − sB+ND+1 = sssN+1
− sssN +1+1
= (A + sN+1D) − (B + (sN + 1)D) = (sN+1 − sN )D + A − B − D
= mD + A − B − D,
i.e.,
(B − A − m) + (D − m(B − A)) ≤ 0. (6)
The inequalities (4)-(6) imply that
B − A = m and D = m(B − A).
Assume that there is some positive integer n such that sn+1 > sn + m. Then
m(m + 1) ≤ m(sn+1 − sn) ≤ ssn+1 − ssn = (A + (n + 1)D) − (A + nD)) = D = m(B − A) = m2
,
a contradiction. Hence s1, s2, . . . is an arithmetic progression with common difference m.
22

A7 JPN (Japan)
Find all functions f from the set of real numbers into the set of real numbers which satisfy for
all real x, y the identity
f(xf(x + y)) = f(yf(x)) + x2
.
Solution 1. It is no hard to see that the two functions given by f(x) = x and f(x) = −x for
all real x respectively solve the functional equation. In the sequel, we prove that there are no
further solutions.
Let f be a function satisfying the given equation. It is clear that f cannot be a constant. Let us
first show that f(0) = 0. Suppose that f(0) = 0. For any real t, substituting (x, y) = (0, t
f(0)
)
into the given functional equation, we obtain
f(0) = f(t), (1)
contradicting the fact that f is not a constant function. Therefore, f(0) = 0. Next for any t,
substituting (x, y) = (t, 0) and (x, y) = (t, −t) into the given equation, we get
f (tf(t)) = f(0) + t2
= t2
,
and
f(tf(0)) = f(−tf(t)) + t2
,
respectively. Therefore, we conclude that
f(tf(t)) = t2
, f(−tf(t)) = −t2
, for every real t. (2)
Consequently, for every real v, there exists a real u, such that f(u) = v. We also see that if
f(t) = 0, then 0 = f(tf(t)) = t2
so that t = 0, and thus 0 is the only real number satisfying
f(t) = 0.
We next show that for any real number s,
f(−s) = −f(s). (3)
This is clear if f(s) = 0. Suppose now f(s) < 0, then we can find a number t for which
f(s) = −t2
. As t = 0 implies f(t) = 0, we can also find number a such that af(t) = s.
Substituting (x, y) = (t, a) into the given equation, we get
f(tf(t + a)) = f(af(t)) + t2
= f(s) + t2
= 0,
and therefore, tf(t + a) = 0, which implies t + a = 0, and hence s = −tf(t). Consequently,
f(−s) = f(tf(t)) = t2
= −(−t2
) = −f(s) holds in this case.
Finally, suppose f(s) > 0 holds. Then there exists a real number t = 0 for which f(s) = t2
.
Choose a number a such that tf(a) = s. Substituting (x, y) = (t, a−t) into the given equation,
we get f(s) = f(tf(a)) = f((a−t)f(t))+t2
= f((a−t)f(t))+f(s). So we have f((a−t)f(t)) = 0,
from which we conclude that (a − t)f(t) = 0. Since f(t) = 0, we get a = t so that s = tf(t)
and thus we see f(−s) = f(−tf(t)) = −t2
= −f(s) holds in this case also. This observation
finishes the proof of (3).
By substituting (x, y) = (s, t), (x, y) = (t, −s−t) and (x, y) = (−s−t, s) into the given equation,
23

we obtain
f(sf(s + t))) = f(tf(s)) + s2
,
f(tf(−s)) = f((−s − t)f(t)) + t2
,
and
f((−s − t)f(−t)) = f(sf(−s − t)) + (s + t)2
,
respectively. Using the fact that f(−x) = −f(x) holds for all x to rewrite the second and the
third equation, and rearranging the terms, we obtain
f(tf(s)) − f(sf(s + t)) = −s2
,
f(tf(s)) − f((s + t)f(t)) = −t2
,
f((s + t)f(t)) + f(sf(s + t)) = (s + t)2
.
Adding up these three equations now yields 2f(tf(s)) = 2ts, and therefore, we conclude that
f(tf(s)) = ts holds for every pair of real numbers s, t. By fixing s so that f(s) = 1, we obtain
f(x) = sx. In view of the given equation, we see that s = ±1. It is easy to check that both
functions f(x) = x and f(x) = −x satisfy the given functional equation, so these are the desired
solutions.
Solution 2. As in Solution 1 we obtain (1), (2) and (3).
Now we prove that f is injective. For this purpose, let us assume that f(r) = f(s) for some
r = s. Then, by (2)
r2
= f(rf(r)) = f(rf(s)) = f((s − r)f(r)) + r2
,
where the last statement follows from the given functional equation with x = r and y = s − r.
Hence, h = (s − r)f(r) satisfies f(h) = 0 which implies h2
= f(hf(h)) = f(0) = 0, i.e., h = 0.
Then, by s = r we have f(r) = 0 which implies r = 0, and finally f(s) = f(r) = f(0) = 0.
Analogously, it follows that s = 0 which gives the contradiction r = s.
To prove |f(1)| = 1 we apply (2) with t = 1 and also with t = f(1) and obtain f(f(1)) = 1 and
(f(1))2
= f(f(1) · f(f(1))) = f(f(1)) = 1.
Now we choose η ∈ {−1, 1} with f(1) = η. Using that f is odd and the given equation with
x = 1, y = z (second equality) and with x = −1, y = z + 2 (fourth equality) we obtain
f(z) + 2η = η(f(zη) + 2) = η(f(f(z + 1)) + 1) = η(−f(−f(z + 1)) + 1)
= −ηf((z + 2)f(−1)) = −ηf((z + 2)(−η)) = ηf((z + 2)η) = f(z + 2). (4)
Hence,
f(z + 2η) = ηf(ηz + 2) = η(f(ηz) + 2η) = f(z) + 2.
Using this argument twice we obtain
f(z + 4η) = f(z + 2η) + 2 = f(z) + 4.
Substituting z = 2f(x) we have
f(2f(x)) + 4 = f(2f(x) + 4η) = f(2f(x + 2)),
24

where the last equality follows from (4). Applying the given functional equation we proceed to
f(2f(x + 2)) = f(xf(2)) + 4 = f(2ηx) + 4
where the last equality follows again from (4) with z = 0, i.e., f(2) = 2η. Finally, f(2f(x)) =
f(2ηx) and by injectivity of f we get 2f(x) = 2ηx and hence the two solutions.
25

C1 Combinatorics 50th IMO 2009
Combinatorics
C1 NZL (New Zealand)
Consider 2009 cards, each having one gold side and one black side, lying in parallel on a long
table. Initially all cards show their gold sides. Two players, standing by the same long side of
the table, play a game with alternating moves. Each move consists of choosing a block of 50
consecutive cards, the leftmost of which is showing gold, and turning them all over, so those
which showed gold now show black and vice versa. The last player who can make a legal move
wins.
(a) Does the game necessarily end?
(b) Does there exist a winning strategy for the starting player?
Solution. (a) We interpret a card showing black as the digit 0 and a card showing gold as the
digit 1. Thus each position of the 2009 cards, read from left to right, corresponds bijectively to
a nonnegative integer written in binary notation of 2009 digits, where leading zeros are allowed.
Each move decreases this integer, so the game must end.
(b) We show that there is no winning strategy for the starting player. We label the cards from
right to left by 1, . . . , 2009 and consider the set S of cards with labels 50i, i = 1, 2, . . . , 40. Let
gn be the number of cards from S showing gold after n moves. Obviously, g0 = 40. Moreover,
|gn − gn+1| = 1 as long as the play goes on. Thus, after an odd number of moves, the non-
starting player ﬁnds a card from S showing gold and hence can make a move. Consequently,
this player always wins.
26

50th IMO 2009 Combinatorics C2
C2 ROU (Romania)
For any integer n ≥ 2, let N(n) be the maximal number of triples (ai, bi, ci), i = 1, . . . , N(n),
consisting of nonnegative integers ai, bi and ci such that the following two conditions are satis-
ﬁed:
(1) ai + bi + ci = n for all i = 1, . . . , N(n),
(2) If i = j, then ai = aj, bi = bj and ci = cj.
Determine N(n) for all n ≥ 2.
Comment. The original problem was formulated for m-tuples instead for triples. The numbers
N(m, n) are then deﬁned similarly to N(n) in the case m = 3. The numbers N(3, n) and
N(n, n) should be determined. The case m = 3 is the same as in the present problem. The
upper bound for N(n, n) can be proved by a simple generalization. The construction of a set
of triples attaining the bound can be easily done by induction from n to n + 2.
Solution. Let n ≥ 2 be an integer and let {T1, . . . , TN } be any set of triples of nonnegative
integers satisfying the conditions (1) and (2). Since the a-coordinates are pairwise distinct we
have
N
i=1
ai ≥
N
i=1
(i − 1) =
N(N − 1)
2
.
Analogously,
N
i=1
bi ≥
N(N − 1)
2
and
N
i=1
ci ≥
N(N − 1)
2
.
Summing these three inequalities and applying (1) yields
3
N(N − 1)
2
≤
N
i=1
ai +
N
i=1
bi +
N
i=1
ci =
N
i=1
(ai + bi + ci) = nN,
hence 3N−1
2
≤ n and, consequently,
N ≤
2n
3
+ 1.
By constructing examples, we show that this upper bound can be attained, so N(n) = 2n
3
+1.
27

We distinguish the cases n = 3k −1, n = 3k and n = 3k +1 for k ≥ 1 and present the extremal
examples in form of a table.
n = 3k − 1
2n
3
+ 1 = 2k
ai bi ci
0 k + 1 2k − 2
1 k + 2 2k − 4
...
...
...
k − 1 2k 0
k 0 2k − 1
k + 1 1 2k − 3
...
...
...
2k − 1 k − 1 1
n = 3k
2n
3
+ 1 = 2k + 1
ai bi ci
0 k 2k
1 k + 1 2k − 2
...
...
...
k 2k 0
k + 1 0 2k − 1
k + 2 1 2k − 3
...
...
...
2k k − 1 1
n = 3k + 1
2n
3
+ 1 = 2k + 1
ai bi ci
0 k 2k + 1
1 k + 1 2k − 1
...
...
...
k 2k 1
k + 1 0 2k
k + 2 1 2k − 2
...
...
...
2k k − 1 2
It can be easily seen that the conditions (1) and (2) are satisfied and that we indeed have
2n
3
+ 1 triples in each case.
Comment. A cute combinatorial model is given by an equilateral triangle, partitioned into
n2
congruent equilateral triangles by n − 1 equidistant parallels to each of its three sides. Two
chess-like bishops placed at any two vertices of the small triangles are said to menace one
another if they lie on a same parallel. The problem is to determine the largest number of
bishops that can be placed so that none menaces another. A bishop may be assigned three
coordinates a, b, c, namely the numbers of sides of small triangles they are off each of the sides
of the big triangle. It is readily seen that the sum of these coordinates is always n, therefore
fulfilling the requirements.
28

Let n be a positive integer. Given a sequence ε1, . . . , εn−1 with εi = 0 or εi = 1 for each
i = 1, . . . , n − 1, the sequences a0, . . . , an and b0, . . . , bn are constructed by the following rules:
a0 = b0 = 1, a1 = b1 = 7,
ai+1 =
2ai−1 + 3ai, if εi = 0,
3ai−1 + ai, if εi = 1,
for each i = 1, . . . , n − 1,
bi+1 =
2bi−1 + 3bi, if εn−i = 0,
3bi−1 + bi, if εn−i = 1,
for each i = 1, . . . , n − 1.
Prove that an = bn.
Solution. For a binary word w = σ1 . . . σn of length n and a letter σ ∈ {0, 1} let wσ =
σ1 . . . σnσ and σw = σσ1 . . . σn. Moreover let w = σn . . . σ1 and let ∅ be the empty word (of
length 0 and with ∅ = ∅). Let (u, v) be a pair of two real numbers. For binary words w we
deﬁne recursively the numbers (u, v)w
as follows:
(u, v)∅
= v, (u, v)0
= 2u + 3v, (u, v)1
= 3u + v,
(u, v)wσε
=
2(u, v)w
+ 3(u, v)wσ
, if ε = 0,
3(u, v)w
+ (u, v)wσ
, if ε = 1.
It easily follows by induction on the length of w that for all real numbers u1, v1, u2, v2, λ1 and
λ2
(λ1u1 + λ2u2, λ1v1 + λ2v2)w
= λ1(u1, v1)w
+ λ2(u2, v2)w
(1)
and that for ε ∈ {0, 1}
(u, v)εw
= (v, (u, v)ε
)w
. (2)
Obviously, for n ≥ 1 and w = ε1 . . . εn−1, we have an = (1, 7)w
and bn = (1, 7)w
. Thus it is
suﬃcient to prove that
(1, 7)w
= (1, 7)w
(3)
for each binary word w. We proceed by induction on the length of w. The assertion is obvious
if w has length 0 or 1. Now let wσε be a binary word of length n ≥ 2 and suppose that the
assertion is true for all binary words of length at most n − 1.
Note that (2, 1)σ
= 7 = (1, 7)∅
for σ ∈ {0, 1}, (1, 7)0
= 23, and (1, 7)1
= 10.
First let ε = 0. Then in view of the induction hypothesis and the equalities (1) and (2), we
obtain
(1, 7)wσ0
= 2(1, 7)w
+ 3(1, 7)wσ
= 2(1, 7)w
+ 3(1, 7)σw
= 2(2, 1)σw
+ 3(1, 7)σw
= (7, 23)σw
= (1, 7)0σw
.
29

Now let ε = 1. Analogously, we obtain
(1, 7)wσ1
= 3(1, 7)w
+ (1, 7)wσ
= 3(1, 7)w
+ (1, 7)σw
= 3(2, 1)σw
+ (1, 7)σw
= (7, 10)σw
= (1, 7)1σw
.
Thus the induction step is complete, (3) and hence also an = bn are proved.
Comment. The original solution uses the relation
(1, 7)αβw
= ((1, 7)w
, (1, 7)βw
)α
, α, β ∈ {0, 1},
which can be proved by induction on the length of w. Then (3) also follows by induction on
the length of w:
(1, 7)αβw
= ((1, 7)w
, (1, 7)βw
)α
= ((1, 7)w
, (1, 7)wβ
)α
= (1, 7)wβα
.
Here w may be the empty word.
30

For an integer m ≥ 1, we consider partitions of a 2m
× 2m
chessboard into rectangles consisting
of cells of the chessboard, in which each of the 2m
cells along one diagonal forms a separate
rectangle of side length 1. Determine the smallest possible sum of rectangle perimeters in such
a partition.
Solution 1. For a k×k chessboard, we introduce in a standard way coordinates of the vertices
of the cells and assume that the cell Cij in row i and column j has vertices (i − 1, j − 1), (i −
1, j), (i, j−1), (i, j), where i, j ∈ {1, . . . , k}. Without loss of generality assume that the cells Cii,
i = 1, . . . , k, form a separate rectangle. Then we may consider the boards Bk = 1≤i<j≤k Cij
below that diagonal and the congruent board Bk = 1≤j<i≤k Cij above that diagonal separately
because no rectangle can simultaneously cover cells from Bk and Bk. We will show that for
k = 2m
the smallest total perimeter of a rectangular partition of Bk is m2m+1
. Then the overall
answer to the problem is 2 · m2m+1
+ 4 · 2m
= (m + 1)2m+2
.
First we inductively construct for m ≥ 1 a partition of B2m with total perimeter m2m+1
. If
m = 0, the board B2m is empty and the total perimeter is 0. For m ≥ 0, the board B2m+1 consists
of a 2m
× 2m
square in the lower right corner with vertices (2m
, 2m
), (2m
, 2m+1
), (2m+1
, 2m
),
(2m+1
, 2m+1
) to which two boards congruent to B2m are glued along the left and the upper
margin. The square together with the inductive partitions of these two boards yield a partition
with total perimeter 4 · 2m
+ 2 · m2m+1
= (m + 1)2m+2
and the induction step is complete.
Let
Dk = 2k log2 k.
Note that Dk = m2m+1
if k = 2m
. Now we show by induction on k that the total perimeter of
a rectangular partition of Bk is at least Dk. The case k = 1 is trivial (see m = 0 from above).
Let the assertion be true for all positive integers less than k. We investigate a ﬁxed rectangular
partition of Bk that attains the minimal total perimeter. Let R be the rectangle that covers the
cell C1k in the lower right corner. Let (i, j) be the upper left corner of R. First we show that
i = j. Assume that i < j. Then the line from (i, j) to (i + 1, j) or from (i, j) to (i, j − 1) must
belong to the boundary of some rectangle in the partition. Without loss of generality assume
that this is the case for the line from (i, j) to (i + 1, j).
Case 1. No line from (i, l) to (i + 1, l) where j < l < k belongs to the boundary of some
rectangle of the partition.
Then there is some rectangle R of the partition that has with R the common side from (i, j)
to (i, k). If we join these two rectangles to one rectangle we get a partition with smaller total
perimeter, a contradiction.
Case 2. There is some l such that j < l < k and the line from (i, l) to (i + 1, l) belongs to the
boundary of some rectangle of the partition.
Then we replace the upper side of R by the line (i + 1, j) to (i + 1, k) and for the rectangles
whose lower side belongs to the line from (i, j) to (i, k) we shift the lower side upwards so that
the new lower side belongs to the line from (i + 1, j) to (i + 1, k). In such a way we obtain a
rectangular partition of Bk with smaller total perimeter, a contradiction.
Now the fact that the upper left corner of R has the coordinates (i, i) is established. Conse-
quently, the partition consists of R, of rectangles of a partition of a board congruent to Bi and
of rectangles of a partition of a board congruent to Bk−i. By the induction hypothesis, its total
31

perimeter is at least
2(k − i) + 2i + Di + Dk−i ≥ 2k + 2i log2 i + 2(k − i) log2(k − i). (1)
Since the function f(x) = 2x log2 x is convex for x > 0, Jensen’s inequality immediately shows
that the minimum of the right hand sight of (1) is attained for i = k/2. Hence the total
perimeter of the optimal partition of Bk is at least 2k + 2k/2 log2 k/2 + 2(k/2) log2(k/2) = Dk.
Solution 2. We start as in Solution 1 and present another proof that m2m+1
is a lower bound
for the total perimeter of a partition of B2m into n rectangles. Let brieﬂy M = 2m
. For
1 ≤ i ≤ M, let ri denote the number of rectangles in the partition that cover some cell from
row i and let cj be the number of rectangles that cover some cell from column j. Note that the
total perimeter p of all rectangles in the partition is
p = 2
M
i=1
ri +
M
i=1
ci .
No rectangle can simultaneously cover cells from row i and from column i since otherwise it
would also cover the cell Cii. We classify subsets S of rectangles of the partition as follows.
We say that S is of type i, 1 ≤ i ≤ M, if S contains all ri rectangles that cover some cell from
row i, but none of the ci rectangles that cover some cell from column i. Altogether there are
2n−ri−ci
subsets of type i. Now we show that no subset S can be simultaneously of type i and of
type j if i = j. Assume the contrary and let without loss of generality i < j. The cell Cij must
be covered by some rectangle R. The subset S is of type i, hence R is contained in S. S is of
type j, thus R does not belong to S, a contradiction. Since there are 2n
subsets of rectangles
of the partition, we infer
2n
≥
M
i=1
2n−ri−ci
= 2n
M
i=1
2−(ri+ci)
. (2)
By applying Jensen’s inequality to the convex function f(x) = 2−x
we derive
1
M
M
i=1
2−(ri+ci)
≥ 2− 1
M
M
i=1(ri+ci)
= 2− p
2M . (3)
From (2) and (3) we obtain
1 ≥ M2− p
2M
and equivalently
p ≥ m2m+1
.
32

