Baltic Way 2002 mathematical team contest

Tartu, November 2, 2002

Problems and solutions

1. Solve the system of equations

 3
 a + 3ab2 + 3ac2 − 6abc = 1
b3 + 3ba2 + 3bc2 − 6abc = 1
 3
c + 3ca2 + 3cb2 − 6abc = 1

in real numbers.
Answer: a = 1, b = 1, c = 1 .
Solution. Denoting the left hand sides of the given equations as A , B and C , the following equalities
can easily be seen to hold:

−A + B + C = (−a + b + c)3 ,
A − B + C = (a − b + c)3 ,
A + B − C = (a + b − c)3 .

Hence, the system of equations given in the problem is equivalent to the following one:

 (−a + b + c)3 = 1
(a − b + c)3 = 1 ,
(a + b − c)3 = 1


which gives

 −a + b + c = 1
a−b+c = 1 .
a+b−c = 1


The unique solution of this system is (a, b, c) = (1, 1, 1) .

2. Let a, b, c, d be real numbers such that

a + b + c + d = −2 ,
ab + ac + ad + bc + bd + cd = 0 .

Prove that at least one of the numbers a, b, c, d is not greater than −1 .

Solution. We can assume that a is the least among a, b, c, d (or one of the least, if some of them are
equal), there are n > 0 negative numbers among a, b, c, d , and the sum of the positive ones is x .
Then we obtain

−2 = a + b + c + d > na + x . (1)

Squaring we get

4 = a 2 + b2 + c 2 + d2

which implies

4 6 n · a 2 + x2 (2)

as the square of the sum of positive numbers is not less than the sum of their squares.

1
Combining inequalities (1) and (2) we obtain

na2 + (na + 2)2 > 4 ,

na2 + n2 a2 + 4na > 0 ,
a2 + na2 + 4a > 0 .

As n 6 3 (if all the numbers are negative, the second condition of the problem cannot be satisfied), we
obtain from the last inequality that

4a2 + 4a > 0 ,
a(a + 1) > 0 .

As a < 0 it follows that a 6 −1 .

Alternative solution. Assume that a, b, c, d > −1 . Denoting A = a + 1 , B = b + 1 , C = c + 1 , D = d + 1
we have A, B, C, D > 0 . Then the first equation gives

A + B + C + D = 2. (3)

We also have

ab = (A − 1)(B − 1) = AB − A − B + 1.

Adding 5 similar terms to the last one we get from the second equation

AB + AC + AD + BC + BD + CD − 3(A + B + C + D) + 6 = 0.

In view of (3) this implies

AB + AC + AD + BC + BD + CD = 0,

a contradiction as all the unknowns A, B, C, D were supposed to be positive.

Another solution. Assume that the conditions of the problem hold:

a + b + c + d = −2 (4)
ab + ac + ad + bc + bd + cd = 0. (5)

Suppose that

a, b, c, d > −1. (6)

If all of a, b, c, d were negative, then (5) could not be satisfied, so at most three of them are nega-
tive. If two or less of them were negative, then (6) would imply that the sum of negative numbers,
and hence also the sum a + b + c + d , is greater than 2 · (−1) = −2 , which contradicts (4). So ex-
actly three of a, b, c, d are negative and one is nonnegative. Let d be the nonnegative one. Then
d = −2 − (a + b + c) < −2 − (−1 − 1 − 1) = 1 . Obviously |a|, |b|, |c|, |d| < 1 . Squaring (4) and subtracting
2 times (5), we get

a2 + b2 + c2 + d2 = 4,

but

a2 + b2 + c2 + d2 = |a|2 + |b|2 + |c|2 + |d|2 < 4,

a contradiction.

2
3. Find all sequences a0 6 a1 6 a2 6 . . . of real numbers such that
am2 +n2 = a2m + a2n
for all integers m, n > 0 .
1
Answer: an ≡ 0 , an ≡ and an = n .
2
Solution. Denoting f (n) = an we have

f (m2 + n2 ) = f 2 (m) + f 2 (n) . (7)

1
Substituting m = n = 0 into (7) we get f (0) = 2f 2 (0) , hence either f (0) = or f (0) = 0 . We consider
2
these cases separately.
1 1
(1) If f (0) = then substituting m = 1 and n = 0 into (7) we obtain f (1) = f 2 (1) + , whence
µ ¶2 2 4
1 1
f (1) − = 0 and f (1) = . Now,
2 2

