INTERNATIONAL MATHEMATICS COMPETITIONS

FOR UNIVERSITY STUDENTS

======================

SELECTION OF
PROBLEMS AND SOLUTIONS

Hanoi, 2009
Contents

1 Questions 6
1.1 Olympic 1994 . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.1 Day 1, 1994 . . . . . . . . . . . . . . . . . . . . . 6
1.1.2 Day 2, 1994 . . . . . . . . . . . . . . . . . . . . . 7
1.2 Olympic 1995 . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.1 Day 1, 1995 . . . . . . . . . . . . . . . . . . . . . 9
1.2.2 Day 2, 1995 . . . . . . . . . . . . . . . . . . . . . 10
1.3 Olympic 1996 . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.1 Day 1, 1996 . . . . . . . . . . . . . . . . . . . . . 12
1.3.2 Day 2, 1996 . . . . . . . . . . . . . . . . . . . . . 14
1.4 Olympic 1997 . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.1 Day 1, 1997 . . . . . . . . . . . . . . . . . . . . . 16
1.4.2 Day 2, 1997 . . . . . . . . . . . . . . . . . . . . . 17
1.5 Olympic 1998 . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5.1 Day 1, 1998 . . . . . . . . . . . . . . . . . . . . . 19
1.5.2 Day 2, 1998 . . . . . . . . . . . . . . . . . . . . . 20
1.6 Olympic 1999 . . . . . . . . . . . . . . . . . . . . . . . . 21
1.6.1 Day 1, 1999 . . . . . . . . . . . . . . . . . . . . . 21
1.6.2 Day 2, 1999 . . . . . . . . . . . . . . . . . . . . . 22
1.7 Olympic 2000 . . . . . . . . . . . . . . . . . . . . . . . . 23
1.7.1 Day 1, 2000 . . . . . . . . . . . . . . . . . . . . . 23
1.7.2 Day 2, 2000 . . . . . . . . . . . . . . . . . . . . . 25
1.8 Olympic 2001 . . . . . . . . . . . . . . . . . . . . . . . . 26
1.8.1 Day 1, 2001 . . . . . . . . . . . . . . . . . . . . . 26

2
CONTENTS 3

1.8.2 Day 2, 2001 . . . . . . . . . . . . . . . . . . . . . 27

1.9 Olympic 2002 . . . . . . . . . . . . . . . . . . . . . . . . 28
1.9.1 Day 1, 2002 . . . . . . . . . . . . . . . . . . . . . 28
1.9.2 Day 2, 2002 . . . . . . . . . . . . . . . . . . . . . 29
1.10 Olympic 2003 . . . . . . . . . . . . . . . . . . . . . . . . 30
1.10.1 Day 1, 2003 . . . . . . . . . . . . . . . . . . . . . 30
1.10.2 Day 2, 2003 . . . . . . . . . . . . . . . . . . . . . 32
1.11 Olympic 2004 . . . . . . . . . . . . . . . . . . . . . . . . 33
1.11.1 Day 1, 2004 . . . . . . . . . . . . . . . . . . . . . 33
1.11.2 Day 2, 2004 . . . . . . . . . . . . . . . . . . . . . 34
1.12 Olympic 2005 . . . . . . . . . . . . . . . . . . . . . . . . 35
1.12.1 Day 1, 2005 . . . . . . . . . . . . . . . . . . . . . 35
1.12.2 Day 2, 2005 . . . . . . . . . . . . . . . . . . . . . 36
1.13 Olympic 2006 . . . . . . . . . . . . . . . . . . . . . . . . 37
1.13.1 Day 1, 2006 . . . . . . . . . . . . . . . . . . . . . 37
1.13.2 Day 2, 2006 . . . . . . . . . . . . . . . . . . . . . 38
1.14 Olympic 2007 . . . . . . . . . . . . . . . . . . . . . . . . 40
1.14.1 Day 1, 2007 . . . . . . . . . . . . . . . . . . . . . 40
1.14.2 Day 2, 2007 . . . . . . . . . . . . . . . . . . . . . 41
1.15 Olympic 2008 . . . . . . . . . . . . . . . . . . . . . . . . 41
1.15.1 Day 1, 2008 . . . . . . . . . . . . . . . . . . . . . 41
1.15.2 Day 2, 2008 . . . . . . . . . . . . . . . . . . . . . 42

2 Solutions 44
2.1 Solutions of Olympic 1994 . . . . . . . . . . . . . . . . . 44
2.1.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 44
2.1.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 47
2.2 Solutions of Olympic 1995 . . . . . . . . . . . . . . . . . 50
2.2.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 50
2.2.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 53
2.3 Solutions of Olympic 1996 . . . . . . . . . . . . . . . . . 58
CONTENTS 4

2.3.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 58
2.3.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 64
2.4 Solutions of Olympic 1997 . . . . . . . . . . . . . . . . . 69
2.4.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 69
2.4.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 75
2.5 Solutions of Olympic 1998 . . . . . . . . . . . . . . . . . 79
2.5.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 79
2.5.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 83
2.6 Solutions of Olympic 1999 . . . . . . . . . . . . . . . . . 87
2.6.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 87
2.6.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 92
2.7 Solutions of Olympic 2000 . . . . . . . . . . . . . . . . . 96
2.7.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 96
2.7.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 100
2.8 Solutions of Olympic 2001 . . . . . . . . . . . . . . . . . 105
2.8.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 105
2.8.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 108
2.9 Solutions of Olympic 2002 . . . . . . . . . . . . . . . . . 113
2.9.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 113
2.9.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 117
2.10 Solutions of Olympic 2003 . . . . . . . . . . . . . . . . . 120
2.10.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 120
2.10.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 126
2.11 Solutions of Olympic 2004 . . . . . . . . . . . . . . . . . 130
2.11.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 130
2.11.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 137
2.12 Solutions of Olympic 2005 . . . . . . . . . . . . . . . . . 140
2.12.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 140
2.12.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 146
2.13 Solutions of Olympic 2006 . . . . . . . . . . . . . . . . . 151
2.13.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 151
CONTENTS 5

2.13.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 156

2.14 Solutions of Olympic 2007 . . . . . . . . . . . . . . . . . 160
2.14.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 160
2.14.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 164
2.15 Solutions of Olympic 2008 . . . . . . . . . . . . . . . . . 170
2.15.1 Day 1 . . . . . . . . . . . . . . . . . . . . . . . . 170
2.15.2 Day 2 . . . . . . . . . . . . . . . . . . . . . . . . 175
Chapter 1

Questions

1.1 Olympic 1994

1.1.1 Day 1, 1994

Problem 1. (13 points)

a) Let A be a n × n, n ≥ 2, symmetric, invertible matrix with real
positive elements. Show that zn ≤ n2 − 2n, where zn is the number of
zero elements in A−1 .
b) How many zero elements are there in the inverse of the n×n matrix
 
1 1 1 1 ... 1
1 2 2 2 . . . 2 
1 2 1 1 . . . 1 
A=1 2 1 2 . . . 2 

. . . . . . . . . . . . . . . . . . 
1 2 1 2 ... ...
Problem 2. (13 points)
Let f ∈ C 1 (a, b), lim f (x) = ∞, lim f (x) = −∞ and f 0 (x)+f 2 (x) ≥
x→a+ x→b−
−1 for x ∈ (a, b). Prove that b − a ≥ π and give an example where
b − a = π.
Problem 3. (13 points)
Give a set S of 2n − 1, n ∈ N, different irrational numbers. Prove
that there are n different elements x1 , x2 , . . . , xn ∈ S such that for all
non-negative rational numbers a1 , a2 , . . . , an with a1 + a2 + · · · an > 0 we
have that a1 x1 + a2 x2 + · · · + an xn , is an irrational number.
Problem 4. (18 points)

6
1.1. Olympic 1994 7

Let α ∈ R\{0} and suppose that F and G are linear maps (operators)
from Rn satisfying F ◦ G − G ◦ F = αF .
a) Show that for all k ∈ N one has F k ◦ G − G ◦ F k = αkF k .
b) Show that there exists k ≥ 1 such that F k = 0.
Problem 5. (18 points)
a) Let f ∈ C[0, b], g ∈ C(R) and let g be periodic with period b. Prove
Rb
that f (x)g(nx)dx has a limit as n → ∞ and
0

Zb Zb Zb
1
lim f (x)g(nx)dx = f (x)dx g(x)dx.
n→∞ b
0 0 0

b) Find
Zπ
sin x
lim dx.
n→∞ 1 + 3 cos2 nx
0
Problem 6. (25 points)
Let f ∈ C 2 [0, N ] and |f 0 (x)| < 1, f ”(x) > 0 for every x ∈ [0, N ]. Let
0 ≤ m0 < m1 < · · · < mk ≤ N be integers such that ni = f (mi ) are also
integers for i = 0, 1, . . . , k. Denote bi = ni - ni-1 and ai = mi - mi-1 for
i = 1,2, ... , k.
a) Prove that
b1 b2 bk
−1 < < < ··· < < 1.
a1 a2 ak
N
b) Prove that for every choice of A > 1 there are no more than
A
indices j such that aj > A.
c) Prove that k ≤ 3N 2/3 (i.e. there are no more than 3N 2/3 integer
points on the curve y = f (x), x ∈ [0, N ]).

1.1.2 Day 2, 1994

Problem 1. (14 points)

1.1. Olympic 1994 8

Let f ∈ C 1 [a, b], f (a) = 0 and suppose that λ ∈ R, λ > 0, is such that

|f 0 (x)| ≤ λ|f (x)|

for all x ∈ [a, b]. Is it true that f (x) = 0 for all x ∈ [a, b]?
Problem 2. (14 points)
2 2
Let f : R2 → R be given by f (x, y) = (x2 − y 2 )e−x −y .
a) Prove that f attains its minimum and its maximum.
∂f ∂f
b) Determine all points (x, y) such that (x, y) = (x, y) = 0 and
∂x ∂y
determine for which of them f has global or local minimum or maximum.
Problem 3. (14 points)
Let f be a real-valued function with n + 1 derivatives at each point
of R. Show that for each pair of real numbers a, b, a < b, such that
 f (b) + f 0 (b) + · · · + f (n) (b) 
ln =b−a
f (a) + f 0 (a) + · · · + f (n) (a)
there is a number c in the open interval (a, b) for which

f (n+1) (c) = f (c).

Note that ln denotes the natural logarithm.

Problem 4. (18 points)
Let A be a n × n diagonal matrix with characteristic polynomial

(x − c1 )d1 (x − c2 )d2 . . . (x − ck )dk ,

where c1 , c2 , . . . , ck are distinct (which means that c1 appears d1 times

on the diagonal, c2 appears d2 times on the diagonal, etc. and d1 + d2 +
· · · + dk = n).
Let V be the space of all n × n matrices B such that AB = BA.
Prove that the dimension of V is

d21 + d22 + · · · + d2k .

Problem 5. (18 points)

1.2. Olympic 1995 9

Let x1 , x2 , . . . , xk be vectors of m-dimensional Euclidian space, such

that x1 + x2 + · · · + xk = 0. Show that there exists a permutation π of
the integers {1, 2, . . . , k} such that
n
X k
X 1/2
2
k xπ(i) k≤ k xi k for each n = 1, 2, . . . , k.
i=1 i=1

Note that k . k denotes the Euclidian norm.

Problem 6. (22 points)
N −2
ln2 N X 1
Find lim . Note that ln denotes the natural
N →∞ N ln k.ln(N − k)
k=2
logarithm.

1.2 Olympic 1995

1.2.1 Day 1, 1995

Problem 1. (10 points)

Let X be a nonsingular matrix with columns Xl , X2 , . . . , Xn . Let Y
be a matrix with columns X2 , X3 , . . . , Xn , 0. Show that the matrices
A = Y X −1 and B = X −1 Y have rank n − 1 and have only 0’s for
eigenvalues.
Problem 2. (15 points)
Let f be a continuous function on [0, 1] such that for every x ∈ [0, 1]
R1 1 − x2 R1 2 1
we have f (t)dt ≥ . Show that f (t)dt ≥ .
x 2 0 3
Problem 3. (15 points)
Let f be twice continuously differentiable on (0, +∞) such that

lim f 0 (x) = −∞
x→0+

and
lim f ”(x) = +∞.
x→0+
Show that
f (x)
lim = 0.
x→0+ f 0 (x)
1.2. Olympic 1995 10

Problem 4. (15 points)

Let F : (1, ∞) → R be the function defined by
Zx2
dt
F (x) := .
lnt
x

Show that F is one-to-one (i.e. injective) and find the range (i.e. set of
values) of F .
Problem 5. (20 points)
Let A and B be real n × n matrices. Assume that there exist n + 1
different real numbers tl , t2 , . . . , tn+1 such that the matrices

Ci = A + ti B, i = 1, 2, . . . , n + 1,

are nilpotent (i.e. Cin = 0).

Show that both A and B are nilpotent.
Problem 6. (25 points)
Let p > 1. Show that there exists a constant Kp > 0 such that for
every x, y ∈ R satisfying |x|p + |y|p = 2, we have

(x − y)2 ≤ Kp (4 − (x + y)2 ).

1.2.2 Day 2, 1995

Problem 1. (10 points)

Let A be 3 × 3 real matrix such that the vectors Au and u are orthog-
onal for each column vector u ∈ R3 . Prove that:
a) AT = −A, where AT denotes the transpose of the matrix A;
b) there exists a vector v ∈ R3 such that Au = v × u for every u ∈ R3 ,
where v × u denotes the vector product in R3 .
Problem 2. (15 points)
Let {bn }∞
n=0 be a sequence of positive real numbers such that b0 =
√ p √
1, bn = 2 + bn−1 − 2 1 + bn−1 . Calculate
∞
X
bn 2n .
n=1
1.2. Olympic 1995 11

Problem 3. (15 points)

Let all roots of an n-th degree polynomial P (z) with complex coeffi-
cients lie on the unit circle in the complex plane. Prove that all roots of
the polynomial
2zP 0 (z) − nP (z)
lie on the same circle.
Problem 4. (15 points)
a) Prove that for every  > 0 there is a positive integer n and real
numbers λ1 , . . . , λn such that
n
X
max x − λk x2k+1 < .
x∈[−1,1]
k=1

b) Prove that for every odd continuous function f on [−1, 1] and for
every  > 0 there is a positive integer n and real numbers µ1 , . . . , µn such
that n
X
max f (x) − µk x2k+1 < .
x∈[−1,1]
k=1
Recall that f is odd means that f (x) = −f (−x) for all x ∈ [−1, 1].
Problem 5. (10+15 points)
a) Prove that every function of the form
N
a0 X
f (x) = + cos x + an cos(nx)
2 n=2

with |a0 | < 1, has positive as well as negative values in the period [0, 2π).
b) Prove that the function
100
3
X
F (x) = cos(n 2 x)
n=1

has at least 40 zeros in the interval (0, 1000).

Problem 6. (20 points)
1.3. Olympic 1996 12

Suppose that {fn }∞ n=1 is a sequence of continuous functions on the

interval [0, 1] such that
Z1 
1 if n = m
fm (x)fn (x)dx =
0 if n 6= m
0

and
sup{|fn (x)| : x ∈ [0, 1] and n = 1, 2, . . .} < +∞.
Show that there exists no subsequence {fnk } of {fn } such that lim fnk (x)
k→∞
exists for all x ∈ [0, 1].

1.3 Olympic 1996

1.3.1 Day 1, 1996

Problem 1. (10 points)

Let for j = 0, ..., n, aj = a0 + jd, where a0 , d are fixed real numbers.
Put
a0 a1 a2 . . . a n
 
 a1 a0 a1 . . . an−1 
A= a
 2 a 1 a0 . . . an−2 
........................

an an−1 an−2 . . . a0
Calculate det(A), where det(A) denotes the determinant of A.
Problem 2. (10 points) Evaluate the definite integral
Zπ
sin nx
dx,
(1 + 2x ) sin x
−π

where n is a natural number.

Problem 3. (15 points)
The linear operator A on the vector space V is called an involution if
A2 = E where E is the identity operator on V . Let dimV = n < ∞.
(i) Prove that for every involution A on V there exists a basis of V
consisting of eigenvectors of A.
1.3. Olympic 1996 13

(ii) Find the maximal number of distinct pairwise commuting involu-

tions on V .
Problem 4. (15 points)
n−1
1X
Let a1 = 1, an = ak an−k for n ≥ 2. Show that
n
k=1
1/n −1/2
(i) lim sup|an | <2 ;
n→∞
2
(ii) lim sup|an |1/n ≥ .
n→∞ 3
Problem 5. (25 points)
i) Let a, b be real number such that b ≤ 0 and 1 + ax + bx2 ≥ 0 for
every x in [0, 1]. Prove that
Z1 (
1
2 − if a < 0
lim n (1 + ax + bx )dx = a
n→∞
0
+∞ if a ≥ 0.

ii) Let f : [0, 1] → [0, ∞) be a function with a continuous second

derivative and let f 00 (x) ≤ 0 for every x in [0,1]. Suppose that L =
R1
lim n (f (x))n dx exists and 0 < L < +∞. Prove that f 0 has a constant
n→∞ 0
sign and minx∈[0,1] |f 0 (x)| = L−1 .
Problem 6. (25 points)
Upper content of a subset E of the plane R is defined as
n
nX o
C(E) = inf diam (Ei )
i=1

where inf is taken over all finite of sets E1 , . . . , En , n ∈ N in R2 such

n
that E ⊂ ∪ Ei . Lower content of E is defined as
i=1

K(E) = sup{length(L) : L is a closed line segment

onto which E can be contracted}

Show that
(a) C(L) = lenght (L) if L is a closed line segment;
(b) C(E) ≥ K(E);
1.3. Olympic 1996 14

(c) the equality in (b) needs not hold even if E is compact.

Hint. If E = T ∪T 0 where T is the triangle with vertices (−2, 2), (2, 2)
and (0, 4), and T 0 is its reflexion about the x-axis, then C(E) = 8 >
K(E).
Remarks: All distances used in this problem are Euclidian. Di-
ameter of a set E is diam (E) = sup{dist (x, y) : x, y ∈ E}. Con-
traction of a set E to a set F is a mapping f : E 7→ F such that
dist (f (x), f (y)) ≤ dist (x, y) for all x, y ∈ E. A set E can be contracted
onto a set F if there is a contraction f of E to F which is onto, i.e., such
that f (E) = F . Triangle is defined as the union of the three segments
joining its vertices, i.e., it does not contain the interior.

1.3.2 Day 2, 1996

Problem 1. (10 points)

Prove that if f : [0, 1] → [0, 1] is a continuous function, then the
sequence of iterates xn+l = f (xn ) converges if and only if

lim (xn+1 − xn ) = 0.
n→∞

Problem 2. (10 points)

et + e−t
Let θ be a positive real number and let cosht = denote the
2
hyperbolic cosine. Show that if k ∈ N and both coshkθ and cosh(k + 1)θ
are rational, then so is coshθ.
Problem 3. (15 points)
Let G be the subgroup of GL2 (R), generated by A and B, where
2 0 1 1
h i h i
A= 0 1 , B= 0 1

a11 a12
 
Let H consist of those matrices a a in G for which a11 = a22 = 1.
21 22
(a) Show that H is an abelian subgroup of G.
(b) Show that H is not finitely generated.
Remarks. GL2 (R) denotes, as usual, the group (under matrix mul-
tiplication) of all 2 × 2 invertible matrices with real entries (elements).
1.3. Olympic 1996 15

Abelian means commutative. A group is finitely generated if there are a

finite number of elements of the group such that every other element of
the group can be obtained from these elements using the group opera-
tion.
Problem 4. (20 points)
Let B be a bounded closed convex symmetric (with respect to the
origin) set in R2 with boundary the curve Γ. Let B have the property
that the ellipse of maximal area contained in B is the disc D of radius 1
centered at the origin with boundary the circle C. Prove that A ∩ Γ 6= ∅
π
for any arcA of C of length l(A) ≥ .
2
Problem 5. (20 points)
(i) Prove that
∞
X nx 1
lim 2 2
= .
n→+∞
n=1
(n + x) 2
(ii) Prove that there is a positive constant c such that for every x ∈
[1, ∞) we have
∞
X nx 1 c
2 + x)2
− ≤ .
n=1
(n 2 x
Problem 6. (Carleman’s inequality) (25 points)
(i) Prove that for every sequence {an }∞
n=1 such that an > 0, n =
∞
P
1, 2, . . . and an < ∞, we have
n=1
∞ ∞
1
X X
(a1 a2 . . . an ) < e
n an ,
n=1 n=1

where e is the natural log base.

(ii) Prove that for every  > 0 there exists a sequence {an }∞
n=1 such
∞
P
that an > 0, n = 1, 2, . . . , an < ∞ and
n=1
∞ ∞
1
X X
(a1 a2 . . . an ) > (e − )
n an .
n=1 n=1
1.4. Olympic 1997 16

1.4 Olympic 1997

1.4.1 Day 1, 1997

Problem 1.
Let {n }∞
n=1 be a sequence of positive real numbers, such that lim n =
n→∞
0. Find n
1 X k 
lim ln + n ,
n→∞ n n
k=1
where ln denotes the natural logarithm.
Problem 2.
P∞
Suppose an converges. Do the following sums have to converge as
n=1
well?
a) a1 + a2 + a4 + a3 + a8 + a7 + a6 + a5 + a16 + a15 + · · · + a9 + a32 + · · ·
b) a1 + a2 + a3 + a4 + a5 + a7 + a6 + a8 + a9 + a11 + a13 + a15 + a10 +
a12 + a14 + a16 + a17 + a19 + · · ·
Justify your answers.
Problem 3.
Let A and B be real n × n matrices such that A2 + B 2 = AB. Prove
that if BA − AB is an invertible matrix then n is divisible by 3.
Problem 4.
Let α be a real number, 1 < α < 2.
a) Show that α has a unique representation as an infinite product
 1  1
α= 1+ 1+ ...
n1 n2
b) Show that α is rational if and only if its infinite product has the
following property:
For some m and all k ≥ m,
nk+1 = n2k .
Problem 5. For a natural n consider the hyperplane
n n
X o
n n
R0 = x = (x1 , x2 , . . . , xn ) ∈ R : xi = 0
i=1
1.4. Olympic 1997 17

and the lattice Z0n = {y ∈ R0n : all yi are integers}. Define the (quasi-
P n 1/p
p
)norm in Rn by k x kp = |xi | if 0 < p < ∞, and k x k∞ =
i=1
max|xi |.
i
a) Let x ∈ R0n be such that

maxxi − minxi ≤ 1.
i i

For every p ∈ [1, ∞] and for every y ∈ Z0n prove that

k x kp ≤k x + y kp .

b) For every p ∈ (0, 1), show that there is an n and an x ∈ R0n with
maxxi − minxi ≤ 1 and an y ∈ Z0n such that
i i

k x kp >k x + y kp .

Problem 6. Suppose that F is a family of finite subsets of N and for

any two sets A, B ∈ F we have A ∩ B 6= ∅.
a) Is it true that there is a finite subset Y of N such that for any
A, B ∈ F we have A ∩ B ∩ Y 6= ∅?
b) Is the statement a) true if we suppose in addition that all of the
members of F have the same size?
Justify your answers.

1.4.2 Day 2, 1997

Problem 1.
Let f be a C 3 (R) non-negative function, f (0) = f 0 (0) = 0, 0 < f 00 (0).
Let
 pf (x) 0
g(x) =
f 0 (x)
for x 6= 0 and g(0) = 0. Show that g is bounded in some neighbourhood
of 0. Does the theorem hold for f ∈ C 2 (R)?
Problem 2.
1.4. Olympic 1997 18

Let M be an invertible matrix of dimension 2n × 2n, represented in

block form as
A B E F
h i h i
−1
M = C D and M = G H .

Show that detM.detH = detA.

Problem 3.
∞ (−1)n−1 sin(logn)
P
Show that converges if and only if α > 0.
n=1 nα
Problem 4.
2
a) Let the mapping f : Mn → R from the space Mn = Rn of n × n
matrices with real entries to reals be linear, i.e.:

f (A + B) = f (A) + f (B), f (cA) = cf (A) (1)

for any A, B ∈ Mn , c ∈ R. Prove that there exists a unique matrix

C ∈ Mn such that f (A) = tr(AC) for any A ∈ Mn . (If A = {aij }ni,j=1
n
P
then tr(A) = aii ).
i=1
b) Suppose in addition to (1) that

f (A.B) = f (B.A) (2)

for any A, B ∈ Mn . Prove that there exists λ ∈ R such that f (A) =

λ.tr(A).
Problem 5.
Let X be an arbitrary set, let f be an one-to-one function mapping
X onto itself. Prove that there exist mappings g1 , g2 : X → X such that
f = g1 ◦ g2 and g1 ◦ g1 = id = g2 ◦ g2 , where id denotes the identity
mapping on X.
Problem 6.
Let f : [0, 1] → R be a continuous function. Say that f ”crosses the
axis” at x if f (x) = 0 but in any neighbourhood of x there are y, z with
f (y) < 0 and f (z) > 0.
a) Give an example of a continuous function that ”crosses the axis”
infiniteley often.
1.5. Olympic 1998 19

b) Can a continuous function ”cross the axis” uncountably often?

Justify your answer.

1.5 Olympic 1998

1.5.1 Day 1, 1998

Problem 1. (20 points)

Let V be a 10-dimensional real vector space and U1 and U2 two linear
subspaces such that U1 ⊆ U2 , dimR U1 = 3 and dimR U2 = 6. Let  be
the set of all linear maps T : V → V which have U1 and U2 as invariant
subspaces (i.e., T (U1 ) ⊆ U1 and T (U2 ) ⊆ U2 ). Calculate the dimension
of  as a real vector space.
Problem 2. Prove that the following proposition holds for n = 3 (5
points) and n = 5 (7 points), and does not hold for n = 4 (8 points).
”For any permutation π1 of {1, 2, . . . , n} different from the identity
there is a permutation π2 such that any permutation π can be obtained
from π1 and π2 using only compositions (for example, π = π1 ◦ π1 ◦ π2 ◦
π1 ).”
Problem 3. Let f (x) = 2x(1 − x), x ∈ R. Define
n
z }| {
f (n) = f ◦ · · · ◦ f .
R1
a) (10 points) Find limn→∞ fn (x)dx
0
R1
b) (10 points) Compute fn (x)dx for n = 1, 2, . . ..
0
Problem 4. (20 points)
The function f : R → R is twice differentiable and satisfies f (0) =
2, f 0 (0) = −2 and f (1) = 1. Prove that there exists a real number
ξ ∈ (0, 1) for which
f (ξ).f 0 (ξ) + f 00 (ξ) = 0.
Problem 5. Let P be an algebraic polynomial of degree n having only
real zeros and real coefficients.
1.5. Olympic 1998 20

a) (15 points) Prove that for every real x the following inequality
holds:
(n − 1)(P 0 (x))2 ≥ nP (x)P 00 (x). (2)
b) (5 points) Examine the cases of equality.
Problem 6. Let f : [0, 1] → R be a continuous function with the
property that for any x and y in the interval,

xf (y) + yf (x) ≤ 1.

a) (15 points) Show that

Z1
π
f (x)dx ≤ .
4
0

b) (5 points) Find a function, satisfying the condition, for which there

is equality.

1.5.2 Day 2, 1998

Problem 1. (20 points)

Let V be a real vector space, and let f, f1 , . . . , fk be linear maps from
V to R Suppose that f (x) = 0 whenever f1 (x) = f2 (x) = · · · = fk (x) =
0. Prove that f is a linear combination of f1 , f2 , . . . , fk .
Problem 2. (20 points) Let
3
X 1
P = {f : f (x) = ak xk , ak ∈ R, |f (±1)| ≤ 1, |f (± )| ≤ 1}
2
k=0

Evaluate
sup max |f 00 (x)|
f ∈P −1≤x≤1
and find all polynomials f ∈ P for which the above ”sup” is attained.
Problem 3. (20 points) Let 0 < c < 1 and
x
 for x ∈ [0, c],
f (x) = 1c − x
 for x ∈ [c, 1].
1−c
1.6. Olympic 1999 21

We say that p is an n-periodic point if

f (f (. . . f ( p))) = p
| {z }
n

and n is the smallest number with this property. Prove that for every
n ≥ 1 the set of n-periodic points is non-empty and finite.
Problem 4. (20 points) Let An = {1, 2, . . . , n}, where n ≥ 3. Let F
be the family of all non-constant functions f : An → An satisfying the
following conditions:
(1) f (k) ≤ f (k + 1) for k = 1, 2, . . . , n − 1,
(2) f (k) = f (f (k + 1)) for k = 1, 2, . . . , n − 1. Find the number of
functions in F.
Problem 5. (20 points)
Suppose that S is a family of spheres (i.e., surfaces of balls of positive
radius) in R2 , n ≥ 2, such that the intersection of any two contains at
most one point. Prove that the set M of those points that belong to at
least two different spheres from S is countable.
Problem 6. (20 points) Let f : (0, 1) → [0, ∞) be a function that is
zero except at the distinct points a1 , a2 , . . .. Let bn = f (an ).
∞
P
(a) Prove that if bn < ∞, then f is differentiable at at least one
n=1
point x ∈ (0, 1).
(b) Prove that for any sequence of non-negative real numbers (bn )∞
n=1
∞
bn = ∞, there exists a sequence (an )∞
P
with n=1 such that the function
n=1
f defined as above is nowhere differentiable.

1.6 Olympic 1999

1.6.1 Day 1, 1999

Problem 1.
a) Show that for any m ∈ N there exists a real m × m matrix A such
that A3 = A + I, where I is the m × m identity matrix. (6 points)
1.6. Olympic 1999 22

b) Show that detA > 0 for every real m × m matrix satisfying A3 =

A + I. (14 points)
Problem 2. Does there exist a bijective map π : N → N such that
∞
X π(n)
< ∞?
n=1
n2

(20 points)
Problem 3. Suppose that a function f : R → R satisfies the inequality
n
X
3k (f (x + ky) − f (x − ky)) ≤ 1 (1)
k=1

for every positive integer n and for all x, y ∈ R. Prove that f is a

constant function. (20 points)
Problem 4. Find all strictly monotonic functions f : (0, +∞) →
 x2 
(0, +∞) such that f ≡ x. (20 points)
f (x)
Problem 5.
Suppose that 2n points of an n × n grid are marked. Show that for
some k > l one can select 2k distinct marked points, say a1 , . . . , a2k , such
that a1 and a2 are in the same row, a2 and a3 are in the same column,
... , a2k−l and a2k are in the same row, and a2k and a1 are in the same
column. (20 points)
Problem 6.
a) For each 1 < p < ∞ find a constant cp < ∞ for which the following
statement holds: If f : [−1, 1] → R is a continuously differentiable
function satisfying f (1) > f (−1) and |f 0 (y)| ≤ 1 for all y ∈ [−1, 1],
then there is an x ∈ [−1, 1] such that f 0 (x) > 0 and |f (y) − f (x)| ≤
cp (f 0 (x))1/p |y − x| for all y ∈ [−1, 1]. (10 points)
b) Does such a constant also exist for p = 1? (10 points)

1.6.2 Day 2, 1999

Problem 1. Suppose that in a not necessarily commutative ring R the

square of any element is 0. Prove that abc + abc = 0 for any three
1.7. Olympic 2000 23

elements a, b, c. (20 points)

Problem 2. We throw a dice (which selects one of the numbers 1, 2, . . . , 6
with equal probability) n times. What is the probability that the sum
of the values is divisible by 5? (20 points)
Problem 3.
n n n
x3i = 0. Prove that
P P
Assume that x1 , . . . , xn ≥ −1 and xi ≤ .
i=1 i=1 3
(20 points)
Problem 4. Prove that there exists no function f : (0, +∞) → (0, +∞)
such that f 2 (x) ≥ f (x + y)(f (x) + y) for any x, y > 0. (20 points)
Problem 5. Let S be the set of all words consisting of the letters x, y, z,
and consider an equivalence relation ∼ on S satisfying the following
conditions: for arbitrary words u, v, w ∈ S
(i) uu ∼ u;
(ii) if v ∼ w, then uv ∼ uw and vu ∼ wu.
Show that every word in S is equivalent to a word of length at most
8. (20 points)
Z 1
Problem 6. Let A be a subset of Zn = containing at most ln n
nZ 100
elements. Define the rth Fourier coefficient of A for r ∈ Zn by
X  2πi 
f (r) = exp sr .
n
s∈A

|A|
Prove that there exists an r 6= 0, such that |f (r)| ≥ . (20 points)
2

1.7 Olympic 2000

1.7.1 Day 1, 2000

Problem 1.
Is it true that if f : [0, 1] → [0, 1] is
a) monotone increasing
b) monotone decreasing then there exists an x ∈ [0, 1] for which
f (x) = x?
1.7. Olympic 2000 24

Problem 2.
Let p(x) = x5 + x and q(x) = x5 + x2 . Find all pairs (w, z) of complex
numbers with w 6= z for which p(w) = p(z) and q(w) = q(z).
Problem 3.
A and B are square complex matrices of the same size and

rank(AB − BA) = 1.

Show that (AB − BA)2 = 0.

Problem 4.
a) Show that if (xi ) is a decreasing sequence of positive numbers then
n n
X 1/2 x
√i .
X
x2i ≤
i=1 i=1
i

b) Show that there is a constant C so that if (xi ) is a decreasing

sequence of positive numbers then
∞ ∞ ∞
X 1  X 2 1/2 X
√ xi ≤C xi .
m=1
m i=m i=1

Problem 5.
Let R be a ring of characteristic zero (not necessarily commutative).
Let e, f and g be idempotent elements of R satisfying e + f + g = 0.
Show that e = f = g = 0.
(R is of characteristic zero means that, if a ∈ R and n is a positive
integer, then na 6= 0 unless a = 0. An idempotent x is an element
satisfying x = x2 .)
Problem 6.
Let f : R → (0, ∞ be an increasing differentiable function for which
lim f (x) = ∞ and f 0 is bounded.
x→∞
Rx
Let F (x) = f . Define the sequence (an ) inductively by
0

1
a0 = 1, an+1 = an + ,
f (an )
1.7. Olympic 2000 25

and the sequence (bn ) simply by bn = F −1 (n). Prove that lim (an −bn ) =
n→∞
0.

1.7.2 Day 2, 2000

Problem 1.
a) Show that the unit square can be partitioned into n smaller squares
if n is large enough.
b) Let d ≥ 2. Show that there is a constant N (d) such that, whenever
n ≥ N (d), a d-dimensional unit cube can be partitioned into n smaller
cubes.
Problem 2. Let f be continuous and nowhere monotone on [0, 1]. Show
that the set of points on which f attains local minima is dense in [0, 1].
(A function is nowhere monotone if there exists no interval where the
function is monotone. A set is dense if each non-empty open interval
contains at least one element of the set.)
Problem 3. Let p(z) be a polynomial of degree n with complex coeffi-
cients. Prove that there exist at least n + 1 complex numbers z for which
p(z) is 0 or 1.
Problem 4. Suppose the graph of a polynomial of degree 6 is tangent
to a straight line at 3 points A1 , A2 , A3 , where A2 lies between A1 and
A3 .
a) Prove that if the lengths of the segments A1 A2 and A2 A3 are equal,
then the areas of the figures bounded by these segments and the graph
of the polynomial are equal as well.
A2 A3
b) Let k = and let K be the ratio of the areas of the appropriate
A1 A2
figures. Prove that
2 5 7
k < K < k5.
7 2
+
Problem 5. Let R be the set of positive real numbers. Find all
functions f : R+ → R+ such that for all x, y ∈ R+

f (x)f (yf (x)) = f (x + y).

1.8. Olympic 2001 26

∞ 1
A
An .
P
Problem 6. For an m × m real matrix A, e is defined as
n=0 n!
(The sum is convergent for all matrices.) Prove or disprove, that for all
real polynomials p and m × m real matrices A and B, p(eAB ) is nilpotent
if and only if p(eBA ) is nilpotent. (A matrix A is nilpotent if Ak = 0 for
some positive integer k.)

1.8 Olympic 2001

1.8.1 Day 1, 2001

Problem 1.
Let n be a positive integer. Consider an n × n matrix with entries
1, 2, . . . , n2 written in order starting top left and moving along each row
in turn left-to-right. We choose n entries of the matrix such that exactly
one entry is chosen in each row and each column. What are the possible
values of the sum of the selected entries?
Problem 2.
Let r, s, t be positive integers which are pairwise relatively prime. If
a and b are elements of a commutative multiplicative group with unity
element e, and ar = bs = (ab)t = e, prove that a = b = e.
Does the same conclusion hold if a and b are elements of an arbitrary
noncommutative group?
P∞ tn
Problem 3. Find lim(1 − t) n
, where t % 1 means that t
t%1 n=1 1 + t
approaches 1 from below.
Problem 4.
Let k be a positive integer. Let p(x) be a polynomial of degree n each
of whose coefficients is −1, 1 or 0, and which is divisible by (x − 1)k . Let
q k
q be a prime such that < . Prove that the complex qth
ln q ln(n + 1)
roots of unity are roots of the polynomial p(x).
Problem 5.
Let A be an n × n complex matrix such that A 6= λI for all λ ∈ C.
1.8. Olympic 2001 27

Prove that A is similar to a matrix having at most one non-zero entry

on the main diagonal.
Problem 6.
Suppose that the differentiable functions a, b, f, g : R → R satisfy

f (x) ≥ 0, f 0 (x) ≥ 0, g(x) > 0, g 0 (x) > 0 for all x ∈ R,

lim a(x) = A > 0, lim b(x) = B > 0, lim f (x) = lim g(x) = ∞,
x→∞ x→∞ x→∞ x→∞

and
f 0 (x) f (x)
0
+ a(x) = b(x).
g (x) g(x)
Prove that
f (x) B
lim = .
x→∞ g(x) A+1

1.8.2 Day 2, 2001

Problem 1.
Let r, s ≥ 1 be integers and a0 , a1 . . . , ar−1 , b0 , b1 , . . . , bs−1 be real non-
negative numbers such that

(a0 + a1 x + a2 x2 + · · · + ar−1 xr−1 + xr )(b0 + b1 x + b2 x2 + · · · + bs−1 xs−1 + xs )

= 1 + x + x2 + · · · + xr+s−1 + xr+s .
Prove that each ai and each bj equals either 0 or 1.
Problem 2.
√ q
2b
pn
p
Let a0 = 2, b0 = 2, an+1 = 2 − 4 − a2n , bn+1 = .
2 + 4 + b2n
a) Prove that the sequences (an ), (bn ) are decreasing and converge to
0.
b) Prove that the sequence (2n an ) is increasing, the sequence (2n bn )
is decreasing and that these two sequences converge to the same limit.
c) Prove that there is a positive constant C such that for all n the
C
following inequality holds: 0 < bn − an < n .
8
Problem 3.
1.9. Olympic 2002 28

Find the maximum number of points on a sphere of radius 1 in Rn

such that the distance between any two of these points is strictly greater
√
than 2.
Problem 4.
Let A = (ak,l )k,l=1,...,n be an n × n complex matrix such that for each
m ∈ {1, . . . , n} and 1 ≤ j1 < · · · < jm ≤ n the determinant of the
matrix (ajk,jl )k,l=1,...,m is zero. Prove that An = 0 and that there exists
a permutation σ ∈ Sn such that the matrix

(aσ(k),σ(l) )k,l=1...,n

has all of its nonzero elements above the diagonal.

Problem 5. Let R be the set of real numbers. Prove that there is no
function f : R → R with f (0) > 0, and such that

f (x + y) ≥ f (x) + yf (f (x)) for all x, y ∈ R.