Five identical empty buckets of 2-liter capacity stand at the vertices of a regular pentagon.
Cinderella and her wicked Stepmother go through a sequence of rounds: At the beginning of
every round, the Stepmother takes one liter of water from the nearby river and distributes it
arbitrarily over the five buckets. Then Cinderella chooses a pair of neighboring buckets, empties
them into the river, and puts them back. Then the next round begins. The Stepmother’s goal
is to make one of these buckets overflow. Cinderella’s goal is to prevent this. Can the wicked
Stepmother enforce a bucket overflow?
Solution 1. No, the Stepmother cannot enforce a bucket overflow and Cinderella can keep
playing forever. Throughout we denote the five buckets by B0, B1, B2, B3, and B4, where Bk
is adjacent to bucket Bk−1 and Bk+1 (k = 0, 1, 2, 3, 4) and all indices are taken modulo 5.
Cinderella enforces that the following three conditions are satisfied at the beginning of every
round:
(1) Two adjacent buckets (say B1 and B2) are empty.
(2) The two buckets standing next to these adjacent buckets (here B0 and B3) have total
contents at most 1.
(3) The remaining bucket (here B4) has contents at most 1.
These conditions clearly hold at the beginning of the first round, when all buckets are empty.
Assume that Cinderella manages to maintain them until the beginning of the r-th round (r ≥ 1).
Denote by xk (k = 0, 1, 2, 3, 4) the contents of bucket Bk at the beginning of this round and
by yk the corresponding contents after the Stepmother has distributed her liter of water in this
round.
By the conditions, we can assume x1 = x2 = 0, x0 + x3 ≤ 1 and x4 ≤ 1. Then, since the
Stepmother adds one liter, we conclude y0 +y1 +y2 +y3 ≤ 2. This inequality implies y0 +y2 ≤ 1
or y1 + y3 ≤ 1. For reasons of symmetry, we only consider the second case.
Then Cinderella empties buckets B0 and B4.
At the beginning of the next round B0 and B4 are empty (condition (1) is fulfilled), due to
y1 +y3 ≤ 1 condition (2) is fulfilled and finally since x2 = 0 we also must have y2 ≤ 1 (condition
(3) is fulfilled).
Therefore, Cinderella can indeed manage to maintain the three conditions (1)–(3) also at the
beginning of the (r + 1)-th round. By induction, she thus manages to maintain them at the
beginning of every round. In particular she manages to keep the contents of every single bucket
at most 1 liter. Therefore, the buckets of 2-liter capacity will never overflow.
Solution 2. We prove that Cinderella can maintain the following two conditions and hence
she can prevent the buckets from overflow:
(1 ) Every two non-adjacent buckets contain a total of at most 1.
(2 ) The total contents of all five buckets is at most 3
2
.
We use the same notations as in the first solution. The two conditions again clearly hold at
the beginning. Assume that Cinderella maintained these two conditions until the beginning of
the r-th round. A pair of non-neighboring buckets (Bi, Bi+2), i = 0, 1, 2, 3, 4 is called critical
33

if yi + yi+2 > 1. By condition (2 ), after the Stepmother has distributed her water we have
y0 + y1 + y2 + y3 + y4 ≤ 5
2
. Therefore,
(y0 + y2) + (y1 + y3) + (y2 + y4) + (y3 + y0) + (y4 + y1) = 2(y0 + y1 + y2 + y3 + y4) ≤ 5,
and hence there is a pair of non-neighboring buckets which is not critical, say (B0, B2). Now,
if both of the pairs (B3, B0) and (B2, B4) are critical, we must have y1 < 1
2
and Cinderella
can empty the buckets B3 and B4. This clearly leaves no critical pair of buckets and the total
contents of all the buckets is then y1 + (y0 + y2) ≤ 3
2
. Therefore, conditions (1 ) and (2 ) are
fulfilled.
Now suppose that without loss of generality the pair (B3, B0) is not critical. If in this case
y0 ≤ 1
2
, then one of the inequalities y0 + y1 + y2 ≤ 3
2
and y0 + y3 + y4 ≤ 3
2
must hold. But then
Cinderella can empty B3 and B4 or B1 and B2, respectively and clearly fulfill the conditions.
Finally consider the case y0 > 1
2
. By y0 +y1 +y2 +y3 +y4 ≤ 5
2
, at least one of the pairs (B1, B3)
and (B2, B4) is not critical. Without loss of generality let this be the pair (B1, B3). Since the
pair (B3, B0) is not critical and y0 > 1
2
, we must have y3 ≤ 1
2
. But then, as before, Cinderella
can maintain the two conditions at the beginning of the next round by either emptying B1 and
B2 or B4 and B0.
Comments on GREEDY approaches. A natural approach for Cinderella would be a GREEDY
strategy as for example: Always remove as much water as possible from the system. It is
straightforward to prove that GREEDY can avoid buckets of capacity 5
2
from overflowing: If
before the Stepmothers move one has x0 + x1 + x2 + x3 + x4 ≤ 3
2
then after her move the
inequality Y = y0 + y1 + y2 + y3 + y4 ≤ 5
2
holds. If now Cinderella removes the two adjacent
buckets with maximum total contents she removes at least 2Y
5
and thus the remaining buckets
contain at most 3
5
· Y ≤ 3
2
.
But GREEDY is in general not strong enough to settle this problem as can be seen in the
following example:
• In an initial phase, the Stepmother brings all the buckets (after her move) to contents
of at least 1
2
− 2 , where is an arbitrary small positive number. This can be done
by always splitting the 1 liter she has to distribute so that all buckets have the same
contents. After her r-th move the total contents of each of the buckets is then cr with
c1 = 1 and cr+1 = 1 + 3
5
· cr and hence cr = 5
2
− 3
2
· 3
5
r−1
. So the contents of each
single bucket indeed approaches 1
2
(from below). In particular, any two adjacent buckets
have total contents strictly less than 1 which enables the Stepmother to always refill the
buckets that Cinderella just emptied and then distribute the remaining water evenly over
all buckets.
• After that phase GREEDY faces a situation like this (1
2
− 2 , 1
2
− 2 , 1
2
− 2 , 1
2
− 2 , 1
2
− 2 )
and leaves a situation of the form (x0, x1, x2, x3, x4) = (1
2
− 2 , 1
2
− 2 , 1
2
− 2 , 0, 0).
• Then the Stepmother can add the amounts (0, 1
4
+ , , 3
4
− 2 , 0) to achieve a situation
like this: (y0, y1, y2, y3, y4) = (1
2
− 2 , 3
4
− , 1
2
− , 3
4
− 2 , 0).
• Now B1 and B2 are the adjacent buckets with the maximum total contents and thus
GREEDY empties them to yield (x0, x1, x2, x3, x4) = (1
2
− 2 , 0, 0, 3
4
− 2 , 0).
• Then the Stepmother adds (5
8
, 0, 0, 3
8
, 0), which yields (9
8
− 2 , 0, 0, 9
8
− 2 , 0).
• Now GREEDY can only empty one of the two nonempty buckets and in the next step the
Stepmother adds her liter to the other bucket and brings it to 17
8
− 2 , i.e. an overflow.
34

A harder variant. Five identical empty buckets of capacity b stand at the vertices of a regular
pentagon. Cinderella and her wicked Stepmother go through a sequence of rounds: At the
beginning of every round, the Stepmother takes one liter of water from the nearby river and
distributes it arbitrarily over the five buckets. Then Cinderella chooses a pair of neighboring
buckets, empties them into the river, and puts them back. Then the next round begins. The
Stepmother’s goal is to make one of these buckets overflow. Cinderella’s goal is to prevent this.
Determine all bucket capacities b for which the Stepmother can enforce a bucket to overflow.
Solution to the harder variant. The answer is b < 2.
The previous proof shows that for all b ≥ 2 the Stepmother cannot enforce overflowing. Now if
b < 2, let R be a positive integer such that b < 2 − 21−R
. In the first R rounds the Stepmother
now ensures that at least one of the (nonadjacent) buckets B1 and B3 have contents of at
least 1 − 21−r
at the beginning of round r (r = 1, 2, . . . , R). This is trivial for r = 1 and if it
holds at the beginning of round r, she can fill the bucket which contains at least 1 − 21−r
liters
with another 2−r
liters and put the rest of her water – 1 − 2−r
liters – in the other bucket.
As Cinderella now can remove the water of at most one of the two buckets, the other bucket
carries its contents into the next round.
At the beginning of the R-th round there are 1 − 21−R
liters in B1 or B3. The Stepmother puts
the entire liter into that bucket and produces an overflow since b < 2 − 21−R
.
35

C6 BGR (Bulgaria)
On a 999×999 board a limp rook can move in the following way: From any square it can move
to any of its adjacent squares, i.e. a square having a common side with it, and every move
must be a turn, i.e. the directions of any two consecutive moves must be perpendicular. A non-
intersecting route of the limp rook consists of a sequence of pairwise different squares that the
limp rook can visit in that order by an admissible sequence of moves. Such a non-intersecting
route is called cyclic, if the limp rook can, after reaching the last square of the route, move
directly to the first square of the route and start over.
How many squares does the longest possible cyclic, non-intersecting route of a limp rook
visit?
Solution. The answer is 9982
− 4 = 4 · (4992
− 1) squares.
First we show that this number is an upper bound for the number of cells a limp rook can
visit. To do this we color the cells with four colors A, B, C and D in the following way: for
(i, j) ≡ (0, 0) mod 2 use A, for (i, j) ≡ (0, 1) mod 2 use B, for (i, j) ≡ (1, 0) mod 2 use C and
for (i, j) ≡ (1, 1) mod 2 use D. From an A-cell the rook has to move to a B-cell or a C-cell. In
the first case, the order of the colors of the cells visited is given by A, B, D, C, A, B, D, C, A, . . .,
in the second case it is A, C, D, B, A, C, D, B, A, . . .. Since the route is closed it must contain
the same number of cells of each color. There are only 4992
A-cells. In the following we will
show that the rook cannot visit all the A-cells on its route and hence the maximum possible
number of cells in a route is 4 · (4992
− 1).
Assume that the route passes through every single A-cell. Color the A-cells in black and white
in a chessboard manner, i.e. color any two A-cells at distance 2 in different color. Since the
number of A-cells is odd the rook cannot always alternate between visiting black and white
A-cells along its route. Hence there are two A-cells of the same color which are four rook-steps
apart that are visited directly one after the other. Let these two A-cells have row and column
numbers (a, b) and (a + 2, b + 2) respectively.
There is up to reflection only one way the rook can take from (a, b) to (a + 2, b + 2). Let this
way be (a, b) → (a, b + 1) → (a + 1, b + 1) → (a + 1, b + 2) → (a + 2, b + 2). Also let without
loss of generality the color of the cell (a, b+1) be B (otherwise change the roles of columns and
rows).
Now consider the A-cell (a, b+2). The only way the rook can pass through it is via (a−1, b+2) →
(a, b + 2) → (a, b + 3) in this order, since according to our assumption after every A-cell the
rook passes through a B-cell. Hence, to connect these two parts of the path, there must be
36

a path connecting the cell (a, b + 3) and (a, b) and also a path connecting (a + 2, b + 2) and
(a − 1, b + 2).
But these four cells are opposite vertices of a convex quadrilateral and the paths are outside of
that quadrilateral and hence they must intersect. This is due to the following fact:
The path from (a, b) to (a, b + 3) together with the line segment joining these two cells form a
closed loop that has one of the cells (a − 1, b + 2) and (a + 2, b + 2) in its inside and the other
one on the outside. Thus the path between these two points must cross the previous path.
But an intersection is only possible if a cell is visited twice. This is a contradiction.
Hence the number of cells visited is at most 4 · (4992
− 1).
The following picture indicates a recursive construction for all n × n-chessboards with n ≡ 3
mod 4 which clearly yields a path that misses exactly one A-cell (marked with a dot, the center
cell of the 15 × 15-chessboard) and hence, in the case of n = 999 crosses exactly 4 · (4992
− 1)
cells.
37

Variant 1. A grasshopper jumps along the real axis. He starts at point 0 and makes 2009
jumps to the right with lengths 1, 2, . . . , 2009 in an arbitrary order. Let M be a set of 2008
positive integers less than 1005 · 2009. Prove that the grasshopper can arrange his jumps in
such a way that he never lands on a point from M.
Variant 2. Let n be a nonnegative integer. A grasshopper jumps along the real axis. He starts
at point 0 and makes n + 1 jumps to the right with pairwise different positive integral lengths
a1, a2, . . . , an+1 in an arbitrary order. Let M be a set of n positive integers in the interval (0, s),
where s = a1 + a2 + · · · + an+1. Prove that the grasshopper can arrange his jumps in such a
way that he never lands on a point from M.
Solution of Variant 1. We construct the set of landing points of the grasshopper.
Case 1. M does not contain numbers divisible by 2009.
We fix the numbers 2009k as landing points, k = 1, 2, . . . , 1005. Consider the open intervals
Ik = (2009(k − 1), 2009k), k = 1, 2, . . . , 1005. We show that we can choose exactly one point
outside of M as a landing point in 1004 of these intervals such that all lengths from 1 to 2009
are realized. Since there remains one interval without a chosen point, the length 2009 indeed
will appear. Each interval has length 2009, hence a new landing point in an interval yields
with a length d also the length 2009 − d. Thus it is enough to implement only the lengths from
D = {1, 2, . . . , 1004}. We will do this in a greedy way. Let nk, k = 1, 2, . . . , 1005, be the number
of elements of M that belong to the interval Ik. We order these numbers in a decreasing way,
so let p1, p2, . . . , p1005 be a permutation of {1, 2, . . . , 1005} such that np1 ≥ np2 ≥ · · · ≥ np1005 .
In Ip1 we do not choose a landing point. Assume that landing points have already been chosen
in the intervals Ip2 , . . . , Ipm and the lengths d2, . . . , dm from D are realized, m = 1, . . . , 1004.
We show that there is some d ∈ D {d2, . . . , dm} that can be implemented with a new landing
point in Ipm+1 . Assume the contrary. Then the 1004 − (m − 1) other lengths are obstructed by
the npm+1 points of M in Ipm+1 . Each length d can be realized by two landing points, namely
2009(pm+1 − 1) + d and 2009pm+1 − d, hence
npm+1 ≥ 2(1005 − m). (1)
Moreover, since |M| = 2008 = n1 + · · · + n1005,
2008 ≥ np1 + np2 + · · · + npm+1 ≥ (m + 1)npm+1 . (2)
Consequently, by (1) and (2),
2008 ≥ 2(m + 1)(1005 − m).
The right hand side of the last inequality obviously attains its minimum for m = 1004 and this
minimum value is greater than 2008, a contradiction.
Case 2. M does contain a number µ divisible by 2009.
By the pigeonhole principle there exists some r ∈ {1, . . . , 2008} such that M does not contain
numbers with remainder r modulo 2009. We fix the numbers 2009(k − 1)+ r as landing points,
k = 1, 2, . . . , 1005. Moreover, 1005 · 2009 is a landing point. Consider the open intervals
38

