Problem j12-I-1/j12-I-4. Differentiable functions f1 , . . .

, fn : R → R are linearly indepen-

dent. Prove that there exist at least n − 1 linearly independent functions among f10 , . . . , fn0 .
(Eötvös Loránd University, Budapest)
Solution. Select a maximal independent set from the derivatives. Without loss of
generality, it can be assumed that this set is f10 , . . . , fm
0
, where m ≤ n. If m ≤ n − 2, then
fn−1 and fn can be expressed as a linear combination of f10 , . . . , fm
0 0 0
; hence, there exist real
numbers a1 , . . . , am , b1 , . . . , bm such that

m m
!0
X X
ai fi0 − fn−1
0
= ai fi − fn−1 =0
i=1 i=1
and
m m
!0
X X
bi fi0 − fn0 = bi fi − fn = 0.
i=1 i=1

Pm Pm
This implies that functions i=1 ai fi − fn−1 and i=1 bi fi − fn are constant. Eliminating
these constants, a linear combination of f1 , . . . , fn is found which vanishes. 

j12-I-1/j12-I-4-1
 2p
Problem j12-I-2/j12-I-9. Let p > 3 be a prime number and n = 2 3−1 . Show that
n divides 2n − 2. (Jagiellonian University in Kraków)
2p
Solution. n = 2 3−1 = 4p−1 +4p−2 +· · ·+1. Hence, in the binary system, n = 1010 . . . 101
(number of 1’s is p). Therefore, in the binary system,

(∗) 2n − 2 = 1111 . . . 110 (number of 1’s is n − 1),

(∗∗) 3n = 1111 . . . 111 (number of 1’s is 2p).

Now if we prove that 2p divides n − 1, then by (∗), (∗∗) and by the rules of multiplication
in the binary system, we will get that 3n divides 2n − 2 — just what we need. But now
observe:
2p | (n − 1) ⇐⇒ (n is odd†) ⇐⇒ p | (n − 1) ⇐⇒
  22p − 1    22p − 4 
⇐⇒ p  − 1 ⇐⇒ p  ⇐⇒
 
3 3
⇐⇒ (p > 3 and prime) ⇐⇒ p | 22p − 4 ⇐⇒


  22p − 4 
⇐⇒ (p > 3 and prime) ⇐⇒ p  ⇐⇒

4
⇐⇒ p | 22p−2 − 1 .

But now from Fermat’s small theorem (p prime and p does not divide a, then ap−1 − 1 ≡ 0
(mod p)), we have 2p−1 ≡ 1 (mod p), hence (2p−1 )2 ≡ 12 (mod p) and finally 22p−2 ≡ 1
(mod p). 

† The sentences in parentheses serve only as justifications of the stated equivalences here.
Thus, e.g., 2p | (n − 1) ⇔ (n is odd) ⇔ p | (n − 1) should be read as “2p divides (n − 1) if
and only if p divides (n − 1) because n is odd” and so on.

j12-I-2/j12-I-9-1
Problem j12-I-3/j12-II-59. Positive numbers x1 , . . . , xn satisfy

1 1 1
+ + ··· + = 1. (1)
1 + x1 1 + x2 1 + xn

Prove that
 
√ √ √ 1 1 1
x1 + x2 + · · · + xn ≥ (n − 1) √ + √ + · · · + √ .
x1 x2 xn

(University of Ostrava)
Solution. It is sufficient to prove that
       
√ 1 √ 1 √ 1 1 1 1
x1 + √ + x2 + √ + ··· + xn + √ ≥ n √ + √ + ··· + √
x1 x2 xn x1 x2 xn

or equivalently (see (1))

    
1 + x1 1 + xn 1 1 1 1 1
√ + ··· + √ + ··· + ≥ n √ + √ + ··· + √ . (2)
x1 xn 1 + x1 1 + xn x1 x2 xn
√
Consider the function f (x) = x + √1x = x+1√ , x ∈ (0, +∞). It is easy to check that the
x
function f is non-decreasing on [1, +∞) and that

f (x) = f ( x1 ) (3)

holds for each x > 0.

Further, it follows from (1) that only x1 can be less than 1 (i.e. xk ≥ 1, k = 2, 3, . . .)
1 1 x1
and 1+x 2
≤ 1 − 1+x 1
= 1+x 1
. Hence
1
x2 ≥ (4)
x1
(a contradiction otherwise). It is now apparent directly (if x1 ≥ 1) or from (3) and (4) (if
x1 < 1) that
f (x1 ) = f ( x11 ) ≤ f (x2 ) ≤ · · · ≤ f (xn ).
n on
This means that the sequence 1+x √ k
xk is non-decreasing. Thus (2) holds according to
k=1  1 n
the well-known Chebyshev’s inequality since the sequence 1+x k k=1
is decreasing.
The equality in (2) holds if and only if

1 1 1 1 + x1 1 + x2 1 + xn
= = ··· = or √ = √ = ··· = √ ,
1 + x1 1 + x2 1 + xn x1 x2 xn

which implies x1 = x2 = · · · = xn . Then we obtain from (1) that x1 = x2 = · · · = xn =

n − 1. 

j12-I-3/j12-II-59-1
Problem j12-I-4/j12-I-5. The numbers 1, 2, . . . , n are assigned to the vertices of a regular
n-gon in an arbitrary order. For each edge compute the product of the two numbers at the
endpoints and sum up these products. What is the smallest possible value of this sum?
(Babeş-Bolyai University, Cluj-Napoca)
Solution. Due to the (a−b)2 = a2 −2ab+b2 identity, it is sufficient to find the maximum
of the sum
n
X
2
σ(k + 1) − σ(k)
k=1

where σ(k) denotes the number from the k th vertex and σ(n + 1) = σ(1). We will give an
inductive algorithm to find an optimal arrangement and so we can find the maximal sum
(or the minimal for the initial problem). Suppose we have an arbitrary arrangement with n
numbers and construct an arrangement with n + 2 numbers in the following way:
  
• Find the maximum
 of σ(k + 1) − σ(k). For such a k, denote x = min σ(k + 1), σ(k)

and y = max σ(k + 1), σ(k) .
• Increase each number by 1.
• Insert the numbers 1 and n + 2 as in figure 1.
If we denote by sn+2 and sn the corresponding distance sums, we have:

2
sn+2 = sn − (x − y)2 + (n + 1) − x + (n + 1)2 + y 2
= sn + 2(n + 1)2 + 2xy − 2x − 2nx.

On the other hand, from the obvious inequalities x ≥ 1 and n + 1 − y ≥ 1, we have

x(n + 1 − y) ≥ 1 and this implies 2xy − 2x − 2nx ≤ −2. Hence

sn+2 = sn + 2(n + 1)2 − 2n = 2n(n + 2).

If yn is the maximal sum, we have yn+2 = yn + 2n(n + 2) (because for

 n = 3 in the maximal

arrangement x = 1, y = 3 and in each step the maximal distance σ(k + 1) − σ(k) occurs
at x = 1 and y = n). For n = 2 and n = 3, we have y2 = 2 and y3 = 6 so from the obtained
recurrence relation we can deduce y2n = 2 + 38 (n − 1)n(n + 1) and thus
Pn
2 k=1 k 2 − y2n 4n3 + 6n2 + 5n − 3
x2n = =
2 3
where xn denotes the minimal sum for the initial problem. Analogously we have

4n3 + 12n2 + 14n + 3

x2n+1 = .
3

For n ∈ {6, 7, 8, 9, 10}, we have illustrated the optimal arrangements on figures 2, 3, 4, 5

(in the exterior we have written the arrangement’s numbers, inside the circle the product of
any two adjacent number and in the inside circle the sum of these products).
Pn Remark. For p >
p 1, the above arrangements will give the maximum of the sum
k=1 σ(k + 1) − σ(k) , and this can be proved by the same method using the inequality

(n + 1)p + (n + 1 − x)p + y p − (y − x)p ≤ np + (n + 1)p + np − (n − 1)p .

j12-I-4/j12-I-5-1
j12-I-4/j12-I-5-2
Problem j12-II-1/j12-II-56. Find all complex solutions of the system
(a + ic)3 + (ia + b)3 + (−b + ic)3 = −6 ,
(a + ic)2 + (ia + b)2 + (−b + ic)2 = 6 ,
(1 + i)a + 2ic = 0 .

(P. J. Šafárik University in Košice)

Solution. Let us notice that the third equation can be written as
(a + ic) + (ia + b) + (−b + ic) = 0;
that is why a natural substitution is
x = a + ic, y = ia + b, z = −b + ic.
Then, our system is
x3 + y 3 + z 3 = −6
x2 + y 2 + z 2 = 6
x +y +z =0

Using symmetric polynomials, we get

x + y + z = σ1 ,
x + y + z = (x + y + z)2 − 2(xy + yz + xz) = σ12 − 2σ2 ,
2 2 2

x3 + y 3 + z 3 = (x + y + z)3 − 3(xy + yz + xz)(x + y + z) + 3xyz = σ13 − 3σ1 σ2 + 3σ3 .

It is a well-known fact that x, y, z must be roots of the cubic polynomial
f (t) = t3 − σ1 t2 + σ2 t − σ3 .
Since σ1 = 0, σ12 − 2σ2 = 6, σ13 − 3σ1 σ2 + 3σ3 = −6, we have
σ1 = 0, σ2 = −3, σ3 = −2.
Rational roots of the polynomial f (t) = t3 − 3t + 2 can only be from the set {−2, −1, 1, 2}.
Trying these, it turns out that t = 1 and t = −2 are roots. Decomposition of the polynomial
then reveals that 1 is a double root.
Thus, we have 
(x, y, z) ∈ (1, 1, −2), (1, −2, 1), (−2, 1, 1) .
Returning back, we solve the system
a + ic = x
ia + b =y
−b + ic = z

Its determinant is  
1 0 i 

|A| =  i 1 0  = i + 1 6= 0,
 0 −1 i
so for each (x, y, z) there is exactly one solution. It is easy to get the inverse matrix:
 
1 + i −1 − i −1 − i
1
A−1 =  1 − i 1 + i −1 + i  .
2
−1 − i 1−i 1−i
Multiplying this matrix by the vectors (x, y, z) gives three solutions (a, b, c):
(1 + i, 2 − i, −1), (1 + i, −1 − i, −1), (−2 − 2i, −1 + 2i, 2).
One can easily verify that all three satisfy the system. 

j12-II-1/j12-II-56-1
Problem j12-II-2/j12-II-52. A ring R (not necessarily commutative) contains at least
one zero divisor and the number of zero divisors is finite. Prove that R is finite.
(Eötvös Loránd University, Budapest)
Solution. Let m be the number of zero divisors and u, v ∈ R two non-zero elements such
that uv = 0.
We generate more zero divisors in the following way. For an arbitrary x ∈ R, the element
xu is either 0 or also a zero divisor, since (xu)v = x(uv) = 0.†
If xu = yu for some different elements x, y ∈ R, then (x − y)u = 0, and x − y is a zero
divisor. This implies that 0 or an arbitrary zero divisor can be obtained at most m + 1 times
in the form xu.‡
Thus, each of 0 and the m zero divisors is obtained at most m times and the number of
elements of R cannot exceed (m + 1)2 . 

† The set {xu; x ∈ R} is finite, its cardinality being ≤ m + 1.

‡ Define an equivalence relation: x ∼ y iff xu = yu. In each class of equivalence, there
are (m + 1) elements at most. Finally, the number of the classes of equivalence is equal to
the cardinality of the set {xu; x ∈ R}, which is finite.

j12-II-2/j12-II-52-1
Problem j12-II-3/j12-II-53. Let E be the set of all continuous functions u: [0, 1] → R
satisfying Z t
u2 (t) ≤ 1 + 4
 
su(s) ds, ∀t ∈ [0, 1].
0

Let ϕ: E → R be defined by
Z 1
u2 (x) − u(x) dx.


ϕ(u) =
0

Prove that ϕ has a maximum value and find it. (Babeş-Bolyai University, Cluj-Napoca)
Solution. Let Z t  
v(t) = 1 + 4 su(s) ds, ∀t ∈ [0, 1].
0

We have s
Z t p
v 0 (t) = 4tu(t) ≤ 4t
   
1+4 su(s) ds ≤ 4t v(t)
0

so
t t
v 0 (s)
p Z Z
v(t) − 1 = p ds ≤ 2s ds = t2
0 2 v(s) 0

therefore  p
u(t) ≤ v(t) ≤ t2 + 1.


If we consider ϕ, we have
 2
u (t) − u(t) = u(t)u(t) − 1 ≤ (t2 + 1)(t2 + 2),
   
Z 1 Z 1
 2 16
(t2 + 1)(t2 + 2) dt =
  
ϕ(u) ≤ u (t) − u(t) dt ≤ .
0 0 5

Equality can be achieved if

u(t) = t2 + 1 u(t) − 1 = t2 + 2.
   
and

This is the case of u(t) = −t2 − 1, which belongs to E. 

j12-II-3/j12-II-53-1
Problem j12-II-4/j12-II-62. Prove that
!
1 √ π2
Z
2 n
lim n 1+ xn dx − 1 = .
n→∞ 0 12

(Sofia University St. Kliment Ohridski)

Solution. We will prove that
∞
!
1 √ (−1)k−1 π2
Z X
2 n
lim n 1+ xn dx − 1 = = .
n→∞ 0 k2 12
k=1

R1 √
P∞ α
k
Let an = n2 0 n 1 + xn dx − 1 . It  αis
kwidely known that (1 + t)α
= k=0

k t  for
any t ∈ [0, 1] and α ∈ (0, 1). Moreover,  k t  = (−1)k−1 αk tk and  αk tk  ≥  αk tk+1  for


k ≥ 1, t ≥ 0 and α ∈ (0, 1). Thus, the following inequalities hold:

2p   2p+1
X 
X α k α α k
t ≤ (1 + t) ≤ t .
k k
k=0 k=0

1
Let us put t = xn and α = n. Integrating on [0, 1], we obtain

2p  1 2p+1
√
 Z X 
X 1/n 1 n 1/n 1
≤ 1+ xn dx ≤ .
k nk + 1 0 k nk + 1
k=0 k=0

Hence,
2p    
X 1/n 1 1/n 1
0 ≤ an − n2 ≤ n2 .
k nk + 1 2p + 1 n(2p + 1) + 1
k=1

A simple calculation gives the following estimation:

 
2 1/n 1 1
n ≤ .
2p + 1 n(2p + 1) + 1 (2p + 1)2

Consequently, as n tends to infinity,

2p
!
X (−1)k 1
0 ≤ lim sup an − ≤
n→∞ k2 (2p + 1)2
k=1

and
2p
!
X (−1)k 1
0 ≤ lim inf an − ≤ .
n→∞ k2 (2p + 1)2
k=1

Letting p → ∞, we obtain the desired result. 

j12-II-4/j12-II-62-1
Problem j13-I-1/j13-I-15. Let d(k) be the number of
all natural divisors of a number
∞
k ∈ N. Prove that for any n0 ∈ N the sequence d(n2 + 1) n=n is not strictly monotone.
0
(Vilnius University)
Solution. Note that d(n2 + 1) < n for all even n. Indeed, the number n2 + 1 is not
square and so it is possible to split the set of all its divisors into pairs { d, (n2 + 1)/d } where
d < n and d is odd. The number of divisors in all such pairs does not exceed n.
Let us assume that starting from some n0 ∈ N, the sequence is strictly monotone. For
d(n2 + 1) is always even, we get

d (n + 1)2 + 1 ≥ d n2 + 1 + 2

or, in general,
d (n + k)2 + 1 ≥ d n2 + 1 + 2k

for any natural numbers n ≥ n0 and k ≥ 1. Let N ≥ n0 (e.g., N = n0 ). Taking any

s ≥ N − d(N 2 + 1) (such that N + s is even), we get

d (N + s)2 + 1 ≥ d N 2 + 1 + 2s ≥ N + s,

which is a contradiction with d (N + s)2 + 1 < N + s. 

j13-I-1/j13-I-15-1
Problem j13-I-2/j13-I-19. Let A = [ai,j ] be anPm × n real matrix with at least one
n
non-zero element. For each i ∈ {1, . . . , m} let Ri := j=1 ai,j (the sum of the i-th row of A)
Pm
and for each j ∈ {1, . . . , n} let Cj := i=1 ai,j (the sum of the j-th column of A). Prove
that there exist indices k ∈ {1, . . . , m} and l ∈ {1, . . . , n} such that

ak,l > 0 , Rk ≥ 0 , Cl ≥ 0 ,
or
ak,l < 0 , Rk ≤ 0 , Cl ≤ 0 .