Problem 6.
For each positive integer n, let fn (ϑ) = sin ϑ. sin(2ϑ). sin(4ϑ) . . . sin(2n ϑ).
For all real ϑ and all n, prove that
2  π 
|fn (ϑ)| ≤ √ fn √ .
3 3

1.9 Olympic 2002

1.9.1 Day 1, 2002

Problem 1. A standard parabola is the graph of a quadratic polynomial

y = x2 +ax+b with leading coefficient 1. Three standard parabolas with
vertices V1 , V2 , V3 intersect pairwise at points A1 , A2 , A3 . Let A 7→ s(A)
be the reflection of the plane with respect to the x axis.
Prove that standard parabolas with vertices s (A1 ), s(A2 ), s(A3 ) in-
tersect pairwise at the points s(V1 ), s(V2 ), s(V3 ).
Problem 2. Does there exist a continuously differentiable function f :
R → R such that for every x ∈ R we have f (x) > 0 and f 0 (x) = f (f (x))?
1.9. Olympic 2002 29

Problem 3. Let n be a positive integer and let

1 k−n
ak = n , bk = 2

, for k = 1, 2, . . . , n.
k

Show that
a1 − b1 a2 − b2 an − b n
+ + ··· + = 0. (1)
1 2 n
Problem 4. Let f : [a, b] → [a, b] be a continuous function and let
p ∈ [a, b]. Define p0 = p and pn+1 = f (pn ) for n = 0, 1, 2, . . .. Suppose
that the set Tp = {pn : n = 0, 1, 2, . . .} is closed, i.e., if x ∈
/ Tp then there
is a δ > 0 such that for all x0 ∈ Tp we have |x0 − x| ≥ δ. Show that Tp
has finitely many elements.
Problem 5. Prove or disprove the following statements:
(a) There exists a monotone function f : [0, 1] → [0, 1] such that for
each y ∈ [0, 1] the equation f (x) = y has uncountably many solutions x.
(b) There exists a continuously differentiable function f : [0, 1] →
[0, 1] such that for each y ∈ [0, 1] the equation f (x) = y has uncountably
many solutions x.
Problem 6. For an n × n matrix M with real entries let k M k=
k M x k2
sup , where k . k2 denotes the Euclidean norm on Rn . As-
x∈Rn \{0} k x k2
sume that an n × n matrix A with real entries satisfies k Ak − Ak−1 k≤
1
for all positive integers k. Prove that k Ak k≤ 2002 for all positive
2002k
integers k.

1.9.2 Day 2, 2002

Problem 1. Compute the determinant of the n × n matrix A = [aij ],


(−1)|i−j| , if i 6= j
aij =
2, if i = j.
Problem 2. Two hundred students participated in a mathematical
contest. They had 6 problems to solve. It is known that each problem
was correctly solved by at least 120 participants. Prove that there must
1.10. Olympic 2003 30

be two participants such that every problem was solved by at least one
of these two students.
Problem 3. For each n ≥ 1 let
∞ ∞
X kn X
kk
n
an = , bn = (−1) .
k! k!
k=0 k=0
Show that an .bn is an integer.
Problem 4. In the tetrahedron OABC, let ∠BOC = α, ∠COA = β
and ∠AOB = γ. Let σ be the angle between the faces OAB and OAC,
and let τ be the angle between the faces OBA and OBC. Prove that
γ > β. cos σ + α cos τ.
Problem 5. Let A be an n×n matrix with complex entries and suppose
that n > 1. Prove that
−1
AA = In ⇔ ∃S ∈ GLn (C such that A = SS .
(If A = [aij ] then A = [aij ], where aij is the complex conjugate of
aij ; GLn (C) denotes the set of all n × n invertible matrices with complex
entries, and In is the identity matrix.)
Problem 6. Let f : Rn → R be a convex function whose gradient ∇f =
 ∂f ∂f 
,..., exists at every point of Rn and satisfies the condition
∂x1 ∂xn
∃L > 0 ∀x1 , x2 ∈ Rn k ∇f (x1 ) − ∇f (x2 ) k≤ L k x1 − x2 k .
Prove that
∀x1 , x2 ∈ Rn k ∇f (x1 ) − ∇f (x2 ) k2 ≤ L < ∇f (x1 ) − ∇f (x2 ), x1 − x2 > .
(1)
In this formula < a, b > denotes the scalar product of the vectors a and
b.

1.10 Olympic 2003

1.10.1 Day 1, 2003

Problem 1.
1.10. Olympic 2003 31

a) Let a1 , a2 , . . . be a sequence of real numbers such that a1 = 1 and

3
an+1 > an for all n. Prove that the sequence
2
an
 n−1
3
2

has a finite limit or tends to infinity. (10 points)

b) Prove that for all α > 1 there exists a sequence a1 , a2 , . . . with the
same properties such that
an
lim  n−1 = α.
3
2

(10 points)
Problem 2. Let a1 , a2 , . . . , a51 be non-zero elements of a field. We
simultaneously replace each element with the sum of the 50 remaining
ones. In this way we get a sequence b1 , . . . , b51 . If this new sequence
is a permutation of the original one, what can be the characteristic of
the field? (The characteristic of a field is p, if p is the smallest positive
{z· · · + x} = 0 for any element x of the field. If
integer such that |x + x +
p
there exists no such p, the characteristic is 0.) (20 points)
Problem 3. Let A be an n × n real matrix such that 3A3 = A2 + A + I
(I is the identity matrix). Show that the sequence Ak converges to an
idempotent matrix. (A matrix B is called idempotent if B 2 = B.) (20
points)
Problem 4. Determine the set of all pairs (a, b) of positive integers for
which the set of positive integers can be decomposed into two sets A and
B such that a.A = b.B. (20 points)
Problem 5. Let g : [0, 1] → R be a continuous function and let fn :
[0, 1] → R be a sequence of functions defined by f0 (x) = g(x) and
Zx
1
fn+1 (x) = fn (t)dt (x ∈ (0, 1], n = 0, 1, 2, . . .).
x
0
1.10. Olympic 2003 32

Determine lim fn (x) for every x ∈ (0, 1]. (20 points)

n→∞
Problem 6. Let f (z) = an z n + an−1 z n−1 + · · · + a1 z + a0 be a polynomial
with real coefficients. Prove that if all roots of f lie in the left half-plane
{z ∈ C : Rez < 0} then

ak ak+3 < ak+1 ak+2

holds for every k = 0, 1, ..., n − 3. (20 points)

1.10.2 Day 2, 2003

Problem 1. Let A and B be n×n real matrices such that AB +A+B =

0. Prove that AB = BA.
2. Evaluate the limit
Z2x
sinm t
lim dt (m, n ∈ N).
x→0+ tn
x

Problem 3. Let A be a closed subset of Rn and let B be the set of all

those points b ∈ Rn for which there exists exactly one point a0 ∈ A such
that
|a0 − b| = inf |a − b|.
a∈A
Prove that B is dense in R ; that is, the closure of B is Rn .
n

Problem 4. Find all positive integers n for which there exists a family
F of three-element subsets of S = {1, 2, . . . , n} satisfying the following
two conditions:
(i) for any two different elements a, b ∈ S, there exists exactly one
A ∈ F containing both a, b;
(ii) if a, b, c, x, y, z are elements of S such that if {a, b, x}, {a, c, y}, {b, c, z} ∈
F, then {x, y, z} ∈ F.
Problem 5. a) Show that for each function f : Q × Q → R there exists
a function g : Q → R such that f (x, y) ≤ g(x) + g(y) for all x, y ∈ Q.
b) Find a function f : R × R → R for which there is no function
g : R → R such that f (x, y) ≤ g(x) + g(y) for all x, y ∈ R.
1.11. Olympic 2004 33

Problem 6. Let (an )n∈N be the sequence defined by

n
1 X ak
a0 = 1, an+1 = .
n+1 n−k+2
k=0

Find the limit n

X ak
lim ,
n→∞ 2k
k=0
if it exists.

1.11 Olympic 2004

1.11.1 Day 1, 2004

Problem 1. Let S be an infinite set of real numbers such that |s1 +

s2 + · · · + sk | < 1 for every finite subset {s1 , s2 , . . . , sk } ⊂ S. Show that
S is countable. [20 points]
Problem 2. Let P (x) = x2 − 1. How many distinct real solutions does
the following equation have:

P (P (. . . (P (x)) . . .)) = 0?
| {z }
2004

[20 points]
n
P
Problem 3. Let Sn be the set of all sums xk , where n ≥ 2, 0 ≤
k=1
π
x1 , x2 , . . . , xn ≤ and
2 n
X
sin xk = 1.
k=1
a) Show that Sn is an interval. [10 points]
b) Let ln be the length of Sn . Find lim ln . [10 points]
n→∞
Problem 4. Suppose n ≥ 4 and let M be a finite set of n points in
R3 , no four of which lie in a plane. Assume that the points can be
coloured black or white so that any sphere which intersects M in at
least four points has the property that exactly half of the points in the
1.11. Olympic 2004 34

intersection of M and the sphere are white. Prove that all of the points
in M lie on one sphere. [20 points]
Problem 5. Let X be a set of 2k−4


k−2 + 1 real numbers, k ≤ 2. Prove
that there exists a monotone sequence {xn }ki=1 ⊇ X such that

|xi+1 − x1 | ≥ 2|xi − x1 |

for all i = 2, . . . , k − 1. [20 points]

Problem 6. For every complex number z 6= {0, 1} define
X
f (z) := (log z)−4 ,

where the sum is over all branches of the complex logarithm.

a) Show that there are two polynomials P and Q such that f (z) =
P (z)
for all z ∈ C\{0, 1}. [10 points]
Q(z)
b) Show that for all z ∈ C\{0, 1}
z 2 + 4z + 1
f (z) = z .
6(z − 1)4
[10 points]

1.11.2 Day 2, 2004

Problem 1. Let A be a real 4 × 2 matrix and B be a real 2 × 4 matrix

such that  1 0 −1 0 
0 1 0 −1
AB = −1 0 1 0  .
0 −1 0 1
Find BA. [20 points]
Problem 2. Let f, g : [a, b] → [0, ∞) be continuous and non-decreasing
functions such that for each x ∈ [a, b] we have
Zx p Zx p
f (t)dt ≤ g(t)dt
a a

Rb p Rb p
and f (t)dt = g(g)dt.
a a
1.12. Olympic 2005 35

Rb p Rb p
Prove that 1 + f (t)dt ≥ 1 + g(t)dt. [20 points]
a a
Problem 3. Let D be the closed unit disk in the plane, and let
p1 , p2 , . . . , pn be fixed points in D. Show that there exists a point p
in D such that the sum of the distances of p to each of p1 , p2 , . . . , pn is
greater than or equal to 1. [20 points]
Problem 4. For n ≥ 1 let M be an n × n complex matrix with dis-
tinct eigenvalues λ1 , λ2 , . . . , λk , with multiplicities m1 , m2 , . . . , mk , re-
spectively. Consider the linear operator LM defined by LM (X) = M X +
XM T , for any complex n × n matrix X. Find its eigenvalues and their
multiplicities. (M T denotes the transpose of M ; that is, if M = (mk,l ),
then M T = (ml,k ).) [20 points]
Problem 5. Prove that
Z1 Z1
dxdy
≤ 1.
x−1 + |ln y| − 1
0 0

[20 points]
Problem 6. For n ≥ 0 define matrices An and Bn as follows: A0 =
B0 = (1) and for every n > 0
An−1 An−1 An−1 An−1
   
An = A and Bn = A .
n−1 Bn−1 n−1 0
Denote the sum of all elements of a matrix M by S(M ). Prove that
S(An−1 n−1
k ) = S(Ak ) for every n, k ≥ 1. [20 points]

1.12 Olympic 2005

1.12.1 Day 1, 2005

Problem 1. Let A be the n × n matrix, whose (i, j)th entry is i + j for

all i, j = 1, 2, . . . , n. What is the rank of A?
Problem 2. For an integer n ≥ 3 consider the sets
Sn = {(x1 , x2 , . . . , xn ) : ∀i xi ∈ {0, 1, 2}}
An = {(x1 , x2 , . . . , xn ) ∈ Sn : ∀i ≤ n − 2 |{xi , xi+1 , xi+2 }| =
6 1}
1.12. Olympic 2005 36

and
Bn = {(x1 , x2 , . . . , xn ) ∈ Sn : ∀i ≤ n − 1 (xi = xi+1 ⇒ xi 6= 0)}.
Prove that |An+1 = 3.|Bn |. (|A| denotes the number of elements of the
set A.)
Problem 3. Let f : R → [0, ∞) be a continuously differentiable func-
tion. Prove that
Z1 Z1  Z1 2
3 2 0
f (x)dx − f (0) f (x)dx ≤ max |f (x)| f (x)dx .
0≤x≤1
0 0 0
Problem 4. Find all polynomials P (x) = an xn + an−1 xn−1 + · · · + a1 x +
a0 (an 6= 0) satisfying the following two conditions:
(i) (a0 , a1 , . . . , an ) is a permutation of the numbers (0, 1, ..., n)
and
(ii) all roots of P (x) are rational numbers.
Problem 5. Let f : (0, ∞) → R be a twice continuously differentiable
function such that
|f 00 (x) + 2xf 0 (x) + (x2 + 1)f (x)| ≤ 1
for all x. Prove that lim f (x) = 0.
x→∞
Problem 6. Given a group G, denote by G(m) the subgroup generated
by the mth powers of elements of G. If G(m) and G(n) are commutative,
prove that G(gcd(m, n)) is also commutative. (gcd(m, n) denotes the
greatest common divisor of m and n.)

1.12.2 Day 2, 2005

Problem 1. Let f (x) = x2 + bx + c, where b and c are real numbers,

and let
M = {x ∈ R : |f (x)| < 1}.
Clearly the set M is either empty or consists of disjoint open intervals.
Denote the sum of their lengths by |M |. Prove that
√
|M | ≤ 2 2.
1.13. Olympic 2006 37

Problem 2. Let f : R → R be a function such that (f (x))n is a

polynomial for every n = 2, 3, . . .. Does it follow that f is a polynomial?
Problem 3. In the linear space of all real n × n matrices, find the
maximum possible dimension of a linear subs pace V such that

∀X, Y ∈ V trace(XY ) = 0.

(The trace of a matrix is the sum of the diagonal entries.)

Problem 4. Prove that if f : R → R is three times differentiable, then
there exists a real number ξ ∈ (−1, 1) such that
f 000 (ξ) f (1) − f (−1)
= − f 0 (0).
6 2
Problem 5. Find all r > 0 such that whenever f : R → R is a differen-
tiable function such that |grad f (0, 0)| = 1 and |grad f (u) − grad f (v)| ≤
|u − v| for all u, v ∈ R2 , then the maximum of f on the disk {u ∈ R2 :
|u| ≤ r} is attained at exactly one point. (grad f (u) = (∂1 f (u), ∂2 f (u))
is the gradient vector of f at the point u. For a vector u = (a, b), |u| =
√
a2 + b2 .)
√
Problem 6. Prove that if pand qare rational numbers and r = p+q 7,
a b 1 0
 
then there exists a matrix c d 6= ± 0 1 with integer entries and
with ad − bc = 1 such that
ar + b
= r.
cr + d

1.13 Olympic 2006

1.13.1 Day 1, 2006

Problem 1. Let f : R → R be a real function. Prove or disprove each

of the following statements.
a) If f is continuous and range(f ) = R then f is monotonic.
b) If f is monotonic and range(f ) = R then f is continuous.
c) If f is monotonic and f is continuous then range(f ) = R.
(20 points)
1.13. Olympic 2006 38

Problem 2. Find the number of positive integers x satisfying the fol-

lowing two conditions:

1. x < 102006 ;

2. x2 − x is divisible by 102006 .

(20 points)
Problem 3. Let A be an n×n-matrix with integer entries and b1 , . . . , bk
be integers satisfying detA = b1 . . . bk . Prove that there exist n × n-
matrices B1 , . . . , Bk with integer entries such that A = B1 . . . Bk and
detBi = bi for all i = 1, . . . , k. (20 points)
Problem 4. Let f be a rational function (i.e. the quotient of two real
polynomials) and suppose that f (n) is an integer for infinitely many
integers n. Prove that f is a polynomial. (20 points)
Problem 5. Let a, b, c, d, e > 0 be real numbers such that a2 + b2 + c2 =
d2 + e2 and a4 + b4 + c4 = d4 + e4 . Compare the numbers a3 + b3 + c3
and d3 + e3 . (20 points)
Problem 6. Find all sequences a0 , a1 , . . . , an of real numbers where
n ≥ 1 and an 6= 0, for which the following statement is true:
If f : R → R is an n times differentiable function and x0 < x1 < · · · <
xn are real numbers such that f (x0 ) = f (x1 ) = · · · = f (xn ) = 0 then
there exists an h ∈ (x0 , xn ) for which

a0 f (h) + a1 f 0 (h) + · · · + an f (n) (h) = 0.

(20 points)

1.13.2 Day 2, 2006

Problem 1. Let V be a convex polygon with n vertices.

a) Prove that if n is divisible by 3 then V can be triangulated (i.e.
dissected into non-overlapping triangles whose vertices are vertices of V )
so that each vertex of V is the vertex of an odd number of triangles.
1.13. Olympic 2006 39

b) Prove that if n is not divisible by 3 then V can be triangulated

so that there are exactly two vertices that are the vertices of an even
number of the triangles.
(20 points)
Problem 2. Find all functions f : R → R such that for any real
numbers a < b, the image f ([a, b]) is a closed interval of length b − a.
(20 points)
Problem 3. Compare tan(sin x) and sin(tan x) for all x ∈ (0, π2 ).
(20 points)
Problem 4. Let v0 be the zero vector in Rn and let v1 , v2 , . . . , vn+1 ∈ Rn
be such that the Euclidean norm |vi − vj | is rational for every 0 ≤ i, j ≤
n + 1. Prove that v1 , . . . , vn+1 are linearly dependent over the rationals.
(20 points)
Problem 5. Prove that there exists an infinite number of relatively
prime pairs (m, n) of positive integers such that the equation

(x + m)3 = nx

has three distinct integer roots.

(20 points)
Problem 6. Let Ai , Bi , Si (i = 1, 2, 3) be invertible real 2 × 2 matrices
such that
1) not all Ai have a common real eigenvector;
2) Ai = Si−1 Bi Si for all i = 1, 2,
 3;
1 0
3) A1 A2 A3 = B1 B2 B3 = 0 1 .
Prove that there is an invertible real 2 × 2 matrix S such that Ai =
S −1 Bi S for all i = 1, 2, 3.
(20 points)
1.14. Olympic 2007 40

1.14 Olympic 2007

1.14.1 Day 1, 2007

Problem 1. Let f be a polynomial of degree 2 with integer coefficients.

Suppose that f (k) is divisible by 5 for every integer k. Prove that all
coefficients of f are divisible by 5.
Problem 2. Let n > 2 be an integer. What is the minimal and maximal
possible rank of an n × n matrix whose n2 entries are precisely the
numbers 1, 2, . . . , n2 ?
Problem 3. Call a polynomial P (x1 , . . . , xk ) good if there exist 2 × 2
real matrices A1 , . . . , Ak such that
k 
P
P (x1 , . . . , xk ) = det xi Ai .
i=1

Find all values of k for which all homogeneous polynomials with k vari-
ables of degree 2 are good.
(A polynomial is homogeneous if each term has the same total degree.)
Problem 4. Let G be a finite group. For arbitrary sets U, V, W ⊂ G,
denote by NU V W the number of triples (x, y, z) ∈ U × V × W for which
xyz is the unity.
Suppose that G is partitioned into three sets A, B and C (i.e. sets
A, B, C are pairwise disjoint and G = A ∪ B ∪ C). Prove that NABC =
NCBA .
Problem 5. Let n be a positive integer and a1 , . . . , an be arbitrary
Pn
integers. Suppose that a function f : Z → R satisfies f (k + ai l) = 0
i=1
whenever k and l are integers and l 6= 0. Prove that f = 0.
Problem 6. How many nonzero coefficients can a polynomial P (z) have
if its coefficients are integers and |P (z) 6 2| for any complex number z
of unit length?
1.15. Olympic 2008 41

1.14.2 Day 2, 2007

Problem 1. Let f : R → R be a continuous function. Suppose that for

any c > 0, the graph of f can be moved to the graph of cf using only a
translation or a rotation. Does this imply that f (x) = ax + b for some
real numbers a and b?
Problem 2. Let x, y and z be integers such that S = x4 + y 4 + z 4 is
divisible by 29. Show that S divisible by 294 .
Problem 3. Let C be a nonempty closed bounded subset of the real
line and f : C → C be a nondecreasing continuous function. Show that
there exists a point p ∈ C such that f (p) = p.
(A set is closed if its complement is a union of open intervals. A
function g is nondecreasing if g(x) 6 g(y) for all x 6 y.)
Problem 4. Let n > 1 be an odd positive integer and A = (aij )i,j=1,...,n
be the n × n matrix with

2 if i = j
aij = 1 if i − j ≡ ±2 (mod n)
0 otherwise.

Find det A.
Problem 6. Let f 6= 0 be a polynomial with real coefficients. Define
the sequence f0 , f1 , f2 , . . . of polynomials by f0 = f and fn+1 = fn + fn0
for every n > 0. Prove that there exists a number N such that for every
n > N , all roots of fn are real.

1.15 Olympic 2008

1.15.1 Day 1, 2008

Problem 1. Find all continuous functions f : R → R such that f (x) −

f (y) is rational for all reals x and y such that x − y is rational.
Problem 2. Denote by V the real vector space of all real polynomials
in one variable, and let P : V → R be a linear map. Suppose that for
1.15. Olympic 2008 42

all f, g ∈ V with P (f g) = 0 we have P (f ) = 0 or P (g) = 0. Prove that

there exist real numbers x0 , c such that P (f ) = cf (x0 ) for all f ∈ V .
Problem 3. Let p be a polynomial with integer coefficients and let
a1 < a2 < · · · < ak be integers.
a) Prove that there exists a ∈ Z such that p(ai ) divides p(a) for all
i = 1, 2, . . . , k.

b) Does there exist an a ∈ Z such that the product p(a1 ).p(a2 ) . . . p(ak )
divides p(a)?
Problem 4. We say a triple (a1 , a2 , a3 ) of nonnegative reals is better
than another triple (b1 , b2 , b3 ) if two out of the three following inequalities
a1 > b1 , a2 > b2 , a3 > b3 are satisfied. We call a triple (x, y, z) special if
x, y, z are nonnegative and x + y + z = 1. Find all natural numbers n for
which there is a set S of n special triples such that for any given special
triple we can find at least one better triple in S.
Problem 5. Does there exist a finite group G with a normal subgroup
H such that |Aut H| > |Aut G|?
Problem 6. For a permutation σ = (i1 , i2 , . . . , in ) of (1, 2, . . . , n) define
Pn
D(σ) = |ik − k|. Let Q(n, d) be the number of permutations σ of
k=1
(1, 2, . . . , n) with d = D(σ). Prove that Q(n, d) is even for d > 2n.

1.15.2 Day 2, 2008

Problem 1. Let n, k be positive integers and suppose that the polyno-

mial x2k − xk + 1 divides x2n + xn + 1. Prove that x2k + xk + 1 divides
x2n + xn + 1.
Problem 2. Two different ellipses are given. One focus of the first
ellipse coincides with one focus of the second ellipse. Prove that the
ellipses have at most two points in common.
Problem 3. Let n be a positive integer. Prove that 2n−1 divides
X  n 
5k .
n 2k + 1
06k< 2
1.15. Olympic 2008 43

Problem 4. Let Z[x] be the ring of polynomials with integer coefficients,

and let f (x), g(x) ∈ Z[x] be nonconstant polynomials such that g(x)
divides f (x) in Z[x]. Prove that if the polynomial f (x) − 2008 has at
least 81 distinct integer roots, then the degree of g(x) is greater than 5.
Problem 5. Let n be a posotive integer, and consider the matrix A =
(aij )16i,j6n , where

1 if i + j is a prime number,
aij =
0 otherwise.

Prove that | det A| = k 2 for some integer k.

Problem 6. Let H be an infinite-dimensional real Hilbert space, let
d > 0, and suppose that S is a set of points (not necessarily countable)
in H such that the distance between any two distinct points in S is equal
to d. Show that there is a point y ∈ H such that
n √2 o
(x − y) : x ∈ S
d
is an orthonormal system of vectors in H.
Chapter 2

Solutions

2.1 Solutions of Olympic 1994

2.1.1 Day 1

Problem 1.
Denote by aij and bij the elements of A and A−1 , respectively. Then
Pn
for k 6= m we have aki bim = 0 and from the positivity of aij we
i=0
conclude that at least one of {bim : 1, 2, . . . , n} is positive and at least
one is negative. Hence we have at least two non-zero elements in every
column of A−1 . This proves part a). For part b) all bij are zero except
b1,1 = 2, bn,n = (−1)n , bi,i+1 = bi+1,i = (−1)i for i = 1, 2, . . . , n − 1.
Problem 2. From the inequality we get
d −1 f 0 (x)
(tan f (x) + x) = +1≥0
dx 1 + f 2 (x)
for x ∈ (a, b). Thus tan−1 f (x) + x is non-decreasing in the interval and
π π
using the limits we get + a ≤ − + b. Hence b − a ≥ pi. One has
2 2
equality for f (x) = cotgx, a = 0, b = π.
Problem 3. Let I be the set of irrational numbers, Q-the set of rational
numbers, Q+ = Q ∪ [0, ∞). We work by induction. For n = 1 the
statement is trivial. Let it be true for n − 1. We start to prove it
for n. From the induction argument there are n − 1 different elements
x1 , x2 , . . . , xn−1 ∈ S such that
a1 x1 + a2 x2 + · · · + an−1 xn−1 ∈ I
f orall a1 , a2 , . . . , an ∈ Q+ with a1 + a2 + · · · + an−1 > 0 (1)

44
2.1. Solutions of Olympic 1994 45

Denote the other elements of S by xn , xn+1 , . . . , x2n−1 . Assume the state-

ment is not true for n. Then for k = 0, 1, . . . , n − 1 there are rk ∈ Q
such that
n−1
X n−1
X
+
bik xi + ck xn+k = rk for some bik , ck ∈ Q , bik + ck > 0. (2)
i=1 i=1

Also
n−1
X n−1
X
+
dk xn+k = R for some dk ∈ Q , dk > 0, R ∈ Q. (3)
k=0 k=0

6 0 and without loss

If in (2) ck = 0 then (2)contradicts (1). Thus ck =
n−1
P
of generality one may take ck = 1. In (2) also bik > 0 in view of
i=1
xn+k ∈ I. Replacing (2) in (3) we get
n−1
X  n−1
X  n−1  X
X n−1 
dk − bik xi + rk = R or dk bik xi ∈ Q,
k=0 i=1 i=1 k=0

which contradicts (1) because of the condition on b0 s and d0 s.

Problem 4. For a) using the assumptions we have
k
X
k k
F ◦G−G◦F = (F k−i+1 ◦ G ◦ F i−1 − F k−i ◦ G ◦ F i ) =
i=1
Xk
= F k−i ◦ (F ◦ G − G ◦ F ) ◦ F i−1 =
i=1
Xk
= F k−i ◦ αF ◦ F i−1 = αkF k .
i=1

b) Consider the linear operator L(F ) = F ◦G−G◦F acting over all n×n
matrices F . It may have at most n2 different eigenvalues. Assuming that
F k 6= 0 for every k we get that L has infinitely many different eigenvalues
αk in view of a) -a contradiction.
Rb
Problem 5. Set k g k1 = |g(x)|dx and
0

ω(f, t) = sup{|f (x) − f (y)| : x, y ∈ [0, b], |x − y| ≤ t}.

2.1. Solutions of Olympic 1994 46

In view of the uniform continuity of f we have ω(f, t) → 0 as t → 0.

Using the periodicity of g we get
Zb bk/n
n
X Z
f (x)g(nx)dx = f (x)g(nx)dx
0 k=1
b(k−1)/n
bk/n bk/n
n
X Z n
X Z
= f (bk/n) g(nx)dx + {f (x) − f (bk/n)}g(nx)dx
k=1 k=1
b(k−1)/n b(k−1)/n

n Zb
1X
= f (bk/n) g(x)dx + O(ω(f, b/n) k g k1 )
n
k=1 0
bk/n Zb
n Z
1X
= f (x)dx g(x)dx
n
k=1 0
b(k−1)/n
bk/n
n
1 Xb
Z  Zb
+ f (bk/n) − f (x)dx g(x)dx + O(ω(f, b/n) k g k1 )
b n
k=1 0
b(k−1)/n
Zb Zb
1
= f (x)dx g(x)dx + O(ω(f, b/n) k g k1 ).
b
0 0

This proves a). For b) we set b = π, f (x) = sin x, g(x) = (1+3 cos2 x)−1 .
From a) and
Zπ Zπ
π
sin xdx = 2, (1 + 3 cos2 x)−1 dx =
2
0 0

we get
Zπ
sin x
lim dx = 1.
n→∞ 1 + 3 cos2 nx
0
Problem 6. a) For i = 1, 2, . . . , k we have

bi = f (mi ) − f (mi−1 ) = (mi − mi−1 )f 0 (xi )

2.1. Solutions of Olympic 1994 47

bi bi
for some xi ∈ (mi−1 , mi ). Hence = f 0 (xi ) and so −1 < < 1.
ai ai
bi
From the convexity of f we have that f 0 is increasing and = f 0 (xi ) <
ai
bi+1
f 0 (xi+1 ) = because of xi < mi < xi+1 .
ai+1
b) Set SA = {j ∈ {0, 1, . . . , k} : aj > A}. Then
k
X X
N ≥ mk − m0 = ai ≥ aj > A|SA |
i=1 j∈SA

N
and hence |SA < .
A
c) All different fractions in (−1, 1) with denominators less or equal A
N
are no more 2A2 . Using b) we get k < + 2A2 . Put A = N 1/3 in the
A
2/3
above estimate and get k < 3N .

2.1.2 Day 2

Problem 1. Assume that there is y ∈ (a, b] such that f (y) 6= 0. Without

loss of generality we have f (y) > 0. In view of the continuity of f there
exists c ∈ [a, y) such that f (c) = 0 and f (x) > 0 for x ∈ (c, y]. For
x ∈ (c, y] we have |f 0 (x)| ≤ λf (x). This implies that the function g(x) =
0 f 0 (x)
ln f (x) − λx is not increasing in (c, y] because of g (x) = − λ ≤ 0.
f (x)
Thus ln f (x) − λx ≥ ln f (y) − λy and f (x) ≥ eλx−λy f (y) for x ∈ (c, y].
Thus
0 = f (c) = f (c + 0) ≥ eλc−λy f (y) > 0
Problem 2. We have f (1, 0) = e−1 , f (0, 1) = −e−1 and te−t ≤ 2e−2
2 2
for t ≥ 2. Therefore |f (x, y)| ≤ (x2 + y 2 )e−x −y ≤ 2e−2 < e−1 for
(x, y) 6= M = {(u, v) : u2 + v 2 ≤ 2} and f cannot attain its minimum
and its maximum outside M . Part a) follows from the compactness of
M and the continuity of f . Let (x, y) be a point from part b). From
∂f 2 2
(x, y) = 2x(1 − x2 + y 2 )e−x −y we get
∂x
x(1 − x2 + y 2 ) = 0. (1)
2.1. Solutions of Olympic 1994 48

Similarly
y(1 + x2 − y 2 ) = 0. (2)
All solutions (x, y) of the system (1), (2) are (0, 0), (0, 1), (0, −1), (1, 0)
and (−1, 0). One has f (1, 0) = f (−1, 0) = e−1 and f has global maxi-
mum at the points (1, 0) and (−1, 0). One has f (0, 1) = f (0, −1) = −e−1
and f has global minimum at the points (0, 1) and (0, −1). The point
2
(0, 0) is not an extrema point because of f (x, 0) = x2 e−x > 0 if x 6= 0
2
and f (y, 0) = −y 2 e−y < 0 if y 6= 0.
Problem 3. Set g(x) = (f (x) + f 0 (x) + · · · + f (n) (x))e−x . From the
assumption one get g(a) = g(b). Then there exists c ∈ (a, b) such that
g 0 (c) = 0. Replacing in the last equality g 0 (x) = (f (n+1) (x) − f (x))e−x
we finish the proof.
Problem 4. Set A = (aij )ni,j=1 , B = (bij )ni,j=1 , AB = (xij )ni,j=1 and
BA = (yij )ni,j=1 . Then xij = aii bij and yij = ajj bij . Thus AB = BA
is equivalent to (aii − ajj )bij for i, j = 1, 2, . . . , n. Therefore bij = 0 if
aii 6= ajj and bij may be arbitrary if aii = ajj . The number of indices
(i, j) for which aii = ajj = cm for some m = 1, 2, . . . , k is d2m . This gives
the desired result.
Problem 5. We define π inductively. Set π(1) = 1. Assume π is defined
for i = 1, 2, . . . , n and also
n
X n
X
2
k xπ(i) k ≤ k xπ(i) k2 . (1)
i=1 i=1

Note (1) is true for n = 1. We choose π(n+1) in a way that (1) is fulfilled
Pn
with n+1 instead of n. Set y = xπ(i) and A = {1, 2, . . . , k}\{π(i) : i =
i=1  P 
1, 2, . . . , n}. Assume that (y, xr ) > 0 for all r ∈ A. Then y, xr > 0
P r∈A
and in view of y + xr = 0 one gets −(y, y) > 0, which is impossible.
r∈A
Therefore there is r ∈ A such that

(y, xr ) ≤ 0. (2)
2.1. Solutions of Olympic 1994 49

Put π(n + 1) = r. Then using (2) and (1) we have

n+1
X
k xπ(i) k2 =k y+xr k2 =k y k2 +2(y, xr )+ k xr k2 ≤k y k2 + k xr k2 ≤
i=1
n
X n+1
X
2 2
≤ k xπ(i) k + k xr k = k xπ(i) k2 ,
i=1 i=1

which verifies (1) for n + 1. Thus we define π for every n = 1, 2, . . . , k.

Finally from (1) we get
n
X n
X k
X
2 2
k xπ(i) k ≤ k xπ(i) k ≤ k xi k2 .
i=1 i=1 i=1

Problem 6. Obviously
N −2
ln 2 N X 1 ln 2 N N − 3 3
AN = ≥ . 2 =1− . (1)
N ln k.ln (N − k) N ln N N
k=2

N 1
Take M, 2 ≤ M < . Then using that is decreasing in
2 ln k.ln (N − k)
[2, N2 ] and the symmetry with respect to N2 one get
M N −M
X−1
ln 2 N n X X o 1
AN = + + N −2 ≤
N ln k.ln (N − k)
k=2 k=M +1 k=N −M
ln 2 N  M −1 N − 2M − 1 o
≤ 2 + ≤
N ln 2.ln (N − 2) ln M.ln (N − M )
2 M ln N  2M  ln N  1 
≤ . + 1− +O
ln 2 N N ln M ln N
h N i
Choose M = + 1 to get
ln 2 N
 2  ln N  1   ln ln N 
AN ≤ 1 − +O ≤ 1+O . (2)
N ln 2 N ln N − 2ln ln N ln N ln N
Estimates (1) and (2) give
N −2
ln 2 N X 1
lim = 1.
N →∞ N ln k.ln (N − k)
k=2
2.2. Solutions of Olympic 1995 50

2.2 Solutions of Olympic 1995

2.2.1 Day 1

Problem 1. Let J = (aij ) be the n × n matrix where aij = 1 if i = j + 1

and aij = 0 otherwise. The rank of J is n−1 and its only eigenvalues are
0’s. Moreover Y = XJ and A = Y X −1 = XJX −l , B = X −1 Y = J. It
follows that both A and B have rank n − 1 with only 0’s for eigenvalues.
Problem 2. From the inequality
Z1 Z1 Z1 Z1
0≤ (f (x) − x)2 dx = f 2 (x)dx − 2 xf (x)dx + x2 dx
0 0 0 0

we get
Z1 Z1 Z1 Z1
1
f 2 (x)dx ≥ 2 xf (x)dx − x2 dx = 2 xf (x)dx − .
3
0 0 0 0

R1 R1 R1 1 − x2 R1
From the hypotheses we have f (t)dtdx ≥ dx or tf (t)dt ≥
0 x 0 2 0
1
. This completes the proof.
3
Problem 3. Since f 0 tends to −∞ and f ” tends to +∞ as x tends to
0+, there exists an interval (0, r) such that f 0 (x) < 0 and f ”(x) > 0 for
all x ∈ (0, r). Hence f is decreasing and f 0 is increasing on (0, r). By
the mean value theorem for every 0 < x < x0 < r we obtain

f (x) − f (x0 ) = f 0 (ξ)(x − x0 ) > 0,

for some ξ ∈ (x, x0 ). Taking into account that f 0 is increasing, f 0 (x) <
f 0 (ξ) < 0, we get
f 0 (ξ) f (x) − f (x0 )
x − x0 < 0 (x − x0 ) = < 0.
f (x) f 0 (x)
Taking limits as x tends to 0+ we obtain
f (x) f (x)
−x0 ≤ lim inf ≤ lim sup ≤ 0.
x→0+ f 0 (x) x→0+ f 0 (x)
2.2. Solutions of Olympic 1995 51

f (x)
Since this happens for all x0 ∈ (0, r) we deduce that limx→0+ exists
f 0 (x)
f (x)
and limx→0+ = 0.
f 0 (x)
Problem 4. From the definition we have
x−1
F 0 (x) = , x > 1.
ln x
Therefore F 0 (x) > 0 for x ∈ (1, ∞). Thus F is strictly increasing and
hence one-to-one. Since
2 1 2 x2 − x
F (x) ≥ (x − x) min{ :x≤t≤x }= →∞
ln t ln x2
as x → ∞, it follows that the range of F is (F (1+), ∞). In order to
determine F (l+) we substitute t = ev in the definition of F and we get
ln x
2Z
ev
F (x) = dv.
v
ln x
Hence
ln x
2Z
1
F (x) < e2lnx dv = x2 ln2
v
ln x
and similarly F (x) > x
ln2. Thus F (1+) = ln 2.
Problem 5. We have that

(A + tB)n = An + tP1 + t2 P2 + · · · + tn−1 Pn−1 + tn B n

for some matrices P1 , P2 , . . . , Pn−l not depending on t.

Assume that a, p1 , p2 , . . . , pn−1 , b are the (i, j)-th entries of the corre-
sponding matrices An , P1 , P2 , . . . , Pn−1 , B n . Then the polynomial

btn + pn−1 tn−1 + · · · + p2 t2 + p1 t + a

has at least n + 1 roots t1 , t2 , . . . , tn+1 . Hence all its coefficients vanish.

Therefore An = 0, B n = 0, Pi = 0; and A and B are nilpotent.
2.2. Solutions of Olympic 1995 52

Problem 6. Let 0 < δ < 1. First we show that there exists Kp,δ > 0
such that
(x − y)2
f (x, y) = ≤ Kp,δ
4 − (x + y)2
for every (x, y) ∈ Dδ = {(x, y) : |x − y| ≥ δ, |x|p + |y|p = 2}.
Since Dδ is compact it is enough to show that f is continuous on
Dδ . For this we show that the denominator of f is different from zero.
x+y p
Assume the contrary. Then |x + y| = 2 and = 1. Since p > 1,
2
x+y p
the function g(t) = |t|p is strictly convex, in other words <
2
|x|p + |y|p x+y p
whenever x 6= y. So for some (x, y) ∈ Dδ we have <
2 2
|x|p + |y|p x+y p
=1= . We get a contradiction.
2 2
If x and y have different signs then (x, y) ∈ Dδ for all 0 < δ < 1
because then |x − y| ≥ max{|x|, |y|} ≥ 1 > δ. So we may further assume
without loss of generality that x > 0, y > 0 and xp +y p = 2. Set x = 1+t.
Then

y = (2 − xp )1/p = (2 − (1 + t)p )1/p

 p(p − 1) 2 2
1/p p(p − 1) 2
= 2 − (1 + pt + t + o(t )) = (1 − pt − t + o(t2 ))1/p
2 2
1  p(p − 1) 2  1 1
 
=1 + − pt − t + o(t2 ) + − 1 (−pt + o(t2 ))2 + o(t2 )
p 2 2p p
p−1 2 p−1 2
=1 − t − t + o(t2 ) − t + o(t2 )
2 2
2 2
=1 − t − (p − 1)t + o(t ).