1
f (2) = f (12 + 12 ) = 2f 2 (1) = ,
2
1
f (8) = f (22 + 22 ) = 2f 2 (2) = ,
2
1 1
etc, implying that f (2i ) = for arbitrarily large natural i and, due to monotonity, f (n) = for every
2 2
natural n .
(2) If f (0) = 0 then by substituting m = 1 , n = 0 into (7) we obtain f (1) = f 2 (1) and hence, f (1) = 0
or f (1) = 1 . This gives two subcases.
(2a) If f (0) = 0 and f (1) = 0 then by the same technique as above we see that f (2 i ) = 0 for arbitrarily
large natural i and, due to monotonity, f (n) = 0 for every natural n .
(2b) If f (0) = 0 and f (1) = 1 then we compute

f (2) = f (12 + 12 ) = 2f 2 (1) = 2 ,

f (4) = f (22 + 02 ) = f 2 (2) = 4 ,
f (5) = f (22 + 12 ) = f 2 (2) + f 2 (1) = 5 .

Now,
f 2 (3) + f 2 (4) = f (25) = f 2 (5) + f 2 (0) = 25 ,

hence f 2 (3) = 25 − 16 = 9 and f (3) = 3 . Further,

f (8) = f (22 + 22 ) = 2f 2 (2) = 8 ,

f (9) = f (32 + 02 ) = f 2 (3) = 9 ,
f (10) = f (32 + 12 ) = f 2 (3) + f 2 (1) = 10 .

From the equalities

f 2 (6) + f 2 (8) = f 2 (10) + f 2 (0) ,

f 2 (7) + f 2 (1) = f 2 (5) + f 2 (5)

we also conclude that f (6) = 6 and f (7) = 7 . It remains to note that

(2k + 1)2 + (k − 2)2 = (2k − 1)2 + (k + 2)2 ,

(2k + 2)2 + (k − 4)2 = (2k − 2)2 + (k + 4)2 ,

and by induction it follows that f (n) = n for every natural n .

3
4. Let n be a positive integer. Prove that
n µ ¶2
X
2 1
xi (1 − xi ) 6 1−
i=1
n

for all nonnegative real numbers x1 , x2 , . . . , xn such that

x1 + x 2 + · · · + x n = 1 .
Solution. Expanding the expressions at both sides we obtain the equivalent inequality
X X 2 1
− x3i + 2 x2i − + 2 >0.
i i
n n

It is easy to check that the left hand side is equal to

Xµ ¶µ ¶2
2 1
2 − − xi xi −
i
n n

and hence is nonnegative.

Alternative solution. First note that for n = 1 the required condition holds trivially, and for n = 2 we
have
µ ¶2 µ ¶2
x + (1 − x) 1 1
x(1 − x)2 + (1 − x)x2 = x(1 − x) 6 = = 1− .
2 4 2

So we may further consider the case n > 3 .

2
Assume first that for each index i the inequality xi < holds. Let f (x) = x(1 − x)2 = x − 2x2 + x3 ,
3 · ¸
00 2
then f (x) = 6x − 4 . Hence, the function f is concave in the interval 0, . Thus, from Jensen’s
3
inequality we have
n n µ ¶ µ ¶
X X x1 + . . . + x n 1
xi (1 − xi )2 = f (xi ) 6 n · f =n·f =
i=1 i=1
n n
µ ¶2 µ ¶2
1 1 1
=n· 1− = 1− .
n n n

2
If some xi > then we have
3
µ ¶2
2 2 1
xi (1 − xi ) 6 1 · 1 − = .
3 9

For the rest of the terms we have

X X 1
xj (1 − xj )2 6 xj = 1 − x i 6 .
3
j6=i j6=i

Hence,
n µ ¶2
X 1 124 1
xi (1 − xi ) 6 + = 6 1−
i=1
9 3 9 n

as n > 3 .
5. Find all pairs (a, b) of positive rational numbers such that

√ √ q
√
a+ b= 2+ 3.