(University of Zagreb)
Solution. Consider the following sets of indices (some of them may be empty):

I + :=

i ∈ {1, . . . , m} | Ri ≥ 0 ,
I−

:= i ∈ {1, . . . , m} | Ri < 0 ,
J+

:= j ∈ {1, . . . , n} | Cj > 0 ,
J−

:= j ∈ {1, . . . , n} | Cj ≤ 0 .

Suppose that the statement of the problem does not hold. Then (but not equivalently) we
+ + − −
have ai,j ≤ 0 for every
P (i, j) ∈ I × J and we have ai,j ≥ 0 for every (i, j) ∈ I × J . Let
us write the sum (i,j)∈I − ×J + ai,j in two different ways:

X n
X X X  X X
ai,j = ai,j − ai,j = Ri − ai,j ≤ 0 ,
(i,j)∈I − ×J + i∈I − j=1 j∈J − i∈I − (i,j)∈I − ×J −
X m
X X X  X X
ai,j = ai,j − ai,j = Cj − ai,j ≥ 0 .
(i,j)∈I − ×J + j∈J + i=1 i∈I + j∈J + (i,j)∈I + ×J +

P
Therefore, ai,j = 0 and we have only equalities in the two formulae above.
(i,j)∈I − ×J +P
This is only possible if i∈I − Ri = 0 and j∈J + Cj = 0, so I − = ∅ and J + = ∅,† which
P
means Ri ≥ 0 for all i = 1, . . . , m and Cj ≤ 0 for all j = 1, . . . , n. Moreover, from
m
X X n
m X n X
X m n
X
0≤ Ri = ai,j = ai,j = Cj ≤ 0,
i=1 i=1 j=1 j=1 i=1 j=1

we conclude Ri = 0 for i = 1, . . . , m and Cj = 0 for j = 1, . . . , n. Since A is a non-zero

matrix, there are indices k and l such that ak,l 6= 0, but Rk = 0 and Cl = 0, which leads to
a contradiction with the assumption that the statement of the problem is false. 

† If I − 6= ∅, then (i,j)∈I − ×J + ai,j ≤ i∈I − Ri < 0 — a contradiction. We can argue

P P

similarly to show J + = ∅.

j13-I-2/j13-I-19-1
Problem j13-I-3/j13-I-9. Find the limit
s r q √
lim 1+2 1+3 · · · + (n − 1) 1 + n .
n→∞

(Dr. Moubinool Omarjee, Paris†)

Solution. Let
s r q √
um,n = 1 + m 1 + (m + 1) · · · + (n − 1) 1 + n .

We have
u2m,n = 1 + mum+1,n ,
u2m,n − (m + 1)2 = m um+1,n − (m + 2) .

Using the equality |a − b| = |a2 − b2 |/|a + b| and inequality um,n + m + 1 ≥ m + 2, we get

 
um,n − m − 1 ≤ m  
um+1,n − (m + 2).
m+2

We deduce that
2 3 n−1
|u2,n − 3| ≤ · · ··· · · |un−1,n − n|,
4 5 n+1
√
q   
6 1
|u2,n − 3| ≤ 1 + (n − 1) 1 + n − n = O .
n(n + 1) n

So we get
lim u2,n = 3.
n→∞

† This problem is formally proposed by the University of Ostrava.

j13-I-3/j13-I-9-1
Problem j13-I-4/j13-I-12. Let A and B be complex hermitian 2 × 2 matrices with pairs
of eigenvalues (α1 , α2 ) and (β1 , β2 ), respectively. Determine all possible pairs (γ1 , γ2 ) of
eigenvalues of the matrix C = A + B. (A matrix A = [ai,j ] is hermitian if and only if
ai,j = aj,i for all i, j.) (Charles University in Prague)
Solution. Recall that all eigenvalues of a hermitian matrix are real numbers and that
there exists an orthonormal basis consisting of eigenvectors of the matrix. As we can add a
suffitiently large multiple of the identity matrix to both matrices A and B, we can suppose
wlog that α1 , α2 , β1 , β2 > 0 and also γ1 , γ2 > 0.
Let us also wlog suppose α1 ≥ α2 , β1 ≥ β2 , γ1 ≥ γ2 and α1 − α2 ≥ β1 − β2 . By easy
arguments, we can see

γ1 + γ2 = Tr C = Tr A + Tr B = α1 + α2 + β1 + β2 .

Further, it holds that

γ1 ≤ α1 + β1 , γ2 ≥ α2 + β2 .
(The first inequality can be seen if we rewrite it slightly: γ1 = kCk ≤ kAk + kBk = α1 + β1 .
The second inequality follows if we consider the equality above and the first inequality
together. — Alternatively, γ1 = max(Cx, x)/(x, x) ≤ max(Ax, x)/(x, x) + max(Bx, x)/
(x, x) = α1 + β1 and γ2 = min(Cx, x)/(x, x) ≥ min(Ax, x)/(x, x) + min(Bx, x)/(x, x) =
α2 + β2 .) Later we will also prove the inequalities

γ1 ≥ α1 + β2 , γ2 ≤ β1 + α2

(in fact, it suffices to prove only the first one because the second one follows if we use the
equality given above).
From these inequalities, we can see that γ1 ∈ [α1 + β2 , α1 + β1 ]. (The value of γ2 has
to be “complementary” to obtain the right value of the sum γ1 + γ2 . It also worths noting
that even if γ1 = α1 + β2 , then still γ1 ≥ γ2 = β1 + α2 . This follows from the assumption
α1 − α2 ≥ β1 − β2 .) We will show that γ1 can assume any value from the given interval
[α1 + β2 , α1 + β1 ]. Consequently, the set of all possible pairs (γ1 , γ2 ) of eigenvalues of the
matrix C = A + B is

(γ1 , γ2 ) : α1 + β2 ≤ γ1 ≤ α1 + β1 , γ1 + γ2 = α1 + α2 + β1 + β2 .

To see this, let us put

     
α1 0 β1 0 cos t sin t
A= , B= , P (t) = .
0 α2 0 β2 − sin t cos t

The matrix A obviously has eigenvalues (α1 , α2 ). The matrix B(t) = P −1 (t)BP (t) obviously
has eigenvalues (β1 , β2 ). If we note that P −1 (t) = P T (t) and define the matrix C(t) =
A + B(t), we have
   
α1 + β1 0 α1 + β2 0
C(0) = A + B = , C( π2 ) = .
0 α2 + β2 0 α2 + β1

The matrix C(0) has the eigenvalue γ1 (0) = α1 + β1 . (Note that γ1 (0) ≥ γ2 (0) = α2 + β2 .)
The matrix C(π/2) has the eigenvalue γ1 (π/2) = α1 + β2 . (Note that γ1 (π/2) ≥ γ2 (π/2) =
α2 + β1 .) As both eigenvalues (γ1 , γ2 ) of a matrix C depend continuously on the coefficients
of the matrix, we deduce that γ1 (t) is a continuous function. Consequently, it assumes every
value from the interval [α1 + β2 , α1 + β1 ], which we wanted to demonstrate.
Now it only remains to prove the inequality γ1 ≥ α1 + β2 for any two complex hermitian
matrices A and B. Let us recall that we still wlog suppose α1 ≥ α2 > 0, β1 ≥ β2 > 0
and γ1 ≥ γ2 > 0. Let v1 and v2 denote the eigenvectors of the matrix A corresponding
to the eigenvalues α1 and α2 , respectively, and let w1 and w2 denote the eigenvectors of B
corresponding to the eigenvalues β1 and β2 , respectively. We can suppose that the bases

j13-I-4/j13-I-12-1
 u11 u12


{v1 , v2 } and {w1 , w2 } are orthonormal. So there exists some unitary matrix U = u21 u22
such that
v1 = u11 w1 + u12 w2 , w1 = u11 v1 + u21 v2 ,
and
v2 = u21 w1 + u22 w2 , w2 = u12 v1 + u22 v2 .
We will estimate γ1 in the following way. First,

γ1 = sup kCxk : kxk = 1 ≥ kCv1 k

where k·k denotes the Euclidean norm. (Let us justify the formula. Recall that γ1 =
maxkxk=1 (Cx, x). Obviously, γ12 is the greater eigenvalue of C 2 . Consequently, it follows
that γ12 = maxkxk=1 (C 2 x, x). As C is hermitian, we have (C 2 x, x) = x∗ CCx = x∗ C ∗ Cx =
(Cx, Cx) = kCxk2 .) Second,

Cv1 = (A + B)v1 = α1 v1 + β1 u11 w1 + β2 u12 w2 = (α1 + β2 )v1 + (β1 − β2 )u11 w1 =

= α1 + β2 + (β1 − β2 )u11 u11 v1 + (β1 − β2 )u11 u21 v2 .

As the vectors v1 and v2 are orthonormal and (β1 − β2 )u11 u11 ≥ 0, we conclude
q 2  2
γ1 ≥ kCv1 k = α1 + β2 + (β1 − β2 )u11 u11  + (β1 − β2 )u11 u21  ≥
q 2
≥ α1 + β2 + (β1 − β2 )u11 u11  ≥ α1 + β2 .

j13-I-4/j13-I-12-2
Problem j13-II-1/j13-II-51. Two real square matrices A and B satisfy the conditions
A2002 = B 2003 = I and AB = BA. Prove that A + B + I is invertible. (The symbol I
denotes the identity matrix.) (University of Belgrade)
Solution. Let (A + B + I)x = 0 for some vector x, i.e., (B + I)x = −Ax. Then
we have −A2 x = A(B + I)x = (B + I)Ax = −(B + I)2 x, and, continuing in this way,
(B + I)k x = (−1)k Ak x. As A2002 = I, we get (B + I)2002 x = x, i.e.,

(B + I)2002 − I x = B 2003 − I x = 0.

(Recall B 2003 = I.) In other words, taking that p(t) = (t + 1)2002 − 1 and q(t) = t2003 − 1
are polynomials, we have just got

p(B)x = q(B)x = 0.

But, since 2003 is a prime, q(t)/(t − 1) is a primitive polynomial for all its roots, and
therefore none of them is a root of the another monic polynomial p(t) of degree 2002; further,
the remained root t = 1 of q(t) is not a root of p(t), which implies that p(t) and q(t) are
coprime.†
Since there exist non-zero polynomials r(t) and s(t) such that r(t)p(t) − s(t)q(t) = 1
(recall the Euclidean algorithm), we can conclude that x = r(B)p(B)x − s(B)q(B)x = 0,
and so A + B + I must be invertible indeed. 

† The polynomials p(t) and q(t) are really coprime (i.e. relatively prime). Here is another
argument: Every polynomial (of degree ≥ 1) can be written as a product of factors of
Q2002
degree 1. In particular, p(t) = (t + 1)2002 − 1 = k=1 (t − zp,k ) and q(t) = t2003 − 1 =
Q2003
k=1 (t − zq,k ), where zp,1 , . . . , zp,2002 and zq,1 , . . . , zq,2003 are the roots of the polynomial
p and q, respectively. Obviously, the polynomials p and q are relatively prime iff they have
no root in common.
It is easy to see that the roots of q lie on the unit circle in the complex plane. Similarly,
it is easy to see that all roots of p are on the circle with radius 1 and its centre at the
point −1.
Thus, the intersections of the two circles,
√ √
2 2
= cos ± 3π 3π π π



2 ±i 2 2 + i sin ± 2 = −1 + cos ± 2 + i sin ± 2 ,

are the only possible common roots of q and p. But none of these two points is a root of q.
It follows that p and q are coprime.

j13-II-1/j13-II-51-1
Problem j13-II-2/j13-I-17. Let {D1 , D2 , . . . , Dn } be a set of disks (a disk is a circle with
its interior) in the Euclidean plane and aij = S(Di ∩ Dj ) be the area of Di ∩ Dj . Prove that
for any numbers x1 , x2 , . . . , xn ∈ R the following inequality holds:
n X
X n
aij xi xj ≥ 0.
i=1 j=1

(Warsaw University)
Solution. Let χDi : R2 → {0, 1} be the characteristic function of the set Di :

1, if (x, y) ∈ Di ,
χDi (x, y) =
0, if (x, y) ∈
/ Di .

We have:
χDi ∩Dj = χDi χDj ,
Z Z
S(Di ) = χDi (x, y) dx dy = χ2Di (x, y) dx dy,
R2 R2
Z Z
S(Di ∩ Dj ) = χDi ∩Dj (x, y) dx dy = χDi (x, y)χDj (x, y) dx dy.
R2 R2

Thus,

X n
n X Z n X
X n
aij xi xj = xi χDi (x, y)xj χDj (x, y) dx dy =
i=1 j=1 R2 i=1 j=1
Z

2
= x1 χD1 (x, y) + · · · + xn χDn (x, y) dx dy ≥ 0.
R2

j13-II-2/j13-I-17-1
Problem j13-II-3/j13-II-70. A sequence (an )∞ n=0 of real numbers is defined recursively
by
an
a0 := 0, a1 := 1, an+2 := an+1 + n , n ≥ 0.
2
Prove that
∞
X 1
lim an = 1 + n(n−1) Qn .
n→∞
n=1 2 2 · k=1 (2k − 1)
(University of Zagreb)
Remark. In fact, we will prove the following:
∞
(a) The sequence (an )P
n=0 is convergent. Q
∞ n

(b) limn→∞ an = 1 + n=1 1/ 2n(n−1)/2 · k=1 (2k − 1) .
(c) The limit limn→∞ an is an irrational number.
Solution. (a) Obviously, an ≥ 0 for every n ≥ 0. The sequence (an )∞ n=0 is increasing
since an+2 − an+1 = an /2n ≥ 0 for every n ≥ 0. It suffices to show that (an )∞
n=0 is bounded
from above. For each n ≥ 0, we have an+2 ≤ an+1 + an+1 /2n = an+1 (1 + 1/2n ). Using the
inequality between geometric and arithmetic mean, for every n ≥ 1 we obtain
n   n    n !n !n
Y 1 Y 1 1 X 1 n+1
an+2 ≤ 1+ k =2 1+ k ≤2 n+ ≤2 ≤ 2e.
2 2 n 2k n
k=0 k=1 k=1
P∞ p √
(b) Consider the power series n=0 an z n . Since lim supn→∞ n |an | ≤ limn→∞ n 2e = 1,
its radius of convergence is R ≥ 1. Therefore, on the P open unit disc, with center at the
∞ n
origin, it converges
Pn to a holomorphic function f (z) := Pa∞n z . Inductively,
n=0 we

obtain
an+2 = 1 + k=0
ak /2 for any n ≥ 0. So limn→∞ an = 1 + k=0 ak /2k = 1 + f 21 and we
k

have to find f 21 .
Now we use the recurrent relation for (an )∞
n=0 to obtain a functional equation for f . We
multiply an+2 := an+1 + an /2n by z n+2 and sum over all n ≥ 0 to get
∞ ∞ ∞
z n
X X X
n+2 n+1 2


an+2 z =z an+1 z +z an 2 ,
n=0 n=0 n=0
that is
f (z) − z = zf (z) + z 2 f z


2 ,
or
(1 − z)f (z) = z 2 f z


2 +z for |z| < 1. (1)

We substitute z = 1/2n for n = 1, . . . , N (where N ≥ 1 is a fixed number) into (1), then

multiply the n-th equality by some constant sn > 0 and finally sum up those N equalities:

2
1 − 21 f 12 = 12 f 14 + 12 ,




| · s1 ,
2
1 − 14 f 14 = 14 f 18 + 14 ,





| · s2 ,
..
.

2
1 − 2n f 2n = 21n f 2n+1
1

1
1
+ 21n ,


| · sn ,
1

1

1

2 1

1
1 − 2n+1 f 2n+1 = 2n+1 f 2n+2 + 2n+1 , | · sn+1 ,
..
.

