Baltic Way 2007

Version: English
Copenhagen, November 3, 2007

Suggested solutions

1. We first prove the lemma that if x1 > · · · > x2n then the grouping

{{x1 , x2 }, . . . , {x2n−1 , x2n }} (1)

gives the largest sum of products of pairs of these numbers.

Let a be the largest and b the second largest among the numbers xi . Consider a grouping of these numbers
into pairs such that a is paired with some c, and b is paired with some d, where c 6= b. Then a 6= d (otherwise
a would be together with b). Furthermore, b > c since otherwise the choice of b implies a = c or b = c which
are both excluded. Now

ab + cd = ac + a(b − c) + bd − (b − c)d
= ac + bd + (a − d)(b − c) > ac + bd,

that is, replacing the pairs {a, c} and {b, d} by the pairs {a, b} and {c, d} makes the sum larger. If the
two largest numbers are paired already, we can do the same to the remaining numbers. So whenever the
grouping is different from (1), the sum of the products of pairs can be made larger.
1 1 1 1
Now it suffices to prove that an = 1 · 2 + ··· + 2n−1 · 2n < 1. We have
1 1 1
an = 1·2 + 3·4 + ··· + (2n−1)·(2n)
2−1 4−3 2n−(2n−1)
= 1·2 ++ · · · + (2n−1)·(2n)
3·4
     
= 11 − 12 + 13 − 14 + · · · + 2n−1
1 1
− 2n
≤ 11 − 12 + 21 − 13 + 13 − 14 + · · · + 2n−11 1





+ 2n
1
=1− 2n
< 1.

2. Assume that such an exact sequence exists. Then by induction we must have a22k = a22k−2 + a4k−2 a2 =
a22k−2 = 0.
Next we prove by induction that a2n+1 = (−1)n . We have a3 = a22 − a21 = −1 and

a4k+1 = a22k+1 − a22k = 1,

a4k+3 = a22k+2 − a22k+1 = −1,

when k ≥ 1. This shows that if such a sequence exists, necessarily a2007 = −1.
It remains to show that the sequence defined by an = 0 for n even, an = 1 when n ≡ 1 (mod 4) and an = −1
when n ≡ 3 (mod 4) is exact:
If n and m have the same parity, then n − m and n + m are both even, and then clearly a2n − a2m = 0 =
an−m an+m .
If n is odd and m is even, n − m ≡ n + m (mod 4), so a2n − a2m = 1 = an−m an+m , since both factors are
either −1 or +1.
Finally, if n is even and m is odd, n − m 6≡ n + m (mod 4), so a2n − a2m = −1 = an−m an+m , since exactly
one factor is −1 and one factor is +1.

— 1 —
3. Consider the polynomial P (x) = G(x) − F (x). It has degree at most 2n + 1. By the condition (1) we
have P (x) ≥ 0 for all real x. By the condition (2) the numbers x1 , x2 , . . ., xn are roots of P . Since P is
non-negative, each of these roots must have even multiplicity, and therefore P must be divisible by (x − xi )2
for i = 1, 2, . . . , n. In other words,
P (x) = Q(x)(x − x1 )2 (x − x2 )2 · · · (x − xn )2
for some polynomial Q. Calculating degrees we see that deg Q = deg P − 2n ≤ 1. On the other hand, we
have Q(x) ≥ 0 for all real x. This can be possible only if Q is constant. Hence
G(x) − F (x) = a(x − x1 )2 (x − x2 )2 · · · (x − xn )2
for some real constant a ≥ 0. Similarly we prove that
H(x) − F (x) = b(x − x1 )2 (x − x2 )2 · · · (x − xn )2
for some b ≥ 0. Now we compute that
F (x) + H(x) − 2G(x) = (b − 2a)(x − x1 )2 (x − x2 )2 · · · (x − xn )2 .
By the assumption (3) the above number is equal to 0 for some value of x = x0 different from x1 , x2 , . . .,
xn . Looking at the right-hand side we see that this forces b − 2a = 0, so the expression becomes identically
zero. In other words, we have F (x) + H(x) − 2G(x) = 0 for all real x, which is what we wanted.
4. Rewrite the two factors on the left-hand side:
2S + n = (a1 + a2 + · · · + an ) + (a2 + a3 + · · · a1 ) + 1 + · · · + 1
2S + a1 a2 + a2 a3 + · · · + an a1 = (a2 + a3 + · · · + a1 ) + (a1 + a2 + · · · an ) + a1 a2 + a2 a3 + · · · + an a1
Applying the Cauchy-Schwarz inequality to the 3n-vectors
√ √ √ √ √ √ √ √ √ √
( a1 , . . . , an , a2 , . . . , a1 , 1, . . . , 1) and ( a2 , . . . , a1 , a1 , . . . an , a1 a2 , . . . , an a1 )
we obtain
n
X √
2
(2S + n)(2S + a1 a2 + a2 a3 + · · · an a1 ) ≥ 3 ai ai+1
i=1