Ik = (2009(k − 1) + r, 2009k + r), k = 1, 2, . . . , 1004. Analogously to Case 1, it is enough to
show that we can choose in 1003 of these intervals exactly one landing point outside of M {µ}
such that each of the lengths of D = {1, 2, . . . , 1004} {r} are implemented. Note that r
and 2009 − r are realized by the first and last jump and that choosing µ would realize these
two differences again. Let nk, k = 1, 2, . . . , 1004, be the number of elements of M {µ} that
belong to the interval Ik and p1, p2, . . . , p1004 be a permutation of {1, 2, . . . , 1004} such that
np1 ≥ np2 ≥ · · · ≥ np1004 . With the same reasoning as in Case 1 we can verify that a greedy
choice of the landing points in Ip2 , Ip3 , . . . , Ip1004 is possible. We only have to replace (1) by
npm+1 ≥ 2(1004 − m)
(D has one element less) and (2) by
2007 ≥ np1 + np2 + · · · + npm+1 ≥ (m + 1)npm+1 .
Comment. The cardinality 2008 of M in the problem is the maximum possible value. For
M = {1, 2, . . . , 2009}, the grasshopper necessarily lands on a point from M.
Solution of Variant 2. First of all we remark that the statement in the problem implies a
strengthening of itself: Instead of |M| = n it is sufficient to suppose that |M ∩ (0, s − a]| ≤ n,
where a = min{a1, a2, . . . , an+1}. This fact will be used in the proof.
We prove the statement by induction on n. The case n = 0 is obvious. Let n > 0 and let the
assertion be true for all nonnegative integers less than n. Moreover let a1, a2, . . . , an+1, s and
M be given as in the problem. Without loss of generality we may assume that an+1 < an <
· · · < a2 < a1. Set
Tk =
k
i=1
ai for k = 0, 1, . . . , n + 1.
Note that 0 = T0 < T1 < · · · < Tn+1 = s. We will make use of the induction hypothesis as
follows:
Claim 1. It suffices to show that for some m ∈ {1, 2, . . . , n + 1} the grasshopper is able to do
at least m jumps without landing on a point of M and, in addition, after these m jumps he
has jumped over at least m points of M.
Proof. Note that m = n + 1 is impossible by |M| = n. Now set n = n − m. Then 0 ≤ n < n.
The remaining n + 1 jumps can be carried out without landing on one of the remaining at
most n forbidden points by the induction hypothesis together with a shift of the origin. This
proves the claim.
An integer k ∈ {1, 2, . . . , n + 1} is called smooth, if the grasshopper is able to do k jumps with
the lengths a1, a2, . . . , ak in such a way that he never lands on a point of M except for the very
last jump, when he may land on a point of M.
Obviously, 1 is smooth. Thus there is a largest number k∗
, such that all the numbers 1, 2, . . . , k∗
are smooth. If k∗
= n + 1, the proof is complete. In the following let k∗
≤ n.
Claim 2. We have
Tk∗ ∈ M and |M ∩ (0, Tk∗ )| ≥ k∗
. (3)
Proof. In the case Tk∗ ∈ M any sequence of jumps that verifies the smoothness of k∗
can be
extended by appending ak∗+1, which is a contradiction to the maximality of k∗
. Therefore we
have Tk∗ ∈ M. If |M ∩ (0, Tk∗ )| < k∗
, there exists an l ∈ {1, 2, . . . , k∗
} with Tk∗+1 − al ∈ M. By
the induction hypothesis with k∗
− 1 instead of n, the grasshopper is able to reach Tk∗+1 − al
39

with k∗
jumps of lengths from {a1, a2, . . . , ak∗+1} {al} without landing on any point of M.
Therefore k∗
+1 is also smooth, which is a contradiction to the maximality of k∗
. Thus Claim 2
is proved.
Now, by Claim 2, there exists a smallest integer k ∈ {1, 2, . . . , k∗
} with
Tk ∈ M and |M ∩ (0, Tk)| ≥ k.
Claim 3. It is sufficient to consider the case
|M ∩ (0, Tk−1]| ≤ k − 1. (4)
Proof. If k = 1, then (4) is clearly satisfied. In the following let k > 1. If Tk−1 ∈ M, then
(4) follows immediately by the minimality of k. If Tk−1 ∈ M, by the smoothness of k − 1, we
obtain a situation as in Claim 1 with m = k − 1 provided that |M ∩ (0, Tk−1]| ≥ k − 1. Hence,
we may even restrict ourselves to |M ∩ (0, Tk−1]| ≤ k − 2 in this case and Claim 3 is proved.
Choose an integer v ≥ 0 with |M ∩ (0, Tk)| = k + v. Let r1 > r2 > · · · > rl be exactly those
indices r from {k + 1, k + 2, . . . , n + 1} for which Tk + ar ∈ M. Then
n = |M| = |M ∩ (0, Tk)| + 1 + |M ∩ (Tk, s)| ≥ k + v + 1 + (n + 1 − k − l)
and consequently l ≥ v + 2. Note that
Tk + ar1 − a1 < Tk + ar1 − a2 < · · · < Tk + ar1 − ak < Tk + ar2 − ak < · · · < Tk + arv+2 − ak < Tk
and that this are k + v + 1 numbers from (0, Tk). Therefore we find some r ∈ {k + 1, k +
2, . . . , n + 1} and some s ∈ {1, 2, . . . , k} with Tk + ar ∈ M and Tk + ar − as ∈ M. Consider the
set of jump lengths B = {a1, a2, . . . , ak, ar} {as}. We have
x∈B
x = Tk + ar − as
and
Tk + ar − as − min(B) = Tk − as ≤ Tk−1.
By (4) and the strengthening, mentioned at the very beginning with k − 1 instead of n, the
grasshopper is able to reach Tk + ar − as by k jumps with lengths from B without landing on
any point of M. From there he is able to jump to Tk + ar and therefore we reach a situation as
in Claim 1 with m = k + 1, which completes the proof.
40

C8 AUT (Austria)
For any integer n ≥ 2, we compute the integer h(n) by applying the following procedure to its
decimal representation. Let r be the rightmost digit of n.
(1) If r = 0, then the decimal representation of h(n) results from the decimal representation
of n by removing this rightmost digit 0.
(2) If 1 ≤ r ≤ 9 we split the decimal representation of n into a maximal right part R that
solely consists of digits not less than r and into a left part L that either is empty or ends
with a digit strictly smaller than r. Then the decimal representation of h(n) consists of the
decimal representation of L, followed by two copies of the decimal representation of R − 1.
For instance, for the number n = 17,151,345,543, we will have L = 17,151, R = 345,543
and h(n) = 17,151,345,542,345,542.
Prove that, starting with an arbitrary integer n ≥ 2, iterated application of h produces the
integer 1 after finitely many steps.
Solution 1. We identify integers n ≥ 2 with the digit-strings, briefly strings, of their decimal
representation and extend the definition of h to all non-empty strings with digits from 0 to
9. We recursively define ten functions f0, . . . , f9 that map some strings into integers for k =
9, 8, . . . , 1, 0. The function f9 is only defined on strings x (including the empty string ε) that
entirely consist of nines. If x consists of m nines, then f9(x) = m + 1, m = 0, 1, . . . . For k ≤ 8,
the domain of fk(x) is the set of all strings consisting only of digits that are ≥ k. We write x
in the form x0kx1kx2k . . . xm−1kxm where the strings xs only consist of digits ≥ k + 1. Note
that some of these strings might equal the empty string ε and that m = 0 is possible, i.e. the
digit k does not appear in x. Then we define
fk(x) =
m
s=0
4fk+1(xs)
.
We will use the following obvious fact:
Fact 1. If x does not contain digits smaller than k, then fi(x) = 4fi+1(x)
for all i = 0, . . . , k − 1.
In particular, fi(ε) = 49−i
for all i = 0, 1, . . . , 9.
Moreover, by induction on k = 9, 8, . . . , 0 it follows easily:
Fact 2. If the nonempty string x does not contain digits smaller than k, then fi(x) > fi(ε) for
all i = 0, . . . , k.
We will show the essential fact:
Fact 3. f0(n) > f0(h(n)).
Then the empty string will necessarily be reached after a finite number of applications of
h. But starting from a string without leading zeros, ε can only be reached via the strings
1 → 00 → 0 → ε. Hence also the number 1 will appear after a finite number of applications of
h.
Proof of Fact 3. If the last digit r of n is 0, then we write n = x00 . . . 0xm−10ε where the xi do
not contain the digit 0. Then h(n) = x00 . . . 0xm−1 and f0(n) − f0(h(n)) = f0(ε) > 0.
So let the last digit r of n be at least 1. Let L = yk and R = zr be the corresponding left and
right parts where y is some string, k ≤ r − 1 and the string z consists only of digits not less
41

than r. Then n = ykzr and h(n) = ykz(r − 1)z(r − 1). Let d(y) be the smallest digit of y. We
consider two cases which do not exclude each other.
Case 1. d(y) ≥ k.
Then
fk(n) − fk(h(n)) = fk(zr) − fk(z(r − 1)z(r − 1)).
In view of Fact 1 this difference is positive if and only if
fr−1(zr) − fr−1(z(r − 1)z(r − 1)) > 0.
We have, using Fact 2,
fr−1(zr) = 4fr(zr)
= 4fr(z)+4fr+1(ε)
≥ 4 · 4fr(z)
> 4fr(z)
+ 4fr(z)
+ 4fr(ε)
= fr−1(z(r − 1)z(r − 1)).
Here we use the additional definition f10(ε) = 0 if r = 9. Consequently, fk(n) − fk(h(n)) > 0
and according to Fact 1, f0(n) − f0(h(n)) > 0.
Case 2. d(y) ≤ k.
We prove by induction on d(y) = k, k −1, . . . , 0 that fi(n)−fi(h(n)) > 0 for all i = 0, . . . , d(y).
By Fact 1, it suffices to do so for i = d(y). The initialization d(y) = k was already treated in
Case 1. Let t = d(y) < k. Write y in the form utv where v does not contain digits ≤ t. Then,
in view of the induction hypothesis,
ft(n) − ft(h(n)) = ft(vkzr) − ft(vkz(r − 1)z(r − 1)) = 4ft+1(vkzr)
− 4ft+1(vkz(r−1)z(r−1))
> 0.
Thus the inequality fd(y)(n) − fd(y)(h(n)) > 0 is established and from Fact 1 it follows that
f0(n) − f0(h(n)) > 0.
Solution 2. We identify integers n ≥ 2 with the digit-strings, briefly strings, of their decimal
representation and extend the definition of h to all non-empty strings with digits from 0 to
9. Moreover, let us define that the empty string, ε, is being mapped to the empty string. In
the following all functions map the set of strings into the set of strings. For two functions f
and g let g ◦ f be defined by (g ◦ f)(x) = g(f(x)) for all strings x and let, for non-negative
integers n, fn
denote the n-fold application of f. For any string x let s(x) be the smallest digit
of x, and for the empty string let s(ε) = ∞. We define nine functions g1, . . . , g9 as follows: Let
k ∈ {1, . . . , 9} and let x be a string. If x = ε then gk(x) = ε. Otherwise, write x in the form
x = yzr where y is either the empty string or ends with a digit smaller than k, s(z) ≥ k and r
is the rightmost digit of x. Then gk(x) = zr.
Lemma 1. We have gk ◦ h = gk ◦ h ◦ gk for all k = 1, . . . , 9.
Proof of Lemma 1. Let x = yzr be as in the definition of gk. If y = ε, then gk(x) = x, whence
gk(h(x)) = gk(h(gk(x)). (1)
So let y = ε.
Case 1. z contains a digit smaller than r.
Let z = uav where a < r and s(v) ≥ r. Then
h(x) =
yuav if r = 0,
yuav(r − 1)v(r − 1) if r > 0
42

and
h(gk(x)) = h(zr) = h(uavr) =
uav if r = 0,
uav(r − 1)v(r − 1) if r > 0.
Since y ends with a digit smaller than k, (1) is obviously true.
Case 2. z does not contain a digit smaller than r.
Let y = uv where u is either the empty string or ends with a digit smaller than r and s(v) ≥ r.
We have
h(x) =
uvz if r = 0,
uvz(r − 1)vz(r − 1) if r > 0
and
h(gk(x)) = h(zr) =
z if r = 0,
z(r − 1)z(r − 1) if r > 0.
Recall that y and hence v ends with a digit smaller than k, but all digits of v are at least r.
Now if r > k, then v = ε, whence the terminal digit of u is smaller than k, which entails
gk(h(x)) = z(r − 1)z(r − 1) = gk(h(gk(x))).
If r ≤ k, then
gk(h(x)) = z(r − 1) = gk(h(gk(x))) ,
so that in both cases (1) is true. Thus Lemma 1 is proved.
Lemma 2. Let k ∈ {1, . . . , 9}, let x be a non-empty string and let n be a positive integer. If
hn
(x) = ε then (gk ◦ h)n
(x) = ε.
Proof of Lemma 2. We proceed by induction on n. If n = 1 we have
ε = h(x) = gk(h(x)) = (gk ◦ h)(x).
Now consider the step from n − 1 to n where n ≥ 2. Let hn
(x) = ε and let y = h(x). Then
hn−1
(y) = ε and by the induction hypothesis (gk ◦ h)n−1
(y) = ε. In view of Lemma 1,
ε = (gk ◦ h)n−2
((gk ◦ h)(y)) = (gk ◦ h)n−2
(gk(h(y))
= (gk ◦ h)n−2
(gk(h(gk(y))) = (gk ◦ h)n−2
(gk(h(gk(h(x)))) = (gk ◦ h)n
(x).
Thus the induction step is complete and Lemma 2 is proved.
We say that the non-empty string x terminates if hn
(x) = ε for some non-negative integer n.
Lemma 3. Let x = yzr where s(y) ≥ k, s(z) ≥ k, y ends with the digit k and z is possibly
empty. If y and zr terminate then also x terminates.
Proof of Lemma 3. Suppose that y and zr terminate. We proceed by induction on k. Let k = 0.
Obviously, h(yw) = yh(w) for any non-empty string w. Let hn
(zr) = . It follows easily by
induction on m that hm
(yzr) = yhm
(zr) for m = 1, . . . , n. Consequently, hn
(yzr) = y. Since y
terminates, also x = yzr terminates.
Now let the assertion be true for all nonnegative integers less than k and let us prove it for k
where k ≥ 1. It turns out that it is suﬃcient to prove that ygk(h(zr)) terminates. Indeed:
Case 1. r = 0.
Then h(yzr) = yz = ygk(h(zr)).
Case 2. 0 < r ≤ k.
We have h(zr) = z(r − 1)z(r − 1) and gk(h(zr)) = z(r − 1). Then h(yzr) = yz(r − 1)yz(r −
43

1) = ygk(h(zr))ygk(h(zr)) and we may apply the induction hypothesis to see that if ygkh(zr))
terminates, then h(yzr) terminates.
Case 3. r > k.
Then h(yzr) = yh(zr) = ygk(h(zr)).
Note that ygk(h(zr)) has the form yz r where s(z ) ≥ k. By the same arguments it is sufficient
to prove that ygk(h(z r )) = y(gk ◦ h)2
(zr) terminates and, by induction, that y(gk ◦ h)m
(zr)
terminates for some positive integer m. In view of Lemma 2 there is some m such that (gk ◦
h)m
(zr) = , so x = yzr terminates if y terminates. Thus Lemma 3 is proved.
Now assume that there is some string x that does not terminate. We choose x minimal. If
x ≥ 10, we can write x in the form x = yzr of Lemma 3 and by this lemma x terminates since
y and zr are smaller than x. If x ≤ 9, then h(x) = (x − 1)(x − 1) and h(x) terminates again
by Lemma 3 and the minimal choice of x.
Solution 3. We commence by introducing some terminology. Instead of integers, we will
consider the set S of all strings consisting of the digits 0, 1, . . . , 9, including the empty string
. If (a1, a2, . . . , an) is a nonempty string, we let ρ(a) = an denote the terminal digit of a and
λ(a) be the string with the last digit removed. We also define λ( ) = and denote the set of
non-negative integers by N0.
Now let k ∈ {0, 1, 2, . . . , 9} denote any digit. We define a function fk : S −→ S on the set of
strings: First, if the terminal digit of n belongs to {0, 1, . . . , k}, then fk(n) is obtained from n
by deleting this terminal digit, i.e fk(n) = λ(n). Secondly, if the terminal digit of n belongs to
{k + 1, . . . , 9}, then fk(n) is obtained from n by the process described in the problem. We also
define fk( ) = . Note that up to the definition for integers n ≤ 1, the function f0 coincides with
the function h in the problem, through interpreting integers as digit strings. The argument will
be roughly as follows. We begin by introducing a straightforward generalization of our claim
about f0. Then it will be easy to see that f9 has all these stronger properties, which means
that is suffices to show for k ∈ {0, 1, . . . , 8} that fk possesses these properties provided that
fk+1 does.
We continue to use k to denote any digit. The operation fk is said to be separating, if the
followings holds: Whenever a is an initial segment of b, there is some N ∈ N0 such that
fN
k (b) = a. The following two notions only apply to the case where fk is indeed separating,
otherwise they remain undefined. For every a ∈ S we denote the least N ∈ N0 for which
fN
k (a) = occurs by gk(a) (because is an initial segment of a, such an N exists if fk is
separating). If for every two strings a and b such that a is a terminal segment of b one has
gk(a) ≤ gk(b), we say that fk is coherent. In case that fk is separating and coherent we call the
digit k seductive.
As f9(a) = λ(a) for all a, it is obvious that 9 is seductive. Hence in order to show that 0 is seduc-
tive, which clearly implies the statement of the problem, it suffices to take any k ∈ {0, 1, . . . , 8}
such that k + 1 is seductive and to prove that k has to be seductive as well. Note that in doing
so, we have the function gk+1 at our disposal. We have to establish two things and we begin with
Step 1. fk is separating.
Before embarking on the proof of this, we record a useful observation which is easily proved by
induction on M.
44