2
1 − 2N f 2N = 21N f 2N1+1 + 21N ,
1

1


| · sN ,
s1 1

sN 1

XN sn
f 2 = f 2N +1
+ .
2 22N n=1 2n
1


To obtain the given result (namely, to achieve cancelling of the terms with f 2n for n =
2, . . . , N ), we had to choose the numbers sn so that
1

2
sn+1 = 21n sn ,


1 − 2n+1 for n ≥ 0. (2a)

j13-II-3/j13-II-70-1
Let us put
s0 := 1. (2b)
It follows that s1 = 2. Equalities (2b) and (2a) lead to

n−1 n−1 1 2 n−1

Y sk+1 Y
2k
Y 1 1
sn = = 1 = k−1 k+1
= n(n−1) Qn
sk 1 − 2k+1 2 (2 − 1) −n k
k=0 k=0 k=0 2 2 k=1 (2 − 1)

for every n ≥ 1. Finally, we have

N N
f 2N1+1


1

sN 1

X sn X 1
f 2 = 2N f 2N +1
+ n
= N (N −1) QN + n(n−1) Qn
.
2 n=1
2 2 2 +N (2k − 1) n=1 2 2 (2k − 1)
k=1 k=1

The first term tends to 0 when N → ∞, so

∞
1

X 1
f 2 = n(n−1) Qn . (3)
k
2 k=1 (2 − 1)
2
n=1

(c) The proof of limn→∞ an ∈ R \ Q is based on the fact that the series in (3) converges
“very rapidly”. Suppose that its sum equals pq for some positive integers p and q. For each
integer N ≥ 1, denote

N N
N (N −1) Y X 1
qN := 2 2 (2k − 1), pN := qN n(n−1) Qn .
2 k − 1)
k=1 (2
2
k=1 n=1

Obviously, pN and qN are positive integers. We manage to estimate pqN − qpN . We have

N N
N (N −1) Y N (N −1) Y 2
qN = 2 2 (2k − 1) < 2 2 2k = 2N
k=1 k=1

and
∞ ∞
p pN X 1 X 1
− = n(n−1) n
≤ n(n−1) Qn
=
q qN Q k k−1
n=N +1 2 k=1 (2 − 1) n=N +1 2 k=1 2
2 2

∞ ∞
X 1 X 1 1 1
= ≤ = N 2 +N −1 < N −1 .
2n(n−1) 2m 2 2 qN
n=N +1 m=N (N +1)
q
Thus, 0 < pqN − qpN < 2N −1 ,† so (pqN − qpN )N ≥1 is a sequence of positive integers that
converges to 0. This is a contradiction and we are done. 

† It is easy to see from the definition of the numbers pN that the sequence pqN


N
is strictly
p pN p
increasing to the limit q . Hence qN < q , qpn < pqN , and 0 < pqN − qpN . As the difference
is integer, we have even 1 ≤ pqN − qpN .

j13-II-3/j13-II-70-2
Problem j13-II-4/j13-I-18. Let f, g: [0, 1] → (0, +∞) be continuous functions such that
f and fg are increasing. Prove that

1
Rx Z 1
f (t) dt
Z
0 f (t)
Rx dx ≤ 2 dt.
0 0
g(t) dt 0 g(t)

(University of Zagreb)
Solution. First, we estimate the expression inside the integral sign on the left side of the
given inequality. By the Chebycheff’s inequality for integrals applied to increasing functions
f and fg on the segment [0, x] (where x ∈ (0, 1] is fixed), we get
 Z x  Z x
1 x
 Z
1 1 g(t)
f (t) dt dt ≤ g(t) dt,
x 0 x 0 f (t) x 0
that is,
Rx
f (t) dt x
R0x ≤ R x g(t) (1)
0
g(t) dt dt
0 f (t)

for every x ∈ (0, 1]. From the integral form of the Cauchy-Schwarz inequality on the segment
[0, x], we have
x x 2 x 2
x4
Z Z  Z
g(t) t f (t)
dt dt ≥ t dt = ,
0 f (t) 0 g(t) 0 4
or
Z x 2
1 4 t f (t)
Rx ≤ dt. (2)
g(t)
dt x4 0 g(t)
0 f (t)

From (1) and (2) we obtain

Rx Z x 2
f (t) dt 4 t f (t)
R0x ≤ 3 dt. (3)
0
g(t) dt x 0 g(t)

Finally, it remains to integrate (3) over x ∈ (0, 1] and to reverse the order of integration.
Z 1 Rx Z 1 Z x 2  Z 1 Z 1 2 
0
f (t) dt 4t f (t) 4t f (t)
Rx dx ≤ 3
dt dx = dx dt =
0 0
g(t) dt 0 0 x g(t) 0 t x3 g(t)
Z 1 2 Z 1  Z 1 2  
4t f (t) dx 4t f (t) 1 1
= 3
dt = − dt =
0 g(t) t x 0 g(t) 2t2 2
Z 1 Z 1
f (t) 2

f (t)
=2 1 − t dt ≤ 2 dt.
0 g(t) 0 g(t)

(Remark. The constant 2 on the right hand side of the given inequality is optimal, i.e.,
the least possible. Consider f (t) := 1 and g(t) := t + ε for some fixed ε > 0. Then
Z 1 Rx Z 1 Z 1
f (t) dt x dx
R0x dx = 1 2 dx = 2 = 2 ln(1 + 2ε) − 2 ln 2 − 2 ln ε
0 0
g(t) dt 0 2 x + εx 0 x + 2ε

and Z 1 Z 1
f (t) dt
dt = = ln(1 + ε) − ln ε.
0 g(t) 0 t+ε
The quotient of these two expressions can be made arbitrarily close to 2 since

2 ln(1 + 2ε) − 2 ln 2 − 2 ln ε − ln(1+2ε)

ln ε + ln 2
ln ε + 1
lim = 2 lim = 2.
ε&0 ln(1 + ε) − ln ε ε&0 − ln(1+ε)
ln ε +1

Therefore, the constant 2 is the best possible one.) 

j13-II-4/j13-I-18-1
Category I
Problem 1. Can the set of positive rationals be split into two nonempty disjoint subsets
Q1 and Q2 , such that both are closed under addition, i.e. p + q ∈ Qk for every p, q ∈ Qk ,
k = 1, 2? Can it be done when addition is exchanged for multiplication, i.e. p · q ∈ Qk for every
p, q ∈ Qk , k = 1, 2?

Solution. (a) No. If pq , rs ∈ Qk then of course ps+qr p

qs ∈ Qk . Adding n times q and m times s
r
nps+mqr
gives qs ∈ Qk for all positive integers n, m, hence ñp+m̃r ∈ Qk for all positive integers ñ, m̃.
So if pqkk , srkk ∈ Qk we get that p1 p2 + r1 r2 ∈ Q1 ∩ Q2 .
(b) Yes, for instance
nm o
Q1 = ∈ Q+ : (m, n) = 1 and 2 | n and Q2 = Q+ \ Q1 .
n

1
Problem 2. Alice has got a circular key ring with n keys, n ≥ 3. When she takes it out
of her pocket, she does not know whether it got rotated and/or flipped. The only way she
can distinguish the keys is by colouring them (a colour is assigned to each key). What is the
minimum number of colors needed?

Solution. Clearly at least two colors are needed in any case to distinguish between at least
two keys. For three, four or five keys on the ring, we will show that three colors are necessary.
For six or more keys on the ring, we will show that two colors suffice. Choose one key and
denote it with k1 . Order all other keys in natural order as they follow each other going from k1
around the ring in one direction. For 1 ≤ i ≤ n denote with c(ki ) color of the key ki . Without
loss of generality let c(k1 ) = 1.
Suppose that two colors suffice for n = 3. Then there are two similar possibilities for coloring
the keys. Either c(k2 ) = c(k3 ) = 2 or c(k2 ) = 1. In the first case one can not distinguish between
keys k2 and k3 . In the second case one can not distinguish between keys k1 and k2 . Hence for
n = 3 we need three colors.
Suppose that two colors suffice for n = 4. Then there are four possibilities for coloring the
keys. If c(k2 ) = c(k3 ) = c(k4 ) = 2, then k2 and k4 can not be distinguished (rotation of the key
ring through the line across k1 and k3 interchanges k2 and k4 ). If c(k2 ) = 1 and c(k3 ) = c(k4 ) = 2
then there is a rotation that interchanges k1 and k2 and also interchanges k3 and k4 (similar is
the case when c(k4 ) = 1 and c(k2 ) = c(k3 ) = 2). If c(k3 ) = 1 and c(k2 ) = c(k4 ) = 2 then there is
a rotation that interchanges k1 and k3 and there is also other rotation that interchanges k2 and
k4 . Hence for n = 4 at least three colors are needed. Consider the following coloring: c(k1 ) = 1,
c(k2 ) = 2, c(k3 ) = 3 and c(k4 ) = 1 (one possibility). Keys k1 and k4 have the same color, but
one can distinguish between them since k1 has a neighbor colored with color 1 and a neighbor
colored with color 2, while k4 has also one neighbor colored with color 1, but the other neighbor
is colored with color 3. Hence three colors suffice for n = 4.
Suppose that two colors suffice for n = 5. Then there are two possibilities for coloring the
keys: all other keys than k1 are colored with color 2 (the similar is the case when one key gets
color 1, only the roles of the colors are interchanged) or one of them gets color 1 and other
three get color 2 (the same is the case when two keys get color 2, only the roles of the colors are
interchanged). In first case one can not distinguish between keys k2 and k5 and also between
keys k3 and k4 (there is a rotation of the key ring where keys in both pairs interchange, while
k1 is fixed). When there is a key other than k1 with color 1 we need to consider two subcases.
If c(k2 ) = 1 (similar is the case when c(k5 ) = 1) we can not distinguish between k1 and k2 (also
between k3 and k5 ). If c(k3 ) = 1 (similar is the case when c(k4 ) = 1) we can not distinguish
between k1 and k3 (also between k4 and k5 ). Hence for n = 5 at least three colors are needed.
Consider the following coloring: c(k1 ) = 1, c(k2 ) = 2, c(k3 ) = 3 and c(k4 ) = c(k5 ) = 2 (one
possibility). Keys k2 , k4 and k5 have the same color, but one can distinguish between them
since k2 is the only one between them that has a neighbor colored with color 1 and a neighbor
colored with color 3, while only k4 has a neighbor colored with color 3 and a neighbor colored
with color 2. Hence three colors suffice for n = 5.
For n ≥ 6 consider the following coloring: c(k1 ) = 1, c(kn ) = 2, c(kn−1 ) = c(kn−2 ) = 1 and
c(ki ) = 2 for 2 ≤ i ≤ n − 3. Then k1 is the only key of color 1 with both neighbors colored with
color 2. Keys kn−1 and kn−2 both have neighbors of two different colors, but the distance (the
smallest of the two numbers: number of the keys lying between the two keys in one and other
direction) between kn−1 and k1 is one while the distance between kn−2 and k1 is two. Hence
one can distinguish between all three keys colored with color 1. Among keys colored with color
2 only kn has both neighbors colored with color 1. All other keys: ki for 2 ≤ i ≤ n − 3 have
either one or two neighbors colored with color 2. But any ki , where 2 ≤ i ≤ n − 3, has a pair of
distances: distance between ki and k1 and distance between ki and kn−2 that is different from
any other pair of distances of some key kj 6= ki for 2 ≤ j ≤ n − 3 . Hence we can distinguish
also between keys colored with color 2.

2
Problem 3. A function f : [0, ∞) → R \ {0} is called slowly changing if for any t > 1 the
limit lim ff(tx)
(x) exists and is equal to 1. Is it true that every slowly changing function has for
x→∞
sufficiently large x a constant sign (that is — it is true that for every slowly changing f there
exists N such that for every x, y > N we have f (x)f (y) > 0?)

Remark. The assumption f (x) 6= 0 is only technical, to avoid explaining what does the limit
mean in the other case, and in reality changes nothing.

Remark. The reader is encouraged to try and solve the problem himself before reading the
solution. The author’s and the proposer’s opinion is that although the solution is simple, it is
not so easy to find it (both tried, both succeeded, but both spent some time on it before getting
the correct idea).

Solution. Take t = 2. Take such a N > 0 that for x > N we have ff(2x)(x) > 0. This means
f (2x) and f (x) are of the same sign for x > N . Suppose that for any x > N we have that
f (x) and f (N ) are of a different sign. Let t = Nx . Then ff(tN )
(N ) < 0, and by easy induction
f (t2k N ) f (tx)
< 0 for any k ∈ N, which contradicts the assumption
f (2k N ) f (x) → 1 when x tends to ∞. The
contradiction proves the thesis.

3
Problem 4. Let f : [0, 1] → [0, ∞) be an arbitrary function satisfying
f (x) + f (y) x + y 
≤f +1 (1)
2 2
for all pairs x, y ∈ [0, 1]. Prove that for all 1 ≤ u < v < w ≤ 1,
w−v v−u
f (u) + f (w) ≤ f (v) + 2 .
w−u w−u

Solution. Let
 
w−v v−u
M (u, w) = sup f (u) + f (w) − f (v) ;
v∈(u,w) w−u w−u

we have to prove M (u, w) ≤ 2. Note that M (u, w) is finite, because

w−v v−u
f (u) + f (w) − f (v) ≤ 1 · f (u) + 1 · f (w) − 0 = f (u) + f (w) .
w−u w−u
Let ε > 0 be an arbitrary positive real number. Choose v such that
w−v v−u
f (u) + f (w) − f (v) > M (u, w) − ε .
w−u w−u
u+w
If v ≤ 2 , then apply (1) for x = u and y = u + 2(v − u) = 2v − u:
f (u) + f (2v − u)
≤ f (v) + 1 ;
2

w−v v−u
M (u, w) − ε < f (u) + f (w) − f (v)
w−u w−u
w−v v−u f (u) + f (2v − u)
≤ f (u) + f (w) − +1
w− u w−u 2 
1 w − (2v − u) (2v − u) − u
= f (u) + f (w) − f (2v − u) + 1
2 w−u w−u
1
≤ M (u, w) + 1 ;
2
M (u, w) ≤ 2 + 2ε .
u+w
Otherwise, if 2 < v, apply x = w − 2(w − v) = 2v − w and y = v in (1):
f (2v − w) + f (w)
≤ f (v) + 1 ;
2

w−v v−u
M (u, w) − ε < f (u) + f (w) − f (v)
w−u w−u
w−v v−u f (2v − w) + f (w)
≤ f (u) + f (w) − +1
w− u w−u 2 
1 w − (2v − w) (2v − w) − u
= f (u) + f (w) − f (2v − w) + 1
2 w−u w−u
1
≤ M (u, w) + 1 ;
2
M (u, w) ≤ 2 + 2ε .
In both cases we obtained M (u, w) ≤ 2 + 2ε. This holds for all ε, therefore M (u, w) ≤ 2.

4
Category II
Problem 1. Construct a set A ⊂ [0, 1] × [0, 1] such that A is dense in [0, 1] × [0, 1] and every
vertical and every horizontal line intersects A in at most one point.
α
Solution. Take α, β 6∈ Q such that β 6∈ Q. Then

A := ({nα}, {nβ}) : n ∈ N ,

where {x} denotes the fractional part of x, fulfills the assumptions.

1
Problem 2. Let A be a real n × n matrix satisfying

A + At = I ,

where At denotes the transpose of A and I the n × n identity matrix. Show that det A > 0.

Solution. The assumption A + At = I is equivalent to saying A = S + 12 I where S denotes

an arbitrary real skew symmetric matrix. In particular, there exists some orthogonal matrix T
that diagonalizes S and for which D := T t ST contains the eigenvalues of S. They are either
zero or purely imaginary and pairwise conjugated, i.e. of the form

r1 i, −r1 i, . . . , rs i, −rs i, 0, . . . , 0

with rk ∈ R for all k = 1, . . . , s. The determinant of A is evaluated as follows:

 1   1 
det A = det S + I = det D + I
2 2
since det(T t T ) = 1 and with the notations from above this expression is
s 
 1 n−2s Y s 
1  1   1 n−2s Y 1 
+ rk i − rk i = + rk2 .
2 2 2 2 4
i=1 i=1

As all factors are strictly positive the result follows.

2
Problem 3. Let f : [0, 1] → R be a continuous function such that f (0) = f (1) = 0. Prove
that the set 
A := h ∈ [0, 1] : f (x + h) = f (x) for some x ∈ [0, 1]
has Lebesgue measure at least 12 .

Solution. Let us observe, that if f is continuous then A is closed, thus A is Lebesgue

measurable. Moreover the set

B := h ∈ [0, 1] : 1 − h ∈ A

has the same Lebesgue measure as the set A. We show that A ∪ B = [0, 1].
For any h ∈ [0, 1] we define a function g : [0, 1] → R by

g(x) = f (x + h) − f (x) if x + h ≤ 1

and
g(x) = f (x + h − 1) − f (x) if x + h > 1 .
From the assumption we have that g is continuous. If f has its minimum and maximum,
respectively, in x0 and x1 , then g(x0 ) ≥ 0 and g(x1 ) ≤ 0. From Darboux property we have that,
there exists x2 such that g(x2 ) = 0, therefore h ∈ A or h ∈ B. This completes the proof.

3
Problem 4. Let S be a finite set with n elements and F a family of subsets of S with the
following property:
A ∈ F, A ⊆ B ⊆ S =⇒ B ∈ F
Prove that the function f : [0, 1] → R given by
X
f (t) := t|A| (1 − t)|S\A|
A∈F

is nondecreasing (|A| denotes the number of elements of A).