with an+1 = a1 .
5. Set y = 1 in the first equation. This gives f (x) = f (x)f (−1) − f (x) + f (1), that is, f (x)(2 − f (−1)) = f (1).
Since f is not constant, we must have f (−1) = 2 and f (1) = 0. Substituting −x instead of x and y = −1 in
the first equation gives f (x) = f (−x)f (1) − f (−x) + f (−1) = −f (−x) + 2. If we let g(x) = 1 − f (x), this
means that g is an odd function.
Rewriting the first equation in terms of g gives
g(xy) = 1 − f (xy) = 1 − ((1 − g(x))(1 − g(−y)) − (1 − g(x)) + (1 − g(y)))
= −g(x)g(−y) + g(−y) + g(y) = g(x)g(y).
Now the second equation gives (since g(1) = −g(−1) = 1)
1 1 g(x)
1 − g(1 − g(x)) = = = ,
1 − g(1/x) 1 − 1/g(x) g(x) − 1
that is,
1
g(1 − g(x)) = .
1 − g(x)

Since f takes all values except 1, g takes all values except 0. By setting y = 1 − g(x) it follows that
g(y) = 1/y for all y 6= 0, that is, f (x) = 1 − g(x) = 1 − 1/x.
It is easily verified that this function satisfies the conditions of the problem.

— 2 —
6. First note that, whenever the numbers are not ordered ascendingly, there exists a pair (i, j) (not necessarily
in the list) such that 1 ≤ i < j ≤ n and the ith number is exactly 1 greater than the jth number. For
proving it, let i be the least number such that the ith number is not i; then the permutation starts with
1, . . . , i − 1. Thus the ith number is greater than i; let it be k. As k − 1 ≥ i, the number k − 1 does not
occur at positions 1 to i in the permutation. Let j be the position of k − 1. Then (i, j) meets the required
condition.
Suppose Freddy chooses such a pair (i, j) on his first move. Obviously, the interchange of the ith and the
jth element does not affect the greater/smaller relationships between elements at other positions. It also
does not affect the greater/smaller relationship between the number at either the ith or the jth position
and numbers at other positions since the number at any other position is either greater than both numbers
interchanged or smaller than both of them. Consequently, the only greater/smaller relationship that changed
is between the ith and the jth position. Hence, after Freddy has completed the action first time, the list
consists of precisely those pairs (i, j) which would have been there if he started the whole process from the
new permutation.
Using this as the loop invariant, one easily deduces that when the pairs in the list are all gone then the
numbers are ordered increasingly.

7. The regular triangle with side length n can be divided into n2 regular triangles with side length 1 having
sides parallel with the original triangle. It is clear that every squiggle must cover exactly six of these smaller
triangles. Thus we get that 6 | n2 , which implies that 6 | n.
Assume now that n is divisible by 6, but not with 12, i.e. n = 12k + 6 for some non-negative integer k.
Colour the large triangle in a “triangular chessboard” fashion with black triangles on the boundary so that
no adjacent triangles have the same colour. Then each squiggle covers either two or four black triangles.
The total number of black triangles is then

(12k + 7)(12k + 6)
n + (n − 1) + · · · + 1 = = (12k + 7)(6k + 3),
2
which is an odd number and hence a covering is impossible to achieve.
It remains to prove that when 12 | n the required division is possible. It is enough to give an example
for n = 12, since triangles with side length 12m can be composed of these for any integer m. A suitable
construction is shown in the figure below.