Claim 1. For any strings A, B and any positive integer M such that fM−1
k (B) = , we have
fM
k (AkB) = AkfM
k (B).
Now we call a pair (a, b) of strings wicked provided that a is an initial segment of b, but there
is no N ∈ N0 such that fN
k (b) = a. We need to show that there are none, so assume that
there were such pairs. Choose a wicked pair (a, b) for which gk+1(b) attains its minimal possible
value. Obviously b = for any wicked pair (a, b). Let z denote the terminal digit of b. Observe
that a = b, which means that a is also an initial segment of λ(b). To facilitate the construction
of the eventual contradiction, we prove
Claim 2. There cannot be an N ∈ N0 such that
fN
k (b) = λ(b).
Proof of Claim 2. For suppose that such an N existed. Because gk+1(λ(b)) < gk+1(b) in view
of the coherency of fk+1, the pair (a, λ(b)) is not wicked. But then there is some N for which
fN
k (λ(b)) = a which entails fN+N
k (b) = a, contradiction. Hence Claim 2 is proved.
It follows that z ≤ k is impossible, for otherwise N = 1 violated Claim 2.
Also z > k +1 is impossible: Set B = fk(b). Then also fk+1(b) = B, but gk+1(B) < gk+1(b) and
a is an initial segment of B. Thus the pair (a, B) is not wicked. Hence there is some N ∈ N0
with a = fN
k (B), which, however, entails a = fN+1
k (b).
We are left with the case z = k + 1. Let L denote the left part and R = R∗
(k + 1) the right
part of b. Then we have symbolically
fk(b) = LR∗
kR∗
k , f2
k (b) = LR∗
kR∗
and fk+1(b) = LR∗
.
Using that R∗
is a terminal segment of LR∗
and the coherency of fk+1, we infer
gk+1(R∗
) ≤ gk+1(LR∗
) < gk+1(b).
Hence the pair ( , R∗
) is not wicked, so there is some minimal M ∈ N0 with fM
k (R∗
) = and
by Claim 1 it follows that f2+M
k (b) = LR∗
k. Finally, we infer that λ(b) = LR∗
= fk(LR∗
k) =
f3+M
k (b), which yields a contradiction to Claim 2.
This final contradiction establishes that fk is indeed separating.
Step 2. fk is coherent.
To prepare the proof of this, we introduce some further pieces of terminology. A nonempty
string (a1, a2, . . . , an) is called a hypostasis, if an < ai for all i = 1, . . . , n − 1. Reading an
arbitrary string a backwards, we easily find a, possibly empty, sequence (A1, A2, . . . , Am) of
hypostases such that ρ(A1) ≤ ρ(A2) ≤ · · · ≤ ρ(Am) and, symbolically, a = A1A2 . . . Am.
The latter sequence is referred to as the decomposition of a. So, for instance, (20, 0, 9) is the
decomposition of 2009 and the string 50 is a hypostasis. Next we explain when we say about
two strings a and b that a is injectible into b. The definition is by induction on the length
of b. Let (B1, B2, . . . , Bn) be the decomposition of b into hypostases. Then a is injectible
into b if for the decomposition (A1, A2, . . . , Am) of a there is a strictly increasing function
H : {1, 2, . . . , m} −→ {1, 2, . . . , n} satisfying
ρ(Ai) = ρ(BH(i)) for all i = 1, . . . , m;
45

λ(Ai) is injectible into λ(BH(i)) for all i = 1, . . . , m.
If one can choose H with H(m) = n, then we say that a is strongly injectible into b. Obviously,
if a is a terminal segment of b, then a is strongly injectible into b.
Claim 3. If a and b are two nonempty strings such that a is strongly injectible into b, then λ(a)
is injectible into λ(b).
Proof of Claim 3. Let (B1, B2, . . . , Bn) be the decomposition of b and let (A1, A2, . . . , Am) be
the decomposition of a. Take a function H exemplifying that a is strongly injectible into b.
Let (C1, C2, . . . , Cr) be the decomposition of λ(Am) and let (D1, D2, . . . , Ds) be the decompo-
sition of λ(Bn). Choose a strictly increasing H : {1, 2, . . . , r} −→ {1, 2, . . . s} witnessing that
λ(Am) is injectible into λ(Bn). Clearly, (A1, A2, . . . , Am−1, C1, C2, . . . , Cr) is the decomposition
of λ(a) and (B1, B2, . . . , Bn−1, D1, D2, . . . , Ds) is the decomposition of λ(b). Then the function
H : {1, 2, . . . , m+r −1} −→ {1, 2, . . . , n+s−1} given by H (i) = H(i) for i = 1, 2, . . . , m−1
and H (m − 1 + i) = n − 1 + H (i) for i = 1, 2, . . . , r exemplifies that λ(a) is injectible into
λ(b), which finishes the proof of the claim.
A pair (a, b) of strings is called aggressive if a is injectible into b and nevertheless gk(a) > gk(b).
Observe that if fk was incoherent, which we shall assume from now on, then such pairs existed.
Now among all aggressive pairs we choose one, say (a, b), for which gk(b) attains its least possible
value. Obviously fk(a) cannot be injectible into fk(b), for otherwise the pair (fk(a), fk(b)) was
aggressive and contradicted our choice of (a, b). Let (A1, A2, . . . , Am) and (B1, B2, . . . , Bn)
be the decompositions of a and b and take a function H : {1, 2, . . . , m} −→ {1, 2, . . . , n}
exemplifying that a is indeed injectible into b. If we had H(m) < n, then a was also injectible
into the number b whose decomposition is (B1, B2, . . . , Bn−1) and by separativity of fk we
obtained gk(b ) < gk(b), whence the pair (a, b ) was also aggressive, contrary to the minimality
condition imposed on b. Therefore a is strongly injectible into b. In particular, a and b have a
common terminal digit, say z. If we had z ≤ k, then fk(a) = λ(a) and fk(b) = λ(b), so that by
Claim 3, fk(a) was injectible into fk(b), which is a contradiction. Hence, z ≥ k + 1.
Now let r be the minimal element of {1, 2, . . . , m} for which ρ(Ar) = z. Then the maximal
right part of a consisting of digits ≥ z is equal to Ra, the string whose decomposition is
(Ar, Ar+1, . . . , Am). Then Ra − 1 is a hypostasis and (A1, . . . , Ar−1, Ra − 1, Ra − 1) is the
decomposition of fk(a). Defining s and Rb in a similar fashion with respect to b, we see that
(B1, . . . , Bs−1, Rb − 1, Rb − 1) is the decomposition of fk(b). The definition of injectibility then
easily entails that Ra is strongly injectible into Rb. It follows from Claim 3 that λ(Ra) =
λ(Ra − 1) is injectible into λ(Rb) = λ(Rb − 1), whence the function H : {1, 2, . . . , r + 1} −→
{1, 2, . . . , s + 1}, given by H (i) = H(i) for i = 1, 2, . . . , r − 1, H (r) = s and H (r + 1) = s + 1
exemplifies that fk(a) is injectible into fk(b), which yields a contradiction as before.
This shows that aggressive pairs cannot exist, whence fk is indeed coherent, which finishes the
proof of the seductivity of k, whereby the problem is finally solved.
46

50th IMO 2009 Geometry G1
Geometry
G1 BEL (Belgium)
Let ABC be a triangle with AB = AC. The angle bisectors of A and B meet the sides BC
and AC in D and E, respectively. Let K be the incenter of triangle ADC. Suppose that
∠BEK = 45◦
. Find all possible values of ∠BAC.
Solution 1. Answer: ∠BAC = 60◦
or ∠BAC = 90◦
are possible values and the only possible
values.
Let I be the incenter of triangle ABC, then K lies on the line CI. Let F be the point, where
the incircle of triangle ABC touches the side AC; then the segments IF and ID have the same
length and are perpendicular to AC and BC, respectively.
A
B CD
E=F
I
KP
Q
R
S
A
B CD
E
F
I
KP
Q
RS
Figure 1 Figure 2
Let P, Q and R be the points where the incircle of triangle ADC touches the sides AD, DC
and CA, respectively. Since K and I lie on the angle bisector of ∠ACD, the segments ID and
IF are symmetric with respect to the line IC. Hence there is a point S on IF where the incircle
of triangle ADC touches the segment IF. Then segments KP, KQ, KR and KS all have the
same length and are perpendicular to AD, DC, CA and IF, respectively. So – regardless of
the value of ∠BEK – the quadrilateral KRFS is a square and ∠SFK = ∠KFC = 45◦
.
Consider the case ∠BAC = 60◦
(see Figure 1). Then triangle ABC is equilateral. Furthermore
we have F = E, hence ∠BEK = ∠IFK = ∠SEK = 45◦
. So 60◦
is a possible value for ∠BAC.
Now consider the case ∠BAC = 90◦
(see Figure 2). Then ∠CBA = ∠ACB = 45◦
. Fur-
thermore, ∠KIE = 1
2
∠CBA + 1
2
∠ACB = 45◦
, ∠AEB = 180◦
− 90◦
− 22.5◦
= 67.5◦
and
∠EIA = ∠BID = 180◦
− 90◦
− 22.5◦
= 67.5◦
. Hence triangle IEA is isosceles and a reﬂection
of the bisector of ∠IAE takes I to E and K to itself. So triangle IKE is symmetric with
respect to this axis, i.e. ∠KIE = ∠IEK = ∠BEK = 45◦
. So 90◦
is a possible value for
∠BAC, too.
If, on the other hand, ∠BEK = 45◦
then ∠BEK = ∠IEK = ∠IFK = 45◦
. Then
• either F = E, which makes the angle bisector BI be an altitude, i.e., which makes triangle
ABC isosceles with base AC and hence equilateral and so ∠BAC = 60◦
,
• or E lies between F and C, which makes the points K, E, F and I concyclic, so 45◦
=
∠KFC = ∠KFE = ∠KIE = ∠CBI + ∠ICB = 2 · ∠ICB = 90◦
− 1
2
∠BAC, and so
∠BAC = 90◦
,
47

G1 Geometry 50th IMO 2009
• or F lies between E and C, then again, K, E, F and I are concyclic, so 45◦
= ∠KFC =
180◦
− ∠KFE = ∠KIE, which yields the same result ∠BAC = 90◦
. (However, for
∠BAC = 90◦
E lies, in fact, between F and C, see Figure 2. So this case does not
occur.)
This proves 90◦
and 60◦
to be the only possible values for ∠BAC.
Solution 2. Denote angles at A, B and C as usual by α, β and γ. Since triangle ABC is
isosceles, we have β = γ = 90◦
− α
2
< 90◦
, so ∠ECK = 45◦
− α
4
= ∠KCD. Since K is the
incenter of triangle ADC, we have ∠CDK = ∠KDA = 45◦
; furthermore ∠DIC = 45◦
+ α
4
.
Now, if ∠BEK = 45◦
, easy calculations within triangles BCE and KCE yield
∠KEC = 180◦
− β
2
− 45◦
− β = 135◦
− 3
2
β = 3
2
(90◦
− β) = 3
4
α,
∠IKE = 3
4
α + 45◦
− α
4
= 45◦
+ α
2
.
So in triangles ICE, IKE, IDK and IDC we have (see Figure 3)
IC
IE
=
sin ∠IEC
sin ∠ECI
=
sin(45◦
+ 3
4
α)
sin(45◦
− α
4
)
,
IE
IK
=
sin ∠EKI
sin ∠IEK
=
sin(45◦
+ α
2
)
sin 45◦ ,
IK
ID
=
sin ∠KDI
sin ∠IKD
=
sin 45◦
sin(90◦
− α
4
)
,
ID
IC
=
sin ∠ICD
sin ∠CDI
=
sin(45◦
− α
4
)
sin 90◦ .
A
B CD
E
I
K
α
2
3
α
4
45˚
45˚
α
4
45˚
45˚
45˚
α
4
45˚
α
2
β
Figure 3
Multiplication of these four equations yields
1 =
sin(45◦
+ 3
4
α) sin(45◦
+ α
2
)
sin(90◦
− α
4
)
.
But, since
sin (90◦
− α
4
) = cos α
4
= cos (45◦
+ 3
4
α) − (45◦
+ α
2
)
= cos 45◦
+ 3
4
α cos (45◦
+ α
2
) + sin (45◦
+ 3
4
α) sin (45◦
+ α
2
),
this is equivalent to
sin(45◦
+ 3
4
α) sin(45◦
+ α
2
) = cos (45◦
+ 3
4
α) cos (45◦
+ α
2
) + sin (45◦
+ 3
4
α) sin (45◦
+ α
2
)
and ﬁnally
cos (45◦
+ 3
4
α) cos (45◦
+ α
2
) = 0.
48

But this means cos (45◦
+ 3
4
α) = 0, hence 45◦
+ 3
4
α = 90◦
, i.e. α = 60◦
or cos (45◦
+ α
2
) = 0,
hence 45◦
+ α
2
= 90◦
, i.e. α = 90◦
. So these values are the only two possible values for α.
On the other hand, both α = 90◦
and α = 60◦
yield ∠BEK = 45◦
, this was shown in
Solution 1.
49

G2 RUS (Russian Federation)
Let ABC be a triangle with circumcenter O. The points P and Q are interior points of the
sides CA and AB, respectively. The circle k passes through the midpoints of the segments BP,
CQ, and PQ. Prove that if the line PQ is tangent to circle k then OP = OQ.
Solution 1. Let K, L, M, B , C be the midpoints of BP, CQ, PQ, CA, and AB, respectively
(see Figure 1). Since CA LM, we have ∠LMP = ∠QPA. Since k touches the segment PQ
at M, we ﬁnd ∠LMP = ∠LKM. Thus ∠QPA = ∠LKM. Similarly it follows from AB MK
that ∠PQA = ∠KLM. Therefore, triangles APQ and MKL are similar, hence
AP
AQ
=
MK
ML
=
QB
2
PC
2
=
QB
PC
. (1)
Now (1) is equivalent to AP · PC = AQ · QB which means that the power of points P and Q
with respect to the circumcircle of ABC are equal, hence OP = OQ.
A
B
B′
C
C′
K
L
M
O
P
Q
k
Figure 1
Comment. The last argument can also be established by the following calculation:
OP2
− OQ2
= OB 2
+ B P2
− OC 2
− C Q2
= (OA2
− AB 2
) + B P2
− (OA2
− AC 2
) − C Q2
= (AC 2
− C Q2
) − (AB 2
− B P2
)
= (AC − C Q)(AC + C Q) − (AB − B P)(AB + B P)
= AQ · QB − AP · PC.
With (1), we conclude OP2
− OQ2
= 0, as desired.
50

Solution 2. Again, denote by K, L, M the midpoints of segments BP, CQ, and PQ, respec-
tively. Let O, S, T be the circumcenters of triangles ABC, KLM, and APQ, respectively (see
Figure 2). Note that MK and LM are the midlines in triangles BPQ and CPQ, respectively, so
−−→
MK = 1
2
−−→
QB and
−−→
ML = 1
2
−→
PC. Denote by prl(−→v ) the projection of vector −→v onto line l. Then
prAB(
−→
OT) = prAB(
−→
OA −
−→
TA) = 1
2
−→
BA − 1
2
−→
QA = 1
2
−−→
BQ =
−−→
KM and prAB(
−−→
SM) = prMK(
−−→
SM) =
1
2
−−→
KM = 1
2
prAB(
−→
OT). Analogously we get prCA(
−−→
SM) = 1
2
prCA(
−→
OT). Since AB and CA are not
parallel, this implies that
−−→
SM = 1
2
−→
OT.
A
B C
K
L
M
O
P
Q
S
T
k
Figure 2
Now, since the circle k touches PQ at M, we get SM ⊥ PQ, hence OT ⊥ PQ. Since T is
equidistant from P and Q, the line OT is a perpendicular bisector of segment PQ, and hence
O is equidistant from P and Q which ﬁnishes the proof.
51