Solution. Without loss of generality assume S = {1, 2, . . . , n}. For each subset A and every
n
(j)
It,A in Rn , where
Q
t ∈ [0, 1] construct a set It,A :=
j=1
(
(j) [0, t) if j ∈ A
It,A :=
[t, 1] if j ∈
/ A.

It’s clear that for any two different subsets A and B the sets It,A and It,B are disjoint.SSince the
c
volume of It,A is equal to t|A| (1 − t)|A | we have that f (t) is equal to the volume of A∈F It,A .
So the claim will be proved if we prove that
[ [
It1 ,A ⊆ It2 ,A for all 0 < t1 < t2 < 1 . (1)
A∈F A∈F

Take an arbitrary x = (x1 , x2 , . . . , xn ) ∈ It1 ,A for some A ∈ F. Construct a set B ⊆ S such that
j ∈ B if and only if xj ≤ t2 . If j ∈ / B then xj > t2 > t1 which implies j ∈ / A. So A ⊆ B and thus
B ∈ F. Moreover, from the definition of B, we have x ∈ It2 ,B . This proves (1) and the problem
is solved.

4
Problem j18-I-1. Find all complex roots (with multiplicities) of the polynomial

2008
X
1004 − |1004 − n| xn .


p(x) =
n=1

Solution. Observe, by comparison of coefficients, that

1003
X 2
p(x) = x xn .
n=0

1003
x1004 −1
xn = we conclude that p has the simple root 0 and the roots exp πin
P
Since x−1 , 502 ,
n=0
n = 1, 2, . . . , 1003, with multiplicity 2. 

2-Apr-2008 j18-I-1 1  12:56

Problem j18-I-2. Find all functions f : (0, ∞) → (0, ∞) such that

f (f (f (x))) + 4f (f (x)) + f (x) = 6x .

Solution. Let a ∈ R+ be arbitrary. Set a0 = a, an = f (an−1 ) for n > 0. Then we obtain

recurrence relation
an+3 + 4an+2 + an+1 − 6an = 0 .
Characteristic equation is
y 3 − 4y 2 + y − 6 = 0
with roots −2, −3 and 1. The general solution of recurrence relation is

an = A(−3)n + B(−2)n + C .

If A or B are not equal to 0, we have a contradiction because in range of f we could find

negative values. So the only possible solution is an = C. Because of a0 = a we have an = a
for all n ∈ N0 . Substituting n = 1 we obtain

f (a) = f (a0 ) = a1 = a ,

so for all a ∈ R+ we have f (a) = a.

The only solution of the equation is f (x) = x, what can be easily checked. 

2-Apr-2008 j18-I-2 1  12:56

Problem j18-I-3. Find all c ∈ R for which there exists an infinitely differentiable function
f : R → R such that for all n ∈ N and x ∈ R we have

f (n+1) (x) > f (n) (x) + c . (1)

Solution. For c ≤ 0 we can take f (x) = e2x . Then f (n+1) (x) = 2n+1 e2x > 2n e2x =
(n)
f (x).
For positive c no function satisfies (1). We begin with two simple lemmas.
Lemma 1. If f satisfies (1), then for any x ∈ R there exists an y ≤ x such that f (y) ≤ − 2c .
Proof. If f (t) > − 2c on (−∞, x], then f 0 (t) > 2c for any t < x, thus
Z x
c
f (y) = f (x) − f 0 (t) dt ≤ f (x) − (x − y)
y 2

for any y < x, thus for sufficiently small y we have f (y) < 0, a contradiction.
Lemma 2. If f satisfies (1), then for any x ∈ R such that f (x) < 2c we have f (y) < 2c for
any y ≤ x.
Proof. Suppose that there exists a y ≤ x such that f (y) ≥ − 2c . Let z := sup t ≤ x :


f (t) ≥ − 2c . By the continuity of f (f is differentiable, thus continuous) we have f (z) ≥ − 2c .

By the assumption upon x we have z 6= x. However by (1) we have f 0 (z) ≥ 2c , thus f 0 is
positive on [z, z + ε] for some ε > 0, f is increasing, thus f (t) ≥ f (z) ≥ − 2c for t ∈ [z, z + ε],
a contradiction with the definition of z. Thus by contradiction the thesis is proved.
Now if f satisfies (1), then obviously f 0 also satisfies (1). Thus by Lemmas 1 and 2, there
exists an x0 such that f 0 (t) < − 2c on (−∞, x0 ]. This, however, means f (t) > f (x0 )+(x0 −t) 2c
0
for t < x0 , so for sufficiently small t0 < x0 we have f (t0 ) > − 3c 2 > f (t0 ) − c, which is a
contradiction with (1). Thus no such f exists. 

2-Apr-2008 j18-I-3 1  12:56

Problem j18-I-4. The numbers of the set {1, 2, . . . , n} are colored with 6 colors. Let
S := (x, y, z) ∈ {1, 2, . . . , n}3 : x + y + z ≡ 0 (mod n)


and x, y, z have the same color
and
D := (x, y, z) ∈ {1, 2, . . . , n}3 : x + y + z ≡ 0 (mod n)


and x, y, z have three different colors .
Prove that
n2
|D| ≤ 2|S| +
.
2
(For a set A, |A| denotes the number of elements in A.)
Solution. Denote by n1 , n2 , n3 , n4 , n5 , n6 the number of occurences of the colors. Clearly
n1 + . . . + n6 = n. We prove that
6
1 X X
|S| − |D| = n2u − nu nv . (1)
2 u=1 1≤u<v≤6

For arbitrary u, v, w ∈ {1, 2, . . . , 6}, denote by Nuvw the number of triples (x, y, z),
satisfying x + y + z ≡ 0 (mod n) and having colors u, v and w, respectively. For any u, v we
P6
obviously have Nuvw = nu nv and therefore
w=1

6
1 X X X
|S| − |D| = Nuuu − Nuvw
2 u=1 1≤u<v≤6 w6=u,v
6 
X X  X
n2u −


= Nuuv − nu nv − Nuuv − Nuvv
u=1 v6=u 1≤u<v≤6
6
X X
= n2u − nu nv .
u=1 1≤u<v≤6

Now, applying the AM-QM inequality,

6 6 6
1 X X 3 X 2 1  X 2
|S| − |D| = n2u − nu nv = nu − nu
2 u=1
2 u=1 2 u=1
1≤u<v≤6
6
1 1 X 2 n2
≥ − nu =− .
4 2 u=1
4


Second solution. We present a different proof for the relation (1). We use the nota-
tion Nuvw as well.
For every u = 1, 2, . . . , 6, let CuPbe the set of those numbers from {1, 2, . . . , n} which
have the uth color and let fu (t) := tx .
x∈Cu
Let ε := e2πi/n . We will use that for every integer s,
n−1 
1 X js 1 if s ≡ 0 (mod n)
ε =
n 0 if s 6≡ 0 (mod n) .
j=0

Then, for arbitrary colors u, v, w,

X X X 1 n−1
X
Nuvw = εj(x+y+z)
n j=0
x∈Cu y∈Cv z∈Cw
n−1 n−1
1 X  X jx  X jy  X jz  1X
= ε ε ε = fu (εj )fv (εj )fw (εj )
n j=0 n j=0
x∈Cu y∈Cv z∈Cw

2-Apr-2008 j18-I-4 1 12:56

and
n−1 6
1 1 X X 3 j X 
|S| − |D| = fu (ε ) − 3 fu (εj )fv (εj )fw (εj )
2 n j=0 u=1 u<v<w
n−1 6 6
1 X X j
X X 
= fu (ε ) fu2 (εj ) − fu (εj )fv (εj )
n j=0 u=1 u=1 u<v
n−1 n 6
X 1 X jx X 2 j X 
= ε fu (ε ) − fu (εj )fv (εj ) .
j=0
n x=1 u=1 u<v

The first factor is 0 except if j = 0. Hence,

6 6
1 X X X X
|S| − |D| = fu2 (1) − fu (1)fv (1) = n2u − nu n v .
2 u=1 u<v u=1 u<v

2-Apr-2008 j18-I-4 2  12:56

Problem j18-II-1. Find all functions f : Z → Z such that

19f (x) − 17f (f (x)) = 2x (1)

for all x ∈ Z.
Solution. Suppose that there exists a function f : Z → Z satisfying the above equation.
Then define a function g: Z → Z by

g(x) = x − f (x) . (2)

Taking into account (1) and (2), we get

17g(f (x)) = 2g(x) . (3)

Let us fix y ∈ Z and let a := g(y). Define a sequence (xn )n≥0 as follows

x0 := y , x1 := f (x0 ) , ..., xn := f (xn−1 ) , ...

for any n ∈ N. Now substituting xn into (3) in turn, we get

17 17n
a = g(x0 ) = g(x1 ) = . . . = n g(xn )
2 2
for any n > 0. Consequently, we infer that

2n a = 17n g(xn )

for any n > 0. Since 2 and 17 are relatively prime, we deduce that 17n | a for any n > 0
and therefore a = 0. Moreover, since y was arbitrary, it follows that g(y) = 0 for any y ∈ Z.
Thus y − f (y) = 0 for any y ∈ Z and hence f (y) = y for any y ∈ Z. This implies that only
one function satisfies the equation (1). So, this completes the solution. 

2-Apr-2008 j18-II-1 1  13:39

Problem j18-II-2. Find all continuously differentiable functions f : [0, 1] → (0, ∞) such
that ff (1)
(0) = e and
Z 1 Z 1
dx
2
+ f 0 (x)2 dx ≤ 2 .
0 f (x) 0

Solution. First, we note that if f is such function, then

Z 1 Z 1 Z 1 0 Z 1
1 2 f (x) dx
0≤ f 0 (x) − dx = f 0 (x)2 dx − 2 dx + 2
0 f (x) 0 0 f (x) 0 f (x)
Z 1 Z 1 Z 1
dx
= f 0 (x)2 dx − 2 (ln f (x))0 dx + 2
0 0 0 f (x)
Z 1 Z 1
f (1) dx
= f 0 (x)2 dx − 2 ln + 2
dx ≤ 0 ,
0 f (0) 0 f (x)

f (1) R1 R1
since f (0) = e and dx
0 f (x)2
+ 0
f 0 (x)2 dx ≤ 2. Therefore

Z 1
1 2
f 0 (x) − dx = 0 . (1)
0 f (x)

Since f is continuously differentiable function on [0, 1], the equality (1) is equivalent to

f 0 (x)f (x) = 1 ∀x ∈ [0, 1] . (2)

√
All positive solutions of the differential equation (2) are in the form f (x) = 2x + C for
some C > 0. Since ff (0)
(1)
= e, we have C = e22−1 , and thus
r
2
f (x) = 2x +
e2 − 1

is the unique function satisfying the conditions from the statement. 

2-Apr-2008 j18-II-2 1  13:39

Problem j18 - -
√ II 3. Find
√ all pairs of natural numbers (n, m) with 1 < n < m such that the
numbers 1, n n and m m are linearly dependent over the field of rational numbers Q.
Solution. The answer is n = 2, m = 4.
We begin with the following
√ √
Lemma. The minimal √ (over Q) polynomial f (X) for n n equals X k − ( n n)k , where k is the
minimal satisfying ( n n)k ∈ N.
√
Proof. n n is a root of X n −n = 0. So there is some nonempty subset A of {0, 1, ..., n−1}
such that Y
f (X) = (X − ζ l ) ,
l∈A

2π 2π



where ζ = cos n + i sin
n .
√ √
The free term of f (X) has an absolute value equal to ( n n)|A| . Hence√( n n)deg f (X) is
integer,√and deg f (X) ≥ k follows (k is as in the lemma). But, clearly n n is a root of
X k − ( n n)k , which has integer coefficients.
√ √
Let us assume that 1, n n, m m√are linearly√ dependent over Q, i.e. there are rational
a, b, c not all equal 0 such that
√ a + b n
n + c m
m = 0. √ √
Case a 6= 0. Then, as n n is irrational, we √ have b, c 6=√0. But a + b n n = −c m m has
the same degree of a minimal √ polynomial√as n n, and as m m. Let k be the degree of the
minimal polynomial for m
m. Then y = n n satisfies
√
(a + by)k = ( m m)k ,
√
but y k and ( m m)k are rational, and√
as a, b 6= 0 we obtain that there is a nonzero polynomial
with rational coefficients vanishing
√
n
n of degree smaller than k, a contradiction.
n
n nm
Case a = 0. Hence m m is rational, and this is equivalent to m
√ n is a mn-th power of

a rational. Let p be any prime, and pa k n, pb k m. So we must have mn | am − bn. But

am−bn ≤ am < mn, in view of a ≤ log2 n < n. In a similar way one obtains am−bn > −mn.
So we must have am = bn, the relation independent of the choice of prime p. Thus

n = mm/n ,
√ √ √ √ √
and n n = m m follows. As the function x x has maximum at x = e, we see that n n = m m
holds only for n = 2, m = 4. 

2-Apr-2008 j18-II-3 1  13:39

Problem j18-II-4. We consider the following game for one person. The aim of the player
is to reach a fixed capital C > 2. The player begins with capital 0 < x0 < C. In each turn
let x be the player’s current capital. Define s(x) as follows:

x if x < 1
(
s(x) := C − x if C − x < 1
1 otherwise.

Then a fair coin is tossed and the player’s capital either increases or decreases by s(x), each
with probability 21 . Find the probability that in a finite number of turns the player wins by
reaching the capital C.
Solution. Let us denote by f (x) the probability that player wins with starting capital x.
If x ≤ 1, then he loses if loses the first turn, and if he wins the first turn, he has
capital 2x. Thus f (x) = 12 f (2x).
If x ≥ C − 1 the player wins if he wins the first turn, and has 2x − C in other case, thus
f (x) = 21 + 12 f (2x − C).
In all other cases there is f (x) = 12 f (x − 1) + f (x + 1) .

We will prove that this implies f (x) = Cx .

Let us define g(x) = f (x) − Cx . It is bounded on [0, C] (as f (x) ∈ [0, 1]), and we have
1 x 1 2x

1

 2 f (2x) − C = 2 f (2x) − C = 2 g(2x) for x ≤ 1,

 1 x


 2 f (x − 1) + f (x + 1) − C




 = 1 f (x − 1) − x−1 + f (x + 1) − x+1


2 C C
g(x) = 1




 = 2 g(x − 1) + g(x + 1) for x ∈ (1, C − 1),

1 1 x

2 + 2 f (2x − C) − C





= 12 f (2x − C) − 2x−C

1
= 2 g(2x − C) for x ≥ C − 1.

C

Obviously g(0) = g(C) = 0. Let K = sup f (t) ∈ [0, ∞). Denote n0 = [C] − 1 ≥ 1.
t∈[0,C]
n
We will prove for any natural 0 < n ≤ n0 and x ∈ (n − 1, n] there is g(x) ≤ 2 2−1
n K.