4
 µ ¶ µ ¶
1 3 3 1
Answer: (a, b) = , or (a, b) = , .
2 2 2 2

Solution. Squaring both sides of the equation gives

√ √
a + b + 2 ab = 2 + 3 (8)

√ √ √
so 2 √ab = r + 3 for some rational number r . Squaring both sides of this gives 4ab = r 2 + 3 + 2r 3 ,
so 2r 3 is rational, which implies µ
r = 0 . ¶Hence ab = 3/4
µ and¶substituting this into (8) gives a + b = 2 .
1 3 3 1
Solving for a and b gives (a, b) = , or (a, b) = , .
2 2 2 2

6. The following solitaire game is played on an m×n rectangular board, m, n > 2 , divided into unit squares.
First, a rook is placed on some square. At each move, the rook can be moved an arbitrary number of
squares horizontally or vertically, with the extra condition that each move has to be made in the 90 ◦
clockwise direction compared to the previous one (e.g. after a move to the left, the next one has to be
done upwards, the next one to the right etc). For which values of m and n is it possible that the rook
visits every square of the board exactly once and returns to the first square? (The rook is considered to
visit only those squares it stops on, and not the ones it steps over.)

Answer: m, n ≡ 0 mod 2 .

Solution. First, consider any row that is not the row where the rook starts from. The rook has to visit
all the squares of that row exactly once, and on its tour around the board, every time it visits this row,
exactly two squares get visited. Hence, m must be even; a similar argument for the columns shows that
n must also be even.

It remains to prove that for any even m and n such a tour is possible. We will show it by an induction-
like argument. Labelling the squares with pairs of integers (i, j) , where 1 6 i 6 m and 1 6 j 6 n , we
start moving from the square (m/2 + 1, 1) and first cover all the squares of the top and bottom rows in
the order shown in the figure below, except for the squares (m/2 − 1, n) and (m/2 + 1, n) ; note that we
finish on the square (m/2 − 1, 1) .

3 7 8 4

2 6 10 1 9 5

The next square to visit will be (m/2 − 1, n − 1) and now we will cover the rows numbered 2 and n − 1 ,
except for the two middle squares in row 2 . Continuing in this way we can visit all the squares except
for the two middle squares in every second row (note that here we need the assumption that m and n
are even):

3 7 8 4
15 19 11 20 16 12
23 27 28 24
35 39 31 40 36 32
34 38 37 33
22 26 30 21 29 25
14 18 17 13
2 6 10 1 9 5

5
 The rest of the squares can be visited easily:

3 7 47 48 8 4
15 19 11 20 16 12
23 27 43 44 28 24
35 39 31 40 36 32
34 38 42 41 37 33
22 26 30 21 29 25
14 18 46 45 17 13
2 6 10 1 9 5

7. We draw n convex quadrilaterals in the plane. They divide the plane into regions (one of the regions is
infinite). Determine the maximal possible number of these regions.
Answer: The maximal number of regions is 4n2 − 4n + 2 .
Solution. One quadrilateral produces two regions. Suppose we have drawn k quadrilaterals Q 1 , . . . , Qk
and produced ak regions. We draw another quadrilateral Qk+1 and try to evaluate the number of regions
ak+1 now produced. Our task is to make ak+1 as large as possible. Note that in a maximal configuration,
no vertex of any Qi can be located on the edge of another quadrilateral as otherwise we could move this
vertex a little bit to produce an extra region.
Because of this fact and the convexity of the Qj ’s, any one of the four sides of Qk+1 meets at most two
sides of any Qj . So the sides of Qk+1 are divided into at most 2k + 1 segments, each of which potentially
grows the number of regions by one (being part of the common boundary of two parts, one of which is
counted in ak ).
But if a side of Qk+1 intersects the boundary of each Qj , 1 6 j 6 k twice, then its endpoints (vertices
of Qk+1 ) are in the region outside of all the Qj -s, and the the segments meeting at such a vertex are
on the boundary of a single new part (recall that it makes no sense to put vertices on edges of another
quadrilaterals). This means that ak+1 − ak 6 4(2k + 1) − 4 = 8k . By considering squares inscribed in a
circle one easily sees that the situation where ak+1 − ak = 8k can be reached.
It remains to determine the expression for the maximal ak . Since the difference ak+1 − ak is lin-
ear in k , ak is a quadratic polynomial in k , and a0 = 2 . So ak = Ak 2 + Bk + 2 . We have
8k = ak+1 − ak = A(2k + 1) + B for all k . This implies A = 4 , B = −4 , and an = 4n2 − 4n + 2 .