Let ABC be a triangle. The incircle of ABC touches the sides AB and AC at the points Z
and Y , respectively. Let G be the point where the lines BY and CZ meet, and let R and S be
points such that the two quadrilaterals BCY R and BCSZ are parallelograms.
Prove that GR = GS.
Solution 1. Denote by k the incircle and by ka the excircle opposite to A of triangle ABC.
Let k and ka touch the side BC at the points X and T, respectively, let ka touch the lines AB
and AC at the points P and Q, respectively. We use several times the fact that opposing sides
of a parallelogram are of equal length, that points of contact of the excircle and incircle to a
side of a triangle lie symmetric with respect to the midpoint of this side and that segments on
two tangents to a circle deﬁned by the points of contact and their point of intersection have
the same length. So we conclude
ZP = ZB + BP = XB + BT = BX + CX = ZS and
CQ = CT = BX = BZ = CS.
A
B C
G
Ia
P
Q
R
S
TX
Y
Z
k
ka
p
q
xx
y
y
y
z
z
y+z
So for each of the points Z, C, their distances to S equal the length of a tangent segment from
this point to ka. It is well-known, that all points with this property lie on the line ZC, which
is the radical axis of S and ka. Similar arguments yield that BY is the radical axis of R and
ka. So the point of intersection of ZC and BY , which is G by deﬁnition, is the radical center
of R, S and ka, from which the claim GR = GS follows immediately.
Solution 2. Denote x = AZ = AY , y = BZ = BX, z = CX = CY , p = ZG, q = GC.
Several lengthy calculations (Menelaos’ theorem in triangle AZC, law of Cosines in triangles
ABC and AZC and Stewart’s theorem in triangle ZCS) give four equations for p, q, cos α
52

and GS in terms of x, y, and z that can be resolved for GS. The result is symmetric in y and
z, so GR = GS. More in detail this means:
The line BY intersects the sides of triangle AZC, so Menelaos’ theorem yields p
q
· z
x
· x+y
y
= 1,
hence
p
q
=
xy
yz + zx
. (1)
Since we only want to show that the term for GS is symmetric in y and z, we abbreviate terms
that are symmetric in y and z by capital letters, starting with N = xy +yz +zx. So (1) implies
p
p + q
=
xy
xy + yz + zx
=
xy
N
and
q
p + q
=
yz + zx
xy + yz + zx
=
yz + zx
N
. (2)
Now the law of Cosines in triangle ABC yields
cos α =
(x + y)2
+ (x + z)2
− (y + z)2
2(x + y)(x + z)
=
2x2
+ 2xy + 2xz − 2yz
2(x + y)(x + z)
= 1 −
2yz
(x + y)(x + z)
.
We use this result to apply the law of Cosines in triangle AZC:
(p + q)2
= x2
+ (x + z)2
− 2x(x + z) cos α
= x2
+ (x + z)2
− 2x(x + z) · 1 −
2yz
(x + y)(x + z)
= z2
+
4xyz
x + y
. (3)
Now in triangle ZCS the segment GS is a cevian, so with Stewart’s theorem we have
py2
+ q(y + z)2
= (p + q)(GS2
+ pq), hence
GS2
=
p
p + q
· y2
+
q
p + q
· (y + z)2
−
p
p + q
·
q
p + q
· (p + q)2
.
Replacing the p’s and q’s herein by (2) and (3) yields
GS2
=
xy
N
y2
+
yz + zx
N
(y + z)2
−
xy
N
·
yz + zx
N
· z2
+
4xyz
x + y
=
xy3
N
+
yz(y + z)2
N
M1
+
zx(y + z)2
N
−
xyz3
(x + y)
N2
−
4x2
y2
z2
N2
M2
=
xy3
+ zx(y + z)2
N
−
xyz3
(x + y)
N2
+ M1 − M2
=
x(y3
+ y2
z + yz2
+ z3
)
N
M3
+
xyz2
N
N2
−
xyz3
(x + y)
N2
+ M1 − M2
=
x2
y2
z2
+ xy2
z3
+ x2
yz3
− x2
yz3
− xy2
z3
N2
+ M1 − M2 + M3
=
x2
y2
z2
N2
+ M1 − M2 + M3,
a term that is symmetric in y and z, indeed.
Comment. G is known as Gergonne’s point of ABC.
53

G4 UNK (United Kingdom)
Given a cyclic quadrilateral ABCD, let the diagonals AC and BD meet at E and the lines AD
and BC meet at F. The midpoints of AB and CD are G and H, respectively. Show that EF
is tangent at E to the circle through the points E, G, and H.
Solution 1. It suﬃces to show that ∠HEF = ∠HGE (see Figure 1), since in circle EGH the
angle over the chord EH at G equals the angle between the tangent at E and EH.
First, ∠BAD = 180◦
−∠DCB = ∠FCD. Since triangles FAB and FCD have also a common
interior angle at F, they are similar.
A
B
C
D
E
F
G
H M
X
Y
Figure 1
Denote by T the transformation consisting of a reﬂection at the bisector of ∠DFC followed
by a dilation with center F and factor of FA
FC
. Then T maps F to F, C to A, D to B, and H
to G. To see this, note that FCA ∼ FDB, so FA
FC
= FB
FD
. Moreover, as ∠ADB = ∠ACB,
the image of the line DE under T is parallel to AC (and passes through B) and similarly the
image of CE is parallel to DB and passes through A. Hence E is mapped to the point X which
is the fourth vertex of the parallelogram BEAX. Thus, in particular ∠HEF = ∠FXG.
As G is the midpoint of the diagonal AB of the parallelogram BEAX, it is also the midpoint
of EX. In particular, E, G, X are collinear, and EX = 2 · EG.
Denote by Y the fourth vertex of the parallelogram DECY . By an analogous reasoning as
before, it follows that T maps Y to E, thus E, H, Y are collinear with EY = 2 · EH.
Therefore, by the intercept theorem, HG XY .
From the construction of T it is clear that the lines FX and FE are symmetric with respect
to the bisector of ∠DFC, as are FY and FE. Thus, F, X, Y are collinear, which together
with HG XY implies ∠FXE = ∠HGE. This completes the proof.
54

Solution 2. We use the following
Lemma (Gauß). Let ABCD be a quadrilateral. Let AB and CD intersect at P, and BC
and DA intersect at Q. Then the midpoints K, L, M of AC, BD, and PQ, respectively, are
collinear.
Proof: Let us consider the points Z that fulfill the equation
(ABZ) + (CDZ) = (BCZ) + (DAZ), (1)
where (RST) denotes the oriented area of the triangle RST (see Figure 2).
A
B
C
D
K
L
M
P
Q
Figure 2
As (1) is linear in Z, it can either characterize a line, or be contradictory, or be trivially fulfilled
for all Z in the plane. If (1) was fulfilled for all Z, then it would hold for Z = A, Z = B, which
gives (CDA) = (BCA), (CDB) = (DAB), respectively, i.e. the diagonals of ABCD would
bisect each other, thus ABCD would be a parallelogram. This contradicts the hypothesis that
AD and BC intersect. Since E, F, G fulfill (1), it is the equation of a line which completes the
proof of the lemma.
Now consider the parallelograms EAXB and ECY D (see Figure 1). Then G, H are the
midpoints of EX, EY , respectively. Let M be the midpoint of EF. By applying the Lemma to
the (re-entrant) quadrilateral ADBC, it is evident that G, H, and M are collinear. A dilation
by a factor of 2 with center E shows that X, Y , F are collinear. Since AX DE and BX CE,
we have pairwise equal interior angles in the quadrilaterals FDEC and FBXA. Since we have
also ∠EBA = ∠DCA = ∠CDY , the quadrilaterals are similar. Thus, ∠FXA = ∠CEF.
Clearly the parallelograms ECY D and EBXA are similar, too, thus ∠EXA = ∠CEY . Con-
sequently, ∠FXE = ∠FXA − ∠EXA = ∠CEF − ∠CEY = ∠Y EF. By the converse of the
tangent-chord angle theorem EF is tangent to the circle XEY . A dilation by a factor of 1
2
completes the proof.
55

Solution 3. As in Solution 2, G, H, M are proven to be collinear. It suﬃces to show that
ME2
= MG · MH. If p =
−→
OP denotes the vector from circumcenter O to point P, the claim
becomes
e − f
2
2
=
e + f
2
−
a + b
2
e + f
2
−
c + d
2
,
or equivalently
4 ef − (e + f)(a + b + c + d) + (a + b)(c + d) = 0. (2)
With R as the circumradius of ABCD, we obtain for the powers P(E) and P(F) of E and F,
respectively, with respect to the circumcircle
P(E) = (e − a)(e − c) = (e − b)(e − d) = e2
− R2
,
P(F) = (f − a)(f − d) = (f − b)(f − c) = f2
− R2
,
hence
(e − a)(e − c) = e2
− R2
, (3)
(e − b)(e − d) = e2
− R2
, (4)
(f − a)(f − d) = f2
− R2
, (5)
(f − b)(f − c) = f2
− R2
. (6)
Since F lies on the polar to E with respect to the circumcircle, we have
4 ef = 4R2
. (7)
Adding up (3) to (7) yields (2), as desired.
56

G5 POL (Poland)
Let P be a polygon that is convex and symmetric to some point O. Prove that for some
parallelogram R satisfying P ⊂ R we have
|R|
|P|
≤
√
2
where |R| and |P| denote the area of the sets R and P, respectively.
Solution 1. We will construct two parallelograms R1 and R3, each of them containing P, and
prove that at least one of the inequalities |R1| ≤
√
2 |P| and |R3| ≤
√
2 |P| holds (see Figure 1).
First we will construct a parallelogram R1 ⊇ P with the property that the midpoints of the
sides of R1 are points of the boundary of P.
Choose two points A and B of P such that the triangle OAB has maximal area. Let a be the
line through A parallel to OB and b the line through B parallel to OA. Let A , B , a and b be
the points or lines, that are symmetric to A, B, a and b, respectively, with respect to O. Now
let R1 be the parallelogram deﬁned by a, b, a and b .
A
A′
BB′
C
D
O
X
X′
X
Y
Y′
a
a′
b
b′
a
R1
R2
R3
*
*
Figure 1
Obviously, A and B are located on the boundary of the polygon P, and A, B, A and B are
midpoints of the sides of R1. We note that P ⊆ R1. Otherwise, there would be a point Z ∈ P
but Z /∈ R1, i.e., one of the lines a, b, a or b were between O and Z. If it is a, we have
|OZB| > |OAB|, which is contradictory to the choice of A and B. If it is one of the lines b, a
or b almost identical arguments lead to a similar contradiction.
Let R2 be the parallelogram ABA B . Since A and B are points of P, segment AB ⊂ P and
so R2 ⊂ R1. Since A, B, A and B are midpoints of the sides of R1, an easy argument yields
|R1| = 2 · |R2|. (1)
Let R3 be the smallest parallelogram enclosing P deﬁned by lines parallel to AB and BA .
Obviously R2 ⊂ R3 and every side of R3 contains at least one point of the boundary of P.
Denote by C the intersection point of a and b, by X the intersection point of AB and OC, and
by X the intersection point of XC and the boundary of R3. In a similar way denote by D
57

the intersection point of b and a , by Y the intersection point of A B and OD, and by Y the
intersection point of Y D and the boundary of R3.
Note that OC = 2·OX and OD = 2·OY , so there exist real numbers x and y with 1 ≤ x, y ≤ 2
and OX = x · OX and OY = y · OY . Corresponding sides of R3 and R2 are parallel which
yields
|R3| = xy · |R2|. (2)
The side of R3 containing X contains at least one point X∗
of P; due to the convexity of
P we have AX∗
B ⊂ P. Since this side of the parallelogram R3 is parallel to AB we have
|AX∗
B| = |AX B|, so |OAX B| does not exceed the area of P confined to the sector defined
by the rays OB and OA. In a similar way we conclude that |OB Y A | does not exceed the
area of P confined to the sector defined by the rays OB and OA . Putting things together we
have |OAX B| = x · |OAB|, |OBDA | = y · |OBA |. Since |OAB| = |OBA |, we conclude that
|P| ≥ 2·|AX BY A | = 2·(x · |OAB| + y · |OBA |) = 4· x+y
2
·|OAB| = x+y
2
·R2; this is in short
x + y
2
· |R2| ≤ |P|. (3)
Since all numbers concerned are positive, we can combine (1)–(3). Using the arithmetic-
geometric-mean inequality we obtain
|R1| · |R3| = 2 · |R2| · xy · |R2| ≤ 2 · |R2|2 x + y
2
2
≤ 2 · |P|2
.
This implies immediately the desired result |R1| ≤
√
2 · |P| or |R3| ≤
√
2 · |P|.
Solution 2. We construct the parallelograms R1, R2 and R3 in the same way as in Solution
1 and will show that |R1|
|P|
≤
√
2 or |R3|
|P|
≤
√
2.
A
A′
BB′
R1
R2
R3
a
bc
Figure 2
Recall that affine one-to-one maps of the plane preserve the ratio of areas of subsets of the
plane. On the other hand, every parallelogram can be transformed with an affine map onto
a square. It follows that without loss of generality we may assume that R1 is a square (see
Figure 2).
Then R2, whose vertices are the midpoints of the sides of R1, is a square too, and R3, whose
sides are parallel to the diagonals of R1, is a rectangle.
Let a > 0, b ≥ 0 and c ≥ 0 be the distances introduced in Figure 2. Then |R1| = 2a2
and
58

|R3| = (a + 2b)(a + 2c).
Points A, A , B and B are in the convex polygon P. Hence the square ABA B is a subset of
P. Moreover, each of the sides of the rectangle R3 contains a point of P, otherwise R3 would
not be minimal. It follows that
|P| ≥ a2
+ 2 ·
ab
2
+ 2 ·
ac
2
= a(a + b + c).
Now assume that both |R1|
|P|
>
√
2 and |R3|
|P|
>
√
2, then
2a2
= |R1| >
√
2 · |P| ≥
√
2 · a(a + b + c)
and
(a + 2b)(a + 2c) = |R3| >
√
2 · |P| ≥
√
2 · a(a + b + c).
All numbers concerned are positive, so after multiplying these inequalities we get
2a2
(a + 2b)(a + 2c) > 2a2
(a + b + c)2
.
But the arithmetic-geometric-mean inequality implies the contradictory result
2a2
(a + 2b)(a + 2c) ≤ 2a2 (a + 2b) + (a + 2c)
2
2
= 2a2
(a + b + c)2
.
Hence |R1|
|P|
≤
√
2 or |R3|
|P|
≤
√
2, as desired.
59

G6 UKR (Ukraine)
Let the sides AD and BC of the quadrilateral ABCD (such that AB is not parallel to CD)
intersect at point P. Points O1 and O2 are the circumcenters and points H1 and H2 are the
orthocenters of triangles ABP and DCP, respectively. Denote the midpoints of segments
O1H1 and O2H2 by E1 and E2, respectively. Prove that the perpendicular from E1 on CD, the
perpendicular from E2 on AB and the line H1H2 are concurrent.
Solution 1. We keep triangle ABP fixed and move the line CD parallel to itself uniformly,
i.e. linearly dependent on a single parameter λ (see Figure 1). Then the points C and D also
move uniformly. Hence, the points O2, H2 and E2 move uniformly, too. Therefore also the
perpendicular from E2 on AB moves uniformly. Obviously, the points O1, H1, E1 and the
perpendicular from E1 on CD do not move at all. Hence, the intersection point S of these
two perpendiculars moves uniformly. Since H1 does not move, while H2 and S move uniformly
along parallel lines (both are perpendicular to CD), it is sufficient to prove their collinearity
for two different positions of CD.
A B
C
D
E1
E2
H1
H2
O1
O2
P
S
Figure 1
Let CD pass through either point A or point B. Note that by hypothesis these two cases
are different. We will consider the case A ∈ CD, i.e. A = D. So we have to show that the
perpendiculars from E1 on AC and from E2 on AB intersect on the altitude AH of triangle
ABC (see Figure 2).
60

A=D
A1
B
B1
C
C1
E1
E2
H
H1
H2
O1
O2
P
S
Figure 2
To this end, we consider the midpoints A1, B1, C1 of BC, CA, AB, respectively. As E1 is the
center of Feuerbach’s circle (nine-point circle) of ABP, we have E1C1 = E1H. Similarly,
E2B1 = E2H. Note further that a point X lies on the perpendicular from E1 on A1C1 if and
only if
XC2
1 − XA2
1 = E1C2
1 − E1A2
1.
Similarly, the perpendicular from E2 on A1B1 is characterized by
XA2
1 − XB2
1 = E2A2
1 − E2B2
1.
The line H1H2, which is perpendicular to B1C1 and contains A, is given by
XB2
1 − XC2
1 = AB2
1 − AC2
1 .
The three lines are concurrent if and only if
0 = XC2
1 − XA2
1 + XA2
1 − XB2
1 + XB2
1 − XC2
1
= E1C2
1 − E1A2
1 + E2A2
1 − E2B2
1 + AB2
1 − AC2
1
= −E1A2
1 + E2A2
1 + E1H2
− E2H2
+ AB2
1 − AC2
1 ,
i.e. it suﬃces to show that
E1A2
1 − E2A2
1 − E1H2
+ E2H2
=
AC2
− AB2
4
.
We have
AC2
− AB2
4
=
HC2
− HB2
4
=
(HC + HB)(HC − HB)
4
=
HA1 · BC
2
.
Let F1, F2 be the projections of E1, E2 on BC. Obviously, these are the midpoints of HP1,
61