8. We may associate with each five-element subset of {1, 2, . . . , n} a sequence a1 , a2 , . . ., an such that exactly
five of the ai s are ones and the rest are zeros. In particular, the non-isolated five-element sets correspond
to sequences, where two of the ones are adjacent and three other ones are adjacent. The number of such
sequences can be computed by considering the number of sequences b1 , b2 , . . ., bn−3 , where one of the bi s is
2, one is 3, and the rest are zeroes. The number of such sequences is (n − 3)(n − 4). But here sequences with
bi = 2, bi+1 = 3 and bi = 3 and bi+1 = 2 correspond to the same subset (having all the elements consecutive).
This means that subsets with all the numbers consecutive have been counted twice. The number of such
subsets equals n − 4. So the total number of non-isolated subsets is (n − 3)(n − 4) − (n − 4) = (n − 4)2 .

— 3 —
 9. Let a be any member of the society and A the board consisting of the candidates chosen by a. If everybody
is happy with A, we are too. Otherwise there is a member b such that the board B consisting of all his
candidates has an empty intersection with A. Let’s divide A = A1 ∪ A2 and B = B1 ∪ B2 , each of A1 , A2 ,
B1 , B2 consisting of five persons. We claim that at least one of the boards A1 ∪ B1 , A1 ∪ B2 , A2 ∪ B1 ,
A2 ∪ B2 will make everybody happy. Suppose, on the contrary, that x1 isn’t happy with A1 ∪ B1 , x2 isn’t
happy with A1 ∪ B2 , x3 isn’t happy with A2 ∪ B1 , and x4 isn’t happy with A2 ∪ B2 . Notice that some xi s
may coincide. It’s easy to check that there isn’t a board consisting of two persons making all the members
a, b, x1 , x2 , x3 , x4 happy – a contradiction.

10. No, it is not possible.

To prove this, note first that for any choice of 16 cells in the 18 × 18 table there exists a 2 × 2 square
containing exactly one of these cells. Indeed, there are more rows in the table than the chosen cells, so we
can choose two neighboring rows R1 , R2 such that R1 contains none of the chosen cells and R2 contains
some of them (but not the whole row, since there are 18 cells in this row and only 16 chosen cells). Thus
there are two neighbouring cells A, B in the row R2 , of which exactly one is chosen. Take also two cells
C, D in the row R1 in the same columns as A, B. Then the numbers A, B, C, D form a 2 × 2 square, in
which exactly one of the cells is among the chosen cells.
Thus if the answer to the problem is positive, it is in particular possible to have exactly one black cell in
some 2 × 2 square. However, it is not hard to see that changing colours of all cells in one column or all cells
in one row in the whole table does not change the parity of the number of black cells in this 2 × 2 square (it
either remains the same if there were two cells of opposite colour in the chosen column or row, or changes
by 2 if these two cells had the same colour). Initially every 2 × 2 square contains an even number of black
cells (namely, zero). Hence it is not possible to have exactly one black cell in any such square after a series
of operations.

11. If P is the circumcentre then Q is the intersection different from A of the circumcircle and the bisector of
∠BAC. If triangle A0 B 0 C 0 is obtained from triangle ABC by a 90◦ rotation in the direction ACB then the
oriented segment EB and hence RQ has the same direction as the oriented segment A0 C 0 , and CF and
hence RS has the same direction as A0 B 0 . The bisector of ∠QRS is then parallel to or coincident with the
bisector of ∠C 0 A0 B 0 and hence perpendicular to the bisector of ∠CAB. Since RQ = RS, the line QS is
then parallel to or coincident with the bisector, and since Q lies on the bisector, S then does so as well.
Let T be the point such that P QRT is a parallelogram. Since P T = QR and T R = P Q, the segments P T
and T R are both equal to the circumradius of triangle ABC. Furthermore, P T has the same direction as
QR and hence BE, end T R has the same direction as P Q and hence AD. From an argument analogous to
that above (interchange A and Q with B and T , respectively) it then follows that S lies on the bisector of
∠CBA. Thus S is the incentre of triangle ABC.
Alternative solution: Let O and I denote the circumcentre and incentre, respectively, of triangle ABC.
Let the bisectors of ∠BAC, ∠ABC and ∠ACB intersect the circumcircle again at K, L and M , respectively.
−−→ −−→ −→ −−→ −−→ −→ −−→ −→ −−→ −→
Then OK = P Q, OL = QR and OM = RS, so we must prove OK + OL + OM = OI. Without
−→ −→ −→ −−→
loss of generality assume ∠ABC ≥ ∠ACB. From ∠(OA, OL) = ∠ABC and ∠(OA, OM ) = ∠ACB
−→ −→ −−→ −→ −→ −−→ −−→ −→ −−→
we get ∠(OA, OL + OM ) = (∠ABC − ∠ACB)/2, whence ∠(OA, OL + OM ) + ∠(OK, OL + OM ) =
−→ −−→ −−→ −−→ −→ −−→ −−→
2×(∠ABC−∠ACB)/2+2×∠ACB+∠BAC = π. Thus we have OL+OM k AK, so if OK+OL+OM = OX,
the point X lies on the bisector of ∠BAC. Similarly X lies on the bisector of ∠ABC, so X = I.