If x ∈ (0, 1] there is g(x) = 21 g(2x) ≤ K

2 .
n−1
Assume, that for x ≤ n − 1 and take x̄ ∈ (n − 1, n]. There is g(x̄ − 1) ≤ 2 2n−1−1 K as
x̄ − 1 ∈ (n − 2, n − 1], and g(x̄ + 1) ≤ K. Thus

1
1  2n−1 − 1  2n − 1
g(x̄) = g(x̄ − 1) + g(x̄ + 1) ≤ K + K = K
2 2 2n−1 2n
as required.
g(x) ≤= 12 g(2x − C) ≤ K 2 for x ≥ C − 1.
Now take x ∈ (n0 , C − 1) (it is empty set for integer C). We have proved that g(x − 1) ≤
2n0 −1 2n0 −1
2n0 K (as x − 1 ∈ (n0 − 1, n0 )) and g(x + 1) ≤ K 2 (x + 1 > C − 1). Thus g(x) ≤ 2n0 K.
n0
Thus we have proved, that g(x) ≤ 2 2n−1 0
K for every x ∈ [0, C], which means that K = 0.
Similarly one can prove, that inf f (t) = 0. Thus g(x) ≡ 0, so f (x) = Cx . 
t∈[0,C]

2-Apr-2008 j18-II-4 1  13:39

 The 19th Annual Vojtěch Jarnı́k
International Mathematical Competition
Ostrava, 1st April 2009
Category I

Problem 1 Let ABC be a non-degenerate triangle in the euclidean plane. Define a sequence (Cn )∞ n=0 of points
as follows: C0 := C, and Cn+1 is the center of the incircle of the triangle ABCn . Find lim Cn .
n→∞
[10 points]
Solution If α is the angle at A, β the angle at B, then the limit is the point on the side AB dividing it in the
ratio α : β. Let αi and βi be the angles at A and B in ABCi , respectively. Since the center of the incircle is
the intersection of the angle bisectors, we have αi+1 = α2i and βi+1 = β2i ; so the limit point will obviously lie
on AB; furthermore, αβii = αβ =: q for all i. Thus, if Ki is the circumcircle of ABCi , S1,i and S2,i the arcs over
|S |
ACi and BCi , respectively, then |S1,i
2,i |
= q for all i. Now, as the Ci approache AB, the arcs converge to the
corresponding sides of the triangle. Hence, the result follows. 
 The 19th Annual Vojtěch Jarnı́k
International Mathematical Competition
Ostrava, 1st April 2009
Category I

Problem 2 Prove that the number

k
22 − 1 − 2k − 1
is composite (not prime) for all positive integers k > 2. [10 points]
Solution Denote k
−1
M = 22 − 2k − 1 .
If k is even then 3 | M , and M is composite, since M > 3 for k > 2.
Suppose k is odd. Then
k
 k−1   k−2   1 
2M = 22 − 1 − 2k+1 + 1 = 22 + 1 22 + 1 . . . 22 + 1 − 2k+1 + 1 .

a
 a


Let k + 1 = 2a q with positive odd integer q and a ≥ 1. Then 22 + 1  2M . Indeed, 22 + 1  2k+1 + 1
and  a    k−1   k−2   1 
22 + 1  22 + 1 22 + 1 . . . 22 + 1 ,


since a ≤ k − 1 for k > 2. 

 The 19th Annual Vojtěch Jarnı́k
International Mathematical Competition
Ostrava, 1st April 2009
Category I

Problem 3 Let k and n be positive integers such that k ≤ n − 1. Let S := {1, 2, . . . , n} and let A1 , A2 , . . . , Ak
be nonempty subsets of S. Prove that it is possible to color some elements of S using two colors, red and blue,
such that the following conditions are satisfied:

(i) Each element of S is either left uncolored or is colored red or blue.

(ii) At least one element of S is colored.
(iii) Each set Ai (i = 1, 2, . . . , k) is either completely uncolored or it contains at least one red and at least one
blue element.
[10 points]
Solution Consider the following system of k linear equations in n real variables x1 , x2 , . . . , xn :
X
xj = 0 , i = 1, 2, . . . , k .
j∈Ai

Since k < n, this system has a nontrivial solution (x1 , x2 , . . . , xn ), i.e. a solution with at least one nonzero xj .
Now color red all elements of the set {j ∈ S : xj > 0}, color blue all elements of the set {j ∈ S : xj < 0}, and
leave uncolored all elements of {j ∈ S : xj = 0}.
P is nontrivial, at least one element is colored. If Ai contains some red element j ∈ S then
Since the solution
xj > 0, and from j∈Ai xj = 0 we see that there exists some j 0 ∈ Ai such that xj 0 < 0, i.e. j 0 is colored blue.
Thus Ai must have elements of both colors. Analogously we argue when Ai contains a blue element. Therefore
we see that the above coloring satisfies all requirements. 
 The 19th Annual Vojtěch Jarnı́k
International Mathematical Competition
Ostrava, 1st April 2009
Category I

Problem 4 Let (an )∞ ∞

n=1 be a sequence of real numbers. We say that the sequence (an )n=1 covers the set of
∞
P k
positive integers if for any positive integer m there exists a positive integer k such that an = m.
n=1

a) Does there exist a sequence of real positive numbers which covers the set of positive integers?
b) Does there exist a sequence of real numbers which covers the set of positive integers?
[10 points]
Solution The answer to the second question is positive.
First we shall prove that for any n there exists a finite sequence (xi )ki=1
n
of real numbers such that
kn
X
x2m+1
i =0 for 0 ≤ m < n
i=1

and
kn
X
x2m+1
i 6= 0 for m ≥ n .
i=1
Pkn 2m+1
For the simplicity of notation we shall write Sm (xi ) for i=1 xi . We shall prove the thesis by induction
upon n. For n = 0 the appropriate sequence is x1 = 1.
Assume the thesis for n. For n + 1 consider the sequence

(yi )3k
i=1 = (−x1 , −x2 , . . . , −xkn , αx1 , αx2 , . . . , αxkn , αx1 , αx2 , . . . , αxkn ) ,
n

where α = 2−1/(2n+1) . As Sm (xi ) = 0 for m < n, we also have Sm (yi ) = 0. We also have

Sn (yi ) = −Sn (xi ) + 2−1 Sn (xi ) + 2−1 Sn (xi ) = 0 .

For m > n we have

Sm (yi ) = (1 − 2 · 2−(2m+1)/(2n+1) )Sm (xi ) 6= 0 .
Thus the induction step is finished, and the thesis is proved. Moreover it is easy to notice that |xi | ≤ 1 and the
length of the sequence is 3n . Denote by x(n) the sequence of length 3n with Sm (x(n)) = 0 for m < n.
Now to give the required sequence (ai ). Our sequence will be a concatenation of multiples of the finite
sequences x(n) given above. We begin with a1 = 1 (that is we begin by taking x(0)). In the n-th step we
assume that we have some finite sequence ai , with Sm (ai ) = m + 1 for m ≤ n. We also assume that the
elements added in the n-th step will be no larger than n1 .
To pass to the (n + 1)-st step let c = n + 2 − Sn+1 (ai ), and let d = Sn+1 (x(n + 1)). Take an integer
N >  (n+1)c
 
, let α = c , |α| < 1 . We add N copies of the sequence αx(n + 1) to the end of ai . This does
d dN n+1
not change Sm (ai ) for m < n + 1 (as Sm (x(n + 1)) = 0, and after the addition we have Sn+1 (ai ) = n + 2. Also
1
all the added elements are of absolute value no larger than n+1 .
Now to prove that for this series we have G2k+1 = k + 1. As Sm (ai ) = m + 1 after every step, no other limit
is possible, we only have to check convergence. Note, however, that after the n–th step we only add sequences
x(m) for m > n, which in Pturn are concatenations of sequences x(n), with some coefficients. Thus every 3n –th
2n+1
partial sum in the series ai is going to be exactly equal to n + 1. The partial sums “in the middle” cannot
n
differ from this value by more than 32 times the value of the maximal element |ai | in the appropriate interval,
and this converges to zero. Thus for any n we do, in fact, have convergence.
For the first question, obviously the same series suffices.
For the last question, the answer is negative. As ai are positive, we may rearrange them in decreasing order.
Take k0 to be the first k for which Gk is finite. For Gk0 to be finite, we have to have ai convergent to zero, thus
only a finite number of terms is larger than 1, assume these P are the first n terms. Note that as for i > n we
∞
have ai ≤ 1, we also have that aki decreases with k, and thus i=n+1 aki decreases with k, and thus is bounded
P∞
by C := i=n+1 aki 0 . As Gk are assumed to attain unbounded values, we have to have terms larger than 1,
thus n > 0.
Assume the first m terms of ai are equal, 1 ≤ m ≤ n. Then for k ≥ k0 and l ≤ k we have

Gl ≤ mak1 + nakm+1 + C .
On the other hand Gl ≥ mak+1
1 for l ≥ k. For sufficiently large k, however, we have

mak+1
1 > mak1 + nakm+1 + C + 2 ,

which means that there is an integer number between mak+1

1 and mak1 + nakm+1 + C which is not the value of
Gk for any k. 
 The 19th Annual Vojtěch Jarnı́k
International Mathematical Competition
Ostrava, 1st April 2009
Category II

Problem 1 A positive integer m is called self-descriptive in base b, where b ≥ 2 is an integer, if:

i) The representation of m in base b is of the form (a0 a1 . . . ab−1 )b
(that is m = a0 bb−1 + a1 bb−2 + · · · + ab−2 b + ab−1 , where 0 ≤ ai ≤ b − 1 are integers).

ii) ai is equal to the number of occurences of the number i in the sequence (a0 a1 . . . ab−1 ).
For example, (1210)4 is self-descriptive in base 4, because it has four digits and contains one 0, two 1s, one 2
and no 3s.
a) Find all bases b ≥ 2 such that no number is self-descriptive in base b.

b) Prove that if x is a self-descriptive number in base b then the last (least significant) digit of x is 0.
[10 points]
Solution
1. For b ≥ 7 it is easy to verify that the number of the form (b − 4)bb−1 + 2bb−2 + bb−3 + b4 is a self descriptive
number (it contains b − 4 instances of digit 0, two instances of digit 1, one instance of digit 2 and one
instance of digit b − 4), and numbers 21200(5) and 2020(4) are self-descriptive numbers in bases 5 and 4,
respectively.
It remains to show that for bases 2, 3 and 6 no self descriptive numbers exist. First note, that a self-
descriptive number (in any admissible base) contains at least one instance of the digit 0. If it does not,
then the first digit is 0, which is a contradiction.
It is easy to prove the claim for b = 2, 3.
Let us prove it for b = 6. Assume there exists x = (b0 b1 b2 b3 b4 b5 )(6) , where x is a self-descriptive number.
We observe the following about x:
P5
(a) i=0 bi = 6
(b) b0 6= 0
P5
(c) i=1 bi = |{bi , bi 6= 0, i ≥ 1}| + 1
(d) Other than the first digit, the set of all other non-zero digits consists of several 1’s and one 2.
Observation 1d implies that all but one of the digits b3 , b4 and b5 are 0, now it is easy to check, that no
such number is self-descriptive, which is a contradiction. Therefore base b = 6 contains no self-descriptive
numbers.

2. Assume that there is in fact a self-descriptive number x in base b that it is b-digits long but not a multiple
of b. The digit at position b − 1 must be at least 1, meaning that there is at least one instance of the digit
b − 1 in x. At whatever position a that digit b − 1 falls, there must be at least b − 1 instances of digit a in
x. Therefore, we have at least one instance of the digit 1, and b − 1 instances of a. If a > 1, then x has
more than b digits, leading to a contradiction of our initial statement. And if a = 0 or a = 1, that also
leads to a contradiction.
3. These numbers are: 1210, 2020, 21200, 3211000, 42101000, 521001000, 6210001000. That these are the
only such numbers, follows from previous observations.

 The 19th Annual Vojtěch Jarnı́k
International Mathematical Competition
Ostrava, 1st April 2009
Category II

Problem 2 Let E be the set of all continuously differentiable real valued functions f on [0, 1] such that f (0) = 0
and f (1) = 1. Define
Z 1
J(f ) = (1 + x2 )(f 0 (x))2 dx .
0

a) Show that J achieves its minimum value at some element of E.

b) Calculate min J(f ).
f ∈E
[10 points]
Solution By the fundamental theorem of Calculus, we have
Z 1 
0 0
1 = |f (1) − f (0)| =  f 00 (x) dx .
 
0

Next, by using the Cauchy-Schwartz inequality, we obtain

Z 1   Z 1 √1 + x 2 
00
f (x) dx = √ f 00 (x) dx
   
 2
0 0 1 + x
Z 1 1/2 Z 1 1
1/2
2 00 2
≤ (1 + x )(f (x)) dx 2
dx
0 0 1+x
Z 1 1/2  1 1/2
= (1 + x2 )(f 00 (x))2 dx arctan x

0 0
Z 1 1/2 √π
= (1 + x2 )(f 00 (x))2 dx .
0 2

Hence Z 1
4
inf (1 + x2 )(f 00 (x))2 dx ≥ .
f ∈E 0 π
Finally, let Z x
4
f (x) := arctan t dt
π 0

for x ∈ [0, 1]. Then f 0 (x) = π4 arctan x (by the fundamental theorem of Calculus) and f 00 (x) = 4 1
π 1+x2 , for
x ∈ [0, 1]. Consequently, we deduce that f ∈ E and
1
16 1 1
Z  4 1 2 Z
16 π 4
J(f ) = (1 + x2 ) dx = dx = 2 · = ,
0 π 1 + x2 π 2 0 1 + x2 π 4 π

which proves that J attains its minimum on E. This completes the solution. 
 The 19th Annual Vojtěch Jarnı́k
International Mathematical Competition
Ostrava, 1st April 2009
Category II

2 2
Problem 3 Let A be an n × n square matrix with integer entries. Suppose that p2 Ap = q 2 Aq + r2 In for
some positive integers p, q, r where r is odd and p2 = q 2 + r2 . Prove that |det A| = 1.
(Here In means the n × n identity matrix.) [10 points]
Solution Consider the function f : R → R.
2 2
f (x) = p2 xp − q 2 xq − r2 . (1)

Observe that   q 4 
2 2
f 0 (x) = p4 xq −1
xr − .
p

4
The roots of equation f 0 (x) = 0 are x1 = 0 and x2 = pq r2 (r 6= 0 and q 6= 1). From f (0) = −r2 < 0 and

4

f pq r2 < 0 we obtain

−1 if x < 1 ,

sgn f (x) = 0 if x = 1 , (2)

1 if x > 1 .


So x = 1 is the only real root of equation f (x) = 0.

Since the matrix A verifies f (A) = On , some eigenvalue λ ∈ σP (A) satisfies the equation f (λ) = 0. Let
λ1 , λ2 , . . . , λn be eigenvalues of the matrix A. We show that |λk | ≤ 1 for all k. The fact f (λ) = 0 can be written
as 2 2
p2 λp = q 2 λq + r2 . (3)
2 2
Passing the relation (3) at modulus we obtain p2 |λ|p ≤ q 2 |λ|q + r2 or

f (|λ|) ≤ 0 . (4)

From (2) and (4) we obtain 0 ≤ |λ| ≤ 1 or 0 ≤ |λk | ≤ 1 for all k = 1, . . . , n. Because f (0) = −r2 6= 0, it results
that λk 6= 0 for all k.
Hence
0 < |λk | ≤ 1 for all k = 1, . . . , n . (5)
From det A = λ1 λ2 · · · λn we obtain

|det A| = |λ1 λ2 · · · λn | = |λ1 ||λ2 | · · · |λn | ≤ 1 . (6)

From (5) and (6) we obtain

0 < |det A| ≤ 1 . (7)
Since A ∈ Mn (Z), it follows that |det A| ∈ N. From (7) we obtain the conclusion that |det A| = 1. 
 The 19th Annual Vojtěch Jarnı́k
International Mathematical Competition
Ostrava, 1st April 2009
Category II

Problem 4 Let k, m, n be positive integers such that 1 ≤ m ≤ n and denote S = {1, 2, . . . , n}. Suppose that
A1 , A2 , . . . , Ak are m-element subsets of S with the following property: for every i = 1, 2, . . . , k there exists a
partition S = S1,i ∪ S2,i ∪ · · · ∪ Sm,i (into pairwise disjoint subsets) such that
(i) Ai has precisely one element in common with each member of the above partition.

(ii) Every Aj , j 6= i is disjoint from at least one member of the above partition.
n−1


Show that k ≤ m−1 . [10 points]
(i)
Solution Without loss of generality assume that 1 ∈ S1 for all i = 1, 2, . . . , k, because otherwise we simply
rename members of each partition.
For every i = 1, 2, . . . , k define the polynomial
m  X
Y 
Pi (x2 , x3 , . . . , xn ) = xs
l=2 s∈Sl
(i)

and regard it as a polynomial over R in variables x2 , x3 , . . . , xn .

Observe that Pi is a homogenous polynomial of degree m − 1 in n − 1 variables. Also observe that all
monomials in Pi are products of different x’s, i.e. there are no monomials with squares or higher powers. The
(i) (i)
last statement follows simply from the
fact that S2 , . . . , Sm are mutually disjoint. Such polynomials form a
n−1
linear space over R of dimension m−1 and polynomials Pi belong to that space. If we prove that polynomials
n−1


Pi , i = 1, 2, . . . , k are linearly independent, the inequality k ≤ m−1 will follow from the dimension argument.
For any i = 1, 2, . . . , k let χi be the characteristic vector of A ∩ {2, 3, . . . , n}. In other words, χi ∈ {0, 1}n−1
where the j-th coordinate of χi equals 1 if j + 1 ∈ A, and 0 otherwise.
(i)
For every i we know that each Ai ∩ Sl has exactly one element and therefore
m m
(i)
Y Y
Pi (χi ) = |Ai ∩ Sl | = 1 = 1.
l=2 l=2

(i) (i)
On the other hand, if j 6= i then either some Aj ∩ Sl , l ≥ 2 is empty, or all Aj ∩ Sl , l ≥ 2 are nonempty but
(i) (i)
Aj ∩ S1 = ∅. In the latter case we must have |Aj ∩ Sl | = 2 for some l ≥ 2. In any case we have at least one
even factor in the following product, and so
m
(i)
Y
Pi (χj ) = |Aj ∩ Sl | ≡ 0 (mod 2) .
l=2

Therefore all diagonal entries in the matrix [Pi (χj )]i,j=1,2,...,k are odd, while all non-diagonal entries are
even. Consequently, its determinant is an odd integer, in particular it is not 0, and thus the matrix is regular.
If polynomials Pi were linearly dependent, we would conclude that rows of [Pi (χj )]i,j=1,2,...,k are also linearly
dependent, but this is not the case. Therefore Pi , i = 1, 2, . . . , k must be linearly independent and this completes
the proof. 
 The 20th Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 25th March 2010
Category I

Problem 1

a) Is it true that for every bijection f : N → N the series

∞
X 1
n=1
nf (n)

is convergent?

b) Prove that there exists a bijection f : N → N such that the series

∞
X 1
n=1
n + f (n)

is convergent.