8. Let P be a set of n > 3 points

µ in ¶ the plane, no three of which are on a line. How many possibilities are
n−1
there to choose a set T of triangles, whose vertices are all in P , such that each triangle in T
2
has a side that is not a side of any other triangle in T ?
Answer: There is one possibility for n = 3 and n possibilities for n > 4 .
Solution. For a fixed pointµ x ∈ P¶, let Tx be the set of all triangles with vertices in P which have x as
n−1
a vertex. Clearly, |Tx | = , and each triangle in Tx has a side which is not a side of any other
2
triangle in Tx . For any x, y ∈ P such that x 6= y , we have Tx 6= Ty if and only if n > 4 . We will show
that any possible set T is equal to Tx for some x ∈ P , i.e. that the answer is 1 for n = 3 and n for
n > 4.
Let
½ µ ¶¾ ½ µ ¶¾
n−1 n−1
T = ti : i = 1, 2, . . . , , S= si : i = 1, 2, . . . ,
2 2

such that T is a set of triangles whose vertices are all µ in¶ P , and si is a side of ti but not of any tj ,
n
j 6= i . Furthermore, let C be the collection of all the triangles whose vertices are in P . Note that
3
µ ¶ µ ¶ µ ¶
n n−1 n−1
|C \ T | = − = .
3 2 3

6
 Let m be the number of pairs (s, t) such that s ∈ S is a side of t ∈ C \ T . Since every s ∈ S is a side
of exactly n − 3 triangles from C \ T , we have
µ ¶ µ ¶
n−1 n−1
m = |S| · (n − 3) = · (n − 3) = 3 · = 3 · |C \ T | .
2 3

On the other hand, every t ∈ C \ T has at most three sides from S . By the above equality, for every
t ∈ C \ T , all its sides must be in S .
Assume that for p ∈ P there is a side s ∈ S such that p is an endpoint of s . Then p is also a vertex
of each of the n − 3 triangles in C \ T which have s as a side. Consequently, p is an endpoint of n − 2
sides in S . Since every side in S has exactly 2 endpoints, the number of points p ∈ P which occur as a
vertex of some s ∈ S is
µ ¶
2 · |S| 2 n−1
= · =n−1.
n−2 n−2 2

Consequently, there is an x ∈ P which is not an endpoint of any s ∈ S , and hence T must be equal to
Tx .
9. Two magicians show the following trick. The first magician goes out of the room. The second magician
takes a deck of 100 cards labelled by numbers 1, 2, . . . , 100 and asks three spectators to choose in turn
one card each. The second magician sees what card each spectator has taken. Then he adds one more
card from the rest of the deck. Spectators shuffle these 4 cards, call the first magician and give him these
4 cards. The first magician looks at the 4 cards and “guesses” what card was chosen by the first spectator,
what card by the second and what card by the third. Prove that the magicians can perform this trick.
Solution. We will identify ourselves with the second magician. Then we need to choose a card in such a
manner that another magician will be able to understand which of the 4 cards we have chosen and what
information it gives about the order of the other cards. We will reach these two goals independently.
Let a , b , c be remainders of the labels of the spectators’ three cards modulo 5. There are three possible
cases.
1) All the three remainders coincide. Then choose a card with a remainder not equal to the remainder
of spectators’ cards. Denote this remainder d .
Note that we now have 2 different remainders, one of them in 3 copies (this will be used by the first
magician to distinguish betwwen the three cases). To determine which of the cards is chosen by us is
now a simple exercise in division by 5. But we must also encode the ordering of the spectators’ cards.
These cards have a natural ordering by their labels, and they are also ordered by their belonging to
the spectators. Thus, we have to encode a permutation of 3 elements. There are 6 permutations of 3
elements, let us enumerate them somehow. Then, if we want to inform the first magician that spectators
form a permutation number k with respect to the natural ordering, we choose the card number 5k + d .
2) The remainders a , b , c are pairwise different. Then it is clear that exactly one of the following
possibilities takes place:

either |b − a| = |a − c|, or |a − b| = |b − c|, or |a − c| = |c − b| (9)

(the equalities are considered modulo 5). It is not hard to prove it by a case study, but one could also
imagine choosing three vertices of a regular pentagon — these vertices always form an isosceles, but not
an equilateral triangle.
Each of these possibilities has one of the remainders distinguished from the other two remainders (these
distinguished remainders are a , b , c , respectively). Now, choose a card from the rest of the deck
having the distinguished remainder modulo 5. Hence, we have three different remainders, one of them
distinguished by (9) and presented in two copies. Let d be the distinguished remainder and s = 5m + d
be the spectator’s card with this remainder.
Now we have to choose a card r with the remainder d such that the first magician would be able to
understand which of the cards s and r was chosen by us and what permutation of spectators it implies.
This can be done easily: if we want to inform the first magician that spectators form a permutation
number k with respect to the natural ordering, we choose the card number s + 5k (mod 100) .