We have
(x − y)2 = (2t + o(t))2 = 4t2 + o(t2 )
and
4 − (x + y)2 = 4 − (2 − (p − 1)t2 + o(t2 ))2
= 4 − 4 + 4(p − 1)t2 + o(t2 ) = 4(p − 1)t2 + o(t2 ).
So there exists δp > 0 such that if |t| < δp we have (x − y)2 < 5t2 , 4 −
2.2. Solutions of Olympic 1995 53

(x + y)2 > 3(p − 1)t2 . Then

5 5
(x − y)2 < 5t2 = .3(p − 1)t2 < (4 − (x + y)2 ) (*)
3(p − 1) 3(p − 1)
if |x − 1| < δp . From the symmetry we have that (*) also holds when
|y − 1| < δp .
To finish the proof it is enough to show that |x − y| ≥ 2δp whenever
|x−1| ≥ δp , |y−1| ≥ δp and xp +y p = 2. Indeed, since xp +y p = 2 we have
 x + y p xp + y p
that max{x, y} ≥ 1. So let x − 1 ≥ δp . Since ≤ =1
2 2
we get x + y ≤ 2. Then x − y ≥ 2(x − 1) ≥ 2δp .

2.2.2 Day 2

Problem 1. a) Set A = (aij ), u = (u1 , u2 , u3 )T . If we use the orthogo-

nallity condition
(Au, u) = 0 (1)
with ui = δik we get akk = 0. If we use (1) with ui = δik + δim we get

akk + akm + amk + amm = 0

and hence akm = −amk .

b) Set v1 = −a23 , v2 = a13 , v3 = −a12 . Then

Au = (v2 u3 − v3 u2 , v3 u1 − v1 u3 , v1 u2 − v2 u1 )T = v × u.

Problem 2. (15 points)

Let {bn }∞
n=0 be a sequence of positive real numbers such that b0 =
√ p √
1, bn = 2 + bn−1 − 2 1 + bn−1 . Calculate
∞
X
bn 2n .
n=1
√
Solution. Put an = 1 + bn for n ≥ 0. Then an > 1, a0 = 2 and
√ √
q
an = 1 + 1 + an−1 − 2 an−1 = an−1 ,
2.2. Solutions of Olympic 1995 54

so an = 22−n . Then
N
X N
X N
X
n 2 n
bn 2 = (an − 1) 2 = [a2n 2n − an 2n+1 + 2n ]
n=1 n=1 n=1
XN
= [(an−1 − 1)2n − (an − 1)2n+1 ]
n=1
−N
1 N +1 22 −1
= (a0 − 1)2 − (aN − 1)2 =2−2 .
2−N
Put x = 2−N . Then x → 0 as N → ∞ and so
∞ −N
X
N
 22 − 1   2x − 1 
bn 2 = lim 2 − 2 −N = lim 2 − 2 = 2 − 2ln2.
n=1
N →∞ 2 x→0 x
Problem 3. It is enough to consider only polynomials with leading
coefficient 1. Let P (z) = (z − α1 )(z − α2 ) . . . (z − αn ) with |αj | = 1,
where the complex numbers α1 , α2 , . . . , αn may coincide.
We have

Pe(z) ≡ 2zP 0 (z) − nP (z) = (z + α1 )(z − α2 ) . . . (z − αn )+

+ (z − α1 )(z + α2 ) . . . (z − αn ) + . . . + (z − α1 )(z − α2 ) . . . (z + αn ).
n
Pe(z) X z + αk z + α |z|2 − |α|2
Hence, = . Since Re = for all complex
P (z) z − αk z−α |z − α|2
k=1
n
Pe(z) X |z|2 − 1
z, α, z 6= α, we deduce that in our case Re = . From
P (z) |z − αk |2
k=1
Pe(z)
|z| =
6 1 it follows that Re 6= 0. Hence Pe(z) = 0 implies |z| = 1.
P (z)
Problem 4. a) Let n be such that (1 − 2 )n ≤ . Then |x(1 − x2 )n < 
for every x ∈ [−1, 1]. Thus one can set λk = (−1)k+1 nk because then

n n  
k n
X X
2k+1
x− λk x = (−1) x2k+1 = x(1 − x2 )n .
k
k=1 k=0
Q
b) From the Weierstrass theorem there is a polynomial, say p ∈ m
such that

max |f (x) − p(x)| < .
x∈[−1,1] 2
2.2. Solutions of Olympic 1995 55

1
Set q(x) = {p(x) − p(−x)}. Then
2
1 1
f (x) − q(x) = {f (x) − p(x)} − {f (−x) − p(−x)}
2 2
and
1 1 
max |f (x) − q(x)| ≤ max |f (x) − p(x)| + max |f (−x) − p(−x)| < .
|x|≤1 2 |x|≤1 2 |x|≤1 2
(1)
Q
But q is an odd polynomial in m and it can be written as
m
X m
X
2k+1
q(x) = bk x = b0 x + bk x2k+1 .
k=0 k=1

If b0 = 0 then (1) proves b). If b0 6= 0 then one applies a) with
2|b0 |
of  to get
n
X 
max b0 x − b0 λk x2k+1 < (2)
|x|≤1 2
k=1
for appropriate n and λ1 , λ2 , . . . , λn . Now b) follows from (1) and (2)
with max{n, m} instead of n.
Problem 5. a) Let us consider the integral
Z2π
f (x)(1 ± cos x)dx = π(a0 ± 1).
0

The assumption that f (x) ≥ 0 implies a0 ≥ 1. Similarly, if f (x) ≤ 0

then a0 ≤ −1. In both cases we have a contradiction with the hypothesis
of the problem.
b) We shall prove that for each integer N and for each real number
h ≥ 24 and each real number y the function
N
3
X
FN (x) = cos(xn 2 )
n=1

changes sign in the interval (y, y + h). The assertion will follow immedi-
ately from here.
2.2. Solutions of Olympic 1995 56

Consider the integrals

Zy+h Zy+h
I1 = FN (x)dx, I2 = FN (x) cos xdx.
y y

If FN (x) does not change sign in (y, y + h) then we have

Zy+h Zy+h
|I2 | ≤ |FN (x)|dx = FN (x)dx = |I1 |.
y y

Hence,it is enough to prove that

|I2 | > |I1 |.

Obviously, for each α 6= 0 we have

Zy+h
2
cos(αx)dx ≤ .
|α|
y

Hence
y+h Z∞ 
N Z N
X 3
X 1  dt
|I1 | = cos(xn 2 dx ≤ 2 3 < 2 1 + 3 = 6. (1)
n=1 n=1 n 2 t 2
y 1

On the other hand we have

y+h
N Z
3
X
I2 = cos x cos(xn 2 )dx
n=1 y

Zy+h
1
= (1 + cos(2x))dx+
2
y

N Zy+h
1X 3 3
+ (cos(x(n 2 − 1)) + cos(x(n 2 + 1)))dx
2 n=2
y
1
= h + 4,
2
2.2. Solutions of Olympic 1995 57

where
N  N
1 X 1 1  1 X 1
|4| ≤ 1+2 3 + 3 ≤ +2 3 .
2 n=2 n 2 − 1 n 2 + 1 2 n=2 n 2 − 1

3 2 3
We use that n 2 − 1 ≥ n 2 for n ≥ 3 and we get
3
N Z ∞
1 2 X 1 1 2 dt
|4| ≤ + 3 +3 3 < + √ + 3 3 < 6.
2 22 − 1 n=3 n 2 2 2 2 − 1 t 2
2

Hence
1
|I2 | > h − 6. (2)
2
We use that h ≥ 24 and inequalities (1), (2) and we obtain |I2 | > |I1 |.
The proof is completed.
Problem 6. It is clear that one can add some functions, say {gm },
which satisfy the hypothesis of the problem and the closure of the finite
linear combinations of {fn } ∪ {gm } is L2 [0, 1]. Therefore without loss of
generality we assume that {fn } generates L2 [0, 1].
Let us suppose that there is a subsequence {nk } and a function f such
that
fnk (x) → f (x) for every x ∈ [0, 1].
k→∞
Fix m ∈ N. From Lebesgue’s theorem we have
Z1 Z1
0= fm (x)fnk (x)dx → fm (x)f (x)dx.
k→∞
0 0

R1
Hence fm (x)f (x)dx = 0 for every m ∈ N, which implies f (x) = 0
0
almost everywhere. Using once more Lebesgue’s theorem we get
Z1 Z1
1= fn2k (x)dx → f 2 (x)dx = 0.
k→∞
0 0

The contradiction proves the statement.

2.3. Solutions of Olympic 1996 58

2.3 Solutions of Olympic 1996

2.3.1 Day 1

Problem 1. Adding the first column of A to the last column we get

that
a0 a1 a2 . . . 1
 
 a1 a0 a1 . . . 1 
det(A) = (a0 + an )det  a
 2 a 1 a0 . . . 1 
.....................

an an−1 an−2 . . . 1
Subtracting the n-th row of the above matrix from the (n + 1)-st one,
(n − 1)-st from n-th,. . ., first from second we obtain that
a0 a1 a2 . . . 1
 
 d −d −d . . . 0
det(A) = (a0 + an )det   d d −d . . . 0

..................
d d d ... 0
Hence,
d −d −d . . . −d
 
d d −d . . . −d
det(A) = (−1)n (a0 + an )det 
d d d . . . −d

...................
d d d ... d
Adding the last row of the above matrix to the other rows we have
2d 0 0 . . . 0
 

n
2d 2d 0 . . . 0 n n−1 n
2d 2d 2d . . . 0 = (−1) (a0 +an )2 d .
det(A) = (−1) (a0 +an )det  
.................
d d d ... d
Problem 2. We have
Zπ
sin nx
In = dx
(1 + 2x ) sin x
−π
Zπ Z0
sin nx sin nx
= dx + dx.
(1 + 2x ) sin x (1 + 2x ) sin x
0 −π
2.3. Solutions of Olympic 1996 59

In the second integral we make the change of variable x = −x and obtain

Zπ Zπ
sin nx sin nx
In = dx + dx
(1 + 2x ) sin x (1 + 2−x ) sin x
0 0
Zπ
(1 + 2x ) sin nx
= dx
(1 + 2x ) sin x
0
Zπ
sin nx
= dx.
sin x
0

For n ≥ 2 we have
Zπ
sin nx − sin(n − 2)x
In − In−2 = dx
sin x
0
Zπ
=2 cos(n − 1)xdx = 0.
0

The answer 
0 if n is even,
In =
π if n is odd
follows from the above formula and I0 = 0, I1 = π.
Problem 3.
1
(i) Let B = (A + E). Then
2
1 1 1
B 2 = (A2 + 2AE + E) = (2AE + 2E) = (A + E) = B.
4 4 2
Hence B is a projection. Thus there exists a basis of eigenvectors for B,
and the matrix of B in this basis is of the form diag(1, . . . , 1, 0 . . . , 0).
Since A = 2B − E the eigenvalues of A are ±1 only.
(ii) Let {Ai : i ∈ I} be a set of commuting diagonalizable operators
on V , and let A1 be one of these operators. Choose an eigenvalue λ of
A1 and denote Vλ = {v ∈ V : A1 v = λv}. Then Vλ is a subspace of
V , and since A1 Ai = Ai A1 for each i ∈ I we obtain that Vλ is invariant
under each Ai . If Vλ = V then A1 is either E or −E, and we can start
2.3. Solutions of Olympic 1996 60

with another operator Ai . If Vλ 6= V we proceed by induction on dimV

in order to find a common eigenvector for all Ai . Therefore {Ai : i ∈ I}
are simultaneously diagonalizable.
If they are involutions then |I| ≤ 2n since the diagonal entries may
equal 1 or −1 only.
Problem 4.
(i) We show by induction that

an ≤ q n for n ≥ 3, (*)

where q = 0.7 and use that 0.7 < 2−1/2 . One has
1 1 11
a1 = 1, a2 = , a3 = , a4 = .
2 3 48
Therefore (*) is true for n = 3 and n = 4. Assume (*) is true for
n ≤ N − 1 for some N ≥ 5. Then
N −3
2 1 1 X
αN = aN −1 + aN −2 + ak aN −k
N N N
k=3

2 N −1 1 N −5 N
≤ q + q N −2 + q ≤ qN
N N N
2 1
because + 2 ≤ 5.
q q
ii) We show by induction that

an ≥ q n for n ≥ 2,
2 1  2 2
where q = . One has a2 = > = q 2 . Going by induction we
3 2 3
have for N ≥ 3.
N −2
2 1 X 2 N −2 N
aN = aN −1 + ak aN −k ≥ q N −1 + q = qN
N N N N
k=2

2
because = 3.
q
Problem 5. i) With a linear change of the variable (i) is equivalent to:
2.3. Solutions of Olympic 1996
2.3. Solutions of Olympic 1996 62

(i) is proved.
R1
ii) Denote In = n (f (x))n dx and M = max f (x).
0 x∈[0,1]
n
For M < 1 we have In ≤ nM → 0, a contradiction.
n→∞
If M > 1 since f is continuous there exists an interval I ⊂ [0, 1] with
|I| > 0 such that f (x) > 1 for every x ∈ I. Then In ≥ n|I| → +∞,
n→∞
0
a contradiction. Hence M = 1. Now we prove that f has a constant
sign. Assume the opposite. Then f 0 (x0 ) = 0 for some x ∈ (0, 1). Then
f (x0 ) = M = 1 because f 00 ≤ 0. For x0 + h in [0, 1], f (x0 + h) = 1 +
h2 00 00 h2
f (ξ), ξ ∈ (x0 , x0 + h). Let m = min f (x). So, f (x0 + h) ≥ 1 + m.
2 x∈[0,1] 2
2
δ
Let δ > 0 be such that 1 + m > 0 and x0 + δ < 1. Then
2
xZ0 +δ Zδ 
n m 2 n
In ≥ n (f (x)) dx ≥ n 1 + h dh → ∞
2 n→∞
x0 0

in view of (i’)-a contradiction. Hence f is monotone and M = f (0) or

M = f (1).
Let M = f (0) = 1. For h in [0, 1]
m
1 + hf 0 (0) ≥ f (h) ≥ 1 + hf 0 (0) + h2 ,
2
where f 0 (0) 6= 0, because otherwise we get a contradiction as above.
Since f (0) = M the function f is decreasing and hence f 0 (0) < 0. Let
m
0 < A < 1 be such that 1 + Af 0 (0) + A2 > 0. Then
2
ZA ZA ZA
m
n (1 + hf 0 (0))ndh ≥ n (f (x))n dx ≥ n (1 + hf 0 (0) + h2 )n dh.
2
0 0 0
1
From (i’) the first and the third integral tend to − as n → ∞,
f 0 (0)
hence so does the second.
R1
Also n (f (x))n dx ≤ n(f (A))n → 0 (f (A) < 1). We ger L =
A n→∞
1
− in this case.
f 0 (0)
2.3. Solutions of Olympic 1996 63

1
If M = f (1) we get in a similar way L = .
f 0 (1)
Problem 6.
Hint. If E = T ∪T 0 where T is the triangle with vertices (−2, 2), (2, 2)
and (0, 4), and T 0 is its reflexion about the x-axis, then C(E) = 8 >
K(E).
Remarks: All distances used in this problem are Euclidian. Di-
ameter of a set E is diam (E) = sup{dist (x, y) : x, y ∈ E}. Con-
traction of a set E to a set F is a mapping f : E 7→ F such that
dist (f (x), f (y)) ≤ dist (x, y) for all x, y ∈ E. A set E can be contracted
onto a set F if there is a contraction f of E to F which is onto, i.e., such
that f (E) = F . Triangle is defined as the union of the three segments
joining its vertices, i.e., it does not contain the interior.
Solution.
n
(a) The choice E1 = L gives C(L) ≤ lenght(L). If E ⊃ ∪ Ei then
i=1
Pn
diam (Ei ) ≥ lenght (L): By induction, n = l obvious, and assuming
i=1
that En+1 contains the end point a of L, define the segment L = {x ∈
L : dist (x, a) ≥ diam (En+1 ) + } and use induction assumption to get
n+1
P
diam (Ei ) ≥ lenght (L ) + diam (En+1 ) ≥ lenght (L) − ; but  > 0 is
i=1
arbitrary.
n n
(b) If f is a contraction of E onto L and E ⊂ ∪ Ei then L ⊂ ∪ f (Ei )
i=1 i=1
Pn Pn
and lenght (L) ≤ diam (f (Ei)) ≤ diam (Ei).
i=1 i=1
(c1) Let E = T ∪T 0 where T is the triangle with vertices (−2, 2), (2, 2)
n
and (0, 4), and T 0 is its reflexion about the x-axis. Suppose E ⊂ ∪ Ei . If
i=1
no set among Ei meets both T and T 0 , then Ei may be partitioned into
covers of segments [(−2, 2), (2, 2)] and [(−2, 2), (2, −2)], both of length
Pn
4, so diam (Ei ) ≥ 8. If at least one set among Ei , say Ek , meets
i=1
both T and T 0 , choose a ∈ Ek ∩ T and b ∈ Ek ∩ T 0 and note that the
sets Ei0 = Ei for i 6= k, Ek0 = Ek ∪ [a, b] cover T ∪ T 0 ∪ [a, b], which is
2.3. Solutions of Olympic 1996 64

a set of upper content at least 8, since its orthogonal projection onto

y-axis is a segment of length 8. Since diam (Ej ) = diam (Ej0 ), we get
Pn
diam (Ei ) ≥ 8.
i=1
(c2) Let f be a contraction of E onto L = [a0 , b0 ]. Choose a =
(al , a2 ), b = (b1 , b2 ) ∈ E such that f (a) = a0 and f (b) = b0 . Since
lenght (L) = dist (a0 , b0 ) ≤ dist (a, b) and since the triangles have di-
ameter only 4, we may assume that a ∈ T and b ∈ T 0 . Observe
that if a2 ≤ 3 then a lies on one of the segments joining some of
the points (−2, 2), (2, 2), (−1, 3), (1, 3); since all these points have dis-
√
tances from vertices, and so from points, of T2 at most 50, we get that
√
lenght (L) ≤ dist (a, b) ≤ 50. Similarly if b2 ≥ −3. Finally, if a2 > 3
and b2 < −3, we note that every vertex, and so every point of T is in the
√
distance at most 10 for a and every vertex, and so every point, of T 0 is
√
in the distance at most 10 of b. Since f is a contraction, the image of
√
T lies in a segment containing a0 of length at most 10 and the image
√
of T 0 lies in a segment containing b0 of length at most 10. Since the
√ √
union of these two images is L, we get lenght (L) ≤ 2 10 ≤ 50. Thus
√
K(E) ≤ 50 < 8.

2.3.2 Day 2

Problem 1. The ”only if” part is obvious. Now suppose that lim (xn+1 −
n→∞
xn ) = 0 and the sequence {xn } does not converge. Then there are two
cluster points K < L. There must be points from the interval (K, L)
in the sequence. There is an x ∈ (K, L) such that f (x) 6= x. Put
|f (x) − x|
 = > 0. Then from the continuity of the function f we
2
get that for some δ > 0 for all y ∈ (x − δ, x + δ) it is |f (y) − y| >
. On the other hand for n large enough it is |xn+1 − xn | < 2δ and
|f (xn ) − xn | = |xn+1 − xn | < . So the sequence cannot come into the
interval (x − δ, x + δ), but also cannot jump over this interval. Then all
cluster points have to be at most x − δ (a contradiction with L being a
2.3. Solutions of Olympic 1996 65

cluster point), or at least x + δ (a contradiction with K being a cluster

point).
Problem 2. First we show that

If cosh t is rational and m ∈ N, then cosh mt is rational. (1)

Since cosh 0.t = cosh 0 = 1 ∈ Q and cosh 1.t = cosh t ∈ Q, (1) follows
inductively from

cosh(m + 1)t = 2 cosh t.coshmt − cosh(m − 1)t.

The statement of the problem is obvious for k = 1, so we consider k ≥ 2.

For any m we have

cosh θ = cosh((m + 1)θ − mθ) =

= cosh(m + 1)θ. cosh mθ − sinh(m + 1)θ. sinh mθ (2)
q p
= cosh(m + 1)θ. cosh mθ − cosh2 (m + 1)θ − 1 cosh2 mθ − 1

Set cosh kθ = a, cosh(k + 1)θ = b, a, b ∈ Q. Then (2) with m = k gives

p p
cosh θ = ab − a − 1 b2 − 1
2

and then
(a2 − 1)(b2 − 1) = (ab − cosh θ)2
(3)
= a2 b2 − 2ab cosh θ + cosh2 θ.
Set cosh(k 2 − 1)θ = A, coshk 2 θ = B. From (1) with m = k − 1 and
t = (k + 1)θ we have A ∈ Q. From (1) with m = k and t = kθ we have
B ∈ Q. Moreover k 2 − 1 > k implies A > a and B > b. Thus AB > ab.
From (2) with m = k 2 − 1 we have
(A2 − 1)(B 2 − 1) = (AB − cosh θ)2
(4)
= A2 B 2 − 2AB cosh θ + cosh2 θ.
So after we cancel the cosh2 θ from (3) and (4) we have a non-trivial
linear equation in cosh θ with rational coefficients.
Problem 3. (a) All of the matrices in G are of the form
∗ ∗
h i
0 ∗ .
2.3. Solutions of Olympic 1996 66

So all of the matrices in H are of the form

1 x
h i
M (x) = 0 1 ,

so they commute. Since M (x)−l = M (−x), H is a subgroup of G.

(b) A generator of H can only be of the form M (x), where x is a
p
binary rational, i.e., x = n with integer p and non-negative integer n.
2
In H it holds

M (x)M (y) = M (x + y)
M (x)M (y)−1 = M (x − y).

The matrices of the form M ( 21n ) are in H for all n ∈ N. With only finite
number of generators all of them cannot be achieved.
Problem 4. Assume the contrary - there is an arcA ⊂ C with length
π
l(A) = such that A ⊂ B\Γ. Without loss of generality we may assume
2
1 1 1 1
that the ends of A are M = ( √ , √ ), N = ( √ , − √ ). A is compact
2 2 2 2
and Γ is closed. From A ∩ Γ = ∅ we get δ > 0 such that dist(x, y) > δ
for every x ∈ A, y ∈ Γ.
x2
Given  > 0 with E we denote the ellipse with boundary: +
(1 + )2
y2
= 1, such that M, N ∈ E . Since M ∈ E we get
b2
(1 + )2
b2 = .
2(1 + )2 − 1
Then we have
(1 + )2
areaE = π p > π = areaD.
2(1 + )2 − 1
In view of the hypotheses, E \ =
6 ∅ for every  > 0. Let S = {(x, y) ∈
R2 : |x| > |y|}. From E \S ⊂ D ⊂ B it follows that E \B ⊂ S. Taking
 < δ we get that

∅=
6 E \B ⊂ E ∩ S ⊂ D1+ ∪ S ⊂ B
2.3. Solutions of Olympic 1996 67

- a contradiction (we use the notation Dt = {(x, y) ∈ R2 : x2 + y 2 ≤ t2 }).

Remark. The ellipse with maximal area is well known as John’s ellipse.
Any coincidence with the President of the Jury is accidental.
Problem 5.
t 1
(1) Set f (t) = , h = √ . Then
(1 + t2 )2 x
∞ ∞ Z ∞
X nx X 1
= h f (nh) → f (t)dt = .
n=1
(n2 + x)2 n=1
h→0 2
0
∞
P
The convergence holds since h f (nh) is a Riemann sum of the integral
n=1
R∞
f (t)dt. There are no problems with the infinite domain because f
0
∞
P R∞
is integrable and f ↓ 0 for x → ∞ (thus h f (nh) ≥ f (t)dt ≥
n=N nN
∞
P
h f (nh)).
n=N +1
(ii) We have
nh+ h2 h
∞
X nx 1
∞ 
X Z  Z2
− = hf (nh) − f (t)dt − f (t)dt
i=1
(n2 + x)2 2 n=1
nh− h2 0
(1)
nh+ h2 h
2
∞
X Z Z
≤ hf (nh) − f (t)dt + f (t)dt
n=1 0
nh− h2

Using twice integration by parts one has

Za+b Zb
1
2bg(a) − g(t)dt = − (b − t)2 (g 00 (a + t) + g 00 (a − t))dt (2)
2
a−b 0

for every g ∈ C 2 [a − b, a + b]. Using f (0) = 0, f ∈ C 2 [0, h/2] one gets

Zh/2
f (t)dt = O(h2 ). (3)
0
2.3. Solutions of Olympic 1996 68

From (1), (2) and (3) we get

nh+ 21
∞ ∞ Z
X nx 1 X
2 2
− ≤ h 2
|f 00 (t)|dt + O(h2 ) =
i=1
(n + x) 2 n=1
nh− 12
Z∞
= h2 |f 00 (t)|dt + O(h2 ) = O(h2 ) = O(x−1 ).
1
2

Problem 6.
(i) Put for n ∈ N
(n + 1)n
cn = (2.1)
nn−1
Observe that c1 c2 . . . cn = (n + 1)n . Hence, for n ∈ N,
1/n (a1 c1 a2 c2 . . . an cn )1/n
(a1 a2 . . . an ) =
(n + 1)
(a1 c1 + · · · + an cn )
≤ .
n(n + 1)
Consequently,
X∞ ∞
X ∞
X 
1/n −1
(a1 a2 . . . an ) ≤ an cn (m(m + 1)) . (2)
n=1 n=1 m=n
Since ∞ ∞ 
X
−1
X 1 1  1
(m(m + 1)) = − =
m=n m=n
m m+1 n
we have
∞ ∞ ∞
X X
−1
 X an cn
an c n (m(m + 1)) =
n=1 m=n n=1
n
∞ ∞
X (n + 1) n X
= an ( ) <e an
n=1
n n=1

(by (1)). Combining the last inequality with (2) we get the result.
(ii) Set an = nn−1 (n + 1)−n for n = 1, 2, . . . , N and an = 2−n for
n > N , where N will be chosen later. Then
1
(a1 . . . an )1/n = (3)
n+1
2.4. Solutions of Olympic 1997 69

for n ≤ N . Let K = K() be such that

 n + 1 n 
>  − for n > K. (4)
n 2
Choose N from the condition
K ∞ N
X X
−n  X 1
an + 2 ≤ , (5)
n=1 n=1
(2e − )(e − ) n
n=K+1

which is always possible because the harmonic series diverges. Using (3),
(4) and (5) we have
∞ K ∞ N
X X X
−n
X 1  n n
an = an + 2 + <
n=1 n=1
n n+1
n=N +1 n=K+1
N N
 X 1   −1 X 1
< + e− =
(2e − )(e − ) n 2 n
n=K+1 n=K+1
N ∞
1 X 1 1 X
= ≤ (a1 . . . an )1/n .
e− n e −  n=1
n=K+1

2.4 Solutions of Olympic 1997

2.4.1 Day 1

Problem 1.
It is well known that
Z1 n
1 X k 
−1 = ln xdx = lim ln
n→∞ n n
0 k=1

(Riemman’s sums). Then

n n
1 X k  1 X k 
ln + n ≥ ln → −1.
n n n n n→∞
k=1 k=1

Given  > 0 there exist n0 such that 0 < n ≤  for all n ≥ n0 . Then
n n
1 X k  1X k 
ln + n ≤ ln + .
n n n n
k=1 k=1
2.4. Solutions of Olympic 1997 70

Since
n
1 X k  Z1
lim ln +  = ln (x + )dx
n→∞ n n
k=1 0
Z1+
= ln xdx


we obtain the result when  goes to 0 and so

n
1 X k
lim ln + n ) = −1.
n→∞ n n
k=1

Problem 2.
∞
P n
P
a) Yes. Let S = an , Sn = ak . Fix  > 0 and a number no
n=1 k=1
such that |Sn − S| <  for n > n0 . The partial sums of the permuted
series have the form L2n−1 +k = S2n−1 + S2n − S2n −k , 0 ≤ k < 2n−1 and for
2n−1 > n0 we have |L2n−1 +k − S| < 3, i.e. the permuted series converges.
(−1)n+1 2n−1
P−1 1
b) No. Take an = √ . Then L3.2n−2 = S2n−1 + √
n k=2n−2 2k + 1
1
and L3.2n−2 − S2n−1 ≥ 2n−2 √ → ∞, so L3.2n−2 → ∞.
2n n→∞ n→∞
Problem 3. √
1 3
Set S = A + ωB, where ω = − + i . We have
2 2
SS = (A + ωB)(A + ωB) = A2 + ωBA + ωAB + B 2
= AB + ωBA + ωAB = ω(BA − AB),

because ω + 1 = −ω. Since det(SS) = detS.detS is a real number and

detω(BA − AB) = ω n and det(BA − AB) 6= 0, then ω n is a real number.
This is possible only when n is divisible by 3.
Problem 4.
a) We construct inductively the sequence {ni } and the ratios
α
θk = Q
k 1
1 (1 + )
ni
2.4. Solutions of Olympic 1997 71

so that
θk > 1 for all k.
Choose nk to be the least n for which
1
n + < θk−1
n
(θ0 = α) so that for each k,
1 1
1+ < θk−1 ≤ 1 + . (1)
nk nk − 1
Since
1
θk−1 ≤ 1 +
nk − 1
we have
1
1+
1 θk−1 nk − 1 1
1+ < θk = ≤ =1+ 2 .
nk+1 1 1 nk − 1
1+ 1+
nk nk
Hence, for each k, nk+1 ≥ n2k .
Since n1 ≥ 2, nk → ∞ so that θk → 1. Hence
∞ 
Y 1
α= 1+ .
1
n k

The uniquness of the infinite product will follow from the fact that on
every step nk has to be determine by (1).
Indeed, if for some k we have
1
1+ ≥ θk−1
nk
then θk ≤ 1, θk+1 < 1 and hence {θk } does not converge to 1.
Now observe that for M > 1,
 1  1  1  1 1 1 1
1+ 1+ 2 1+ 4 · · · = 1+ + 2 + 3 +· · · = 1+ .
M M M M M M M −1
(2)
Assume that for some k we have
1
1+ < θk−1 .
nk − 1
2.4. Solutions of Olympic 1997 72

Then we get
α θk−1
1 1 = 1 1
(1 + n1 )(1 + n2 ) · · · (1 + nk )(1 + nk+1 ) · · ·
θk−1 θk−1
≥ = >1
(1 + n1k )(1 + 1
n2k
)··· 1 + nk1−1

- a contradiction,
b) From (2) α is rational if its product ends in the stated way.
p
Conversely, suppose α is the rational number , Our aim is to show
q
that for some m,
nm
θm−1 = .
nm − 1
Suppose this is not the case, so that for every m,
nm
θm−1 < . (3)
nm − 1
For each k we write
pk
θk =
qk
as a fraction (not necessarily in lowest terms) where

p0 = p, q0 = q

and in general
pk = pk−1 nk , qk = qk−1 (Nk + 1).
The numbers pk − qk are positive integers: to obtain a contradiction it
suffices to show that this sequence is strictly decreasing. Now,

pk − qk − (pk−1 − qk−1 ) = nk pk−1 − (nk + 1)qk−1 − pk−1 + qk−1

= (nk − 1)pk−1 − nk qk−1

and this is negative because

pk−1 nk
= θk−1 <
qk−1 nk − 1
by inequality (3).
2.4. Solutions of Olympic 1997 73

Problem 5.
a) For x = a the statement is trivial. Let x 6= 0. Then maxxi > 0
i
and minxi < 0. Hence k x k∞ < 1. From the hypothesis on x it follows
i
that:
i) If xj ≤ 0 then maxxi ≤ xj + 1.
i
ii) If xj ≥ 0 then minxi ≥ xj − 1.
i
Consider y ∈ Z0n , y 6= 0. We split the indices {1, 2, . . . , n} into five
sets:

I(0) = {i : yi = 0},
I(+, +) = {i : yi > 0, xi ≥ 0}, I(+, −) = {i : yi > 0, xi < 0},
I(−, +) = {i : yi < 0, xi > 0}, I(−, −) = {i : yi < 0, xi ≤ 0}

As least one of the last four index sets is not empty. If I(+, +) 6= ∅ or
I(−, −) 6= ∅ then k x + y k∞ ≥ 1 >k x k∞ . If I(+, +) = I(−, −) = ∅
P
then yi = 0 implies I(+, −) 6= ∅ and I(−, +) 6= ∅. Therefore i) and
ii) give k x + y k∞ ≥k x k∞ which completes the case p = ∞.
Now let 1 ≤ p < ∞. Then using i) for every j ∈ I(+, −) we get
|xj + yj | = yj − 1 + xj + 1 ≥ |yj | − 1 + maxxi . Hence
i

|xj + yj |p ≥ |yj | − 1 + |xk |p for every k ∈ I(−, +) and j ∈ I(+, −).

Similarly

|xj + yj |p ≥ |yj | − 1 + |xk |p for every k ∈ I(+, −) and j ∈ I(−, +);

|xj + yj |p ≥ |yj | + |xj |p for every j ∈ I(+, +) ∪ I(−, −).
2.4. Solutions of Olympic 1997 74
P P
Assume that 1≥ 1. Then
j∈(+,−) j∈I(−,+)

k x + y kpp − k x kpp
X  X X 
p p
= (|xj + yj | − |xj | ) + |xj + yj |p − p
|xk |
j∈I(+,+)∪I(−,−) j∈I(+,−) k∈I(−,+)
 X X 
+ |xj + yj |p − |xk |p
j∈I(−,+) k∈I(+,−)
X X
≥ |yj | + (|yj | − 1)
j∈I(+,+)∪I(−,−) j∈I(+,−)
 X X X 
+ (|yj | − 1) − 1+ 1
j∈I(−,+) j∈I(+,−) j∈I(−,+)
n
X X X X
= |yi | − 2 1=2 (yj − 1) + 2 yj ≥ 0.
i=1 j∈I(+,−) j∈I(+,−) j∈I(+,+)
P P
The case 1≤ 1 is similar. This proves the statement.
j∈I(+,−) j∈I(−,+)
b) Fix p ∈ (0, 1) and a rational t ∈ ( 12 , 1). Choose a pair of positive
integers m and l such that mt = l(1 − t) and set n = m + l. Let

xi = t, i = 1, 2, . . . , m; xi = t − 1, i = m + 1, m + 2, . . . , n;
yi = −1, i = 1, 2, . . . , m; ym+1 = m; yi = 0, i = m + 2, . . . , n.

Then x ∈ R0n , maxxi − minxi = 1, y ∈ Z0n and

i i

k x kpp − k x + y kpp = m(tp − (1 − t)p ) + (1 − t)p − (m − 1 + t)p ,

which is possitive for m big enough.

Problem 6.
a) No.
Consider F = {A1 , B1 , . . . , An , Bn , . . .}, where An = {1, 3, 5, . . . , 2n −
1, 2n}, Bn = {2, 4, 6, . . . , 2n, 2n + 1}.
b) Yes.
We will prove inductively a stronger statement:
Suppose F, G are two families of finite subsets of N such that:
2.4. Solutions of Olympic 1997 75

1) For every A ∈ F and B ∈ G we have A ∩ B 6= ∅;

2) All the elements of F have the same size r, and elements of G- size
s. (we shall write #(F ) = r, #(G) = s).
Then there is a finite set Y such that A ∪ B ∪ Y 6= ∅ for every A ∈ F
and ∈ G.
The problem b) follows if we take F = G.
Proof of the statement: The statement is obvious for r = s = 1.
Fix the numbers r, s and suppose the statement is proved for all pairs
F 0 , G0 with #(F 0 ) < r, #(G0 ) < s. Fix A0 ∈ F, B0 ∈ G. For any subset
C ⊂ A0 ∪ B0 , denote

F (C) = {A ∈ F : A ∩ (A0 ∪ B0 ) = C}.

Then F = ∪ F (C). It is enough to prove that for any pair of

∅6=C⊂A0 ∪B0
non-empty sets C, D ⊂ A0 ∪ B0 the families F (C) and G(D) satisfy the
statement.
Indeed, if we denote by YC,D the corresponding finite set, then the
finite set ∪ YC,D will satisfy the statement for F and G. The
C,D⊂A0 ∪B0
proof for F (C) and G(D).
If C ∩ D 6= ∅, it is trivial.
If C ∩ D = ∅, then any two sets A ∈ F (C), B ∈ G(D) must meet
outside A0 ∪ B0 . Then if we denote Fe(C) = {A\C : A ∈ F (C)}, G(D)
e =
{B\D : B ∈ G(D)}, then Fe(C) and G(D) e satisfy the conditions 1) and
2) above, with #(Fe(C)) = #(F )−#C < r, #(G(D))
e = #(G)−#D < s,
and the inductive assumption works.

2.4.2 Day 2

Problem 1.
1
Let c = f 00 (0). We have
2
(f 0 )2 − 2f f 00
g= √ ,
2(f 0 )2 f
2.4. Solutions of Olympic 1997 76

where

f (x) = cx2 + O(x3 ), f 0 (x) = 2cx + O(x2 ), f 00 (x) = 2c + O(x).

Therefore (f 0 (x))2 = 4c2 x2 + O(x3 ),

2f (x)f 00 (x) = 4c2 x2 + O(x3 )

and
p p
2(f 0 (x))2 f (x) = 2(4c2 x2 + O(x3 ))|x| c + O(x).
g is bounded because
p
2(f 0 (x))2 f (x)
3
→ 8c5/2 6= 0
|x| x→0

and f 0 (x)2 − 2f (x)f ”(x) = O(x3 ).

The theorem does not hold for some C 2 -functions.
p
Let f (x) = (x + |x|3/2 )2 = x2 + 2x2 |x| + |x|3 , so f is C 2 . For x > 0,
1 1 0 1 1 3 1
g(x) = 3√ =− · 3√ 2 · · √ → −∞.
2 1+ 2 x 2 (1 + 2 x) 4 x x→0
Problem 2.
Let I denote the identity n × n matrix. Then
A B I F A 0
h i h i h i
detM.detH = det C D .det 0 H = det C I = detA.