(N is the set of all positive integers.) [10 points]

Solution a) Yes. Applying the inequality, if 0 ≤ a1 ≤ · · · ≤ an and 0 ≤ b1 ≤ · · · ≤ bn and σ : {1, . . . , n} →
{1, . . . , n} is a permutation, then
Xn Xn
aj bσ(j) ≤ aj bj ,
j=1 j=1

for every n we get

n n ∞
X 1 X 1 X 1
≤ ≤ .
j=1
jf (j) j=1 j 2 j=1
j2
Pn 1


Since the sequence is increasing and bounded, it converges.
j=1 jf (j)
b) No. We will construct a permutation f : N → N such that the series
∞
X 1
n=1
n + f (n)

is convergent. Let f : N → N be given in the following way: f (1) = 4 and for [(n!)2 + 1, ((n + 1)!)2 ] ∩ N we put
n−1
X
f ((n!)2 + k) = [(n + 2)!]2 − (k − 1) if 1 ≤ k < [(n + 1)!]2 − 1 − (−1)j [(n − j)!]2 .
j=0

and
n−1
X
f ([(n + 1)!]2 − k) = [(n − 1)!]2 + k + 1 if 0 ≤ k ≤ 1 + (−1)j [(n − j)!]2 .
j=0

Then
[(n+1)!]2
X 1 ((n + 1)!)2 − (n!)2 (n!)2 − [(n − 1)!]2
≤ +
n + f (n) (n!)2 + [(n + 2)!]2 + 1 [(n + 1)!]2 + [(n − 1)!]2 + 1
j=(n!)2 +1
1 1
<+ .
(n + 2)2 (n + 1)2
Pn 1


Thus we show that the sequence j=1 j+f (j) is bounded. Since it is increasing, it converges. 
 The 20th Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 25th March 2010
Category I

Problem 2 Let A and B be two complex 2 × 2 matrices such that AB − BA = B 2 . Prove that AB = BA.
[10 points]
Solution We may concludethat  AB = BA  only if 2 6= 0 in F (that is, char F 6= 2).
 if and
1 1 0 0
If char F = 2, take B = ,A= .
0 1 1 0
   2 
a b 2 a + bc b(a + d)
Assume that char F 6= 2. Let B = , then B = . We have a2 + d2 + 2bc =
c d c(a + d) d2 + bc
trace B 2 = trace AB − trace BA = 0. If B is invertible, then A = B(A + B)B −1 , hence

trace A = trace(B(A + B)B −1 ) = trace(A + B) = trace A + trace B,

so trace B = 0, d = −a, trace B 2 = 2(a2 + bc) = 0. Since char F 6= 2, it implies a2 + bc = 0, hence B 2 = 0 and
AB = BA. If B is not invertible, then det B = ad − bc = 0, so (a + d)2 = a2 + d2 + 2bc = 0, a + d = 0, a = −d,
a2 + bc = −ad + bc = 0, so B 2 = 0. 
 The 20th Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 25th March 2010
Category I

Problem 3 Prove that there exist positive constants c1 and c2 with the following properties:
a) For all real k > 1,
Z 1 p  c1
1 − x2 cos(kx) dx < 3/2 .
 

0 k
b) For all real k > 1,
Z 1 p  c
2
1 − x2 sin(kx) dx > .
 
k

0
[10 points]
√
Solution Put f (x) = 1 − x2 .
1. Integrating by parts, we have
Z 1 i1 Z 1
h 1 1
f (x) · cos kx dx = f (x) · sin kx − f 0 (x) · sin kx dx .
0 k 0 0 k
The first term is 0 − 0 = 0. The second term is (−1/k) times
Z √1−1/k Z 1
0
f (x) · sin kx dx + √ f 0 (x) · sin kx dx . (1)
0 1−1/k

Here the first term equals

√
h 1 i√1−1/k Z 1−1/k
1
0
−f (x) · cos kx + f 00 (x) · cos kx dx ,
k 0 0 k
whose absolute value is p
2 p
2 1 − 1/k 2
≤ − f0 1 − 1/k = p <√ .
k k 1/k k
The absolute value of the second term in (1) is
Z 1
1
≤ √ |f 0 (x)| dx = −[f (x)]1√ =√ .
1−1/k k
1−1/k

Thus, we may choose c1 = 2 + 1 = 3.

2. Integrating by parts, we have
Z 1 i1 Z 1
h 1 1
f (x) · sin kx dx = − f (x) · cos kx + f 0 (x) · cos kx dx .
0 k 0 0 k
The first term is 1/k. The second term is (1/k) times
Z √ 1−1/k Z 1
f 0 (x) · cos kx dx + √ f 0 (x) · cos kx dx . (2)
0 1−1/k

Here the first term equals

√
h 1 i√1−1/k Z 1−1/k
1
0
f (x) · sin kx − f 00 (x) · sin kx dx ,
k 0 0 k
whose absolute value is p
2 p
2 1 − 1/k 2
≤ − f0 1 − 1/k = p <√ .
k k 1/k k
The absolute value of the second term in (2) is
Z 1
1
≤ √ |f 0 (x)| dx = −[f (x)]1√ =√ .
1−1/k k
1−1/k
Thus, Z 1  
1 3
f (x) · sin kx dx > 1− √ .
0 k k
This proves the desired claim for k ≥ 3π.
The integral has a positive lower bound for k < 3π as well, since
Z 1 Z 1

1 − cos kx
f (x) · sin kx dx = −f 0 (x) · dx > 0 .
0 0 k


 The 20th Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 25th March 2010
Category I

Problem 4 For every positive integer n let σ(n) denote the sum of all its positive divisors. A number n is
called weird if σ(n) ≥ 2n and there exists no representation

n = d1 + d2 + · · · + dr ,

where r > 1 and d1 , . . . , dr are pairwise distinct positive divisors of n.

Prove that there are infinitely many weird numbers. [10 points]
Solution The idea is to show that given a weird number, one can construct a sequence of weird numbers
tending to infinity.
We claim that for weird n and p a prime greater than σ(n) and coprime to n, the number pn is also weird.
In fact, if 1 = d1 , d2 , . . . , dk = n are the positive divisors of n, the ones of pn are d1 , d2 , . . . , dk , pd1 , . . . , pdk and
they are pairwise distinct as (p, n) = 1. Suppose now that we have

pn = di1 + · · · + dir + p(dj1 + · · · + djs )

with ik , jl ∈ {1, . . . , k}. Then we have

di1 + · · · + dir = p(n − dj1 − · · · − djs ) .

Note that n ∈/ {dj1 , . . . , djs } as the representation must have more than only one summand and the assumption
that n is weird implies n − dj1 − . . . − djs 6= 0. Hence as the right hand expression is divisible by p and non
zero, so must be di1 + · · · + dir which is impossible as p > σ(n).
It remains to find a weird number. A possible reasoning could be: look for a number n with σ(n) = 2n + 4
that is not divisible by 3 and 4. Then the smallest possible divisors are 1, 2, 5 so that it will be impossible to
represent 4, and hence n, as a sum of pairwise distinct divisors of n. Checking for numbers with three distinct
prime factors 2, p, q yields
σ(2pq) = 3(p + 1)(q + 1) = 3pq + 3p + 3q + 3
and hence we need
3pq + 3p + 3q + 3 = 4pq + 4 ⇐⇒ (p − 3)(q − 3) = 8 .
This equality is solved by p = 5 and q = 7 which yields the weird number n = 70. 
 The 20th Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 25th March 2010
Category II

Problem 1 Let a and b be given positive coprime integers. Then for every integer n there exist integers x, y
such that
n = ax + by .
Prove that n = ab is the greatest integer for which xy ≤ 0 in all such representations of n. [10 points]
Solution The greatest such integer is a · b.
If ab = ax + by, then a | y and b | x. Thus if x > 0, then x ≥ b and by = ab − ax ≤ ab − ab = 0, so y ≤ 0.
Now let n > ab. Let n = ax + by be the representation such that x is positive and as small as possible. Then
since n = a(x − b) + b(y + a) is another representation of n, x − b must not be positive and therefore x ≤ b.
Hence by = n − ax ≥ n − ab > 0, so y > 0. 
 The 20th Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 25th March 2010
Category II

Problem 2 Prove or disprove that if a real sequence (an ) satisfies an+1 − an → 0 and a2n − 2an → 0 as n → ∞,
then an → 0. [10 points]
Solution The proposition is true.
From the condition an+1 −an → 0 we conclude by Cesaro’s lemma that ann → 0. Since the sequence a2n −2an
must be bounded, we know that
C := sup{|a2n − 2an | : n ∈ N} < ∞ .
Considering the identity
m
an a m+1 X an·2k an·2k+1 
− n·2m+1 = −
n n·2 n · 2k n · 2k+1
k=0

we conclude by letting m → ∞ and n fixed that

∞
an X an·2k a k+1 
= k
− n·2k+1 .
n n·2 n·2
k=0

Now from
∞ 
a  X ∞
 n  an·2k an·2k+1  X C C
 ≤ − ≤ =
n n · 2k n · 2k+1 n · 2k+1 n
 
k=0 k=0

we infer that |an | ≤ C, i. e. the sequence (an ) must be bounded.

Now suppose that (an ) does not converge to 0. Then, by Bolzano’s theorem, there must exist a subsequence
(ank ) converging to some number a 6= 0. From the hypothesis we conclude in turn that

a2nk → 2a ,
a4nk → 4a ,
..
.

which would result in an unbounded set of accumulation points a, 2a, 4a, . . . of (an ) in contradiction to (an )
being bounded. 
 The 20th Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 25th March 2010
Category II

Problem 3 Let A and B be two n × n matrices with integer entries such that all of the matrices

A, A+B, A + 2B , A + 3B , ... , A + (2n)B

are invertible and their inverses have integer entries, too. Show that A + (2n + 1)B is also invertible and that
its inverse has integer entries. [10 points]

Solution Suppose that the n × n matrix M has integer entries and M has inverse matrix M −1 with integer
entries. Then M · M −1 = I implies det M · det M −1 = 1. Thus det M = 1 or det M = −1. Set M (t) = A + tB.
The determinant of the matrix M (t)

det M (t) = det (A + tB) = det A + · · · + tn det B

is the polynomial of degree n in t. The polynomial det M (t) takes values 1 or −1 at points t = 0, 1, 2, . . . , 2n.
Hence det M (t) takes the value 1 or the value −1 at least n + 1 times. This implies that det M (t) is a
constant polynomial: M (t) = 1 or M (t) = −1 for all t. Consequently, det M (2n + 1) = ±1. Hence the matrix
A + (2n + 1)B is invertible. By Cramer’s formula, the inverse matrix has integer entries, since the determinant
is equal to 1 or −1. 
 The 20th Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 25th March 2010
Category II

Problem 4 Let f : [0, 1] → R be a function satisfying

|f (x) − f (y)| ≤ |x − y|
for every x, y ∈ [0, 1]. Show that for every ε > 0 there exists a countable family of rectangles (Ri ) of dimensions
ai × bi , ai ≤ bi , in the plane such that
 [ X
(x, f (x)) : x ∈ [0, 1] ⊂ Ri and ai < ε .
i i

(The edges of the rectangles are not necessarily parallel to the coordinate axes.) [10 points]
Solution Assume without loss of generality that f (0) = 0, thus |f (x)| ≤ 1 for x ∈ [0, 1].
First notice that if C ⊂ [0, 1] is a set of Lebesgue measure no larger than ε/3, then it can be covered by a
countable family of intervals Ii of total measure at most ε/2, and thus {(x, f (x) : x ∈ C} is covered by rectangles
Ii × [−1, 1], and their total width is at most ε/2.
Notice that as we are interested in only one dimension of the rectangle, and the graph we are to covered is
bounded, we may as well think in terms of covering with strips instead of rectangles.
For now on fix ε > 0. We shall introduce a few definitions. Let x, y ∈ [0, 1]. We say that the interval [x, y]
is covered, if |f (z) − α(z)| < ε|x − y| for all z ∈ [x, y], where α is the linear function meeting f at x and y. The
inclination of an interval [x, y], denoted i(x, y), is the number |f (x) − f (y)|/|x − y|. Notice the inclination of
any interval cannot be larger than 1 as f is 1-Lipschitz.
Now we prove the following lemma.
Lemma There exists a constant δ > 0 such that the following holds. Consider any interval [x, y] ⊂ [0, 1]. Then
either [x, y] is covered, or there exists a subinterval [x0 , y 0 ] ⊂ [x, y] of length |y 0 − x0 | > δ|x − y| and inclination
at least i(x, y) + ε.
Proof The proof is pretty simple. If [x, y] is not covered, then there exists a point z ∈ [x, y] with |f (z)−α(z)| >
ε|x − y|. Without loss of generality assume f (x) < f (y) and f (z) − α(z) > ε|x − y|. The interval [x, z] in this
case has inclination
f (y)−f (x)
f (z) − f (x) α(z) + ε(y − x) − f (x) y−x (z − x) + ε(y − x)
i(x, z) = |f (x) − f (z)|/|x − z| = ≥ =
z−x z−x z−x
f (y) − f (x) y−x
= +ε ≥ i(x, y) + ε.
x−y z−x
The cases of f (x) > f (y) and (or) f (z) − α(z) < −ε|x − y| are similar. Moreover we have
f (z) > α(z) + ε|x − y| = f (x) ± i(x, y)(z − x) + ε(y − x) .
Thus
2|z − x| ≥ |f (z) − f (x)| + i(x, y)|z − x| ≥ f (z) − f (x) ± i(x, y)(z − x) ≥ ε|x − y| ,
thus |z − x| ≥ ε|x−y|
2 , which finishes the proof of the lemma with δ = ε/2. 

Take a constant n > 1/ε. If begin with an interval [x, y] and apply the lemma n times, we end up with
an interval of length at least |x − y|δ n , which is either covered, or has inclination at least nε — the second
is impossible, however, as the inclination of any interval is at most 1. Thus for any interval we can find its
subinterval of length at least δ n times the length of the original, which is covered. Thus we have the following
corollary: for any interval [x, y] ⊂ [0, 1] there exists a covered subinterval [x0 , y 0 ] of [x, y] of length at least
c|x − y| for some fixed constant c.
Now we are ready to solve the S problem. We shall construct a family of disjoint intervals Ci ⊂ [0, 1], with
the Lebesgue measure of [0, 1] \ Ci no larger than ε. Each of these intervals will be covered, and thus we
shall be able to cover the whole graph of f by rectangles — each interval is covered, and thus the appropriate
piece of the graph is contained in a rectangle of width at most 2ε, while the remaining part can be covered by
a countable family of vertical rectangles of total width at most 2ε. As ε was arbitrary, this will end the proof.
The construction of Ci s follows directly from the corollary — we choose C0 = [x0 , y0 ] to be the interval given
by the corollary for [0, 1], then C1 and C2 the intervals for [0, x0 ] and [y0 , 1] respectively, then (in the third
step), C3 , C4 , C5 and C6 are given for [0, x1 ], [y1 , x0 ], [y0 , x2 ] and [y2 , 1] respectively, and so on. In each step a
constant fraction of measure is removed, thus after sufficiently many steps no more than ε measure remains. 
 The 21st Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 31st March 2011
Category I

Problem 1
(a) Is there a polynomial P (x) with real coefficients such that
1 k+2
P = ,
k k
for all positive integers k?

(b) Is there a polynomial P (x) with real coefficients such that

1 1
P = ,
k 2k + 1
for all positive integers k?
Solution (a) YES. It suffices to define a polynomial W (x) as follows

W (x) = 2x + 1.