7
 The decoding procedure is easy: if we have two numbers p and q that have the same remainder modulo 5,
calculate p − q (mod 100) and q − p (mod 100) . If p − q (mod 100) > q − p (mod 100) then r = q is our
card and s = p is the spectator’s card. (The case p − q (mod 100) = q − p (mod 100) is impossible since
the sum of these numbers is equal to 100 , and one of them is not greater than 6 · 5 = 30 .)
3) Two remainders (say, a and b ) coincide. Let us choose a card with the remainder d = (a+c)/2 mod 5 .
Then |a − d| = |d − c| mod 5 , so the remainder d is distinguished by (9). Hence we have three different
remainders, one of them distinguished by (9) and one of the non-distinguished remainders presented in
two copies. The first magician will easily determine our card, and the rule to choose the card in order to
enable him also determine the order of spectators is similar to the one in the 1-st case.
Alternative solution. This solution gives a non-constructive proof that the trick is possible. For this,
we need to show there is an injective mapping from the set of ordered triples to the set of unordered
quadruples that additionally respects inclusion.
To prove that the desired mapping exists, let’s consides a bipartite graph such that the set of ordered
triples T and the set of unordered quadruples Q form the two disjoint sets of vertices and there is an
edge between a triple and a quadruple if and only if the triple is a subset of the quadruple.
For each triple t ∈ T , we can add any of the remaining 97 cards to it, and thus we have 97 different
quadruples connected to each triple in the graph. Conversely, for each quadruple q ∈ Q , we can remove
any of the 4 cards from it, and reorder the remaining 3 cards in 3! = 6 different ways, and thus we have
24 different triples connected to each quadruple in the graph.
According to the Hall’s theorem, a bipartite graph G = (T, Q, E) has a perfect matching if and only if
for each subset T 0 ⊆ T the set of neighbours of T 0 , denoted N (T 0 ) , satisfies |N (T 0 )| > |T 0 | .
To prove that this condition holds for our graph, consider any subset T 0 ⊆ T . Because we have 97
quadruples for each triple, and there can be at most 24 copies of each of them in the multiset of neighbours,
97 0
we have |N (T 0 )| > |T | > 4|T 0 | , which is even much more than we need.
24
Thus, the desired mapping is guaranteed to exist.
Another solution. Let the three chosen numbers be (x1 , x2 , x3 ) . At least one of the sets {1, 2, . . . , 24} ,
{25, 26, . . . , 48} , {49, 50, . . . , 72} and {73, 74, . . . , 96} should contain none of x 1 , x2 and x3 , let S be
such set. Next we split S into 6 parts: S = S1 ∪ S2 ∪ . . . ∪ S6 so that 4 first elements of S are in S1 ,
four next in S2 , etc. Now we choose i ∈ {1, 2, . . . , 6} corresponding to the order of numbers x 1 , x2 and
x3 (if x1 < x2 < x3 then i = 1 , if x1 < x3 < x2 then i = 2 , . . . ,if x3 < x2 < x1 then i = 6 ). At last
let j be the number of elements in {x1 , x2 , x3 } that are greater than elements of S (note that any xk ,
k ∈ {1, 2, 3} , is either greater or smaller than all the elements of S ). Now we choose x 4 ∈ Si so that
x1 + x2 + x3 + x4 ≡ j mod 4 and add the card number x4 to those three cards.
Decoding of {a, b, c, d} is straightforward. We first put the numbers into increasing order and then
calculate a+b+c+d mod 4 showing the added card. The added card belongs to some S i ( i ∈ {1, 2, . . . , 6} )
for some S and i shows us the initial ordering of cards.