HP2, where P1, P2 are the midpoints of PB and PC respectively. Then
E1A2
1 − E2A2
1 − E1H2
+ E2H2
= F1A2
1 − F1H2
− F2A2
1 + F2H2
= (F1A1 − F1H)(F1A1 + F1H) − (F2A1 − F2H)(F2A1 + F2H)
= A1H · (A1P1 − A1P2)
=
A1H · BC
2
=
AC2
− AB2
4
,
which proves the claim.
Solution 2. Let the perpendicular from E1 on CD meet PH1 at X, and the perpendicular
from E2 on AB meet PH2 at Y (see Figure 3). Let ϕ be the intersection angle of AB and CD.
Denote by M, N the midpoints of PH1, PH2 respectively.
A
B
C
D
E1
E2
H1
H2
M
N
P
Q
X
Y
α
β
ϕ
ψ
Figure 3
We will prove now that triangles E1XM and E2Y N have equal angles at E1, E2, and supple-
mentary angles at X, Y .
In the following, angles are understood as oriented, and equalities of angles modulo 180◦
.
Let α = ∠H2PD, ψ = ∠DPC, β = ∠CPH1. Then α + ψ + β = ϕ, ∠E1XH1 = ∠H2Y E2 = ϕ,
thus ∠MXE1 + ∠NY E2 = 180◦
.
By considering the Feuerbach circle of ABP whose center is E1 and which goes through M,
we have ∠E1MH1 = ψ + 2β. Analogous considerations with the Feuerbach circle of DCP
yield ∠H2NE2 = ψ + 2α. Hence indeed ∠XE1M = ϕ − (ψ + 2β) = (ψ + 2α) − ϕ = ∠Y E2N.
It follows now that
XM
ME1
=
Y N
NE2
.
Furthermore, ME1 is half the circumradius of ABP, while PH1 is the distance of P to the
orthocenter of that triangle, which is twice the circumradius times the cosine of ψ. Together
62

with analogous reasoning for DCP we have
ME1
PH1
=
1
4 cos ψ
=
NE2
PH2
.
By multiplication,
XM
PH1
=
Y N
PH2
,
and therefore
PX
XH1
=
H2Y
Y P
.
Let E1X, E2Y meet H1H2 in R, S respectively.
Applying the intercept theorem to the parallels E1X, PH2 and center H1 gives
H2R
RH1
=
PX
XH1
,
while with parallels E2Y , PH1 and center H2 we obtain
H2S
SH1
=
H2Y
Y P
.
Combination of the last three equalities yields that R and S coincide.
63

Let ABC be a triangle with incenter I and let X, Y and Z be the incenters of the triangles
BIC, CIA and AIB, respectively. Let the triangle XY Z be equilateral. Prove that ABC is
equilateral too.
Solution. AZ, AI and AY divide ∠BAC into four equal angles; denote them by α. In
the same way we have four equal angles β at B and four equal angles γ at C. Obviously
α + β + γ = 180◦
4
= 45◦
; and 0◦
< α, β, γ < 45◦
.
A
B C
I
X
YZ
α
β
γ
Easy calculations in various triangles yield ∠BIC = 180◦
− 2β − 2γ = 180◦
− (90◦
− 2α) =
90◦
+2α, hence (for X is the incenter of triangle BCI, so IX bisects ∠BIC) we have ∠XIC =
∠BIX = 1
2
∠BIC = 45◦
+ α and with similar aguments ∠CIY = ∠Y IA = 45◦
+ β and
∠AIZ = ∠ZIB = 45◦
+ γ. Furthermore, we have ∠XIY = ∠XIC + ∠CIY = (45◦
+ α) +
(45◦
+ β) = 135◦
− γ, ∠Y IZ = 135◦
− α, and ∠ZIX = 135◦
− β.
Now we calculate the lengths of IX, IY and IZ in terms of α, β and γ. The perpendicular
from I on CX has length IX · sin ∠CXI = IX · sin (90◦
+ β) = IX · cos β. But CI bisects
∠Y CX, so the perpendicular from I on CY has the same length, and we conclude
IX · cos β = IY · cos α.
To make calculations easier we choose a length unit that makes IX = cos α. Then IY = cos β
and with similar arguments IZ = cos γ.
Since XY Z is equilateral we have ZX = ZY . The law of Cosines in triangles XY I, Y ZI yields
ZX2
= ZY 2
=⇒ IZ2
+ IX2
− 2 · IZ · IX · cos ∠ZIX = IZ2
+ IY 2
− 2 · IZ · IY · cos ∠Y IZ
=⇒ IX2
− IY 2
= 2 · IZ · (IX · cos ∠ZIX − IY · cos ∠Y IZ)
=⇒ cos 2
α − cos 2
β
L.H.S.
= 2 · cos γ · (cos α · cos (135◦
− β) − cos β · cos (135◦
− α))
R.H.S.
.
A transformation of the left-hand side (L.H.S.) yields
L.H.S. = cos 2
α · sin 2
β + cos 2
β − cos 2
β · sin 2
α + cos 2
α
= cos 2
α · sin 2
β − cos 2
β · sin 2
α
64

= (cos α · sin β + cos β · sin α) · (cos α · sin β − cos β · sin α)
= sin (β + α) · sin (β − α) = sin (45◦
− γ) · sin (β − α)
whereas a transformation of the right-hand side (R.H.S.) leads to
R.H.S. = 2 · cos γ · (cos α · (− cos (45◦
+ β)) − cos β · (− cos (45◦
+ α)))
= 2 ·
√
2
2
· cos γ · (cos α · (sin β − cos β) + cos β · (cos α − sin α))
=
√
2 · cos γ · (cos α · sin β − cos β · sin α)
=
√
2 · cos γ · sin (β − α).
Equating L.H.S. and R.H.S. we obtain
sin (45◦
− γ) · sin (β − α) =
√
2 · cos γ · sin (β − α)
=⇒ sin (β − α) ·
√
2 · cos γ − sin (45◦
− γ) = 0
=⇒ α = β or
√
2 · cos γ = sin (45◦
− γ).
But γ < 45◦
; so
√
2 · cos γ > cos γ > cos 45◦
= sin 45◦
> sin(45◦
− γ). This leaves α = β.
With similar reasoning we have α = γ, which means triangle ABC must be equilateral.
65

G8 BGR (Bulgaria)
Let ABCD be a circumscribed quadrilateral. Let g be a line through A which meets the
segment BC in M and the line CD in N. Denote by I1, I2, and I3 the incenters of ABM,
MNC, and NDA, respectively. Show that the orthocenter of I1I2I3 lies on g.
Solution 1. Let k1, k2 and k3 be the incircles of triangles ABM, MNC, and NDA, respec-
tively (see Figure 1). We shall show that the tangent h from C to k1 which is different from
CB is also tangent to k3.
A
B
C
D
H
I1
I2
I3
L1
L3
M
N
X
g
h
k1
k2
k3
Figure 1
To this end, let X denote the point of intersection of g and h. Then ABCX and ABCD are
circumscribed quadrilaterals, whence
CD − CX = (AB + CD) − (AB + CX) = (BC + AD) − (BC + AX) = AD − AX,
i.e.
AX + CD = CX + AD
which in turn reveals that the quadrilateral AXCD is also circumscribed. Thus h touches
indeed the circle k3.
Moreover, we find that ∠I3CI1 = ∠I3CX + ∠XCI1 = 1
2
(∠DCX + ∠XCB) = 1
2
∠DCB =
1
2
(180◦
− ∠MCN) = 180◦
− ∠MI2N = ∠I3I2I1, from which we conclude that C, I1, I2, I3 are
concyclic.
Let now L1 and L3 be the reflection points of C with respect to the lines I2I3 and I1I2 respec-
tively. Since I1I2 is the angle bisector of ∠NMC, it follows that L3 lies on g. By analogous
reasoning, L1 lies on g.
Let H be the orthocenter of I1I2I3. We have ∠I2L3I1 = ∠I1CI2 = ∠I1I3I2 = 180◦
− ∠I1HI2,
which entails that the quadrilateral I2HI1L3 is cyclic. Analogously, I3HL1I2 is cyclic.
66

Then, working with oriented angles modulo 180◦
, we have
∠L3HI2 = ∠L3I1I2 = ∠I2I1C = ∠I2I3C = ∠L1I3I2 = ∠L1HI2,
whence L1, L3, and H are collinear. By L1 = L3, the claim follows.
Comment. The last part of the argument essentially reproves the following fact: The Simson
line of a point P lying on the circumcircle of a triangle ABC with respect to that triangle bisects
the line segment connecting P with the orthocenter of ABC.
Solution 2. We start by proving that C, I1, I2, and I3 are concyclic.
AB
C
D
I
I1
I2
I3
M
N
Z
g
αβ
γ
δ
Figure 2
To this end, notice ﬁrst that I2, M, I1 are collinear, as are N, I2, I3 (see Figure 2). Denote by
α, β, γ, δ the internal angles of ABCD. By considerations in triangle CMN, it follows that
∠I3I2I1 = γ
2
. We will show that ∠I3CI1 = γ
2
, too. Denote by I the incenter of ABCD. Clearly,
I1 ∈ BI, I3 ∈ DI, ∠I1AI3 = α
2
.
Using the abbreviation [X, Y Z] for the distance from point X to the line Y Z, we have because
of ∠BAI1 = ∠IAI3 and ∠I1AI = ∠I3AD that
[I1, AB]
[I1, AI]
=
[I3, AI]
[I3, AD]
.
Furthermore, consideration of the angle sums in AIB, BIC, CID and DIA implies ∠AIB +
∠CID = ∠BIC + ∠DIA = 180◦
, from which we see
[I1, AI]
[I3, CI]
=
I1I
I3I
=
[I1, CI]
[I3, AI]
.
Because of [I1, AB] = [I1, BC], [I3, AD] = [I3, CD], multiplication yields
[I1, BC]
[I3, CI]
=
[I1, CI]
[I3, CD]
.
By ∠DCI = ∠ICB = γ/2 it follows that ∠I1CB = ∠I3CI which concludes the proof of the
67

above statement.
Let the perpendicular from I1 on I2I3 intersect g at Z. Then ∠MI1Z = 90◦
− ∠I3I2I1 =
90◦
− γ/2 = ∠MCI2. Since we have also ∠ZMI1 = ∠I2MC, triangles MZI1 and MI2C are
similar. From this one easily proves that also MI2Z and MCI1 are similar. Because C, I1, I2,
and I3 are concyclic, ∠MZI2 = ∠MI1C = ∠NI3C, thus NI2Z and NCI3 are similar, hence
NCI2 and NI3Z are similar. We conclude ∠ZI3I2 = ∠I2CN = 90◦
− γ/2, hence I1I2 ⊥ ZI3.
This completes the proof.
68

50th IMO 2009 Number Theory N1
Number Theory
N1 AUS (Australia)
A social club has n members. They have the membership numbers 1, 2, . . . , n, respectively.
From time to time members send presents to other members, including items they have already
received as presents from other members. In order to avoid the embarrassing situation that a
member might receive a present that he or she has sent to other members, the club adds the
following rule to its statutes at one of its annual general meetings:
“A member with membership number a is permitted to send a present to a member with
membership number b if and only if a(b − 1) is a multiple of n.”
Prove that, if each member follows this rule, none will receive a present from another member
that he or she has already sent to other members.
Alternative formulation: Let G be a directed graph with n vertices v1, v2, . . . , vn, such that
there is an edge going from va to vb if and only if a and b are distinct and a(b − 1) is a multiple
of n. Prove that this graph does not contain a directed cycle.
Solution 1. Suppose there is an edge from vi to vj. Then i(j − 1) = ij − i = kn for some
integer k, which implies i = ij−kn. If gcd(i, n) = d and gcd(j, n) = e, then e divides ij−kn = i
and thus e also divides d. Hence, if there is an edge from vi to vj, then gcd(j, n)| gcd(i, n).
If there is a cycle in G, say vi1 → vi2 → · · · → vir → vi1 , then we have
gcd(i1, n)| gcd(ir, n)| gcd(ir−1, n)| . . . | gcd(i2, n)| gcd(i1, n),
which implies that all these greatest common divisors must be equal, say be equal to t.
Now we pick any of the ik, without loss of generality let it be i1. Then ir(i1 −1) is a multiple of
n and hence also (by dividing by t), i1 − 1 is a multiple of n
t
. Since i1 and i1 − 1 are relatively
prime, also t and n
t
are relatively prime. So, by the Chinese remainder theorem, the value of
i1 is uniquely determined modulo n = t · n
t
by the value of t. But, as i1 was chosen arbitrarily
among the ik, this implies that all the ik have to be equal, a contradiction.
Solution 2. If a, b, c are integers such that ab − a and bc − b are multiples of n, then also
ac − a = a(bc − b) + (ab − a) − (ab − a)c is a multiple of n. This implies that if there is an
edge from va to vb and an edge from vb to vc, then there also must be an edge from va to vc.
Therefore, if there are any cycles at all, the smallest cycle must have length 2. But suppose
the vertices va and vb form such a cycle, i. e., ab − a and ab − b are both multiples of n. Then
a − b is also a multiple of n, which can only happen if a = b, which is impossible.
Solution 3. Suppose there was a cycle vi1 → vi2 → · · · → vir → vi1 . Then i1(i2 − 1)
is a multiple of n, i. e., i1 ≡ i1i2 mod n. Continuing in this manner, we get i1 ≡ i1i2 ≡
i1i2i3 ≡ i1i2i3 . . . ir mod n. But the same holds for all ik, i. e., ik ≡ i1i2i3 . . . ir mod n. Hence
i1 ≡ i2 ≡ · · · ≡ ir mod n, which means i1 = i2 = · · · = ir, a contradiction.
69

N1 Number Theory 50th IMO 2009
Solution 4. Let n = k be the smallest value of n for which the corresponding graph has a
cycle. We show that k is a prime power.
If k is not a prime power, it can be written as a product k = de of relatively prime integers
greater than 1. Reducing all the numbers modulo d yields a single vertex or a cycle in the
corresponding graph on d vertices, because if a(b − 1) ≡ 0 mod k then this equation also holds
modulo d. But since the graph on d vertices has no cycles, by the minimality of k, we must
have that all the indices of the cycle are congruent modulo d. The same holds modulo e and
hence also modulo k = de. But then all the indices are equal, which is a contradiction.
Thus k must be a prime power k = pm
. There are no edges ending at vk, so vk is not contained
in any cycle. All edges not starting at vk end at a vertex belonging to a non-multiple of p, and
all edges starting at a non-multiple of p must end at v1. But there is no edge starting at v1.
Hence there is no cycle.
Solution 5. Suppose there was a cycle vi1 → vi2 → · · · → vir → vi1 . Let q = pm
be a prime
power dividing n. We claim that either i1 ≡ i2 ≡ · · · ≡ ir ≡ 0 mod q or i1 ≡ i2 ≡ · · · ≡ ir ≡
1 mod q.
Suppose that there is an is not divisible by q. Then, as is(is+1 − 1) is a multiple of q, is+1 ≡
1 mod p. Similarly, we conclude is+2 ≡ 1 mod p and so on. So none of the labels is divisible by
p, but since is(is+1 − 1) is a multiple of q = pm
for all s, all is+1 are congruent to 1 modulo q.
This proves the claim.
Now, as all the labels are congruent modulo all the prime powers dividing n, they must all be
equal by the Chinese remainder theorem. This is a contradiction.
70

N2 PER (Peru)
A positive integer N is called balanced, if N = 1 or if N can be written as a product of an
even number of not necessarily distinct primes. Given positive integers a and b, consider the
polynomial P defined by P(x) = (x + a)(x + b).
(a) Prove that there exist distinct positive integers a and b such that all the numbers P(1), P(2),
. . . , P(50) are balanced.
(b) Prove that if P(n) is balanced for all positive integers n, then a = b.
Solution. Define a function f on the set of positive integers by f(n) = 0 if n is balanced and
f(n) = 1 otherwise. Clearly, f(nm) ≡ f(n) + f(m) mod 2 for all positive integers n, m.
(a) Now for each positive integer n consider the binary sequence (f(n+1), f(n+2), . . . , f(n+
50)). As there are only 250
different such sequences there are two different positive integers
a and b such that
(f(a + 1), f(a + 2), . . . , f(a + 50)) = (f(b + 1), f(b + 2), . . . , f(b + 50)).
But this implies that for the polynomial P(x) = (x+a)(x+b) all the numbers P(1), P(2),
. . . , P(50) are balanced, since for all 1 ≤ k ≤ 50 we have f(P(k)) ≡ f(a+k)+f(b+k) ≡
2f(a + k) ≡ 0 mod 2.
(b) Now suppose P(n) is balanced for all positive integers n and a < b. Set n = k(b − a) − a
for sufficiently large k, such that n is positive. Then P(n) = k(k + 1)(b − a)2
, and this
number can only be balanced, if f(k) = f(k + 1) holds. Thus, the sequence f(k) must
become constant for sufficiently large k. But this is not possible, as for every prime p we
have f(p) = 1 and for every square t2
we have f(t2
) = 0.
Hence a = b.
Comment. Given a positive integer k, a computer search for the pairs of positive integers
(a, b), for which P(1), P(2), . . . , P(k) are all balanced yields the following results with
minimal sum a + b and a < b:
k 3 4 5 10 20
(a, b) (2, 4) (6, 11) (8, 14) (20, 34) (1751, 3121)
Therefore, trying to find a and b in part (a) of the problem cannot be done by elementary
calculations.
71