(b) NO. Suppose that such a polynomial W (x) exists. Define a polynomial F (x) as follows

F (x) = (x + 2)W (x) − x.

Then 1  1 1

1
F +2 W
= − = 0,
k k k k
for all k ∈ N. Hence, the polynomial F (x) admits infinitely many zeros. Consequently,

(x + 2)W (x) − x = 0,

for all x ∈ R. But this implies that

x
W (x) = ,
x+2
for all x ∈ R – a contradiction. 
 The 21st Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 31st March 2011
Category I

Problem 2 Let (an )∞n=1 be unbounded and strictly increasing sequence of positive reals such that the arithmetic
mean of any four consecutive terms an , an+1 , an+2 , an+3 belongs to the same sequence. Prove that the sequence
an+1 /an converges and find all possible values of its limit.
Solution Since an < an+1 < an+2 < an+3 , one has
1
an < (an + an+1 + an+2 + an+3 ) < an+3 ,
4
thus (an + an+1 + an+2 + an+3 )/4 ∈ {an+1 , an+2 }. Hence for any n ∈ N precisely one of the two identities

an + an+1 + an+2 + an+3 = 4an+1 (1)

or
an + an+1 + an+2 + an+3 = 4an+2 (2)
holds. Let A be the set of indices n ∈ N for which (1) holds and let B be the set of indices n ∈ N for which (2)
holds. Clearly, A ∪ B = N, A ∩ B = ∅. We shall prove that one of A or B is finite. Indeed, suppose the contrary,
that both A and B are infinite. Since A and B partition N, there exists a positive integer k, such that k ∈ B,
k + 1 ∈ A. From (1) and (2), it follows that

ak + ak+1 + ak+2 + ak+3 = 4ak+2 and ak+1 + ak+2 + ak+3 + ak+4 = 4ak+2 .

Hence ak = ak+4 , which contradicts the fact that an is strictly increasing. We now consider two cases.
Case 1) The set A is infinite, the set B is finite. By (1), the sequence an satisfies a linear recurrence
an − 3an+1 + an+2 + an+3 = 0 for all n > n0 . The characteristic polynomial of the linear recurrence

φ(λ) = λ3 + λ2 − 3λ + 1 = (λ − 1)(λ2 + 2λ − 1)
√ √
has roots λ1 = 1, λ2 = −1 − 2, λ3 = −1 + 2. Hence
√ √
an = C1 + C2 (−1 − 2)n + C3 (−1 + 2)n , C1 , C2 , C3 ∈ R, n > n0 .

Observe that λ2 < −1, 0 < λ3 < 1. If C2 6= 0, then limn→∞ |an | = ∞ and an alternates in sign for n sufficiently
large which contradicts the monotonicity property. If C2 = 0, then the sequence an is bounded, which leads to
the contradiction again. Thus we reject the case one.
Case 2) The set A is finite, the set B is infinite. By (1), the sequence an satisfies a linear recurrence
an + an+1 − 3an+2 + an+3 = 0 for all n > n0 . The characteristic polynomial of the linear recurrence

φ(λ) = λ3 − 3λ2 + λ + 1 = (λ − 1)(λ2 − 2λ − 1)

√ √
has roots λ1 = 1, λ2 = 1 − 2, λ3 = 1 + 2. Hence
√ √
an = C1 + C2 (1 − 2)n + C3 (1 + 2)n , C1 , C2 , C3 ∈ R, n > n0 .

Note that −1 < λ2 < 0, λ3 > 1. If C3 ≤ 0, then the sequence an is bounded from above. Hence C3 > 0 so
an ∼ C3 λn3 as n → ∞ . The standard limit calculation now shows that bn converges and has limit value
an+1 √
lim bn = lim = λ3 = 1 + 2.
n→∞ n→∞ an


 The 21st Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 31st March 2011
Category I

Problem 3 Prove that

∞ ∞
X
k 1 + x2k+2 X
k xk
x = (−1)
(1 − x2k+2 )2 (1 − xk+1 )2
k=0 k=0

for all x ∈ (−1, 1).

Solution We use the binomial series
∞
1 X
2
= (j + 1)uj , |u| < 1
(1 − u) j=0

to get
∞ ∞ ∞ ∞ X ∞
X 1 + x2k+2 X X X
xk = x k
(1 + x 2k+2
) (j + 1)x j(2k+2)
= xk (1 + x2k+2 )(j + 1)xj(2k+2) =
(1 − x2k+2 )2 j=0 j=0 k=0
k=0 k=0
∞ ∞ ∞
x2
 
X
2j
X
k 2k+2 j2k
X
2j 1
= (j + 1)x x (1 + x )x = (j + 1)x + =
j=0 j=0
1 − x2j+1 1 − x2j+3
k=0
∞ ∞ ∞ ∞
X (j + 1)x2j X jx2j X (2j + 1)x2j d X
= + = =− log(1 − x2j+1 )
j=0
1 − x2j+1 j=1
1−x 2j+1
j=0
1 − x 2j+1 dx j=0

and
∞ ∞ ∞ ∞ ∞ ∞
X (−x)k X
k
X
(k+1)j
X
j
X
k kj
X (j + 1)xj
= (−x) (j + 1)x = (j + 1)x (−x) x = =
(1 − xk+1 )2 j=0 j=0 j=0
1 + xj+1
k=0 k=0 k=0
∞
d X
= log(1 + xj+1 ).
dx j=0

The proposition now follows by logarithmic differentiation of the classical identity

∞ ∞
Y 1 Y
2n+1
= (1 + xn ),
n=0
1 − x n=1

which can be proved as follows:

∞ ∞ Q∞ Q∞ ∞
Y Y
n 1 − x2n n=1 (1 − x2n ) n=1 (1 − x2n ) Y 1
(1 + x ) = n
= Q∞ n)
= Q∞ 2n )
Q∞ 2n−1 )
= 2n−1
.
n=1 n=1
1 − x n=1 (1 − x n=1 (1 − x n=1 (1 − x n=1
1 − x


 The 21st Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 31st March 2011
Category I

Problem 4 Let a, b, c be elements of finite order in some group. Prove that if a−1 ba = b2 , b−2 cb2 = c2 and
c−3 ac3 = a2 , then a = b = c = e, where e is the unit element.
Solution Let r(g) denote the rank of g ∈ G. Assume that the assertion does not hold. Let p be the smallest
prime number dividing r(a)r(b)r(c). Without loss of generality we can assume that p | r(b) (if p | r(a) or
p | r(c), then the reasoning is the same). Then there exists k such that r(b) = pk. Let d := bk . Then r(d) = p.
m
Lemma For any m ∈ N, a−m dam = d2 .
Proof First we prove that
a−1 da = d2 .
Indeed, multiplying the equation a−1 ba = b2 k-times with itself we get

(a−1 ba)(a−1 ba) · · · (a−1 ba) = b2 b2 · · · b2 ;

and hence
a−1 bk a = (b2 )k = (bk )2 .
Now, the assertion of the above lemma follows from the following calculations:
2 2 3 m
d = ad2 a−1 = a(ad2 a−1 )2 a−1 = a2 d2 a−2 = a2 (ad2 a−1 )2 a−2 = a3 d2 a−3 = · · · = am d2 a−m . (1)

Observe that Fermat’s little theorem implies that 2p ≡ 2 (mod p). Consequently,
p
a−p dap = d2 = d2 = a−1 da. (2)

Since gcd(r(a), p − 1) = 1, there exist integers r and s such that

r · r(a) + s · (p − 1) = 1. (3)

From (2) we get

a−l(p−1) dal(p−1) = d,
for all l ∈ Z (see the calculations in (1)). Finally, putting l := s, we obtain
(3)
d = a−s(p−1) das(p−1) = arr(a)−1 da−rr(a)+1 = a−1 da = d2 ,

which implies that d = e, a contradiction. 

 The 21st Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 31st March 2011
Category II

Problem 1 Let n > k and let A1 , . . . , Ak be real n × n matrices of rank n − 1. Prove that

A1 · . . . · Ak 6= 0 .
g f
Solution Consider two linear operators V → V → V of an n-dimensional vector space V. If Ker(f ) ⊂ Im(g),
then dim (Im (f g)) = dim (Im(g)) − dim (Ker(f )) . But we have the inequality

dim (Im(f g)) ≥ dim (Im(g)) − dim (Ker(f ))

in the general case. Applying the correspondence between linear operators and matrices, we obtain the inequality
rank (AB) ≥ rank B − (n − rank A) for every two matrices A and B. The inequality rank (A1 · . . . · Ak ) ≥
(rank (A1 ) + . . . + rank (Ak )) − (k − 1) n can be deduced from the inequality rank (AB) ≥ rank A + rank B − n
by the simple induction. We obtain the inequality rank (A1 · . . . · Ak ) ≥ k (n − 1) − (k − 1) n = n − k in our
case. Thus, if k < n then rank (A1 · . . . · Ak ) ≥ 1 and the product A1 · . . . · Ak can not be equal to zero. 
 The 21st Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 31st March 2011
Category II

Problem 2 Let k be a positive integer. Compute

∞ X
∞ ∞
X X 1
··· .
n1 =1 n2 =1 nk
n n
=1 1 2
. . . nk (n 1 + . . . + nk + 1)

Solution
∞ X
∞ ∞ ∞ X ∞ ∞ Z 1
X X 1 X X 1
··· = ··· xn1 +...+nk dx =
n1 =1 n2 =1 nk =1
n 1 n 2 . . . nk (n1 + . . . + n k + 1) n1 =1 n2 =1 nk =1
n 1 n 2 . . . n k 0
Z 1 X ∞ X ∞ ∞ n1 +...+nk Z 1 Z ∞
X x −u
= ··· dx = k
(− log(1−x)) dx = [1−x = e ] = uk e−u du = Γ(k+1) = k!
0 n =1 n =1 n n . . . nk
n =1 1 2 0 0
1 2 k


 The 21st Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 31st March 2011
Category II

p(z)
Problem 3 Let p and q be complex polynomials with deg p > deg q and let f (z) = . Suppose that all
q(z)
roots of p lie inside the unit circle |z| = 1 and that all roots of q lie outside the unit circle. Prove that

deg p − deg q
max |f 0 (z)| > max |f (z)|.
|z|=1 2 |z|=1

Solution Without loss of generality we can assume that the maximum of |f | is attained at 1.
n1
Q n2
Q
Let p(z) = a (z − ck ) and q(z) = b (z − d` ) where n1 = deg p and n2 = deg q. Then
k=1 `=1

n n
f 0 (z) X1
1 X2
1
= − .
f (z) z − ck z − d`
k=1 `=1

Since |ck | < 1 and |d` | > 1 for all k and `, we have
1 1
Re >
1 − ck 2
and
1 1
Re < .
1 − dk 2
Therefore,
|f 0 (1)| f 0 (1) 1 1 deg p − deg q
≥ Re > n1 · − n2 · =
|f (1)| f (1) 2 2 2
and
|f 0 (1)| deg p − deg q
max |f 0 (z)| ≥ |f 0 (1)| = · |f (1)| ≥ max |f (z)| .
|z|=1 |f (1)| 2 |z|=1


 The 21st Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 31st March 2011
Category II

Problem 4 Let Q[x] denote the vector space over Q of polynomials with rational coefficients in one variable
x. Find all Q-linear maps Φ : Q[x] → Q[x] such that for any irreducible polynomial p ∈ Q[x] the polynomial
Φ(p) is also irreducible.
(A polynomial p ∈ Q[x] is called irreducible if it is non-constant and the equality p = q1 q2 is impossible for
non-constant polynomials q1 , q2 ∈ Q[x].)
Solution
The answer is Φ(p(x)) = ap(bx + c) for some non-zero rationals a, b and some rational c. It is clear that
such operators preserve irreducibility. Let’s prove that any irreducibility-preserving operator is of such form.
We start with the following
Lemma 1 Assume that f, g ∈ Π are two polynomials such that for all rational numbers c the polynomial f + cg
is irreducible. Then either g ≡ 0, or f is non-constant linear polynomial and g is non-zero constant.
Proof Let g(x0 ) 6= 0 for some rational x0 . Then for c = −f (x0 )/g(x0 ) we have (f + cg)(x0 ) = 0, so the
polynomial f + cg is divisible by x − x0 . Hence f + cg = C(x − x0 ) for some non-zero rational C. Choose
x1 6= x0 such that g(x1 ) 6= 0. Then for c1 = −f (x1 )/g(x1 ) 6= c (since f (x1 ) + cg(x1 ) = C(x1 − x0 ) 6= 0) we have
f + c1 g = C1 (x − x1 ). Subtracting we get that (c1 − c)g is linear, hence g is linear, hence f too. If f (x) = ax + b,
g(x) = a1 x + b1 , then a 6= 0 (since f is irreducible) and if a1 6= 0, then for c = −a/a1 the polynomial f + cg is
constant, hence not irreducible. So a1 = 0 and we are done. 

Now denote gk = Φ(xk ).

Lemma 2 g0 is non-zero constant and g1 is non-constant linear function.
Proof Since x+c is irreducible for any rational c, we get that g1 +cg0 is irreducible for any rational c. By Lemma
1 we have that either g0 = 0 or g0 is constant and g1 is linear non-constant. Assume that g0 = 0. Note that for
any rational α one may find rational β such that x2 + αx + β is irreducible, hence g2 + αg1 = Φ(x2 + αx + β) is
irreducible for any rational α. It follows by Lemma 1 that g1 is constant, hence not irreducible. A contradiction,
hence g0 6= 0 and we are done. 

Denote g0 = C, g1 (x) = Ax + B. Consider the new operator p(x) → C −1 Φ(p(A−1 Cx − A−1 B)). This
operator of course preserves irreducibility, consider it instead Φ.
Now g0 = 1, g1 (x) = x and our goal is to prove that gn = xn for all positive integers n. We use induction by
n. Assume that n ≥ 2 and gk (x) = xk is already proved for k = 0, 1, . . . , n − 1. Denote h(x) = gn (x) − xn and
assume that h is not identical 0. For arbitrary monic irreducible polynomial f of degree n we have Φ(f ) = f +h,
hence f + h is irreducible aswell. Choose rational x0 such that h(x0 ) 6= 0, our goal is to find irreducible f such
that f (x0 ) = −h(x0 ) and hence f + h has a root in x0 .
There are many ways to do it, consider one of them, via Eisenstein’s criterion. Recall it.
Eisenstein’s criterion Assume that f (x) = an xn + · · · + a0 is a polynomial with rational coefficients and p is a
prime number so that ak = bk /ck with coprime integers bk , ck such that bk is divisible by p for k = 0, 1, . . . , n−1,
both bn and cn are not divisible by p and b0 is not divisible by p2 . Then f is irreducible.
Without loss of generality, x0 = 0 (else denote x − x0 by new variable). Then we want to find an irreducible
polynomial f (x) = xn + an−1 xn−1 + · · · + a1 x − h(0). Denote −h(0) = u/v for coprime positive integer v and
non-zero integer u. Take L = 6uv and consider the prime divisor p of the number vLn /u − 1. Clearly, p does
not divide 6uvL. Then consider the polynomial (x + L)n − Ln + u/v. If vLn /u − 1 is not divisible by p2 , then
we are done by Eisenstein’s criterion (with new variable y = x + L). If vLn /u − 1 is divisible by p2 , then add
px to our polynomial and now Eisenstein’s criterion works.
Unless h(x) = −xn + . . . , the polynomial f + h is not linear and so is not irreducible. If n ≥ 3, then we may
add px2 or 2px2 to our polynomial f and get non-linear f + h (but still irreducible f ). Finally, if n = 2, and
h(x) = −x2 + ax + b, then choose irreducible polynomial of the form f (x) = x2 − ax + c and get f + h being
constant (hence not irreducible).
The induction step and the whole proof are finished. 
 The 22nd Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 30th March 2012
Category I

Problem 1 Let f : [0, 1] → [0, 1] be a differentiable function such that |f 0 (x)| 6= 1 for all x ∈ [0, 1]. Prove that
there exist unique points α, β ∈ [0, 1] such that f (α) = α and f (β) = 1 − β.
Solution Existence: Since f is derivable in [0, 1], then f is continuous in [0, 1]. Considering the functions
g(x) = f (x) − x and h(x) = f (x) − (1 − x) that are continuous in [0, 1] and applying Bolzano’s theorem we get
that exists α ∈ [0, 1] such that g(α) = 0 and β ∈ [0, 1] with h(β) = 0. That is, there exist α, β ∈ [0, 1] for which
f (α) = α and f (β) = 1 − β.
Uniqueness: Suppose that there exist α, α0 ∈ [0, 1], α < α0 ) such that f (α) = α and f (α0 ) = α0 . On account
of Lagrange’s theorem, there exists θ ∈ (α, α0 ) ⊂ [0, 1] such that

f (α0 ) − f (α) α0 − α
f 0 (θ) = = =1
α0 − α α0 − α
contradiction. Likewise, if we assume that there exist β, β 0 ∈ [0, 1], ( β < β 0 ) such that f (β) = 1 − β and
f (β 0 ) = 1 − β 0 . On account of Lagrange’s theorem, there exists θ0 ∈ (β, β 0 ) ⊂ [0, 1] such that

f (β 0 ) − f (β) (1 − β 0 ) − (1 − β)
f 0 (θ0 ) = 0
= = −1
β −β β0 − β
contradiction. This completes the proof. 
 The 22nd Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 30th March 2012
Category I

Problem 2 Determine all 2 × 2 integer matrices A having the following properties:

1. the entries of A are (positive) prime numbers,
2. there exists a 2 × 2 integer matrix B such that A = B 2 and the determinant of B is the square of a prime
number.
Solution Let  
p1 p2
A= = B2,
p3 p4
and d = det(B) = q 2 with p1 , p2 , p3 , p4 , q ∈ P; here P denotes the set of positive prime numbers.
By Cayley-Hamilton Theorem,
B 2 = tr(B)B − det(B)E,
where E is the 2 × 2 identity matrix. Without loss of generality, we assume that tr(B) ≥ 0, otherwise, replace
B by −B. The equality  
2 p1 + d p2
tr(B)B = B + dE = A + dE =
p3 p4 + d
implies that tr(B) divides the numbers p2 and p3 . Moreover,

(tr(B))2 = tr(tr(B)B) = p1 + p4 + 2d ≥ 2 + 2 + 8 = 12 =⇒ tr(B) > 3.