10. Let N be a positive integer. Two persons play the following game. The first player writes a list of
positive integers not greater than 25, not necessarily different, such that their sum is at least 200. The
second player wins if he can select some of these numbers so that their sum S satisfies the condition
200 − N 6 S 6 200 + N . What is the smallest value of N for which the second player has a winning
strategy?
Answer: N = 11 .
Solution. If N = 11 , then the second player can simply remove numbers from the list, starting with the
smallest number, until the sum of the remaining numbers is less than 212 . If the last number removed
was not 24 or 25 , then the sum of the remaining numbers is at least 212 − 23 = 189 . If the last
number removed was 24 or 25 , then only 24 -s and 25 -s remain, and there must be exactly 8 of them
since their sum must be less than 212 and not less than 212 − 24 = 188 . Hence their sum S satisfies
8 · 24 = 192 6 S 6 8 · 25 = 200 . In any case the second player wins.
On the other hand, if N 6 10 , then the first player can write 25 two times and 23 seven times. Then
the sum of all numbers is 211 , but if at least one number is removed, then the sum of the remaining ones
is at most 188 — so the second player cannot win.

8
11. Let n be a positive integer. Consider n points in the plane such that no three of them are collinear and
no two of the distances between them are equal. One by one, we connect each point to the two points
nearest to it by line segments (if there are already other line segments drawn to this point, we do not
erase these). Prove that there is no point from which line segments will be drawn to more than 11 points.
Solution. Suppose there exists a point A such that A is connected to twelve points. Then there exist
three points B , C and D such that ∠BAC 6 60◦ , ∠BAD 6 60◦ and ∠CAD 6 60◦ .
We can assume that |AD| > |AB| and |AD| > |AC| . By the cosine law we have

|BD|2 = |AD|2 + |AB|2 − 2|AD||AB| cos ∠BAD

< |AD|2 + |AB|2 − 2|AB|2 cos ∠BAD
= |AD|2 + |AB|2 (1 − 2 cos ∠BAD)
6 |AD|2

since 1 6 2 cos(∠BAD) . Hence |BD| < |AD| . Similarly we get |CD| < |AD| . Hence A and D should
not be connected which is a contradiction.
Comment. It would be interesting to know whether 11 can be achieved or the actual bound is lower.
12. A set S of four distinct points is given in the plane. It is known that for any point X ∈ S the remaining
points can be denoted by Y , Z and W so that

|XY | = |XZ| + |XW |.

Prove that all the four points lie on a line.

Solution. Let S = {A, B, C, D} and let AB be the longest of the six segments formed by these four
points (if there are several longest segments, choose any of them). If we choose X = A then we must also
choose Y = B . Indeed, if we would, for example, choose Y = C , we should have |AC| = |AB| + |AD|
contradicting the maximality of AB . Hence we get

|AB| = |AC| + |AD| . (10)

Similarly, choosing X = B we must choose Y = A and we obtain

|AB| = |BC| + |BD| . (11)

On the other hand, from the triangle inequality we know that

|AB| 6 |AC| + |BC| ,

|AB| 6 |AD| + |BD| ,

where at least one of the inequalities is strict if all the four points are not on the same line. Hence, adding
the two last inequalities we get

2|AB| < |AC| + |BC| + |AD| + |BD| .

On the other hand, adding (10) and (11) we get

2|AB| = |AC| + |AD| + |BC| + |BD| ,

a contradiction.
13. Let ABC be an acute triangle with ∠BAC > ∠BCA , and let D be a point on side AC such that
|AB| = |BD| . Furthermore, let F be a point on the circumcircle of triangle ABC such that line F D
is perpendicular to side BC and points F , B lie on different sides of line AC . Prove that line F B is
perpendicular to side AC .
Solution. Let E be the other point on the circumcircle of triangle ABC such that |AB| = |EB| . Let
D0 be the point of intersection of side AC and the line perpendicular to side BC , passing through E .
Then ∠ECB = ∠BCA and the triangle ECD 0 is isosceles. As ED 0 ⊥ BC , the triangle BED 0 is also
isosceles and |BE| = |BD 0 | implying D = D 0 . Hence, the points E , D , F lie on one line. We now have

∠EF B + ∠F DA = ∠BCA + ∠EDC = 90◦ .

9
 The required result now follows.

PSfrag replacements E

A D C

14. Let L , M and N be points on sides AC , AB and BC of triangle ABC , respectively, such that BL
is the bisector of angle ABC and segments AN , BL and CM have a common point. Prove that if
∠ALB = ∠M N B then ∠LN M = 90◦ .

Solution. Let P be the intersection point of lines M N and AC . Then ∠P LB = ∠P N B and the
quadrangle P LN B is cyclic. Let ω be its circumcircle. It is sufficient to prove that P L is a diameter
of ω .