Problem 3.
sin(log t)
Set f (t) = . We have
tα
−α cos(log t)
f 0 (x) = sin(log t) + .
tα+1 tα+1
1+α
So |f 0 (t)| ≤ for α > 0. Then from Mean value theorem for some
tα+1
1+α P1+α
θ ∈ (0, 1) we get |f (n+1)−f (n)| = |f 0 (n+θ)| ≤ α+1 . Since α+1
<
∞
n ∞
n
(−1)n−1 f (n) =
P P
+∞ for α > 0 and f (n) → 0 we get that (f (2n −
n→∞ n=1 n=1
1) − f (2n)) converges.
2.4. Solutions of Olympic 1997 77

sin(log n)
Now we have to prove that does not converge to 0 for α ≤ 0.
nα
It suffices to consider α = 0.
We show that an = sin(log n) does not tend to zero. Assume the
h 1 1i
contrary. There exist kn ∈ N and λn ∈ − , for n > e2 such that
2 2
log n
= kn + λn . Then |an | = sin π|λn |. Since an → 0 we get λn → 0.
π
We have

kn+1 − kn =
log (n + 1) − log n 1  1
= − (λn+1 − λn ) = log 1 + − (λn+1 − λn ).
π π n
Then |kn+1 − kn | < 1 for all n big enough. Hence there exists no so that
log n
kn = kn0 for n > n0 . So = kn0 + λn for n > n0 . Since λn → 0 we
π
get contradiction with log n → ∞.
Problem 4.
a) If we denote by Eij the standard basis of Mn consisting of elemen-
tary matrix (with entry 1 at the place (i, j) and zero elsewhere), then
the entries cij of C can be defined by cij = f (Eji ).
b) Denote by L the n2 −1-dimensional linear subspace of Mn consisting
of allmatrices with zero trace. The elements Eij with i 6= j and the
elements Eii − Enn , i = 1, . . . , n − 1 form a linear basis for L. Since

Eij = Eij .Ejj − Ejj Eij , i 6= j

Eii − Enn = Ein Eni − Eni Ein , i = 1, . . . , n − 1,

then the property (2) shows that f is vanishing identically on L. Now,

1
for any A ∈ Mn we have A − tr(A).E ∈ L, where E is the identity
n
1
matrix, and therefore f (A) = f (E).tr(A).
n
Problem 5.
Let f n = f ◦ f ◦ · · · ◦ f , f 0 = id, f −n = (f −1 )n for every natural num-
| {z }
n times
ber n. Let T (x) = {f n (x) : n ∈ Z} for every x ∈ X. The sets T (x) for
different x0 s either coinside or do not intersect. Each of them is mapped
2.4. Solutions of Olympic 1997 78

by f onto itself. It is enough to prove the theorem for every such set. Let
A = T (x). If A is finite, then we can think that A is the set of all vertices
2π
of a regular n polygon and that f is rotation by . Such rotation can
n
be obtained as a composition of 2 symmetries mapping the n polygon
π
onto itself (if n is even then there are axes of symmetry making angle;
n
2π
if n = 2k + 1 then there are axes making k angle). If A is infinite then
n
we can think that A = Z and f (m) = m + 1 for every m ∈ Z. In this
1
case we define g1 as a symmetry relative to , g2 as a symmetry relative
2
to 0.
Problem 6.
1
a) f (x) = x sin .
x
b) Yes. The Cantor set is given by
∞
X
C = {x ∈ [0, 1) : x = bj 3−j , bj ∈ {0, 2}}.
j=1

There is an one-to-one mapping f : [0, 1) → C. Indeed, for x =

∞ ∞
−j
(2aj )3−j . Hence C is uncount-
P P
aj 2 , aj ∈ {0, 1} we set f (x) =
j=1 j=1
able.
For k = 1, 2, . . . and i = 0, 1, 2, . . . , 2k−1 − 1 we set
 Xk−2   Xk−2 
−k j −k
ak,i =3 6 aj 3 + 1 , bk,i = 3 6 aj 3 j + 2 ,
j=0 j=0

k−2
aj 2j , aj ∈ {0, 1}. Then
P
where i =
j=0
k−1
[0, 1]\C = ∪∞ 2
k=1 ∪i=0
−1
(ak,i , bk,i ),

i.e. the Cantor set consists of all points which have a trinary representa-
tion with 0 and 2 as digits and the points of its compliment have some
2k−1 −1
0
1 s in their trinary representation. Thus, ∪ (ak,i , bk,i ) are all points
i=0
(exept ak,i ) which have 1 on k-th place and 0 or 2 on the j-th (j < k)
places.
2.5. Solutions of Olympic 1998 79

Noticing that the points with at least one digit equals to 1 are every-
where dence in [0, 1] we set
∞
X
f (x) = (−1)k gk (x).
k=1

where gk is a piece-wise linear continuous functions with values at the

knots a + b 
k,i k,i
gk = 2−k ,
2
gk (0) = gk (1) = gk (ak,i ) = gk (bk,i ) = 0, i = 0, 1, . . . , 2k−1 − 1.
Then f is continuous and f ”crosses the axis” at every point of the
Cantor set.

2.5 Solutions of Olympic 1998

2.5.1 Day 1

Problem 1. First choose a basis {v1 , v2 , v3 } of U1 . It is possible to

extend this basis with vectors v4 , v5 and v6 to get a basis of U2 . In the
same way we can extend a basis of U2 with vectors v7 , . . . , v10 to get as
basis of V .
Let T ∈  be an endomorphism which has U1 and U2 as invariant
subspaces. Then its matrix, relative to the basis {v1 , . . . , v10 } is of the
form
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
 
 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
 ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗
0 0 0 ∗ ∗ ∗ ∗ ∗ ∗ ∗
 
0 0 0 ∗ ∗ ∗ ∗ ∗ ∗ ∗
0 0 0 ∗ ∗ ∗ ∗ ∗ ∗ ∗
 
 0 0 0 0 0 0 ∗ ∗ ∗ ∗
 
 0 0 0 0 0 0 ∗ ∗ ∗ ∗
0 0 0 0 0 0 ∗ ∗ ∗ ∗
 
0 0 0 0 0 0 ∗ ∗ ∗ ∗
So dimR  = 9 + 18 + 40 = 67.
Problem 2.
Let Sn be the group of permutations of {1, 2, . . . , n}.
2.5. Solutions of Olympic 1998 80

1) When n = 3 the proposition is obvious: if x = (12) we choose

y = (123); if x = (123) we choose y = (12).
2) n = 4. Let x = (12)(34). Assume that there exists y ∈ Sn , such
that S4 =< x, y >. Denote by K the invariant subgroup

K = {id, (12)(34), (13)(24), (14)(23)}.

By the fact that x and y generate the whole group S4 , it follows that
the factor group S4 /K contains only powers of y = yK, i.e., S4 /K is
cyclic. It is easy to see that this factor-group is not comutative (some-
thing more this group is not isomorphic to S3 ).
3) n = 5
a) If x = (12), then for y we can take y = (12345).
b) If x = (123), we set y = (124)(35). Then y 3 xy 3 = (125) and
y 4 = (124). Therefore (123), (124), (125) ∈< x, y >- the subgroup gen-
erated by x and y. From the fact that (123), (124), (125) generate the
alternating subgroup A5 , it follows that A5 ⊂< x, y >. Moreover y is an
odd permutation, hence < x, y >= S5 .
c) If x = (123)(45), then as in b) we see that for y we can take the
element (124).
d) If x = (1234), we set y = (12345). Then (yx)3 = (24) ∈< x, y >
, x2 (24) = (13) ∈< x, y > and y 2 = (13524) ∈< x, y >. By the fact
(13) ∈< x, y > and (13524) ∈< x, y >, it follows that < x, y >= S5 .
e) If x = (12)(34), then for y we can take y = (1354). Then y 2 x =
(125), y 3 x = (124)(53) and by c) S5 =< x, y >.
f) If x = (12345), then it is clear that for y we can take the element
y = (12).
Problem 3. a) Fix x = x0 ∈ (0, 1). If we denote xn = fn (x0 ), n =
1, 2, . . . it is easy to see that x1 ∈ (0, 1/2], x1 ≤ f (x1 ) ≤ 1/2 and xn ≤
f (xn ) ≤ 1/2 (by induction). Then (xn )n is a bounded nondecreasing
sequence and, since xn+1 = 2xn (1 − xn ), the limit l = limn→∞ xn satisfies
l = 2l(1 − l), which implies l = 1/2. Now the monotone convergence
2.5. Solutions of Olympic 1998 81

theorem implies that

Z1
1
lim fn (x)dx = .
n→∞ 2
0
b) We prove by induction that
1 2n −1
 1 2n
fn (x) = − 2 1− (1)
2 2
holds for n = 1, 2, . . .. For n = 1 this is true, since f (x) = 2x(1 − x) =
1 1
− 2(x − )2 . If (1) holds for some n = k, then we have
2 2
1 k 1 1 1 k
fk+1 (x) = fk (f (x)) = − 22 −1 ( − 2(x − )2 ) − )2
2 2 2 2
1 k 1 k
= − 22 −1 (−2(x − )2 )2
2 2
1 k+1 1 k+1
= − 22 −1 (x − )2
2 2
which is (2) for n = k + 1.
Using (1) we can compute the integral,
Z1 n
22 −1 
h1 1 2n +1 i1 1 1
fn (x)dx = x − n x− = − .
2 2 +1 2 x=0 2 2(2n + 1)
0

Problem 4. Define the function

1
g(x) = f 2 (x) + f 0 (x).
2
Because g(0) = 0 and

f (x).f 0 (x) + f 00 (x) = g 0 (x),

it is enough to prove that there exists a real number 0 < η ≤ 1 for which
g(η) = 0.
a) If f is never zero, let
x 1
h(x) = − .
2 f (x)
2.5. Solutions of Olympic 1998 82

1
Because h(0) = h(1) = − , there exists a real number 0 < η < 1 for
2
which h0 (η) = 0. But g = f 2 .h0 , and we are done.
b) If f has at least one zero, let z1 be the first one and z2 be the last
one. (The set of the zeros is closed.) By the conditions, 0 < z1 ≤ z2 < 1.
The function f is positive on the intervals [0, z1 ) and (z2 , 1]; this
implies that f 0 (z1 ) ≤ 0 and f 0 (z2 ) ≥ 0. Then g(z1 ) = f 0 (z1 ) ≤ 0 and
g(z2 ) = f 0 (z2 ) ≥ 0, and there exists a real number η ∈ [z1 , z2 ] for which
g(η) = 0.
2
Remark. For the function f (x) = the conditions hold and
x+1
f.f 0 + f ” is constantly 0.
Problem 5. Observe that both sides of (2) are identically equal to zero
if n = 1. Suppose that n > 1. Let x1 , . . . , xn be the zeros of P . Clearly
(2) is true when x = xi , i ∈ {1, . . . , n}, and equality is possible only if
P 0 (xi ) = 0, i.e., if xi is a multiple zero of P . Now suppose that x is not
a zero of P . Using the identities
n
P 0 (x) X 1 P 00 (x) X 2
= , = ,
P (x) i=1
x − xi P (x) 1≤i<j≤n
(x − x i )(x − x j )

we find
 P 0 (x) 2 n
P 00 (x) X n − 1 X 2
(n − 1) −n = 2
− .
P (x) P (x) i=1
(x − xi ) 1≤i<j≤n
(x − x i )(x − x j )

But this last expression is simply

X  1 1 2
− ,
1≤i<j≤n
x − xi x − xj

and therefore is positive. The inequality is proved. In order that (2)

holds with equality sign for every real x it is necessary that x1 = x2 =
. . . = xn . A direct verification shows that indeed, if P (x) = c(x − x1 )n ,
then (2) becomes an identity.
2.5. Solutions of Olympic 1998 83

Problem 6. Observe that the integral is equal to

π
Z2
f (sin θ) cos θdθ
0

and to π
Z2
f (cos θ) sin θdθ
0
So, twice the integral is at most
π
Z2
π
1dθ = .
2
0
√
Now let f (x) = 1 − x2 . If x = sin θ and y = sin φ then

xf (y) + yf (x) = sin θ cos φ + sin φ cos θ = sin(θ + φ) ≤ 1.

2.5.2 Day 2

Problem 1. We use induction on k. By passing to a subset, we may

assume that f1 , . . . , fk are linearly independent.
Since fk is independent of f1 , . . . , fk−1 , by induction there exists a
vector ak ∈ V such that f1 (ak ) = · · · = fk−l (ak ) = 0 and fk (ak ) =6=
0. After normalising, we may assume that fk (ak ) = 1. The vectors
a1 , . . . , ak−1 are defined similarly to get

1 if i = j
fi (aj ) =
0 if i 6= j.
For an arbitrary x ∈ V and 1 ≤ i ≤ k, fi (x − f1 (x)a1 − f2 (x)a2 − · · · −
k
P
fk (x)ak ) = fi (x) − fj (x)fi (aj ) = fi (x) − fi (x)fi (ai ) = 0, thus f (x −
j=1
f1 (x)a1 − · · · − fk (x)ak ) = 0. By the linearity of f this implies f (x) =
f1 (x)f (a1 ) + · · · + fk (x)f (ak ), which gives f (x) as a linear combination
of f1 (x), . . . , fk (x).
2.5. Solutions of Olympic 1998 84

1 1
Problem 2. Denote x0 = 1, x1 = − , x2 = , x3 = 1,
2 2
3
Y
ω(x) = (x − xi ),
i=0
ω(x)
ωk (x) = , k = 0, . . . , 3,
x − xk
ωk (x)
lk (x) = .
ωk (xk )
Then for every f ∈ P
3
X
00
f (x) = lk00 (x)f (xk ),
k=0
3
X
00
|f (x)| ≤ |lk00 (x)|.
k=0

Since f ” is a linear function max−1≤x≤1 |f 00 (x)| is attained either at x =

−1 or at x = 1. Without loss of generality let the maximum point is
x = 1. Then
X 3
00
sup max |f (x)| = |lk00 (1)|.
f ∈P −1≤x≤1
k=0
In order to have equality for the extremal polynomial f∗ there must hold

f∗ (xk ) = signlk00 (1), k = 0, 1, 2, 3.

It is easy to see that {lk00 (1)}3k=0 alternate in sign, so f∗ (xk ) = (−1)k−1 , k =

0, . . . , 3. Hence f∗ (xk ) = T3 (x) = 4x3 − 3x, the Chebyshev polynomial
of the first kind, and f∗00 (1) = 24. The other extremal polynomial, corre-
sponding to x = −1, is −T3 .
Problem 3. Let fn (x) = f (f (. . . f ( p))). It is easy to see that fn (x) is
| {z }
n
a picewise monotone function and its graph contains 2n linear segments;
one endpoint is always on {(x, y) : 0 ≤ x ≤ 1, y = 0}, the other is
on {(x, y) : 0 ≤ x ≤ 1, y = 1}. Thus the graph of the identity function
2.5. Solutions of Olympic 1998 85

intersects each segment once, so the number of points for which fn (x) = x
is 2n .
Since for each n-periodic points we have fn (x) = x, the number of
n-periodic points is finite.
A point x is n-periodic if fn (x) = x but fk (x) 6= x for k = 1, . . . , n−1.
But as we saw before fk (x) = x holds only at 2k points, so there are at
most 21 + 22 + · · · + 2n−1 = 2n − 2 points x for which fk (x) = x for at
least one k ∈ {1, 2, . . . , n − 1}. Therefore at least two of the 2n points
for which fn (x) = x are n-periodic points.
Problem 4. It is clear that id : An → An given by id(x) = x, does not
verify condition (2). Since id is the only increasing injection on An , F
does not contain injections. Let us take any f ∈ F and suppose that
#(f −1 (k)) ≥ 2. Since f is increasing, there exists i ∈ An such that
f (i) = f (i + 1) = k. In view of (2), f (k) = f (f (i + 1)) = f (i) = k. If
{i < k : f (i) < k} = ∅, then taking j = max{i < k : f (i) < k} we get
f (j) < f (j + 1) = k = f (f (j + 1)), a contradiction. Hence f (i) = k for
i ≤ k. If #(f −1 ({l})) ≥ 2 for some l ≥ k, then the similar consideration
shows that f (i) = l = k for i ≤ k. Hence #(f −1 {i}) = 0 or 1 for every
i > k. Therefore f (i) ≤ i for i > k. If f (l) = l, then taking j = max{i <
l : f (i) < l} we get f (j) < f (j + 1) = l = f (f (j + 1)), a contradiction.
Thus, f (i) ≤ i − 1 for i > k. Let m = max{i : f (i) = k}. Since f is non-
constant m ≤ n−1. Since k = f (m) = f (f (m+1)), f (m+1) ∈ [k+1, m].
If f (l) > l − 1 for some l > m + 1, then l − 1 and f (l) belong to f −1 (f (l))
and this contradicts the facts above. Hence f (i) = i − 1 for i > m + 1.
Thus we show that every function f in F is defined by natural numbers
k, l, m, where 1 ≤ k < l = f (m + 1) ≤ m ≤ n − 1.

k if i ≤ m
f (i) = l if i = m
i − 1 if i > m + 1

Then  
n
#(F) = .
3
2.5. Solutions of Olympic 1998 86

Problem 5. For every x ∈ M choose spheres S, T ∈ S such that S 6= T

and x ∈ S ∩ T ; denote by U, V, W the three components of Rn \(S ∪ T ),
where the notation is such that ∂U = S, ∂V = T and x is the only point
of U ∩ V , and choose points with rational coordinates u ∈ U, v ∈ V ,
and w ∈ W . We claim that x is uniquely determined by the triple
< u, v, w >; since the set of such triples is countable, this will finish the
proof.
To prove the claim, suppose, that from some x0 ∈ M we arrived to the
same < u, v, w > using spheres S 0 , T 0 ∈ S and components U 0 , V 0 , W 0 of
Rn \(S 0 ∪T 0 ). Since S∩S 0 contains at most one point and since U ∩U 0 6= ∅,
we have that U ⊂ U 0 or U 0 ⊂ U ; similarly for V 0 s and W 0 s. Exchanging
the role of x and x0 and/or of U 0 s and V 0 s if necessary, there are only
two cases to consider: (a) U ⊃ U 0 and V ⊃ V 0 and (b) U ⊂ U 0 , V ⊃ V 0
and W ⊃ W 0 . In case (a) we recall that U ∩ V contains only x and
that x0 ∈ U 0 ∩ V 0 , so x = x0 . In case (b) we get from W ⊂ W 0 that
U 0 ⊂ U ∪ V ; so since U 0 is open and connected, and U ∩ V is just one
point, we infer that U 0 = U and we are back in the already proved case
(a).
Problem 6.
a) We first construct a sequence cn of positive numbers such that
P∞ 1 P∞
cn → ∞ and cn bn < . Let B = bn , and for each k = 0, 1, . . .
n=1 2 n=1
denote by Nk the first positive integer for which
∞
X B
bn ≤ k .
4
n=Nk

2k
Now set cn = for each n, Nk ≤ n < Nk+1 . Then we have cn →∞
5B
and
∞ ∞ ∞ ∞ ∞
X X X X 2k X X 2k B 2
cn b n = cn bn ≤ bn ≤ . k = .
5B 5B 4 5
n=1 k=0 Nk ≤n<Nk+1 k=0 n=Nk k=0

Consider the intervals In = (an − cn bn , an + cn bn ). The sum of their

P
lengths is 2 cn bn < 1, thus there exists a point x0 ∈ (0, 1) which is
2.6. Solutions of Olympic 1999 87

not contained in any In . We show that f is differentiable at x0 , and

f 0 (x0 ) = 0. Since x0 is outside of the intervals In , x0 6= an an for any n
and f (x0 ) = 0. For arbitrary x ∈ (0, 1)\{x0 }, if x = an for some n, then
f (x) − f (x0 ) f (an ) − 0 bn 1
= ≤ = ,
x − x0 |an − x0 | cn b n cn
f (x) − f (x0 )
otherwise = 0. Since cn → ∞, this implies that for arbi-
x − x0
trary  > 0 there are only finitely many x ∈ (0, 1)\{x0 } for which
f (x) − f (x0 )
<
x − x0
does not hold, and we are done.
Remark. The variation of f is finite, which implies that f is differen-
tiable almost everywhere.
b) We remove the zero elements from sequence bn . Since f (x) = 0
except for a countable subset of (0, 1), if f is differentiable at some point
x0 , then f (x0 ) and f 0 (x0 ) must be 0.
It is easy to construct a sequence βn satisfying 0 < βn ≤ bn , bn → 0
P∞
and βn = ∞.
n=1
Choose the numbers a1 , a2 , . . . such that the intervals In = (an −
βn , an + βn )(n = 1, 2, . . .) cover each point of (0, 1) infinitely many times
P
(it is possible since the sum of lengths is 2 bn = ∞. Then for arbitrary
x0 ∈ (0, 1), f (x0 ) = 0 and  > 0 there is an n for which βn <  and c0 ∈ In
which implies
|f (an ) − f (x0 )| bn
> ≥ 1.
|an − x0 | βn

2.6 Solutions of Olympic 1999

2.6.1 Day 1

Problem 1.
2.6. Solutions of Olympic 1999 88

a) The diagonal matrix

 
λ 0
A = λI =  ... 
0 λ

is a solution for equation A3 = A + I if and only if λ3 = λ + 1, because

A3 − A − I = (λ3 − λ − 1)I. This equation, being cubic, has real solution.
b) It is easy to check that the polynomial p(x) = x3 − x − 1 has
a positive real root λ1 (because p(0) < 0) and two conjugated complex
roots λ2 and λ3 (one can check the discriminant of the polynomial, which
 1 3  1 2 23
is − + − = > 0, or the local minimum and maximum
3 2 108
of the polynomial).
If a matrix A satisfies equation A3 = A + I, then its eigenvalues can
be only λ1 , λ2 and λ3 . The multiplicity of λ2 and λ3 must be the same,
because A is a real matrix and its characteristic polynomial has only
real coefficients. Denoting the multiplicity of λ1 by α and the common
multiplicity of λ2 and λ3 by β,

detA = λα1 λβ2 λβ3 = λα1 .(λ2 λ3 )β .

Because λ1 and λ2 λ3 = |λ2 |2 are positive, the product on the right side
has only positive factors.
Problem 2. No. For, let π be a permutation of N and let N ∈ N. We
shall argue that
3N
X π(n) 1
> .
n2 9
n=N +1

In fact, of the 2N numbers π(N + 1), . . . , π(3N ) only N can be ≤ N so

that at least N of them are > N . Hence
3N 3N
X π(n) 1 X 1 1
≥ π(n) > · N.N = .
n2 (3N )2 9N 2 9
n=N +1 n=N +1

Solution 2. Let π be a permutation of N. For any n ∈ N, the

numbers π(1), . . . , π(n) are distinct positive integers, thus π(1) + · · · +
2.6. Solutions of Olympic 1999 89

n(n + 1)
π(n) ≥ 1 + · · · + n = . By this inequality,
2
∞ ∞
X π(n) X 1 1 
2
= (π(1) + · · · + p(n)) 2 − 2
≥
n=1
n n=1
n (n + 1)
∞ ∞ ∞
X n(n + 1) 2n + 1 X 2n + 1 X 1
≥ · 2 = ≥ = ∞.
n=1
2 n (n + 1)2 n=1
2n(n + 1) n=1
n + 1

Problem 3. Writing (1) with n − 1 instead of n,

n−1
X
3k (f (x + ky) − f (x − ky)) ≤ 1. (2)
k=1

From the difference of (1) and (2),

|3n (f (x + ny) − f (x − ny))| ≤ 2;

which means
2
|f (x + ny) − f (x − ny)| ≤ . (3)
3n
For arbitrary u, v ∈ R and n ∈ N one can choose x and y such that
u+v v−u
x − ny = u and x + ny = v, namely x = and y = , Thus, (3)
2 2n
yields
2
|f (u) − f (v)| ≤ n
3
2
for arbitrary positive integer n. Because n can be arbitrary small, this
3
implies f (u) = f (v).
f (x)  x 
Problem 4. Let g(x) = . We have f = g(x). By induction
 x  x g(x)
it follows that g n = g(x), i.e.
g (x)
 x  x
f n = n−1 , n ∈ N. (1)
g (x) g (x)
 x2 
On the other hand, let substitute x by f (x) in f = x. From the
f (x)
f 2 (x)
injectivity of f we get = x, i.e. g(xg(x)) = g(x). Again by
f (f (x))
2.6. Solutions of Olympic 1999 90

induction we deduce that g(xg n (x)) = g(x) which can be written in the
form
f (xg n (x)) = xg n−1 (x), n ∈ N. (2)
Set f (m) = f ◦ f ◦ · · · ◦ f . It follows from (1) and (2) that
| {z }
m times

f (m) (xg n (x)) = xg n−m (x), m, n ∈ N. (3)

Now, we shall prove that g is a constant. Assume g(x1 ) < g(x2 ). Then
we may find n ∈ N such that x1 g n (x1 ) ≤ x2 g n (x2 ). On the other hand,
if m is even then f (m) is strictly increasing and from (3) it follows that
xm1 g
n−m
(x1 ) ≤ xm
2 g
n−m
(x2 ). But when n is fixed the opposite inequality
holds ∀m  1. This contradiction shows that g is a constant, i.e. f (x) =
Cx, C > 0.
Conversely, it is easy to check that the functions of this type verify
the conditions of the problem.
Problem 5.
We prove the more general statement that if at least n + k points are
marked in an n × k grid, then the required sequence of marked points
can be selected.
If a row or a column contains at most one marked point, delete it.
This decreases n + k by 1 and the number of the marked points by at
most 1, so the condition remains true. Repeat this step until each row
and column contains at least two marked points. Note that the condition
implies that there are at least two marked points, so the whole set of
marked points cannot be deleted.
We define a sequence b1 , b2 , . . . of marked points. Let b1 be an arbitrary
marked point. For any positive integer n, let b2n be an other marked
point in the row of b2n−1 and b2n+1 be an other marked point in the
column of b2n .
Let m be the first index for which bm is the same as one of the earlier
points, say bm = bl , l < m.
2.6. Solutions of Olympic 1999 91

If m − l is even, the line segments bl bl+1 , bl+1 bl+2 , . . . , bm−1 bl = bm−1 bm

are alternating horizontal and vertical. So one can choose 2k = m − l,
and (a1 , . . . , a2k ) = (bl , . . . , bm−1 ) or (a1 , . . . , a2k ) = (bl+1 , . . . , bm ) if l is
odd or even, respectively.
If m − l is odd, then the points b1 = bm , bl+1 and bm−1 are in the same
row/column. In this case chose 2k = m − l − 1. Again, the line seg-
ments bl+1 bl+2 , bl+2 bl+3 , . . . , bm−1 bl+1 are alternating horizontal and verti-
cal and one can choose (a1 , . . . , a2k ) = (bl+1 , . . . , bm−1 ) or (a1 , . . . , a2k ) =
bl+2 , . . . , bm−1 , bl+1 if l is even or odd, respectively.
Solution 2. Define the graph G in the following way: Let the vertices
of G be the rows and the columns of the grid. Connect a row r and a
column c with an edge if the intersection point of r and c is marked.
The graph G has 2n vertices and 2n edges. As is well known, if a
graph of N vertices contains no circle, it can have at most N − 1 edges.
Thus G does contain a circle. A circle is an alternating sequence of rows
and columns, and the intersection of each neighbouring row and column
is a marked point. The required sequence consists of these intersection
points.
Problem 6.
0
R1 0 R1
a) Let g(x) = max(0, f (x)). Then 0 < f (x)dx = g(x)dx +
−1 −1
R1 R1 R1 R1
(f 0 (x)−g(x))dx, so we get |f 0 (x)|dx = g(x)dx+ (g(x)−f 0 (x))dx <
−1 −1 −1 −1
R1
2 g(x)dx. Fix p and c (to be determined at the end). Given any
−1
t > 0, choose for every x such that g(x) > t an interval Ix = [x, y] such
that |f (y) − f (x)| > cg(x)1/p |y − x| > ct1/p |Ix | and choose disjoint Ixi .
that cover at least one third of the measure of the set {g > t}. For
1/p
R 0 R1 0 R1
I = ∪Ii we thus have ct |I| ≤ f (x)dx ≤ |f (x)|dx < 2 g(x)dx;
i I −1 −1
Z1
6 −1/p
so |{g > t}| ≤ 3|I| < t g(x)dx. Integrating the inequality, we
c
−1
2.6. Solutions of Olympic 1999 92

Z1
R1 R1 6 p
get g(x)dx = |{g > t}|dt < g(x)dx; this is a contradiction
−1 0 cp−1
−1
e.g. for cp = (6p)/(p − 1).
1
b) No. Given c > 1, denote α = and choose 0 <  < 1 such
 1 +  −α c
1
that < . Let g : [−1, 1] → [−1, 1] be continuous, even,
2 4  |x| +  −α−1
g(x) = −1 for |x| ≤  and 0 ≤ g(x) < α for  < |x| ≤
2
Z1 
R1  |x| +  −α−1 
1 is chosen such that g(t)dt > − + α dt = − +
 2 2 2

  1 +  −α  R
2 1 − > . Let f = g(t)dt. Then f (1) − f (−1) ≥
2
R1
−2 + 2 g(t)dt > 0. If  < x < 1 and y = −, then |f (x) − f (y)| ≥

Rx Rx  t +  −α−1  x +  −α |x − y|
2 − g(t)dt ≥ 2 − α = 2 > g(x) =
  2 2 α
|x − y|
f 0 (x) ; symmetrically for −1 < x < − and y = .
α

2.6.2 Day 2

Problem 1. From 0 = (a + b)2 = a2 + b2 + ab + ba = ab + ba, we have

ab = −(ba) for arbitrary a, b, which implies

abc = a(bc) = −((bc))a = −(b(ca)) = (ca)b = c(ab) = −((ab)c) = −abc.

Problem 2. For all nonnegative integers n and modulo 5 residue class

(r)
r, denote by pn the probability that after n throwing the sum of values
(0) (1) (2)
is congruent to r modulo n. It is obvious that p0 = 1 and p0 = p0 =
(3) (4)
p0 = p0 = 0.
Moreover, for any n > 0 we have
6
X 1 (r−i)
p(r)
n = pn−1 . (1)
i=1
6

From this recursion we can compute the probabilities for small values
2.6. Solutions of Olympic 1999 93

(r) 1 4
of n and can conjecture that pn = + if n ≡ r (mod 5) and
5 5.6n
(r) 1 1
pn = − otherwise. From (1), this conjecture can be proved by
5 5.6n
induction.
Solution 2. Let S be the set of all sequences consisting of digits
1, . . . , 6 of length n. We create collections of these sequences.
Let a collection contain sequences of the form

66 . . . 6} XY1 . . . Yn−k−1 ,
| {z
k

where X ∈ {1, 2, 3, 4, 5} and k and the digits Y1 , . . . , Yn−k−1 are fixed.

Then each collection consists of 5 sequences, and the sums of the digits
of sequences give a whole residue system mod 5.
Except for the sequence 66 . . . 6, each sequence is the element of one
collection. This means that the number of the sequences, which have
1
a sum of digits divisible by 5, is (6n − 1) + 1 if n is divisible by 5,
5
1 n
otherwise (6 − 1).
5
1 4
Thus, the probability is + if n is divisible by 5, otherwise it is
5 5.6n
1 1
− .
5 5.6n
Solution 3. For arbitrary positive integer k denote by pk the proba-
bility that the sum of values is k. Define the generating function
∞
X  x + x2 + x3 + x4 + x5 + x6 n
k
f (x) = pk x = .
6
k=1

(The last equality can be easily proved by induction.)

P∞ 2π 2π
Our goal is to compute the sum p5k . Let  = cos + i sin ; be
k=1 5 5
the first 5th root of unity. Then
∞
X f (1) + f () + f (2 ) + f (3 ) + f (4 )
p5k = .
5
k=1

jn j
Obviously f (1) = 1, and f ( ) = n for j = 1, 2, 3, 4. This implies that
6
2.6. Solutions of Olympic 1999 94

4 −1
f () + f (2 ) + f (3 ) + f (4 ) is n if n is divisible by 5, otherwise it is n .
∞
6 6
P 1 1 1 1
Thus, p5k is + if n is divisible by 5, otherwise it is − .
k=1 5 5.6n 5 5.6n
Problem 3.
The inequality
3 3 1  1 2
0 ≤ x − x + = (x + 1) x −
4 4 2
holds for x ≥ −1.
Substituting x1 , . . . , xn , we obtain
n  n n n
X
3 3 1 X 3 3 X n 3X n
0≤ xi − xi + = xi − xi + = 0 − xi + ,
i=1
4 4 i=1
4 i=1 4 4 i=1 4
n
P n
so xi ≤ . Remark. Equailty holds only in the case when n = 9k, k
i=1 3
1
of the x1 , . . . , xn are −1, and 8k of them are .
2
Problem 4. Assume that such a function exists. The initial inequality
f 2 (x) f (x)y
can be written in the form f (x)−f (x+y) ≥ f (x)− = .
f (x) + y f (x) + y
Obviously, f is a decreasing function. Fix x > 0 and choose n ∈ N such
that nf (x + 1) ≥ 1. For k = 0, 1, . . . , n − 1 we have
 k  k + 1 f (x + nk ) 1
f x+ −f x+ ≥ ≥ .
n n nf (x + nk ) + 1 2n
1
The additon of these inequalities gives f (x + 1) ≥ f (x) − . From this
2
it follows that f (x + 2m) ≤ f (x) − m for all m ∈ N. Taking m ≥ f (x),
we get a contradiction with the conditon f (x) > 0.
Problem 5. Let S be the set of all words consisting of the letters x, y, z,
and consider an equivalence relation ∼ on S satisfying the following
conditions: for arbitrary words u, v, w ∈ S
(i) uu ∼ u;
(ii) if v ∼ w, then uv ∼ uw and vu ∼ wu.
Show that every word in S is equivalent to a word of length at most
8. (20 points)
2.6. Solutions of Olympic 1999 95

Solution. First we prove the following lemma: If a word u ∈ S

contains at least one of each letter, and v ∈ S is an arbitrary word, then
there exists a word w ∈ S such that uvw ∼ u.
If v contains a single letter, say x, write u in the form u = u1 xu2 , and
choose w = u2 . Then uvw = u1 xu2 )xu2 = u1 ((xu2 )(xu2 )) ∼ u1 (xu2 ) =
u.
In the general case, let the letters of v be a1 , . . . , ak . Then one can
choose some words w1 , . . . , wk such that (ua1 )w1 ) ∼ u, (ua1 a2 )w2 ∼
ua1 , . . . , (ua1 . . . ak )wk ∼ ua1 . . . ak−1 . Then u ∼ ua1 w1 ∼ ua1 a2 w2 w1 ∼
· · · ∼ ua1 . . . ak wk . . . w1 = uv(wk . . . w1 ), so w = wk . . . w1 is a good
choice.
Consider now an arbitrary word a, which contains more than 8 digits.
We shall prove that there is a shorter word which is equivalent to a.
If a can be written in the form uvvw, its length can be reduced by
uvvw ∼ uvw. So we can assume that a does not have this form.
Write a in the form a = bcd, where band d are the first and last four
letter of a, respectively. We prove that a ∼ bd.
It is easy to check that b and d contains all the three letters x, y and z,
otherwise their length could be reduced. By the lemma there is a word
e such that b(cd)e ∼ b, and there is a word f such that def ∼ d. Then
we can write

a = bcd ∼ bc(def ) ∼ bc(dedef ) = (bcde)(def ) ∼ bd.

Remark. Of course, it is enough to give for every word of length 9 an

shortest shorter word. Assuming that the first letter is x and the second
is y, it is easy (but a little long) to check that there are 18 words of
length 9 which cannot be written in the form uvvw.
For five of these words there is a 2-step solution, for example

xyxzyzxzy ∼ xyxzyzxzyzy ∼ xyxzyzy ∼ xyxzy.

In the remaining 13 cases we need more steps. The general algorithm

given by the Solution works for these cases as well, but needs also
2.7. Solutions of Olympic 2000 96

very long words. For example, to reduce the length of the word a =
xyzyxzxyz, we have set b = xyzy, c = x, d = zxyz, e = xyxzxzyxyzy, f =
zyxyxzyxzxzxzxyxyzxyz. The longest word in the algorithm was

bcdedef =

= xyzyxzxyzxyxzxzyxyzyzxyzxyxzxzyxyzyzyxyxzyxzxzxzxyxyzxyz,
which is of length 46. This is not the shortest way: reducing the length
of word a can be done for example by the following steps:

xyzyxzxyz ∼ xyzyxzxyzyx ∼ xyzyxzxyzyxyzyz ∼

xyzyxzxyzyxzyxyzyz ∼∼ xyzyxzyxyz ∼ xyzyxyz.

(The last example is due to Nayden Kambouchev from Sofia University.)

Problem 6. Let A = {a1 , . . . , ak }. Consider the k-tuples
 2πia1 t 2πiak t 
exp , . . . , exp ∈ Ck , t = 0, 1, . . . , n − 1.
n n
Each component is in the unit circle |z| = 1. Split the circle into 6 equal
arcs. This induces a decomposition of the k-tuples into 6k classes. By
1
the condition k ≤ ln we have n > 6k , so there are two k-tuples in
100
the same class say for t1 < t2 . Set r = t2 − t1 . Then
2πiaj r  2πa t
j 2 2πaj t1  π 1
Reexp = cos − ≥ cos =
n n n 3 2
for all j, so
k
|f (r)| ≥ Ref (r) ≥ .
2

2.7 Solutions of Olympic 2000

2.7.1 Day 1

Problem 1.
a) Yes.
Proof: Let A = {x ∈ [0, 1] : f (x) > x}. If f (0) = 0 we are done, if
not then A is non-empty (0 is in A) bounded, so it has supremum, say
a. Let b = f (a).
2.7. Solutions of Olympic 2000 97

I. case: a < b. Then, using that f is monotone and a was the sup, we
get b = f (a) ≤ f ( (a+b) a+b
2 ) ≤ 2 , which contradicts a < b.
II. case: a > b. Then we get b = f (a) ≥ f ( a+b a+b
2 ) > 2 contradiction.
Therefore we must have a = b.
b) No. Let, for example,
x 1
f (x) = 1 − if x ≤
2 2
and
1 x 1
− if x >
f (x) =
2 2 2
This is clearly a good counter-example.
Problem 2. Short solution. Let
p(x) − p(y)
P (x, y) = = x4 + x3 y + x2 y 2 + xy 3 + y 4 + 1
x−y
and
q(x) − q(y)
Q(x, y) = = x4 + x3 y + x2 y 2 + xy 3 + y 4 + x + y.
x−y
We need those pairs (w, z) which satisfy P (w, z) = Q(w, z) = 0.
From P − Q = 0 we have w + z = 1. Let c = wz. After a short
calculation we obtain c2 − 3c + 2 = 0, which has the solutions c = 1 and
c = 2. From the system w + z = 1, wz = c we obtain the following pairs:
 1 ± √3i 1 ∓ √3i   1 ± √7i 1 ∓ √7i 
, and , .
2 2 2 2
Problem 3.
A and B are square complex matrices of the same size and

rank(AB − BA) = 1.

Show that (AB − BA)2 = 0.