12. Triangles AM X and CM Y are similar: ∠M XA = ∠M Y C = 90◦ and ∠M AX = ∠M CY as inscribed

angles. Therefore there exists a spiral homothety H that brings X to A and Y to C: Its centre is M , its
AM CM −−→ 1 −−→ −−→
angle is ∠XM A = ∠Y M C, and its coefficient is M X = M Y . By the definition M N = 2 (M X + M Y )
−−→ −−→ −−→ −−→ −−→ −−→
and M K = 12 (M A + M C). As H is a linear transformation we get H(M N ) = 12 (H(M X) + H(M Y )) =
1 −−→ −−→ −−→
2 (M A + M C) = M K, from which the desired follows.

13. If all the lines are parallel to each other, choose some plane perpendicular to them. Then the points may be
chosen to be the intersections of the lines with this plane.

— 4 —
 Otherwise, denote the acute or right angle between ti and ti+1 by αi , i = 1, . . . , k − 1, and the acute or
right angle between tk and t1 by αk . At least one of these angles is not 0, so cos α1 cos α2 · · · cos αk < 1.
Clearly, Pi , i = 2, . . . , k, is determined by a choice of P1 , so let Pk+1 be the projection of Pk on t1 . We must
prove that P1 may be chosen such that P1 and Pk+1 coincide.
If P1 moves along t1 with constant speed v, then Pk+1 moves along t1 with constant speed v · cos α1 ·
cos α2 · · · cos αk < v. Hence at some moment P1 and Pk+1 coincide.

14. The triangle DEB is right-angled (∠DEB = 90◦ ). Hence if the line EF passes through the midpoint of the
hypotenuse BD, we must have ∠F EB = ∠DBE. On the other hand, the lines BE and DC are parallel
and we have ∠DBE = ∠CDB. Thus ∠F EB = ∠CDB. But since ∠AEB = ∠AF B = 90◦ , the points
A, E, F , B lie on a circle and ∠F EB = ∠F AB = ∠CAB. Hence we see that ∠CDB = ∠CAB, and the
assertion follows.
D C

E B

15. Let B1 = D, and similarly let the incircle touch the sides AB and BC at the points C1 and A1 , respectively.
Let the second circle touch the ray BC at the point M . Let x = B1 C, y = AB1 . Obviously, A1 M = C1 A =
AB1 = y and A1 C = B1 C = x. Hence

CM = A1 M − A1 C = y − x.

On the other hand, CM is a tangent and CA is a secant to the second circle. Therefore, by powers with
respect to this circle,

CM 2 = CB1 · CA = x(x + y).

So we have the equation

(y − x)2 = x(x + y).

From this equation, we get y/x = 3.

16. Let a = m n
k and b = k where k is the least positive common denominator of a and b. Then k, m, n are
relatively prime, otherwise we could obtain a smaller positive common denominator by dividing them all by
their greatest common divisor.
m+n m2 +n2
By the conditions of the problem, s = k = k2 , giving

(m + n)k = m2 + n2 . (2)

This representation shows that each prime that divides both k and m divides also n and therefore would be
a common divisor of k, m, n. Analogous consideration can be made about primes dividing both k and n.
Thus there cannot be such primes, i.e., gcd(k, m) = gcd(k, n) = 1.
As the denominator in the representation of s as an irreducible fraction obviously divides k, it suffices to
prove that gcd(k, 6) = 1. For that, we prove that 3 - k and 2 - k.