Let Q denote the second intersection point of the line AB and ω . Then ∠P QB = ∠P LB and

∠QP L = ∠QBL = ∠LBN = ∠LP N ,

and the triangles P AQ and BAL are similar. Therefore,

|P Q| |BL|
= . (12)
|P A| |BA|

We see that the line P L is a bisector of the inscribed angle N P Q . Now in order to prove that P L is
a diameter of ω it is sufficient to check that |P N | = |P Q| .

The triangles N P C and LBC are similar, hence

|P N | |BL|
= . (13)
|P C| |BC|

Note also that

|AB| |AL|
= . (14)
|BC| |CL|

by the properties of a bisector. Combining (12), (13) and (14) we have

|P N | |AL| |CP |
= · .
|P Q| |AP | |CL|

We want to prove that the left hand side of this equality equals 1. This follows from the fact that the
quadruple of points (P, A, L, C) is harmonic, as can be proven using standard methods (e.g. considering
the quadrilateral M BN S , where S = M C ∩ AN ).

10
PSfrag replacements
A
B B
C N
D M
E
F

C P
L A

Q
15. A spider and a fly are sitting on a cube. The fly wants to maximize the shortest path to the spider
along the surface of the cube. Is it necessarily best for the fly to be at the point opposite to the spider?
(“Opposite” means “symmetric with respect to the center of the cube”.)
Answer: no.
Solution. Suppose that the side of the cube is 1 and the spider sits at the middle of one of the edges.
Then the shortest path to the middle of the opposite edge has length 2 . However, if the fly goes to a
point on this edge at distance s from the middle, then the length of the shortest path is
 s 
µ ¶2
p 9 3
min  4 + s2 , + −s  .
4 2
√
If 0 < s < (3 − 7)/2 then this expression is greater than 2 .
16. Find all nonnegative integers m such that
¢2
am = 22m+1 + 1
¡

is divisible by at most two different primes.

Answer: m = 0, 1, 2 are the only solutions.
Solution. Obviously m = 0, 1, 2 are solutions as a0 = 5 , a1 = 65 = 5 · 13 , and a2 = 1025 = 25 · 41 . We
show that these are the only solutions.
Assume that m > 3 and that am contains at most two different prime factors. Clearly, am = 42m+1 + 1
is divisible by 5 , and
am = 22m+1 + 2m+1 + 1 · 22m+1 − 2m+1 + 1 .
¡ ¢ ¡ ¢

The two above factors are relatively prime as they are both odd and their difference is a power of 2 .
Since both factors are larger than 1 , one of them must be a power of 5 . Hence,
2m+1 · (2m ± 1) = 5t − 1 = (5 − 1) · (1 + 5 + · · · + 5t−1 )
for some positive integer t , where ± reads as either plus or minus. For odd t the right hand side is not
divisible by 8 , contradicting m > 3 . Therefore, t must be even and

2m+1 · (2m ± 1) = (5t/2 − 1) · (5t/2 + 1) .

Clearly, 5t/2 + 1 ≡ 2 ( mod 4) . Consequently, 5t/2 − 1 = 2m · k for some odd k , and 5t/2 + 1 = 2m · k + 2
divides 2(2m ± 1) , i.e.

2m−1 · k + 1 | 2m ± 1 .
This implies k = 1 , finally leading to a contradiction since
2m−1 + 1 < 2m ± 1 < 2(2m−1 + 1)
for m > 3 .

11
17. Show that the sequence
µ ¶ µ ¶ µ ¶
2002 2003 2004
, , ,... ,
2002 2002 2002
considered modulo 2002, is periodic.
Solution. Define
µ ¶
n
xkn =
k
and note that
µ ¶ µ ¶ µ ¶
n+1 n n
xkn+1 − xkn = − = = xnk−1 .
k k k−1

Let m be any positive integer. We will prove by induction on k that the sequence {x kn }∞n=k is periodic
modulo m . For k = 1 it is obvious that xkn = n is periodic modulo m with period m . Therefore it
will suffice to show that the following is true: the sequence {xn } is periodic modulo m if its difference
sequence, dn = xn+1 − xn , is periodic modulo m .
Furthermore, if t then the period of {xn } is equal to ht where h is the smallest positive integer such
that h(xt − x0 ) ≡ 0 modulo m .
Indeed, let t be the period of {dn } and h be the smallest positive integer such that h(xt − x0 ) ≡ 0
modulo m . Then
 
n+ht−1
X n−1
X t−1
X
xn+ht = x0 + dj = x 0 + dj + h  dj  =
j=0 j=0 j=0

= xn + h(xt − x0 ) ≡ xn (mod m)

for all n , so the sequence {xn } is in fact periodic modulo m (with a period dividing ht ).
18. Find all integers n > 1 such that any prime divisor of n6 − 1 is a divisor of (n3 − 1)(n2 − 1) .
Solution. Clearly n = 2 is such an integer. We will show that there are no others.
Consider the equality
n6 − 1 = (n2 − n + 1)(n + 1)(n3 − 1) .