Let 0 = AB − BA. Since rankC = 1, at most one eigenvalue of C
is different from 0. Also trC = 0, so all the eigevalues are zero. In the
Jordan canonical form there can only be one 2 × 2 cage and thus C 2 = 0.
2.7. Solutions of Olympic 2000 98

Problem 4.
a)
n n n i n n
X xi 2 X xi xj X xi X xi X xi xi X
√ = √√ ≥ √ √ ≥ √ i√ = x2i
i=1
i i,j
i j i=1
i j=1 j i=1
i i i=1

b)
∞ ∞ ∞ ∞
X 1  X 2 1/2 X 1 X xi
√ xi ≤ √ √
m=1
m i=m m=1
m i=m i − m + 1
by a)
∞ i
X X 1
= xi √ √
i=1 m=1
m i−m+1
You can get a sharp bound on
i
X 1
sup √ √
i m=1
m i−m+1
by checking that it is at most
Zi+1
1
√ √ dx = π
x i+1−x
0

Alternatively you can observe that

i i/2
X 1 X 1
√ √ =2 √ √ ≤
m=1
m i + 1 − m m=1
m i + 1 − m
i/2 r
1 X 1 1 i
≤ 2q √ ≤ 2 q .2 =4
i
m=1
m i 2
2 2

Problem 5.
Suppose that e + f + g = 0 for given idempotents e, f, g ∈ R. Then

g = g 2 = (−(e + f )2 = e + (ef + f e) + f = (ef + f e) − g,

i.e. ef + f e = 2g, whence the additive commutator

[e, f ] = ef − f e = [e, ef + f e] = 2[e, g] = 2[e, −e − f ] = −2[e, f ],

2.7. Solutions of Olympic 2000 99

i.e. ef = f e (since R has zero characteristic). Thus ef + f e = 2g

becomes ef = g, so that e + f + ef = 0. On multiplying by e, this
yields e + 2ef = 0, and similarly f + 2ef = 0, so that f = −2ef = e,
hence e = f = g by symmetry. Hence, finaly, 3e = e + f + g = 0, i.e.
e = f = g = 0.
For part (i) just omit some of this.
Problem 6.
From the conditions it is obvious that F is increasing and lim bn = ∞.
n→∞
By Lagrange’s theorem and the recursion in (1), for all k ≥ 0 integers
there exists a real number ξ ∈ (ak , ak+1 ) such that
f (ξ)
F (ak+1 ) − F (ak ) = f (ξ)(ak+1 − ak ) = . (2)
f (ak )
By the monotonity, f (ak ) ≤ f (ξ) ≤ f (ak+1 ), thus
f (ak+1 ) f (ak+1 ) − f (ak )
1 ≤ F (ak+1 ) − F (ak ) ≤ =1+ . (3)
f (ak ) f (ak )
Summing (3) for k = 0, . . . , n − 1 and substituting F (bn ) = n, we have
n−1
X f (ak+1 ) − f (ak )
F (bn ) < n + F (a0 ) ≤ F (an ) ≤ F (bn ) + F (a0 ) + . (4)
f (ak )
k=0

From the first two inequalities we already have an > bn and lim an = ∞.
n→∞
Let  be an arbitrary positive number. Choose an integer K such
2
that f (aK ) > . If n is sufficiently large, then

n−1
X f (ak+1 ) − f (ak )
F (a0 ) + =
f (ak )
k=0
K  −1
 X f (ak+1 ) − f (ak )  X f (ak+1 ) − f (ak )
= F (a0 ) + + n−1 < (5)
f (ak ) f (ak )
k=0 k=K
n−1
1 X
< O (1) + (f (ak+1 ) − f (ak )) <
f (aK )
k=K

< O (1) + (f (an ) − f (aK )) < f (an ).
2
2.7. Solutions of Olympic 2000 100

Inequalities (4) and (5) together say that for any positive , if n is suffi-
ciently large,
F (an ) − F (bn ) < f (an ).
Again, by Lagrange’s theorem, there is a real number ζ ∈ (bn , an ) such
that
F (an ) − F (bn ) = f (ζ)(an − bn ) > f (bn )(an − bn ), (6)
thus
f (bn )(an − bn ) < f (an ). (7)
Let B be an upper bound for f 0 . Apply f (an ) < f (bn ) + B(an − bn ) in
(7):

f (bn )(an − bn ) < (f (bn ) + B(an − bn )),

(f (bn ) − B)(an − bn ) < f (bn ). (8)

Due to lim f (bn ) = ∞, the first factor is positive, and we have

n→∞

f (bn )
an − bn <  < 2 (9)
f (bn ) − B
for sufficiently large n.
Thus, for arbitrary positive  we proved that 0 < an − bn < 2 if n is
sufficiently large.

2.7.2 Day 2

Problem 1.
We start with the following lemma: If a and b be coprime positive
integers then every sufficiently large positive integer m can be expressed
in the form ax + by with x, y non-negative integers.
Proof of the lemma. The numbers 0, a, 2a, . . . , (b−1)a give a complete
residue system modulo b. Consequently, for any m there exists a 0 ≤
x ≤ b−1 so that ax ≡ m (mod b). If m ≥ (b−1)a, then y = (m−ax)/b,
for which x + by = m, is a non-negative integer, too.
2.7. Solutions of Olympic 2000 101

Now observe that any dissection of a cube into n smaller cubes may
be refined to give a dissection into n+(ad −1) cubes, for any a ≥ 1. This
refinement is achieved by picking an arbitrary cube in the dissection, and
cutting it into ad smaller cubes. To prove the required result, then, it
suffices to exhibit two relatively prime integers of form ad − 1. In the 2-
dimensional case, a1 = 2 and a2 = 3 give the coprime numbers 22 −1 = 3
and 32 − 1 = 8. In the general case, two such integers are 2d − 1 and
(2d − l)d − 1, as is easy to check.
Problem 2. Let (x − α, x + α) ⊂ [0, 1] be an arbitrary non-empty open
interval. The function f is not monoton in the intervals [x − α, x] and
[x, x + α], thus there exist some real numbers x − α ≤ p < q ≤ x, x ≤
r < s ≤ x + α so that f (p) > f (q) and f (r) < f (s).
By Weierstrass’ theorem, f has a global minimum in the interval
[p, s]. The values f (p) and f (s) are not the minimum, because they are
greater than f (q) and f (s), respectively. Thus the minimum is in the
interior of the interval, it is a local minimum. So each nonempty interval
(x − α, x + α) ⊂ [0, 1] contains at least one local minimum.
Problem 3. The statement is not true if p is a constant polynomial.
We prove it only in the case if n is positive.
For an arbitrary polynomial q(z) and complex number c, denote by
µ(q, c) the largest exponent α for which q(z) is divisible by (z − c)α .
(With other words, if c is a root of q, then µ(q, c) is the root’s multiplicity.
Otherwise 0.)
Denote by S0 and S1 the sets of complex numbers z for which p(z) is
0 or 1, respectively. These sets contain all roots of the polynomials p(z)
and p(z) − 1, thus
X X
µ(p, c) = µ(p − 1, c) = n. (1)
c∈S0 c∈S1

The polynomial p0 has at most n − 1 roots (n > 0 is used here). This

2.7. Solutions of Olympic 2000 102

implies that
X
µ(p0 , c) ≤ n − 1. (2)
c∈S0 ∪S1
If p(c) = 0 or p(c) − 1 = 0, then

µ(p, c) − µ(p0 c) = 1 or µ(p − 1, c) − µ(p0 c) = 1, (3)

respectively. Putting (1), (2) and (3) together we obtain

X X
0
|S0 | + |S1 | = (µ(p, c) − µ(p , c)) + (µ(p − 1, c) − µ(p0 , c)) =
c∈S c∈S
X X0 X 1

0
= µ(p, c) + µ(p − 1, c) − µ(p , c) ≥ n + n − (n − 1) = n + 1.
c∈S0 c∈S1 c∈S0 ∪S1

Problem 4.
a) Without loss of generality, we can assume that the point A2 is the
origin of system of coordinates. Then the polynomial can be presented
in the form

y = (a0 x4 + a1 x3 + a2 x2 + a3 x + a4 )x2 + a5 x,

where the equation y = a5 x determines the straight line A1 A3 . The

abscissas of the points A1 and A3 are −a and a, a > 0, respectively.
Since −a and a are points of tangency, the numbers −a and a must be
double roots of the polynomial a0 x4 + a1 x3 + a2 x2 + a3 x + a4 . It follows
that the polynomial is of the form

y = a0 (x2 − a2 )2 + a5 x.

The equality follows from the equality of the integrals

Z0 Za
a0 (x2 − a2 )x2 dx = a0 (x2 − a2 )x2 dx
−a 0

due to the fact that the function y = a0 (x2 − a2 ) is even.

b) Without loss of generality, we can assume that a0 = 1. Then the
function is of the form

y = (x + a)2 (x − b)2 x2 + a5 x,
2.7. Solutions of Olympic 2000 103

where a and b are positive numbers and b = ka, 0 < k < ∞. The areas
of the figures at the segments A1 A2 and A2 A3 are equal respectively to
Z0
2 2 2 a7
(x + a) (x − b) x dx = (7k 2 + 7k + 2)
210
−a

and
Zb
a7
(x + a)2 (x − b)2 x2 dx = (2k 2 + 7k + 7)
210
0
Then
2
5 2k
+ 7k + 7
K=k .
7k 2 + 7k + 2
2k 2 + 7k + 7
The derivative of the function f (k) = 2 is negative for 0 <
7k + 7k + 2
7 2
k < ∞. Therefore f (k) decreases from to when k increases from 0
2 7
2 2k 2 + 7k + 7 7
to ∞. Inequalities < 2 < imply the desired inequalities.
7 7k + 7k + 2 2
Problem 5.
First solution. First, if we assume that f (x) > 1 for some xR+ ,
x
setting y = gives the contradiction f (x) = 1. Hence f (x) ≤ 1
f (x) − 1
for each x ∈ R+ , which implies that f is a decreasing function.
If f (x) = 1 for some x ∈ R+ , then f (x + y) = f (y) for each y ∈ R+ ,
and by the monotonicity of f it follows that f ≡ 1.
Let now f (x) < 1 for each x ∈ R+ . Then f is strictly decreasing
function, in particular injective. By the equalities

f (x)f (yf (x)) = f (x + y) =

= f (yf (x) + x + y(1 − f (x))) = f (yf (x))f ((x + y(1 − f (x)))f (yf (x)))

we obtain that x = (x + y(1 − f (x)))f (yf (x)). Setting x = 1, z = xf (1)

1 − f (1) 1
and a = , we get f (z) = .
f (1) 1 + az
1
Combining the two cases, we conclude that f (x) = for each
1 + ax
2.7. Solutions of Olympic 2000 104

x ∈ R+ , where a ≥ 0. Conversely, a direct verification shows that the

functions of this form satisfy the initial equality.
Second solution. As in the first solution we get that f is a decreasing
function, in particular differentiable almost everywhere. Write the initial
equality in the form
f (x + y) − f (x) f (yf (x)) − 1
= f 2 (x) .
y yf (x)
It follows that if f is differentiable at the point x ∈ R+ , then there
f (z) − 1
exists the limit lim+ =: −a. Therefore f 0 (x) = −af 2 (x) for
z→0
 1z 0 1
each x ∈ R+ , i.e. = a, which means that f (x) = .
f (x) ax + b
Substituting in the initial relaton, we find that b = 1 and a ≥ 0.
Problem 6. First we prove that for any polynomial q and m × m
matrices A and B, the characteristic polinomials of q(eAB ) and q(eBA )
∞
are the same. It is easy to check that for any matrix X, q(eX ) = cn X n
P
n=0
with some real numbers cn which depend on q. Let
∞
X ∞
X
n−1
C= cn .(BA) B= cn .B(AB)n−1 .
n=1 n=1

Then q(eAB ) = c0 I + AC and q(eBA ) = c0 I + CA. It is well-known that

the characteristic polynomials of AC and CA are the same; denote this
polynomial by f (x). Then the characteristic polynomials of matrices
q(eAB ) and q(eBA ) are both f (x − c0 ).
Now assume that the matrix p(eAB ) is nilpotent, ie. (p(eAB ))k = 0 for
some positive integer k. Chose q = pk . The characteristic polynomial
of the matrix q(AB ) = 0 is xm , so the same holds for the matrix q(eBA ).
By the theorem of Cayley and Hamilton, this implies that (q(eBA ))m =
(p(eBA ))km = 0. Thus the matrix q(eBA ) is nilpotent, too.
2.8. Solutions of Olympic 2001 105

2.8 Solutions of Olympic 2001

2.8.1 Day 1

Problem 1. Since there are exactly n rows and n columns, the choice
is of the form
{(j, σ(j) : j = 1, . . . , n}
where σ ∈ Sn is a permutation. Thus the corresponding sum is equal to
n
X n
X n
X n
X
n(j − 1) + σ(j) = nj − n+ σ(j)
j=1 j=1 j=1 j=1
n n n
X X X n(n + 1) n(n2 + 1)
=n j− n+ j = (n + 1) − n2 = .
j=1 j=1 j=1
2 2

which shows that the sum is independent of σ.

Problem 2.
1. There exist integers u and v such that us + vt = 1. Since ab = ba,
we obtain

ab = (ab)us+vt = (ab)us ((ab)t )v = (ab)us e = (ab)us = aus (bs )u = aus e = aus .

Therefore, br = ebr = ar br = (ab)r = ausr = (ar )us = e. Since xr+ys = 1

for suitable integers x and y,

b = bxr+ys = (br )x (bs )y = e.

It follows similarly that a = e as well.

2. This is not true. Let a = (123) and b = (34567) be cycles of the
permutation group S7 of order 7. Then ab = (1234567) and a3 = b5 =
(ab)7 = e.
Problem 3.
∞ ∞
X tn 1−t X tn
lim (1 − t) n
= lim (−ln t) n
=
t→1−0 1 + t t→1−0 −ln t 1 + t
n=1 n=1
∞ ∞ Z∞
X 1 X 1 dx
= lim (−ln t) = lim h = = ln 2.
t→1−0 1 + e−nln t h→+0 1 + e nh 1 + ex
n=1 n=1 0
2.8. Solutions of Olympic 2001 106

Problem 4. Let p(x) = (x−1)k r(x) and j = e1πij/q (j = 1, 2, . . . , q−1).

As is well-known, the polynomial xq−1 +xq−2 +· · ·+x+1 = (x−1 ) . . . (x−
q−1 ) is irreducible, thus all 1 , . . . , q−1 are roots of r(x), or none of them.
Suppose that none of 1 , . . . , q−1 is a root of r(x). Then q−1
Q
j=1 r(j ) is
a rational integer, which is not 0 and
q−1
Y q−1
Y q−1
Y
q−1 k
(n + 1) ≥ |p(j )| = (1 − j ) r(j ) ≥
j=1 j=1 j=1
q−1
Y k
≥ (1 − j ) = (1q−1 + 1q−2 + · · · + 11 + 1)k = q k .
j=1

q k
This contradicts the condition < .
ln q ln (n + 1)
Problem 5.
The statement will be proved by inductionh on n.
i For n = 1, there is
a b
6 0, and c 6= 0
nothing to do. In the case n = 2, write A = c d . If b =
or b = c = 0 then A is similar to
 h i   
1 0 a b 1 0 0 b
a/b 1 c d −a/b 1 = c − ad/b a + d
or  h i   
1 −a/c a b 1 a/c 0 b − ad/c
0 1 c d 0 1 = c a+d
respectively. If b = c = 0 and a 6= d, then A is similar to
1 1 a 0 1 −1 0 d−a
h ih ih i h i
0 1 0 d 0 1 = 0 d ,

and we can perform the step seen in the case b 6= 0 again.

Assume now that n> 3 and the problem has been solved for all
0
A ∗
n0 < n. Let A = ∗ β , where A0 is (n − 1) × (n − 1) matrix. Clearly
n
0
we may assume that A 6
= λ0 I, so the induction provides a P with, say,
0 ∗
h i
P −1 A0 P = ∗ α . But then the matrix
n−1
  h i  −1 0 
−1 A0 ∗ P 0 P A P ∗
B = P0 10 ∗ β 0 1 = ∗ β
2.8. Solutions of Olympic 2001 107

is similar to A and its diagonal i is (0, 0, . . . , 0, α, β). On the other hand,

0 ∗
h
we may also view B as ∗ C , where C is an (n − 1) × (n − 1) matrix
with diagonal (0, . . . , 0, α, β). If the inductive hhypothesis is applicable
0 ∗
i
to C, we would have Q−1 CQ = D, with D = ∗ γ so that finally
n−1
the matrix
  i  h
1 0 1 0 1 0 0 ∗ 1 0 0 ∗
h ih i h i
E = 0 Q−1 .B. 0 Q = 0 Q−1 ∗ C 0 Q = ∗ D

is similar to A and its diagonal is (0, 0, . . . , 0, γ), as required.

The inductive argument can fail only when n−1 = 2 and the resulting
matrix applying P has the form
" #
0 a b
P −1 AP = c d 0
e 0 d
where d 6= 0. The numbers a, b, c, e cannot be 0 at the same time. If,
say, b 6= 0, A is similar to
−b
" #" #" # " #
1 0 0 0 a b 1 0 0 a b
0 1 0 c d 0 0 1 0 = c d 0 .
1 0 1 e 0 d −1 0 1 e−b−d a b+d
Performing half of the induction step again, the diagonal of the resulting
matrix will be (0, d − b, d + b) (the trace is the same) and the induction
step can be finished. The cases a 6= 0, c 6= 0 and e 6= 0 are similar.
Problem 6.
Let 0 <  < A be an arbitrary real number. If x is sufficiently large
then f (x) > 0, g(x) > 0, |a(x) − A| < , |b(x) − B| <  and
f 0 (x) f (x) f 0 (x) f (x)
B −  < b(x) = 0 + a(x) < 0 + (A + ) <
g (x) g(x) g (x) g(x)
(A + )(A + 1) f 0 (x)(g(x))A + A.f (x).(g(x))A−1 .g 0 (x)
< · = (1)
A (A + 1).(g(x))A .g 0 (x)
(A + )(A + 1) (f (x).(g(x))A )0
= · ,
A ((g(x))A+1 )0
thus
(f (x).(g(x))A )0 A(B − )
> . (2)
((g(x))A+1 )0 (A + )(A + 1)
2.8. Solutions of Olympic 2001 108

It can be similarly obtained that, for sufficiently large x,

(f (x).(g(x))A )0 A(B + )
< . (3)
((g(x))A+1 )0 (A − )(A + 1)
From  → 0, we have
(f (x).(g(x))A )0 B
lim = .
x→∞ ((g(x))A+1 )0 A+1
By l’Hospital’s rule this implies
f (x) f (x).(g(x))A B
lim = lim = .
x→∞ g(x) x→∞ (g(x))A+1 A+1

2.8.2 Day 2

Problem 1.
Multiply the left hand side polynomials. We obtain the following
equalities:
a0 b0 = 1, a0 b1 + a1 b0 = 1, . . .
Among them one can find equations

a0 + a1 bs−1 + a2 bs−2 + · · · = 1

and
b0 + b1 ar−1 + b2 ar−2 + · · · = 1.
From these equations it follows that a0 , b0 ≤ 1. Taking into account that
a0 b0 = 1 we can see that a0 = b0 = 1.
Now looking at the following equations we notice that all a0 s must be
less than or equal to 1. The same statement holds for the b0 s. It follows
from a0 b1 + a1 b0 = 1 that one of the numbers a1 , b1 equals 0 while the
other one must be 1. Follow by induction.
p √ √
Problem 2. Obviously a2 =p 2 − 2 < 2.
√
Since the function f (x) = 2 − 4 − x2 is increasing on the interval
[0, 2] the inequality a1 > a2 implies that a2 > a3 . Simple induction ends
2.8. Solutions of Olympic 2001 109

the proof of monotonicity of (an ). In the same way we prove that (bn )
decreases (just notice that
2x 2
g(x) = √ = q .
2 + 4 + x2 2
+ 1+ 4
x x2

It is a matter of simple manipulation to prove that 2f (x) > x for all

x ∈ (0, 2), this implies that the sequence (2n an ) is strictly increasing.
The inequality 2g(x) < x for x ∈ (0, 2) implies that the sequence (2n bn )
2 4b2
strictly decreases. By an easy induction one can show that an =
4 + b2n
for positive integers n. Since the limit of the decreasing sequence (2n bn )
of positive numbers is finite we have
4.4n b2n
lim 4n a2n = lim 2
= lim 4n b2n .
4 + bn
We know already that the limits lim 2n an and lim 2n bn are equal. The first
of the two is positive because the sequence (2n an ) is strictly increasing.
The existence of a number C follows easily from the equalities

n n

n 2 4n+1 b2n  n n (2n bn )4 1 1
2 b n − 2 an = 4 b n − /(2 b n + 2 a n ) = · ·
4 + b2n 4 + b2n 4n 2n (bn + an )
and from the existence of positive limits lim 2n bn and lim 2n an .
Remark. The last problem may be solved in a much simpler way by
someone who is able to make use of sine and cosine. It is enough to
π π
notice that an = 2 sin n+1 and bn = 2tan n+1 .
2 2
Problem 3.
The unit sphere in Rn is defined by
n n
X o
n 2
Sn−1 = (x1 , . . . , xn ) ∈ R xk = 1 .
k=1

The distance between the points X = (x1 , . . . , xn ) and Y = (y1 , . . . , yn )

is: n
X
2
d (X, Y ) = (xk − yk )2 .
k=1
2.8. Solutions of Olympic 2001 110

We have
√
d(X, Y ) > 2 ⇔ d2 (X, Y ) > 2
Xn Xn n
X
2 2
⇔ xk + yk + 2 x k yk > 2
k=1 k=1 k=1
Xn
⇔ x k yk < 0
k=1
Taking account of the symmetry of the sphere, we can suppose that
A1 = (−1, 0, . . . , 0).
n
P
For X = A1 , xk yk < 0 implies y1 > 0, ∀Y ∈ Mn .
k=1
Let X = (x1 , X), Y = (y1 , Y ) ∈ Mn \{A1 }, X, Y ∈ Rn−1 .
We have
Xn n−1
X n−1
X
x k yk < 0 ⇒ x 1 y1 + xk yk < 0 ⇔ x0k yk0 < 0,
k=1 k=1 k=1
where
xk y
x0k = pP , y 0
k = pPk .
2
xk yk 2
therefore
(x01 , . . . , x0n−1 ), (y10 , . . . , yn−1
0
) ∈ Sn−2
n
P
and verifies xk yk < 0.
k=1
If an is the search number of points in Rn we obtain an ≤ 1 + an−1
and a1 = 2 implies that an ≤ n + 1.
We show that an = n + 1, giving an example of a set Mn with (n + 1)
elements satisfying the conditions of the problem.
A1 = (−1, 0, 0, 0, . . . , 0, 0)
1 
A2 = , −c1 , 0, 0, . . . , 0, 0
n
1 1 
A3 = , · c1 , −c2 , 0, . . . , 0, 0
n n−1
1 1 1 
A4 = , · c1 , · c2 , −c3 , . . . , 0, 0
n n−1 n−1
2.8. Solutions of Olympic 2001 111

1
1 1 1 
An−1 = · c1 ,
, · c2 , · c3 , . . . , −cn−2 , 0
n n−1 n−2 n−3
1 1 1 1 1 
An = , · c1 , · c2 , · c3 , . . . , · cn−2 , −cn−1
n n−1 n−2 n−3 2
1 1 1 1 1 
An+1 = , · c1 , · c2 , · c3 , . . . , · cn−2 , cn−1
n n−1 n−2 n−3 2
where r
1  1 
ck = 1+ 1− , k = 1, n − 1.
n n−k+1
n 1 n
x2k = 1, ∀X, Y ∈ {A1 , . . . , An+1 }.
P P
We have xk yk = − < 0 and
k=1 n k=1
These points are on the unit sphere in Rn and the distance between any
two points is equal to
√ 1 √
r
d = 2 1 + > 2.
n
Remark. For n = 2 the points form an equilateral triangle in the unit
circle; for n = 3 the four points from a regular tetrahedron and in Rn
the points from an n dimensional regular simplex.
Problem 4.
We will only prove (2), since it implies (1). Consider a directed graph
G with n vertices V1 , . . . , Vn and a directed edge from Vk to Vl when
ak,l 6= 0. We shall prove that it is acyclic.
Assume that there exists a cycle and take one of minimum length m.
Let j1 < · · · < jm be the vertices the cycle goes through and let σ0 ∈ Sn
be a permutation such that ajk ,jσ0 (k) 6= 0 for k = 1, . . . , m. Observe that
for any other σ ∈ Sn we have ajk ,jσ(k) = 0 for some k ∈ {1, . . . , m},
otherwise we would obtain a different cycle through the same set of
vertices and, consequently, a shorter cycle. Finally

0 = det(ajk ,jl )k,l=1,...,m

m
Y X m
Y
= (−1)signσ0 ajk ,jσ0 (k) + (−1) signσ
ajk ,jσ (k) 6= 0,
k=1 σ6=σ0 k=1

which is a contradiction.
2.8. Solutions of Olympic 2001 112

Since G is acyclic there exists a topological ordering i.e. a permutation

(σ ∈ Sn such that k < l whenever there is an edge from Vσ(k) to Vσ(l) . It
is easy to see that this permutation solves the problem.
Problem 5. Suppose that there exists a function satisfying the inequal-
ity. If f (f (x)) ≤ 0 for all x, then f is a decreasing function in view
of the inequalities f (x + y) ≥ f (x) + yf (f (x)) ≥ f (x) for any y ≤ 0.
Since f (0) > 0 ≥ f (f (x)), it implies f (x) > 0 for all x, which is a
contradiction. Hence there is a z such that f (f (z)) > 0. Then the in-
equality f (z + x) ≥ f (z) + xf (f (z)) shows that lim f (x) = +∞ and
x→∞
therefore f (f (x)) = +∞. In particular, there exist x, y > 0 such that
x→∞
x+1
f (x) ≥ 0, f (f (x)) > 1, y ≥ and f (f (x + y + 1)) ≥ 0. Then
f (f (x)) − 1
f (x + y) ≥ f (x) + yf (f (x)) ≥ x + y + 1 and hence
f (f (x + y)) ≥ f (x + y + 1) + (f (x + y) − (x + y + 1))f (f (x + y + 1)) ≥
≥ f (x + y + 1) ≥ f (x + y) + f (f (x + y)) ≥
≥ f (x) + yf (f (x)) + f (f (x + y)) > f (f (x + y)).
This contradiction completes the solution of the problem.
Problem 6. We prove that g(ϑ) = | sin ϑ|| sin(2ϑ)|1/2 attains its max-
√ k
imum value ( 23 )3/2 at points 2 3π (where k is a positive integer). This
can be seen by using derivatives or a classical bound like
√ q
1/2 2 4 √ 2
|g(ϑ)| = | sin ϑ|| sin(2ϑ)| = √ 4
| sin ϑ|.| sin ϑ|.| sin ϑ|.| 3 cos ϑ|
√ 3 √
2 3 sin2 ϑ + 3 cos2 ϑ  3 3/2
≤√ · = .
4
3 4 2
Hence
fn (ϑ)
=
fn (π/3)
g(ϑ).g(2ϑ)1/2 .g(4ϑ)3/4 · · · g(2n−1 ϑ)E sin(2n ϑ) 1−E/2
= ·
g(π/3).g(2π/3)1/2 .g(4π/3)3/4 · · · g(2n−1 π/3)E sin(2n π/3)
sin(2n ϑ) 1−E/2  1 1−E/2 2
≤ ≤ √ ≤ √ .
sin(2n π/3) 3/2 3
where E = 23 (1 − (− 12 )n ). This is exactly the bound we had to prove.
2.9. Solutions of Olympic 2002 113

2.9 Solutions of Olympic 2002

2.9.1 Day 1

Problem 1. First we show that the standard parabola with vertex V

contains point A if and only if the standard parabola with vertex s(A)
contains point s(V ).
Let A = (a, b) and V = (v, w). The equation of the standard parabola
with vertex V = (v, w) is y = (x − v)2 + w, so it contains point A if and
only if b = (a − v)2 + w. Similarly, the equation of the parabola with
vertex s(A) = (a, −b) is y = (x−a)2 −b; it contains point s(V ) = (v, −w)
if and only if −w = (v − a)2 − b. The two conditions are equivalent. Now
assume that the standard parabolas with vertices V1 and V2 , V1 and V3 , V2
and V3 intersect each other at points A3 , A2 , A1 , respectively. Then, by
the statement above, the standard parabolas with vertices s(A1 ) and
s(A2 ), S(A1 ) and s(A3 ), s(A2 ) and S(A3 ) intersect each other at points
V3 , V2 , V1 , respectively, because they contain these points.
Problem 2. Assume that there exists such a function. Since f 0 (x) =
f (f (x)) > 0, the function is strictly monotone increasing.
By the monotonity, f (x) > 0 implies f (f (x)) > f (0) for all x. Thus,
f (0) is a lower bound for f 0 (x), and for all x < 0 we have f (x) < f (0) +
x.f (0) = (1 + x)f (0). Hence, if x ≤ −1 then f (x) ≤ 0, contradicting the
property f (x) > 0.
So such function does not exist.
Problem 3. Let n be a positive integer and let
1
ak = n
, bk = 2k−n , for k = 1, 2, . . . , n.
k
Show that
a1 − b1 a2 − b2 an − b n
+ + ··· + = 0. (1)
1
2
n
Solution. Since k nk = n n−1
k−1 for all k ≥ 1, (1) is equivalent to

2n h 1 1 1 i 21 22 2n
+ + · · · + = + + · · · + . (2)
n n−1 n−1 n−1




0 1 n−1
1 2 n
2.9. Solutions of Olympic 2002 114

We prove (2) by induction. For n = 1, both sides are equal to 2.

Assume that (2) holds for some n. Let
2n h 1 1 1 i
xn = + n−1 + · · · + n−1
;
n n−1



0 1 n−1

then
n n−1
2n+1 X 1 2n  X 1 1  
xn+1 = n =

1+ n +

n

+1 =
n+1 k
n + 1 k k+1
k=0 k=0
n−1 n−k k+1 n−1
2n X
n + n 2n+1 n X
2 1 2n+1 2n+1
= n−1

+ = n−1 +

= xn + .
n+1 k
n+1 n k
n + 1 n + 1
k=0 k=0

This implies (2) for n + 1.

Problem 4. If for some n > m the equality pm = pn holds then Tp is
a finite set. Thus we can assume that all points p0 , p1 , . . . are distinct.
There is a convergent subsequence pnk and its limit q is in Tp . Since f is
continuous pnk +1 = f (pnk ) → f (q), so all, except for finitely many, points
pn are accumulation points of Tp . Hence we may assume that all of them
are accumulation points of Tp . Let d = sup{|pm − pn | : m, n ≥ 0}. Let δn
P∞ d
be positive numbers such that δn < . Let In be an interval of length
n=0 2
less than δn centered at pn such that there are there are infinitely many
n
k 0 s such that pk ∈
/ ∪ Ij , this can be done by induction. Let n0 = 0
j=0
nm
and nm+1 be the smallest integer k ¿ nm such that pk ∈
/ ∪ Ij . Since
j=0
Tp is closed the limit of the subsequence (pnm ) must be in Tp but it is
impossible because of the definition of In0 s, of course if the sequence (pnm )
is not convergent we may replace it with its convergent subsequence. The
proof is finished.
Remark. If Tp = {p1 , p2 , . . .} and each pn is an accumulation point
of Tp , then Tp is the countable union of nowhere dense sets (i.e. the
single-element sets {pn }). If T is closed then this contradicts the Baire
Category Theorem.
Problem 5.
2.9. Solutions of Olympic 2002 115

a. It does not exist. For each y the set {x : y = f (x)} is either empty
or consists of 1 point or is an interval. These sets are pairwise disjoint,
so there are at most count ably many of the third type.
b. Let f be such a map. Then for each value y of this map there is
an xo such that y = f (x) and f 0 (x) = 0, because an uncountable set
{x : y = f (x)} contains an accumulation point x0 and clearly f 0 (x0 ) = 0.
For every  > 0 and every x0 such that f 0 (x0 ) = a there exists an open
interval Ix0 such that if x ∈ Ix0 then |f 0 (x)| < . The union of all
these intervals Ix0 may be written as a union of pairwise disjoint open
intervals Jn . The image of each Jn is an interval (or a point) of length
<  length (Jn ) due to Lagrange Mean Value Theorem. Thus the image
of the interval [0, 1] may be covered with the intervals such that the sum
of their lengths is .1 = . This is not possible for  < 1.
Remarks. 1. The proof of part b is essentially the proof of the easy
part of A. Sard’s theorem about measure of the set of critical values of
a smooth map.
2. If only continuity is required, there exists such a function, e.g. the
first co-ordinate of the very well known Peano curve which is a continuous
map from an interval onto a square.
Problem 6.
Lemma 1. Let (an )n≥0 be a sequence of non-negative numbers such
that a2k − a2k+1 ≤ a2k , a2k+1 − a2k+2 ≤ ak ak+1 for any k ≥ 0 and
1 √
lim sup nan < . Then lim sup n an < 1.
4
Proof. Let cl = supn≥2l (n + 1)an for l ≥ 0. We will show that
cl+1 ≤ 4c2l . Indeed, for any integer n ≥ 2l+1 there exists an integer
k ≥ 2l such that n = 2k or n = 2k + 1. In the first case there is
2 c2l 4c2l 4c2l
a2k − a2k+1 ≤ ak ≤ ≤ − , whereas in the second
(k + 1)2 2k + 1 2k + 2
c2l 4c2l 4c2l
case there is a2k+1 −a2k+2 ≤ ak ak+1 ≤ ≤ − .
(k + 1)(k + 2) 2k + 2 2k + 3
4c2l
Hence a sequence (an − ) l+1 is non-decreasing and its terms
n + 1 n≥2
2.9. Solutions of Olympic 2002 116

4c2l
are non-positive since it converges to zero. Therefore an ≤ for n ≥
n + 1 −l
2l+1 , meaning that c2l+1 ≤ 4c2l . This implies that a sequence ((4cl )2 )l≥0
is nonincreasing and therefore bounded from above by some number
q ∈ (0, 1) since all its terms except finitely many are less than 1. Hence
l
cl ≤ q 2 for l large enough. For any n between 2l and 2l+1 there is
cl l √ √ √
an ≤ ≤ q 2 ≤ ( q)n yielding lim sup n an ≤ q < 1, yielding
n√ +1 √
lim sup an ≤ 1 < 1, which ends the proof.
n

Lemma 2. Let T be a linear map from Rn into itself. Assume that

1
lim sup n k T n+1 − T n k< . Then lim sup k T n+1 − T n k1/n < 1. In
4
particular T n converges in the operator norm and T is power bounded.
Proof. Put an =k T n+1 − T n k. Observe that

T k+m+1 − T k+m = (T k+m+2 − T k+m+1 ) − (T k+1 − T k )(T m+1 − T m )

implying that ak+m ≤ ak+m+1 + ak am . Therefore the sequence (am )m≥0

satisfies assumptions of Lemma 1 and the assertion of Proposition 1
follows.
Remarks. 1. The theorem proved above holds in the case of an
operator T which maps a normed space X into itself, X does not have
to be finite dimensional.
1
2. The constant in Lemma 1 cannot be replaced by any greater num-
4
1
ber since a sequence an = satisfies the inequality ak+m − ak+m+1 ≤
4n
ak am for any positive integers k and m whereas it does not have expo-
nential decay.
1
3. The constant in Lemma 2 cannot be replaced by any number
4
1
greater that . Consider an operator (T f )(x) = xf (x) on L2 ([0, 1]).
e
1
One can easily check that lim sup k T n+1 − T n k= , whereas T n does
e
not converge in the operator norm. The question whether in general
lim sup n k T n+1 − T n k< ∞ implies that T is power bounded remains
open.
2.9. Solutions of Olympic 2002 117

Remark. The problem was incorrectly stated during the competi-

1
tion: instead of the inequality k Ak − Ak−1 k≤ , the inequality
2002k
k k−1 1 
1 

k

1 k

k A −A k≤ was assumed. If A = 0 1 then A = 0 1 .
2002n 
0 

k k−1
Therefore A − A = 0 0 , so for sufficiently small  the condition
is satisfied although the sequence (k Ak k) is clearly unbounded.

2.9.2 Day 2

Problem 1. Adding the second row to the first one, then adding the
third row to the second one, ..., adding the nth row to the (n − l)th, the
determinant does not change and we have
2 −1 +1 ... ±1 ∓1 1 1 0 0 ... 0 0
−1 2 −1 ... ∓1 ±1 0 1 1 0 ... 0 0
+1 −1 2 ... ±1 ∓1 0 0 1 1 ... 0 0
det(A) = .. .. .. ... .. .. = .. .. .. .. . . . ... .. .
. . . . . . . . . .
∓1 ±1 ∓1 ... 2 −1 0 0 0 0 ... 1 1
±1 ∓1 ±1 ... −1 2 ±1 ∓1 ±1 ∓1 . . . −1 2
Now subtract the first column from the second, then subtract the result-
ing second column from the third, ..., and at last, subtract the (n − 1)th
column from the nth column. This way we have
1 0 0 ... 0 0
0 1 0 ... 0 0
det(A) = ... ..
.
..
. ... ..
.
..
. = n + 1.
0 0 0 ... 1 0
0 0 0 ... 0 n+1
Problem 2. For each pair of students, consider the set of those problems
which was not solved by them. There exist 200


2 = 19900 sets; we have
to prove that at least one set is empty.
For each problem, there are at most 80 students who did not solve it.
From these students at most 80


2 = 3160 pairs can be selected, so the
problem can belong to at most 3160 sets. The 6 problems together can
belong to at most 6.3160 = 18960 sets.
Hence, at least 19900 − 18960 = 940 sets must be empty.
2.9. Solutions of Olympic 2002 118

an
Problem 3. We prove by induction on n that and bn e are integers, we
e
prove this for n = 0 as well. (For n = 0, the term 00 in the definition of
the sequences must be replaced by 1.) From the power series of ex , an =
1
e1 = e and bn = e−1 = .
e
Suppose that for some n ≥ 0, a0 , a1 , . . . , an and b0 , b1 , . . . , bn are all
1
multipliers of e and , respectively. Then, by the binomial theorem,
e
n ∞ ∞ n  
X (k + 1)n+1 X (k + 1)n X X n k m
an+1 = = = =
(k + 1)! k! m k!
k=0 k=0 k=0 m=0
n 
X n Xk  ∞ m n
X n  
= = am
m=0
m k! m=0
m
k=0

and similarly
n ∞
X + 1)n+1
k+1 (k
X
k (k + 1)
n
bn+1 = (−1) =− (−1)
(k + 1)! k!
k=0 k=0
∞ n   m
X X n k
=− (−1)k
m=0
m k!
k=0
n  X ∞ m n  
X n kk
X n
=− (−1) =− bm .
m=0
m k! m=0
m
k=0

The numbers an+1 and bn+1 are expressed as linear combinations of the
previous elements with integer coefficients which finishes the proof.
Problem 4. We can assume OA = OB = OC = 1. Intersect the unit
sphere with center O with the angle domains AOB, BOC and COA; the
1 1 1
intersections are ”slices” and their areas are γ, α and β, respectively.
2 2 2
Now project the slices AOC and COB to the plane OAB. Denote by
C 0 the projection of vertex C, and denote by A0 and B 0 the reflections of
vertices A and B with center 0, respectively. By the projection, OC 0 < 1.
The projections of arcs AC and BC are segments of ellipses with long
axes AA0 and BB 0 , respectively. (The ellipses can be degenerate if σ or
τ is right angle.) The two ellipses intersect each other in 4 points; both
half ellipses connecting A and A0 intersect both half ellipses connecting
2.9. Solutions of Olympic 2002 119

B and B 0 . There exist no more intersection, because two different conics

cannot have more than 4 common points.
The signed areas of the projections of slices AOC and COB are
1 1
α. cos τ and β. cos σ respectively. The statement says thet the sum of
2 2
these signed areas is less than the area of slice BOA.
There are three significantly different cases with respect to the signs
of cos σ and cos τ (see Figure). If both signs are positive (case (a)),
then the projections of slices OAC and OBC are subsets of slice OBC
without common interior point, and they do not cover the whole slice
OBC; this implies the statement. In cases (b) and (c) where at least
one of the signs is negative, projections with positive sign are subsets of
the slice OBC, so the statement is obvious again.
−1
Problem 5. The direction ⇐ is trivial, since if A = SS , then
−1
AA = SS = In .
For the direction ⇒, we must prove that there exists an invertible
matrix S such that AS = S.
Let w be an arbitrary complex number which is not 0. Choosing
S = wA + wIn we have AS = A(wA + wIn ) = wIn + wA = S. If S is
1 w  w
singular, then S = A− In is singular as well, so is an eigenvalue
w w w
w
of A. Since A has finitely many eigenvalues and can be any complex
w
number on the unit circle, there exist such w that S is invertible.
Problem 6. Let g(x) = f (x)−f (x1 )− < ∇f (x1 ), x−x1 >. It is obvious
that g has the same properties. Moreover, g(x1 ) = ∇g(x1 ) = 0 and, due
to convexity, g has 0 as the absolute minimum at x1 . Next we prove that
1
g(x2 ) ≥ k ∇g(x2 ) k2 . (2)
2L
2.10. Solutions of Olympic 2003 120

1
Let y0 = x2 − k ∇g(x2 ) k and y(t) = y0 + t(x2 − y0 ). Then
L
Z1
g(x2 ) = g(y0 ) + < ∇g(y(t)), x2 − y0 > dt =
0
Z1
= g(y0 )+ < ∇g(x2 ), x2 − y0 > − < ∇g(x2 ) − ∇g(y(t)), x2 − y0 > dt ≥
0
Z1
1
≥0+ k ∇g(x2 ) k2 − k ∇g(x2 ) − ∇g(y(t)) k . k x2 − y0 k dt ≥
L
0
Z1
1
≥ k ∇g(x2 ) k2 − k x2 − y0 k L k x2 − g(y) k dt =
L
0
Z1
1 1
= k ∇g(x2 ) k2 −L k x2 − y0 k2 tdt = k ∇g(x2 ) k2 .
L 2L
0
Substituting the definition of g into (2), we obtain
1
f (x2 ) − f (x1 )− < ∇f (x1 ), x2 − x1 >≥ k ∇f (x2 ) − ∇f (x1 ) k2 ,
2L
2
k ∇f (x2 ) − ∇f (x1 ) k ≤ 2L < ∇f (x1 ), x1 − x2 > +2L(f (x2 ) − f (x1 )).
(3)
Exchanging variables x1 and x2 , we have
k ∇f (x2 )−∇f (x1 ) k2 ≤ 2L < ∇f (x2 ), x2 −x1 > +2L(f (x1 )−f (x2 )). (4)
The statement (1) is the average of (3) and (4).