— 5 —
 Suppose that 3 | k. Then 3 - m and 3 - n, giving m2 ≡ n2 ≡ 1 (mod 3). Hence the left-hand side of
equation (2) is divisible by 3 while the right-hand side is not, a contradiction.
Suppose that 2 | k. Then 2 - m and 2 - n, giving both 2 | m + n and m2 ≡ n2 ≡ 1 (mod 4). Hence the
left-hand side of equation (2) is divisible by 4 while the right-hand side is not, a contradiction.

17. If we denote x = dx1 , y = dy1 , z = dz1 , we get that

x+1 y+1 z+1
S= + +
y z x
d3 (x1 y12 + y1 z12 + z1 x21 ) + d2 (x1 y1 + y1 z1 + z1 x1 )
= .
d3 x1 y1 z1
xy+yz+zx
As S is an integer d is a divisor of x1 y1 + y1 z1 + z1 x1 . Therefore d ≤ d2 , from which the desired
follows.

18. No. Indeed, if integers x, y, z, t satisfy x2 + y 2 − 3z 2 − 3t2 = 0, then x2 + y 2 is a multiple of 3. But squares
are congruent to 0 or 1 modulo 3, so this is possible only if both x, y are divisible by 3. Then the number
x2 + y 2 = 3(z 2 + t2 ) is divisible by 9, the number z 2 + t2 is divisible by 3, and analogously we see that z,
t are divisible by 3. Hence any integer solution to the equation x2 + y 2 − 3z 2 − 3t2 = 0 has all variables
divisible by 3, and since the equation is homogeneous, this implies x = y = z = t = 0.

19. There exist infinitely many nice numbers, so we can find a nice number with at least 10k + 1 digits in its
decimal representation. Let c1 , c2 , . . ., c10k+1 be consecutive digits of this nice number.
By the definition of a nice number, the following two numbers are divisible by r:

a = 10k−1 c1 + 10k−2 c2 + · · · + 10ck−1 + ck ,

b = 10k−1 c2 + 10k−2 c3 + · · · + 10ck + ck+1 .

It follows that the number

10a − b = 10k c1 − ck+1

is divisible by r as well. If we denote di = cik+1 (i = 0, 1, . . . , 10), then by similar calculations we obtain

the divisibilities

r | 10k di − di+1 for i = 0, 1, . . . , 9. (3)

Observe that d0 , d1 , . . ., d10 is a sequence of 11 digits, so some of them must be equal. Consequently,
there exist indices 0 ≤ i < j ≤ 10 such that the digits di , di+1 , . . ., dj−1 are pairwise different and di = dj .
Therefore using (3) we see that the number

(10k di − di+1 ) + (10k di+1 − di+2 ) + · · · + (10k dj−1 − dj )

= (10k di − di+1 ) + (10k di+1 − di+2 ) + · · · + (10k dj−1 − di )
= (10k − 1)(di + di+1 + · · · + dj−1 )

is divisible by r. But the factor di + di+1 + · · · + dj−1 is a sum of distinct digits, so it does not exceed
0 + 1 + 2 + · · · + 9 = 45. Hence this factor is relatively prime to r. This proves that the k-digit number
10k − 1 is divisible by r, and it is therefore a nice number.

20. It it is sufficient to check that for any prime p the maximal power of p that divides ab equals p3m .
Let pk be a maximal power of p that divides a, and p` be a maximal power of p that divides b.

(1) If k = ` then a3 + b3 + ab is divisible by at most p2k , and ab(a − b) is divisible by at least p3k . Therefore,
this case is impossible.

— 6 —
(2) If k > ` then ab(a − b) is divisible by at least pk+2` . The three summands in the first number is
divisible by p3k , p3` and pk+` . Since 3` and k + ` are less than k + 2`, the divisibility of the given
numbers is possible if and only if k + ` = 3` (in this case the sum b3 + ab could be divisible by a power
of p greater than pk+` ). Therefore, we have k = 2`, and hence the maximal power of p that divides ab
is p3` .

— 7 —