It follows that    
1 p1 + d p2 a 1
tr(B) = p2 = p3 , and B = =
tr(B) p3 p4 + d 1 b
for some positive integers a and b. Hence,

a2 + 1
 
a+b
A = B2 = .
a+b b2 + 1

The numbers a2 + 1, b2 + 1, a + b cannot all be odd, thus, one of them equals 2. Since ab = d + 1 = q 2 + 1 ≥ 5
we have max(a, b) ≥ 3. Hence, a + b ≥ 3 + 1 > 2.
Now we assume that a2 + 1 ≤ b2 + 1. Then a2 + 1 = 2 and a = 1. Note that d = ab − 1 = b − 1 and
0 < p2 = a+b = b+1 = d+2 = q 2 +2. If q 6= 3 then q 2 ≡ 1 mod 3 =⇒ p2 ≡ 0 mod 3 =⇒ p2 = 3 =⇒ q 2 = 1,
which is impossible. Hence, q = 3, b = p2 − a = 32 + 2 − 1 = 10,
   
1 1 2 11
B= , and A = B 2 = .
1 10 11 101

Similarly, if a2 + 1 > b2 + 1 we obtain the matrix

 
101 11
A= .
11 2

Answer:    
2 11 101 11
A= , and A = .
11 101 11 2

 The 22nd Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 30th March 2012
Category I

Problem 3 Determine the smallest real number C such that the inequality
x 1 y 1 z 1
√ · +√ · +√ · ≤C
yz x + 1 zx y + 1 xy z + 1

holds for all positive real numbers x, y and z with

1 1 1
+ + = 1.
x+1 y+1 z+1
Solution In what follows we shall deal with the harder version of the problem only.
1. We consider the case x = y = t. Then
2 1
+ =1
t+1 z+1
that is
2
z= .
t−1
Thus the inequality under consideration becomes
2
2t 1 1
r · + √t−1 · 2 ≤C
2 t+1 t·t t−1+1
t·
t−1
that is √ √ √
2· t· t−1 2
+ ≤ C.
t+1 t(t + 1)
√
Letting here t → ∞ leads to C ≥ 2.

2. We are now going to prove that always

x 1 y 1 z 1 √
√ · +√ · +√ · < 2.
yz x + 1 zx y + 1 xy z + 1

In order to achieve this goal we make use of the following transformation

1 1 1
a= , b= , c= .
x+1 y+1 z+1

Then the three new variables satisfy a, b, c ∈ (0; 1) and are subject to the condition a + b + c = 1.
Furthermore
1−a 1−b 1−c
x= , y= , z=
a b c
that is (due to 1 − a = b + c, etc.)

b+c c+a a+b

x= , y= , z=
a b c
yield for the claimed inequality
√ √ √
(a + b) ab (b + c) bc
(c + a) ca √
p +p +p < 2.
(b + c)(c + a) (c + a)(a + b) (a + b)(b + c)

Upon clearing fractions this inequality becomes

p p p p
(a + b) ab(a + b) + (b + c) bc(b + c) + (c + a) ca(a + c) < 2(a + b)(b + c)(c + a).
We smuggle the condition 1 = a + b + c into the inequality and get
p p p p
(a + b) ab(a + b) + (b + c) bc(b + c) + (c + a) ca(a + c) < 2(a + b)(b + c)(c + a)(a + b + c).

Next, we deal with the right-hand expressions. For them we have

p p
(a + b)(b + c)(c + a) = ab(a + b) + bc(b + c) + ca(c + a) + 2abc

and √ p p
2(a + b + c) = 2(a + b + c)2 = (a + b)2 + (a + c)2 + (b + c)2 + 2(ab + bc + ca)
But, employing the Cauchy-Schwarz inequality yields for our inequality
p p p
(a + b) ab(a + b) + (b + c) bc(b + c) + (c + a) ca(a + c) ≤
p p
(a + b)2 + (b + c)2 + (c + a)2 · ab(a + b) + ac(a + c) + bc(b + c).

This together with the two previously stated equations completes the proof. It is also evident that there
cannot exist any triples (a, b, c), and thus also (x, y, z), yielding equality.

 The 22nd Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 30th March 2012
Category I

Problem 4 Find all positive integers n for which there exists a positive integer k such that the decimal
representation of nk starts and ends with the same digit.
Solution The number nk ends with zero whenever n is divisible by 10 and starts with nonzero digit. We show
that the claim is true for all other n’s.
It can be easily shown that all the numbers

n, n5 , n9 , . . . , n4m+1 , . . . (1)

ends with the same digit. In fact, n5 − n = n(n − 1)(n + 1)(n2 + 1) is even and for each possible reminder of n
modulo 5 there is a factor divisible by 5 in this product. Thus n5 − n is divisible by 10 and in the same fashion
we can show this for n9 − n5 , n13 − n9 , . . .
Now it suffices to show that for any nonzero digit c there is a number in the sequence (1) which starts with
c. For any nonnegative integer m put dm = n4m+1 /10l , where l is the greatest integer for which 10l ≤ n4m+1 .
Thus 1 ≤ dm < 10 and bdm c is the first digit of n4m+1 . Clearly all the dm ’s are different, since for m0 > m we
have 0 0 0
dm0 n4m +1 /10l n4(m −m)
= 4m+1 = 6= 1
dm n /10l 10l0 −l
(the numerator is not a power of 10 for n not divisible by 10).
The sequence (dm )∞ 2 ε
m=1 has the following property: If dm+i = dm · q, then dm+2i = dm · q · 10 , where
ε ∈ {−1, 0, 1}. This is true since when
0 00
dm = n4m+1 /10l , dm+i = n4(m+i)+1 /10l and dm+2i = n4(m+2i)+1 /10l ,
0
we have q = dm+i /dm = n4i /10l −l and so
00 0 00
−l
dm+2i /dm = n8i /10l = q 2 · 102l −l−l = q 2 · 10ε

for some integer ε. But dm , dm · q, dm · q 2 · 10ε ∈ [1, 10), i. e. ε ∈ {−1, 0, 1}.

Since all the terms of the sequence (dm )∞ m=1 are different and all lie in the interval [1, 10), there have to
be two terms dm and dm0 such that |dm0 − dm | < 10 1
. Without loss of generality let m0 > m. There are two
possibilities.
1
Let dm0 > dm . Then we have dm0 < dm + 10 . Thus
1
dm + 10 1 1
1 < q = dm0 /dm < =1+ ≤1+ 10 .
dm 10dm

By previous remark dm ·q 2 lies in the studied sequence, whenever it lies in the interval [1, 10). Repeating this idea
we have the numbers dm , dm · q, dm · q 2 , dm · q 3 , . . . , dm · q i all lying in the studied sequence and after overrunning
the value 10 we have the numbers dm · q i+1 /10, dm · q i+2 /10, . . . in the sequence, and so on. Computing the
difference of two consecutive terms in this recurrence we get

dm · q j+1 − dm · q j = dm · q j (q − 1) < dm · q j · 1
10 <1 for j < i,
j+1 j j j 1
dm · q /10 − dm · q /10 = dm · q (q − 1)/10 < dm · q /10 · 10 <1 for j > i

and for the first term after overrunning 10 we obtain

dm · q i+1 /10 = dm · q i /10 · q < 10/10 · (1 + 1

10 ) = 11
10 < 2.

Since the difference is less then 1 and after overrunning we jump into the interval [1, 2), we must get at least
one dm+j(m0 −m) in the interval [c, c + 1) for every nonzero digit c.
1
Let dm0 < dm . Then we have dm < dm0 + 10 . Thus
1
dm0 + 10 1 1
1 < q = dm /dm0 < =1+ ≤1+ 10 .
dm0 10dm0
In the very similar way as in the previous case (new terms are generated by dividing instead of multiplying by
q) we obtain the new sequence of terms with consecutive differences less then 1 and after underrunning 1 we
jump to
10
dm /q i+1 · 10 = dm /q i · 10/q > 1 · 100
1 = 11 > 9.
1 + 10
Thus also in this case we must obtain some dm+j(m0 −m) in the interval [c, c + 1) for every nonzero digit c.
This ends the proof.
Answer. Integers satisfying the given conditions are all integers not divisible by 10. 
 The 22nd Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 30th March 2012
Category II

Problem 1 Let f : [1, ∞) → (0, ∞) be a non-increasing function such that

f (2n+1 ) 1
lim sup < .
n→∞ f (2n ) 2

Prove that Z ∞
f (x) dx < ∞ .
1
Solution Since
2n+1 f (2n+1 )
lim sup < 1,
n→∞ 2n f (2n )
then by ratio test we obtain that the series
∞
X
2n f (2n )
n=1

converges. Using Cauchy condensation test we obtain that

∞
X
f (n)
n=1

converges. Now, by integral test for convergence we have

Z ∞
f (x) dx < ∞.
1


 The 22nd Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 30th March 2012
Category II

Problem 2 Let M be the (tridiagonal) 10 × 10 matrix

 
−1 3 0 ··· ··· ··· 0
.. 
 

 3 2 −1 0 .
.. 
 
 ..
 0
 −1 2 −1 . . 
M =  ... .. ..  .

 0 −1 2 . 0 .
 . .. .. ..
 ..

 . . . −1 0 
 .
 ..

 0 −1 2 −1  
0 ··· ··· ··· 0 −1 2

Show that M has exactly nine positive real eigenvalues (counted with multiplicities).
Solution Let xT = (0, x1 , ..., x9 ). Then the direct calculation shows that

xT M x = x21 + (x2 − x1 )2 + · · · + (x9 − x8 )2 + x29 . (1)

Let λmin := min{λ | λ ∈ σ(M )} (recall that if a matrix M is symmetric then σ(M ) ⊂ R). Moreover, since M is
symmetric, there exists an orthogonal matrix C such that C T M C = diag{λmin , λ1 ..., λ9 }. Hence we infer that
y T (λmin I − M )y ≤ 0 for y ∈ R10 . Let y T = (1, −1, 0, ..., 0). Then 2λmin ≤ y T M y = −5. Thus λmin < 0.
Let V1 = {(0, x1 , ..., x9 ) | xi ∈ R} ⊂ R10 . Then, in view of (1), we have

yT M y ≥ 0 (2)

for any y ∈ V1 and y T M y = 0 if and only if y = 0.

Suppose on the contrary that M admits at least two nonpositive eigenvalues λ1 , λ2 ∈ σ(M ). Consequently,
there exist y1 , y2 ∈ R10 such that y1 ⊥ y2 , y1T y1 = y2T y2 = 1 and M yi = λi yi (i = 1, 2). Put V2 := span{y1 , y2 }.
Then for any y = α1 y1 + α2 y2 ∈ V2 one has

y T M y = α12 · λ1 + α22 · λ2 ≤ 0. (3)

Finally, we obtain that

dim V1 + dim V2 = 9 + 2 = 11 > 10.
Therefore V1 ∩V2 6= {0}. Take 0 6= y ∈ V1 ∩V2 . Then, in view of (2), y T M y > 0. But (3) implies that y T M y ≤ 0
– a contradiction. 
 The 22nd Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 30th March 2012
Category II

Problem 3 Let (A, +, ·) be a ring with unity, having the following property: for all x ∈ A either x2 = 1 or
xn = 0 for some n ∈ N. Show that A is a commutative ring.
Solution Denote by U (A) the multiplicative group of units of the ring A (U (A) = {x | x is invertible}). Note
first that (U (A), ·) is commutative, because if x, y ∈ U (A), (xy)2 = 1 ⇒ xy · xy = 1, and multiplying by x to
the left and by y to the right and using also the fact that x2 = 1 = y 2 , we get that

xy = yx. (1)

We now show that if

x∈
/ U (A) then 1 − x ∈ U (A).
Assume, by contradiction, that
∃x ∈
/ U (A) so y = 1 − x ∈
/ U (A). (2)
By hypothesis,
∃n and m ∈ N so xn = 0; y m = 0
and as
xy = x(1 − x) = x − x2 = (1 − x)x = yx
we get that X
(x + y)n+m = i
Cn+m xi y j = 0,
i+j=n+m

Note that whenever i + j = n + m we have

i ≥ n or j ≥ m and so xi = 0 or y j = 0;

So
1=x+y ∈
/ U (A),
which is a contradiction; thus (2) is proved.
Commutativity in A follows now from (1) and (2) with a case by case analysis: x, y ∈ A,

1. if x ∈ U (A), y ∈ U (A) then (1) ⇒ xy = yx ;

2. if x ∈ U (A), y ∈
/ U (A) then (2) ⇒ 1 − y ∈ U (A) and from (1) we have x(1 − y) = (1 − y)x ⇒ xy = yx ;
3. if x ∈
/ U (A), y ∈ U (A) analogous to the case 2 and
4. if x ∈
/ U (A), y ∈
/ U (A) then (2) ⇒ 1 − x, 1 − y ∈ U (A)

and using
(1 − x)(1 − y) = (1 − y)(1 − x) ⇔ 1 − x − y + xy = 1 − y − x + xy ⇔ xy = yx.
Now cases 1 → 4 above show that A is a commutative ring. 
 The 22nd Annual Vojtěch Jarník
International Mathematical Competition
Ostrava, 30th March 2012
Category II

Problem 4 Let a, b, c, x, y, z, t be positive real numbers with 1 ≤ x, y, z ≤ 4. Prove that

x y z y+z−x z+x−y x+y−z
+ + ≥ + + .
(2a)t (2b)t (2c)t (b + c)t (c + a)t (a + b)t

Solution We will use the following variant of Schur’s inequality.

Lemma 1 For arbitrary A, B, C > 0,

x(A − B)(A − C) + y(B − A)(B − C) + z(C − A)(C − B) ≥ 0.

Proof Without loss of generality we can assume A ≤ B ≤ C. Let U = B − A and V = C − B. Then

LHS = xU (U + V ) − yU V + z(U + V )V ≥ U (U + V ) − 4U V + (U + V )V = (U − V )2 ≥ 0.

Lemma 2 For every p > 0, Z ∞

1 1
k
= tk−1 e−pt dt.
p Γ(k) 0

Proof Substituting u = pt,

Z ∞ Z ∞
k−1 −pt 1 Γ(k)
t e dt = k uk−1 e−u du = .
0 p 0 pk


Now, applying Lemma 1 to A = e−at , B = e−bt , and C = e−ct , the statement can be proved as
Z ∞  
k−1 −at −bt −at −ct −bt −at −bt −ct −ct −at −ct −bt
0≤ t x(e − e )(e − e ) + y(e − e )(e − e ) + z(e − e )(e − e ) dt
0
Z ∞  
k−1 −2at −2bt −2ct −(b+c)t −(c+a)t −(a+b)t
= t xe + ye + ze − (y + z − x)e − (z + x − y)e − (x + y − x)e dt
0
 
x y z y+z−x z+x−y x+y−z
= Γ(k) + + − − − .
(2a)k (2b)k (2c)k (b + c)k (c + a)k (a + b)k