2.10 Solutions of Olympic 2003

2.10.1 Day 1

Problem 1.
an 3
a) Let bn =  n−1 . Then an+1 > an is equivalent to bn+1 > bn , thus
3 2
2
the sequence (bn ) is strictly increasing. Each increasing sequence has a
finite limit or tends to infinity.
2.10. Solutions of Olympic 2003 121

b) For all α > 1 there exists a sequence 1 = b1 < b2 < . . . which con-
 n−1
verges to α. Choosing an = 32 bn , we obtain the required sequence
(an ).
Problem 2. Let S = a1 + a2 + · · · + a51 . Then b1 + b2 + · · · + b51 = 50S.
Since b1 , b2 , . . . , b51 is a permutation of a1 , a2 , . . . , a51 , we get 50S = S,
so 49S = 0. Assume that the characteristic of the field is not equal
to 7. Then 49S = 0 implies that S = 0. Therefore bi = −ai for i =
1, 2, . . . , 51. On the other hand, bi = aϕ(i) , where ϕ ∈ S51 . Therefore, if
the characteristic is not 2, the sequence a1 , a2 , . . . , a51 can be partitioned
into pairs {ai , aϕ(i) } of additive inverses. But this is impossible, since 51
is an odd number. It follows that the characteristic of the field is 7 or 2.
The characteristic can be either 2 or 7. For the case of 7, x1 = · · · =
x51 = 1 is a possible choice. For the case of 2, any elements can be
chosen such that S = 0, since then bi = −ai = ai .
Problem 3. The minimal polynomial of A is a divisor of 3x3 − x2 −
x − 1. This polynomial has three different roots. This implies that
A is diagonalizable: A = C −l DC where D is a diagonal matrix. The
eigenvalues of the matrices A and D are all roots of polynomial 3x3 −x2 −
x − 1. One of the three roots is 1, the remaining two roots have smaller
absolute value than 1. Hence, the diagonal elements of Dk , which are
the kth powers of the eigenvalues, tend to either 0 or 1 and the limit
M = lim Dk is idempotent. Then lim Ak = C −1 M C is idempotent as
well.
Problem 4. Clearly a and b must be different since A and B are disjoint.
Let {a, b} be a solution and consider the sets A, B such that a.A =
b.B. Denoting d = (a, b) the greatest common divisor of a and b, we have
a = d.a1 , b = d.b1 , (a1 , b1 ) = 1 and a1 A = b1 B. Thus {a1 , b1 } is a solution
and it is enough to determine the solutions {a, b} with (a, b) = 1.
If 1 ∈ A then a ∈ a.A = b.B, thus b must be a divisor of a. Similarly,
if 1 ∈ B, then a is a divisor of b. Therefore, in all solutions, one of
numbers a, b is a divisor of the other one.
2.10. Solutions of Olympic 2003 122

Now we prove that if n ≥ 2, then (1, n) is a solution. For each positive

integer k, let f (k) be the largest non-negative integer for which nf (k) |k.
Then let A = {k : f (k) is odd} and B = {k : f (k) is even}. This is a
decomposition of all positive integers such that A = n.B.
Problem 5.
B. We shall prove in two different ways that lim fn (x) = g(0) for
n→∞
every x ∈ (0, 1]. (The second one is more lengthy but it tells us how to
calculate fn directly from g.)
Proof I. First we prove our claim for non-decreasing g. In this case, by
induction, one can easily see that
1. each fn is non-decrasing as well, and
2. g(x) = f0 (x) ≥ f1 (x) ≥ f2 (x) ≥ · · · ≥ g(0) (x ∈ (0, 1]).
Then (2) implies that there exists

h(x) = lim fn (x) (x ∈ (0, 1]).

n→∞

Clearly h is non-decreasing and g(0) ≤ h(x) ≤ fn (x) for any x ∈

(0, 1], n = 0, 1, 2, . . .. Therefore to show that h(x) = g(0) for any
x ∈ (0, 1], it is enough to prove that h(1) cannot be greater than g(0).
Suppose that h(1) > g(0). Then there exists a 0 < δ < 1 such that
h(1) > g(δ). Using the definition, (2) and (1) we get
Z1 Zδ Z1
fn+1 (1) = fn (t)dt ≤ g(t)dt + fn (t)dt ≤ δg(δ) + (1 − δ)fn (1).
0 0 δ

Hence

fn (1) − fn+1 (1) ≥ δ(fn (1) − g(δ)) ≥ δ(h(1) − g(δ)) > 0,

so fn (1) → −∞, which is a contradiction.

Similarly, we can prove our claim for non-increasing continuous func-
tions as well.
Now suppose that g is an arbitrary continuous function on [0, 1]. Let

M (x) = sup g(t), m(x) = inf g(t) (x ∈ [0, 1])

t∈[0,x] t∈[0,x]
2.10. Solutions of Olympic 2003 123

Then on [0, 1] m is non-increasing, M is non-decreasing, both are con-

tinuous, m(x) ≤ g(x) ≤ M (x) and M (0) = m(0) = g(0). Define the
sequences of functions Mn (x) and mn (x) in the same way as fn is de-
fined but starting with M0 = M and m0 = m.
Then one can easily see by induction that mn (x) ≤ fn (x) ≤ Mn (x).
By the first part of the proof, limn mn (x) = m(0) = g(0) = M (0) =
limn Mn (x) for any x ∈ (0, 1]. Therefore we must have limn fn (x) = g(0).
Proof II. To make the notation clearer we shall denote the variable of
fj by xj . By definition (and Fubini theorem) we get that
x
Zn+1 Zxn x
Zn−1 Zx2 Zx1
1 1 1 1
fn+1 (xn+1 ) = ··· g(x0 )dx0 dx1 . . . dxn
xn+1 xn xn−1 x1
Z0 0
Z 0 0 0
1 dx0 dx1 . . . dxn
= g(x0 )
xn+1 x1 . . . xn
0≤x0 ≤x1 ≤···≤xn ≤xn+1
x
Zn+1 Z Z
1  dx1 . . . dxn 
= g(x0 ) dx0 .
xn+1 x1 . . . xn
0 x0 ≤x1 ≤···≤xn ≤xn+1

Therefore with the notation

Z Z
dx1 . . . dxn
hn (a, b) =
x1 . . . xn
a≤x1 ≤···≤xn ≤b
and x = xn+1 , t = x0 we have
Zx
1
fn+1 (x) = g(t)hn (t, x)dt.
x
0
Using that hn (a, b) is the same for any permutation of x1 , . . . , xn and the
fact that the integral is 0 on any hyperplanes (xi = xj ) we get that
Z Z Zb Zb
dx1 . . . dxn dx1 . . . dxn
n!hn (a, b) = = ···
x1 . . . xn x1 . . . xn
a≤x1 ,...,xn ≤b a a

 Zb dx n  b n
= = log .
x a
a
2.10. Solutions of Olympic 2003 124

Therefore
Zx
1 (log (x/t))n
fn+1 (x) = g(t) dt.
x n!
0
Note that if g is constant then the definition gives fn = g. This implies
on one hand that we must have
Zx
1 (log (x/t))n
dt = 1
x n!
0

and on the other hand that, by replacing g by g − g(0), we can suppose

that g(0) = 0.
Let x ∈ (0, 1] and  > 0 be fixed. By continuity there exists a 0 <
δ < x and an M such that |g(t)| <  on [0, δ] and |g(t)| ≤ M on [0, 1] .
Since
(log (x/δ))n
lim =0
n→∞ n!
log (x/δ)n
there exists an n0 sucht that <  whenever n ≥ n0 . Then, for
n!
any n ≥ n0 , we have
Zx
1 (log (x/t))n
|fn+1 (x)| ≤ |g(t)| dt
x n!
0
Zδ n Zx
1 (log (x/t)) 1 (log (x/δ))n
≤  dt + |g(t)| dt
x n! x n!
0 δ
Zx Zx
1 (log (x/t))n 1
≤  dt + M dt
x n! x
0 δ
≤  + M .

Therefore limn f (x) = 0 = g(0).

Problem 6. The polynomial f is a product of linear and quadratic
factors, f (z) = i (ki z + li ). j (pj z 2 + qj z + rj ), with ki , ki , pj , qj ∈ R.
Q Q

Since all roots are in the left half-plane, for each i, ki and li are of the
same sign, and for each j, pj , qj , rj are of the same sign, too. Hence,
2.10. Solutions of Olympic 2003 125

multiplying f by −1 if necessary, the roots of f don’t change and f

becomes the polynomial with all positive coefficients.
For the simplicity, we extend the sequence of coefficients by an+1 =
an+2 = · · · = 0 and a−1 = a−2 = · · · = 0 and prove the same statement
for −1 ≤ k ≤ n − 2 by induction.
For n ≤ 2 the statement is obvious: ak+1 and ak+2 are positive and at
least one of ak−1 and ak+3 is 0; hence, ak+1 ak+2 > ak ak+3 = 0.
Now assume that n ≥ 3 and the statement is true for all smaller
values of n. Take a divisor of f (z) which has the form z 2 + pz + q where
p and q are positive real numbers. (Such a divisor can be obtained from
a conjugate pair of roots or two real roots.) Then we can write

f (z) = (z 2 + pz + q)(bn−2 z n−2 + · · · + b1 z + b0 ) = (z 2 + pz + q)g(x). (1)

The roots polynomial g(z) are in the left half-plane, so we have bk+1 bk+2 <
bk bk+3 for all −1 ≤ k ≤ n − 4. Defining bn−1 = bn = · · · = 0 and
b−1 = b−2 = · · · = 0 as well, we also have bk+1 bk+2 ≤ bk bk+3 for all
integer k.
Now we prove ak+1 ak+2 > ak ak+3 . If k = −1 or k = n − 2 then
this is obvious since ak+1 ak+2 is positive and ak ak+3 = 0. Thus, assume
0 ≤ k ≤ n − 3. By an easy computation,

ak+1 ak+2 − ak ak+3 =

= (qbk+1 + pbk + bk−1 )(qbk+2 + pbk+1 + bk )−
−(qbk + pbk−1 + bk−2 )(qbk+3 + pbk+2 + bk+1 ) =
= (bk−1 bk − bk−2 bk+1 ) + p(b2k − bk−2 bk+2 ) + q(bk−1 bk+2 − bk−2 bk+3 )+
+p2 (bk bk+1 − bk−1 bk+2 ) + q 2 (bk+1 bk+2 − bk bk+3 ) + pq(b2k+1 − bk−1 bk+3 ).

We prove that all the six terms are non-negative and at least one is
positive. Term p2 (bk bk+1 − bk−1 bk+2 is positive since 0 ≤ k ≤ n − 3. Also
terms bk−1 bk − bk−2 bk+1 and q 2 (bk+1 bk+2 − bk bk+3 ) are non-negative by the
induction hypothesis.
2.10. Solutions of Olympic 2003 126

To check the sign of p(b2k − bk−2 bk+2 ) consider

bk−1 (b2k − bk−2 bk+2 )

= bk−2 (bk bk+1 − bk−1 bk+2 ) + bk (bk−1 bk − bk−2 bk+1 ) ≥ 0.

If bk−1 > 0 we can divide by it to obtain b2k − bk−2 bk+2 ≥ 0. Otherwise,
if bk−1 = 0, either bk−2 = 0 or bk+2 = 0 and thus b2k − bk−2 bk+2 = b2k ≥ 0.
Therefore, p(b2k − bk−2 bk+2 ) ≥ 0 for all k. Similarly, pq(b2k+1 − bk−1 bk+3 ) ≥
0.
The sign of q(bk−1 bk+2 − bk−2 bk+3 ) can be checked in a similar way.
Consider

bk+1 (bk−1 bk+2 −bk−2 bk+3 ) = bk−1 (bk+1 bk+2 −bk bk+3 )+bk+3 (bk−1 bk −bk−2 bk+1 ) ≥ 0.

If bk+1 > 0, we can divide by it. Otherwise either bk−2 = 0 or bk+3 = 0.

In all cases, we obtain bk−1 bk+2 − bk−2 bk+3 ≥ 0.
Now the signs of all terms are checked and the proof is complete.

2.10.2 Day 2
sin t
Problem 1. We use the fact that is decreasing in the interval
t
sin t
(0, π) and lim = 1. For all x ∈ (0, π2 ) and t ∈ [x, 2x] we have
t→0+0 t
sin 2x sin t
x< < 1, thus
2 t
 sin 2x m Z2x tm Z2x
sinm t
Z2x
tm
< dt < dt,
2x tn tn tn
x x x
Z2x Z2
tm
dt = xm−n+1 um−n du.
tn
x 1
 sin 2x m
The factor tends to 1. If m − n + 1 < 0, the limit of xm−n+1 is
2x
m−n+1
R2
infinity; if m−n+1 > 0 then 0. If m−n+1 = 0 then x um−n du =
1
2.10. Solutions of Olympic 2003 127

ln 2. Hence,

Z2x 0, m≥n
sinm t
lim dt = ln 2, n − m = 1
x→0+0 tn +∞, n − m > 1.
x

Problem 3. Let b0 ∈ / A (otherwise b0 ∈ A ⊂ B, % = inf |a − b0 |. The

a∈A
intersection of the ball of radius %+1 with centre b0 with set A is compact
and there exists a0 ∈ A : |a0 − b0 | = %.
Denote by Br (a) = {x ∈ Rn : |x − a| ≤ r} and ∂Br (a) = {x ∈ Rn :
|x−a| = r} the ball and the sphere of center a and radius r, respectively.
If a0 is not the unique nearest point then for any point a on the
open line segment (a0 , b0 ) we have B|a−a0 | (a) ⊂ B% (b0 ) and ∂B|a−a0 | (a) ∩
∂B% (b0 ) = {a0 }, therefore (a0 , b0 ) ⊂ B and b0 is an accumulation point
of set B.
Problem 4. The condition (i) of the problem allows us to define a
(well-defined) operation * on the set S given by

a ∗ b = c if and only if {a, b, c} ∈ F, where a 6= b.

We note that this operation is still not defined completely (we need to
define a ∗ a), but nevertheless let us investigate its features. At first,
due to (i), for a 6= b the operation obviously satisfies the following three
conditions:
a) a 6= a ∗ b 6= b;
b) a ∗ b = b ∗ a;
c) a ∗ (a ∗ b) = b.
What does the condition (ii) give? It claims that
e’) x∗(a∗c) = x∗y = z = b∗c = (x∗a)∗c for any three different x, a, c,
i.e. that the operation is associative if the arguments are different. Now
we can complete the definition of *. In order to save associativity for
nondifferent arguments, i.e. to make b = a ∗ (a ∗ b) = (a ∗ a) ∗ b hold, we
will add to S an extra element, call it 0, and define
d) a ∗ a = 0 and a ∗ 0 = 0 ∗ a = a.
2.10. Solutions of Olympic 2003 128

Now it is easy to check that, for any a, b, c ∈ S ∪ {0}, (a),(b),(c) and

(d), still hold, and
e) a ∗ b ∗ c := (a ∗ b) ∗ c = a ∗ (b ∗ c).
We have thus obtained that (S ∪ {0}, ∗) has the structure of a finite
Abelian group, whose elements are all of order two. Since the order of
every such group is a power of 2, we conclude that |S ∪{0}| = n+1 = 2m
and n = 2m − 1 for some integer m ≥ 1.
Given n = 2m − 1, according to what we have proven till now, we will
construct a family of three-element subsets of S satisfying (i) and (ii).
Let us define the operation * in the following manner:
if a = a0 + 2a1 + · · · + 2m−1 am−1 and b = b0 + 2b1 + · · · + 2m−1 bm−1 ,
where ai , bi are either 0 or 1, we put a ∗ b = |a0 − b0 | + 2|a1 − b1 | + · · · +
2m−1 |am−1 − bm−1 |.
It is simple to check that this * satisfies (a),(b),(c) and (e’). Therefore,
if we include in F all possible triples a, b, a ∗ b, the condition (i) follows
from (a),(b) and (c), whereas the condition (ii) follows from (e’)
The answer is: n = 2m − 1.
Problem 5.
a) Let ϕ : Q → N be a bijection. Define g(x) = max{|f (s, t)| : s, t ∈
Q, ϕ(s) ≤ ϕ(x), ϕ(t) ≤ ϕ(x)}. We have f (x, y) ≤ max{g(x), g(y)} ≤
g(x) + g(y).
1
b) We shall show that the function defined by f (x, y) = for
|x − y|
x 6= y and f (x, x) = 0 satisfies the problem. If, by contradiction there
1
exists a function g as above, it results, that g(y) ≥ − f x) for
|x − y|
x, y ∈ R, x 6= y; one obtains that for each x ∈ R, lim g(y) = ∞. We
y→x
show, that there exists no function g having an infinite limit at each
point of a bounded and closed interval [a, b]. For each k ∈ N+ denote
Ak = {x ∈ [a, b] : |g(x)| ≤ k}.
∞
We have obviously [a, b] = ∪ Ak . The set [a, b] is uncountable, so
k=1
at least one of the sets Ak is infinite (in fact uncountable). This set
2.10. Solutions of Olympic 2003 129

Ak being infinite, there exists a sequence in Ak having distinct terms.

This sequence will contain a convergent subsequence (xn )n∈N convergent
to a point x ∈ [a, b]. But lim g(y) = ∞ implies that g(xn) → ∞, a
y→x
contradiction because |g(xn)| ≤ k, ∀n ∈ N.
Second solution for part (b). Let S be the set of all sequences of real
2
numbers. The cardinality of S is |S| = |R|N0 = 2N0 = 2N0 = |R|. Thus,
there exists a bijection h : R → S. Now define the function f in the
following way. For any real x and positive integer n, let f (x, n) be the
nth element of sequence h(x). If y is not a positive integer then let
f (x, y) = 0. We prove that this function has the required property.
Let g be an arbitrary R → R function. We show that there exist
real numbers x, y such that f (x, y) > g(x) + g(y). Consider the sequence
(n+g(n))∞ ∞
n=1 . This sequence is an element of S, thus (n+g(n))n=1 = h(x)
for a certain real x. Then for an arbitrary positive integer n, f (x, n) is
the nth element, f (x, n) = n + g(n). Choosing n such that n > g(x), we
obtain f (x, n) = n + g(n) > g(x) + g(n).
∞
an xn . By
P
Problem 6. Consider the generating function f (x) =
n=0
induction 0 < an ≤ 1, thus this series is absolutely convergent for |x| <
1, f (0) = 1 and the function is positive in the interval [0, 1). The goal is
to compute f ( 21 ).
By the recurrence formula,
∞ ∞ X
n
0
X
n
X ak
f (x) = (n + 1)an+1 x = xn =
n=0 n=0 k=0
n−k+2
∞ ∞ ∞
X
k
X xn−k X xm
= ak x = f (x) .
n−k+2 m=0
m + 2
k=0 n=k
2.11. Solutions of Olympic 2004 130

Then
Zx ∞
f0 X xm+1
ln f (x) = ln f (x) − ln f (0) = = =
f m=0
(m + 1)(m + 2)
0
∞  ∞
X xm+1 xm+1   1  X xm+1  1 1
= − =1+ 1− =1+ 1− ln ,
m=1
(m + 1 (m + 2) x m=1
(m + 1) x 1 − x
1
ln f = 1 − ln 2,
x
1 e
and thus f = .
2 2

2.11 Solutions of Olympic 2004

2.11.1 Day 1
 
1
Problem 1. Let Sn = S ∩ n, ∞ for any integer n > 0. It follows from
the inequality that |Sn | < n. Similarly, if we define S−n = S ∩(−∞, − n1 ),
then |S−n | < n. Any nonzero x ∈ S is an element of some Sn or S−n ,
1 1
because there exists an n such that x > or x < − . Then S ⊂
n n
{0} ∪ ∪ (Sn ∪ S−n ), S is a countable union of finite sets, and hence
n∈N
countable.
Problem 2. Put Pn (x) = P (P (. . . (P (x)) . . .)). As P1 (x) ≥ −1, for
| {z }
n
each x ∈ R, it must be that Pn+1 (x) = P1 (Pn (x)) ≥ −1, for each n ∈ N
and each x ∈ R. Therefore the equation Pn (x) = a, where a < −1 has
no real solutions.
Let us prove that the equation Pn (x) = a, where a > 0, has exactly
two distinct real solutions. To this end we use mathematical induction by
n. If n = 1 the assertion follows directly. Assuming that the assertion
holds for a n ∈ N we prove that it must also hold for n + 1. Since
Pn+1 (x) = a is equivalent to P1 (Pn (x)) = a, we conclude that Pn (x) =
√ √ √ √
a + 1 or Pn (x) = − a + 1. The equation Pn (x) = a + 1, as a + 1 >
1, has exactly two distinct real solutions by the inductive hypothesis,
√
while the equation Pn (x) = − a + 1 has no real solutions (because
2.11. Solutions of Olympic 2004 131
√
− a + 1 < −1). Hence the equation Pn+1 (x) = a, has exactly two
distinct real solutions.
Let us prove now that the equation Pn (x) = 0 has exactly n + 1
distinct real solutions. Again we use mathematical induction. If n = 1
the solutions are x = ±1, and if n = 2 the solutions are x = 0 and
√
x = ± 2, so in both cases the number of solutions is equal to n + 1.
Suppose that the assertion holds for some n ∈ N . Note that Pn+2 (x) =
P2 (Pn (x)) = Pn2 (x)(Pn2 (x) − 2), so the set of all real solutions of the
equation Pn+2 = 0 is exactly the union of the sets of all real solutions
√ √
of the equations Pn (x) = 0, Pn (x) = 2 and Pn (x) = − 2. By the
inductive hypothesis the equation Pn (x) = 0 has exactly n + 1 distinct
√ √
real solutions, while the equations Pn (x) = 2 and Pn (x) = − 2 have
two and no distinct real solutions, respectively. Hence, the sets above
being pairwise disjoint, the equation Pn+2 (x) = 0 has exactly n + 3
distinct real solutions. Thus we have proved that, for each n ∈ N , the
equation Pn (x) = 0 has exactly n+1 distinct real solutions, so the answer
to the question posed in this problem is 2005.
Problem 3.
a) Equivalently, we consider the set

Y = {y = (y1 , y2 , . . . , yn )|0 ≤ y1 , y2 , . . . , yn ≤ 1, y1 +y2 +· · ·+yn = 1} ⊂ Rn

and the image f (Y ) of Y under

f (y) = arcsin y+ arcsin y2 + · · · + arcsin yn .

Note that f (Y ) = Sn . Since Y is a connected subspace of Rn and f is

a continuous function, the image f (Y ) is also connected, and we know
that the only connected subspaces of R are intervals. Thus S n is an
interval.
b) We prove that
1 π
n arcsin ≤ x1 + x2 + · · · + xn ≤ .
n 2
2.11. Solutions of Olympic 2004 132

π
Since the graph of sin x is concave down for x ∈ [0, ], the chord joining
π 2
the points (0, 0) and ( , 1) lies below the graph. Hence
2
2x π
≤ sin x for all x ∈ [0, ]
π 2
and we can deduce the right-hand side of the claim:
2
(x1 + x2 + · · · + xn ) ≤ sin x1 + sin x2 + · · · + sin xn = 1.
π
π
The value 1 can be reached choosing x1 = and x2 = · · · = xn = 0.
2
The left-hand side follows immediately from Jensen’s inequality, since
π x1 + x2 + · · · + xn π
sin x is concave down for x ∈ [0, ] and 0 ≤ <
2 n 2
1 sin x1 + sin x2 + · · · + sin xn x1 + x2 + · · · + xn
= ≤ sin .
n n n
1
Equality holds if x1 = · · · = xn = arcsin .
n
Now we have computed the minimum and maximum of interval Sn ;
1 π π
we can conclude that Sn = [n arcsin , ]. Thus ln = − n and
n 2 2
π arcsin(1/n) π
− lim
lim ln = = − 1.
n→∞ 2 n→∞ 1/n 2

−1, if X is white
Problem 4. Define f : M → {−1, 1}, f (X) = .
P 1, if X is black
The given condition becomes f (X) = 0 for any sphere S which passes
X∈S
through at least 4 points of M . For any 3 given points A, B, C in M ,
denote by S(A, B, C) the set of all spheres which pass through A, B, C
and at least one other point of M and by |S(A, B, C)| the number of
P P
these spheres. Also, denote by the sum f (X).
X∈M
We have
X X X
0= f (X) = (|S(A, B, C)| − 1)(f (A) + f (B) + f (C)) +
S∈S(A,B,C) X∈S
(1)
2.11. Solutions of Olympic 2004 133

since the values of A, B, C appear |S(A, B, C)| times each and the other
values appear only once.
If there are 3 points A, B, C such that |S(A, B, C)| = 1, the proof is
finished.
If |S(A, B, C)| > 1 for any distinct points A, B, C in M , we will prove
P
at first that = 0.
P
Assume that > 0. From (1) it follows thatf (A) + f (B) + f (C) < 0
and summing by all n3 possible choices of (A, B, C) we obtain that


n P

P
3 < 0, which means < 0 (contradicts the starting assumption).
P
The same reasoning is applied when assuming < 0.
P
Now, from = 0 and (1), it follows that f (A) + f (B) + f (C) = 0
for any distinct points A, B, C in M . Taking another point D ∈ M , the
following equalities take place

f (A) + f (B) + f (C) = 0

f (A) + f (B) + f (D) = 0
f (A) + f (C) + f (D) = 0
f (B) + f (C) + f (D) = 0

which easily leads to f (A) = f (B) = f (C) = f (D) = 0, which contra-

dicts the definition of f .
Problem 5. We prove a more general statement:
Lemma. Let k, l ≥ 2, let X be a set of k+l−4


k−2 real numbers. Then
either X contains an increasing sequence {xi }ki=1 ⊆ X of length k and

|xi+1 − x1 | ≥ 2|xi − x1 | ∀i = 2, . . . , k − 1,

or X contains a decreasing sequence {xi }li=1 ⊆ X of length l and

|xi+1 − x1 | ≥ 2|xi − x1 | ∀i = 2, . . . , l − 1.

Proof of the lemma. We use induction on k + l. In case k = 2 or l = 2

the lemma is obviously true.
2.11. Solutions of Olympic 2004 134

Now let us make the induction step. Let m be the minimal element
of X, M be its maximal element. Let
m+M m+M
Xm = {x ∈ X : x ≤ }, XM = {x ∈ X : x > }.
2 2
(k−1)+l−4
Since k+l−4 k+(l−1)−4




k−2 = k−2 + (k−1)−2
, we can see that either
   
(k − 1) + l − 4 k + (l − 1) − 4
|Xm | ≥ + 1, or |XM | ≥ + 1.
(k − 1) − 2 k−2
In the first case we apply the inductive assumption to Xm and either
obtain a decreasing sequence of length l with the required properties
(in this case the inductive step is made), or obtain an increasing se-
quence {xi }k−1 i=1 ⊆ Xm of length k − 1. Then we note that the sequence
{x1 , x2 , . . . , xk−1 , M } ⊆ X has length k and all the required properties.
In the case |XM | ≥ k+(l−1)−4


k−2 the inductive step is made in a similar
way. Thus the lemma is proved.
The reader may check that the number k+l−4


k−2 + 1 cannot be smaller
in the lemma.
Problem 6. It is clear that the left hand side is well defined and inde-
pendent of the order of summation, because we have a sum of the type
P −4
n , and the branches of the logarithms do not matter because all
branches are taken. It is easy to check that the convergence is locally
uniform on C\{0, 1}; therefore, f is a holomorphic function on the com-
plex plane, except possibly for isolated singularities at 0 and 1. (We
omit the detailed estimates here.)
The function log has its only (simple) zero at z = 1, so f has a
quadruple pole at z = 1.
Now we investigate the behavior near infinity. We have Re(log (z)) =
2.11. Solutions of Olympic 2004 135

log |z|, hence (with c := log |z|)

X X X
| (log z)−4 | ≤ |log z|−4 = (log |z| + 2πin)−4 + O(1)
Z∞
= (x + 2πix)−4 dx + O(1)
−∞
Z∞
= c−4 (1 + 2πix/c)−4 dx + O(1)
−∞
Z∞
= c−3 (1 + 2πit)−4 dt + O(1)
−∞
≤ α(log |z|)−3
for a universal constant α. Therefore, the infinite sum tends to 0 as
|z| → ∞. In particular, the isolated singularity at ∞ is not essential,
but rather has (at least a single) zero at ∞.
The remaining singularity is at z = 0. It is readily verified that
f (1/z) = f (z) (because log (1/z) = −log (z)); this implies that f has a
zero at z = 0.
We conclude that the infinite sum is holomorphic on C with at most
one pole and without an essential singularity at ∞, so it is a rational
function, i.e. we can write f (z) = P (z)/Q(z) for some polynomials P
and Q which we may as well assume coprime. This solves the first part.
Since j has a quadruple pole at z = 1 and no other poles, we have
Q(z) = (z − 1)4 up to a constant factor which we can as well set equal to
1, and this determines P uniquely. Since f (z) → 0 as z → ∞, the degree
of P is at most 3, and since P (0) = 0, it follows that P (z) = z(az 2 +bz+c)
for yet undetermined complex constants a, b, c.
There are a number of ways to compute the coefficients a, b, c, which
turn out to be a = c = 61 , b = 23 . Since f (z) = f ( zl ), it follows easily that
a = c. Moreover, the fact limz→1 (z − 1)4 f (z) = 1 implies a + b + c = 1
(this fact follows from the observation that at z = 1, all summands can-
cel pairwise, except the principal branch which contributes a quadruple
2.11. Solutions of Olympic 2004 136

pole). Finally, we can calculate

X X
f (−1) = π −4 n−4 = 2π −4 n−4
nodd n≥1odd

−4
X
−4
X
−4
 1
= 2π n − n = .
n≥1 n≥1even
48

This implies a − b + c = − 31 . These three equations easily yield a, b, c.

Moreover, the function f satisfies f (z)+f (−z) = 16f (z 2 ): this follows
because the branches of log (z 2 ) = log ((−z)2 ) are the numbers 2log (z)
and 2log (−z). This observation supplies the two equations b = 4a and
a = c, which can be used instead of some of the considerations above.
P 1
Another way is to compute g(z) = first. In the same
(log z)2
dz
way, g(z) = . The unknown coefficient d can be computed from
(z − 1)2
lim(z − 1)2 g(z) = 1; it is d = 1. Then the exponent 2 in the denominator
z→1
can be increased by taking derivatives (see Solution 2). Similarly, one
can start with exponent 3 directly.
A more straightforward, though tedious way to find the constants is
computing the first four terms of the Laurent series of f around z = 1.
1
For that branch of the logarithm which vanishes at 1, for |w| < we
2
have
w2 w3 w4
log (1 + w) = w − + − + O(|w5 |);
2 3 4
after some computation, one can obtain
1 −4 −2 7 −2 1 −1
= w + 2w + w + w + O(1).
log (1 + w)4 6 6
The remaining branches of logarithm give a bounded function. So
7 1
f (1 + w) = w−4 + 2w−2 + w−2 + w−1
6 6
(the remainder vanishes) and
1 + 2(z − 1) + 67 (z − 1)2 + 61 (z − 1)3 z(z 2 + 4z + 1)
f (z) = = .
(z − 1)4 6(z − 1)4
2.11. Solutions of Olympic 2004 137

Solution 2. From the well-known series for the cotangent function,

N
X 1 i iw
lim = cot
N →∞ w + 2πi.k 2 2
k=−N
and
N ilog z
X 1 i ilog z i 22i 2 + 1 1 1
lim = cot = .i = + .
n→∞ log z + 2πi.k 2 2 2 2i ilog z 2 z−1
k=−N 2 2 −1
Taking derivatives we obtain
X 1 1 1 0 z
= −z. + = ,
(log z)2 2 z−1 (z − 1)2
X 1 z 1 0 z(z + 1)
= − . =
(log z)3 2 (z − 1)2 2(z − 1)3
and
X 1 z  z(z + 1) 0 z(z 2 + 4z + 1)
=− = .
(log z)4 3 2(z − 1)3 2(z − 1)4

2.11.2 Day 2

A1


Problem 1. Let A = A 2
and B = (B1 B2 ) where A1 , A2 , B1 , B2 are
2 × 2 matrices. Then
 1 0 −1 0 
 
0 1 0 −1  = A1 (B1 B2 ) = A1 B1 A1 B2
 
−1 0 1 0 A2 B1 A2 B2

A2
0 −1 0 1
therefore, A1 B1 = A2 B2 = I2 and A1 B2 = A2 B1 = −I2 Then B1 =
A−1 −1 −1
1 , B2 = −A1 and A2 = B2 = −A1 . Finally,
 
A1 
2 0

BA = (B1 B2 ) = B1 A1 + B2 A2 = 2I2 = 0 2
A2
Rx p Rx p
Problem 2. Let F (x) = f (t)dt and G(x) = g(t)dt. The func-
a a
tions F, G are convex, F (a) = 0 = G(a) and F (b) = G(b) by the hypoth-
esis. We are supposed to show that
Zb p Zb p
1 + (F 0 (t))2 dt ≥ 1 + (G0 (t))2 dt
a a
2.11. Solutions of Olympic 2004 138

i.e. The length at the graph of F is ≥ the length of the graph of G. This
is clear since both functions are convex, their graphs have common ends
and the graph of F is below the graph of G - the length of the graph of F
is the least upper bound of the lengths of the graphs of piecewise linear
functions whose values at the points of non-differentiability coincide with
the values of F , if a convex polygon P1 is contained in a polygon P2 then
the perimeter of P1 is ≤ the perimeter of P2 .
Problem 3. Considering as vectors, thoose p to be the unit vector
Pn
which points into the opposite direction as pi . Then, by the triangle
i=1
inequality,
n
X n
X n
X
|p − pi | ≥ np − pi = n + pi ≥ n.
i=1 i=1 i=1

Problem 4. We first solve the problem for the special case when the
eigenvalues of M are distinct and all sums λr +λs are different. Let λr and
λs be two eigenvalues of M and → −
v r, →
−
v s eigenvectors associated to them,
i.e. M → −v j = λ→ −v j for j = r, s. We have M → −v r (→
−
v s )T + → −v r (→
−
v s )T M T =
(M → −
v r )(→
−v s )T + →
−v r (M →
−
v s )T = λr →
−
v r (→
−
v s )T + λs →−
v r (→
−v s )T , so →
−
v r (→−
v s ) is
an eigenmatrix of LM with the eigenvalue λr + λs .
Notice that if λr 6= λs then vectors → −
u ,→−w , ware linearly independent
and matrices → −
u (→−
w )T and → −
w (→
−
u T are linearly independent, too. This
implies that the eigenvalue λr + λs is double if r 6= s.
The map LM maps n2 -dimensional linear space into itself, so it has
at most n2 eigenvalues. We already found n2 eigenvalues, so there exists
no more and the problem is solved for the special case.
In the general case, matrix M is a limit of matrices M1 , M2 , . . . such
that each of them belongs to the special case above. By the continuity
of the eigenvalues we obtain that the eigenvalues of LM are

• 2λr with multiplicity m2r (r = 1, . . . , k);

• λr + λs with multiplicity 2mr ms (1 ≤ r < s ≤ k).

2.11. Solutions of Olympic 2004 139

(It can happen that the sums λr + λs are not pairwise different; for those
multiple values the multiplicities should be summed up.)
Problem 5. First we use the inequality

x−1 − 1 ≥ |ln x|, x ∈ (0, 1],

which follows from

(x−1 − 1)|x=1 = |ln x||x=1 = 0,

1 1
(x−1 − 1)0 = − 2 ≤ − = |ln x|0 , x ∈ (0, 1].
x x
Therefore
Z1 Z1 Z1 Z1 Z1 Z1
dxdy dxdy dxdy
≤ = .
x−1 + |ln y| − 1 |ln x| + |ln y| |ln (x.y)|
0 0 0 0 0 0
u
Substituting y = , we obtain
x
Z1 Z1 Z1  Z1 Z1
dxdy dx  du du
= = |lnu| = 1.
|ln (x.y)| x |ln u| |ln u|
0 0 0 u 0

Solution 2. Substituting s = x−1 − 1 and u = s − ln y,

Z1 Z1 Z∞ Z∞ s−u Z∞  Zu  e−u
dxdy e es
= duds = ds dsdu.
x−1 + |ln y| − 1 (s + 1)2 u (s + 1)2 u
0 0 0 s 0 0

es
Since the function is convex,
(s + 1)2
Zu
es u  eu 
ds ≤ +1
(s + 1)2 2 (u + 1)2
0
so
Z1 Z1 Z∞   e−u
dxdy u eu
≤ +1 du
x−1 + |ln y| − 1 2 (u + 1)2 u
0 0 0
2.12. Solutions of Olympic 2005 140

Z∞ Z∞
1 du −u

= + e du = 1.
2 (u + 1)2
0 0

Problem 6. The quantity S(Ak−1 n ) has a special combinatorical mean-

ing. Consider an n × k table filled with 0’s and 1’s such that no 2 × 2
contains only 1’s. Denote the number of such fillings by Fnk . The filling
of each row of the table corresponds to some integer ranging from 0 to
2n − 1 written in base 2. Fnk equals to the number of k-tuples of integers
such that every two consecutive integers correspond to the filling of n×2
table without 2 × 2 squares filled with 1’s.
Consider binary expansions of integers i and jin in−1 . . . i1 and jn jn−1 . . . j1 .
There are two cases:
1. If in jn = 0 then i and j can be consecutive iff in−1 . . . i1 and
jn−1 . . . j1 can be consequtive.
2. If in = jn = 1 then i and j can be consecutive iff in−1 jn−1 = 0 and
in−2 . . . i1 and jn−2 . . . j1 can be consecutive.
Hence i and j can be consecutive iff (i+1, j+1)-th entry of An is 1. De-
n n
2P −1 2P −1
k−1
noting this entry by ai,j , the sum S(An ) = ··· ai1 i2 ai2 i3 . . . aik−1 ik
i1 =0 ik =0
k−1
counts the possible fillings. Therefore Fnk = S(An ).
The the obvious statement Fnk = Fkn completes the proof.