N3 EST (Estonia)
Let f be a non-constant function from the set of positive integers into the set of positive integers,
such that a − b divides f(a) − f(b) for all distinct positive integers a, b. Prove that there exist
infinitely many primes p such that p divides f(c) for some positive integer c.
Solution 1. Denote by vp(a) the exponent of the prime p in the prime decomposition of a.
Assume that there are only finitely many primes p1, p2, . . . , pm that divide some function value
produced of f.
There are infinitely many positive integers a such that vpi
(a) > vpi
(f(1)) for all i = 1, 2, . . . , m,
e.g. a = (p1p2 . . . pm)α
with α sufficiently large. Pick any such a. The condition of the problem
then yields a| (f(a + 1) − f(1)). Assume f(a + 1) = f(1). Then we must have vpi
(f(a + 1)) =
vpi
(f(1)) for at least one i. This yields vpi
(f(a + 1) − f(1)) = min {vpi
(f(a + 1)), vpi
(f(1))} ≤
vp1 (f(1)) < vpi
(a). But this contradicts the fact that a| (f(a + 1) − f(1)).
Hence we must have f(a + 1) = f(1) for all such a.
Now, for any positive integer b and all such a, we have (a + 1 − b)|(f(a + 1) − f(b)), i.e.,
(a + 1 − b)|(f(1) − f(b)). Since this is true for infinitely many positive integers a we must have
f(b) = f(1). Hence f is a constant function, a contradiction. Therefore, our initial assumption
was false and there are indeed infinitely many primes p dividing f(c) for some positive integer
c.
Solution 2. Assume that there are only finitely many primes p1, p2, . . . , pm that divide some
function value of f. Since f is not identically 1, we must have m ≥ 1.
Then there exist non-negative integers α1, . . . , αm such that
f(1) = pα1
1 pα2
2 . . . pαm
m .
We can pick a positive integer r such that f(r) = f(1). Let
M = 1 + pα1+1
1 pα2+1
2 . . . pαm+1
m · (f(r) + r).
Then for all i ∈ {1, . . . , m} we have that pαi+1
i divides M − 1 and hence by the condition of the
problem also f(M) − f(1). This implies that f(M) is divisible by pαi
i but not by pαi+1
i for all i
and therefore f(M) = f(1).
Hence
M − r > pα1+1
1 pα2+1
2 . . . pαm+1
m · (f(r) + r) − r
≥ pα1+1
1 pα2+1
2 . . . pαm+1
m + (f(r) + r) − r
> pα1
1 pα2
2 . . . pαm
m + f(r)
≥ |f(M) − f(r)|.
But since M − r divides f(M) − f(r) this can only be true if f(r) = f(M) = f(1), which
contradicts the choice of r.
Comment. In the case that f is a polynomial with integer coefficients the result is well-known,
see e.g. W. Schwarz, Einführung in die Methoden der Primzahltheorie, 1969.
72

N4 PRK (Democratic People’s Republic of Korea)
Find all positive integers n such that there exists a sequence of positive integers a1, a2, . . . , an
satisfying
ak+1 =
a2
k + 1
ak−1 + 1
− 1
for every k with 2 ≤ k ≤ n − 1.
Solution 1. Such a sequence exists for n = 1, 2, 3, 4 and no other n. Since the existence of
such a sequence for some n implies the existence of such a sequence for all smaller n, it suﬃces
to prove that n = 5 is not possible and n = 4 is possible.
Assume ﬁrst that for n = 5 there exists a sequence of positive integers a1, a2, . . . , a5 satisfying
the conditions
a2
2 + 1 = (a1 + 1)(a3 + 1),
a2
3 + 1 = (a2 + 1)(a4 + 1),
a2
4 + 1 = (a3 + 1)(a5 + 1).
Assume a1 is odd, then a2 has to be odd as well and as then a2
2 + 1 ≡ 2 mod 4, a3 has to be
even. But this is a contradiction, since then the even number a2 + 1 cannot divide the odd
number a2
3 + 1.
Hence a1 is even.
If a2 is odd, a2
3 + 1 is even (as a multiple of a2 + 1) and hence a3 is odd, too. Similarly we must
have a4 odd as well. But then a2
3 + 1 is a product of two even numbers (a2 + 1)(a4 + 1) and
thus is divisible by 4, which is a contradiction as for odd a3 we have a2
3 + 1 ≡ 2 mod 4.
Hence a2 is even. Furthermore a3 +1 divides the odd number a2
2 +1 and so a3 is even. Similarly,
a4 and a5 are even as well.
Now set x = a2 and y = a3. From the given condition we get (x+1)|(y2
+1) and (y+1)|(x2
+1).
We will prove that there is no pair of positive even numbers (x, y) satisfying these two conditions,
thus yielding a contradiction to the assumption.
Assume there exists a pair (x0, y0) of positive even numbers satisfying the two conditions
(x0 + 1)|(y2
0 + 1) and (y0 + 1)|(x2
0 + 1).
Then one has (x0 + 1)|(y2
0 + 1 + x2
0 − 1), i.e., (x0 + 1)|(x2
0 + y2
0), and similarly (y0 + 1)|(x2
0 + y2
0).
Any common divisor d of x0 + 1 and y0 + 1 must hence also divide the number
(x2
0 + 1) + (y2
0 + 1) − (x2
0 + y2
0) = 2. But as x0 + 1 and y0 + 1 are both odd, we must have d = 1.
Thus x0 + 1 and y0 + 1 are relatively prime and therefore there exists a positive integer k such
that
k(x + 1)(y + 1) = x2
+ y2
has the solution (x0, y0). We will show that the latter equation has no solution (x, y) in positive
even numbers.
Assume there is a solution. Pick the solution (x1, y1) with the smallest sum x1 +y1 and assume
x1 ≥ y1. Then x1 is a solution to the quadratic equation
x2
− k(y1 + 1)x + y2
1 − k(y1 + 1) = 0.
73

Let x2 be the second solution, which by Vieta’s theorem fulfills x1 + x2 = k(y1 + 1) and
x1x2 = y2
1 − k(y1 + 1). If x2 = 0, the second equation implies y2
1 = k(y1 + 1), which is
impossible, as y1 + 1 > 1 cannot divide the relatively prime number y2
1. Therefore x2 = 0.
Also we get (x1 + 1)(x2 + 1) = x1x2 + x1 + x2 + 1 = y2
1 + 1 which is odd, and hence x2 must
be even and positive. Also we have x2 + 1 =
y2
1+1
x1+1
≤
y2
1+1
y1+1
≤ y1 ≤ x1. But this means that the
pair (x , y ) with x = y1 and y = x2 is another solution of k(x + 1)(y + 1) = x2
+ y2
in even
positive numbers with x + y < x1 + y1, a contradiction.
Therefore we must have n ≤ 4.
When n = 4, a possible example of a sequence is a1 = 4, a2 = 33, a3 = 217 and a4 = 1384.
Solution 2. It is easy to check that for n = 4 the sequence a1 = 4, a2 = 33, a3 = 217 and
a4 = 1384 is possible.
Now assume there is a sequence with n ≥ 5. Then we have in particular
a2
2 + 1 = (a1 + 1)(a3 + 1),
a2
3 + 1 = (a2 + 1)(a4 + 1),
a2
4 + 1 = (a3 + 1)(a5 + 1).
Also assume without loss of generality that among all such quintuples (a1, a2, a3, a4, a5) we have
chosen one with minimal a1.
One shows quickly the following fact:
If three positive integers x, y, z fulfill y2
+ 1 = (x + 1)(z + 1) and if y is even, then
x and z are even as well and either x < y < z or z < y < x holds. (1)
Indeed, the first part is obvious and from x < y we conclude
z + 1 =
y2
+ 1
x + 1
≥
y2
+ 1
y
> y,
and similarly in the other case.
Now, if a3 was odd, then (a2 + 1)(a4 + 1) = a2
3 + 1 ≡ 2 mod 4 would imply that one of a2 or
a4 is even, this contradicts (1). Thus a3 and hence also a1, a2, a4 and a5 are even. According
to (1), one has a1 < a2 < a3 < a4 < a5 or a1 > a2 > a3 > a4 > a5 but due to the minimality of
a1 the first series of inequalities must hold.
Consider the identity
(a3 +1)(a1 +a3) = a2
3 −1+(a1 +1)(a3 +1) = a2
2 +a2
3 = a2
2 −1+(a2 +1)(a4 +1) = (a2 +1)(a2 +a4).
Any common divisor of the two odd numbers a2 + 1 and a3 + 1 must also divide (a2 + 1)(a4 +
1) − (a3 + 1)(a3 − 1) = 2, so these numbers are relatively prime. Hence the last identity shows
that a1 + a3 must be a multiple of a2 + 1, i.e. there is an integer k such that
a1 + a3 = k(a2 + 1). (2)
Now set a0 = k(a1 + 1) − a2. This is an integer and we have
(a0 + 1)(a2 + 1) = k(a1 + 1)(a2 + 1) − (a2 − 1)(a2 + 1)
= (a1 + 1)(a1 + a3) − (a1 + 1)(a3 + 1) + 2
= (a1 + 1)(a1 − 1) + 2 = a2
1 + 1.
74

Thus a0 ≥ 0. If a0 > 0, then by (1) we would have a0 < a1 < a2 and then the quintuple
(a0, a1, a2, a3, a4) would contradict the minimality of a1.
Hence a0 = 0, implying a2 = a2
1. But also a2 = k(a1 + 1), which ﬁnally contradicts the fact
that a1 + 1 > 1 is relatively prime to a2
1 and thus cannot be a divisior of this number.
Hence n ≥ 5 is not possible.
Comment 1. Finding the example for n = 4 is not trivial and requires a tedious calculation,
but it can be reduced to checking a few cases. The equations (a1 + 1)(a3 + 1) = a2
2 + 1 and
(a2 + 1)(a4 + 1) = a2
3 + 1 imply, as seen in the proof, that a1 is even and a2, a3, a4 are odd. The
case a1 = 2 yields a2
2 ≡ −1 mod 3 which is impossible. Hence a1 = 4 is the smallest possibility.
In this case a2
2 ≡ −1 mod 5 and a2 is odd, which implies a2 ≡ 3 or a2 ≡ 7 mod 10. Hence we
have to start checking a2 = 7, 13, 17, 23, 27, 33 and in the last case we succeed.
Comment 2. The choice of a0 = k(a1 +1)−a2 in the second solution appears more natural if
one considers that by the previous calculations one has a1 = k(a2+1)−a3 and a2 = k(a3+1)−a4.
Alternatively, one can solve the equation (2) for a3 and use a2
2 + 1 = (a1 + 1)(a3 + 1) to get
a2
2 − k(a1 + 1)a2 + a2
1 − k(a1 + 1) = 0. Now a0 is the second solution to this quadratic equation
in a2 (Vieta jumping).
75

N5 HUN (Hungary)
Let P(x) be a non-constant polynomial with integer coefficients. Prove that there is no function
T from the set of integers into the set of integers such that the number of integers x with
Tn
(x) = x is equal to P(n) for every n ≥ 1, where Tn
denotes the n-fold application of T.
Solution 1. Assume there is a polynomial P of degree at least 1 with the desired property
for a given function T. Let A(n) denote the set of all x ∈ Z such that Tn
(x) = x and let
B(n) denote the set of all x ∈ Z for which Tn
(x) = x and Tk
(x) = x for all 1 ≤ k < n. Both
sets are finite under the assumption made. For each x ∈ A(n) there is a smallest k ≥ 1 such
that Tk
(x) = x, i.e., x ∈ B(k). Let d = gcd(k, n). There are positive integers r, s such that
rk − sn = d and hence x = Trk
(x) = Tsn+d
(x) = Td
(Tsn
(x)) = Td
(x). The minimality of k
implies d = k, i.e., k|n. On the other hand one clearly has B(k) ⊂ A(n) if k|n and thus we
have A(n) = d|n B(d) as a disjoint union and hence
|A(n)| =
d|n
|B(d)|.
Furthermore, for every x ∈ B(n) the elements x, T1
(x), T2
(x), . . . , Tn−1
(x) are n distinct
elements of B(n). The fact that they are in A(n) is obvious. If for some k < n and
some 0 ≤ i < n we had Tk
(Ti
(x)) = Ti
(x), i.e. Tk+i
(x) = Ti
(x), that would imply
x = Tn
(x) = Tn−i
(Ti
(x)) = Tn−i
(Tk+i
(x)) = Tk
(Tn
(x)) = Tk
(x) contradicting the minimality
of n. Thus Ti
(x) ∈ B(n) and Ti
(x) = Tj
(x) for 0 ≤ i < j ≤ n − 1.
So indeed, T permutes the elements of B(n) in (disjoint) cycles of length n and in particular
one has n |B(n)|.
Now let P(x) = k
i=0 aixi
, ai ∈ Z, k ≥ 1, ak = 0 and suppose that |A(n)| = P(n) for all n ≥ 1.
Let p be any prime. Then
p2
|B(p2
)| = |A(p2
)| − |A(p)| = a1(p2
− p) + a2(p4
− p2
) + . . .
Hence p|a1 and since this is true for all primes we must have a1 = 0.
Now consider any two different primes p and q. Since a1 = 0 we have that
|A(p2
q)| − |A(pq)| = a2(p4
q2
− p2
q2
) + a3(p6
q3
− p3
q3
) + . . .
is a multiple of p2
q. But we also have
p2
q |B(p2
q)| = |A(p2
q)| − |A(pq)| − |B(p2
)|.
This implies
p2
q |B(p2
)| = |A(p2
)| − |A(p)| = a2(p4
− p2
) + a3(p6
− p3
) + · · · + ak(p2k
− pk
).
Since this is true for every prime q we must have a2(p4
−p2
)+a3(p6
−p3
)+· · ·+ak(p2k
−pk
) = 0
for every prime p. Since this expression is a polynomial in p of degree 2k (because ak = 0) this
is a contradiction, as such a polynomial can have at most 2k zeros.
76

Comment. The last contradiction can also be reached via
ak = lim
p→∞
1
p2k
a2(p4
− p2
) + a3(p6
− p3
) + · · · + ak(p2k
− pk
) = 0.
Solution 2. As in the first solution define A(n) and B(n) and assume that a polynomial P
with the required property exists. This again implies that |A(n)| and |B(n)| is finite for all
positive integers n and that
P(n) = |A(n)| =
d|n
|B(d)| and n |B(n)|.
Now, for any two distinct primes p and q, we have
P(0) ≡ P(pq) ≡ |B(1)| + |B(p)| + |B(q)| + |B(pq)| ≡ |B(1)| + |B(p)| mod q.
Thus, for any fixed p, the expression P(0) − |B(1)| − |B(p)| is divisible by arbitrarily large
primes q which means that P(0) = |B(1)| + |B(p)| = P(p) for any prime p. This implies that
the polynomial P is constant, a contradiction.
77

N6 TUR (Turkey)
Let k be a positive integer. Show that if there exists a sequence a0, a1, . . . of integers satisfying
the condition
an =
an−1 + nk
n
for all n ≥ 1,
then k − 2 is divisible by 3.
Solution 1. Part A. For each positive integer k, there exists a polynomial Pk of degree k − 1
with integer coefficients, i. e., Pk ∈ Z[x], and an integer qk such that the polynomial identity
xPk(x) = xk
+ Pk(x − 1) + qk (Ik)
is satisfied. To prove this, for fixed k we write
Pk(x) = bk−1xk−1
+ · · · + b1x + b0
and determine the coefficients bk−1, bk−2, . . . , b0 and the number qk successively. Obviously, we
have bk−1 = 1. For m = k − 1, k − 2, . . . , 1, comparing the coefficients of xm
in the identity (Ik)
results in an expression of bm−1 as an integer linear combination of bk−1, . . . , bm, and finally
qk = −Pk(−1).
Part B. Let k be a positive integer, and let a0, a1, . . . be a sequence of real numbers satisfying
the recursion given in the problem. This recursion can be written as
an − Pk(n) =
an−1 − Pk(n − 1)
n
−
qk
n
for all n ≥ 1,
which by induction gives
an − Pk(n) =
a0 − Pk(0)
n!
− qk
n−1
i=0
i!
n!
for all n ≥ 1.
Therefore, the numbers an are integers for all n ≥ 1 only if
a0 = Pk(0) and qk = 0.
Part C. Multiplying the identity (Ik) by x2
+x and subtracting the identities (Ik+1), (Ik+2) and
qkx2
= qkx2
therefrom, we obtain
xTk(x) = Tk(x − 1) + 2x Pk(x − 1) + qk − (qk+2 + qk+1 + qk),
where the polynomials Tk ∈ Z[x] are defined by Tk(x) = (x2
+x)Pk(x)−Pk+1(x)−Pk+2(x)−qkx.
Thus
xTk(x) ≡ Tk(x − 1) + qk+2 + qk+1 + qk mod 2, k = 1, 2, . . .
78