The integer n2 − n + 1 = n(n − 1) + 1 clearly has an odd divisor p . Then p | n3 + 1 . Therefore, p does
not divide n3 − 1 and consequently p | n2 − 1 . This implies that p divides (n3 + 1) + (n2 − 1) = n2 (n + 1) .
As p does not divide n , we obtain p | n + 1 . Also, p | (n2 − 1) − (n2 − n + 1) = n − 2 . From p | n + 1
and p | n − 2 it follows that p = 3 , so n2 − n + 1 = 3r for some positive integer r .
The discriminant of the quadratic n2 − n + (1 − 3r ) must be a square of an integer, hence

1 − 4(1 − 3r ) = 3(4 · 3r−1 − 1)

must be a squareof an integer. Since for r > 2 the number 4 · 3r−1 − 1 is not divisible by 3 , this is
possible only if r = 1 . So n2 − n − 2 = 0 and n = 2 .
19. Let n be a positive integer. Prove that the equation
1 1
x+y+ + = 3n
x y
does not have solutions in positive rational numbers.
p r
Solution. Suppose x = and y = satisfy the given equation, where p, q, r, s are positive integers
q s
and gcd(p, q) = 1 , gcd(r, s) = 1 . We have
p r q s
+ + + = 3n ,
q s p r

12
 or

(p2 + q 2 )rs + (r 2 + s2 )pq = 3npqrs ,

so rs | (r 2 + s2 )pq . Since gcd(r, s) = 1 , we have gcd(r 2 + s2 , rs) = 1 and rs | pq . Analogously pq | rs ,

so rs = pq and hence there are either two or zero integers divisible by 3 among p, q, r, s . Now we have

(p2 + q 2 )rs + (r 2 + s2 )rs = 3n(rs)2 ,

p2 + q 2 + r2 + s2 = 3nrs ,

but 3nrs ≡ 0 (mod 3) and p2 + q 2 + r2 + s2 is congruent to either 1 or 2 modulo 3 , a contradiction.

20. Does there exist an infinite non-constant arithmetic progression, each term of which is of the form a b ,
where a and b are positive integers with b > 2 ?
Answer: no.
Solution. For an arithmetic progression a1 , a2 , . . . with difference d the following holds:

1 1 1 1 1 1
Sn = + + ... + = + + ... + >
a1 a2 an+1 a1 a1 + d a1 + nd
µ ¶
1 1 1 1
> + + ... + ,
m 1 2 n+1

where m = max(a1 , d) . Therefore Sn tends to infinity when n increases.

On the other hand, the sum of reciprocals of the powers of a natural number x 6= 1 is
1
1 1 x2 1
+ 3 + ... = = .
x2 x 1 x(x − 1)
1−
x
Hence, the sum of reciprocals of the terms of the progression required in the problem cannot exceed
µ ¶
1 1 1 1 1 1 1
+ + + ... = 1 + − + − + ... = 2 ,
1 1·2 2·3 1 2 2 3

a contradiction.
Alternative solution. Let ak = a0 + dk , k = 0, 1, . . . . Choose a prime number p > d and set
k 0 ≡ (p − a0 )d−1 mod p2 . Then ak0 = a0 + k 0 d ≡ p mod p2 and hence, ak0 can not be a power of
a natural number.
√ √
Another solution. There can be at most b nc squares in the set {1, 2, . . . , n} , at most b 3 nc cubes in
the same set, etc. The greatest power that can occur in the set {1, 2, . . . , n} is blog 2 nc and thus there
are no more than
√ √ √
b nc + b 3 nc + . . . + b blog2 nc nc

powers among the numbers 1, 2, . . . , n . Now we can estimate this sum above:
√ √ √ √
b nc + b 3 nc + . . . + b blog2 nc nc 6 b nc(blog2 nc − 1) <
√
< b nc · blog2 nc = o(n).

This means that every arithmetic progression grows faster than the share of powers.

13