2.12 Solutions of Olympic 2005

2.12.1 Day 1

Problem 1. For n = 1 the rank is 1. Now assume n ≥ 2. Since

A = (i)ni,j=1 + (j)ni,j=1 , matrix A is the sum of two matrixes of rank 1.
Therefore, the rank of A is at most 2. The determinant of the top-left
2 × 2 minor is -1, so the rank is exactly 2. Therefore, the rank of A is 1
for n = 1 and 2 for n ≥ 2.
Solution 2. Consider the case n ≥ 2. For i = n, n−1, . . . , 2, subtract
the (i − 1)th row from the nth row. Then subtract the second row from
2.12. Solutions of Olympic 2005 141

all lower rows.

2 3 ... n + 1 2 3 ... n+1
   
3 4 . . . n + 2 1 1 ... 1 
rank  .. .. ..  = rank  .. .. ..  =

. . ... . . . ... .
n+1 n+2 . . . 2n 1 1 ... 1
 
1 2 ... n
1 1 ... 1
= rank 0 0 ... 0  = 2.

 ... ..
. ... .. 
.
0 0 ... 0
Problem 2. Extend the definitions also for n = 1, 2. Consider the
following sets

A0n = {(x1 , x2 , . . . , xn ) ∈ An : xn−1 = xn }, A00n = An \A0n ,

Bn0 = {(x1 , x2 , . . . , xn ) ∈ Bn : xn = 0}, Bn00 = Bn \Bn0

and denote an = |An |, a0n = |A0n |, a00n = |A00n |, bn = |Bn |, b0n = |Bn0 |, b00n =
|Bn00 |.
It is easy to observe the following relations between the a-sequences

an = a0n + a00n
a0n+1 = a00n ,
a00 0 00
n+1 = 2an + 2an

which lead to an+1 = 2an + 2an−1 .

For the b-sequences we have the same relations

bn = b0n + b00n
b0n+1 = b00n ,
b00 0 00
n+1 = 2bn + 2bn

therefore bn+1 = 2bn + 2bn−1 .

By computing the first values of (an ) and (bn ) we obtain

a1 = 3, a2 = 9, a3 = 24
b1 = 3, b2 = 8
which leads to 
a2 = 3b1
a3 = 3b2
2.12. Solutions of Olympic 2005 142

Now, reasoning by induction, it is easy to prove that an+1 = 3bn for

every n ≥ 1.
Solution 2. Regarding xi to be elements of Z3 and working ”modulo
3” , we have that

(x1 , x2 , . . . , xn ) ∈ An ⇒ (x1 + 1, x2 + 1, . . . , xn + 1) ∈ An ,
(x1 + 2, x2 + 2, . . . , xn + 2 ∈ An
1
which means that of the elements of An start with 0. We establish a
3
bijection between the subset of all the vectors in An+1 which start with
0 and the set Bn by

(0, x1 , x2 , . . . , xn ) ∈ An+1 7→ (y1 , y2 , . . . , yn ) ∈ Bn

y1 = x1 , y2 = x2 − x1 , y3 = x3 − x2 , . . . , yn = xn − xn−1

(if yk = yk+1 = 0 then xk − xk−1 = xk+1 − xk = 0 (where x0 = 0), which

gives xk−1 = xk = xk+1 , which is not possible because of the definition
of the sets Ap ; therefore, the definition of the above function is correct).
The inverse is defined by

(y1 , y2 , . . . , yn ) ∈ Bn 7→ (0, x1 , x2 , . . . , xn ) ∈ An+1

x1 = y1 , x2 = y1 + y2 , . . . , xn = y1 + y2 + · · · + yn

Problem 3. Let M = max |f 0 (x)|. By the inequality −M ≤ f 0 (x) ≤

0≤x≤1
M, x ∈ [0, 1] it follows:

−M f (x) ≤ f (x)f 0 (x) ≤ M f (x), x ∈ [0, 1].

By integration
Zx Zx
1 1
−M f (t)dt ≤ f 2 (x) − f 2 (0) ≤ M f (t)dt, x ∈ [0, 1]
2 2
0 0
Zx Zx
1 1
−M f (x) f (t)dt ≤ f 3 (x) − f 2 (0)f (x) ≤ M f (x) f (t)dt, x ∈ [0, 1].
2 2
0 0
2.12. Solutions of Olympic 2005 143

Integrating the last inequality on [0, 1] it follows that

 Z1 2 Z1 Z1  Z1 2
3 2
−M f (x)dx ≤ f (x)dx−f (0) f (x)dx ≤ M f (x)dx ⇔
0 0 0 0
Z1 Z1
 Z1 2
3 2
⇔ f (x)dx − f (0) f (x)dx ≤ M f (x)dx .
0 0 0

R1
Solution 2. Let M = max |f 0 (x)| and F (x) = − f ; then F 0 =
0≤x≤1 x
R1
f, F (0) = − f and F (1) = 0. Integrating by parts,
0

Z1 Z1 Z1
f3 = f 2 .F 0 = [f 2 F ]10 − (f 2 )0 F =
0 0 0
Z1 Z1 Z1
= f 2 (1)F (1) − f 2 (0)F (0) − 2F f f 0 = f 2 (0) f− 2F f f 0 .
0 0 0

Then
Z1 Z1 Z1 Z1
f 3 (x)dx − f 2 (0) f (x)dx = 2F f f 0 ≤ 2F f |f 0 | ≤
0 0 0 0
Z1  Z1 2
≤M 2F f = M.[F 2 ]10 = M f .
0 0

Problem 4. Note that P (x) does not have any positive root because
P (x) > 0 for every x > 0. Thus, we can represent them in the form
αi , i = 1, 2, . . . , n, where αi ≥ 0. If a0 6= 0 then there is a k ∈ N, 1 ≤ k ≤
n − 1, with ak = 0, so using Viete’s formulae we get

α1 α2 . . . αn−k−1 αn−k + α1 α2 . . . αn−k−1 αn−k+1 + · · · + αk+1 αk+2 . . . αn−1 αn

ak
= = 0,
an
2.12. Solutions of Olympic 2005 144

which is impossible because the left side of the equality is positive.

Therefore a0 = 0 and one of the roots of the polynomial, say αn , must
be equal to zero. Consider the polynomial Q(x) = an xn−l + an−1 xn−2 +
· · · + a1 . It has zeros −αi , i = 1, 2, . . . , n − 1. Again, Viete’s formulae,
for n ≥ 3, yield:
a1
α1 α2 . . . αn−1 = (1)
an
a2
α1 α2 . . . αn−2 + α1 α2 . . . αn−3 αn−1 + · · · + α2 α3 . . . αn−1 = (2)
an
an−1
α1 + α2 + · · · + αn−1 = . (3)
an
Dividing (2) by (1) we get
1 1 1 a2
+ + ··· + = . (4)
α1 α2 αn−1 a1
From (3) and (4), applying the AM-HM inequality we obtain
an−1 α1 + α2 + · · · + αn−1 1 (n − 1)a1
= ≥ 1 1 1 = ,
(n − 1)an n−1 α1 + α2 + ··· + αn−1
a2

a2 an−1 2 n2 a2 an−1
therefore ≥ (n − 1) . Hence ≥ ≥ (n − 1)2 , implying
a1 an 2 a1 an
n ≤ 3. So, the only polynomials possibly satisfying (i) and (ii) are those
of degree at most three. These polynomials can easily be found and they
are P (x) = x, P (x) = x2 + 2x, P (x) = 2x2 + x, P (x) = x3 + 3x2 + 2x and
P (x) = 2x3 + 3x2 + x.
Solution 2. Consider the prime factorization of P in the ring Z[x].
Since all roots of P are rational, P can be written as a product of n
linear polynomials with rational coefficients. Therefore, all prime factor
of P are linear and P can be written as
n
Y
P (x) = (bk x + ck )
k=1

where the coefficients bk , ck are integers. Since the leading coefficient of

P is positive, we can assume bk > 0 for all k. The coefficients of P are
2.12. Solutions of Olympic 2005 145

nonnegative, so P cannot have a positive root. This implies ck ≥ 0. It

is not possible that ck = 0 for two different values of k, because it would
imply a0 = a1 = 0. So ck > 0 in at least n − 1 cases.
Now substitute x = 1.
n
n(n + 1) Y
P (1) = an + · · · + a0 = 0 + 1 + · · · + n = = (bk + ck ) ≥ 2n−1 ;
2
k=1
n(n + 1)
therefore it is necessary that 2n−1 ≤ , therefore n ≤ 4. More-
2
n(n + 1)
over, the number can be written as a product of n − 1 integers
2
greater than 1.
If n = 1, the only solution is P (x) = 1x + 0.
If n = 2, we have P (1) = 3 = 1.3, so one factor must be x, the other
one is x + 2 or 2x + 1. Both x(x + 2) = 1x2 + 2x + 0 and x(2x + 1) =
2x2 + 1x + 0 are solutions.
If n = 3, then P (1) = 6 = 1.2.3, so one factor must be x, another one
is x + 1, the third one is again x + 2 or 2x + 1. The two polynomials are
x(x+1)(x+2) = 1x3 +3x2 +2x+0 and x(x+1)(2x+1) = 2x3 +3x2 +1x+0,
both have the proper set of coefficients.
n(n + 1)
In the case n = 4, there is no solution because cannot be
2
written as a product of 3 integers greater than 1.
Altogether we found 5 solutions: 1x+0, 1x2 +2x+0, 2x2 +1x+0, 1x3 +
3x2 + 2x + 0 and 2x3 + 3x2 + 1x + 0.
Problem 5. Let g(x) = f 0 (x) + xf (x); then f 00 (x) + 2xf 0 (x) + (x2 +
1)f (x) = g 0 (x) + xg(x).
We prove that if h is a continuously differentiable function such that
h0 (x) + xh(x) is bounded then limh = 0. Applying this lemma for h = g
∞
then for h = f , the statement follows.
2
Let M be an upper bound for |h0 (x) + xh(x)| and let p(x) = h(x)ex /2 .
2
(The function e−x /2 is a solution of the differential equation u0 (x) +
xu(x) = 0.) Then
2 2
|p0 (x)| = |h0 (x) + xh(x)|ex /2
≤ M ex /2
2.12. Solutions of Olympic 2005 146

and
Rx 0
Rx 2
p(0) + p |p(0)| + M ex /2
dx
p(x) 0 0
|h(x)| = x2 /2 = x 2 /2 ≤ x 2 /2 .
e e e
Rx 2
ex /2
dx
2 0
Since lim ex /2 = ∞ and lim = 0 (by L’Hospital’s rule), this
x→∞ ex2 /2
implies lim h(x) = 0.
x→∞
2
f (x)ex /x
Solution 2. Apply L’Hospital rule twice on the fraction .
ex2 /x
(Note that L’Hospital rule is valid if the denominator converges to infin-
ity, without any assumption on the numerator.)
2 2
f (x)ex /2 (f 0 (x) + xf (x))ex /2
lim f (x) = lim = lim
x→∞ x→∞ ex2 /2 x→∞ xex2 /2
2
(f 00 (x) + 2xf 0 (x) + (x2 + 1)f (x))ex /2
= lim =
x→∞ (x2 + 1)ex2 /2
f 00 (x) + 2xf 0 (x) + (x2 + 1)f (x)
= lim = 0.
x→∞ x2 + 1
Problem 6. Write d = gcd(m, n). It is easy to see that < G(m), G(n) >=
G(d); hence, it will suffice to check commutativity for any two elements
in G(m) ∪ G(n), and so for any two generators am and bn . Consider their
commutator z = a−m b−n am bn ; then the relations

z = (a−m bam )−n bn = a−m (b−n abn )m

show that z ∈ G(m) ∩ G(n). But then z is in the center of G(d). Now,
from the relation am bn = bn am z, it easily follows by induction that
2
aml bnl = bnl aml z l .
2 2
Setting l = md and l = n
d we obtain z (m/d) = z (n/d) = e, but this implies
that z = e as well.

2.12.2 Day 2
2
Problem 1. Write f (x) = (x + 2b )2 + d where d = c − b4 . The absolute
minimum of f is d.
2.12. Solutions of Olympic 2005 147

If d ≥ 1 then f (x) ≥ 1 for all x, M = ∅ and |M | = 0.

If −1 < d < 1 then f (x) > −1 for all x,
 b 2 b √
−1 < x + +d<1 ⇔ x+ < 1−d
2 2
so  b √ b √ 
M= − − 1 − d, − + 1 − d
2 2
and
√ √
|M | = 2 1 − d < 2 2.
If d ≤ −1 then
 b 2 p b p
−1 < x + +d<1 ⇔ |d| − 1 < x + < |d| + 1
2 2
so
p p p p
M = (− |d| + 1, − |d| − 1) ∪ ( |d| − 1, |d| + 1)
and
p p (|d| + 1) − (|d| − 1)
|M | = 2( |d| + 1 − |d| − 1) = 2 p p ≤
|d| + 1 + |d| − 1
2 √
≤ 2√ √ = 2 2.
1+1+ 1−0
Problem 2. Yes, it is even enough to assume that f 2 and f 3 are poly-
nomials.
Let p = f 2 and q = f 3 . Write these polynomials in the form of

p = a.pa21 . . . pakk , q = b.q1b1 . . . qlbl ,

where a, b ∈ R, a1 , . . . , ak , b1 , . . . , bl are positive integers and p1 , . . . , pk ,

q1 , . . . , ql are irreducible polynomials with leading coefficients 1. For
p3 = q 2 and the factorisation of p3 = q 2 is unique we get that a3 =
b2 , k = 1 and for some (i1 , . . . , ik ) permutation of (1, . . . , k) we have
p1 = qi1 , . . . , pk = qik and 3a1 = 2bi1 , . . . , 3ak = 2bik . Hence b1 , . . . , bl are
b /3 b /3
divisible by 3 let r = b1/3 .q11 . . . ql l be a polynomial. Since r3 = q = f 3
we have f = r.
2.12. Solutions of Olympic 2005 148

p
Solution 2. Let be the simplest form of the rational function
q
f3 p2
. Then the simplest form of its square is . On the other hand
f2 q2
p2  f 3  2
2
= 2
= f 2 is a polynomial therefore q must be a constant and so
q f
3
f p
f = 2 = is a polynomial.
f q
Problem 3. If A is a nonzero symmetric matrix, then trace(A2 ) =
trace(At A) is the sum of the squared entries of A which is positive. So
V cannot contain any symmetric matrix but 0.
Denote by S the linear space of all real n × n symmetric matrices;
n(n + 1)
dimV = . Since V ∩ S = {0}, we have dimV + dimS ≤ n2 and
2
n(n + 1) n(n − 1)
thus dimV ≤ n2 − = .
2 2
n(n − 1)
The space of strictly upper triangular matrices has dimension
2
and satisfies the condition of the problem.
n(n − 1)
Therefore the maximum dimension of V is .
2
Problem 4. Let
f (−1) 2 f (1) 2
g(x) = − x (x−1)−f (0)(x2 −1)+ x (x+1)−f 0 (0)x(x−1)(x+1).
2 2
It is easy to check that g(±1) = f (±1), g(0) = f (0) and g 0 (0) = f 0 (0).
Apply Rolle’s theorem for the function h(x) = f (x) − g(x) and its
derivatives. Since h(−1) = h(0) = h(1) = 0, there exist η ∈ (−1, 0) and
ϑ ∈ (0, 1) such that h0 (η) = h0 (ϑ) = 0. We also have h0 (0) = 0, so there
exist % ∈ (η, 0) and σ ∈ (0, ϑ) such that h00 (%) = h00 (σ) = 0. Finally,
there exists a ξ ∈ (%, σ) ⊂ (−1, 1) where h000 (ξ) = 0. Then
f (−1) f (1) f (1) − f (−1) 0
f 000 (ξ) = g 000 (ξ) = − .6−f (0).0+ .6−f 0 (0).6 = −f (0).
2 2 2
f (1) − f (−1)
Solution 2. The expression − f 0 (0) is the divided
2
difference f [−l, 0, 0, 1] and there exists a number ξ ∈ (−1, 1) such that
f 000 (ξ)
f [−1, 0, 0, 1] = .
3!
2.12. Solutions of Olympic 2005 149

x2
Problem 5. To get an upper bound for r, set f (x, y) = x − +
2
y2
. This function satisfies the conditions, since grad f (x, y) = (1 −
2
x, y), grad f (0, 0) = (1, 0) and (|grad f (x1 , y1 ) − grad f (x2 , y2 )| = |(x2 −
x1 , y1 − y2 )| = |(x1 , y1 ) − (x2 , y2 )|
In the disk Dr = {(x, y) : x2 + y 2 ≤ r2 }
x2 + y 2  1 1 r2 1
f (x, y) = − x − )2 + ≤ + .
2 2 4 2 4
1 r2 1
If r > then the absolute maximum is + , attained at the points
r 2 2 4
1 1  1 1
, ± r2 − . Therefore, it is necessary that r ≤ because if r >
2 4 2 2
then the maximum is attained twice.
1
Suppose now that r ≤ and that f attains its maximum on Dr at
2
u, v, u 6= v. Since |grad f (z) − grad f (0)| ≤ r, |grad f (z)1 ≤ 1 − r > 0 for
all z ∈ Dr . Hence f may attain its maximum only at the boundary of Dr ,
so we must have |u| = |v| = r and grad f (u) = au and grad f (v) = bv,
where a, b ≥ 0. Since au = grad f (u) and bv = grad f (v) belong to
the disk D with centre grad f (0) and radius r, they do not belong to the
interior of Dr . Hence |grad f (u)−grad f (v)| = |au−bv| ≥ |u−v| and this
inequality is strict since D ∩ Dr contains no more than one point. But
this contradicts the assumption that |grad f (u) − grad f (v)| ≤ |u − v|.
1
So all r ≤ satisfies the condition.
2
Problem 6. First consider the case when q = 0 and r is rational.
Choose a positive integer t such that r2 t is an integer and set
  2

a b 1 + rt −r t .

c d = t 1 − rt
Then
a b
  ar + b (1 + rt)r − r2 t
det c d and = = r.
cr + d tr + (1 − rt)
Now assume q 6= 0. Let the minimal polynomial of r in Z[x] be ux2 +
√
vx + w. The other root of this polynomial is r = p − q 7, so v =
2.12. Solutions of Olympic 2005 150

−u(r + r) = −2up and w = urr = u(p2 − 7q 2 ). The discriminant is

v 2 − 4uw = 7.(2uq)2 . The left-hand side is an integer, implying that also
4 = 2uq is an integer.
ar + b
The equation = r is equivalent to cr2 + (d − a)r − b = 0. This
cr + d
must be a multiple of the minimal polynomial, so we need

c = ut, d − a = vt, −b = wt

for some integer t 6= 0. Putting together these equalities with ad−bc = 1

we obtain that

(a + d)2 = (a − d)2 + 4ad = 4 + (v 2 − 4uw)t2 = 4 + 742 t2 .

Therefore 4 + 742 t2 must be a perfect square. Introducing s = a + d,

we need an integer solution (s, t) for the Diophantine equation

s2 − 742 t2 = 4 (1)

such that t 6= 0.
The numbers s and t will be even. Then a + d = s and d − a = vt
will be even as well and a and d will be really integers.
√ √
Let (8 ± 3 7)n = kn ± ln 7 for each integer n. Then kn2 − 7ln2 =
√ √ √ √
(kn + ln 7)(kn − ln 7) = ((8 + 3 7)n (8 − 3 7))n = 1 and the sequence
(ln ) also satisfies the linear recurrence ln+1 = 16ln − ln−1 . Consider the
residue of ln modulo 4. There are 42 possible residue pairs for (ln , ln+1 )
so some are the same. Starting from such two positions, the recurrence
shows that the sequence of residues is periodic in both directions. Then
there are infinitely many indices such that ln ≡ l0 = 0 (mod 4).
Taking such an index n, we can set s = 2kn and t = 2ln /4.
Remarks. 1. It is well-known that if D > 0 is not a perfect square
then the Pell-like Diophantine equation

x2 + Dy 2 = 1

has infinitely many solutions. Using this fact the solution can be gener-
alized to all quadratic algebraic numbers.
2.13. Solutions of Olympic 2006 151

2. It is also known that the continued fraction of a real number r is

periodic from a certain point if and only if r is a root of a quadratic
equation. This fact can lead to another solution.

2.13 Solutions of Olympic 2006

2.13.1 Day 1

Problem 1.
a) False. Consider function f (x) = x3 −x. It is continuous, range(f ) =
R but, for example, f (0) = 0, f ( 21 ) = − 38 and f (1) = 0, therefore
f (0) > f ( 12 ), f ( 12 ) < f (1) and f is not monotonic.
b) True. Assume first that f is non-decreasing. For an arbitrary
number a, the limits limf and limf exist and limf ≤ limf . If the two
a− a+ a− a+
limits are equal, the function is continuous at a. Otherwise, if limf =
a−
b < limf = c, we have f (x) ≤ b for all x < a and f (x) ≥ c for all x > a;
a+
therefore range(f ) ⊂ (−∞, b) ∪ (c, ∞) ∪ {f (a)} cannot be the complete
R.
For non-increasing f the same can be applied writing reverse relations
or g(x) = −f (x).
c) False. The function g(x) = arctan x is monotonic and continuous,
but range(g) = (− π2 , π2 ) 6= R.
Problem 2. Let Sk = {0 < x < 10k |x2 − x is divisible by 10k } and
s(k) = |Sk |, k ≥ 1. Let x = ak+1 ak . . . a1 be the decimal writing of an
integer x ∈ Sk+1 , k ≥ 1. Then obviously y = ak . . . a1 ∈ Sk . Now, let
y = ak . . . a1 ∈ Sk be fixed. Considering ak+1 as a variable digit, we have
x2 − x = (ak+1 10k + y)2 − (ak+1 10k + y) = (y 2 − y) + ak+1 10k (2y − 1) +
a2k+1 102k . Since y 2 − y = 10k z for an iteger z, it follows that x2 − x is
divisible by 10k+1 if and only if z + ak+1 (2y − 1) ≡ 0 (mod 10). Since
y ≡ 3 (mod 10) is obviously impossible, the congruence has exactly one
solution. Hence we obtain a one-to-one correspondence between the sets
Sk+1 and Sk for every k ≥. Therefore s(2006) = s(1) = 3, because
2.13. Solutions of Olympic 2006 152

S1 = {1, 5, 6}.
Solution 2. Since x2 − x = x(x − 1) and the numbers x and x − 1
are relatively prime, one of them must be divisible by 22006 and one of
them (may be the same) must be divisible by 52006 . Therefore, x must
satisfy the following two conditions:

x ≡ 0 or 1 (mod 22006 );
x ≡ 0 or 1 (mod 52006 ).

Altogether we have 4 cases. The Chinese remainder theorem yields that

in each case there is a unique solution among the numbers 0, 1, . . . , 102006 −
1. These four numbers are different because each two gives different
residues modulo 22006 or 52006 . Moreover, one of the numbers is 0 which
is not allowed.
Therefore there exist 3 solutions.
Problem 3. By induction, it is enough to consider the case m = 2. Fur-
thermore, we can multiply A with any integral matrix with determinant
1 from the right or from the left, without changing the problem. Hence
we can assume A to be upper triangular.
Lemma. Let A be an integral upper triangular matrix, and let b, c be
integers satisfying detA = bc. Then there exist integral upper triangular
matrices B, C such that detB = b, detC = c, A = BC.
Proof. The proof is done by induction on n, the case n = 1 being
obvious. Assume the statement is true for n − 1. Let A, b, c as in the
statement of the lemma. Define Bnn to be the greatest common divisor
Ann
of b and Ann , and put Cnn = . Since Ann divides bc, Cnn divides
Bnn
b b
c, which divides c. Hence Cnn divides c. Therefore, b0 = and
Bnn Bnn
c
c0 = are integers. Define A0 to be the upper-left (n − 1) × (n − 1)-
Cnn
submatrix of A; then detA0 = b0 c0 . By induction we can find the upper-
left (n−1)×(n−1)-part of B and C in such a way that detB = b, detC = c
and A = BC holds on the upper-left (n − 1) × (n − 1)-submatrix of A.
2.13. Solutions of Olympic 2006 153

It remains to define Bi,n and Ci,n such that A = BC also holds for the
(i, n)-th entry for all i < n.
First we check that Bii and Cnn are relatively prime for all i < n.
Since Bii divides b0 , it is certainly enough to prove that b0 and Cnn are
relatively prime, i.e.
 b Ann 
gcd , = 1,
gcd(b, Ann ) gcd(b, Ann )
which is obvious. Now we define Bj,n and Cj,n inductively: Suppose we
have defined Bi,n and Ci,n for all i = j + 1, j + 2, . . . , n − 1. Then Bj,n
and Cj,n have to satisfy

Aj,n = Bj,j Cj,n + Bj,j+1 Cj+1,n + · · · + Bj,n Cn,n

Since Bj,j and Cn,n are relatively prime, we can choose integers Cj,n and
Bj,n such that this equation is satisfied. Doing this step by step for all
j = n − 1, n − 2, . . . , 1, we finally get B and C such that A = BC.
Problem 4. Let S be an infinite set of integers such that rational
p(x)
function f (x) is integral for all X ∈ S. Suppose that f (x) = where
q(x)
p is a polynomial of degree k and q is a polynomial of degree n. Then
p, q are solutions to the simultaneous equations p(x) = q(x)f (x) for all
x ∈ S that are not roots of q. These are linear simultaneous equations
in the coefficients of p, q with rational coefficients. Since they have a
solution, they have a rational solution.
Thus there are polynomials p0 , q 0 with rational coefficients such that
p0 (x) = q 0 (x)f (x) for all x ∈ S that are not roots of q. Multiplying this
with the previous equation, we see that p0 (x)q(x)f (x) = p(x)q 0 (x)f (x)
for all x ∈ S that are not roots of q. If x is not a root of p or q, then
f (x) 6= 0, and hence p0 (x)q(x) = p(x)q 0 (x) for all x ∈ S except for
finitely many roots of p and q. Thus the two polynomials p0 q and pq 0 are
equal for infinitely many choices of value. Thus p0 (x)q(x) = p(x)q 0 (x).
p(x)
Dividing by q(x)q 0 (x), we see that df racp0 (x)q 0 (x) = = f (x). Thus
q(x)
2.13. Solutions of Olympic 2006 154

f (x) can be written as the quotient of two polynomials with rational

coefficients. Multiplying up by some integer, it can be written as the
quotient of two polynomials with integer coefficients.
p00 (x)
Suppose f (x) = 00 where p00 and q 00 both have integer coeffi-
q (x)
cients. Then by Euler’s division algorithm for polynomials, there ex-
ist polynomials s and r, both of which have rational coefficients such
that p00 (x) = q 00 (x)s(x) + r(x) and the degree of r is less than the de-
r(x)
gree of q 00 . Dividing by q 00 (x), we get that f (x) = s(x) + 00 . Now
q (x)
there exists an integer N such that N s(x) has integral coefficients. Then
N f (x) − N s(x) is an integer for all x ∈ S. However, this is equal to
Nr
the rational function 00 , which has a higher degree denominator than
q
numerator, so tends to 0 as x tends to ∞. Thus for all sufficiently large
x ∈ S, N f (x) − N s(x) = 0 and hence r(x) = 0. Thus r has infinitely
many roots, and is 0. Thus f (x) = s(x), so f is a polynomial.
Problem 5. Without loss of generality a ≥ b ≥ c, d ≥ e. Let c2 =
e2 + 4, 4 ∈ R. Then d2 = a2 + b2 + 4 and the second equation implies

4 4 2 2 2 2 2 a2 b 2
4
a + b + (e + 4) = (a + b + 4) + e , 4 = − 2 .
a + b2 − e2
2 1 1
(Here a2 + b2 − e2 ) ≥ (a2 + b2 + c2 ) − (d2 + e2 ) = (d2 + e2 ) > 0).
3 2 6
2 2 2 2 2 2
a b (a − e )(e − b )
Since c2 = e2 − 2 = > 0 then a > e > b.
a + b2 − e2 a2 + b2 − e2
a2 b2
2 2
Therefore d = a + b − 2 2
2 2
< a2 and a > d ≥ e > b ≥ c.
a +b −e
Consider a function f (x) = ax + bx + cx − dx − ex , x ∈ R. We shall
prove that f (x) has only two zeroes x = 2 and x = 4 and changes the
sign at these points. Suppose the contrary. Then Rolle’s theorem implies
that f 0 (x) has at least two distinct zeroes. Without loss of generality
a = 1. Then f 0 (x) = ln b.bx + ln c.cx − ln d.dx − ln e.ex , x ∈ R. If f 0 (x1 ) =
f 0 (x2 ) = 0, x1 < x2 , then ln b.bxi + ln c.cxi = ln d.dxi + ln e.exi , i = 1, 2,
2.13. Solutions of Olympic 2006 155

but since 1 > d ≥ e > b ≥ c we have

(−ln b).bx2 + (−ln c).cx2 x2 −x1 x2 −x1 (−ln d).dx2 + (−ln e).ex2
≤b <e ≤ ,
(−ln b).bx1 + (−ln c).cx1 (−ln d).dx1 + (−ln e).ex1
a contradiction. Therefore f (x) has a constant sign at each of the inter-
vals (−∞, 2), (2, 4) and (4, ∞). Since f (0) = 1 then f (x) > 0, x ∈
(−∞, 2) ∪ (4, ∞) and f (x) < 0, x ∈ (2, 4). In particular, f (3) =
a3 + b3 + c3 − d3 − e3 < 0.
Problem 6. Let A(x) = a0 + a1 x + · · · + an xn . We prove that se-
quence a0 , . . . , an satisfies the required property if and only if all zeros
of polynomial A(x) are real.
a) Assume that all roots of A(x) are real. Let us use the following
notations. Let I be the identity operator on R → R functions and D
be differentiation operator. For an arbitrary polynomial P (x) = p0 +
p1 x + · · · + pn xn , write P (D) = p0 I + p1 D + p2 D2 + · · · + pn Dn . Then
the statement can written as (A(D)f )(ξ) = 0.
First prove the statement for n = 1. Consider the function
a0
g(x) = e a1 f (x).

Since g(x0 ) = g(x1 ) = 0, by Rolle’s theorem there exists aξ ∈ (x0 , x1 )

for which
a0
a1 ξ
0 a0 aa0 ξ a0
ξ 0 e
g (ξ) = e 1 f (ξ) + e a1 f (ξ) = (a0 f (ξ) + a1 f 0 (ξ)) = 0.
a1 a1
Now assume that n > 1 and the statement holds for n − 1. Let
A(x) = (x − c)B(x) where c is a real root of polynomial A. By the
n = 1 case, there exist y0 ∈ (x0 , x1 ), y1 ∈ (x1 , x2 ), . . . , yn−1 ∈ (xn−1 , xn )
such that f 0 (yj ) − cf (yj ) = 0 for all j = 0, 1, . . . , n − 1. Now apply
the induction hypothesis for polynomial B(x), function g = f 0 − cf and
points y0 , . . . , yn−1 . The hypothesis says that there exists a ∈ (y0 , yn−1 ⊂
(x0 , xn ) such that

(B(D)g)(ξ) = (B(D)(D − cI)f )(ξ) = (A(D)f )(ξ) = 0.

2.13. Solutions of Olympic 2006 156

b) Assume that u + vi is a complex root of polynomial A(x) such that

v 6= 0. Consider the linear differential equation an g (n) +· · ·+a1 g 0 +g = 0.
A solution of this equation is g1 (x) = eux sin vx which has infinitely many
zeros.
Let k be the smallest index for which ak 6= 0. Choose a small  > 0 and
set f (x) = g1 (x) + xk . If  is sufficiently small then g has the required
number of roots but a0 f + a1 f 0 + · · · + an f (n) = ak  6= 0 everywhere.

2.13.2 Day 2

Problem 1. Apply induction on n. For the initial cases n = 3, 4, 5,

chose the triangulations shown in the Figure to prove the statement.
Now assume that the statement is true for some n = k and con-
sider the case n = k + 3. Denote the vertices of V by P1 , . . . , Pk+3
Apply the induction hypothesis on the polygon P1 P2 . . . Pk in this tri-
angulation each of vertices P1 , . . . , Pk belong to an odd number of tri-
angles, except two vertices if n is not divisible by 3. Now add trian-
gles P1 Pk Pk+2 , Pk Pk+1 Pk+2 and P1 Pk+2 Pk+3 . This way we introduce two
new triangles at vertices P1 and Pk so parity is preserved. The vertices
Pk+l , Pk+2 and Pk+3 share an odd number of triangles. Therefore, the
number of vertices shared by even number of triangles remains the same
as in polygon P1 P2 . . . Pk .
Problem 2. The functions f (x) = x + c and f (x) = −x + c with some
constant c obviously satisfy the condition of the problem. We will prove
now that these are the only functions with the desired property.
Let f be such a function. Then f clearly satisfies |f (x) − f (y)| ≤
|x − y| for all x, y; therefore, f is continuous. Given x, y with x < y, let
a, b ∈ [x, y] be such that f (a) is the maximum and f (b) is the minimum
of f on [x, y]. Then f ([x, y]) = [f (b), f (a)]; hence

y − x = f (a) − f (b) ≤ |a − b| ≤ y − x

This implies {a, b} = {x, y}, and therefore f is a monotone function.

2.13. Solutions of Olympic 2006 157

Suppose f is increasing. Then f (x) − f (y) = x − y implies f (x) − x =

f (y)−y, which says that f (x) = x+c for some constant c. Similarly, the
case of a decreasing function f leads to f (x) = −x + c for some constant
c.
Problem 3. Let f (x) = tan(sin x) − sin(tan x). Then

0 cos x cos(tan x) cos3 x − cos(tan x). cos2 (sin x)

f (x) = − =
cos2 (sin x) cos2 x cos2 (x). cos2 (tan x)
Let 0 < x < arctan π2 . It follows from the concavity of cosine on (0, π2 )
that
p3 1
cos(tan x). cos2 (sin x) < [cos(tan x) + 2 cos(sin x)]
3
h tan x + 2 sin x i
≤ cos < cos x,
3
the last inequality follows from
h tan x + 2 sin x i0 1 h 1 i r 1
3
= + 2 cos x ≥ . cos x. cos x = 1.
3 3 cos2 x cos2 x
This proves that cos3 x − cos(tan x). cos2 (sin x) > 0, so f 0 (x) > 0, so f
increases on the interval [0, arctan π2 ]. To end the proof it is enough to
notice that (recall that 4 + π 2 < 16)
h  π i π/2 π
tan sin arctan = tan p > tan = 1.
2 1 + π 2 /4 4

This implies that if x ∈ [arctan π2 , π2 ] then tan(sin x) > 1 and therefore

f (x) > 0.
Problem 4. By passing to a subspace we can assume that v1 , . . . , vn
are linearly independent over the reals. Then there exist λ1 , . . . , λn ∈ R
satisfying
X n
vn+1 = λj vj
j=1

We shall prove that λj is rational for all j. From

−2 < vi , vj >= |vi − vj |2 − |vi |2 − |vj |2

2.13. Solutions of Olympic 2006 158

we get that < vi , vj > is rational for all i, j. Define A to be the rational
n × n-matrix Aij =< vi , vj >, w ∈ Qn to be the vector wi =< vi , vn+1 >,
and λ ∈ Rn to be the vector (λi )i Then,
n
X
< vi , vi+1 >= λj < vi , vj >
j=1

gives Aλ = w. Since v1 , . . . , vn are linearly independent, A is invertible.

The entries of A−1 are rationals, therefore λ = A−1 w ∈ Qn , and we are
done.
Problem 5. Substituting y = x + m, we can replace the equation by

y 3 − ny + mn = 0.

Let two roots be u and v; the third one must be w = −(u + v) since the
sum is 0. The roots must also satisfy

uv + uw + vw = −(u2 + uv + v 2 ) = −n, i.e. u2 + uv + v 2 = n

and
uvw = −uv(u + v) = mn.
So we need some integer pairs (u, v) such that uv(u + v) is divisible by
u2 + uv + v 2 . Look for such pairs in the form u = kp, v = kq. Then

u2 + uv + v 2 = k 2 (p2 + pq + q 2 ),

and
uv(u + v) = k 3 pq(p + q).
Chosing p, q such that they are coprime then setting k = p2 + pq + q 2 we
uv(u + v)
have 2 2
= p2 + pq + q 2 .
u + uv + v
Substituting back to the original quantites, we obtain the family of
cases
m = (p2 + pq + q 2 )3 , m = p2 q + pq 2 ,
and the three roots are

x1 = p3 , x2 = q 3 , x3 = −(p + q)3 .
2.13. Solutions of Olympic 2006 159

Problem 6. We note that the problem is trivial if Aj = λI for some j,

so suppose this is not the case. Consider then first the situation where
some Aj , sayA3 , has
 two distinct real eigenvalues. We may assume  that
λ a b

A3 = B3 = µ by conjugating both sides. Let A2 = c d and
 
a0 b 0
B2 = c0 d0 . Then

a + d = T rA2 = T rB2 = a0 + d0
aλ + dµ = T r(A2 A3 ) = T r(A−1 −1 0 0
1 = T rB1 = T r(B2 B3 ) = a λ + d µ.

Hence a = a0 and d = d0 and so also bc = b0 c0 . Now we cannot have

c = 0 or b = 0, for then (1, 0)T or (0, T
 1) would be a common eigenvector
c0 −1
of all Aj . The matrix S = c conjugates A2 = S B2 S, and as S
commutes with A3 = B3 , it follows that Aj = S −1 Bj S for all j.
If the distinct eigenvalues of A3 = B3 are not real, we know from
above that Aj = S −1 Bj S for some S ∈ GL2 C unless all Aj have a com-
mon eigenvector over C. Even if they do, say Aj v = λj v, by taking the
conjugate square root it follows that A0j s can  be simultaneously diago-
0 0
a a b
 
nalized. If A2 = d and B2 = c0 d0 , it follows as above that
a = a0 , d = d0 and so b0 c0 = 0. Now B2 and B3 (and hence B1 too)
have a common eigenvector over C so they too can be simultaneously
diagonalized. And so SAj = Bj S for some S ∈ GL2 C in either case.
Let S0 = ReS and S1 = ImS. By separating the real and imaginary
components, we are done if either S0 or S1 is invertible. If not, So may
x 0
be conjugated to some T −1 S0 T = y 0 , with (x, y)T 6= (0, 0)T , and it
follows that all Aj have a common eigenvector T (0, 1)T , a contradiction.
We are left with the case when n0 Aj has distinct eigenvalues; then
these eigenvalues by necessity are real. Byconjugation and division
1 b
by scalars we may assume that A3 = 1 and b 6= 0. By further
conjugation by upper-triangular matrices (which preservesthe shape of
0 u

A3 up to the value of b) we can also assume that A2 = 1 v . Here
2.14. Solutions of Olympic 2007 160
 
−(b + v)/u 1
v 2 = T r2 A2 = 4detA2 = −4u. Now A1 = A−1 −1
3 A2 = 1/u ,
b + v)2 4
and hence 2
= T r2 A1 = 4detA1 = − . Comparing these two it
u u
follows that b = −2v. What we have done is simultaneously reduced all
Aj to matrices whose all entries depend on u and v (= −detA2 and T rA2 ,
respectively) only, but these themselves are invariant under similarity.
So Bj ’s can be simultaneously reduced to the very same matrices.