Comparing the degrees, we easily see that this is only possible if Tk is the zero polynomial
modulo 2, and
qk+2 ≡ qk+1 + qk mod 2 for k = 1, 2, . . .
Since q1 = −1 and q2 = 0, these congruences finish the proof.
Solution 2. Part A and B. Let k be a positive integer, and suppose there is a sequence
a0, a1, . . . as required. We prove: There exists a polynomial P ∈ Z[x], i. e., with integer
coefficients, such that an = P(n), n = 0, 1, . . . , and xP(x) = xk
+ P(x − 1).
To prove this, we write P(x) = bk−1xk−1
+ · · · + b1x + b0 and determine the coefficients
bk−1, bk−2, . . . , b0 successively such that
xP(x) − xk
− P(x − 1) = q,
where q = qk is an integer. Comparing the coefficients of xm
results in an expression of bm−1
as an integer linear combination of bk−1, . . . , bm.
Defining cn = an − P(n), we get
P(n) + cn =
P(n − 1) + cn−1 + nk
n
, i. e.,
q + ncn = cn−1,
hence
cn =
c0
n!
− q ·
0! + 1! + · · · + (n − 1)!
n!
.
We conclude limn→∞ cn = 0, which, using cn ∈ Z, implies cn = 0 for sufficiently large n.
Therefore, we get q = 0 and cn = 0, n = 0, 1, . . . .
Part C. Suppose that q = qk = 0, i. e. xP(x) = xk
+ P(x − 1). To consider this identity for
arguments x ∈ F4, we write F4 = {0, 1, α, α + 1}. Then we get
αPk(α) = αk
+ Pk(α + 1) and
(α + 1)Pk(α + 1) = (α + 1)k
+ Pk(α),
hence
Pk(α) = 1 · Pk(α) = (α + 1)αPk(α)
= (α + 1)Pk(α + 1) + (α + 1)αk
= Pk(α) + (α + 1)k
+ (α + 1)αk
.
Now, (α + 1)k−1
= αk
implies k ≡ 2 mod 3.
Comment 1. For k = 2, the sequence given by an = n+1, n = 0, 1, . . . , satisfies the conditions
of the problem.
79

Comment 2. The first few polynomials Pk and integers qk are
P1(x) = 1, q1 = −1,
P2(x) = x + 1, q2 = 0,
P3(x) = x2
+ x − 1, q3 = 1,
P4(x) = x3
+ x2
− 2x − 1, q4 = −1,
P5(x) = x4
+ x3
− 3x2
+ 5, q5 = −2,
P6(x) = x5
+ x4
− 4x3
+ 2x2
+ 10x − 5, q6 = 9,
q7 = −9, q8 = −50, q9 = 267, q10 = −413, q11 = −2180.
A lookup in the On-Line Encyclopedia of Integer Sequences (A000587) reveals that the sequence
q1, −q2, q3, −q4, q5, . . . is known as Uppuluri-Carpenter numbers. The result that qk = 0
implies k ≡ 2 mod 3 is contained in
Murty, Summer: On the p-adic series ∞
n=0 nk
· n!. CRM Proc. and Lecture Notes 36, 2004.
As shown by Alexander (Non-Vanishing of Uppuluri-Carpenter Numbers, Preprint 2006),
Uppuluri-Carpenter numbers are zero at most twice.
Comment 3. The numbers qk can be written in terms of the Stirling numbers of the second
kind. To show this, we fix the notation such that
xk
=Sk−1,k−1x(x − 1) · · · (x − k + 1)
+ Sk−1,k−2x(x − 1) · · · (x − k + 2) (∗)
+ · · · + Sk−1,0x,
e. g., S2,2 = 1, S2,1 = 3, S2,0 = 1, and we define
Ωk = Sk−1,k−1 − Sk−1,k−2 + − · · · .
Replacing x by −x in (∗) results in
xk
=Sk−1,k−1x(x + 1) · · · (x + k − 1)
− Sk−1,k−2x(x + 1) · · · (x + k − 2)
+ − · · · ± Sk−1,0x.
Defining
P(x) =Sk−1,k−1(x + 1) · · · (x + k − 1)
+ (Sk−1,k−1 − Sk−1,k−2)(x + 1) · · · (x + k − 2)
+ (Sk−1,k−1 − Sk−1,k−2 + Sk−1,k−3)(x + 1) · · · (x + k − 3)
+ · · · + Ωk,
we obtain
xP(x) − P(x − 1) = Sk−1,k−1x(x + 1) · · · (x + k − 1)
− Sk−1,k−2x(x + 1) · · · (x + k − 2)
+ − · · · ± Sk−1,0x − Ωk
= xk
− Ωk,
hence qk = −Ωk.
80

N7 MNG (Mongolia)
Let a and b be distinct integers greater than 1. Prove that there exists a positive integer n such
that (an
− 1)(bn
Solution 1. At first we notice that
(1 − α)
1
2 (1 − β)
1
2 = 1 − 1
2
· α − 1
8
· α2
− · · · 1 − 1
2
· β − 1
8
· β2
− · · ·
=
k, ≥0
ck, · αk
β for all α, β ∈ (0, 1), (1)
where c0,0 = 1 and ck, are certain coefficients.
For an indirect proof, we suppose that xn = (an − 1)(bn − 1) ∈ Z for all positive integers n.
Replacing a by a2
and b by b2
if necessary, we may assume that a and b are perfect squares,
hence
√
ab is an integer.
At first we shall assume that aµ
= bν
for all positive integers µ, ν. We have
xn = (
√
ab)n
1 −
1
an
1
2
1 −
1
bn
1
2
=
k, ≥0
ck,
√
ab
akb
n
. (2)
Choosing k0 and 0 such that ak0
>
√
ab, b 0
>
√
ab, we define the polynomial
P(x) =
k0−1, 0−1
k=0, =0
(ak
b x −
√
ab) =:
k0· 0
i=0
dixi
with integer coefficients di. By our assumption, the zeros
√
ab
akb
, k = 0, . . . , k0 − 1, = 0, . . . , 0 − 1,
of P are pairwise distinct.
Furthermore, we consider the integer sequence
yn =
k0· 0
i=0
dixn+i, n = 1, 2, . . . (3)
By the theory of linear recursions, we obtain
yn =
k, ≥0
k≥k0 or ≥ 0
ek,
√
ab
akb
n
, n = 1, 2, . . . , (4)
with real numbers ek, . We have
|yn| ≤
k, ≥0
k≥k0 or ≥ 0
|ek, |
√
ab
akb
n
=: Mn.
81

Because the series in (4) is obtained by a finite linear combination of the absolutely convergent
series (1), we conclude that in particular M1 < ∞. Since
√
ab
akb
≤ λ := max
√
ab
ak0
,
√
ab
b 0
for all k, ≥ 0 such that k ≥ k0 or ≥ 0,
we get the estimates Mn+1 ≤ λMn, n = 1, 2, . . . Our choice of k0 and 0 ensures λ < 1, which
implies Mn → 0 and consequently yn → 0 as n → ∞. It follows that yn = 0 for all sufficiently
large n.
So, equation (3) reduces to k0· 0
i=0 dixn+i = 0.
Using the theory of linear recursions again, for sufficiently large n we have
xn =
k0−1, 0−1
k=0, =0
fk,
√
ab
akb
n
for certain real numbers fk, .
Comparing with (2), we see that fk, = ck, for all k, ≥ 0 with k < k0, < 0, and ck, = 0 if
k ≥ k0 or ≥ 0, since we assumed that aµ
= bν
for all positive integers µ, ν.
In view of (1), this means
(1 − α)
1
2 (1 − β)
1
2 =
k0−1, 0−1
k=0, =0
ck, · αk
β (5)
for all real numbers α, β ∈ (0, 1). We choose k∗
< k0 maximal such that there is some i
with ck∗,i = 0. Squaring (5) and comparing coefficients of α2k∗
β2i∗
, where i∗
is maximal with
ck∗,i∗ = 0, we see that k∗
= 0. This means that the right hand side of (5) is independent of α,
which is clearly impossible.
We are left with the case that aµ
= bν
for some positive integers µ and ν. We may assume
that µ and ν are relatively prime. Then there is some positive integer c such that a = cν
and
b = cµ
. Now starting with the expansion (2), i. e.,
xn =
j≥0
gj
√
cµ+ν
cj
n
for certain coefficients gj, and repeating the arguments above, we see that gj = 0 for sufficiently
large j, say j > j0. But this means that
(1 − xµ
)
1
2 (1 − xν
)
1
2 =
j0
j=0
gjxj
for all real numbers x ∈ (0, 1). Squaring, we see that
(1 − xµ
)(1 − xν
)
is the square of a polynomial in x. In particular, all its zeros are of order at least 2, which
implies µ = ν by looking at roots of unity. So we obtain µ = ν = 1, i. e., a = b, a contradiction.
82

Solution 2. We set a2
= A, b2
= B, and zn = (An − 1)(Bn − 1). Let us assume that zn
is an integer for n = 1, 2, . . . Without loss of generality, we may suppose that b < a. We
determine an integer k ≥ 2 such that bk−1
≤ a < bk
, and deﬁne a sequence γ1, γ2, . . . of rational
numbers such that
2γ1 = 1 and 2γn+1 =
n
i=1
γiγn−i for n = 1, 2, . . .
It could easily be shown that γn = 1·1·3...(2n−3)
2·4·6...2n
, for instance by reading Vandermondes con-
volution as an equation between polynomials, but we shall have no use for this fact.
Using Landaus O–Notation in the usual way, we have
(ab)n
− γ1
a
b
n
− γ2
a
b3
n
− · · · − γk
a
b2k−1
n
+ O
b
a
n 2
= An
Bn
− 2γ1An
−
k
i=2
2γi −
i−1
j=1
γjγi−j
A
Bi−1
n
+ O
A
Bk
n
+ O (Bn
)
= An
Bn
− An
+ O (Bn
) ,
whence
zn = (ab)n
− γ1
a
b
n
− γ2
a
b3
n
− · · · − γk
a
b2k−1
n
+ O
b
a
n
.
Now choose rational numbers r1, r2, . . . , rk+1 such that
(x − ab) · (x − a
b
) . . . (x − a
b2k−1 ) = xk+1
− r1xk
+ − · · · ± rk+1,
and then a natural number M for which Mr1, Mr2, . . . Mrk+1 are integers. For known reasons,
M(zn+k+1 − r1zn+k + − · · · ± rk+1zn) = O
b
a
n
for all n ∈ N and thus there is a natural number N which is so large, that
zn+k+1 = r1zn+k − r2zn+k−1 + − · · · rk+1zn
holds for all n N. Now the theory of linear recursions reveals that there are some rational
numbers δ0, δ1, δ2, . . . , δk such that
zn = δ0(ab)n
− δ1
a
b
n
− δ2
a
b3
n
− · · · − δk
a
b2k−1
n
for suﬃciently large n, where δ0 > 0 as zn > 0. As before, one obtains
An
Bn
− An
− Bn
+ 1 = z2
n
= δ0(ab)n
− δ1
a
b
n
− δ2
a
b3
n
− · · · − δk
a
b2k−1
n 2
= δ2
0An
Bn
− 2δ0δ1An
−
i=k
i=2
2δ0δi −
j=i−1
j=1
δjδi−j
A
Bi−1
n
+ O
A
Bk
n
.
Easy asymptotic calculations yield δ0 = 1, δ1 = 1
2
, δi = 1
2
j=i−1
j=1 δjδi−j for i = 2, 3, . . . , k−2, and
then a = bk−1
. It follows that k > 2 and there is some P ∈ Q[X] for which (X −1)(Xk−1
−1) =
P(X)2
. But this cannot occur, for instance as Xk−1
− 1 has no double zeros. Thus our
83

assumption that zn was an integer for n = 1, 2, . . . turned out to be wrong, which solves the
problem.
Original formulation of the problem. a, b are positive integers such that a·b is not a square of
an integer. Prove that there exists a (inﬁnitely many) positive integer n such that (an
−1)(bn
−1)
is not a square of an integer.
Solution. Lemma. Let c be a positive integer, which is not a perfect square. Then there exists
an odd prime p such that c is not a quadratic residue modulo p.
Proof. Denoting the square-free part of c by c , we have the equality c
p
= c
p
of the corre-
sponding Legendre symbols. Suppose that c = q1 · · · qm, where q1 < · · · < qm are primes.
Then we have
c
p
=
q1
p
· · ·
qm
p
.
Case 1. Let q1 be odd. We choose a quadratic nonresidue r1 modulo q1 and quadratic residues
ri modulo qi for i = 2, . . . , m. By the Chinese remainder theorem and the Dirichlet theorem,
there exists a (inﬁnitely many) prime p such that
p ≡ r1 mod q1,
p ≡ r2 mod q2,
...
...
p ≡ rm mod qm,
p ≡ 1 mod 4.
By our choice of the residues, we obtain
p
qi
=
ri
qi
=
−1, i = 1,
1, i = 2, . . . , m.
The congruence p ≡ 1 mod 4 implies that qi
p
= p
qi
, i = 1, . . . , m, by the law of quadratic
reciprocity. Thus
c
p
=
q1
p
· · ·
qm
p
= −1.
Case 2. Suppose q1 = 2. We choose quadratic residues ri modulo qi for i = 2, . . . , m. Again,
by the Chinese remainder theorem and the Dirichlet theorem, there exists a prime p such
that
p ≡ r2 mod q2,
...
...
p ≡ rm mod qm,
p ≡ 5 mod 8.
By the choice of the residues, we obtain p
qi
= ri
qi
= 1 for i = 2, . . . , m. Since p ≡ 1 mod 4 we
have qi
p
= p
qi
, i = 2, . . . , m, by the law of quadratic reciprocity. The congruence p ≡ 5 mod 8
84

implies that 2
p
= −1. Thus
c
p
=
2
p
q2
p
· · ·
qm
p
= −1,
and the lemma is proved.
Applying the lemma for c = a · b, we find an odd prime p such that
ab
p
=
a
p
·
b
p
= −1.
This implies either
a
p−1
2 ≡ 1 mod p, b
p−1
2 ≡ −1 mod p, or a
p−1
2 ≡ −1 mod p, b
p−1
2 ≡ 1 mod p.
Without loss of generality, suppose that a
p−1
2 ≡ 1 mod p and b
p−1
2 ≡ −1 mod p. The second
congruence implies that b
p−1
2 −1 is not divisible by p. Hence, if the exponent νp(a
p−1
2 −1) of p in
the prime decomposition of (a
p−1
2 − 1) is odd, then (a
p−1
2 − 1)(b
p−1
2 − 1) is not a perfect square.
If νp(a
p−1
2 − 1) is even, then νp(a
p−1
2
p
− 1) is odd by the well-known power lifting property
νp a
p−1
2
p
− 1 = νp a
p−1
2 − 1 + 1.
In this case, (a
p−1
2
p
− 1)(b
p−1
2
p
Comment 1. In 1998, the following problem appeared in Crux Mathematicorum:
Problem 2344. Find all positive integers N that are quadratic residues modulo all primes
greater than N.
The published solution (Crux Mathematicorum, 25(1999)4) is the same as the proof of the
lemma given above, see also http://www.mathlinks.ro/viewtopic.php?t=150495.
Comment 2. There is also an elementary proof of the lemma. We cite Theorem 3 of Chapter 5
and its proof from the book
Ireland, Rosen: A Classical Introduction to Modern Number Theory, Springer 1982.
Theorem. Let a be a nonsquare integer. Then there are infinitely many primes p for which a is
a quadratic nonresidue.
Proof. It is easily seen that we may assume that a is square-free. Let a = 2e
q1q2 · · · qn, where
qi are distinct odd primes and e = 0 or 1. The case a = 2 has to be dealt with separately. We
shall assume to begin with that n ≥ 1, i. e., that a is divisible by an odd prime.
Let 1, 2, . . . , k be a finite set of odd primes not including any qi. Let s be any quadratic
nonresidue modqn, and find a simultaneous solution to the congruences
x ≡ 1 mod i, i = 1, . . . , k,
x ≡ 1 mod 8,
x ≡ 1 mod qi, i = 1, . . . , n − 1,
x ≡ s mod qn.
Call the solution b. b is odd. Suppose that b = p1p2 · · · pm is its prime decomposition. Since
85

b ≡ 1 mod 8 we have 2
b
= 1 and qi
b
= b
qi
by a result on Jacobi symbols. Thus
a
b
=
2
b
e q1
b
· · ·
qn−1
b
qn
b
=
b
q1
· · ·
b
qn−1
b
qn
=
1
q1
· · ·
1
qn−1
s
qn
= −1.
On the other hand, by the definition of a
b
, we have a
b
= a
p1
a
p2
· · · a
pm
. It follows that
a
pi
= −1 for some i.
Notice that j does not divide b. Thus pi /∈ { 1, 2, . . . , k}.
To summarize, if a is a nonsquare, divisible by an odd prime, we have found a prime p, outside
of a given finite set of primes {2, 1, 2, . . . , k}, such that a
p
= −1. This proves the theorem
in this case.
It remains to consider the case a = 2. Let 1, 2, . . . , k be a finite set of primes, excluding 3, for
which 2
i
= −1. Let b = 8 1 2 · · · k + 3. b is not divisible by 3 or any i. Since b ≡ 3 mod 8
we have 2
b
= (−1)
b2−1
8 = −1. Suppose that b = p1p2 · · · pm is the prime decomposition of
b. Then, as before, we see that 2
pi
= −1 for some i. pi /∈ {3, 1, 2, . . . , k}. This proves the
theorem for a = 2.
This proof has also been posted to mathlinks, see http://www.mathlinks.ro/viewtopic.
php?t=150495 again.
86

Imo2009

More Related Content

What's hot (19)

Similar to Imo2009 (20)

More from Θανάσης Δρούγας (20)

Recently uploaded (20)

Imo2009