2.14 Solutions of Olympic 2007

2.14.1 Day 1

Problem 1. Let f (x) = ax2 + bx + c. Substituting x = 0, x = 1 and

x = −1, we obtain that 5|f (0) = c, 5|f (1) = (a + b + c) and 5|f (−1) =
(a − b + c). Then 5|f (1) + f (−1) − 2f (0) = 2a and 5|f (1) − f (−1) = 2b.
Therefore 5 divides 2a, 2b and c and the statement follows.
Solution 2. Consider f (x) as a polynomial over the 5-element field (i.e.
modulo 5). The polynomial has 5 roots while its degree is at most 2.
Therefore f ≡ 0 (mod 5) and all of its coefficients are divisible by 5.
Problem 2. The minimal rank is 2 and the maximal rank is n. To
prove this, we have to show that the rank can be 2 and n but it cannot
be 1.
(i) The rank is at least 2. Consider an arbitrary matrix A = [aij ] with
entries 1, 2, . . . , n2 in some order. Since permuting rows or columns of a
matrix does not change its rank, we can assume that 1 = a11 < a21 <
· · · < an1 and 1 = a11 < a12 < · · · < a1n . Hence an1 > n hand a1n > i n
a11 a1n
and at least one of these inequalities is strict. Then det a a <
n1 nn
a a
h i
1.n2 − n.n = 0 so rk(A) > rk a 11 a 1n > 2.
n1 nn
2.14. Solutions of Olympic 2007 161

(ii) The rank can be 2. Let

1 2 ... n
 
 n+1 n+2 . . . 2n
T = .. .. . . . ... 
. .
2 2
n −n+1 n −n+2 . . . n2
The i-th row is (1, 2, . . . , n) + n(i − 1).(1, 1, . . . , 1) so each row is in
the two-dimensional subspace generated by the vectors (1, 2, . . . , n) and
(1, 1, . . . , 1). We already proved that the rank is at least 2, so rk(T ) = 2.
(iii) The rank can be n, i.e. the matrix can be nonsingular. Put odd
numbers into the diagonal, only even numbers above the diagonal and
arrange the entries under the diagonal arbitrarily. Then the determinant
of the matrix is odd, so the rank is complete.
Problem 3. The possible values for k are 1 and 2. 
1 0

2
If k = 1 then P (x) = αx and we can choose A1 = 0 α .
If k= 2 then P (x, y)= αx2 + 2
 βy + γxy and we can choose matrices
1 0 1 β

A1 = 0 α and A2 = −1 γ .
k
x2i
P
Now let k > 3. We show that the polynomial P (x1 , . . . , xk ) =
k  i=0
P
is not good. Suppose that P (x1 , . . . , xk ) = det xi Ai . Since the
i=0
first columns of A1 , . . . , Ak are linearly dependent, the first column of
some non-trivial linear combination y1 A1 + · · · + yk Ak is zero. Then
det(y1 A1 + · · · + yk Ak ) = 0 but P (y1 , . . . , yk ) 6= 0, a contradiction.
Problem 4. We start with three preliminary observations.
Let U, V be two arbitrary subsets of G. For each x ∈ U and y ∈ V
there is a unique z ∈ G for which xyz = e. Therefore,

NU V G = |U × V | = |U |.|V |.

Second, the equation xyz = e is equivalent to yzx = e and zxy = e.

For arbitrary sets U, V, W ⊂ G, this implies

{(x, y, z) ∈ U × V × W : xyz = e} = {(x, y, z) ∈ U × V × W : yzx = e}

= {(x, y, z) ∈ U × V × W : zxy = e}
2.14. Solutions of Olympic 2007 162

and therefore
NU V W = NV W U = NW U V .
Third, if U, V ⊂ G and W1 , W2 , W3 are disjoint sets and W = W1 ∪
W2 ∪ W3 then, for arbitrary U, V ⊂ G,

{(x, y, z) ∈ U × V × W : xyz = e} = {(x, y, z) ∈ U × V × W1 : xyz = e}∪

∪{(x, y, z) ∈ U × V × W2 : xyz = e} ∪ {(x, y, z) ∈ U × V × W3 : xyz = e}

so
NU V W = NU V W1 + NU V W2 + NU V W3 .
Applying these observations, the statement follows as

NABC = NABG − NABA − NABB = |A|.|B| − NBAA − NBAB

= NBAG − NBAA − NBAB = NBAC = NCBA .

Problem 5. Let us define a subset I of the polynomial ring R[X] as

follows:
m
X m
X
j

I = P (X) = bj X : bj f (k + jl) = 0 for all k, l ∈ Z, l 6= 0 .
j=0 j=0

This is a subspace of the real vector space R[X]. Furthermore, P (X) ∈ I

implies X.P (X) ∈ I. Hence, I is an ideal, and it is non-zero, because
n
X ai belongs to I. Thus, I is generated (as
P
the polynomial R(X) =
i=1
an ideal) by some non-zero polynomial Q.
If Q is constant then the definition of I implies f = 0, so we can
assume that Q has a complex zero c. Again, by the definition of I, the
polynomial Q(X m ) belongs to I for every natural number m > 1; hence
Q(X) divides Q(X m ). This shows that all the complex numbers

c, c2 , c3 , c4 , . . .

are roots of Q. Since Q can have only finitely many roots, we must have
cN = 1 for some N > 1; in particular, Q(1) = 0, which implies P (1) = 0
2.14. Solutions of Olympic 2007 163

n
X ai ∈ I, and
P
for all P ∈ I. This contradicts the fact that R(X) =
i=1
we are done.
Problem 6. We show that the number of nonzero coefficients can be 0, 1
and 2. These values are possible, for example the polynomials P0 (z) =
0, P1 (z) = 1 and P2 (z) = 1 + z satisfy the conditions and they have 0, 1
and 2 nonzero terms, respectively.
Now consider an arbitrary polynomial P (z) = a0 + a1 z + · · · + an z n
satisfying the conditions and assume that it has at least two nonzero
coefficients. Dividing the polynomial by a power of z and optionally
replacing p(z) by −p(z), we can achieve a0 > 0 such that conditions are
not changed and the numbers of nonzero terms is preserved. So, without
loss of generality, we can assume that a0 > 0.
Let Q(z) = a1 z + · · · + an−1 z n−1 . Our goal is to show that Q(z) = 0.
Consider those complex numbers w0 , w1 , . . . , wn−1 on the unit circle
for which an wkn = |an |; namely, let
( 2kπi
e n if an > 0
wk = (2k+1)πi (k = 0, 1, . . . , n).
e n if an < 0

Notice that
n−1 n−1 n−1 n−1  k
X X 2kπi
X X 2jπi
aj w0j


Q(wk ) = Q w0 e n = e n = 0.
k=0 k=0 j=1 k=0

Taking the average of polynomial P (z) at the points wk , we obtained

n−1 n−1
1X 1X
P (wk ) = (a0 + Q(wk ) + an wkn ) = a0 + |an |
n n
k=0 k=0

and
n−1 n−1
1X 1X
2> |P (wk )| > P (wk ) = a0 + |an | > 2.
n n
k=0 k=0
This obviously implies a0 = |an | = 1 and |P (wk )| = |2 + Q(wk )| = 2 for
all k. Therefore, all values of Q(wk ) must lie on the circle |2 + z| = 2,
2.14. Solutions of Olympic 2007 164

while their sum is 0. This is possible only if Q(wk ) = 0 for all k. Then
polynomial Q(z) has at least n distinct roots while its degree is at most
n−1. So Q(z) = 0 and P (z) = a0 +an z n has only two nonzero coefficients.
Remark. From Parseval’s formula (i.e. integrating |P (z)|2 = P (z)P (z)
on the unit circle) it can be obtained that
Z2π Z2π
1 1
|a0 |2 + · · · + |an |2 = |P (eit )|2 dt 6 4dt = 4. (2.2)
2π 2π
0 0

Hence, there cannot be more than four nonzero coefficients, and if there
are more than one nonzero term, then their coefficients are ±1.
It is also easy to see that equality in (2.2) cannot hold two or more
nonzero coefficients, so it is sufficient to consider only polynomials of the
form 1 ± xm ± xn . However, we do not know any simpler argument for
these cases than the proof above.

2.14.2 Day 2

Problem 1. No. The function f (x) = ex also has this property since
cex = ex+log c .
Problem 2. We claim that 29|x, y, z. Then, x4 + y 4 + z 4 is clearly
divisible by 294 .
Assume, to the contrary, that 29 does not divide all of the numbers
x, y, z. Without loss of generality, we can suppose that 29 - x. Since the
residue classes modulo 29 form a field, there is some w ∈ Z such that
xw ≡ 1 (mod 29). Then (xw)4 + (yw)4 + (zw)4 is also divisible by 29.
So we can assume that x ≡ 1 (mod 29).
Thus , we need to show that y 4 +z 4 ≡ −1 (mod 29), i.e. y 4 ≡ −1−z 4
(mod 29), is impossible. There are only eight fourth powers modulo 29,
2.14. Solutions of Olympic 2007 165

0 ≡ 04 ,
1 ≡ 14 ≡ 124 ≡ 174 ≡ 284 (mod 29),
7 ≡ 84 ≡ 94 ≡ 204 ≡ 214 (mod 29),
16 ≡ 24 ≡ 54 ≡ 244 ≡ 274 (mod 29),
20 ≡ 64 ≡ 144 ≡ 154 ≡ 234 (mod 29),
23 ≡ 34 ≡ 74 ≡ 224 ≡ 264 (mod 29),
24 ≡ 44 ≡ 104 ≡ 194 ≡ 254 (mod 29),
25 ≡ 114 ≡ 134 ≡ 164 ≡ 184 (mod 29).

The differences −1−z 4 are congruent to 28, 27, 21, 12, 8, 5, 4 and 3. None
of these residue classes is listed among the fourth powers.
Problem 3. Suppose f (x) 6= x for all x ∈ C. Let [a, b] be the smallest
closed interval that contains C. Since C is closed, a, b ∈ C. By our
hypothesis f (a) > a and f (b) < b. Let p = sup{x ∈ C : f (x) > x}.
Since C is closed and f is continuous, f (p) > p, so f (p) > p. For all


x > p/, x ∈ C we have f (x) < x. Therefore f f (p) < f (p) contrary to
the fact that f is non-decreasing. 
1 if i − j ≡ ±1 (mod n)
Problem 4. Notice that A = B 2 , with bij = .
0 otherwise
So it is sufficient to find det B.
To find det B, expand the determinant with respect to the first row,
and then expad both terms with respect to the first columns.
0 1 1
1 0 1
1 0 1
det B = 1 ... ...
... 0 1
1 0 1
1 1 0
2.14. Solutions of Olympic 2007 166

1 1 1 0 1
0 1 1 0 1
.
1 .. .. . 1 ... ...
=− ... 0 1 + ... 0 1
1 0 1 1 0
1 1 0 1 1
0 1 1
 
 1 ... ... 0 1
. .

= −
 ... 0 1 − 1 . . . . +


1 0 1 . .. 0 1 
1 0 1 0 1
1 0 1 0 1
 
. .
1 .. .. 1 0 1
. .
 
+
 ... 0 1 − 1 . . . .  = −(0 − 1) + (1 − 0) = 2,


1 0 . .. 0 1 
1 1 0
since the second and the third matrices are lower/upper triangular, while
in the first and the fourth matrices we have row1 − row3 + row5 − · · · ±
rown−2 = 0.
So det B = 2 and thus det A = 4.
Problem 5. The answer is nk = 2k . In that case, the matrices can
be constructed as follows: Let V be the n-dimensional real vector space
with basis elements [S], where S runs through all n = 2k subsets of
{1, 2, . . . , k}. Define Ai as an endomorphism of V by

0 if i ∈ S
Ai [S] =
[S ∪ {i}] if i ∈
/S

for all i = 1, 2, . . . , k and S ⊂ {1, 2, . . . , k}. Then A2i = 0 and Ai Aj =

Aj Ai . Furthermore,

A1 A2 . . . Ak [∅] = [{1, 2, . . . , k}],

and hence A1 A2 . . . Ak 6= 0.
Now let A1 , A2 , . . . , Ak be n × n matrices satisfying the conditions of
the problem; we prove that n > 2k . Let v be a real vector satisfying
2.14. Solutions of Olympic 2007 167

A1 A2 . . . Ak v 6= 0. Denote by P the set of all subsets of {1, 2, . . . , k}.

Choose a complete ordering ≺ on P with the property

X ≺ Y ⇒ |X| 6 |Y | for all X, Y ∈ P.

For every element X = {x1 , x2 , . . . , xr } ∈ P, define AX = Ax1 Ax2 . . . Axr

and vX = AX v. Finally, write X = {1, 2, . . . , k} \ X for the complement
of X.
Now take X, Y ∈ P with X  Y . Then AX annihilates vY , because
X  Y implies the existence of some y ∈ Y \ X = Y ∩ X, and

AX vY = AX\{y} Ay Ay vY \{y} = 0,

since A2y = 0. So AX annihilates the span of all the vY with X  Y . This

implies that vX does not lie in this span, because AX vX = v{1,2,...,k} 6= 0.
Therefore, the vectors vX (with X ∈ P) are linearly independent; hence
n > |P| = 2k .
Problem 6. For the proof, we need the following
Lemma 1. For any polynomial g, denote by d(g) the minimum
distance of any two of its real zeros (d(g) = ∞ if g has at most one
real zero). Assume that g and g + g 0 both are of degree k > 2 and have
k distinct real zeros. Then d(g + g 0 ) > d(g).
Proof of Lemma 1: Let x1 < x2 < · · · < xk be the roots of g. Suppose
a, b are roots of g + g 0 satisfying 0 < b − a < d(g). Then a, b cannot be
roots of g and
g 0 (a) g 0 (b)
= = −1. (2.3)
g(a) g(b)
g0
Since is strictly decreasing between consecutive zeros of g, we must
g
have a < xj < b for some j.
For all i = 1, 2, . . . , k − 1 we have xi+1 − xi > b − a, hence a − xi >
b − xi+1 . If i < j, both sides of this inequality are negative; if i > j,
2.14. Solutions of Olympic 2007 168

1 1
both sides are positive. In any case, < , and hence
a − xi b − xi+1
k−1 k−1
g 0 (a) X 1 1 X 1 1 g 0 (b)
= + < + =
g(a) i=1
a − x i a − x k
| {z } i=1 b − x i+1 b − x
| {z }1 g(b)
<0 >0

This contrdicts (2.3).

Now we turn to the proof of the stated problem. Denote by m the
degree of f . We will prove by induction on m that fn has m distinct
real zeros for sufficient large n. The cases m = 0, 1 are trivial; so we
assume m > 2. Without loss of generality we can assume that f is
monic. By induction, the result holds for f 0 , and by ignoring the first
few terms we can assume that fn0 has m − 1 distinct real zeros for all n.
(n) (n) (n)
Let denote these zeros by x1 > x2 > · · · > xm−1 . Then fn has minima
(n) (n) (n) (n) (n) (n)
in x1 , x3 , x5 , . . . , and maxima in x2 , x4 , x6 , . . . . Note that in the
(n) (n) 0
interval (xi+1 , xi ), the function fn+1 = fn0 + fn00 must have a zero (this
follows by applying Rolle’s theorem to the function ex fn0 (x)); the same is
(n)
true for the interval (−∞, xm−1 ). Hence, in each of these m − 1 intervals,
0
fn+1 has exactly one zero. This shows that
(n) (n+1) (n) (n+1) (n) (n+1)
x1 > x1 > x2 > x2 > x3 > x3 > ... (2.4)
(n)
(n)

Lemma 2. We have lim fn xj = −∞ if j is odd, and lim fn xj =
n→∞ n→∞
+∞ if j is even.
Lemma 2 immediately implies the result: For sufficiently large n, the
values of all maxima of fn are positive, and the values of all minima of
fn are negative; this implies that fn has m distinct zeros.
Proof of Lemma 2: Let d = min{d(f 0 ), 1}; then by Lemma 1, d(fn0 ) >
(m − 1)dm−1
g for all n. Define  = ; we will show that
mm−1
(n+1)
(n)

fn+1 xj > fn xj +  for j even. (2.5)

(The corresponding result for odd j can be shown similarly.) Do to so,

(n)
write f = fn , b = xj , and choose a satisfying d 6 b − a 6 1 such that
2.14. Solutions of Olympic 2007 169

1
f 0 has no zero inside (a, b). Define ξ by the relation b − ξ = (b − a);
m
then ξ ∈ (a, b). We show that f (ξ) + f 0 (ξ) > f (b) + .
Notice that
m−1
f 00 (ξ) X 1
=
f 0 (ξ) (n)
i=1 ξ − xi
X 1 1 X 1
= (n)
+ +
i<j ξ − xi
ξ − b i>j ξ − x(n)
| {z } | {z i }
1
< ξ−a <0

1 1
< (m − 1) + = 0.
ξ−a ξ−b
f 00
The last equality holds by definition of ξ. Since f 0 is positive and is
f0
decreasing in (a, b), we have that f 00 is negative on (ξ, b). Therefore,
Zb Zb
f (b) − f (ξ) = f 0 (t)dt 6 f 0 (ξ)dt = (b − ξ)f 0 (ξ).
ξ ξ

Hence,

f (ξ) + f 0 (xi) > f (b) − (b − ξ)f 0 (ξ) + f 0 (ξ)

= f (b) + (1 − (ξ − b))f 0 (ξ)
1
= f (b) + 1 − (b − a) f 0 (ξ)


m
1
0
> f (b) + 1 − f (ξ).
m
Together with
m−1
0 0
Y (n) m−1 dm−1
f (ξ) = |f (ξ)| = m |ξ − xi | > m|ξ − b| > m−1
i=1
| {z } m
>|ξ−b|

we get
f (ξ) + f 0 (ξ) > f (b) + .
Together with (2.4) this shows (2.5). This finishes the proof of Lemma
2.
2.15. Solutions of Olympic 2008 170

2.15 Solutions of Olympic 2008

2.15.1 Day 1

Problem 1. We prove that f (x) = ax + b where a ∈ Q and b ∈ R.

These functions obviously satify the conditions.
Suppose that a function f (x) fulfills the required properties. For an
arbitrary rational q, consider the function gq (x) = f (x + q) − f (x). This
is a continuous function which attains only rational values, therefore gq
is constant.
Set a = f (1) − f (0) and b = f (0). Let n be an arbitrary positive
1
1
1

integer and let r = f −f (0). Since f x+ −f (x) = f −f (0) = r
n n n
for all x, we have
k   1
  2
1

f − f (0) = f − f (0) + f −f
n n n n
 k
k − 1

+··· + f −f = kr
n n
and
k
 1
  1
2

f − − f (0) = − f (0) − f − − f − −f −
n n n n
 k − 1
k

−··· − f − −f − = −kr for k > 1.
n n
a
In the case k = n we get a = f (1) − f (0) = nr, so r = . Hence,
n
k
ak k
k
f − f (0) = kr = and then f = a. + b for all integer k and
n n n n
n > 0.
So, we have f (x) = ax + b for all rational x. Since the function f is
continuous and the rational numbers form a dense subset of R, the same
holds for all real x.
Problem 2. We can assume that P 6= 0.
Let f ∈ V be such that P (f ) 6= 0. Then P (f 2 ) 6= 0, and therefore
P (f 2 ) = aP (f ) for some non-zero real a. Then 0 = P (f 2 − af ) =
P (f (f − a)) implies P (f − a) = 0, so we get P (a) 6= 0. By rescaling,
2.15. Solutions of Olympic 2008 171

we can assume that P (1) = 1. Now P (X + b) = 0 for b = −P (X).

Replacing P by Pb given as

Pb(f (X)) = P (f (X + b))

we can assume that P (X) = 0.

Now we are going to prove that P (X k ) = 0 for all k > 1. Suppose
this is true for all k < n. We know that P (X n + e) = 0 for e = −P (X n ).
From the induction hypothesis we get

P ((X + e)(X + 1)n−1 ) = P (X n + e) = 0,

and therefore P (X + e) = 0 (since P (X + 1) = 1 6= 0). Hence e = 0 and

P (X n ) = 0, which completes the inductive step. From P (1) = 1 and
P (X k ) = 0 for k > 1 we immediately get P (f ) = f (0) for all f ∈ V .
Problem 3. The theorem is obvious if p(ai ) = 0 for some i, so assume
that all p(ai ) are nonzero and pairwise different.
There exist numbers s, t such that s|p(a1 ), t|p(a2 ), st = lcm(p(a1 ), p(a2 ))
and gsd(s, t) = 1.
As s, t are relatively prime numbers, there exist m, n ∈ Z such that
a1 + sn = a2 + tm =: b2 . Obviously s|p(a1 + sn) − p(a1 ) and t|p(a2 +
tm) − p(a2 ), so st|p(b2 ).
Similarly one obtains b3 such that p(a3 )|p(b3 ) and p(b2 )|p(b3 ) thus also
p(a1 )|p(b3 ) and p(a2 )|p(b3 ).
Reasoning inductively we obtain the existence of a = bk as required.
The polynomial p(x) = 2x2 + 2 shows that the second part of the
problem is not true, as p(0) = 2, p(1) = 4 but no value of p(a) is divisible
by 8 for integer a.
Remark. One can assume that the p(ai ) are nonzero and ask for a such
that p(a) is a nonzero multiple of all p(ai ). In the solution above, it can
happen that p(a) = 0. But every number p(a + np(a1 )p(a2 ) . . . p(ak )) is
also divisible by every p(ai ), since the polynomial is nonzero, there exists
n such that p(a + np(a1 )p(a2 ) . . . p(ak )) satisfies the modified thesis.
2.15. Solutions of Olympic 2008 172

Problem 4. The answer is n > 4.

Consider the following set of special triples:
 8 7  2 3   3 2   2 11 2 
0, , , , 0, , , ,0 , , , .
15 15 5 5 5 5 15 15 15
We will prove that any special triple (x, y, z) is worse than one of
these (triple a is worse than triple b if triple b is better than triple a).
We suppose that some special triple (x, y, z) is actually not worse than
the first three of the triples from the given set, derive some conditions
on x, y, z and prove that, under these conditions, (x, y, z) is worse than
the fourth triple from the set.
8 7 8


Triple (x, y, z) is not worse than 0, 15 , 15 means that y > 15 or
7
. Triple (x, y, z) is not worse than 52 , 0, 35 — x > 52 or z > 35 .


z > 15
Triple (x, y, z) is not worse than 53 , 25 , 0 — x > 35 or z > 25 . Since

x + y + z = 1, then it is impossible that all inequalities x > 52 , y > 52

7
and z > 15 are true. Suppose that x < 25 , then y > 52 and z > 53 . Using
x + y + z = 1 and x > 0 we get x = 0, y = 25 , z = 35 . We obtain the triple
0, 52 , 35 which is worse than 15 2 11 2
, 15 , 15 . Suppose that y < 25 , then x > 53



7
and y > 15 and this is a contradiction to the admissibility of (x, y, z).
7
Suppose that z < 15 , then x > 25 and y > 15 8
. We get (by admissibility,
1
and y 6 35 . The last inequalities imply that 15 2 11 2


again) that z 6 15 , 15 , 15
is better than (x, y, z).
We will prove that for any given set of three special triples one can
find a special triple which is not worse than any triple from the set.
Suppose we have a set S of three special triples

(x1 , y1 , z1 ), (x2 , y2 , z2 ), (x3 , y3 , z3 ).

Denote

a(S) = min(x1 , x2 , x3 ), b(S) = min(y1 , y2 , y3 ), c(S) = min(z1 , z2 , z3 ).

2.15. Solutions of Olympic 2008 173

It is easy to check that S1 :

 x −a 1 y1 − b z1 − c 
, ,
1−a−b−c 1−a−b−c 1−a−b−c
 x2 − a y2 − b z2 − c 
, ,
1−a−b−c 1−a−b−c 1−a−b−c
 x3 − a y3 − b z3 − c 
, ,
1−a−b−c 1−a−b−c 1−a−b−c

is a set of three special triples also (we may suppose that a + b + c < 1,
because otherwise all three triples are equal and our statement is trivial).
If there is a special triple (x, y, z) which is not worse than any triple
from S1 , then the triple

((1 − a − b − c)x + a, (1 − a − b − c)y + b, (1 − a − b − c)z + c)

is special and not worse than any triple from S. We also have a(S1 ) =
b(S1 ) = c(S1 ) = 0, so we may suppose that the same holds for our
starting set S.
Suppose that one element of S has two entries equal to 0.
Note that one of the two remaining triples from S is not worse than
the other. This triple is also not worse than all triples from S because
any special triple is not worse than itself and the triple with two zeroes.
So we have a = b = c = 0 but we may suppose that all triples from
S contain at most one zero. By transposing triples and elements in
triples (elements in all triples must be transposed simultaneously) we
may achieve the following situation x1 = y2 = z3 = 0 and x2 > x3 . If
z2 > z1 , then the second triple (x2 , 0, z2 ) is not worse than the other two
triples from S. So we may assume that z1 > z2 . If y1 > y3 , then the first
triple is not worse than the second and the third and we assume y3 > y1 .
Consider the three pairs of numbers x2 , y1 ; z1 , x3 ; y3 , z2 . The sum of all
these numbers is three and consequently the sum of the numbers in one
of the pairs is less than or equal to one. If it is the first pair then the
triple (x2 , 1 − x2 , 0) is not worse than all triples from S, for the second
2.15. Solutions of Olympic 2008 174

we may take (1−z1 , 0, z1 ) and for the third — (0, y3 , 1−y3 ). So we found
a desirable special triple for any given S.
Problem 5. Yes. Let H be the commutative group H = F32 , where
F2 ∼
= Z/2Z is the field with two elements. The group of automorphisms
of H is the general linear group GL3 F2 ; it has

(8 − 1).(8 − 2).(8 − 4) = 7.6.4 = 168

elements. One of them is the shift operator φ : (x1 , x2 , x3 ) 7→ (x2 , x3 , x1 ).

Now let T = {a0 , a1 , a2 } be a group of order 3 (written multiplica-
tively); it acts on H by τ (a) = φ. Let G be the semidirect product
G = H oτ T . In other words, G is the group of 24 elements

G = {bai : b ∈ H, i ∈ (Z/3Z)}, ab = φ(b)a.

G has one element e of order 1 and seven elements b, b ∈ H, b 6= e of

order 2.
If g = ba, we find that g 2 = baba = bφ(b)a2 6= e, and that

g 3 = bφ(b)a2 ba = bφ(b)aφ(b)a2 = bφ(b)φ2 (b)a3 = ψ(b),

where the homomorphism ψ : H → H is defined as ψ : (x1 , x2 , x3 ) 7→

(x1 + x2 + x3 )(1, 1, 1). It is clear that g 3 = ψ(b) = e for 4 elements b ∈ H,
while g 6 = ψ 2 (b) = e for all b ∈ H.
We see that G has 8 elements of order 3, namely ba and ba2 with
b ∈ ker ψ, and 8 elements of order 6 namely ba and ba2 with b ∈ / ker ψ.
That accounts for orders of all elements of G.
Let b0 ∈ H \ ker ψ be arbitrary; it is easy to see that G is generated by
b0 and a. As every automorphism of G is fully determined by its action
on b0 and a, it follows that G has no more than 7.8 = 56 automorphisms.
Remark. G and H can be equivalently presented as subgroups of S6 ,
namely as H = (135)(246), (12) .
2.15. Solutions of Olympic 2008 175

Problem 6. Consider the n × n determinant

1 x . . . xn−1
x 1 . . . xn−2
∆(x) = .. .. ..
. . ... .
xn−1 xn−2 ... 1
where the ij-th entry is x|i−j| . From the definition of the determinant
we get
X
∆(x) = (−1)inv(i1 ,i2 ,...,in ) xD(i1 ,i2 ,...,in )
(i1 ,i2 ,...,in )∈Sn

where Sn is the set of all permutations of (1, 2, . . . , n) and inv(i1 , i2 , . . . , in )

denotes the number of inversions in the sequence (i1 , i2 , . . . , in ). So
Q(n, d) has the same parity as the coefficients of xd in ∆(x).
It remains to evaluate ∆(x). In order to eliminate the entries below
the diagonal, subtract the (n − 1)-th row, multipled by x, from the n-th
row. Then subtract the (n − 2)-th row, multipled by x, from the (n − 1)-
th and so on. Finally, subtract the first row, multipled by x, from the
second row.
1 x . . . xn−2 xn−1
x 1 . . . xn−3 xn−2
∆(x) = .. .. ... .. ..
. . . .
xn−2 xn−3 ... 1 x
xn−1 xn−2 ... x 1
1 x ... xn−2 xn−1
0 1 − x2 . . . xn−3 − xn−1 xn−2 − xn
= · · · = ... ..
. ...
..
.
..
. = (1 − x2 )n−1 .
0 0 ... 1 − x2 x − x2
0 0 ... 0 1 − x2
For d > 2n, the coefficient of xd is 0 so Q(n, d) is even.

2.15.2 Day 2

Problem 1. Let f (x) = x2n + xn + 1, g(x) = x2k − xk + 1, h(x) =

π π



x2k + xk + 1. The complex number x1 = cos 3k + i sin 3k is a root of
g(x).
2.15. Solutions of Olympic 2008 176

Let α = πn 3k . Since g(x) divides f
(x), f (x1 ) = g(x1 )
= 0. So, 0 =

x2n n
1 + x1 + 1 = cos(2α) + i sin(2α) + cos α + i sin α + 1 = 0, and


(2 cos α + 1) cos α + i sin α = 0. Hence 2 cos α + 1 = 0, i.e. α =
± 2π3 + 2πc, where c ∈ Z.
3k
−1
Let x2 be a root of the polynomial h(x). Since h(x) = xxk −1 , the roots
of the polynomial h(x) are distinct and they are x2 = cos 2πs 2πs
3k + i sin 3k ,
where s = 3a ± 1, a ∈ Z. It is enough to prove that f (x2 ) = 0. We have



f (x2 ) = x2n2 +x n
2 +1 = cos(4sα)+sin(4sα) + cos(2sα)+sin(2sα) +1 =



2 cos(2sα) + 1 cos(2sα) + i sin(2sα) = 0 (since 2 cos(2sα) + 1 =
2 cos 2s(± 2π 4πs 4π




3 + 2πc) + 1 = 2 cos 3 + 1 = 2 cos 3 (3a ± 1) + 1 = 0).
Problem 2. It is well known that an ellipse might be defined by a
focus (a point) and a directrix (a straight line), as a locus of points
such that the distance to the focus divided by the distance to directrix
is equal to a given number e < 1. So, if a point X belongs to both
ellipses with the same focus F and directrices l1 .l2 , then e1 .l1 X = F X =
e2 .l2 X (here we denote by l1 X, l2 X distances between the corresponding
line and the point X). The equation e1 .l1 X = e2 .l2 X defines two lines
whose equations are linear combinations with coefficients e1 , ±e2 of the
normalized equations of lines l1 , l2 but of those two only one is relevant,
since X and F should lie on the same side of each directrix. So, we have
that all possible points lie on one line. The intersection of a line and an
ellipse consists of at most two points.
Problem 3.  As is known, theFibonacci numbers Fn can be expressed
√
√

1 1+ 5 n 1− 5 n
as Fn = 5 √ − . Expanding this expression, we obtain
2
2

1 n n n l−1



that Fn = 2n−1 1 + 3 5 + · · · + l 5 2 , where l is the greatest odd
numbers such that l 6 n and s = l−1 n
2 6 2.
s
1 n n

k n−1


5k .
P P
So, Fn = 2n−1 2k+1 5 , which implies that 2 divides 2k+1
k=0 06k6 n2
Problem 4. Let f (x) = g(x)h(x) where h(x) is a polynomial with
integer coefficients.
Let a1 , . . . , a81 be distinct integer roots of the polynomial f (x) −
2.15. Solutions of Olympic 2008 177

2008. Then f (ai ) = g(ai )h(ai ) = 2008 for i = 1, . . . , 81. Hence,

g(a1 ), . . . , g(a81 ) are integer divisors of 2008.
Since 2008 = 23 .251 (2, 251 are primes) then 2008 has exactly 16
distinct integer divisors (including the negative divisors as well). By
the pigeonhole principle, there are at least 6 equal numbers among
g(a1 ), . . . , g(a81 ) (because 81 > 16.5). For example, g(a1 ) = g(a2 ) =
· · · = g(a6 ) = c. So g(x) − c is a nonconstant polymial which has at least
6 distinct roots (namely a1 , . . . , a6 ). Then the degree of the polynomial
g(x) − c is at least 6.
Problem 5. Call a square matrix of type (B), if it is of the form
0 b12 0 . . . b1,2k−2 0
 
 b21 0 b23 . . . 0 b2,2k−1 
 0 b32 0 . . . b3,2k−2 0 
 . .. .. ... .. ..
 ..

. . . . 
b2k−2,1 0 b2k−2,3 . . . 0 b2k−2,2k−1
 
0 b2k−1,2 0 . . . b2k−1,2k−2 0
Note that every matrix of this form has determinant zero, because it has
k columns spanning a vector space of dimension at most k − 1.
Call a square matrix of type (C), if it is of the form
0 c11 0 c12 . . . 0 c1,k
 
 c11 0 c12 0 . . . c1,k 0 
 0 c21 0 c22 . . . 0 c2,k 
C0 =  c21 0 c22 0 . . . c2,k 0 
 
 .. .. .. .. . . . .. .. 
 . . . . . . 
 0 c 0 ck,2 . . . 0 ck,k 
k,1
ck,1 0 ck,2 0 . . . ck,k 0
By permutations of rows and columns, we see that

C 0
 
0
| det C | = det 0 C = | det C|2 ,

where C denotes the k × k-matrix with coefficients ci,j . Therefore, the

determinant of any matrix of type (C) is a perfect square (up to a sign).
Now let X 0 be the matrix obtained from A by replacing the first
row by (1 0 0 . . . 0), and let Y be the matrix obtained from A by
2.15. Solutions of Olympic 2008 178

replacing the entry a11 by 0. By multi-linearity of the determinant,

det(A) = det(X 0 ) + det(Y ). Note that X 0 can be written as
1 0
 
0
X = v X

for some (n − 1) × (n − 1)-matrix X and some column vector v. Then

det(A) = det(X) + det(Y ). Now consider two cases. If n is old, then X
is of type (C), and Y is of type (B). Therefore, | det(A)| = | det(X)| is a
perfect square. If n is even, then X is of type (B), and Y is of type (C);
hence | det(A)| = | det(Y )| is a perfect square.
The set of primes can be replaced by any subset of {2}∪{3, 5, 7, 9, 11, . . . }.
Problem 6. It is clear that, if B is an orthonormal system in a Hilbert
space H, then { √d2 e : e ∈ B} is a set of points in H, any two of which
are at distance d apart. We need to show that every set S of equidistant
points is a translate of such a set.
We begin by noting that, if x1 , x2 , x3 , x4 ∈ S are four distinct points,
then

x2 − x1 , x2 − x1 = d2 ,
1
1
x2 − x1 , x3 − x1 = kx2 − x1 k2 + kx3 − x1 k2 − kx2 − x3 k2 = d2 ,
2 2
x2 − x1 , x4 − x3 = x2 − x1 , x4 − x1 − x2 − x1 , x3 − x1
1 1
= d2 − d2 = 0.
2 2
This shows that scalar products among vectors which are finite linear
combinations of the form

λ1 x1 + λ2 x2 + · · · + λn xn ,

where x1 , x2 , . . . , xn are distinct points in S and λ1 , λ2 , . . . λn are integers

with λ1 + λ1 + · · · + λ1 = 0, are universal across all such sets S in all
Hilbert spaces H; in particular, we may conveniently evaluate them using
examples of our choosing, such as the canonical example above in Rn . In
fact this property trivially follows also when coefficients λi are rational,
and hence by continuity any real numbers with sum 0.
2.15. Solutions of Olympic 2008 179

If S = {x1 , x2 , . . . , xn } is a finite set, we form

1
x= (x1 + x2 + · · · + xn ),
n
pick a nonzero vector z ∈ [Span(x1 − x, x2 − x, . . . , xn − x)]⊥ and seek y
in the form y = x + λz for a suitable λ ∈ R. We find that

x1 − y, x2 − y = x1 − x − λz, x2 − x − λz = x1 − x, x2 − x + λ2 kzk2 ,

x1 − x, x2 − x may be computed by our remark above as

* +
2  ⊥ ⊥
d 1 1 1 1   1 1 1 1 
x1 − x, x2 − x = − 1, , , . . . , , , − 1, , . . . ,
2 n n n n n n n n
Rn
d2 2 1
   n−2  d2
= −1 + 2 =− .
2 n n n 2n
√
d 2
So the choice λ = √ will make all vectors (xi − y) orthogonal
2nkzk d
to each other; it is easily checked as above that they will also be of length
one.
Let now S be an infinite set. Pick an infinite sequence

T = {x1 , x2 , . . . , xn , . . . }

of distinct points in S. We claim that the sequence

1
(x1 + x2 + · · · + xn )
yn =
n
is a Cauchy sequence in H. (This is the crucial observation). Indeed, for
m > n, the norm kym − yn k may be computed by the above remark as

2 d2  1 1 1 1 1 1 > 2
kym − yn k = − ,..., − , ,
2 m n m n m m Rm
2 2
d n(m − n) m−n 
= +
2 m2 n2 m2
2
d (m − n)(m − n + n) d2 m − n
= =
2 m2 n 2 mn
2
d 1 1 
= − → 0, m, n → ∞.
2 n m
2.15. Solutions of Olympic 2008 180

By completeness of H, it follows that there exists a limit

y = lim yn ∈ H.
n→∞

We claim that y satisfies all conditions of the problem. For m > n > p,
with n, p fixed, we compute
2 d2  1 1 1 1 1 > 2
kxn − ym k = − ,...,− ,1 − ,− ,...,−
2 m m m m m Rm
d2 h m − 1 (m − 1)2 i d2 m − 1 d2
= + = → , m → ∞,
2 m2 m2 2 m 2
showing that kxn − yk = √d2 , as well as
*
d2  1 1 1 1 ⊥
xn − ym , xp − ym = − ,...,− ,...,1 − ,...,− ,
2 m m m m
+
 1 1 1 1  ⊥
− ,...,1 − ,...,− ,...,−
m m m m
Rm
d2 h m − 2 2 1
i d2
= − 1− =− → 0, m → ∞,
2 m2 m m 2m
showing that xn − y, xp − y = 0, so that
(√ )
2
(xn − y) : n ∈ N
d
is indeed an orthonormal system of vectors.
This completes the proof in case when T = S, which we can always
take if S is countable. If it is not, let x0 , x00 be any two distinct points in
S \ T . Then applying the above procedure to the set

T 0 = {x0 , x00 , x1 , x2 , . . . , xn , . . . }

it follows that
x0 + x00 + x1 + x2 + · · · + xn x1 + x2 + · · · + xn
lim = lim =y
n→∞ n+2 n→∞ n
satisfies that
(√ √ ) (√ )
2 0 2 00 2
(x − y), (x − y) ∪ (xn − y) : n ∈ N
d d d
2.15. Solutions of Olympic 2008 181

is still an orthonormal system.

This it true for any distinct x0 , x00 ∈ S \ T ; it follows that the entire
system (√ )
2
(x − y) : x ∈ S
d
is an orthonormal system of vectors in H, as required.